Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized FULL REPORT A Poverty and Equity Assessment of Djibouti TO INCLUSIVE GROWTH CHALLENGES A March 2019 FULL REPORT CHALLENGES TO INCLUSIVE GROWTH A Poverty and Equity Assessment of Djibouti TABLE OF CONTENTS Acknowledgments___________________________________________________________________ XII Acronyms____________________________________________________________________________XIII OVERVIEW__________________________________________________________________________XIV Welfare and poverty are distributed unequally in Djibouti_________________________________________________XIV Location shapes access to services, with large coverage gaps between urban and rural areas, but the labor market opportunities remain low for urban and rural poor alike______________________ XX Economic growth in recent years has been accompanied by an improvement in well-being of households, as measured by nonmonetary indicators___________________________________ XXIII Looking forward, investments in human capital may contribute to poverty reduction if the government ensures the benefits are evenly distributed in the population___________________________ XXVI The full potential of the labor market as a sustainable driver out of poverty is still untapped—over half of the working-age population is not in the labor force__________________________ XXIX While the public sector continues to draw highly skilled individuals, the rest of the employed workforce is informal_________________________________________________________XXXII The employment profile highlights the need to continue addressing labor demand-side issues___________________________________________________________________________XXXV Djibouti must take full advantage of its growing economy to strengthen the poverty reduction angle in public policies_________________________________________________________ XXXVI References__________________________________________________________________________ XL FULL REPORT III Chapter 1 : Welfare and Poverty in Djibouti_____________________________________________ 1 Introduction and macroeconomic context_____________________________________________________ 1 Analysis of welfare and monetary poverty_____________________________________________________ 3 Main source of data__________________________________________________________________ 3 The welfare aggregate________________________________________________________________ 4 Poverty lines_______________________________________________________________________ 7 Poverty in Djibouti in 2017_____________________________________________________________ 8 Poverty profile________________________________________________________________________ 13 Location characteristics_____________________________________________________________ 13 Demographic characteristics__________________________________________________________ 13 Nonmonetary indicators_____________________________________________________________ 15 Economic characteristics____________________________________________________________ 23 Trends in well-being________________________________________________________________ 25 Inequality in Djibouti____________________________________________________________________ 31 Monetary inequality_________________________________________________________________ 31 Deprivation in Djibouti: Human opportunity index___________________________________________ 35 Comparing the urban poor and nonpoor in the capital: The case of Balbala_______________________ 38 References___________________________________________________________________________ 45 Chapter 2 : Human Capital and Labor Market__________________________________________47 Human capital: Educational outcomes in Djibouti______________________________________________ 47 Intergenerational transformation in Djibouti___________________________________________________ 49 Labor market outcomes_________________________________________________________________ 53 References___________________________________________________________________________ 64 Chapter 3 : Selected Topics to Inform Public Policy in Djibouti__________________________67 Introducing competition to the telecom sector________________________________________________ 67 Provision of ICT services in Djibouti_____________________________________________________ 69 Potential impact of a new entrant on retail prices in Djibouti___________________________________ 72 Simulating increased competition in the ICT sector in Djibouti_________________________________ 74 Conclusion_______________________________________________________________________ 76 Improved targeting of social programs______________________________________________________ 77 A proxy means test (PMT) application using EDAM4-IS______________________________________ 79 Conclusion_______________________________________________________________________ 86 Nomads and pastoralists________________________________________________________________ 87 References___________________________________________________________________________ 93 IV CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Appendix A. Note on EDAM4-IS 2017 and Sampling____________________________________95 The Djibouti Household Survey (EDAM4-IS)__________________________________________________ 95 Sampling and extrapolation coefficients of the EDAM4-IS________________________________________ 96 Data collection____________________________________________________________________ 97 Sampling weights__________________________________________________________________ 97 Does the EDAM represent the actual Djiboutian population?__________________________________ 98 References__________________________________________________________________________ 100 Appendix B. Note on Poverty Measurement Methodology____________________________ 101 The aggregate of consumption as a measure of well-being______________________________________ 101 Expenditures on food products_______________________________________________________ 101 Expenditure on nonfood products_____________________________________________________ 104 Durable goods___________________________________________________________________ 104 Rent and housing services___________________________________________________________ 106 Analysis of the consumption aggregate____________________________________________________ 109 Household composition_____________________________________________________________ 110 Temporal and spatial adjustment______________________________________________________ 110 Estimation of poverty line_______________________________________________________________ 111 Estimation of the food poverty line_____________________________________________________ 111 Food poverty threshold_____________________________________________________________ 113 Extreme poverty threshold___________________________________________________________ 113 Overall poverty threshold____________________________________________________________ 113 References__________________________________________________________________________ 116 FULL REPORT V FIGURES Figure O.1 Annual Consumption per Capita per Decile (Djiboutian francs, DF)_________________________XIV Figure O.2 Extreme Poverty Rates, by EDAM4 Representative Domains______________________________ XV Figure O.3 Household Characteristics, by Region and Year_______________________________________ XXII Figure O.4 Literacy Rates by Age Groups, 2017 (percent)_______________________________________ XXIV Figure O.5 Educational Attainment by Age Cohort, 2017 (percent)__________________________________ XXV Figure O.6 Educational Mobility of Individuals 25 Years and Older Compared to Their Fathers (percent)____ XXV Figure O.7 Distribution of Population in the Labor Force (percent)_________________________________ XXVII Figure O.8 Employment Rate Among Labor Market Participants, by Educational Attainment____________XXVIII Figure BO.2.1 Distribution of Youth (15–24 years) across Education and Labor Force Activities_____________ XXIX Figure 1.1 Macroeconomic Indicators in Djibouti_______________________________________________ 2 Figure 1.2 Annual per Adult Equivalent Consumption, by Area and District in Djibouti City (DF)____________ 6 Figure 1.3 Share of Expenditure on Each Category_____________________________________________ 7 Figure 1.4 Extreme Poverty Rates, by EDAM4 Representative Domains_____________________________ 10 Figure 1.5 Poverty Rates Based on International (US$1.90 and US$3.20 2011 PPP per day) and Societal Poverty Lines______________________________________________________ 12 Figure 1.6 Poverty Rate (percent), by Number of Household Members_____________________________ 14 Figure 1.7 Dwelling Characteristics and Access to Services by Population Groups____________________ 16 Figure 1.8 Educational Attainment of Adults Age 25 and Older (percent)____________________________ 18 Figure 1.9 Extreme Poverty Rates, by Educational Attainment of Household Head_____________________ 19 Figure 1.10 Education Expenditure per Capita and Share of Expenditure on Education in Total Household Expenditure, 2017________________________________________________ 19 Figure 1.11 Deprivation score for Population subgroups_________________________________________ 22 Figure 1.12 Employment Status of All Age 15 and Older, by Poverty Status (percent)____________________ 23 Figure 1.13 Type and Sector of Employment (percent of those employed, 15 years and older)_____________ 23 Figure 1.14 Access to Services, by Year (percent of population)___________________________________ 25 Figure 1.15 Dwelling Characteristics and Access to Goods, by Year (percent of population)_______________ 26 Figure 1.16 Comparison of Poverty Rates between 2013 and 2017, Djibouti City_______________________ 30 Figure 1.17 Annual Consumption per Capita, per Decile_________________________________________ 31 Figure 1.18 Annual Consumption per Capita, across Regions and Districts of Djibouti City (DF)____________ 32 Figure 1.19 Gini Coefficient across the World, circa 2015________________________________________ 33 Figure 1.20 Self-Reported Welfare Category, by Quintiles of Consumption per Capita (percent)____________ 35 Figure 1.21 Coverage Rates and Human Opportunity Index for Children’s Opportunities_________________ 37 Figure 1.22 Inequality Decomposition of Disparities in Access to Opportunities________________________ 38 Figure 1.23 Household Characteristics by District in Djibouti City__________________________________ 39 VI CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Figure 2.1 Literacy Rates among People Age 15 Years and Older, 2017 (percent)______________________ 48 Figure 2.2 Percentage of Population Age 15 Years and Older with at Least Primary Education, 2017_______ 48 Figure 2.3 Literacy Rates, 2017 (percent)___________________________________________________ 49 Figure 2.4 Educational Attainment by Age Cohort, 2017________________________________________ 50 Figure 2.5 Educational Mobility of Individuals 25 Years and Older Compared to Their Fathers’ (percent)_____ 51 Figure 2.6 Distribution of Population in the Labor Market (percent)________________________________ 53 Figure 2.7 Employment Rate among Labor Market Participants, by Educational Attainment______________ 54 Figure 2.8 Distribution of Youth Ages 15–24 Years across Education and Labor Force Activities___________ 55 Figure 3.1 Selected ICT World Rankings for MENA and SSA countries_____________________________ 68 Figure 3.2 Digital gap between the richest and poorest households by quintile_______________________ 70 Figure 3.3 Digital gap between urban and rural households______________________________________ 70 Figure 3.4 Travel Distance from Telephone Service Infrastructure__________________________________ 71 Figure 3.5 Social Assistance Expenditure as a Percentage of GDP________________________________ 77 Figure 3.6 Errors of Inclusion and Exclusion for PMT Model in Djibouti Compared to Other Countries (cutoff point at 30th percentile)___________________________________________________ 82 Figure 3.7 Average Temperature in Djibouti (Celsius)___________________________________________ 88 Figure 3.8 Percentage of Population Owning and/or Raising Animals_______________________________ 89 Figure 3.9 Percentage of Population Practicing Transhumance and Is Pastoralist______________________ 90 Figure 3.10 Extreme (Monetary) Poverty Rate among Various Population Groups_______________________ 91 FULL REPORT VII TABLES Table O.1 Characteristics of Households, by Poverty Status_____________________________________ XVII Table O.2 Characteristics of the Population by Poverty Status and Location__________________________XIX Table O.3 Perceptions on the Evolution of Poverty (percent population by response category)___________XXIII Table O.4 Education Transition Matrix of Poor Women (25 years and older)_________________________ XXVI Table BO.2.1 Distribution of Youth (15–24 years old) in and out of the Labor Force______________________ XXIX Table O.5 Distribution of Employed Individuals by Branch of Activity and Sector Of Employment (percent)_____ XXX Table O.6 Distribution of Employed Individuals by Firm Size and Sector Of Employment (percent)______ XXXI Table O.7 Distribution of Employed Individuals by Educational Attainment and Sector of Employment____ XXXII Table O.8 Characteristics by Sector Of Employment___________________________________________ XXXII Table 1.1 Distribution of the EDAM4 Sample and Estimated Population, by Region_____________________ 4 Table 1.2 Poverty Indicators in Djibouti, 2017__________________________________________________ 10 Table 1.3 Indicators of Poverty in Djibouti City, 2017____________________________________________ 11 Table 1.4 Distribution of Extreme Poor and Overall Population across Locations (% population)__________ 14 Table 1.5 Demographics and Characteristics of Extreme Poor, Nonpoor, and Overall Population__________ 15 Table 1.6 Extreme Poverty Rate, by Population Group___________________________________________ 16 Table 1.7 Asset Ownership Rates by Poverty Status____________________________________________ 17 Table 1.8 Education Indicators among the Extreme Poor, Nonpoor, and Overall Population______________ 18 Table 1.9 Health Indicators among the Extreme Poor, Nonpoor, and Overall Population_________________ 20 Table B1.3.1 Dimensions and Their Corresponding Weights_________________________________________ 21 Table 1.10 Percentage of Population Deprived on Various Dimensions of the MPI______________________ 22 Table 1.11 Percentage of Population with Access to Any Source to Finance Consumption_______________ 24 Table 1.12 Perceptions on the Evolution of Poverty (percent population by response category)____________ 27 Table 1.13 Perceptions on the Evolution of Poverty by Location and Year_____________________________ 28 Table 1.14 Indicators of Inequality___________________________________________________________ 33 Table 1.15 Access to Services, By Quantiles___________________________________________________ 34 Table 1.16 Opportunities and Circumstances Used in Human Opportunity Index Calculation______________ 36 Table 1.17 Characteristics of Poor and Nonpoor, by Area of Residence______________________________ 40 Table 2.1 Education Transition Matrices of Individuals 25 Years and Older Based on Their Father’s Education_ 52 Table B2.1.1 DistriDimensions and Their Corresponding Weights_____________________________________ 53 Table 2.2 Labor Force Participation Rates, by Groups of Population________________________________ 53 Table 2.3 Reasons for Not Wanting to Find Employment, among the Youth 15–24 Years Old (percent)_____ 55 Table 2.4 Distribution of Employed Individuals by Branch of Activity and Sector of Employment (percent)___ 56 Table 2.5 Distribution of Employed Individuals by Firm Size and Sector of Employment (percent)_________ 57 Table 2.6 Distribution of Employed individuals by Educational Attainment and Sector of Employment (percent)__ 58 Table 2.7 Wages Earned in Private and Public Enterprises by Status of Employment___________________ 59 VIII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Table 2.8 Returns to Education_____________________________________________________________ 60 Table 2.9 Characteristics by Sector of Employment of Household Heads____________________________ 61 Table 2.10 Share of Various Income Sources in Total Income_______________________________________ 61 Table 3.1 Djibouti Telecom Costs, by Service (Djiboutian francs, DF)________________________________ 71 Table 3.2 Average per Capita Consumption (DF by Quintile_______________________________________ 72 Table 3.3 Summary Statistics on Telecom Expenditures and Users_________________________________ 74 Table 3.4 WELCOM Simulations____________________________________________________________ 75 Table 3.5 Estimated Increase in Probability Due to Welfare Gains__________________________________ 76 Table 3.6 Social Safety Net Programs in Djibouti_______________________________________________ 78 Table 3.7 Inclusion and Exclusion Errors______________________________________________________ 79 Table 3.8 Performance of National and Flexible Model___________________________________________ 81 Table 3.9 Performance of Flexible Model for Subsets of Population________________________________ 83 Table 3.10 Errors of Exclusion and Inclusion for Truncated Model with Predictions for the Whole and for the Restricted Sample________________________________________________ 83 Table 3.11 Simulation Results on Poverty Rate after Hypothetical Expansion of PNSF___________________ 85 Table 3.12 Simulation Results on Poverty Rate after Alternative Hypothetical Expansion of PNSF__________ 86 Table 3.13 Characteristics of Households in the Regions That Practice Transhumance and Those That Do Not___ 91 FULL REPORT IX MAPS Map O.1 Djibouti’s Location in Africa________________________________________________________ XIII Map O.2 Catchment Areas of a One Kilometer Radius of Primary Schools and Hospitals in Djibouti City___ XX Map 1.1 Catchment Areas of Primary Schools in Djibouti City (One Kilometer)_______________________ 41 Map 1.2 Catchment Areas of Hospitals in Djibouti City (One Kilometer)_____________________________ 42 Map 1.3 Clusters and Density of Clusters for Regions Outside Djibouti City_________________________ 43 Map 3.1 Major Livestock Migration Routes in Djibouti__________________________________________ 87 X CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI BOXES Box O.1 The Fourth Djiboutian Household Survey (EDAM4)______________________________________XVI Box O.2 The Djiboutian Youth Needs Support to Fulfill Their Potential____________________________ XXIX Box 1.1 The Population Covered by EDAM4__________________________________________________ 5 Box 1.2 Issues with Comparability of Welfare and Monetary Poverty across Time in Djibouti_____________ 9 Box 1.3 Multidimensional Poverty Index (MPI)________________________________________________ 21 FULL REPORT XI ACKNOWLEDGMENTS The World Bank greatly appreciates the collaboration analyses, and Walker Bradley produced geospatial analytics with the government of Djibouti through the Ministry included in the report. John Van Dyck, Amr Moubarak, and of Finance, the Commissariat au Plan Chargé des Xavier Decoster contributed to this report. Statistiques, and the Directorate of Statistics and The team received guidance and comments from Demographic Studies (DISED) in the preparation of Carolina Sanchez-Paramo, Benu Bidani, Marina Wes, this report. The team benefitted from an extensive Asad Alam, Poonam Gupta, Tracey Lane, Atou Seck, collaboration with DISED that made the implementation Mamadou Ndione, Kadar Mouhoumed Omar, of the recent household survey possible. The team also Nistha Sinha, Aziz Atamanov, Carlos Rodriguez-Castelán, wishes to acknowledge DISED’s close partnership in Eduardo Malasquez, Axel Rifón Perez and Guido Licciardi. the development and record-breaking publication of the The team thanks Samuel Freije Rodriguez and Javier Baez most recent welfare indicators. The report was prepared for their support to the team as peer reviewers of the by the Poverty and Equity Global Practice. The work Djibouti Programmatic Poverty Work. was led by Gabriel Lara Ibarra and Vibhuti Mendiratta. Marco Santacroce provided inputs and contributed to the XII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI ACRONYMS DISED Direction de la Statistiques et des Etudes Démographiques DF Djibouti francs DT Djibouti Telecom EDAM Enquete Djiboutienne Aupres des Menages (Djiboutian Household Survey) EDAM-IS Enquete Djiboutienne Aupres des Menages- Indicateurs Sociaux (Djiboutian Household Survey) EDAM-BC Enquete Djiboutienne Aupres des Menages- Budget Consommation (Djiboutian Household Survey) GDP gross domestic product GoD government of Djibouti HOI human opportunity index ICT information and communication technologies IPL international poverty line ITU International Telecommunications Union MENA Middle East and North Africa MPI multidimensional poverty index ONEAD Office National de l’Eau et de l’Assainissement PMT proxy means test PNSF Programme National Solidarite Familiale PPP purchasing power parity SDG Sustainable Development Goal SSN social safety net WDI World Development Indicators FULL REPORT XIII OVERVIEW The objective of this report is to present a snapshot of the welfare landscape in Djibouti that will help stakeholders understand poverty’s determinants. Exploiting data from the 2017 household expenditure survey, as well as administrative and geospatial data, it provides a detailed analysis of poverty in the country, its strong link to labor market outcomes, and the differential access to opportunities between urban and rural areas. Djibouti has recently experienced a period of unprecedented economic development, allowing it to grow its economy by an average of 7 percent per year between 2013 and 2016. This report shows that over a fifth of the Djiboutian population continues to live in extreme poverty and that the country has high levels of inequality, ranking 71 out of 95 countries with information on Gini available circa 2015. The report also focuses on the nexus of (monetary) poverty and the labor market. With a dual labor market, divided between the public sector and informality, a large share of the population— especially those who are vulnerable—continue to face high risks, as they have neither the skills nor opportunities to lift themselves out of poverty. Djibouti needs to double its efforts in pursuing an inclusive growth strategy. This strategy will need to be sustained by a strong labor market that allows individuals to capitalize on recent investments in education and closes the gaps in human capital across all groups of the population. WELFARE AND POVERTY ARE DISTRIBUTED UNEQUALLY IN DJIBOUTI The Republic of Djibouti is located in the east of the Horn Sabieh to 77 percent in Tadjourah. The remainder live in of Africa. It shares its borders with Ethiopia in the west the country’s other cities and towns. and the southwest, Eritrea in the northwest, Somalia in Djibouti is a relatively small, lower-middle income economy the southeast, and the Gulf of Aden in the east. Its area is with a nominal gross domestic product (GDP) equivalent to 23,200 square kilometers and it is divided into six regions: US$2.6 billion in 2016 (DISED). The country, however, has Ali Sabieh, Dikhil, Tadjourah, Obock, Arta, and the city of experienced impressive economic growth in recent years. Djibouti, the capital (map O.1). Djibouti city comprises five GDP real growth averaged 4.5 percent per year from 2003 districts. About three-quarters of the population lives in to 2013 and 7 percent between 2013 and 2016. Djibouti’s the capital city. Rural areas are the residence of about 15 growth in each year of 2013–16 surpassed that of percent of the population, with significant variation across countries with similar characteristics, such as Cabo Verde, regions, ranging from 40 percent of the population in Ali XIV CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI MAP O.1 Djibouti’s Location in Africa Source: World Bank. OVERVIEW XV Comoros, Sao Tome, and Principe.1 It also surpassed that an estimated per capita consumption level more than of countries that had a similar GDP per capita in 2013, 16 times the average per capita consumption in the such as Albania, Guinea-Bissau, Mauritius, or Mali. And it first decile (over DF 670,000 compared to about DF surpassed many of its Common Market for Eastern and 40,500). In addition, the richest decile has a per capita Southern Africa partners, including Madagascar, Rwanda, consumption twice as high as that of the ninth decile. This and Uganda. Unfortunately, the recent growth spell—as distribution leads to an estimated Gini index of 0.42— noted, driven mainly by capital intensive investments— among the highest in the Middle East and North Africa does not appear to have been inclusive and has had (MENA) region. limited strength at eradicating poverty in the country. The low levels of consumption among certain groups lead to Nevertheless, pursuing its ambition to position itself as a a nonnegligible share of the Djiboutian population not able regional digital hub, the country is re-shifting the growth to cover its basic needs. A little over a fifth (21.1 percent) engine back to exports of transportation and logistics of the population in Djibouti was considered extremely services after a temporary domination of public investment poor, according to estimates based on the official poverty with high import content financed mainly through debt line (about $2.18 a day 2011 PPP). Poverty rates based accumulation. It is also increasingly the case that Djibouti’s on the international poverty line of $1.90 a day (2011 PPP) growth is closely tied to Ethiopia’s economy. are similar at 17.1 percent. However, for an economy In 2017, the average consumption per capita in Djibouti at Djibouti’s level of development, a more informative was estimated at about DF 208,000 per year, equivalent benchmark can be made using the World Bank’s (2018b) to about US$5.08 per day in 2011 purchasing power poverty line for lower middle-income economies of $3.20 parity (PPP). Beyond this average, wide disparities in 2 a day (2011 PPP). In this case, the poverty rate would be the levels of well-being of the Djiboutian population are closer to 32 percent. found (figure O.1). Individuals of the richest decile have FIGURE O.1 Annual Consumption per Capita per Decile (Djiboutian francs, DF) 800,000 700,000 600,000 500,000 400,000 300,000 200,000 100,000 0 1 2 3 4 5 6 7 8 9 10 Source: Calculations based on the fourth Enquete Djiboutienne Aupres des Menages- Indicateurs Sociaux (Fourth Djiboutian Household Survey for Social Indicators or EDAM4-IS). Note: Deciles are calculated based on per capita consumption. 1  Based on World Development Indicators (WDI) and the Find-My-Friends tool developed by the World Bank. 2 Several parts of the analyses in this report are based on the Fourth Djiboutian Household Survey (EDAM4) conducted in 2017. See box O.1 and Appendix A for details. XVI CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI There are significant gaps in welfare between the residents across the international border. The adjacent regions of of the capital and those of other regions. In Djibouti city, Afar (zone 1) and Somali (Shinile zone) in Ethiopia are the extreme poverty rate is estimated at 13.6 percent, classified as a Developing Regional State by the Ethiopian while in the other regions, it is more than three times Government—a classification identifying regions where higher at 45 percent (figure O.2). Spatial differences in poverty rates are higher and performance on social Djiboutian households’ levels of well-being are evident, indicators is poorer than national averages. Large parts of with regions further away from the capital showing higher these regions in Ethiopia also face arid or semiarid climatic levels of deprivation. Tadjourah region has the highest conditions, much like Djibouti’s border regions. Finally, percentage of population living below the extreme poverty in the capital, disparities are also found: the fourth and line, followed by the Dikhil region. These two regions in fifth districts (jointly known as the community of Balbala) the west side of the country are also hosts of a relatively have extreme poverty rates of 18.3 and 15.8 percent, large rural population. Furthermore, it is notable that respectively. These rates are more than three times higher this relatively high level of deprivation can also be found than the 1st district (4.7 percent). FIGURE O.2 Extreme Poverty Rates, by EDAM4 Representative Domains 70% 60% 50% 40% 30% 20% 10% 0% s h n t h ct t t t ck ty l ta il ric l on ric ra ric ric ie ba na ra kh tri ci Ar bo ab Ru st ou st gi st st ur io is Di ti di re di di di O iS at td ou dj er N d d Ta h h er Al th ib 1s 2n 4t 5t 3r th Dj O O Source: Calculations based on EDAM4-IS. Note: Line across graph denotes the national extreme poverty rate. Together with location, several demographic and economic don’t have (or use) appropriate means to dispose of their characteristics appear to be correlated with poverty status garbage, and they have less access to water, electricity, (table O.1). Compared to the average Djiboutian, and the and sanitation services. Poor families are also found to nonpoor population, households considered to be extreme have limited capacity to purchase and own assets: a much poor tend to be larger and have more dependents, and lower percentage of them own mobile phones, computers, their heads have lower literacy rates and have fewer years televisions, refrigerators, air-conditioning units, washing of education. The dwellings inhabited by poor households machines, and cars. are of lesser quality than those among the nonpoor, they OVERVIEW XVII Box O.1 The Fourth Djiboutian Household Survey (EDAM4) The EDAM4 is a nationally representative survey conducted in 2017 in Djibouti. The sampling strategy of the EDAM4-IS was designed to produce representative indicators at several levels: national, urban, rural, by regions and by 5 districts in Djibouti city. In a first for the Djiboutian Statistical Office (DISED), the EDAM4- IS collected information on sedentary ordinary households and nomads, with a total of 4,474 households interviewed. As it is common in household surveys, the population in hospitals, prisons, military and homeless were not included. The survey adopted current best practices in several aspects of questionnaire design and the analytical work subsequently done on welfare and poverty measurement. These changes include better collection of information on food expenditures via the use of a well-crafted list of representative items and better estimation of nonfood expenditures by collecting information on health and education in dedicated modules. The data collected also allowed for the updating of the welfare measure used for poverty estimation in two key ways. First, by incorporating detailed information on purchase of durables and tenure, the calculation of the flow of services of such items was incorporated. Second, a hedonic model was developed to capture the value of housing services (i.e. imputed rent) for owner-occupied dwellings. While the adoption of such changes led to incomparability with earlier surveys, the DISED used this survey as a new baseline from which indicators of well-being will be monitored in the future. The population in extreme poverty may also face obstacles the fact that almost 18 percent of the poor population to escaping poverty, as they lag behind other groups in declares it does not have a primary school available. In their capacity to build and sustain human capital. Children addition, the extreme poor have lower rates of coverage by in poor households are less likely to attend school at both health facilities. Higher percentages of the poor population the primary and lower secondary levels, and a higher declare that a health center (24.5 percent), hospital (26.8 percentage of these children ages 6–14 have never percent), or maternity clinic (39.1 percent) is not available, attended school (a population sometimes referred to as as compared to the nonpoor (6.5 percent, 6.7 percent, descolarisés). These findings may be partly explained by and 17.5 percent, respectively). XVIII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE O.1 Characteristics of Households, by Poverty Status Nonpoor Extreme poor National Household size 6.2 7.0 6.4 Dependency ratio (%)* 85 111 90 Household head characteristics Age** 48.0 48.2 48.0 Married 80% 84% 81% Female 21% 20% 20% Djiboutian*** 96% 94.6% 95.8% Literate 46% 12% 39% Years of education 4.2 0.8 3.5 Dwelling characteristics Sheet metal roof 82% 47% 74% Cement floor 53% 31% 49% Open defecation 5% 42% 12% Appropriate garbage disposal 86% 46% 78% Access to electricity 72% 16% 60% Access to water 96% 69% 90% Asset ownership rate Mobile phone 87.1% 55.4% 80.4% Laptop/personal computer 13.3% 0.3% 10.5% Television 61.4% 9.4% 50.4% Satellite dish 39.7% 5.8% 32.5% Refrigerator/freezer 44.6% 6.1% 36.4% Air-conditioning 17.5% 0.7% 14.0% Washing machine 16.0% 0.8% 12.8% Private car 3.8% 0.3% 3.1% Education indicators Literacy rates among 15-year-olds and older 58.2% 31.8% 53.1% Percentage of children of 6–10 years attending school 83.8% 63.3% 78.4% Percentage of children of 11–14 years attending school 88.5% 71.1% 84.8% Percentage of 6- to 14-year-olds that never attended school 11.1% 30.8% 15.9% Maximum years of education among all adults in household 10.0 5.3 8.3 Percentage of population that declare no primary school is available 5.7% 17.9% 8.3% Health indicators Percentage of households that declare no health center is available 6.5% 24.5% 10.2% Percentage of households that declare no hospital is available 6.7% 26.8% 10.9% Percentage of households that declare no maternity clinic is available 17.5% 39.1% 22.0% Source: Calculations based on EDAM4-IS. Notes: Access to electricity is defined as the use of electricity as the main source of lighting. Access to water is defined as the availability of water in the household in the form of running water (Office National de l’Eau et de l’Assainissement, or ONEAD, indoor connection), direct connection from a borehole, outdoor connection by pipe provided by ONEAD, public fountain, or drilling (with a pump). Appropriate garbage disposal is defined as availability of garbage collector-OVD (public dump), availability of garbage collector-private, and garbage deposited in a designated place. * Only households with at least one working age (15–64) individual are included. ** The sample comprises 4,359 household heads. For the remaining households, the age of the household head is unknown. *** Nationality is self-declared. OVERVIEW XIX LOCATION SHAPES ACCESS TO SERVICES, WITH LARGE COVERAGE GAPS BETWEEN URBAN AND RURAL AREAS, BUT THE LABOR MARKET OPPORTUNITIES REMAIN LOW FOR URBAN AND RURAL POOR ALIKE Poverty in Djibouti is both a rural and urban phenomenon. Djibouti city is the wealthiest region in the country, but Rural areas comprise only 15 percent of the population but the disparities among the population living in the capital are home to 45 percent of the extreme poor population. are multifaceted. In fact, the poor living in Djibouti city Meanwhile, the capital city is host to half of the extreme have characteristics similar to rural poor households (and poor, mainly due to its large share of the overall population. sometimes rural households in general) across many The districts that make up the Balbala community in the dimensions. Individuals in poor households living in both capital account for 37 percent of the poor population (and Balbala and other districts of the capital have employment 48 percent of the total population). As a comparison, the rates as low as the rural poor, although more individuals are other districts in Djibouti city are host to 12 percent of the participating in the labor force. Heads of poor households poor, but to 32 percent of the overall population. in the capital city have similar educational attainment and formality levels in the private sector as the heads of poor Rural poor households are the most deprived group in the households in rural areas. The level of consumption of population (table O.2). Showing the highest dependency the poor in the capital city is also much closer to that of ratios, they have the lowest participation in the labor force the poor in rural areas than to the people living in the same (among individuals age 15 and above), and their heads district as they are. The deprivation score among poor urban have very low levels of employment, a low likelihood of households is lower than those living in rural areas (reflecting being employed in the public sector, and very low levels better service coverage), but it still lags their nonpoor of formality among those working in the private sector.3 neighbors’ deprivation score. This evidence suggests that the In terms of access to services, they perform worse than low levels of human capital and consequently the types of jobs other groups of the population, with 41 percent of the poor held by the urban poor do not allow them to escape poverty. population having access to an improved water source, 10 percent having access to sanitation, and a mere 3 percent A major distinction between the poor in Djibouti city and having access to electricity. A little more than a third of the the rural poor is that the former have better access to rural poor are close (less than one kilometer away) to an public services, perhaps by virtue of being in the capital. elementary school and only 10 percent live less than one Compared to the rural poor, poor in the capital have kilometer away from any health facility. Taking all these better coverage rates of water, sanitation, and electricity, elements together, it is no surprise that the deprivation and more families live close to an elementary school and score is 0.83 for this subgroup of the population. 4 a health facility. The urban poor also are more likely to send their children to school. There is much scope for improvement beyond just access, however. Map O.2 3 The sector of employment is divided into public (when the main branch of activity is public administration) and private (when the main branch of activity is anything except the public administration). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. 4 A deprivation score is calculated looking at deprivations along 3 dimensions: monetary poverty, education, and access to services. The final score ranges from 0 to 1, where a score of 1 indicates deprivation on all 6 indicators and a score 0 indicates no deprivation. XX CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI uses geospatial data to show catchment areas of one call for attention in examining whether these areas are kilometer of primary schools and hospitals in Djibouti city. underserved and consequently if the quality of service The majority of the city dwellings are covered by at least delivery in these areas is affected. For schools, this is one primary school (80.6 percent) and a hospital (71.3 less likely to be a problem as a result of the introduction percent). However, the estimated population served 5 of double flows in public schools in the capital. Using an by these facilities is expected to be high and may thus alternate catchment area of 1.5 kilometers would increase TABLE O.2 Characteristics of the Population by Poverty Status and Location Nonpoor population Extreme poor population Districts Districts 1-2-3 Balbala Rural 1-2-3 Balbala Rural Dependency ratio* 0.71 0.87 1.08 0.75 0.82 1.35 Percentage of population older than 15 48% 46% 41% 49% 41% 33% in labor force Percentage of individuals employed** 29% 25% 22% 13% 12% 13% Head has no education 44% 61% 81% 87% 87% 94% Head has at least secondary education 19% 15% 3% 4% 2% 1% Head is employed 57% 61% 41% 31% 42% 26% Head is employed in public sector 43% 48% 45% 32% 46% 27% (as % of all employed) Head is employed in private formal sector 19% 10% 5% 2% 3% 4% (as % of all employed) Access to water 99% 99% 66% 92% 97% 41% Access to sanitation 48% 54% 23% 28% 32% 10% Access to electricity 90% 68% 10% 59% 18% 3% Household lives less than one kilometer away 39% 41% 41% 41% 47% 37% from an elementary school Percentage of 6- to 14-year-olds enrolled 84% 87% 69% 70% 79% 53% in school Household lives less than one kilometer away 26% 25% 15% 28% 29% 10% from any health facility Average consumption per capita 323,598 212,859 151,326 69,392 67,669 49,051 Deprivation score 0.13 0.16 0.44 0.47 0.50 0.83 Source: Calculations based on EDAM4-IS. Notes: Access to electricity is defined as the use of electricity as the main source of lighting. Access to water is defined as the availability of water in the household in the form of running water (Office National de l’Eau et de l’Assainissement, or ONEAD, indoor connection), direct connection from a borehole, ONEAD outdoor connection by pipe provided, public fountain, and drilling (with a pump). Access to sanitation is defined as un- shared access to a water closet with flush, or latrine with slab. Private sector comprises all branches of activity except public administration and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered work- ing in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. * Only households with at least one working age (15–64) individual are included. ** Includes all individuals 15 years and older. 5 Recent school openings in Balbala during 2018 are not reflected in the map or calculations. Georeferenced data is currently unavailable to estimate the impact on the estimated coverage rates. Regarding health, coverage rates of other health facilities like community health centers are not included—partly explaining the difference from rates obtained from the survey. OVERVIEW XXI MAP O.2 Catchment Areas of a One Kilometer Radius of Primary Schools and Hospitals in Djibouti City a. Elementary schools b. Hospitals Note: Data on primary schools is obtained from Ministère de l’Education Nationale et de la Formation Professionnelle site. Data on hos- pitals is obtained from OpenStreetMap and do not include military hospitals or doctor’s offices. Data for dwellings (marked as blue dots) are obtained from the building footprint of Djibouti produced by the World Bank. the coverage of public elementary schools and health quarters (73 percent) are in very low-density areas (two facilities to 89.7 percent and 86.8 percent, respectively. buildings or less per square kilometer). In Dikhil, where These elements point to the need to boost coverage in extreme poverty is 53 percent, these shares are 77 and 86 certain areas of the capital (if the areas are found to be percent, respectively. underserved), especially as these areas seem to host poor The second aspect—the rough arid climate—further and vulnerable populations. affects inhabitants of rural areas, as they will require a Finally, location plays an important role in two other strategy that supports them in the face of increased climate aspects: it poses challenges to implementing public change risk. Due to the series of droughts that Djibouti has investments in sparsely populated areas and increases experienced in recent years, the livelihoods of nomadic the urgency for policies to help high-risk populations due and pastoralists have come under threat. Moreover, this to climate change. On the first aspect, lack of access to population is estimated to have dwindled substantially by several services is evident in rural areas of Djibouti, but a flight to neighboring countries or becoming sedentary in the realistic government of Djibouti (GoD) strategy to expand outskirts of villages and cities. Still, in 2017, more than half coverage of basic services will need to find innovative of the population in Tadjourah and Obock practiced some ways to reach the populations in the poorest regions, as sort of animal husbandry, as did 42, 31, and 25 percent of significant portions of the population live far apart from the population in Arta, Ali Sabieh and Dikhil, respectively. each other. In Tadjourah, the region with the highest Survey estimates suggest that transhumance or poverty rate, geospatial analysis shows that close to 60 pastoralism is still practiced by 6 percent of the population percent of dwellings are in extremely low-density areas outside of Djibouti city. Poverty rates in this population (one building per square kilometer) and almost three- appear to be much higher than the national rate. XXII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI ECONOMIC GROWTH IN RECENT YEARS HAS BEEN ACCOMPANIED BY AN IMPROVEMENT IN WELL-BEING OF HOUSEHOLDS, AS MEASURED BY NONMONETARY INDICATORS The recent period of economic growth in Djibouti has been 2012. Improvements are observed in other regions too, capital intensive and driven mostly by large investments, with the exception of Tadjourah. It seems that proximity to including an investment of US$3.4 billion on the railroad the capital and the predominance of rural population are to Ethiopia (Borgen Project 2017) as well as other port important indicators that can help explain the difference projects in Doraleh (cost of US$590 million), Goubet (cost 6 in access to services between regions. In addition, we of US$64 millions) and Tadjourah (US$160 million). A key find that these improvements in access are highest for the policy question is whether the extent to which some of the bottom two quintiles of the population. economic growth has translated into improvements in the In terms of dwelling characteristics (potentially reflecting well-being of households. Due to the lack of comparable the ability to invest in one’s residence), it is found that a data along monetary indicators, evidence of the potential lower percentage of the Djiboutian population is living improvement must be found in nonmonetary indicators.7 in tents or spontaneous dwellings (makeshift dwellings) Survey data from 2012 and 2017 suggest that the in 2017 than in 2012. There is also a higher proportion Djiboutian population experienced improvements in their of the Djiboutian population living in houses with sheet well-being. Indicators on access to services, dwelling metal as roof. These results are largely driven by changes characteristics, or ownership of assets indicate an increase in the capital city, Ali Sabieh, and Arta. It seems that in households’ livelihoods (figure O.3). Unfortunately, many of these characteristics have not changed in regional disparities in such nonmonetary indicators, with Tadjourah, Dikhil, and Obock. The percentage of the regions further away from the capital city performing population living in dwellings with a cement floor has also worse, are an enduring feature. As evidenced by data from increased, indicating that the precarity of the dwellings 2012 and 2017, improvements of recent years had limited has decreased. Higher mobile phone ownership over effects in closing the gaps. Tadjourah region is found to time is also found across all the regions of Djibouti, with have the lowest coverage of services, followed by Obock the exception of Tadjourah. Nationally, this percentage and Dikhil in both years. Only one-fifth of the population increased from 64 percent in 2012 to 80 percent in 2017. has access to electricity in the regions of Tadjourah, Dikhil, Besides objective nonmonetary indicators that correlate and Obock, and this has hardly changed over time. In with households’ well-being, measures relying on Djibouti city, access to water has improved and is almost subjective data also point to a relative improvement in universal. Djibouti city is followed by Ali Sabieh and Arta—a living conditions. When asked about their perceptions, higher percentage of the population declared having members of the Djiboutian population appear to have a access to water in these regions in 2017 as compared to positive view of the evolution of poverty in 2017, especially 6  See http://www.portdedjibouti.com/wp-content/uploads/2017/05/Djibouti-Article.pdf, accessed January 16, 2019. 7 Recurrent changes in methodology and survey instruments prevent obtaining a reliable comparison of monetary welfare across years. With the incorporation of current good practices in the latest household survey, the EDAM4-IS is expected to be the start of a new series to track progress on poverty reduction. See chapter 1 and DISED (2018) for more details. OVERVIEW XXIII FIGURE O.3 Household Characteristics, by Region and Year a. Access to electricity b. Access to improved water sources 2012 2017 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% ity ity ah ah h h l l k k na na il il ta ta ie ie c c kh kh c c ur ur ab ab bo bo Ar Ar io io ti ti Di Di o o ou ou at at iS iS O O dj dj N N Ta Ta ib ib Al Al Dj Dj c. Living in a tent, spontaneous habitat, similar d. Sheet metal as roof material 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% ty ty h h h h l l ck ck na na il il ta ta ra ra ie ie ci ci kh kh ab ab bo bo Ar Ar ou ou io io ti ti Di Di ou ou at at iS iS O O dj dj N N Ta Ta ib ib Al Al Dj Dj e. Cement as floor material f. Ownership of mobile phone 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% ty ity h h h h al l ck ck na il il ta ta ra ra ie ie ci kh kh ic n ab ab bo bo Ar Ar ou ou io io ti Di Di ut ou at at iS iS O O dj dj o N N Ta Ta ib ib Al Al Dj Dj Source: Calculations using EDAM 2012 and EDAM4-IS. Notes: Access to electricity reflects the use of electricity as the main source of lighting. Access to water in 2017 and 2012 is defined as availability of water in running water (ONEAD indoor connection), ONEAD outdoor connection by pipe, public fountain, and borehole (with a pump). Access to water in 2017 had one additional category not included in 2012—direct connection from a borehole. XXIV CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI when compared to their sentiment in 2012. About a third be a belief that things got better in the period 2012–2017. of the population perceived a reduction in poverty in the It is notable that that there is also inequality in people’s previous five years in 2017 (table O.3). This percentage prospects. Breaking down the indicator on perceptions of was only 13 percent in 2012. About 41 percent felt that poverty by location shows the positive expectations are poverty would decrease in the next five years, while only driven by residents of Djibouti city, with regions showing 25 percent believed so in 2012. Overall, there seems to more tame improvements in their views. TABLE O.3 Perceptions on the Evolution of Poverty (percent population by response category) a. 2017 “In the next five years, do you think poverty in your community will . . .” Decrease Same Increase Total Decrease 31 3 1 35 In the past five years, do you “ Same 5 23 3 31 think that in your community poverty has . . .” Increase 4 6 23 33 Total 41 32 27 100 b. 2012 “In the next five years, do you think poverty in your community will . . .” Decrease Same Increase Total Decrease 13 3 1 17 In the past five years, do you “ Same 4 14 6 24 think that in your community poverty has . . .” Increase 7 7 45 59 Total 25 23 52 100 Source: Calculations based on EDAM 2012 and EDAM-IS 2017. Notes: For 2017, only households who answered both questions were included (3,480 households). For 2012, only households who answered both questions were included (3,436 households). OVERVIEW XXV LOOKING FORWARD, INVESTMENTS IN HUMAN CAPITAL MAY CONTRIBUTE TO POVERTY REDUCTION IF THE GOVERNMENT ENSURES THE BENEFITS ARE EVENLY DISTRIBUTED IN THE POPULATION The GoD has made significant efforts to increase access of 91 percent for the same cohort (UNESCO 2018). The to schooling in recent years. The number of public primary gains in literacy have benefitted both boys and girls and schools increased from 84 in 2004–05 to 113 in 2010–11 the gender gap in literacy rates that is observed for those and to 136 in 2016–17 (DISED 2012a, 2014a, 2017a). At born around the mid-1950s and mid-1970s has mostly the same time, the number of public middle and secondary closed. Among the population of 40–60 years, there is a schools also increased threefold, from 11 in 2004–5 to 24 percentage point (p.p.) difference between men’s and 29 in 2010–11 and to 36 in 2016–17 (DISED 2012, 2014, women’s literacy rates. This gap is only 10 p.p. and nearly 2017).8 These investments appear to be bearing fruit by 2 p.p. for 15–24 years and 10–14 years old, respectively. boosting the educational attainment of younger cohorts as Improvements in educational attainment beyond basic compared to previous generations. literacy have also accumulated. A quarter of those 15–24 years old have attained some secondary education, Cohorts of individuals 10–14 years old and 15–24 years compared to less than 3 percent among the 61+ years old old have remarkably higher literacy rates as compared to and above (figure O.5). Gender disparities are evident as older age groups (figure O.4). Not only that, the cohort of women are more likely to have no formal education and 15–24 years old has a literacy rate of 80 percent which is only 38 percent have completed at least primary education significantly higher than the national average of 53 percent as opposed to 57 percent of the men. though still lower than the average of the MENA region FIGURE O.4 Literacy Rates by Age Groups, 2017 (percent) 85 86 85 84 80 75 68 66 58 53 49 42 41 30 28 19 18 10 All Men Women 10–14 years 15–24 years 25–39 years 40–60 years 61 years and above Average Source: Calculations based on EDAM4-IS. 8 According to WDI (n.d.), the GoD spent 12.3 percent of its total expenditure (or 4.5 percent of GDP) in education in 2010—both shares lower than in 2007 (22.5 and 8.4 percent, respectively). More recent estimates are not available. XXVI CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE O.5 Educational Attainment by Age Cohort, 2017 (percent) 8 3 18 13 11 23 25 22 2 18 25 3 29 25 34 4 4 49 5 4 84 72 58 49 6 48 39 21 All Males Females 15–24 years 25–39 years 40–60 years 61 years and above Secondary and above Primary and above, but less than secondary Less than primary No education/ own education Source: Calculations based on EDAM4-IS. Note: Only individuals above the age of 15 years are included. Educational mobility, or the rate at which a group of At the same time, about 60 percent attained the same individuals is able to attain a higher level of education than level, and a mere 4 percent performed worse than their their parents, is positive for the overall population. Using father: overall, leading to a net gain in the population’s the population 25 years old and older as reference, the educational attainment. The upward mobility phenomenon share of individuals with higher educational qualifications was not evenly distributed among all groups, however. than their fathers’ is 36 percent in Djibouti (figure O.6). Nonpoor men experienced the largest gains in education: FIGURE O.6 Educational Mobility of Individuals 25 Years and Older Compared to Their Fathers (percent) 44 60 65 78 4 87 4 6 53 36 3 29 3 19 10 All Nonpoor men Nonpoor women Poor men Poor women Upward mobility Downward mobility No change Source: Calculations based on EDAM4-IS. Notes: We use educational attainment divided into four categories: no education, primary education or less, between primary education and secondary education, and secondary education or more. If the educational attainment of the child is higher than that of the father, he or she is considered upwardly mobile. Shares below 3 percent are not labeled. Data on education of self or father were missing for 7 percent of all individuals. OVERVIEW XXVII over half were able to surpass their father’s education distribution with countries like Egypt, Hungary, or China. level. In contrast, poor women are found to be the group Poor women’s mobility is close to the bottom and similar with the lowest upward mobility (10 percent). To put to those in Mali and South Sudan. The poor women were this gap in context, a recent study (Narayan et al. 2018) found to least mobile, even though none of this group’s on intergenerational educational mobility with upward fathers had more than primary education (table O.4). At mobility estimates for close to 150 economies can this rate, in the absence of better targeted interventions, it provide informative benchmarks. Upward mobility among would take several generations for poor women to catch Djiboutian nonpoor men is around the median of the up to the national average in educational attainment. TABLE O.4 Education Transition Matrix of Poor Women (25 years and older) Father’s education More than No Primary primary or More than Individual’s education education or less secondary secondary Total No education 87 2 0 0 89 Primary or less 7 0 0 0 7 More than primary or secondary 3 0 0 0 3 More than secondary 0 0 0 0 0 Total 97 3 0 0 100 Source: Calculations based on EDAM4-IS. Note: Numbers represent the distribution of the population 25 years and older in percent. Data on education of self or father were missing for 7 percent of all individuals. Economic growth experienced by Djibouti in recent years and ultimately support breaking the intergenerational has been accompanied by investments that contributed transmission of poverty. A better educated workforce to human capital accumulation, although these have 9 presents a good opportunity to transform the labor market not been inclusive. Still, the marked improvements in and further contribute to economic development and educational attainment can play a pivotal role in allowing poverty reduction in the country. Unfortunately, the country individuals to access better jobs, improve their well-being, still faces challenges to unleashing this potential. 9 There is mixed evidence of improvements in other human capital indicators, such as health. Projections for infant mortality, for instance, show a continuous decline, dropping to 51.5 in 2017 from 80.3 in 2000. However, indicators on stunting rates appear to be unchanged, and wasting rates appear to have increased between 1996 and 2012 (World Bank 2018a). Thus, the gains in education may not fully translate into gains in productivity among the labor force. XXVIII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI THE FULL POTENTIAL OF THE LABOR MARKET AS A SUSTAINABLE DRIVER OUT OF POVERTY IS STILL UNTAPPED— OVER HALF OF THE WORKING-AGE POPULATION IS NOT IN THE LABOR FORCE The trickle-down effect of high economic growth leading participate in the labor market. Among people in prime to the creation of a dynamic labor market and a job- working years of age—25–39 and 40–60 years—the generating private sector is yet to be seen in Djibouti. participation rate is 55 and 53 percent, respectively. Labor force participation rates in Djibouti are low. Overall, A sluggish labor market is evidenced not only by the low among the population of 15 years and above, about 45 levels of labor market attachment but also by the high percent participate in the labor market—a ratio practically rates of unemployment. Furthermore, the distribution of job unchanged from the earlier estimate of 46.3 percent in opportunities across the active population (among those 1996 (World Bank 1997) despite the positive economic 15 years and older) varies substantially between men and growth during the past 20 years. Labor force participation women and across age groups (figure O.7). Women and shows substantial variation across gender and age, with the youngest cohort (15–24 years) are the groups less patterns similar to those encountered in other MENA likely to be employed. Unemployment rates are also high countries. Men’s labor force participation stands at 59 among prime-age workers, with about half of them not percent, whereas for women it is only 32 percent. Less having a job. than a third of individuals in the 15–24 year age range FIGURE O.7 Distribution of Population in the Labor Force (percent) 26 38 47 47 63 86 43 32 30 28 25 30 25 30 23 7 11 6 All Men Women 15–24 years 25–39 years 40–60 years Employed in public sector Employed elsewhere Unemployed Source: Calculations based on EDAM4-IS. Notes: Only the population 15 years and older is included. Public sector includes those employed declaring public administration as the main branch of activity. OVERVIEW XXIX Employment rates and education levels show a slight third of women are. This is true across all educational positive correlation in Djibouti but attaining higher levels levels. Employment rates are positively correlated with age, of education does not guarantee more opportunities in although even among individuals between 40 and 60 years the labor market. Individuals with a secondary education of age, only about three-quarters are employed (figure O.7 or more show higher employment rates than those with and figure O.8). The lowest rates are found among the no education. However, the differences in employment youngest cohorts, with only 14 percent being employed. between those with no education and individuals that While in some contexts this result points to a delay in attain only primary education are not always substantial. the school-to-work transition, it appears that there is a The employment-education patterns are similar across nonnegligible share (28 percent) of Djiboutian youth that men and women, but men have much higher employment are disconnected from both the education system and the rates. Two-thirds of men are employed, while only one- labor market (see box O.2). FIGURE O.8 Employment Rate Among Labor Market Participants, by Educational Attainment 68% 70% 65% 62% 55% 53% 54% 49% 52% 41% 42% 39% 37% 26% 21% All Men Women 92% 80% 71% 69% 73% 74% 53% 49% 47% 44% 20% 21% 10% 8% 14% 15–24 years 25–39 years 40–60 years No education/own education Less than primary Primary and above, but less than secondary Secondary and above Average Source: Calculations based on EDAM4-IS. Notes: Employment rate is defined as those employed among those active in the labor force. Both employed and unemployed individu- als are considered active in the labor force. XXX CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Box O.2 The Djiboutian Youth Needs Support to Fulfill Their Potential Djibouti is a young country, with slightly less than 28 percent of its population between 16 and 30 years old (and over 35 percent under 15 years old). Thus, capitalizing on the improved human capital investments the country has made presents a great opportunity to promote poverty reduction and boost shared prosperity. The cohort of 15–24 years has higher rates of literacy and a greater share of individuals with at least primary education than any other age group in the country. Nonetheless, their low labor participation and employment rates highlight the hardships they face in transitioning from school to work. On average, about 45 percent of those age 15–24 years are studying and about 5 percent are engaged in work (figure BO.2.1). Close to a fifth of the youth (22 percent) are unemployed. Worryingly, nearly 28 percent are not in education, employment, or in training (NEET). This distribution of activities and lack of engagement is replicated across gender, although women show slightly higher rates of disconnection to productive activities. Among young women, a third are NEET, implying that they are neither active in the labor force nor studying. Youth in the regions are estimated to have worse outcomes than in Djibouti city, with about 43 percent of them considered to be NEET. FIGURE BO.2.1 Distribution of Youth (15–24 years) across Education and Labor Force Activities a. All a. Women 28 33 45 43 Source: Calculations based on EDAM4-IS. 22 20 Note: NEET refers to youth not in employment, 5 3 education or training and not “unemployed” (those individuals who are seeking a job). Employed/ Studying Employed/ in training Unemployed NEET training and studying shares are negligible. Young women’s labor force participation decreased between 2012 and 2017 (table BO.2.1). This drop in labor force participation is driven mostly by a higher percentage of women reporting familial obligations as a reason to stay out of the labor force in 2017 as compared to 2012. That said, the main reason for not seeking participation in the labor market is the pursuit of education—about 40 percent report studying as the main reason. For men, there has been a drop in the percentage reporting studying as the main reason for not being active in the labor force. These results highlight a need for stronger policy efforts to keep the Djiboutian youth engaged. While some of the shifts observed may be attributed to changes in preferences, the lack of human capital accumulation or stagnant labor market experience can only increase the dependency of these young individuals in the future. Without tools to attain and keep a good job and generate income by themselves, this group will be dependent on help from private or public sources to be able to stay out of poverty. TABLE BO.2.1 Distribution of Youth (15–24 years old) in and out of the Labor Force 2017 2012 Males Females Males Females In the labor force 32 26 33 33 Out of the labor force Studying 44 40 53 42 Family obligations 7 19 4 13 Others / missing 17 16 10 13 Total 100 100 100 100 Source: Calculations based on EDAM 2012 and EDAM4-IS. Notes: Students seeking employment are counted as part of the labor force. Trainees without other activity are counted out of the labor force. Shares may differ slightly from figure BO.2.1. OVERVIEW XXXI WHILE THE PUBLIC SECTOR CONTINUES TO DRAW HIGHLY SKILLED INDIVIDUALS, THE REST OF THE EMPLOYED WORKFORCE IS INFORMAL The working population in Djibouti faces a markedly dual individuals. Formal labor market opportunities are less labor market. Workers are largely concentrated in the private available for working women than for working men. Informal informal sector or public administration (Table O.5). These private sector employs 63 percent of women, while a third two sectors host 90 percent of those employed, with 43 of women work in the public administration (in contrast to percent working in the public sector. Meanwhile, the formal the 48 percent estimated for men). sector employs the remaining 10 percent. In the private Half of Djiboutian workers are employed in small firms sector, agriculture and manufacturing play a relatively small with 10 or fewer employees, but informal workers and role, as the services sector employs the large majority of women and overrepresented in this category (table O.6). TABLE O.5 Distribution of Employed Individuals by Branch of Activity and Sector Of Employment (percent) Private Public Total of informal Private formal administration branch activity Overall population Agriculture 1.3 0.2 - 1.5 Manufacturing 4.1 0.9 - 5.0 Services 24.0 3.4 - 27.4 Other 16.9 5.5 43.8 66.2 Total 46.3 9.9 43.8 100.0 Women population Agriculture 0.9 0.2 - 1.1 Manufacturing 0.5 0.3 - 0.8 Services 48.8 1.8 - 50.5 Other 13.0 3.6 31.1 47.7 Total 63.1 5.8 31.1 100.0 Source: Calculations based on EDAM4-IS. Notes: The sample includes 3,139 individuals that answered to all corresponding questions. Private sector comprises all branches of activity except public administration: agriculture, manufacturing, services (including private administration), and others (international organizations, military bases and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. . XXXII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Individuals working in the informal private sector are own) a firm that has at most 10 employees. In the informal predominantly one-person firms, pointing to largely self- sector, however, micro- and small firms account for 9 of employed individuals working as microentrepreneurs. every 10 jobs held by women. These numbers for women are even higher (47 percent). The public sector continues to have enormous appeal to In the informal sector, firms with 10 or fewer employees highly qualified individuals (table O.7). About half of the account for slightly more than three-quarters of workers public employees (45 percent) have at least a secondary working. In the case of women, 71 percent work in (or TABLE O.6 Distribution of Employed Individuals by Firm Size and Sector Of Employment (percent) Private Private Public informal formal administration Size total Overall population 1 worker 20.8 0.6 1.1 22.5 2–3 8.7 0.5 2.2 11.4 4–10 7.8 2.1 6 15.9 11+ 9.9 7.2 33 50.2 Total 47.3 10.4 42.3 100 Women population 1 worker 46.8 0.7 0.5 48 2–3 8.8 0.3 2.4 11.5 4–10 5.5 1 5.3 11.8 11+ 4.3 3.9 20.5 28.6 Total 65.4 5.9 28.7 100 Source: Calculations based on EDAM4-IS. Notes: The sample includes 2,806 individuals with complete information. Private sector comprises all branches of activity except public administration: agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector education. Moreover, the public administration hires 7 out per month while those in the (informal) private sector earn of every 10 Djiboutian workers with at least secondary DF 84,221 per month (table O.8). Controlling for factors education. Thus, the pool of available talent in the private such as education and experience, it is estimated that on sector is constituted of a population where individuals with average, one more year of formal education increases labor low skills are overrepresented. The high wages offered by earnings by about 7 percent. Nonetheless, working in the the public sector help explain the skewed distribution of public sector still carries a premium of 18 percent even after educated workforce between the private and public sectors. controlling for other factors. On average, workers in the public sector earn DF 104,161 OVERVIEW XXXIII TABLE O.7 Distribution of Employed Individuals by Educational Attainment and Sector of Employment Private Private Public informal formal administration Total No education/own education 30.8 2.5 11.5 44.8 Less than primary 2.5 0.4 1.3 4.2 Primary but less than secondary 9.1 3.2 11.0 23.3 Secondary and above 3.9 3.9 19.9 27.6 Total 46.3 9.9 43.8 100 Source: Calculations based on EDAM4-IS. Note: The sample includes 3,132 individuals with complete information. Private sector comprises all branches of activity except public administration: agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. Cells show the percentage of the working population. TABLE O.8 Characteristics by Sector Of Employment Private Private Public informal formal administration National average monthly wages of those employed (DF) 84,221 118,442 104,161 Based on sector of employment of household head Average monthly wages of head (DF) 87,081 121,576 101,077 Extreme poverty rate 18% 3% 11% Extreme poverty rate when poverty line is 5% higher 20% 6% 12% Extreme poverty rate when poverty line is 10% higher 22% 6% 13% Percentage of households that experienced a health problem 28% 24% 27% Health expenses as a percentage of total: average 3% 2.5% 1.9% Health expenses as a percentage of total: 90th percentile 7.2% 6.7% 4.9% Source: Calculations based on EDAM4-IS. Notes: Wage information is available for 2,415 individuals and 1,531 heads. Private sector comprises all branches of activity except public administration: agriculture, manufacturing, services (including private administration), and others (international organizations, military bases and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. Cells show the percentage of the working population XXXIV CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Djibouti’s high economic growth rate has not been informal private sector as compared to when s/he works accompanied by the creation of a dynamic formal private in the formal sector (table O.8). Households with a head sector. A lagging formal private sector may have important employed informally also spend more on health-related negative consequences for the economy going forward, expenses, although they fall ill at the same rate as according to the relevant literature. In fact, it has been other groups, pointing to vulnerability to health shocks. argued that formal firms are the driver of economic Vulnerability is higher among those with a head working development, while informal firms are a by-product of in the informal sector, as evidenced by the higher poverty poverty (Rauch 1991; La Porta and Schleifer 2014). The rates found after small changes to the poverty line. Finally, literature points out that as an economy modernizes, among households in the bottom of the distribution, the formal firms grow, and informality becomes less prevalent. vulnerability of informal incomes is further compounded Countries with an abundance of informal firms suffer by the variability of other unpredictable sources of from low aggregate productivity, too. Thus, efforts to income. The income of households in the bottom quintile encourage the growth of the formal sector in Djibouti may includes of 40 percent from informal work, 20 percent be important to stimulate job-driven growth. from transfers (public or private), and another 14 percent from agricultural activity. Formal work (including the public Tackling informality directly can also have positive effects sector) accounts for a quarter of total income, on average. on the population. In Djibouti, informality is correlated For the overall population, half of income comes from to poverty and vulnerability. Poverty rates are six times formal or public sector jobs. higher when the household head is employed in the THE EMPLOYMENT PROFILE HIGHLIGHTS THE NEED TO CONTINUE ADDRESSING LABOR DEMAND-SIDE ISSUES A sustainable path of economic growth accompanied with According to the enterprise survey conducted in 2013,10 poverty reduction cannot continue to rely on a labor market among formal firms in Djibouti, certain aspects of the where about 80 percent of individuals are working either in regulatory environment and quality of service were flagged the public sector or in micro/small informal firms. Djibouti’s as barriers for development of business. For instance, high economic growth rate has had limited spillovers into the tax rates were found to be a hindrance for 15 percent of creation of a dynamic formal private sector with competitive the small and medium-size firms, while the paucity and wages. Recent investments in education seem to be paying quality of electricity as well as corruption were also cited off in building the human capital and, ultimately, boosting as constraints among a third of large firms. Per the Doing the quality of the labor supply. However, to develop the Business (DB) 2014 report, Djibouti ranked 160th of all labor market and absorb this increasingly educated countries evaluated, with several areas of improvement population, obstacles in the labor demand side should also identified: procedures for starting a business (a rank of be studied and addressed. 127), getting electricity (144), registering property (133), and getting credit (180), among others (World Bank 2014). 10 The sample comprised 266 formal nonagricultural private firms. Public utilities, government services, health care, and financial services sectors are not included in the sample. OVERVIEW XXXV Regarding the process of setting up a (formal) business to 99th, as opposed to 154th a year before (World Bank and getting it up and running, there were two areas that 2019). The largest gains came from changes in protecting deserved particular attention. First, the costs associated minority investors, registering property, and getting with starting a business was estimated at 184.7 percent of electricity. Other improvements were seen in resolving the country’s income per capita for all the permits required insolvency and starting a business. The current cost of for the benchmark case. Second, securing an electrical opening the business was estimated at 41.9 percent of connection to a business took 180 days at a cost of 7,487 income per capita.11 Nonetheless, there is still room for percent of income per capita. It shouldn’t be a surprise that improvement. The DB 2019 report notes that the costs informality remains pervasive throughout the economy. of starting a business in the MENA region average 22.6 percent (as a share of income per capita) and that of OECD There is, however, more recent evidence that the GoD high-income countries is at 3.1 percent. Steps in the right has begun tackling some of these issues. In the DB 2019 direction have been taken, but there is more to be achieved report, Djibouti made significant strides, jumping 55 places to maximize the potential of Djibouti’s labor market. DJIBOUTI MUST TAKE FULL ADVANTAGE OF ITS GROWING ECONOMY TO STRENGTHEN THE POVERTY REDUCTION ANGLE IN PUBLIC POLICIES Djibouti’s policies need to take better advantage of the government needs to be adopted. The large potential of growing economy to eradicate poverty and boost shared Djibouti, given its geostrategic location coupled with the prosperity in the country. No comparable estimates for GoD’s objective to become a regional hub for trade and monetary poverty from previous years are available. commerce, sets up a natural environment where priority However, despite the strong economic growth of recent should be given to investments in service provision and years, extreme poverty still affects more than a fifth of human capital of the population, and the provision of the population and about two-thirds of rural households. incentives to develop a dynamic labor market. Inequality is also high—and among the highest in the Rural and urban poor face a distinct set of concerns, and world. The richest decile in Djibouti enjoys 32 percent separate, targeted sets of policies must be considered of the total consumption and shows more than 16 for each of these subgroups. Rural areas require a times the level of consumption of the poorest bottom comprehensive strategy that invests in infrastructure 10 percent (whose consumption share is 1.9 percent). and service provision. Rural Djibouti is the residence There is suggestive evidence of improvements in certain of 15 percent of the population yet hosts close to 45 nonmonetary indicators and individual’s perceptions. Still, percent of the poor population. Households in these Djibouti seems to have significant scope at leveraging their areas suffer from low access to public services such as booming economy into a poverty-reducing strategy. electricity, water, and sanitation. Electricity access—a key The heterogeneous Djiboutian population requires a broad service that has been shown to stimulate consumption variety of supports. A multifaceted welfare-enhancing and income and enable better education and health— strategy that cuts across the policies implemented by the especially needs to be addressed, as only 6 percent of 11  Roughly, the costs summed to about DF 129,000, not including costs for stamp duties. XXXVI CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI the rural population has access to it. Meanwhile, open Programme for International Student Assessment or defecation rates are high in regions with high poverty Trends in International Mathematics and Science Study. rates and with a high rural presence, such as Tadjourah Employment policies need to be rethought as an essential and Dikhil (69 and 48 percent, respectively). These call part of the GoD’s poverty reduction strategy. Expanding for strong investments in sanitation services as well as employment opportunities will be key to unlocking the disease management, especially as these areas are also potential of the Djiboutian population. With just 45 percent affected by low coverage of health facilities. Achieving of the working-age population participating in the labor universal coverage in these areas will face the obstacle of market, a push must be made to increase the incentives of reaching a very sparsely populated region. the population to join the labor force. This detachment may The urban poor are more numerous and have much lower be partly explained by the distinct dual market Djiboutian consumption than others in their community. Crucially, individuals face. This report shows that in Djibouti’s they also show lower educational attainment and worse labor market, an individual’s employment sector is highly employment outcomes. The estimated population served by determined by the individual’s skill level. On the one side, schools and hospitals is expected to be high in these areas there is the public administration. The public sector hires and may thus call for attention in examining whether these two-fifths of employed individuals, with a large share areas are underserved and consequently if the quality of of them being highly skilled workers. These individuals service delivery in these areas is affected. For schools, this are likely to be drawn to the public sector due to the is less likely to be a problem as a result of the introduction high wages this sector offers: even controlling for other of double flows in public schools in the capital. A policy tool characteristics such as education and experience, public kit to promote better education and a boost in employment administration pays an average 18 percent more than the opportunities in the country is critical, going forward. private sector. On the other side, workers with low skills are mostly left with the option of working in the informal Education policies need to reinforce recent public sector. More than 40 percent of employed individuals are investments to help close the disparities that persist. informal, as are two-thirds of employed women. GoD investments in education have paid off in increasing the literacy rates and educational attainment among Boosting employment in the country will require ensuring the the younger generation. These investments need to be elements of a vibrant private sector exist. The improvements more balanced and allow the poor to catch up. Women on measures such as the Doing Business Index are steps in poverty are a particularly vulnerable group in terms in the right direction, but there is more to be done. On the of human capital accumulation and low mobility. More labor supply side, a better educated workforce would be schools have opened in the poorer areas of Djibouti city able to fill vacancies from new investments. On the labor and double classes have been introduced in schools in demand side, the pervasiveness of the informal sector the capital but there is a gap that needs to be filled in suggests that barriers for formalization are varied. Firms understanding how the increased educational attainment (or small entrepreneurs) may be showing a preference for reflects actual learning. Learning based on acquired skills staying informal or reacting to a simple cost-benefit analysis promotes employment, income, health, and poverty showing the large costs of registering, getting basic services reduction (World Bank 2018c). Djibouti is among the like electricity, and strong competition from those who few countries that do not participate in internationally decide to stay informal and avoid the increased overhead. comparable testing efforts of students, such as the However, the corresponding variability of income and OVERVIEW XXXVII vulnerability to shocks (in both short and long terms) may boosted household real consumption by 11 percent not be fully internalized in such decisions. A broad range (Beuermann, McKelvey, and Vakis 2012). The ICT sector of policies that spurs a better business climate (including has a great capability to enable job creation and innovation taxation, labor regulations, and costs of key inputs such as in the private sector, with a corresponding increase in electricity or telecom services) and investment in labor- economic opportunities for the population. Moreover, intensive sectors are the best bet. The youth in Djibouti the Internet is also an enabler of higher labor productivity present a particularly vulnerable group and will need support by lowering information and search costs. For example, from private (or public) networks if they continue to be the introduction of mobile phones in the grain markets detached from studying or working. of Niger led to farmers obtaining grain price information over the phone, thereby reducing search costs by 50 This assessment explores three themes that should percent (Aker 2010) and reduced dispersion of grain be considered in the design of the GoD’s public policy prices across markets by 10 percent (Aker and Mbiti strategy going forward. The first topic that we investigate 2010). People’s perceptions also reveal that the Internet is the strengthening of social safety net (SSN) programs, has led to an increase in consumer welfare by making such as cash transfer programs. SSN is a powerful policy several products and services available digitally. From the tool to support the poor and vulnerable population. standpoint of both the government and the private sector, Recently, Djibouti has made efforts to move away from a the Internet can bring major benefits to the provision and largely donor-driven initiative, which was mainly focused delivery of services. Digital identification can also improve on providing food to vulnerable populations, and to participation of disadvantaged groups and help them investing in adaptive social safety nets and incentivizing integrate into the economy. The advantages of a well- households to invest in human development. Moreover, developed ICT sector are manifold and transformational. Djibouti has begun laying the groundwork for a social protection system through the national expansion of Djibouti is particularly well placed to take advantage of the the Programme National Solidarite Familiale (PNSF) and digital economy with its regional comparative advantage the establishment of a social registry. Simulations show as the landing site of the undersea fiber optic cables. that with increasing efforts in targeted programs such Unfortunately, Djibouti Telecom’s (the single ICT operator) as the PNSF, either through the application of a proxy outward strategy has not exploited the full development means test (PMT) to identify and target households that potential of the sector, leading to lower than expected experience a high level of deprivation, or a combination of penetration rates of ICT services, very low ownership a PMT with geographic targeting, there is strong potential rates among rural households, and prices for broadband for poverty reduction, especially among rural households. services that are considered higher than in comparator Programs that connect those in beneficiary households countries. Based on international case studies, we with the opportunity to work will also help, provided there hypothesize that increased competition in the telecom are work opportunities available. sector could help improve service delivery, productivity and innovation, and overall performance in Djibouti’s The second topic studied in this report is the continuous ICT sector. Simulations of the welfare effects of price development of information and communication changes in the telecom sector (that could result from technologies (ICT) services, which hold enormous potential more competition in the sector) show there are potential for economic growth and poverty alleviation. For example, welfare gains for the population and an increase in the in rural Peru, the expansion of mobile network coverage likelihood of adoption of new technologies. In view of XXXVIII CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI this, an extraordinary collective, national effort must be sedentary on the outskirts of villages and cities. Survey considered in Djibouti to achieve the goal of expanding estimates suggest that transhumance or pastoralism is affordable access to the Internet and transmission of practiced by only 6 percent of the population outside of data. This should be accompanied by the promotion of Djibouti city. Poverty rates in this population appear to modern finance and payment systems supporting financial be much higher than the national rate. Individuals living transactions that can be carried out electronically. Such a in households practicing transhumance also have fewer large-scale effort is likely to foster productivity, competition, years of education and are more likely to be illiterate. entrepreneurship, and growth of businesses. This would It, thus, becomes imperative to protect this vulnerable especially be an appealing prospect for the youth, who are group, especially in light of the persistent threat of climate more Internet savvy and currently seem not able to find change. Equally importantly, the statistical systems jobs or are stuck in low wage employment. in Djibouti must redouble efforts to be able to better observe this population. The upcoming population census The final theme investigated is the livelihoods of (expected to be conducted in 2020) is an opportunity that pastoralists in Djibouti. About 20 percent of the Djiboutian should not be missed to improve our understanding of this population was found to be nomadic in 2009, according to important subgroup of the population. It would also be the country’s 2009 census. Due to the series of droughts worthwhile to investigate resilience strategies that could be that Djibouti has experienced in recent years, the nomadic employed by this population as well as the policy actions population is estimated to have dwindled substantially that may facilitate the promotion of these strategies and because of flight to neighboring countries or by becoming alternative livelihoods. OVERVIEW XXXIX REFERENCES Aker, J. C. 2010. “Information from Markets Near and La Porta, R., and A. Shleifer. 2014. “Informality and Far: Mobile Phones and Agricultural Markets in Niger.” Development.” Journal of Economic Perspectives 28 (3): American Economic Journal: Applied Economics 2 (3): 109–26. 46–59. Ministere de l’Education Nationale et de la Formation Aker, J. C., and I. M. Mbiti. 2010. “Mobile Phones and Professionnelle. 2018. “Cartographie des ecoles de base.” Economic Development in Africa.” Journal of Economic http://qgiscloud.com/Destini/projet_ETS_djibouti/ Perspectives 24 (3): 207–32. ?bl=mapnik&st=&l=ecole-djibouti%2Cregion-djibouti&t= projet_ETS_djibouti&e=4790662%2C1291018%2C48109 Beuermann, D. W., C. McKelvey, and R. Vakis. 2012. 82%2C1300946. “Mobile Phones and Economic Development in Rural Peru.” Journal of Development Studies 48 (11): 1617–28. Narayan, Ambar, Roy Van der Weide, Alexandru Cojocaru, Christoph Lakner, Silvia Redaelli, Daniel Gerszon Borgen Project. 2017. “Ethiopia-Djibouti Water Mahler, Rakesh Gupta N. Ramasubbaiah, and Stefan Pipeline.” Borgen blog. Seattle, WA. December. https:// Thewissen. 2018. Fair Progress?: Economic Mobility borgenproject.org/tag/ethiopia-djibouti-water-pipeline/. Across Generations Around the World. Washington, DC: DISED (Direction de la Statistiques et des Etudes World Bank. https://openknowledge.worldbank.org/ Démographiques). 2012a. Annuaire Statistique 2012. handle/10986/28428. DISED. 2012b. EDAM 3-IS. Djibouti: DISED. Rauch, J. E. 1991. “Modelling the Informal Sector Formally.” Journal of development Economics 35 (1): DISED. 2014a. Annuaire Statistique 2014. Djibouti: DISED. 33–47. DISED. 2014b. “Mesures de la pauvreté et des inégalités” UNESCO Institute for Statistics. 2018a. “Literacy Rate, Photocopy. Djibouti: DISED. Adult Total (% of People Ages 15 and Above).” World Bank Open Data. https://data.worldbank.org/indicator/SE.ADT. DISED. 2017a. Annuaire Statistique 2017. Djibouti: DISED. LITR.ZS. DISED. 2017b. EDAM4-IS. Djibouti: DISED. UNESCO Institute for Statistics. 2018b. “Literacy Rate, DISED (Direction de la Statistiques et des Etudes Youth Total (% of People Ages 15–24).” World Bank Démographiques.) 2018. Résultats de la Quatrième Open Data. https://data.worldbank.org/indicator/ Enquête Djiboutienne Auprès de Ménages pour les SE.ADT.1524.LT.ZS. Indicateurs Sociaux (EDAM4-IS). Djibouti city: DISED. WDI (World Development Indicators). n.d. Database. http://www.dised.dj/Rapport1_resultats_EDAM4.pdf. Washington, DC: World Bank. https://datacatalog. Financial Times. 2018. Djibouti ready to pay compensation worldbank.org/dataset/world-development-indicators. to settle DP World dispute. XL CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI World Bank. 1997. Djibouti: Crossroads of the Horn of Africa. Poverty Assessment Report 16543-DJI. Washington, DC: World Bank. http://documents. worldbank.org/curated/en/557421468748480282/ Djibouti-Crossroads-of-the-Horn-of-Africa-poverty- assessment World Bank. 2014. Doing Business. Washington, DC: World Bank. World Bank. n.d. “Tables, Charts, and Technical Notes.” Global Consumption Database. http://datatopics. worldbank.org/consumption/detail. World Bank. 2018a. Economic Transformation in Djibouti: Systematic Country Diagnostic. Washington, DC: World Bank. World Bank. 2018b. Poverty and Shared Prosperity 2018: Piecing Together the Poverty Puzzle. Washington, DC: World Bank. World Bank. 2018c. World Development Report 2018: Learning to Realize Education’s Promise. Overview booklet. Washington, DC: World Bank. World Bank and OECD (Organisation for Economic Co-operation and Development). 2018. “GDP Growth (Annual %).” World Bank Open Data. https://data. worldbank.org/indicator/NY.GDP.MKTP.KD.ZG. World Bank. 2019. Doing Business. Washington, DC: World Bank. OVERVIEW XLI WELFARE AND POVERTY CHAPTER 1 IN DJIBOUTI INTRODUCTION AND MACROECONOMIC CONTEXT The Republic of Djibouti is located in the east of the an important security role in the region, hosting refugees Horn of Africa. It shares its borders with Ethiopia in west fleeing conflicts and environmental risks in neighboring and southwest, Eritrea in the northwest, Somalia in the countries. According to United Nations High Commissioner southeast, and the Gulf of Aden in the east. Its area is for Refugees, the population of concern (refugees, asylum 23,200 square kilometers. Djibouti is divided into six seekers, and others) was 27,000 in 2017.2 regions: five regions of the interior (Ali Sabieh, Dikhil, Djibouti is a relatively small lower-middle income economy Tadjourah, Obock, and Arta) and the city of Djibouti, with a nominal gross domestic product (GDP) equivalent the capital. Djibouti city comprises three municipalities to US$2.6 billion in 2016 (DISED). The country, however, (Rasdika, Boualos, and Balbala), which are further has experienced impressive economic growth in recent subdivided into five districts. years. GDP real growth has averaged 4.5 percent per Djibouti is situated in a very important geostrategic location year from 2003 to 2013 and 7 percent between 2013 and for trade and security in the region. The port in Djibouti 2016. Djibouti’s growth in each year of 2013–16 surpassed is located at the Red Sea and serves as an important that of countries that are similar in characteristics, such as connection point between Asia, Europe, and Africa. It is Cabo Verde, Comoros, Sao Tome and Principe;3 countries estimated that about 30 percent of world shipping cargo that had a similar GDP per capita in 2013, such as Albania, transits through the Red Sea basin (Financial Times, 2018). Guinea-Bissau, Mauritius, Mali, Jamaica, and Republic of The country also serves as the main gateway of trade the Congo; and many of its Common Market for Eastern for Ethiopia, its landlocked neighbor. Indeed, about 70 and Southern Africa partners, including Madagascar, percent of the cargo at the port of Djibouti is shipped from Rwanda, Uganda, Tunisia and Egypt. or to Ethiopia (Meseret 2016). What’s more, it accounts Several large investments on infrastructure projects for 95 percent of Ethiopia’s foreign trade (Maasho 2011).1 (such as a railway line connecting to Ethiopia) have The port’s location at the Gulf of Aden has also made it an propelled Djibouti’s growth. From 2014 to 2016, the fast important military outpost for the French navy, the US navy, implementation of large public infrastructure projects Japan’s only foreign military base, an Italian base, and added 3 to 5 percentage points to real growth. As many more recently the Chinese navy. The country also plays 1 Ethiopia’s and Djibouti’s economies are reliant on each other, with about 70 percent of all trade through Djibouti’s port coming from its landlocked neighbor. 2 This number could underestimate the true magnitude of this population, as some immigrants are not registered in camps. Other estimates (IMF 2017) suggest that there are about 60,000 refugees, asylum seekers, and migrants. According to estimates from the Ministry of Social Affairs, the population of unregistered migrants and refugees in the country is 150,000, and this may have consequences for the socio-economic situation in the country as well as for public policy decisions. 3  Based on WDI (n.d.) and Find-My-Friends tool developed by the World Bank. 1 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE 1.1 Macroeconomic Indicators in Djibouti a. GDP growth (%) 9.0 9.0 8.0 8.0 7.0 7.0 6.0 6.0 5.0 5.0 4.0 4.0 3.0 3.0 2.0 2.0 1.0 1.0 0.0 0.0 2015 2016 2017 2015 2016 2017 b. Share of economic sectors in GDP (%) 100% 100% 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Agriculture Industry Services Agriculture Industry Services Source: DISED. infrastructure projects come to an end, the rate of GDP been gradually increasing over time. Meanwhile, agriculture growth slowed to 4.1 percent in 2017. Output is expected plays an insignificant role in Djibouti’s economy, given the to increase by 7 percent in 2019 as international trade arid climatic conditions in the country (Figure 1.1). normalizes in Ethiopia following the successful political This impressive economic growth has been highly capital transition and the devaluation of the birr by 15 percent, intensive such as an investment of US$3.4 billion on the the Ethiopian currency, in October 2017. Growth will be railroad to Ethiopia (Borgen Project, 2017). According to driven by export of transportation and logistics services the Port of Djibouti website,4 the country’s investments supported by the newly commissioned infrastructures in large-scale projects have included the construction of (railway, port, and free zone). The tertiary sector dominates several new ports, such as the port of Doraleh (cost of economic activity in Djibouti with an estimated share of US$590 million), port of Goubet (US$64 million) and port more than 80 percent of GDP in 2017. This share has also of Tadjourah (US$160 million). Finally, Djibouti’s revenues 4  See http://www.portdedjibouti.com/wp-content/uploads/2017/05/Djibouti-Article.pdf, accessed January 16, 2019. CHAPTER 1 2 are also dependent on port activities and military bases.5 Djibouti. Relying on a household consumption survey fielded However, pursuing its ambition to position itself as a in 2017, this chapter fills a knowledge gap by shedding regional digital hub, the country is re-shifting the growth light on the most recent state of affairs for the Djiboutian engine back to exports of transportation and logistics population. We introduce the household expenditure services after a temporary domination of public investment survey conducted in 2017 and the poverty measurement with high import content, financed mainly through debt methodology for Djibouti. A full demographic and economic accumulation. It is also increasingly the case that Djibouti’s profile of the poor population and how it compares to that growth is closely tied to Ethiopia’s economy. Past growth of the nonpoor population follows. The report then explores driven by capital intensive investments is likely to not have the evolution of welfare in the country by comparing several been conducive of an inclusive growth path, as large-scale indicators of well-being between 2013 and 2017. The main infrastructure and logistics investments are more likely to objective of this chapter is to help understand whether the create demand for skilled workers. high economic growth that Djibouti has experienced has percolated down to the population and consequently led to It is against this macroeconomic backdrop that this chapter improvements in welfare. presents a detailed update of welfare and poverty in ANALYSIS OF WELFARE AND MONETARY POVERTY One of the main components to designing programs and conditions, to monitor and evaluate social development policies that help alleviate poverty is to identify the poor and poverty alleviation policies, and to monitor progress and vulnerable population in the country. This section toward the Sustainable Development Goals (SDGs). A key presents results on welfare and poverty in Djibouti. Most implication of these objectives is the ability to provide an of the information presented is drawn from the Enquete updated profile of poverty and well-being of the Djiboutian Djiboutienne Aupres des Menages- Indicateurs Sociaux population. (henceforth referred to as EDAM4-IS) conducted in 2017 The EDAM4-IS integrates several themes: demographic by the Directorate of Statistics and Demographic Studies characteristics of household members, education, (DISED) under the supervision of the Planning Commission health, employment, migration, housing characteristics, for Statistics. The structure of this section is as follows: possession of livestock, food expenses (consumed at we first elaborate on the (monetary-based) estimate home and meals taken outside), nonfood expenses, of consumption that is used as the measuring bar of sources of income (private and public transfers), shocks households’ well-being (Deaton and Zaidi 2002), followed and survival mechanisms, perceptions of poverty, by a brief presentation of the estimation of the poverty line. governance, access to services, and income from The main results on poverty in Djibouti close this section. agricultural and livestock activities. MAIN SOURCE OF DATA The sampling strategy of the EDAM4-IS was designed to Several parts of analysis in this report are based on the produce indicators at several levels of representativeness: EDAM4-IS survey from 2017.6 This survey was designed to national, urban, rural, regional, and by districts in Djibouti provide recent data on household consumption and living city. The EDAM4-IS collected information on both 5 For instance, about US$63 million is received annually from the United States and US$20 million from China (Jacobs and Perlez 2017). 6  See appendix A for a more detailed description of the survey. 3 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 1.1 Distribution of the EDAM4 Sample and Estimated Population, by Region Distribution of sampled a.  b. Distribution of population using sampling weights households Population Households weighted Population in Region interviewed Region representation rural areas Djibouti city 2,035 Djibouti city 76% 0% Ali Sabieh 495 Ali Sabieh 5.4% 40% Dikhil 496 Dikhil 6.6% 64.2% Tadjourah 493 Tadjourah 5.5% 77% Obock 475 Obock 2.4% 63.7% Arta 480 Arta 4.2% 73.5% National 4,474 National 100% 15.2% Source: Calculations based on EDAM4-IS. sedentary ordinary households and nomads. A total of THE WELFARE AGGREGATE 4,474 households were interviewed. Table 1.1 shows the The measure of a household’s standard of living—its decomposition of the total sample size. The sampling welfare—is typically based on an estimate of the level weights were obtained from a two-step sampling approach consumption or income of a household. The choice and corrected for nonresponse. between the two depends on many factors, such as the The EDAM4 relied for its framework on (i) the distribution of availability of data, the design of the survey, as well as the population according to the 2009 census, (ii) the listing the context of the country. Consumption is preferred to exercise of all the enumeration areas that were selected income due to several reasons. Consumption is typically for the survey data collection, and (iii) the population smooth over the duration of the year, while income may projections made by DISED. Finally, the population that be subject to more variability and patterns of seasonality. was covered in the survey could be drawn only from those It may also be difficult to gather accurate data on income households that could be listed and whose members when there is a high level of informality. Thus, the income were likely to be found during the interview. Thus, a observed in a survey (over a short time frame) is less likely direct implication is that there will be a gap between to reflect accurately the living standard of the household, the population that is represented in the survey and the or the average level of well-being the household enjoys by overall population in the country. Current estimates of the smoothing their consumption throughout the year. In the Djiboutian population are about 1 million, but it includes case of a country like Djibouti, the recording of income is subgroups such as homeless people and refugees living difficult because of underreporting, the informality of the in camps that are not part of the survey. Moreover, these labor market, and the high variation in labor income. A estimates are based on extrapolations of the distribution measure that relies on consumption expenditure can then of the population from the 2009 census that may be considered as a better approximation of well-being overestimate the evolution of the nomadic population (see than income. Box 1.1). While there is evidence that several statistics The calculation of the welfare aggregate in EDAM4-IS is from the EDAM4 replicate the patterns of other official data based on the addition of several components of household sources (see appendix A), this difference should be kept in consumption (measured by associated expenditures). mind while reading the results. The included components of well-being are food CHAPTER 1 4 Box 1.1 The Population Covered by EDAM4 There are two important points to make in understanding the population covered by the EDAM4 and how this estimation may not be comparable to total population estimates. The first is that, due to the nature of a data collection exercise of surveys, households that are included in the survey must be identifiable during the listing exercise, and they must be available in the same physical place during the survey interview. Thus, the homeless population is not covered in the survey, nor is the population that is living in very precarious dwellings. Other populations are also not typically included in surveys. The EDAM4 did not include individuals in boarding schools, orphanages, prisons, hospitals, hotels, military and paramilitary camps, housing for foreign laborers, and refugees in camps. The second point worth highlighting is that unlike earlier surveys (2012 EDAM3-IS, the 2012 EDSF/ PAPFAM, and the 2015 Employment and Informal Sector Survey), the EDAM4 also covered the nomadic population. Incorporating appropriate estimates of this population required several sources of information. Since 2008, Djibouti has been experiencing longer and more frequent cycles of drought, accompanied by continued desertification and significant losses of livestock and pastures (map B.1.1). In 2009, the government of Djibouti adopted (Décret n°2009-0113/PRE) the National Strategy for Food and Nutrition Security. In volume 1 of that strategy’s report, it is mentioned that “[due to the drought] . . . nomadic pastoralists are led to move their animals . . . in the interior of the country, especially to high-altitude pastures and to other countries in the region. . . . More substantial and more regular humanitarian aid to Ethiopian pastoralists has helped set up pastoralists in Ethiopia, whose home lands were traditionally in Djibouti. Therefore, there is a double hemorrhage of the rural population,” Thus, resilience strategies developed by the nomadic populations against the negative effects of the drought have been (i) to migrate to regions, sometimes beyond the country’s borders, or (ii) to settle around villages or other points, abandoning traditional transhumance in favor of nearby pastures. MAP B1.1.1 Djibouti Vegetation and Urbanization a. Djibouti 2007 a. Djibouti 2012 Source: DISED 2018. For EDAM4, DISED’s listings were complemented with information from the National Strategy for Food and Nutritional Security and thus estimated a decrease in the nomadic population (as defined in the 2009 General Population and Housing Census and excluding settled pastoralists households). This decrease was estimated at nearly three quarters between 2009 and 2017. 5 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI expenditures, housing expenditures such as water and express all expenditures in May 2017 prices. Finally, to electricity services, transportation and communications, take into account the possible differences in needs across clothing purchases, hotel and food purchases, purchase members of the household by age, as well as the potential of household items, khat, tobacco and alcohol purchases, for economies of scale, the welfare aggregate is calculated education, health, recreation, and general services. In in per adult equivalent terms. Appendix B provides more order to better capture well-being derived from durable details on the construction of the consumption aggregate goods, the flow of services based on the current market for Djibouti using EDAM4-IS. value of each property and the depreciation rate for each Results from the survey show that the estimated per adult durable good is estimated. Finally, a hedonic model was equivalent consumption at the national level is DF 261,391 estimated to allocate the rental value of the dwellings per year, or approximately US$1,468 per year.8 In line with among home owners.7 Expenses for ceremonies, previous studies, Djibouti city is found to be a relatively richer transfers, or investments were not taken into account. zone than the five other regions in Djibouti (Figure 1.2). In Temporal adjustments must be made to household the regions, consumption is about 57 percent of the amount consumption to ensure that poverty measures are spent in Djibouti city on average, while per adult equivalent comparable across different data collection periods. Thus, consumption in rural areas is very low (DF 114,909). Among the national monthly consumer price index in 2017 is used the five districts of the city of Djibouti, the fourth and fifth to adjust the expenditures recorded in the survey and districts have the lowest levels of consumption. FIGURE 1.2 Annual per Adult Equivalent Consumption, by Area and District in Djibouti City (DF) a. Consumption across major regions b. Consumption across Djibouti city districts 500,000 500,000 450,000 450,000 400,000 400,000 350,000 350,000 300,000 300,000 250,000 250,000 200,000 200,000 150,000 150,000 100,000 100,000 50,000 50,000 - - Djibouti Other Rural National 1st 2nd 3rd 4th 5th city regions district district district district district Source: Calculations based on EDAM4-IS. 7 Imputations were done according to two hedonic models: one for Djibouti city and another for regions. Imputations for rental values among tenants that were considered outliers were also carried out. 8 The exchange rate assumed is US$1 = DF 178. A simple transformation using the household size and welfare aggregate at the household level suggest that the average consumption per capita is about US$5.08 per day in 2011 purchasing power parity. CHAPTER 1 6 The distribution of expenditures of the basket of goods and rural areas in Djibouti is roughly in line with the results consumed in Djibouti is in line with other international in Anker (2011) where the average gap is 20.6 percentage examples. The Djiboutian population spends about 40 points among 11 countries analyzed.10 Nonfood percent of its expenditures on food (Figure 1.3). This consumption per capita spending is higher in Djibouti percentage is higher for households in the interior regions city (61 percent) as compared to the interior regions (52 (47 percent) and in rural areas (56 percent), while that percent) and rural areas in the regions (44 percent), mainly of Djibouti city is 39 percent. According to World Bank’s due to rents, electricity, and water costs. Rents are a large Global Consumption Database,9 it is found that food component of the overall consumption expenditures, share in total consumption at the national level in Djibouti especially in Djibouti city. is lower than some countries in the region such as Egypt (48 percent), Yemen (49 percent), or neighboring Ethiopia POVERTY LINES (58 percent), as well as other comparable countries such The second component to calculate poverty rates is as Sao Tome (68 percent) and Cabo Verde (50 percent). the poverty threshold or poverty line. The poverty lines Nonetheless, such national differences may follow the are based on the consumption basket of the Djiboutian varying degrees of the countries’ urbanization. The food population in 2017, and therefore they reflect a reliable share of Djibouti city compares reasonably well with the estimate of the minimum cost needed to cover the needs food share in urban areas of other countries in the region of Djiboutian households. To estimate the overall poverty such as Egypt (44 percent) and Yemen (40 percent). The line and the extreme poverty line, the approach suggested differences in the food expenditure share between urban by Ravallion (1998) was used. This consists of estimating FIGURE 1.3 Share of Expenditure on Each Category 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% National Djibouti city Other regions Rural Food and food away from home Housing services Water and electricity Tobacco, khat, alcohol, narcotics Others Source: Calculations based on EDAM4-IS. Notes: Other expenditures include those on education, health, flow of services from durables, fuel, transportation, communication, well-being, housing repair, clothing and footwear, and services. 9  See http://datatopics.worldbank.org/consumption/detail, accessed November 21, 2018. 10  They use data for 11 developing countries, including Bangladesh, Pakistan, Samoa, and Vietnam, for which both national and urban expenditure weights are available. 7 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI a food poverty line to then construct an extreme or “lower” all essential food and nonfood needs. A complementary poverty line, followed by an overall or “upper” poverty line. measure of poverty is obtained by calculating the overall poverty line. To obtain this line, we first identify households The food threshold or poverty line is defined as the product whose food expenditures are close to the food poverty of the minimum caloric intake that a basket of food line. Thus, the nonfood expenditure of this group is consumption must guarantee by the unit cost of acquiring considered to indicate the spending required to meet a a kilocalorie. This is to estimate the cost of a basket of household’s nonfood needs. Adding the food poverty line food items necessary to maintain day-to-day activities. and this nonfood component yields the overall poverty line. Using the cost-of-basic-needs methodology, 2,115 In earlier DISED reports, this upper threshold has been kilocalories per day is considered the minimum necessary referred to as a global poverty line. To avoid confusion to meet nutritional needs. A representative basket of the with other international indicators, we will refer to it as the consumer distribution community was used to obtain the overall poverty line. cost per calorie that determines the food poverty line. Adjustments to obtain the structure of the household The main parameters and results of the exercise members were based on an adult equivalent formula. concerning the estimation of the poverty line are as follows. Using the EDAM4-IS survey and the reference Once the food poverty line is set, Ravallion’s (1998) consumption basket, the extreme poverty line is estimated approach allows us to determine a poverty line, referred at DF 111,783 per year (DF 306 per day) and the overall to as the extreme poverty line. The nonfood component poverty line is estimated at DF 151,391 per year (DF of the extreme threshold is calculated by considering only 415 per day). These lines translate to US$2.17 in 2011 households whose total consumption corresponds to the purchasing power parity (PPP) and US$2.96 in 2011 PPP food poverty line. Based on this, nonfood expenditures per day, respectively.11 of these households are observed. Although these households have a total consumption equal to the food POVERTY IN DJIBOUTI IN 2017 threshold that would enable them to meet their basic food needs, these households nevertheless choose to DJIBOUTI POVERTY RATES BASED ON THE NATIONAL POVERTY LINE divide their consumption between food and nonfood. Using the adult equivalent consumption expenditure as As a result, these households consider that the part of a measure of household welfare and poverty line just the expenditure devoted to the acquisition of nonfood described, various indicators of poverty can be obtained. items further improves their level of satisfaction. These Table 1.2 presents the results. Measured by consumption nonfood expenses are then considered indispensable as per adult equivalent, the extreme poverty rate for the households end up sacrificing meeting their food needs whole country is estimated at 21.1 percent in 2017. In in order to cover for these nonfood items. This level of line with previous years’ studies (Box 1.2 nevertheless nonfood expenditure is added to the food threshold to highlights that the welfare aggregate and consequently obtain the extreme poverty line. poverty are not comparable over time), there appears to As in many countries, the extreme poverty line is used be a significant gap between the well-being of the capital in Djibouti to present the official poverty rates. The and other regions. Indeed, in Djibouti city, the extreme information from the survey, however, allows the estimation poverty rate is estimated at 13.6 percent, while in the of an overall poverty line. This line represents the value other regions, at 45 percent, it is more than two times that would enable households to cover, without sacrifice, higher than the national rate. Using the overall poverty 11 To go from an adult equivalent scale to a per capita scale, the poverty line in adult equivalent is adjusted by dividing it by the mean adult equivalent and multiplying it by the average household size as obtained from EDAM4. CHAPTER 1 8 Box 1.2 Issues with Comparability of Welfare and Monetary Poverty across Time in Djibouti In 2013, DISED revised the methodology for monitoring poverty. Using data from Enquete Djiboutienne Aupres des Menages- Budget Consommation (EDAM-BC) 2013, a basic-cost-of-needs approach was adopted to define food and total poverty lines. After defining the threshold, conventional price adjustments were done to find poverty lines corresponding to the previous 2002 and 2012 EDAMs. Estimated extreme poverty rates were 24.1 percent in 2002, 16 percent in 2012, and 23 percent in 2013. Such figures should not be used together with the results presented here. Previous EDAMs have gone through significant methodological changes, including changes to the consumption aggregate used to measure well-being, thus rendering it not strictly comparable across surveys. Three key components of the aggregate of well- being must be highlighted: 1. Food component: Over time, EDAM’s consumption data collection efforts have evolved by increasing the level of detail of the questions. These variations between the EDAMs limit temporal comparability. In EDAM2-IS 2002, the food component of household well-being was identified using a single question. For the 2012 EDAM3-IS, a 21-question module was developed to improve the data collected on food consumption. In addition, the recall period had been set at a week or a month. For EDAM-BC 2013, a log was used to collect consumption data instead of the recall. At the same time, information on a much more detailed list of articles was collected during the data collection. (It was possible to register 208 food items.) 2. Nonfood components: Like the food aggregate, there were significant differences in the data collection method and the level of detail of nonfood items. For example, in EDAM2-IS 2002, household nonfood expenditures were captured for about 10 items, while the list in EDAM3-IS 2012 comprised about 90 items. In EDAM-BC 2013, more than 500 nonfood items were collected at different recall periods. 3. Housing services: the well-being of neighborhoods where households live is one of the key elements in understanding their living conditions. In addition, the share of housing services in household welfare becomes more important as countries develop. In the case of Djibouti, the EDAM surveys of 2002 and 2012 have collected information on actual rent and shadow rent that are considered for housing services. For EDAM-BC 2013, the approach was very different, because housing and household characteristics were used to estimate a hedonic model. This model predicted housing services for owner occupied households. The EDAM4-IS 2017 questionnaire incorporated improvements based on current good practices to better capture household food and nonfood expenditures and turn it into a true multitopic tool. Despite the loss of comparability over time, DISED decided to seize the opportunity of the survey and to estimate well- being with the most recent methodologies. EDAM4-IS 2017 will become the baseline for monitoring the evolution of poverty in the country. Source: Adapted from DISED, 2018. 9 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 1.2 Poverty Indicators in Djibouti, 2017 National Djibouti city Other regions Other urban Rural 21.1% 13.6% 45.0% 14.8% 62.6% Extreme poverty (0.90) (1.10) (1.25) (2.04) (1.33) 7.1% 3.5% 18.6% 5.2% 26.4% Poverty gap (0.34) (0.38) (0.68) (1.07) (0.80) 3.4% 1.3% 10.1% 2.6% 14.4% Severity of poverty (0.19) (0.49) (0.19) (0.78) (0.60) 35.8% 28.2% 59.8% 27.6% 78.4% Overall poverty (1.08) (1.37) (1.24) (2.32) (1.07) Source: Calculations based on EDAM4-IS. Note: Standard errors in parentheses. line, the proportion of the population considered poor extreme poverty line, followed by the Dikhil region. In the increases considerably. Across the country, 35.8 percent capital, disparities are also highlighted: the fourth and fifth are unable to cover their food and nonfood needs. Among districts have extreme poverty rates three times higher households in other regions, this rate is even higher, at than the 1st district. Overall, poverty also remains high in 59.8 percent. the fourth and fifth districts of the capital city, with nearly a third of the population living below the overall poverty line Figure 1.4 shows spatial differences in Djiboutian in these districts (Table 1.3). Tadjourah and Dikhil are the population’s levels of well-being. Tadjourah region has regions that show the highest estimated extreme poverty the highest percentage of population living below the rates. This could be partly explained by the large share FIGURE 1.4 Extreme Poverty Rates, by EDAM4 Representative Domains 70% 60% 50% 40% 30% 20% 10% 0% ty s n l h l ah ck ta ct t t t t ra i l ric ric ric ric kh na on ba ie ci Ar tri bo Ru ur ab st st st st Di io gi ur is ti o O di di di di at ou re td iS dj er N d d h h Ta ib er Al 1s th 4t 5t 2n 3r Dj th O O Source: Calculations based on EDAM4-IS. Note: Line across graph denotes the national extreme poverty rate. CHAPTER 1 10 TABLE 1.3 Indicators of Poverty in Djibouti City, 2017 Djibouti city 1st dist. 2nd dist. 3rd dist. 4th dist. 5th dist. 13.6% 4.7% 9.8% 8.5% 18.3% 15.8% Extreme poverty (1.10) (2.21) (1.81) (1.78) (2.34) (2.26) 3.5% 1.0% 2.3% 2.1% 4.8% 4.2% Poverty gap (0.38) (0.51) (5.63) (0.58) (0.89) (0.81) 28.2% 13.0% 22.0% 15.4% 35.2% 33.6% Overall poverty (1.37) (2.92) (2.64) (2.20) (2.71) (2.72) Source: Calculations based on EDAM4-IS. Note: Standard errors in parentheses. of population living in rural areas in these two regions (77 possible the realities and costs of basic needs among percent and 64 percent, respectively) and consequently the Djiboutian population. The approach to estimate the limited income opportunities that their residents have the extreme and overall poverty lines relied heavily on access to. In fact, it is notable that this relatively high level the expertise of DISED. These national poverty lines are of deprivation can also be found across the international calculated using the basket of goods that are reflective border in Ethiopia. The regions of Afar (zone 1) and Somali of the consumption patterns of the Djiboutian population (Shinile zone) in Ethiopia share the national border with the in 2017 using data from EDAM4-IS. Nonetheless, to gain regions of Tadjourah and Dikhil in Djibouti. Both the Afar a broader view of the state of poverty in the country, it and Somali regions are classified as a “developing regional is important to use complementary benchmarks that state” by the Ethiopian Government. This classification help contextualize the results already shown. Here we identifies regions where poverty rates are higher and present results by applying alternative poverty lines to the performance on social indicators is poorer as compared to distribution of consumption in Djibouti in 2017. national averages. Indeed, the poverty rate using national A key target of SDG 1 of ending poverty in all its forms poverty line is higher than the national average of 23.4 everywhere is the eradication of extreme poverty by 2030. percent—38.5 percent in the Shinile zone of Somali region In order to assess the progress made against this SDG, it and 24.2 percent in zone 1 of Afar region in 2016. This is is necessary to monitor poverty based on an internationally not surprising, however, given the similar characteristics comparable benchmark. The global poverty rate is defined of border regions in the two countries. In Afar and Somali as the percentage of the population living below the regions, about 60 percent and 38 percent respectively rely international poverty line (IPL). The IPL was originally based on livestock as the main occupation of household head.12 on the International Comparison Program’s (ICP’s) 2005 PPP Large parts also face arid or semiarid climatic conditions, and valued at US$1.25 per person per month. Recently, the much like Djibouti’s border regions. ICP’s revision of 2011 PPP and new data led to the updated value of the IPL.13 Currently, the IPL is estimated to be DJIBOUTI POVERTY RATES BASED ON INTERNATIONALLY DEFINED POVERTY LINES US$1.90 (in 2011 PPP) per person per day.14 The results on poverty presented thus far are based on The World Bank’s (2018a) Poverty and Shared Prosperity national poverty lines calculated to reflect in the best way 2018 report recognizes that while the IPL is informative, 12  Comprehensive Food Security and Vulnerability Analysis (CFSVA), Ethiopia, March 2014 13  See Ferreira and Sánchez-Páramo (2017) for a review and a brief discussion. 14 The International Comparison Program adjustment factor in 2011 is 101.4806. This means that in 2011, US$1 was equivalent in purchasing power as DF 101.48. 11 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI the analytical underpinnings are derived from surveys metric called the societal poverty line, which helps take conducted before 2005. The relatively outdated foundation these considerations into account. The measure of societal (of average assessments of basic needs in low-income poverty line is equal to US$1.90 in 2011 PPP per day or countries) can limit its ability to measure poverty in a world US$1 in 2011 PPP plus half of the median consumption of where a high proportion of the population and global poor a person in the country, whichever is greater. The societal are now living in middle- and high-income countries. To poverty line accounts for extreme poverty, which is fixed account for these factors, the report develops additional for everyone at US$1.90 in 2011 PPP and adds a relative higher value lines, including the poverty line for lower- dimension of welfare. As a country progresses and the middle-income countries, the poverty line for upper- median consumption per person increases, the societal middle-income countries, and a societal poverty line. poverty line would increase as well, reflecting the change in the cost of basic needs. In the case of Djibouti, the Two more thresholds are explored to complement the societal poverty line is estimated at US$2.82 in 2011 PPP. picture of poverty in Djibouti. First, as the country is Djibouti’s societal poverty line is greater than the IPL, which classified as a lower middle-income country by the World reflects that the IPL may longer adequately reflect the cost Bank, the lower middle-income class poverty line with of basic needs in Djibouti. a value of US$3.20 (2011 PPP) becomes a relevant threshold. Second, in using the IPL, one assumes that Figure 1.5 presents the poverty rates associated with the the cost of basic needs is the same across all countries. IPL, the lower middle-income class poverty line, and the However, that is empirically shown not to be the case. As 15 societal poverty line for Djibouti and its regions.16 About 17 countries progress, US$1.90 in 2011 PPP may no longer percent of the Djiboutian population is found to live below reflect the cost of basic needs. Due to these reasons, the IPL, and 40 percent has consumption per capita below Poverty and Shared Prosperity (2018a) develops another US$3.2 in 2011 PPP. As expected, the pattern of poverty FIGURE 1.5 Poverty Rates Based on International (US$1.90 and US$3.20 2011 PPP per day) and Societal Poverty Lines 83% 77% 69% 66% 61% 61% 60% 54% 53% 51% 48% 42% 40% 32% 35% 34% 25% 28% 23% 17% 10% Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta National Poverty rate at 1.9$ 2011 PPP Poverty rate at 3.2$ 2011 PPP Poverty rate at societal poverty line Source: Calculations based on EDAM4-IS. 15  Refer to Poverty and Shared Prosperity (World Bank 2018b) for more details 16  Note that the welfare aggregate used in this exercise is in per capita terms and adjusted to 2011 PPP. CHAPTER 1 12 rates using these poverty lines across regions is the same from societal poverty in Djibouti, and the patterns across as poverty rates calculated using the national poverty line. regions persist. Taken together, these results suggest that Tadjourah registers the highest percentage of population the poverty-reducing agenda in Djibouti is still important. living below the IPL, followed by Dikhil. About two-fifths of Despite the strong period of economic growth of recent the population lived below US$3.2 in 2011 PPP nationally years, welfare levels are low, especially in certain regions in 2017. Meanwhile, about a third of the population suffers with a large rural population. POVERTY PROFILE The multitopic nature of the EDAM4-IS makes it is possible on inequality presents more details about the differences in to create a detailed profile of different groups of the the poor living in Balbala and the poor living in rural areas. population. In this section, we present results concerning The strong correlation between welfare and location the poorest population, those considered as extreme poor: (especially where areas with rural populations are more the 21.1 percent of the Djiboutian population with a level prominent) will become a common thread in the different of annual consumption below the extreme poverty line. pieces of analysis considered in this report. Djibouti city is The location of the poor population, the kind of dwellings entirely urban and constitutes three-fourths of the country’s they live in, and their educational and employment population. Tadjourah has the bulk of the population living in outcomes are emphasized. We contrast the outcomes of rural areas, followed closely by Dikhil and Obock. Since the poor and nonpoor populations in order to assess the key urban environment is more affluent than rural areas, regions distinguishing characteristics of the former in Djibouti. with a high proportion of the population living in rural areas will be shown to have weaker monetary and nonmonetary LOCATION CHARACTERISTICS indicators when compared to more urbanized regions. Table 1.4 shows the distribution of the poor and overall population across the country. About 12 percent of DEMOGRAPHIC CHARACTERISTICS country’s population lives in Tadjourah and Dikhil but these Nonpoor households differ in demographic characteristics regions host a disproportionate share of the country’s from the extreme poor households. The latter households poor. A third of the country’s extreme poor reside in these tend to be bigger, with an average of 7 household two regions. Based on the distribution of the population members, and have a higher dependency ratio than the from EDAM4, the rural population represents 15 percent nonpoor households. In fact, household size seems to be of the total population, but it hosts about 45 percent of highly correlated with poverty. Figure 1.6 shows that as the the population considered as extreme poor. On the other household size increases, the poverty rate of the population hand, 76 percent of the total population lives in Djibouti city living in these households is also higher. Households living and 49 percent of the poor population lives in Djibouti city. in extreme poverty are also younger and comprise a larger Finally, in terms of magnitude, we find a high concentration share of children (younger than 15 years of age). Among the of the extreme poor population in some districts of the extreme poor, this group constitutes 39 percent, whereas it capital. The fourth and fifth districts—belonging to the is 35 percent among the nonpoor (Table 1.5). Extreme poor community of Balbala—are the residence of nearly 37 households also have much higher dependency ratios (111 percent of the poor population in Djibouti. Taken together, percent) than nonpoor households. the poor population of the country is concentrated in rural areas (45 percent) and Balbala (37 percent). The section 13 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 1.4 Distribution of Extreme Poor and Overall Population across Locations (% population) Extreme poor Overall population By regions (outside Djibouti city) Djibouti city 49 76 Ali Sabieh 7 5 Dikhil 16 7 Tadjourah 17 5 Obock 5 2 Arta 6 4 National 100 100 By location Urban 55 85 Rural 45 15 National 100 100 Within Djibouti city Overall 49 76 1st district 2 8 2nd district 8 17 3rd district 2 5 4th district 20 24 5th district 17 22 Source: Calculations based on EDAM4-IS. FIGURE 1.6 Poverty Rate (percent), by Number of Household Members 30% 25% 20% 15% 10% 5% 0% 1 2 3 4 5 6 7 8 9 and above Source: Calculations based on EDAM4-IS. CHAPTER 1 14 In terms of age, marital status, and gender, household NONMONETARY INDICATORS heads in extreme poor and nonpoor households do not There are marked differences in access to services across seem to differ substantially. Extreme poor households different groups of the population. Figure 1.7 shows the are slightly more likely to have a non-Djiboutian head. In differences between the extreme poor and the nonpoor contrast, characteristics of heads are found to be different population, as well as the indicators of the overall on other dimensions. Nonpoor household heads have an population. The dwellings inhabited by the poor seem to average literacy rate of 46 percent as compared to be more precarious that those occupied by the nonpoor. 12 percent of the poor household heads. Not only that, the The poor population, as compared to nonpoor, is more former also have an average of 4.2 years of education as likely to live in a tent and less likely to live in a house with compared to 0.8 years among the latter. brick walls, sheet metal roof, and cement floor. Among Table 1.6 presents extreme poverty rates of children (0–17 the very poor, the percentage with access to electricity years) and men and women (18 years and over). There are 17 is 16 percent, while for the rest of the population it is slight disparities between the population groups analyzed. estimated at 72 percent. The gap in access to water is In particular, children do not seem to have a higher level of less pronounced: 69 percent and 96 percent, respectively. deprivation than adults in Djibouti city. Outside Djibouti city, Finally, disparities in access to garbage disposal services however, the differences are slightly larger. For example, in and toilets services are also evident. Among the very rural areas, the poverty rate among children is 67 percent poor population, 42 percent practices open defecation, and that among adults is about 58 percent. No gender compared with just 5 percent among the nonpoor differences in poverty rates are found among adults. population. Access to services such as electricity and TABLE 1.5 Demographics and Characteristics of Extreme Poor, Nonpoor, and Overall Population Nonpoor Extreme Poor Total Household size 6.2 7.0 6.4 Dependency ratio (%)** 85 111 90 Percentage in each group Children ages 0–14 years 35 39 36 Individuals ages 15–39 years 45 42 44 Individuals ages 40 years and older 21 19 20 Household head characteristics Age* 48.0 48.2 48.0 Married 80% 84% 81% Female 21% 20% 20% Djiboutian*** 96% 95% 96% Literate 46% 12% 39% Years of education 4.2 0.8 3.5 Source: Calculations based on EDAM4-IS. * The sample comprises 4,359 household heads. For the remaining households, the age of the household head is unknown. ** Only households with at least one working age (15–64) individual are included. *** Self-declared. 17 Extreme poverty rates are derived from household-level poverty classification and associated equivalent adult consumption. Therefore, the rates represent the percentage of children, men, or women who live in extreme poor households. 15 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 1.6 Extreme Poverty Rate, by Population Group National Djibouti city Other regions Rural Poverty rate among children (0–17 years) 23% 14% 48% 67% Poverty rate among men (18+ years) 19% 14% 42% 59% Poverty rate among women (18+ years) 19% 13% 41% 58% Global population 21.1% 13.6% 45.0% 62.6% Source: Calculations based on EDAM4-IS. FIGURE 1.7 Dwelling Characteristics and Access to Services by Population Groups a. Extremely Poor, Nonpoor, and Total Population 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% nt pl , k r n rb te al tri o ty er of d/ ick sp e oo an tio ec t os di ag ci ga ria te at ro el ess oo r ca w w in b op a t al en c fe to in et pr Ac de em ls ng tm ss Ap al ce n vi C W ee pe Li Ac Sh O Nonpoor Extreme poor All a. Urban, Rural and Total Population 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% nt pl , k r n rb te al tri o ty er of d/ ick sp e oo an tio ec t os di ag ci ga ria te at ro el ess oo r ca w w in b op a t al en c fe to in et pr Ac de em ls ng tm ss Ap al ce n vi C W ee pe Li Ac Sh O Urban Rural All Source: Calculations based on EDAM4-IS. Note: Walls in brick, wood/ plank include 3 categories: adobe bricks (cement block) in cement, wood/plank and brick/ wood. Access to electricity is defined as the use of electricity as the main source of lighting. Access to water is defined as the availability of water in the household in the form of running water (Office National de l’Eau et de l’Assainissement [ONEAD] indoor connection), direct connection from a borehole, and ONEAD outdoor connection by pipe, public fountain, and drilling (with a pump). Appropriate garbage disposal is defined as availability of garbage collector-OVD (public dump), availability of garbage collector-private, and garbage deposited in a designated place. CHAPTER 1 16 TABLE 1.7 Asset Ownership Rates by Poverty Status Nonpoor Extreme poor Overall Mobile phone 87.1% 55.4% 80.4% Laptop/Personal computer 13.3% 0.3% 10.5% Tablet 2.8% 0.0% 2.2% Radio 13.3% 8.0% 12.2% Television 61.4% 9.4% 50.4% Satellite dish 39.7% 5.8% 32.5% Refrigerator/freezer 44.6% 6.1% 36.4% Air conditioning 17.5% 0.7% 14.0% Washing machine 16.0% 0.8% 12.8% Microwave/oven 1.5% 0.0% 1.2% Private car 3.8% 0.3% 3.1% Source: Calculations based on EDAM4-IS. sanitation seem to be a priority focus area for public ownership rate is less than one percent for the following policy, going forward. Electricity access has been shown items: laptops, tablets, air conditioning units, washing to contribute to well-being in various ways. It has been machines, and cars. shown to stimulate consumption and income (Chakravorty, In terms of human capital accumulation and the ability of Pelli, and Marchand 2014; Van de Walle et al. 2017), individuals to progress economically, differences between promote income generating activities and entrepreneurship the extreme poor population and the rest of the population (Chowdhury 2010; Dinkelman 2011), enable better are also evident. There is a 20 percentage point difference education (Chandrasekhar and Amin 2012), and enable in the literacy rates of extreme poor and nonpoor (Table better health (World Bank 2008). 1.8). Large differences are also found in attendance at It must be kept in mind that about 45 percent of the poor school among those 6–14 years old (corresponding to live in rural areas while another 37 percent live in Balbala primary and secondary school ages) between the poor area of the capital. Thus, these results on extreme poor and nonpoor population groups. The poor of 6–14 years are driven by those living in rural areas (Figure 1.7). For are much more likely to have never attended school. example, the open defecation rates among those living in There are also gaps in access to primary school, with rural areas is 65 percent, while only 3 percent for urban only 6 percent of the nonpoor and 18 percent of the poor areas. Similarly, about 70 percent of rural residents are declaring unavailability of school. Additionally, 45 percent found to be living in tents while only 1 percent in the urban declare having a primary school less than 1 kilometer areas are found to be doing so. away. About a third among the extreme poor declare that a high school does not exist, while only 7 percent of the There are also marked differences in the assets owned nonpoor declare so. The nonpoor spend nearly four times by households based on their poverty status (Table on education per capita as compared to the poor. 1.7). Likely a reflection of the low capacity to save and purchase durable goods, or have access to loans for credit About 85 percent of individuals older than 25 years in the purchases, the poor have much lower rates of ownership very poor population have no education, and 2 percent of of radios, TVs, refrigerators, and even mobile phones. It this population has attended school but not attained primary is notable that among the extreme poor, the estimated education (Figure 1.8). Among the nonpoor population, the 17 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI corresponding percentages are 55 percent and 4 percent. attainment. There is again evidence of a strong correlation In addition, 18 percent of the nonpoor population has at between education and poverty status: the poverty rate least completed high school education. Figure 1.9 presents stands at 30 percent when the head has no education the poverty rate of the household head by educational whatsoever and drops sharply for heads with less than TABLE 1.8 Education Indicators among the Extreme Poor, Nonpoor, and Overall Population Nonpoor Extreme poor Total Literacy among those age 15 years and older 58.2% 31.8% 53.1% Percentage of children of 6–10 years attending school 83.8% 63.3% 78.4% Percentage of children of 11–14 years attending school 88.5% 71.1% 84.8% Percentage of children of 15–17 years attending school 72.2% 52.5% 68.2% Percentage of children of 6–14 years that never attended school 11.1% 30.8% 15.9% Percentage of households where household head has less than 61.8% 93.3% 67.8% primary (including no) education Maximum years of education among all adults in household 10.0 5.3 8.3 Percentage of population that declares no primary school is available 5.7% 17.9% 8.3% Percentage of population with a primary school within one kilometer 41.8% 41.7% 41.8% Percentage of population that declares no high school is available 7.4% 31.1% 12.3% Percentage of population with a high school within one kilometer 21.8% 13.7% 20.1% Annual per capita spending on education (DF) 9,126 2,197 7,662 Share of consumption expenditure on education 4.4% 3.5% 4.2% Source: Calculations based on EDAM4-IS. Notes: Education expenditures reflect households’ expenses on school subscription fee, cost of books and furniture associated with schooling, school and sports uniform, transport to and from school, and others (cost of accommodation, repetitions of class, special cours- es, and other services). FIGURE 1.8 Educational Attainment of Adults Age 25 and Older (percent) 2 18 11 15 2 21 23 4 4 85 55 61 Non- poor Extreme poor All No education/ own education Less than primary Primary and above, but less than secondary Secondary and above Source: Calculations based on EDAM4-IS. CHAPTER 1 18 FIGURE 1.9 Extreme Poverty Rates, by Educational Attainment of Household Head 30% 13% 7% 3% No education/ Less than primary Primary and above, Secondary and above own education but less than secondary Source: Calculations based on EDAM4-IS. Notes: Line indicates national poverty rate. primary education. The poverty rate is extremely low when educational outcomes may occur insofar as private, the head has completed at least secondary education. school-related expenditures affect children’s learning. These significant disparities across the poor and nonpoor This is due to the fact that regional disparities seen in on educational outcomes is important and must be kept in monetary indicators are mirrored by households’ education mind when we explore labor market outcomes. expenditures (Figure 1.10). The average household in Djibouti city spends slightly more than DF 9,000 per capita A majority of children (95 percent) attending school in on education, and in Tadjourah, about DF 1,200. Not Djibouti go to public schools. However, inequities in only is the difference in the levels of per capita education FIGURE 1.10 Education Expenditure per Capita and Share of Expenditure on Education in Total Household Expenditure, 2017 6.0% 10,000 9,000 5.0% 4.8% 8,000 4.0% 4.2% 7,000 6,000 3.0% 2.7% 5,000 2.7% 2.7% 4,000 2.0% 3,000 1.1% 1.0% 2,000 0.9% 1,000 0.0% - Djibouti city Ali-Sabieh Dikhil Tadjourah Obock Arta Total Education expenditure per capita Share of education expenditure in total Source: Calculations using EDAM4-IS Notes: Education expenditures reflect households’ expenses on school subscription fee, cost of books and furniture associated with schooling, school and sports uniform, transport to and from school and others (cost of accommodation, repetitions of class, special courses, and other services). 19 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI expenditure highest across regions, but also the share of (and their ability to spend) to treat their ailments. While these expenditures in total follows the same trend. It is over 77 percent of people reporting being ill go to a public thus not surprising that an average household spends 5 health facility for consultation, the nonpoor spend nearly percent of their total expenditure on education in Djibouti three times as much on health per capita than the poor do. city and only 1 percent in Tadjourah. The preceding sections have explored the concept of About 8 percent of the population reports having had an monetary poverty or the cost for a household to achieve a illness during the preceding year, with a lower share of the level of well-being considered as the minimum necessary extreme poor reporting being ill (Table 1.9). Meanwhile, and that allows the members to satisfy their food and the percentage of people that declared having certain nonfood needs. Separately, we explore its correlation with physical limitations is essentially the same across poor other nonmonetary indicators. As a way to track several and nonpoor. Access to health centers and hospitals 18 margins of deprivation and broaden our understanding appears to be limited among the poor population, with of poverty, it is important to supplement the monetary nearly a fourth of them declaring that a health center does measure of well-being with nonmonetary measures that not exist nearby. About 40 percent of the poor declare may improve welfare. In light of this, a deprivation score that a maternity clinic does not exist nearby, while only is calculated based on three dimensions (poverty, access 18 percent of the nonpoor do so. In addition to primary to education and access to services), and a total of six schools, access to health facilities such as health centers, potential deprivations that households can face (see Box hospitals, and maternity clinics is another public facility 1.3). The deprivation score ranges from 0 to 1: a score of for which the poor show lower coverage rates. Finally, 1 indicates deprivation on all 6 indicators and a score pf 0 disparities emerge when we look at households spending indicates no deprivation. TABLE 1.9 Health Indicators among the Extreme Poor, Nonpoor, and Overall Population Nonpoor Extreme poor Total Percentage of population reporting being ill in the prior 30 days 5.9% 4.4% 5.6% Percentage of population reporting being ill in the past year 8.4% 6.7% 8% Percentage of population declaring certain limitations (visual, 8.7% 8.6% 8.7% listening, walking, memory, and/or understanding) Percentage of households that declare no health center is available 6.5% 24.5% 10.2% Percentage of households with a health center within one kilometer 34.0% 31.1% 33.4% Percentage of households that declare no hospital is available 6.7% 26.8% 10.9% Percentage of households with a hospital within one kilometer 23.1% 19.3% 22.3% Percentage of households that declare no maternity clinic is available 17.5% 39.1% 22.0% Percentage of households with a maternity clinic within one kilometer 21.8% 15.8% 20.6% Annual health expenditures per capita (‘000s DF) 1,517 538 1,310 Share of household expenditure on health 0.6% 0.9% 0.7% Source: Calculations based on EDAM4-IS. Notes: Health expenditures include households’ expenses on consultation fees, medical exams, medications, hospitalization, and vaccination. 18  This estimate is based on the set of questions used by Washington Group on Disability Statistics to identify disability statistics. CHAPTER 1 20 Box 1.3 Multidimensional Poverty Index (MPI) Following the methodology proposed in Poverty and Shared Prosperity 2018 (World Bank 2018a), the dimensions used in the estimation of the deprivation score are given in table B1.3.1 below. The EDAM4- IS captures information for each of the indicators required. Thus, we first calculate each of the six indicators for all households surveyed in Djibouti. In the education component, the first indicator is a check on school attendance of all children ages 6–14 in the household. The second indicator is a check on whether any adult age 15 years and older in the household has completed primary school education. Access to electricity reflects the use of electricity as the main source of lighting. Access to water reflects the availability of water via running water (ONEAD indoor connection), direct connection from a borehole, ONEAD outdoor connection by pipe, public fountain. and borehole (with a pump). Improved sanitation services are defined as a water closet with flush and latrines with slab, as long as these are not shared with other households. Next, we apply the corresponding weights to each of the indicators. Finally, the summation of all the weighted components produces the deprivation score. TABLE B1.3.1 Dimensions and Their Corresponding Weights Dimensions Indicators Weight Monetary poverty Daily consumption per capita less than US$1.9 1/3 Education Any school aged child up to age of grade 8 is not attending school 1/6 No adult of grade nine and above has completed primary education 1/6 Access to services No access to improved water sources 1/9 No access to improved sanitation facilities 1/9 No access to electricity 1/9 Source: Poverty and Shared Prosperity 2018 (World Bank 2018a). Monetary poverty, children not attending school, adults Aggregating the information on all dimensions, we find without primary education, and lack of access to water, that richer households and richer regions have a lower sanitation, or electricity constitute the six deprivations deprivation score, but certain deprivations remain (Figure included in the deprivation score. In Djibouti, only about 1.11). Among the richest 20 percent of the population, a fourth of the population lives in households that face the deprivation score stands at 0.1. The results by region none of these deprivations, and about 30 percent lives in show a similar pattern as that of monetary poverty. Thus, households deprived in one of the indicators (Table 1.10). not only do the regions of Tadjourah and Obock have the The most common deprivations are access to sanitation highest percentage of extreme (monetary) poor population, and electricity, as 56 and 40 percent of the population they also have the highest deprivation score. This implies does not have access to these services, respectively. that coverage of basic services and education outcomes There is close to 30 percent of the population that is (proxied by attainment) are worse in these regions. In line deprived on three or more indicators, thus highlighting with the results shown earlier, we find lower deprivation the need to establish a multipronged strategy to address more typically in urban than rural settings. Notably, we find poverty in Djibouti. strong evidence supporting the importance of education in 21 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI reducing poverty and deprivation. Households headed by head does not appear to be linked to the households’ a highly educated person (secondary to tertiary education) level of deprivation. These findings point to the relevance show a fivefold reduction in the deprivation score of disparities in education, labor opportunities, and compared to uneducated ones and exhibit a near perfect location in explaining the level of poverty across Djiboutian lack of deprivation (0.09). The gender of the household households. TABLE 1.10 Percentage of Population Deprived on Various Indicators of the MPI Number of Percentage of deprivations population 0 24.7 1 30.3 2 15.8 3 11.5 4 7.8 Source: Calculations based on EDAM4-IS. 5 5.6 Note: Analysis is restricted to 4,418 households for which information 6 4.3 was found on all six dimensions. FIGURE 1.11 Deprivation score for Population Subgroups a. By quintiles of consumption per capita b. By gender of the household head 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 Male HHH Female HHH n im n y se ss but nd n y ab y e ar ar d ar ov tio pr tha co tha an nd a le y uc ss ar co im Le ed Se Pr o N c. By education of household head d. By region 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 Lowest 2 3 4 Highest ity h il ah ck ta kh ie Ar c bo ur ab Di ti jo O ou i-S d Ta ib Al Dj Source: Calculations based on EDAM4-IS. Note: Analysis is restricted to 4,418 households for which information was found on all 6 dimensions. HHH = head of household. Deprivation scores on y-axis. CHAPTER 1 22 ECONOMIC CHARACTERISTICS (Figure 1.12). About 54 percent of the nonpoor are out of the labor force, compared to 62 percent of the poor. The nonpoor population differs in important ways from the poor population in economic characteristics. An Among those employed, about a half of the nonpoor individual living in a poor household, on average, incurs a population work as salaried employees. Nearly 35 percent per capita consumption expenditure of about DF 59,000 of the poor are self-employed, and another third work annually, less than a quarter of the corresponding level as salaried employees (Figure 1.13). Among both the of a nonpoor individual, which is closer to DF 250,000. poor and nonpoor population, employment is primarily This low consumption expenditure could be directly in the tertiary or services sector. Nonpoor individuals linked to the household earnings, particularly employment work in the public sector almost at the same rate as the outcomes. Looking at the composition of the labor for all overall working population (46 percent and 44 percent, individuals of or older than the age of 15 years, we find respectively), whereas among the extreme poor only 31 that almost double of the nonpoor population is employed percent have a job in the public sector. as compared to the extreme or very poor population FIGURE 1.12 Employment Status of All Age 15 and Older, by Poverty Status (percent) 54 62 55 20 21 26 26 24 13 Nonpoor Extreme poor Total Employed Unemployed Out of labor force Source: Calculations based on EDAM4-IS. FIGURE 1.13 Type and Sector of Employment (percent of those employed, 15 years and older) a. Type of employment activity b. Sector of employment c. Percentage working in public sector 46% 44% 3 4 3 8 13 8 33 48 47 31% 11 5 6 87 76 85 22 35 23 21 21 5 4 5 18 1 7 1 Nonpoor Extreme Total Nonpoor Extreme Total Nonpoor Extreme Total poor poor poor Employer Self-employed Primary Secondary Worker Salaried Other Tertiary Other Source: Calculations based on EDAM4-IS. Note: Salaried employment is comprised of several types of employment. The sector of employment is calculated based on the branch of activity declared by the individual. Public sector includes those employed declaring public administration as the main branch of activity. 23 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI The low levels of labor market participation and employment gifts from family or friends, highlighting the importance of raise the question of what other sources of monetary and solidarity in the community. Finally, a possibility also exists in-kind resources are available to both poor and nonpoor that the household engages in primary sector economic populations. It is particularly important to understand the activities such as agriculture, rearing of animals, collecting income-generating activities of the poor population and the wood or performing other forestry activities, and fishing. networks on which they rely to finance their consumption Table 1.11 shows each of these different possibilities and and day-to-day activities. The information collected in the sources. Results indicate that 72 percent of the nonpoor EDAM4-IS questionnaire on a variety of topics provides a and 46 percent of poor rely on labor income, as one or good starting point to explore the sources and ways people more of its members is employed. Primary sector activities in Djibouti finance their consumption. such as agriculture, forestry and fishing are less prevalent than owning and/or raising animals. About 30 percent of The typical channel through which to finance day-to-day the poor population own animals, whose by-products are needs is income generated from employment. Another presumably consumed directly by them and/or sold. About potential source could be transfers. Private transfers can 11 percent of the poor also produce their own food. In take the form of cash and gifts received from family and terms of transfers, the poor are more likely to rely on both friends or income earned from physical and financial assets. private and public transfers to finance their consumption Public transfers include aid received from any of the social than nonpoor households. Overall, about 81 percent of the programs listed in the questionnaire. This population could nonpoor and 72 percent of the poor engage in any income also be adopting other measures to make ends meet, such generating activities. as self-production of food or receiving food in the form of TABLE 1.11 Percentage of Population with Access to Any Source to Finance Consumption Sources of income Nonpoor Extreme poor Total Any adult (15 years and older) employed in household 72% 46% 67% Engages in agriculture, gardening, forestry, or fishing 1% 5% 2% Owns animals/engages in animal rearing 6% 30% 11% Food is self-produced 10% 11% 10% Public transfer 5% 7% 6% Private transfers 10% 9% 10% Any of the above income-generating activities 81% 72% 79% Other sources and coping strategies Any household member retired or recently unemployed 6% 3% 6% Food received as gift 5% 8% 6% Employing money-generating coping strategies 8% 19% 10% Any of the above Sources of income, other sources and coping strategies 84% 78% 83% Source: Calculations using EDAM4-IS Note: The variable “any of the above sources” includes several sources: any adult (15 years and older) employed in the household; any household member who is retired or has become unemployed in the past two months; household using other money-generating coping strategy; households receiving private transfers or transfers from government social programs; households engaging in self-production of food or receiving it as a gift; households engaging in agriculture, gardening, forestry, or fishing activities; and households owning or engaging in animal rearing. CHAPTER 1 24 In addition to income-generating activities, other monetary COMPARING NONMONETARY INDICATORS OVER TIME and in-kind resources may be available to the population. A While monetary measures may be hard to compare across household might finance consumption by relying on income survey rounds, due to changes made to the questionnaire from retired members and/or from severance payments as well as the change in methodology to calculate poverty, of those who recently became unemployed. Only a small it is still possible to compare nonmonetary indicators of percentage of households use this source of income. Some well-being, such as access to services, amenities, and households may also resort to informal networks and certain assets. The similar coverage of the surveys, as well as coping mechanisms to generate money. About 8 percent the ability to define the indicators in a similar way, provide of the poor population declare receiving food as gifts. At confidence on their comparability across time. The the same time, an important share of the poor population evolution of these nonmonetary indicators would help us appears to rely on such coping strategies as selling assets, emphasize the trends in the evolution of welfare of the borrowing, buying food on credit, and begging, among Djiboutian population in the context of high economic others, to generate income, providing further evidence growth that the country has experienced in recent years. of their vulnerability. No explanation regarding the source Survey data from 2012 and 2017 suggest that the of financing is found for about 22 percent of the poor Djiboutian population experienced improvements in well- population in the different sections covered by EDAM4- being. Unfortunately, regional disparities that emerged IS survey. These results likely underrepresent the set of previously are also found among nonmonetary indicators, available income sources for households. Other potential with regions further away from the capital city performing ways to make ends meet that may not have been accurately worse. Furthermore, as evidenced by data from 2012 captured in the survey due to underreporting include and 2017, improvements of recent years had limited pursuing illegal activities, having children earn income (child effects in closing the gaps (Figure 1.14). The Tadjourah labor), or performing very infrequent tasks or jobs. region appears to have the worst access to services, TRENDS IN WELL-BEING followed by Obock and Dikhil in both years. Only one- fifth of the population has access to electricity in the In this section, the evolution of welfare in Djibouti in recent regions of Tadjourah, Dikhil, and Obock, and this has years is investigated as a way to understand whether hardly changed over time. In Djibouti city, access to water economic growth—through capital accumulation, mostly— has improved and is almost universal. Djibouti city is has led to tangible improvements in the livelihood of the followed by Ali Sabieh and Arta—a higher percentage of Djiboutian population. The main results use information the population declared having access to water in these from the EDAM 2012—the most recent nationwide data regions in 2017 as compared to 2012, too. Improvements collection effort for a household consumption survey. The are observed in other regions, also, with the exception focus is on nonmonetary dimensions and the perceptions of Tadjourah. It seems that proximity to the capital and of the population on what is happening in their community. the predominance of a rural population are important Regarding monetary welfare, an attempt is made to indicators that can help explain the difference in access compare poverty rates using EDAM-BC 2013 and 2017. to services between regions. In addition, it is found that However, given the differences in the questionnaire and these improvements in access are highest for the bottom the poverty measurement methodology, this exercise faces two quintiles of the population. issues and should be considered suggestive at best. 25 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE 1.14 Access to Services, by Year (percent of population) a. Access to electricity b. Access to improved water sources 98% 100% 100% 90% 82% 83% 80% 70% 80% 60% 60% 60% 49% 60% 52% 46% 40% 40% 23% 22% 21% 22% 20% 20% 0% 0% l ty h l ah ck ta l ty h l ah ck ta na hi na hi ie ie ci ci Ar Ar k k bo bo ur ur io ab io ab Di Di ti ti o o at at O O ou ou iS iS dj dj N N Ta Ta ib ib Al Al Dj Dj 2012 2017 Source: Calculations based on EDAM 2012 and EDAM4-IS. Note: Access to electricity reflects the use of electricity as the main source of lighting. Access to water in 2017 and 2012 is defined as availability of water in running water (ONEAD indoor connection) and ONEAD outdoor connection by pipe, public fountain, and borehole (with a pump). Access to water in 2017 had one additional category that was not included in 2012—direct connection from a borehole. FIGURE 1.15 Dwelling Characteristics and Access to Goods, by Year (percent of population) a. Living in a tent, spontaneous habitat, others b. Sheet metal as material of roof 82% 74% 74% 71% 65% 68% 57% 44% 32% 30% 28% 26% 14% 3% l ty h il h k ta l ty eh il h k ta na na kh kh c c ie ra ra ci ci Ar Ar bo bo bi io ab io ou ou Di Di ti ti Sa at at O O ou ou iS dj dj N N Ta i Ta ib ib Al Al Dj Dj c. Cement as material of floor d. Ownership of mobile phone 91% 80% 55% 63% 49% 52% 41% 45% 46% 35% 36% 33% 24% 15% l ty eh l ah ck ta l ty h il h ck ta na i na kh kh ie ra ci ci Ar Ar bo bo bi ur io io ab ou Di Di ti ti Sa o at at O O ou ou iS dj dj N N i Ta Ta ib ib Al Al Dj Dj 2012 2017 Source: Calculations based on EDAM 2012 and EDAM4-IS. CHAPTER 1 26 Dwelling characteristics also reflect improvements over To complement this section, we explore whether the time and are consistent with earlier regional patterns improvements observed in access to services, dwelling (Figure 1.15). It is found that a lower percentage of the characteristics, and to some extent accumulation of Djiboutian population is living in tents or spontaneous assets are also reflected in people’s perceptions. We habitats in 2017 as compared to 2012. There is also a exploit a perception question related to the evolution higher proportion of the Djiboutian population living in of poverty in a household’s community that was asked houses with sheet metal roofs. These results are largely both in a 2012 survey and in EDAM4 in 2017. Along the driven by the capital city, Ali Sabieh, and Arta. It seems lines of the indicators just shown, there seems to be a that many of these characteristics have not changed positive perception about the evolution of poverty in 2017 in Tadjourah, Dikhil, and Obock. The percentage of the as compared to 2012 (Table 1.12). About a third of the population living in dwellings with a cement floor has population perceived a reduction in poverty in the previous also increased, pointing to the fact that the precarity of five years and expected continuing improvement in the next the dwellings has decreased. A higher penetration of five years in 2017, as compared to 13 percent in 2012. mobile phones is also found across all the regions of About 41 percent perceived that poverty would decrease in Djibouti, with the exception of Tadjourah. This percentage the next five years, while only 25 percent believed so in 2012. increased from 64 percent in 2012 to 80 percent in 2017. Thus, overall, there seems to be an expectation or perception of things getting better in 2017, compared to 2012. TABLE 1.12 Perceptions on the Evolution of Poverty (percent population by response category) a. 2017 “In the next five years, do you think poverty in your community will . . .” Decrease Be the same Increase Total Decreased 31 3 1 35 In the past five years, do you “ Not changed 5 23 3 31 think that in your community poverty has . . .” Increased 4 6 23 33 Total 41 32 27 100 b. 2012 “In the next five years, do you think poverty in your community will . . .” Decrease Be the same Increase Total Decreased 13 3 1 17 In the past five years, do you “ Not changed 4 14 6 24 think that in your community poverty has . . .” Increased 7 7 45 59 Total 25 23 52 100 Source: For panel a, calculations based on EDAM4-IS. For panel b, calculations are based on EDAM 2012. Notes: Only households that answered both questions were included: for panel a, 3,480 households; for panel b, 3,436 households. 27 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI The positive expectations of the population, as the distribution may be less optimistic than the average. reflected by subjective indicators could be driven by the Breaking down the indicator on perceptions of poverty by known improvements in access to services, dwellings location shows evidence of these disparities. The positive characteristics, and certain spillovers from the high prospects about poverty between 2012 and 2017 are driven growth rates Djibouti experienced in recent years. Thus, by residents of Djibouti city (Table 1.13). In 2012, 53 percent perceptions at the national level may also be composed of those residing in Djibouti city thought that poverty would of heterogeneous views across groups of the population. increase in the next five years. However, in 2017, this share If economic growth has been concentrated in only certain had halved to 24 percent. Much less improvement in positive groups (as it can happen in high-inequality contexts), perceptions was seen for those living in regions. then one could expect that expectations in the bottom of TABLE 1.13 Perceptions on the Evolution of Poverty by Location and Year In the next five years, do you think poverty in your community will . . . a. 2017 Djibouti city Regions Decrease Decrease Increase Increase Be the Be the same same Total Total “ In the past five Decreased 37 3 0 40 13 4 3 20 years, do you Not changed 4 23 3 30 9 19 5 33 think that in your community Increased 4 5 20 29 6 9 32 47 poverty has . . .” Total 45 30 24 99 27 33 40 100 b. 2012 Djibouti city Regions Decrease Decrease Increase Increase Be the Be the same same Total Total “ In the past five Decreased 12 3 1 16 16 1 2 18 years, do you Not changed 4 5 7 16 5 5 4 14 think that in your community Increased 5 6 45 56 15 9 44 67 poverty has . . .” Total 21 15 53 89 35 15 50 100 Source: Calculations based on EDAM4-IS and EDAM 2012. Notes: Only households who answered both questions were included (3,480 households in 2017 and 3,436 in 2012). Cells correspond to the percentage of the population. CHAPTER 1 28 COMPARING MONETARY MEASURES ACROSS 3. Nonfood consumption: Like the food aggregate, EDAM3-IS (2013) AND EDAM4-IS (2017) there were significant differences in the data collection There have been many changes in the poverty method and the level of detail of nonfood items. measurement methodology among the various household For EDAM-BC 2013, a log or diary was used to consumption surveys implemented Djibouti (see Box collect consumption data instead of recall, which 1.2). This hinders the ability to make straightforward was used in EDAM4-IS. In EDAM-BC 2013, more comparisons of data from multiple rounds of surveys, as than 500 nonfood items were collected at different is typically done in other contexts. Comparing welfare recall periods. This included expenditures on over time and understanding whether the living standards health, education, electricity, and water. However, of the Djiboutian population have improved over time in EDAM4-IS, these expenditures were moved to requires making strong assumptions in the production different sections and hence were better placed in the of estimates. In this subsection, an attempt is made to questionnaire. study the evolution of poverty by adjusting the welfare aggregates and poverty lines over the years to make 4. Durables: In EDAM-BC 2013, the purchase value of them comparable. The results must be considered at best durable goods acquired by the household during the illustrative and not conclusive of the evolution of monetary reference year was included in the welfare aggregate. welfare in Djibouti. In contrast, the welfare measure obtained in EDAM4 incorporated an estimate of the flow of services from There are several reasons why welfare and poverty rates in the current market value of each durable good. To 2017 are not comparable to those in 2013. The key points account for this difference, expenditures on durables related to the noncomparability of data are as follows: in both 2013 and 2017 are not included. 1. Complementarity: The EDAM-BC survey was 5. Temporal adjustments: Welfare and poverty lines administered only in Djibouti city in 2013. However, are estimates in current prices for each survey round to produce national estimates, data from EDAM3- and represent the cost of basic needs at the time of IS conducted in 2012 was used to supplement implementation. Adjustments are required to take into the expenditure information for regions and rural account differences over time in the cost of living in areas. The process that was used to implement this the country. adjustment is unclear. There is no bulletproof way to convincingly address all the 2. Food consumption: For EDAM-BC 2013, a points just noted. Instead, this exercise provides a set of log or diary was used instead of recall to collect alternative assumptions that would lead to the estimation consumption data. At the same time, information on of pseudo-comparable welfare aggregates and poverty a much more detailed list of articles was collected lines for both 2013 and 2017. This yields four imperfect during the data collection (it was possible to register approaches: more than 200 food items). In EDAM4-IS 2017, consumption of 100 food items was collected using Approach 1: Using data from EDAM-BC 2013, a seven-day recall period. As noted in Backiny-Yetna, a basic cost of needs approach was adopted to Steele, and Djima (2014) and Ahmed Brzozowski, define food and total poverty lines in 2013. A similar and Crossley (2006), these differences in the data methodology was used in EDAM4-IS 2017, and the collection process may lead to differences in the poverty lines were defined based on the consumption actual data collected. basket of the Djiboutian population in 2017. In 29 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI order to make poverty lines comparable, the food Thus, the food expenditures in 2013 are adjusted by poverty line in 2017 is adjusted by 11.3 percent increasing them by 10 percent. and the nonfood part is adjusted by 0.54 percent, Approach 4: Following the same adjustments is corresponding to temporal adjustments between May approach 3, the final approach allows the food 2013 and May 2017. expenditures in 2013 to be adjusted downward. Approach 2: The poverty line of 2017 is used and the Food expenditures in this approach are adjusted by food part of the welfare aggregate in 2013 is adjusted decreasing them by 5 percent. by 11.3 percent and the nonfood part is adjusted by Figure 1.16 presents the results for poverty rates in 0.54 percent, corresponding to temporal adjustments Djibouti city of all four approaches. To the extent that the between May 2013 and May 2017. welfare aggregates are comparable, it is found that the Approach 3: In order to make poverty lines extreme poverty rate seems to have dropped in Djibouti comparable, the food poverty line in 2017 is adjusted city between 2013 and 2017, irrespective of the approach by 11.3 percent and the nonfood part is adjusted by applied. Under approaches 2 and 4, there appears to have 0.54 percent, corresponding to temporal adjustments been a drop of about 5 percentage points, whereas under between May 2013 and May 2017. There is no approach 3 the decrease is of about 2 percentage points. consensus in the literature as to the magnitude as well Considering all the caveats, this may be interpreted as as the direction of the difference with respect to diary suggestive evidence that, similar to the results shown in and recall as a mode of collecting food expenditures. the section on nonmonetary indicators, the living standards While some research reports higher expenditures of the Djiboutian population may have improved slightly in using a diary method, others report the opposite. the past few years. FIGURE 1.16 Comparison of Poverty Rates between 2013 and 2017, Djibouti City 22% 20% 18% 16% 14% 12% 10% 8% 6% 4% 2% 0% 2013 (1) 2013 (2) 2013 (3) 2013 (4) 2017 Source: Calculations based on EDAM3-BC and EDAM4-IS. Notes: The approach used for 2013 estimation is labeled in parenthesis. See text for explanation of the approaches. In approach 1, the food poverty and the nonfood poverty lines in 2017 are adjusted separately, corresponding to temporal adjustments between May 2013 and May 2017. In approach 2, the welfare aggregate of 2013 is temporally adjusted separately for food and nonfood parts, and the pov- erty line of 2017 is used. In approach 3, food expenditures are increased by 10 percent. In addition, the food poverty and the nonfood poverty lines in 2017 are adjusted separately, corresponding to temporal adjustments between May 2013 and May 2017. In approach 4, food expenditures are decreased by 5 percent In addition, the food poverty and the nonfood poverty lines in 2017 are adjusted sepa- rately, corresponding to temporal adjustments between May 2013 and May 2017. Vertical bars show confidence intervals. CHAPTER 1 30 INEQUALITY IN DJIBOUTI To provide a better view of welfare and development in a complement, we also discuss inequality form the angle the country, it is important to understand the depth of of disparities in access to opportunities for children. The inequality. As discussed in the literature, there are several section concludes with an in-depth look at the gaps in reasons to worry about inequality. For instance, inequality well-being between the poor of urban and nonurban areas. can be associated with lower growth, and policies that tackle inequality may turn into the main drivers of poverty MONETARY INEQUALITY reduction, especially as economies reach higher levels of The welfare aggregate obtained from EDAM4 is helpful in development (Olinto, Lara Ibarra, and Saavedra-Chanduvi understanding monetary inequality in Djibouti. Following 2014; Marrero and Serven 2018; Brueckner and Lederman common practice, we use per capita consumption to 2015). In order to continue promoting poverty reduction explore differences in consumption levels across the at the pace needed to meet SDG 1, it will be as important population, as opposed to the per adult equivalent to enact policies that help tackle inequality as policies that measure that was used to measure poverty in the earlier boost economic growth. Inequality has also been shown sections. There are wide disparities in the levels of well- to be correlated with negative well-being perceptions, to being of Djiboutian population. Figure 1.17 shows the lead to social tensions (Cramer 2003), and in some cases annual consumption per capita by decile. The population be concurrent with low mobility (Dang and Ianchovichina of the richest decile has an estimated per capita 2018). While the disparities in levels of consumption consumption level that is more than 16 times the average presented so far provide evidence that there are certain household consumption in the first decile. In addition, segments of the population that continue to live in extreme the richest decile has a consumption twice as high per poverty and have lower outcomes in other nonmonetary capita as the 9th decile. Figure 1.18 also highlights these indicators, it is still important to assess the magnitude of disparities in consumption per capita geographically. some of these systemic and structural barriers to evaluate Mirroring high poverty rates, rural areas have the lowest how excluded certain vulnerable groups are. consumption per capita, less than half of the national average. Djibouti city has a high consumption per capita, We begin by showing (traditional) indicators related although disparities exist. The fourth and the fifth districts to monetary poverty, before presenting results on the of the capital city together make up the area of Balbala, population’s subjective perception about well-being. As which has the lowest consumption per capita. FIGURE 1.17 Annual Consumption per Capita, per Decile 800,000 700,000 600,000 500,000 400,000 300,000 200,000 100,000 0 1 2 3 4 5 6 7 8 9 10 Source: Calculations based on EDAM4-IS. Note: Deciles are calculated based on per capita consumption. 31 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE 1.18 Annual Consumption per Capita, across Regions and Districts of Djibouti City (DF) a. Consumption across major regions b. Consumption across Djibouti city districts 400,000 400,000 350,000 350,000 Djiboutian Francs Djiboutian Francs 300,000 300,000 250,000 250,000 200,000 200,000 150,000 150,000 100,000 100,000 50,000 50,000 - - Djibouti Other Rural National 1st 2nd 3rd 4th 5th city regions district district district district district Source: Calculations based on EDAM4-IS. The Gini coefficient of the consumption per capita on durables in both 2013 and 2017 are not included in distribution is estimated at 0.42 (Table 1.14). It is notable the welfare aggregate for the purpose of this analysis. that, compared to the capital city, the rest of the country Next, the food part of the welfare aggregate in 2013 is has higher levels of inequality—probably linked to the adjusted by 11.3 percent and the nonfood part is adjusted marked differences between the main cities in the regions by 0.54 percent, corresponding to temporal adjustments and the welfare levels in the region’s rural areas. Within between May 2013 and May 2017. Nevertheless, the regions, Tadjourah and Dikhil show the largest monetary results presented must be considered as illustrative and inequality as measured by the Gini coefficient. Also, not conclusive of the evolution of inequality in Djibouti. We these regions show the largest disparities between the find a very slight decline in the Gini coefficient in 2017 in 90th and the 10th percentile of consumption per capita. Djibouti city as compared to in 2013 (Table 1.14). In Tadjourah, households in the 90th percentile have a A decomposition of inequality using the Theil index shows consumption level 8.2 times higher than those in the 10th that the majority of inequality in Djibouti is coming from percentile. In Dikhil, this ratio stands at 7.6. disparities in consumption in the capital city. With a national- We make an attempt to study the evolution of inequality level Theil index estimated at 0.326, inequality within Djibouti by adjusting the welfare aggregates between 2013 and city explains about 78 percent of this inequality. Inequality 2017 to make them comparable for Djibouti city only. In between the capital, other urban areas, and the rural areas EDAM-BC 2013, the purchase value of durable goods explain about 5 percent of total inequality.19 Focusing only acquired by the household during the reference year was in the capital, districts 2 and 5 explain the largest share included in the welfare aggregate. In contrast, the welfare of inequality in Djibouti city, representing 23.7 percent of measure obtained in EDAM4 incorporated an estimate of total inequality each. This results from the combination the flow of services from the current market value of each of (relatively) high population shares, high share of total durable good. To account for this difference, expenditures aggregate consumption, and high inequality. 19  This does not imply that inequality is low in each of the regions, but that its contributions to total inequality are low CHAPTER 1 32 TABLE 1.14 Indicators of Inequality a. By location National Djibouti city Other regions Other urban Rural Gini 0.42 0.40 0.42 0.34 0.38 p90/p10 6.70 5.51 8.20 5.51 6.18 p90/p50 2.53 2.52 2.83 2.23 2.42 p10/p50 0.38 0.46 0.35 0.41 0.39 p75/p25 2.52 2.34 3.04 2.27 2.43 b. By region Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta Gini 0.40 0.33 0.42 0.52 0.38 0.30 p90/p10 5.51 6.52 7.68 8.27 7.51 5.62 p90/p50 2.52 2.24 2.56 2.79 3.09 2.47 p10/p50 0.46 0.34 0.33 0.34 0.41 0.44 p75/p25 2.34 2.80 2.92 2.90 3.08 2.47 c. Comparison of Gini coefficient over time, Djibouti city 2017 2013 Gini coefficient 0.396 0.406 Source: Calculations using EDAM3-BC and EDAM4-IS. Notes: p90/p10 shows the ratio of per capita consumption of individuals from the 90th percentile to those from the 10th. Percentiles are calculated based on per capita consumption FIGURE 1.19 Gini Coefficient across the World, circa 2015 Tunisia 70 Egypt, Arab Rep. West Bank and Gaza 60 Yemen, Rep. Iran, Islamic Rep. Morocco 50 Djibouti 40 30 20 10 0 Slovenia Czech Republic Slovak Republic Belarus Finland Belgium Denmark Timor-Leste Sweden Hungary Tunisia Germany Egypt, Arab Rep. Mongolia Armenia Mauritania Estonia Liberia Pakistan Luxembourg Tajikistan Niger Burkina Faso Portugal Romania Greece Spain Yemen, Rep. Lithuania Russian Federation Iran, Islamic Rep. Morocco Uruguay Kenya Cote d'Ivoire Djibouti Uganda Togo Peru Ecuador Cameroon Paraguay Benin Costa Rica Panama Brazil Zambia South Africa Source: WDI, accessed October 19, 2018; and EDAM4-IS. 33 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Using consumption data from EDAM4, we calculate The high monetary inequality indexes, as well as the the Gini coefficient for Djibouti and compare it to Gini disparities between urban and rural areas, provide coefficient for 95 countries as obtained from the World suggestive evidence that the benefits from economic Development Indicators website (Figure 1.19). The growth might be reaped by only certain segments of the coefficients are obtained from country surveys circa 2015. population in Djibouti. As far as the evolution of inequality Djibouti is in the 71st place out of the 95 countries shown. is concerned, there are no comparable consumption or Monetary inequality in the country, as measured by the income indicators over time that would allow a study. Gini coefficient, is higher than that of other Middle East and Nevertheless, changes in access to services and dwelling North Africa (MENA) countries and Ethiopia. Uganda has a characteristics (as seen in the subsection on trends in similar estimated Gini (0.43). well-being) could shed some light in this regard. Table 1.15 TABLE 1.15 Access to Services, By Quantiles Access to Appropriate Access to public electricity Access to water garbage disposal sanitation network Quantiles 2012 2017 2012 2017 2012 2017 2012 2017 1 9% 13% 63% 67% 41% 43% 4% 14% 2 33% 45% 86% 91% 59% 79% 11% 23% 3 60% 71% 93% 97% 72% 85% 17% 30% 4 80% 81% 94% 98% 78% 90% 20% 29% 5 91% 92% 95% 99% 87% 93% 28% 35% Total 55% 60% 86% 90% 67% 78% 16% 26% Source: Calculations based on EDAM4-IS and EDAM 2012. Notes: Quantiles are calculated based on per capita consumption. Access to electricity is defined as the use of electricity as the main source of lighting. Access to water is defined as the availability of water in the household in the form of running water (ONEAD indoor connection), direct connection from a borehole, ONEAD outdoor connection by pipe, public fountain and drilling (with a pump). Access to water in 2017 had one additional category that was not included in 2012—direct connection from a borehole. Appropriate garbage disposal is defined as availability of garbage collector-OVD (public dump), availability of garbage collector-private, and garbage deposited in a designated place. Access to public sanitation network was elaborated in 2017 to include public sanitation networks (trench, open concrete channel/slab, river network), and sanitation networks directly linked to the collective network from the house. presents the evolution of nonmonetary indicators along around inequality. As far as the self-perceived welfare is the 2012 and 2017 consumption distribution points. concerned, as one moves from the bottom 20 percent There seem to be large gains in access to electricity and (bottom quintile), or the poorest households, to the top appropriate garbage disposal in the middle deciles, while 20 percent (top quintile), or the richest households, the improvements in access to water are found throughout declared welfare category in which the household places the consumption distribution. The highest gains in access itself improves as well (Figure 1.20). For example, 39 and to public sanitation network are found in the lower and 44 percent of the bottom quintile perceive themselves to middle deciles up until decile 9. The findings show that be very poor and poor, respectively. On the other hand, 62 nonmonetary indicators saw improvements across the percent of the top 20 percent perceive themselves to be distribution without uniquely benefitting the richest. middle class, while only 14 percent consider themselves to be rich or very rich. Thus, there seems to be some Perceptions among members of the population about coherence between the distribution of the population their own welfare and how they place themselves in the across income groups according to both objective and consumption distribution are closely tied to the perceptions subjective measures of welfare. CHAPTER 1 34 FIGURE 1.20 Self-Reported Welfare Category, by Quintiles of Consumption per Capita (percent) 1 2 2 2 4 16 10 34 50 58 44 62 43 37 32 39 20 21 11 7 4 Bottom 20% 2 3 4 Top 20% Very poor Poor Middle Rich Very rich Source: Calculations based on EDAM4-IS. Notes: Bottom 20 percent represents the first quintile of the consumption per capita distribution. DEPRIVATION IN DJIBOUTI: HUMAN goods or services that everyone would agree should be OPPORTUNITY INDEX universal. Access (measured by coverage rates) paints only part of the picture of how a certain good or service In addition to the conventional measures of well-being, is distributed across the population, especially when a such as poverty and inequality, it is worth investigating particular subgroup of the population is consistently left access to opportunities and resources among the out of coverage. The HOI is an index that starts with vulnerable to assess mobility or the chances that they the coverage rate of a good or a service and applies have to escape poverty and achieve outcomes at par a correction that reflects the extent to which coverage with the rest of the population. Labor markets have been is equally distributed across different subgroups of the found to be crucial for mobility, in addition to access to population (defined by circumstances like gender or essential services early on in life such as health, education, location). The more that certain subgroups have lower and infrastructure (Krishnan et al. 2016). The existence coverage than the average, the higher the penalty and the of inequality in opportunities may help explain the lower the HOI. disadvantage that certain groups face that may determine outcomes later on in life. Table 1.16 provides a list of indicators chosen as the opportunities for children in Djibouti and the circumstances To complement the view on inequality in Djibouti, we used to create the subgroups of interest. We follow estimate the human opportunity index (HOI) using the literature (Ferreira et al. 2008; Krishnan et al. 2016; the EDAM4-IS data, in line with the widely applied Narayan et al 2018) in defining the opportunities set. methodology found in the literature (Paes de Barros et Certainly, these should not be taken as a comprehensive al. 2009; Ferreira et al. 2008; Krishnan et al. 2016). This list of the opportunities that should be available to a child index rests on the principle that the circumstances into to achieve his or her potential in life. The opportunities which a child is born should not determine his or her considered for the HOI estimation fall into three main access to opportunities, where opportunities are basic categories: education, basic housing services, and 35 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI household poverty. For education, we consider three drivers of public health, as they are known to reduce the opportunities: the ability to read and write, attendance incidence of diarrhea and its long-term consequences, or enrolment in school, and achievement of at least such as malnutrition. Finally, we study the opportunity of a primary education. For basic housing services, access to child to be raised in a family that is not multidimensionally a safe and improved water source, improved sanitation poor, defined as a deprivation score less than 0.33. All the facilities (water closet with flush or latrine with slab, and opportunities are analyzed for children ages 6 to 14 years, not shared with other households), and electricity are inclusive. The age reference for literacy is 10 to 14 years of the opportunities assessed. Water and sanitation are key age and completion of primary education is 12 to 16 years. TABLE 1.16 Opportunities and Circumstances Used in Human Opportunity Index Calculation Category Opportunity Description Opportunities Basic housing services Water Dwelling has improved water access (children ages 6–14 years) Sanitation Dwelling has improved sanitation access (children ages 6–14 years) Electricity Dwelling has electricity (children ages 6–14 years) Education Literacy Child can read and write (children ages 10–14 years) Attending school Child is currently attending school (children ages 6–14 years) Completed primary Child completed at least elementary school (children ages 12–16 education years) Not multidimensionally Child is not in a multidimensionally poor household (children ages Household poverty poor 6–14 years) Circumstances Child characteristics Gender Binary variable equals 1 if child is male Children in the Household characteristics Number of members ages 0–14 years living in the same household household Binary variable equals 1 if there is at least one household member Elderly presence age 65 years or older Single parent household Binary variable equals 1 if household head’s partner is not present Head of household characteristics Age Age of household head (years) Education of father broken into four categories by level of Educational attainment attainment Binary variable equals 1 if household head works in the public Public worker sector Location Rural Binary variable equals 1 if household is located in a rural community Region Series of binary variables that refer to region of residence Household consumption per capita mapped to a quintile of the Socioeconomic status Consumption quintile consumption per capita distribution in the country CHAPTER 1 36 We use a list of circumstances (that is, characteristics Among the seven opportunities analyzed, improved water of a child that he or she cannot control) categorized access has the highest coverage rate and HOI (Figure into five main groups: child characteristics, household 1.21). The second highest “scores” are found in literacy, characteristics, household head’s (or father’s) as 85 percent of all children in Djibouti ages 10–14 years characteristics, location, and socioeconomic status. The are literate, with a corresponding inequality penalty of 4 gender of the child is the first circumstance used. For percent. School attendance is still widely available among household’s characteristics, three circumstances are Djiboutian children (ages 6–14), with a coverage rate of looked at: the number of children ages 0–14 years, the 82 percent. In education, the lowest scores are found presence of at least one elderly member (age 65 years or in completion of primary education, with a coverage of older), and whether the household is headed by a single 68 percent and with access broadly distributed across parent. For characteristics of the head of household, all circumstance groups. In terms of dwelling facilities, age and employment in public sector are used, as well access to improved sanitation and access to electricity as the educational attainment of the child’s father. For are, not only limited, but unequally distributed among the location, urban or rural and region of residence are the population. Access to electricity as a basic service has circumstance groups. Socioeconomic status, defined as the second highest penalty of circumstance groups out of the wealth quintile (per capita expenditures) that a child is all opportunities analyzed in this report, reaching an HOI born into, is the last circumstance group. of just 44 percent. Finally, about two-thirds of children are not “MPI poor,” but there are large differences across circumstance groups: the HOI is only 51 percent. FIGURE 1.21 Coverage Rates and Human Opportunity Index for Children’s Opportunities 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Improved water Improved Electricity Literate School Completed Not MPI poor access sanitation access attendance primary access education HOI Coverage Source: Calculations based on EDAM4-IS. Notes: Definition of opportunities in Table 1.16. 37 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE 1.22 Inequality Decomposition of Disparities in Access to Opportunities 100% 90%100% 80% 90% 70% 80% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% Improved Improved ImprovedElectricity Improved Electricity Literate Literate School Completed School Completed Not MPI Not MPI poor poor water access water sanitation access sanitation access access attendance attendance primary primary access access education education Child characteristics HHHH characteristics Child characteristics characteristics HHH HHHcharacteristics characteristics Location Location Social status status Source: Calculations based on EDAM4-IS. Notes: Definition of opportunities in Table 1.16. HH = household; HHH = head of household. To better understand the nature of the inequalities in the Improvements in educational outcomes such as school distributions of opportunities, a decomposition of the HOI’s attendance and primary school completion appear to be D-index is estimated to explain the marginal contribution of linked to the income level of the household. Thus, efforts each circumstance to the overall inequality of opportunity. 20 that improve economic opportunities—and hence the In general, socioeconomic status tends to explain the income-generating capacity of households—are likely largest variation in inequality, followed by location and to have second order effects on improving educational head of household characteristics (Figure 1.22). Child outcomes of those children and improving their potential characteristics (gender) and household characteristics for future income generation. Finally, improvements in often explain little of the inequality in opportunity. However, provision of services in harder to reach rural areas in inequality in literacy and school attendance are explained regions is an important policy action going forward. by household characteristics for 24 and 22 percent of their observed inequality, respectively. Socioeconomic status COMPARING THE URBAN POOR AND explains about 49 percent or more of inequality in access NONPOOR IN THE CAPITAL: THE CASE to sanitation, 55 percent in access to electricity, 53 percent OF BALBALA in multidimensional poverty, and 46 percent completion The community of Balbala, comprising the fourth and of primary school. Location is the biggest determinant for fifth districts of Djibouti city, has experienced important lack of improved water access, and it explains between transformations in recent years, including a major influx a third and a fourth of overall inequality in children’s of people from both inside and outside the country. This opportunities. demographic change can partly explain the proliferation of slums and, as described from anecdotal evidence, an increase in the demand for a wide range of services. An 20 Broadly speaking, the D-index denotes the percentage of goods and/or services that would have to be redistributed so that all circumstance groups would attain the same coverage. CHAPTER 1 38 analysis of geospatial data via a building footprint–drawing With poverty rates between 15.8 percent (fifth district) exercise suggests that the population in these districts is and 18.3 percent (fourth district), the Balbala community denser than in other districts of Djibouti city. In addition, as is host to about 37 percent of the poor population of the shown in the analysis so far, there are marked disparities country. Thus, it is important to better understand the within the capital: the fourth and fifth districts show the characteristics of the poor who reside in this area and lowest levels of consumption per capita, the highest to understand how they fare compared to the nonpoor poverty rates, and the highest poverty gaps (that is, the residents in the neighborhood and in Djibouti city. We extreme poor are on average much further away from the also contrast these characteristics with those of the poor poverty line than poor in other districts). and nonpoor population in rural areas to compare the opportunities in these two areas separately. Populations Other nonmonetary indicators provide further evidence living in rural areas have lower consumption per capita of welfare gaps in this urban area. Figure 1.23 shows on average and lack many amenities; there is a strong that households in Balbala are relatively more deprived correlation between welfare and the prominence of rural than those in other areas of the city. Household groups in population in the interior regions of the country. Since this commune are more likely to reside in a dwelling with the urban environment is typically more affluent than rural dirt floors. They are less likely to own such assets as a areas, regions with a high proportion of the population refrigerator. Households in this area could also be more living in rural areas have been shown to have weaker at risk to shocks. As evidenced by the consequences monetary and nonmonetary indicators when compared to of tropical storm Sagar in May 2018, households in more urbanized regions. the capital are at risk of suffering from floods. Among households that self-report residing in an area that is prone Table 1.17 presents a series of characteristics of the poor to flooding, those in Balbala have a high risk of flooding and nonpoor populations, differentiating among those living occurring inside their home. in Balbala, those in Djibouti city outside Balbala, and those FIGURE 1.23 Household Characteristics by District in Djibouti City a. Percentage of households b. Percentage reporting risk of c. Percentage who own a with dirt floors flooding inside home refrigerator 74 61 55 57 55 54 40 38 33 33 30 20 18 10 4 1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th district district district district district district district district district district district district district district district Source: Calculations based on EDAM4-IS. 39 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI living in rural areas. The poor in Balbala appear to share in rural areas. The deprivation score among poor urban characteristics with the poor elsewhere in the country. In households is also closer to the levels of the monetary poor in particular, the educational attainment and formality of jobs of rural areas than to the deprivation score of their neighbors. the household heads of the extreme poor population living in Nevertheless, the poor in Balbala appear to be more Balbala is similar to those residing in rural areas. The average attached to the labor market and have higher access to consumption per capita among the poor in Balbala is much certain services than the poor in rural areas. The percentage lower than the nonpoor in Balbala and also lower than that of the heads who are employed among the poor population TABLE 1.17 Characteristics of Poor and Nonpoor, by Area of Residence Nonpoor population Extreme poor population Districts Districts Balbala Rural Balbala Rural 1-2-3 1-2-3 Dependency ratio* 0.71 0.87 1.08 0.75 0.82 1.35 Percentage of those age15 or more in labor 48% 46% 41% 49% 41% 33% force Percentage of individuals employed** 29% 25% 22% 13% 12% 13% Head has no education 44% 61% 81% 87% 87% 94% Head has at least secondary education 19% 15% 3% 4% 2% 1% Head is employed 57% 61% 41% 31% 42% 26% Head is employed in public sector (as a % of all 43% 48% 45% 32% 46% 27% employed) Head is employed in private formal sector (as a 19% 10% 5% 2% 3% 4% % of all employed) Access to water 99% 99% 66% 92% 97% 41% Access to sanitation 48% 54% 23% 28% 32% 10% Access to electricity 90% 68% 10% 59% 18% 3% Household lives less than one kilometer away 39% 41% 41% 41% 47% 37% from an elementary school Children ages 6–14 enrolled in school 84% 87% 69% 70% 79% 53% Household lives less than one kilometer away 26% 25% 15% 28% 29% 10% from any health facility Average consumption per capita (DF) 323,598 212,859 151,326 69,392 67,669 49,051 Deprivation score 0.13 0.16 0.44 0.47 0.50 0.83 Source: Calculations based on EDAM4-IS. Note: Access to electricity is defined as the use of electricity as the main source of lighting. Access to water is defined as the availability of water through an ONEAD indoor connection, direct connection from a borehole, ONEAD outdoor connection by pipe, public fountain and drilling (with a pump). Access to sanitation is defined as unshared access to a water closet with flush, or latrine with slab. Private sector comprises all branches of activity except public administration and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. * Only households with at least one working age (15–64) individual are included. ** Includes all individuals 15 years and older. CHAPTER 1 40 in Balbala is 42 percent, while only a quarter of poor rural one kilometer is nearly similar for the poor and nonpoor in household heads are employed. By virtue of being in the the capital and rural areas. About 47 percent of the poor in capital, the poor in Balbala perform better in terms of Balbala live within one kilometer of the school. The urban access to services than the rural poor: poor households in poor children ages 6–14 years have relatively high rates of the capital have better coverage rates of water, sanitation, school attendance, while only 53 percent of the rural poor and electricity and more families live close to an elementary children ages of 6–14 are likely to go to school. school and health facility. While access to water does not Other data sources can help better capture the (simply seem like a concern among the Balbala poor, access to defined) access to certain services by the population of electricity and sanitation are much lower than among the the capital city. Evidence from geospatial data shows nonpoor. Expansion of electricity access in this area should the catchment areas of primary schools and hospitals in remain a primary concern. Access to primary schools within MAP 1.1 Catchment Areas of Primary Schools in Djibouti City (One Kilometer) Note: Data on primary schools is obtained from the website of the Ministère de l’Education Nationale et de la Formation Professionnelle (accessed December 27, 2018). Data for dwellings (marked in blue) is obtained from the building footprint of Djibouti produced by the World Bank. 41 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Djibouti city in the radius of one kilometer (Map 1.1 and the capital. Using an alternate catchment area of 1.5 Map 1.2). The majority of the city dwellings are covered kilometers would increase the coverage of public primary by at least one primary school (80.6 percent) and hospital schools and hospitals to 89.7 percent and 86.9 percent, (71.3 percent). However, the estimated density (population respectively. These elements point to the need to boost served) of these facilities is expected to be high in Djibouti coverage in certain areas of the capital that host poor city and may thus call for attention in examining whether and vulnerable populations. It would also be important these areas are underserved and consequently if the to consider improvements in service delivery (if the areas quality of service delivery in these areas is affected. For are found to be underserved) as well as strategies to schools, this is less likely to be a problem as a result of encourage parents to have children stay in school longer. the introduction of double classes in public schools in MAP 1.2 Catchment Areas of Hospitals in Djibouti City (One Kilometer) Note: Data on hospitals is obtained from OpenStreetMap, https://www.openstreetmap.org. Military hospitals and doctor’s clinics have been left out. Data for dwellings (marked in blue) are obtained from the building footprint of Djibouti produced by the World Bank. CHAPTER 1 42 Rural and urban poor face a distinct set of concerns, and Each has an area of about 7,000 square kilometers. Using a separate and targeted set of policies must be thought geospatial data, we conduct the following exercise. First, out for each of these subgroups of population. The we identify all the buildings via satellite imagery. Second, urban poor have much lower consumption than others using these building as proxies for dwellings or inhabited in their community and have much worse education and spaces, we create a one-kilometer radius around each employment outcomes. In urban areas, especially in dwelling. Third, we create clusters of dwellings to represent Balbala, the quality of service delivery may be poor due inhabited areas, like communities or villages. To build these to the high density of population served in these areas. clusters, we draw a one-kilometer radius around each Meanwhile, rural areas require a comprehensive strategy dwelling and then all the radii where their areas overlap. that invests in the provision of access to infrastructure For instance, if two dwelling are near each other, they and amenities. Households in these areas suffer from will form a cluster shaped like a peanut. And fourth, we low access to services such as electricity, water, and calculate the dwelling density (number of dwellings per sanitation. Electricity access is especially a concern to be square kilometer) for all the clusters identified. addressed, as only 6 percent of the rural population has Map 1.3 shows the results of this exercise for the regions. access. Finally, open defecation rates are high in regions In Tadjourah, the region with the highest poverty and and coverage of health facilities is low. lowest coverage rates of certain services, about 60 There is one more aspect of location that is important to percent of dwellings are in extremely low-density areas highlight in the Djiboutian context. Location plays a major (one building per square kilometer) and almost three role as it poses challenges to the implementation of public quarters (73 percent) are in very low-density areas (two investments such as building of schools and clinics or the buildings or less per square kilometer). Meanwhile, in distribution of transfers to vulnerable populations. The Dikhil, where extreme poverty is 53 percent, these shares lack of access to several services is evident in rural areas are 77 and 86 percent, respectively. As a comparison, of Djibouti, but a realistic government strategy to expand Djibouti city has all dwellings (80,818) under one cluster coverage of basic services will need to find innovative with an estimated density of 293.3 buildings per square ways to reach the populations in the poorest regions: most kilometer. Achieving universal coverage of basic services of the population residing in the regions live far apart from in rural areas will certainly require additional resources, but each other. We illustrate this by looking at Tadjourah and they will have to be coupled with a well-crafted strategy to Dikhil, which have the highest poverty rates in the country. reach the neediest in an efficient manner. 43 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI MAP 1.3 Clusters and Density of Clusters for Regions Outside Djibouti City Source: Calculations made using geospatial data created by the World Bank. The underlying map is taken from OpenStreetMap. Clusters refer to groups of one-kilometer-radius neighborhoods around all buildings identified through a building footprint exercise. Djibouti city cluster not shown. CHAPTER 1 44 REFERENCES Ahmed, N., M. Brzozowski, and T. F. Crossley. 2006. Deaton, A., and S. Zaidi. 2002. Guidelines for Constructing “Measurement Errors in Recall Food Consumption Data.” Consumption Aggregates for Welfare Analysis. Vol. 135. IFS Working Papers 06/21. Institute for Fiscal Studies, Washington, DC: World Bank. London. Dinkelman, T. 2011. “The Effects of Rural Electrification on Anker, R. 2011. “Engel’s Law around the World 150 Years Employment: New Evidence from South Africa.” American Later.” Political Economy Research Institute Working Paper Economic Review 101 (7): 3078–108. 247. Amherst, MA. DISED (Direction de la Statistiques et des Etudes Backiny-Yetna, P., D. Steele, and I. Y. Djima. 2014. The Démographiques). 2018. Résultats de la Quatrième impact of Household Food Consumption Data Collection Enquête Djiboutienne Auprès de Ménages pour les Methods on Poverty and Inequality Measures in Niger. Indicateurs Sociaux (EDAM4-IS). Djibouti city: DISED. Washington, DC: World Bank. http://www.dised.dj/Rapport1_resultats_EDAM4.pdf. Borgen Project. 2017. “Ethiopia-Djibouti Water Ferreira, F. H., J. R. Molinas Vega, R. Paes de Barros, Pipeline.” Borgen blog. Seattle, WA. December. https:// and J. Saavedra Chanduvi. 2008. Measuring inequality borgenproject.org/tag/ethiopia-djibouti-water-pipeline/. of Opportunities in Latin America and the Caribbean. Washington, DC: World Bank. Brueckner, M., and D. Lederman. 2015. Effects of Income Inequality on Aggregate Output. Washington, DC: World Ferreira, Francisco, and Carolina Sánchez-Páramo. 2017. Bank. “A Richer Array of International Poverty Lines.” Let’s Talk Development, World Bank blog. Washington, DC. Chakravorty, U., M. Pelli, and B. Ural Marchand. 2014. http://blogs.worldbank.org/developmenttalk/richer-array- “Does the Quality of Electricity Matter? Evidence international-poverty-lines. from Rural India.” Journal of Economic Behavior and Organization, 107, pp.228-247. IMF (International Monetary Fund). 2017. “Djibouti 2016 Article IV Consultation—Press Release; Staff Report; and Chandrasekhar, S., and S. Amin. 2012. “Looking beyond Statement by the Executive Director for Djibouti.” IMF Universal Primary Education: Gender Differences in Country Report no. 17/87. Washington, DC: IMF. Time Use among Children in Rural Bangladesh.” Asian population studies, 8(1), 23-38. Jacobs, Andrew, and Jane Perlez. 2017. “U.S. Wary of Its New Neighbor in Djibouti: A Chinese Naval Base.” Chowdhury, S. K. 2010. “Impact of Infrastructures on New York Times. February 25. https://www.nytimes. Paid Work Opportunities and Unpaid Work Burdens on com/2017/02/25/world/africa/us-djibouti-chinese-naval- Rural Women in Bangladesh.” Journal of International base.html. Development 22 (7): 997–1017. Krishnan, N., G. Lara Ibarra, A, Narayan, S. Tiwari, and Cramer, C. 2003. “Does Inequality Cause Conflict?” T. Vishwanath. 2016. Uneven Odds, Unequal Outcomes: Journal of International Development 15 (4): 397–412. Inequality of Opportunity in the Middle East and North Dang, H. A. H., and E. Ianchovichina. 2018. “Welfare Africa. Washington, DC: World Bank. Dynamics with Synthetic Panels: The Case of the Arab World in Transition.” Review of Income and Wealth 64: S114–S144. 45 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Maasho, Aaron. 2011. “Ethiopia Signs Djibouti Railway Washington Group on Disability Statistics. n.d. http:// Deal with China.” Reuters. December 17. https://www. www.washingtongroup-disability.com/washington-group- reuters.com/article/ethiopia-railway/ethiopia-signs-djibouti- question-sets/ railway-deal-with-china-idUSL6E7NH07M20111217?feedT WDI (World Development Indicators). n.d. Database. ype=RSS&feedName=industrialsSector Washington, DC: World Bank. https://datacatalog. Marrero, G. A., and L. Serven. 2018. “Growth, Inequality, worldbank.org/dataset/world-development-indicators. and Poverty: A Robust Relationship?” The World Bank. World Bank. n.d. “Tables, Charts, and Technical Notes.” Meseret, Elias. 2016. “Ethiopia’s New Coastal Rail Link Global Consumption Database. http://datatopics. Runs through Restive Region.” Associated Press. October worldbank.org/consumption/detail. 5. https://www.voanews.com/a/ethiopia-new-coastal- World Bank. 2008. The Welfare Impact of Rural rail/3537554.html Electrification: A Reassessment of the Costs and Benefits; Narayan, Ambar, Roy Van der Weide, Alexandru Cojocaru, An IEG Impact Evaluation. Washington, DC: World Bank. Christoph Lakner, Silvia Redaelli, Daniel Gerszon World Bank. 2018a. Poverty and Shared Prosperity 2018: Mahler, Rakesh Gupta N. Ramasubbaiah, and Stefan Piecing Together the Poverty Puzzle. Washington, DC: Thewissen. 2018. Fair Progress?: Economic Mobility World Bank. Across Generations Around the World. Washington, DC: World Bank. https://openknowledge.worldbank.org/ handle/10986/28428. Olinto, P., G. Lara Ibarra, and J. Saavedra-Chanduvi. 2014. Accelerating Poverty Reduction in a Less Poor World: The roles of Growth and Inequality. Washington, DC: World Bank. Paes de Barros, Ricardo, Francisco H. G. Ferreira, Jose R. Molinas Vega, and Jaime Saavedra Chanduvi. 2009. Measuring Inequality of Opportunities in Latin America and the Caribbean. Washington, DC: World Bank. https:// openknowledge.worldbank.org/handle/10986/2580. Ravallion, M. 1998. “Poverty Lines in Theory and Practice.” LSMS Working Paper 133. World Bank: Washington DC. UNHCR (United Nations High Commissioner for Refugees). https://www.unhcr.org/djibouti.html Van de Walle, D., M. Ravallion, V. Mendiratta, and G. Koolwal. 2017. “Long-term Gains from Electrification in Rural India.” World Bank Economic Review 31 (2): 385–411. CHAPTER 1 46 HUMAN CAPITAL AND CHAPTER 2 LABOR MARKET A key objective of a poverty reduction strategy in Djibouti the biggest contributor to changes in poverty. Thus, it is should be to unleash the potential of its (current and future) expected that a sustainable approach to generate income labor force, by providing people with skills to obtain or and eradicate poverty in the country is to promote a well- generate good jobs that allow them to grow economically. educated labor force that is competitive in the global arena Understanding the labor market opportunities of the and that this workforce can grow in a system welcoming Djiboutian workforce requires looking into the population’s private sector investments. education, as well as their market outcomes. The framework There is suggestive evidence that the government of developed in Lopez-Calva and Rodríguez-Castelán (2016) Djibouti (GoD) has made important strides in investing in argues that the capacity of an individual (or the household its population and boosting its educational attainment. she belongs to) to generate income is based on the assets Nonetheless, the concentration of economic growth in (or all kinds of capital) and opportunities that she owns, and capital-intensive sectors has missed the opportunity to her ability to use them to obtain returns. Of key interest, unleash the potential of the improvements in educational then, is how human capital (measured, for instance, by attainment of the Djiboutian labor force. In this chapter, educational attainment) is accumulated in the population we provide analytics of a snapshot of the labor market in and how the population is able to reap benefits from it. In Djibouti. We start with a brief look at the human capital addition, earnings from labor is integral to economic growth of the labor force, followed by an analysis of the labor and to escaping poverty. As shown in Azevedo et al. (2013) outcomes. Finally, we present some findings on the types for a broad set of countries, Barros et al. (2006) in Brazil, of firms existing in Djibouti to shed some light on the labor and several other studies in the literature, labor income is demand side. A brief discussion of the policies concludes. often associated with more and better paying jobs and is HUMAN CAPITAL: EDUCATIONAL OUTCOMES IN DJIBOUTI Across the country’s regions, educational outcomes in disparities in literacy rates are also evident—59 percent Djibouti follow a pattern similar to that found in welfare of the population in the capital city is literate as opposed and other nonmonetary indicators. Using self-declared to 29 percent in the regions, with wide disparities across information from EDAM4-IS, an estimate of literacy rates, regions. The literacy rates are lowest in the region of or the ability to read and write among individuals age 15 Tadjourah, where only about a fifth of the population can years and older, is calculated. In 2017, a bit more than read and write, followed closely by Obock and Dikhil. half (53 percent) of the population over 15 years of age of Similar regional trends are also observed for men and Djibouti could read and write (Figure 2.1). Literacy rates women, with women having lower literacy rates than men vary by gender, with nearly three-fifths of all men being in every region. literate as compared to 43 percent of women. Geographic 47 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Looking at another indicator of education, defined as on education. This might also point to education being a the percentage of the population age 15 years and older normal good in Djibouti, wherein spending on the good that has completed at least primary schooling (Figure increases as income increases. Gender disparities are also 2.2), we find that about 40 percent of the population has present in this indicator, while following the same pattern completed primary schooling. This number stands at 44 of regional disparities. About a third of Djiboutian women percent in the capital city and is lowest in the regions of have finished primary education compared to almost Tadjourah and Obock, at about 15 percent. Differences half of the men. This gap is present in every region of the in primary schooling completion across regions seen country, with Tadjourah and Obock showing the smallest here may be explained by these differences in spending differences (although probably due to the low overall rate). FIGURE 2.1 Literacy Rates among People Age 15 Years and Older, 2017 (percent) 69 63 59 53 54 49 42 43 34 36 30 32 29 27 26 25 21 23 19 19 16 All Men Women All Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta Source: Calculations based on EDAM4-IS. FIGURE 2.2 Percentage of Population Age 15 Years and Older with at Least Primary Education, 2017 53 48 44 39 37 35 29 30 31 24 26 23 20 19 20 15 17 15 15 12 11 All Men Women All Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta Source: Calculations based on EDAM4-IS. CHAPTER 2 48 INTERGENERATIONAL TRANSFORMATION IN DJIBOUTI Behind the results of low aggregate literacy rates, there is Some of these efforts appear to be bearing fruit as some evidence of an educational transformation underway the younger cohorts perform much better on a host of in Djibouti. For instance, in 2010, the GoD spent 12.3 educational outcomes and women seem to be catching percent of its total expenditure (equivalent to 4.49 percent up to men, but certain groups of the population are being of GDP) in education. While more recent estimates are 21 left behind. Comparing literacy rates across age cohorts, not available, other indicators suggest that the GoD has findings reveal that the younger cohorts of ages 10–14 made significant efforts to increase access to schooling years and 15–24 years have remarkably higher literacy in recent years. The number of public primary schools rates as compared to older cohorts of 25 years and older increased from 84 in 2004–5 to 136 in 2016–17 (DISED (Figure 2.3). The cohort of 15–24 years has a literacy 2012a, 2017a). Further, the number of public middle rate of 80 percent, significantly higher than the national and secondary schools also increased threefold, from average of 53 percent but still lower than the average 11 in 2004–05 to 36 in 2016–17 (DISED 2012a, 2017a). of the MENA region of 91 percent for the same cohort As higher schooling is likely to lead to higher earnings (The World Bank, 2018c). Another interesting finding is (Ashenfelter and Krueger 1994; Card and Krueger 1992; that the gap in literacy rates among men and women is Duflo 2001; Heckman et al. 2006; Psacharopoulos and closing for younger cohorts. Among the population aged Patrinos 2018), this increased availability of access to 25–39 years and older, there is a 26 percentage point schooling holds vast potential for poverty reduction in the difference in literacy rate between men and women, while medium and long terms. this gap is only 10 percentage points and not quite 2 percentage points for those age 15–24 years and 6–14 years old, respectively. FIGURE 2.3 Literacy Rates, 2017 (percent) 85 86 85 84 80 75 68 53 42 41 30 28 19 18 10 All Men Women 10–14 years 15–24 years 25–39 years 40–60 years 61 years and above Source: Calculations based on EDAM4-IS. 21 These expenditures represented a decrease from previous years. In 2010, the GoD spent 22.5 percent of its total expenditure (or 8.4 percent of GDP) in education See WDI, https://data.worldbank.org/indicator/SE.XPD.TOTL.GD.ZS?locations=DJ, accessed December 27, 2018. 49 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE 2.4 Educational Attainment by Age Cohort, 2017 8 3 18 13 11 23 25 22 18 2 25 3 29 25 34 4 4 49 5 4 84 72 58 49 6 48 39 21 All Males Females 15–24 years 25–39 years 40–60 years 61 years and above No education/ own education Less than primary Primary and above, but less than secondary Secondary and above Source: Calculations based on EDAM4-IS. Note: Only individuals older than the age of 15 years are included Figure 2.4 shows similar results on educational attainment. out of three people ages 11–39 years has never been About 49 percent of those age 15–24 years have to school. The results are worse for women (39 percent) completed at least primary education, and this percentage and the population outside Djibouti city (52 percent). As deteriorates dramatically with age. Among those age 40 per EDAM4, the main reason cited by respondents for years and older, three-fourths have no education. Another never going to school is found to be lack of interest, while interesting finding is that the percentage of population the fact of being a woman is cited as the second most with less than primary education is relatively small and is important reason. not changing much across age cohorts. Thus, it seems The concept of economic mobility rests on the idea that that the transition of education runs from no education access to opportunity depends little on one’s starting to primary education or more. Women are more likely to point and that an individual has the opportunity to improve have no education whatsoever and only 38 percent have his or her economic status over time. In developing completed at least primary education, as opposed to 57 countries, a large body of research has focused on the percent of the men. intergenerational transmission of poverty and mobility Another noteworthy finding is that nearly 81 percent of the out of it. For example, Gong et al. (2012) show that the children ages 6–14 were enrolled in school in 2017. In the relative placement of children in the income distribution is regions outside Djibouti city, this percentage is 71 percent. a result of their parent’s income in urban China. Further, There is, thus, an indication that the Djiboutian population they establish that link between the two runs through considers education to be important for its children, and the channel of education and occupation, among this may have consequences for Djibouti’s human capital others. Many emerging economies such as South Africa, stock in the future. Among children ages 6–14 going to Brazil, Chile, and China appear to have low levels of school, 96 percent have been or are going to a French intergenerational mobility due to high inequality that public school. On the other hand, approximately one persists in these countries (Gong, Leigh, and Meng, 2012; CHAPTER 2 50 Nunez and Miranda 2010; Piraino 2015). In addition, it no education, and about 10 percent of poor women can be contended that education can play a pivotal role in attain some education when their father has none. Poor lifting people out of poverty, and consequently, breaking its men follow the lead with a corresponding number of 78 intergenerational transmission. While we cannot establish percent. The most upwardly mobile group comprises a link between parents’ and children’s income due to the nonpoor men—53 percent of them have educational unavailability of such disaggregated data, we can explore attainment higher than their father’s. To better understand the intergenerational mobility of education. the gaps among these groups, international benchmarks can be obtained from a recent study (Narayan et al. 2018) Based on the results noted, there is evidence that an on intergenerational educational mobility that provided intergenerational movement to higher education levels estimates of upward mobility for close to 150 economies.22 in Djibouti has taken place. Indeed, the percentage of Upward mobility among Djiboutian nonpoor men is around individuals (age 25 years and older) with higher educational the median of the distribution with such countries as China, qualifications than their father is 36 percent in Djibouti Egypt, and Hungary. Poor women’s mobility is close to the (Figure 2.5). A third of individuals over the age of 25 years bottom and similar to those in Mali and South Sudan. whose father has no education have attained some form of education (Table 2.1). However, certain disparities remain Some of these results point to Djibouti becoming well among men and women. The poor women are found to placed to reap the benefits of high economic growth it has be the most immobile group; 87 percent of them have been experiencing over the past decade. A key challenge no improvements in educational outcomes with respect will be to ensure that all groups of the population are to their father’s. Almost all fathers in this subgroup have capable of participating in a more developed economy FIGURE 2.5 Educational Mobility of Individuals 25 Years and Older Compared to Their Fathers’ (percent) 44 60 65 78 4 87 4 6 53 36 3 29 19 3 10 All Nonpoor men Nonpoor women Poor men Poor women Upward mobility Downward mobility No change Source: Calculations based on EDAM4-IS. Notes: We use educational attainment as divided into four categories: no education, primary education or less, between primary edu- cation and secondary education, and secondary education or more. If the educational attainment of the child is higher than that of the father, he or she is considered upwardly mobile. Data on education of self or father were missing for 7 percent of all individuals. 22 To obtain a comparable estimate within the estimates from Narayan et al. (2018), we use the Global Database on Intergenerational Mobility. In particular, we obtain the average estimated absolute upward mobility (across all available cohorts) for fathers and all children for each country. 51 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 2.1 Education Transition Matrices of Individuals 25 Years and Older Based on Their Father’s Education Father’s education More than No Primary primary or More than All education or less secondary secondary Total No education 57 3 0 0 61 Primary or less 12 1 0 0 14 More than primary or secondary 14 2 1 0 17 More than secondary 6 1 1 1 8 Total 88 8 2 1 100 Father’s education More than No Primary or primary or More than Poor women education less secondary secondary Total No education 87 2 0 0 89 Primary or less 7 0 0 0 7 More than primary or secondary 3 0 0 0 3 More than secondary 0 0 0 0 0 Total 97 3 0 0 100 Source: Calculations based on EDAM4-IS. Data on education of self or father were missing for 7 percent of all individuals. and reaping returns of increased educational attainment International Student Assessment or Trends in International in the labor market. Certain groups of the population Mathematics and Science Study, have not been appear to already be ahead of the curve and surpassing conducted in Djibouti. Second, there is mixed evidence the educational levels of their parents. However, stronger of improvements in other human capital indicators, such efforts will be needed for growth to be truly inclusive and all as health. Regarding this more long-term view of human groups to prosper along with the country’s growth. capital investments, it is harder to evaluate the implemented policies in recent years. For instance, projections for infant Finally, it is important to note that this section on human mortality show a continuous decline, reaching 51.5 in capital takes a narrow view on educational attainment. 2017 from 80.3 in 2000. However, indicators on stunting There are two main reasons. First, there are no data on rates appear to be unchanged and wasting rates appear the extent of actual learning obtained via the educational to have increased between 1996 and 2012 (World Bank system in Djibouti. Learning, as opposed to obtaining a 2018a). There are no health indicators available with more certificate, could be a better measure of the skills obtained recent surveys.23 With these in mind, it must be clear that by an individual and her potential to be a productive part the gains in education may not fully translate into gains in of the economy. Typical proxies for quality of learning, productivity among the Djiboutian labor force. including such international tests as the Programme for 23 Information on the government’s expenditure on health dates from 2015. According to WDI (accessed February 13, 2019), the domestic general government health expenditure represented 2.4 percent of GDP in 2015. CHAPTER 2 52 LABOR MARKET OUTCOMES The high rates of economic growth that Djibouti has about 45 percent participate in the labor market—a ratio experienced in recent years could have spillover effects in practically unchanged from the earlier estimate of 46.3 the form of job creation and hence contribute to poverty percent in 1996 (World Bank 1997), despite the positive reduction. Economic well-being is intrinsically tied to economic growth during the past 20 years. Similar to employment, and this is more so for the poor. In chapter 1, the patterns found in other MENA countries, labor force a strong correlation was found between poverty status and participation in Djibouti shows substantial variation across labor force outcomes. In this section, it is shown that labor gender and age. Men’s labor force participation stands force outcomes are also highly correlated with location, at 59 percent, whereas for women it is only 32 percent. gender, and age. We provide a snapshot of labor force Less than a third of individuals in the 15–24 age range outcomes in Djibouti in 2017, followed by an exploration of participate in the labor market. Among people in prime age the links between education and employment outcomes. working years (25–39 and 40–60), the participation rate is For education to break the intergenerational transmission 55 and 53 percent, respectively. of poverty, the underlying assumption is that improved Figure 2.6 shows the distribution of the population of working education outcomes must translate into better employment age (15 years and older) across gender and age cohorts. outcomes. We also give a closer look at those employed, It appears that women and the youngest cohort of 15–24 the sectors of employment, and wages in these sectors. years are less likely to be active in the labor force. Among Labor force participation rates in Djibouti are low (Table those who are active in these two groups, a large percentage 2.2). Overall, among the population 15 years and older, remain unemployed. For example, 13 percent of those TABLE 2.2 Characteristics of the Population by Poverty Status and Location Gender Age group (in years) Overall Male Female 15–24 25–39 40–60 61+ Labor force participation 44.7% 58.5% 31.5% 28.6% 55.3% 53.4% 15.8% Source: Calculations based on EDAM4-IS. Notes: Only the population 15 years and older is included. FIGURE 2.6 Distribution of Population in the Labor Market (percent) 26 24 47 38 47 63 86 43 32 51 30 28 25 23 30 7 25 30 25 11 6 All Men Women 15–24 years 25–39 years 40–60 years 61 years and older Employed in public sector Employed elsewhere Unemployed Source: Calculations based on EDAM4-IS. Notes: Only the population 15 years and older is included. Public sector includes those employed declaring public administration as the main branch of activity. 53 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI active among the youngest cohort are employed. About 62 have much higher level of employment rate as compared percent of all active working age men are employed, and 30 to women. Seniority and years of experience also seem percent are employed in the public sector. What is striking to increase the probability of being employed. About 74 is that even though the population ages 15–24 years have percent in the age range of 40–60 years are employed, higher educational attainment than the older cohorts, a large while this number stands at 53 percent for individuals percentage of those active among them strive to find a job. ages 25–39 years. It seems that the high education levels among 15–24 years old have not translated to Based on educational attainment, employment rates are employment opportunities yet. We study the population highest for individuals with no education and those with ages 15–24 year in a bit more detail to explore this finding. secondary education or more (Figure 2.7). This pattern is observed for both men and women, although the former FIGURE 2.7 Employment Rate among Labor Market Participants, by Educational Attainment 68% 70% 65% 62% 55% 53% 54% 52% 49% 41% 42% 39% 37% 26% 21% All Men Women 92% 91% 84% 80% 74% 76% 76% 71% 73% 69% 53% 49% 47% 44% 36% 20% 21% 14% 10% 8% 15–24 years 25–39 years 40–60 years 61 years and older No education/own education Less than primary Primary and above, but less than secondary Secondary and above Average Source: Calculations based on EDAM4-IS. Notes: Employment rate is defined as those employed among those active in the labor force. Both employed and unemployed individu- als are considered active in the labor force. CHAPTER 2 54 With the lowest employment shares, it is worth exploring 45 percent of those ages 15–24 years are studying, 22 how the youth (15–24 years old) in Djibouti occupy their percent are unemployed, 5 percent are employed, and time. We do this by distinguishing among four activities: 28 percent are NEET (Figure 2.8). Meanwhile, a third of studying, employed or in training, unemployed (within young women are NEET. Youth in the regions have worse those seeking a job), and neither studying nor employed, outcomes: about 43 percent are NEET and thus are nor in training, nor unemployed (NEET). Nationally, about disengaged from any productive activity. FIGURE 2.8 Distribution of Youth Ages 15–24 Years across Education and Labor Force Activities 0 0 28 33 Employed/ training and studying 45 43 Studying Employed/ in training 22 Unemployed NEET 20 5 3 Source: Calculations based on EDAM4-IS. Note: NEET refers to youth not in employment, education or training neither unemployed (those individuals who are seeking a job). One of the reasons for not seeking participation in the are out of the labor force in 2017 and more of them also labor market could be that more women are in school for seem to report family obligations as the key reason. Thus, a longer period of time (as noted in the findings related to not only are women less likely to participate in the labor education mobility). Table 2.3 present the reasons for not force in 2017 than men, this phenomenon is becoming wanting to find employment for men and women in the worse over time for them but not for men. Better education age group of 15–24 years. Indeed, studying is cited as one outcomes for young women do not seem to be translating of the key reasons, however less so by men in 2017 as to higher labor market attachment rates for them. compared to 2012. As compared to 2012, more women TABLE 2.3 Reasons for Not Wanting to Find Employment, among the Youth 15–24 Years Old (percent) 2017 2012 Males Females Males Females In the labor force 32 26 33 33 Out of the labor force: Source: Calculations based on Studying 44 40 53 42 EDAM 2012 and EDAM4-IS. Family obligations 7 19 4 13 Note: Students seeking employ- Others 11 11 9 9 ment are counted as part of the labor force. Trainees without Missing 6 5 1 4 other activity are counted out of the labor force. Shares may Total 100 100 100 100 differ slightly from Figure 2.8. 55 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI For the 53 percent of the population employed nationally, contractors and independent workers are considered a dual market is the norm, with the public sector being a working in the formal sector when they have accounting major employer, on one side, and informal jobs or working records, a trading license (commonly referred to as in informal firms, on the other. Table 2.4 shows that public “patente”) and registration in the chamber of commerce. administration is the branch of activity for close to half Within the private sector, salaried individuals, family helpers of the employed population (44 percent). Services follow and apprentices are considered working in the formal with nearly 27 percent of those employed. We divide the sector if the worker has a contract and is registered in sector of employment into public administration (when social security. All remaining individuals working in the the main branch of activity is public administration) and private sector are considered as working in the informal private (when the main branch of activity is anything except sector. The findings indicate that most of the private sector the public administration, with exceptions). Within the jobs are informal, with 46 percent of those employed private sector, we distinguish between formal and informal engaging in them. Among the women who are employed, private sector jobs. Within the private sector, employers, this number is even higher at 63 percent. TABLE 2.4 Distribution of Employed Individuals by Branch of Activity and Sector of Employment (percent) Private Public informal Private formal Administration Total Overall population Agriculture 1.3 0.2 0.0 1.5 Manufacturing 4.1 0.9 0.0 5.0 Services 24.0 3.4 0.0 27.4 Public administration 0.0 0.0 43.8 43.8 Other 16.9 5.5 0.0 22.4 Total 46.3 9.9 43.8 100.0 Women’s population Agriculture 0.9 0.2 0.0 1.1 Manufacturing 0.5 0.3 0.0 0.8 Services 48.8 1.8 0.0 50.5 Public administration 0.0 0.0 31.1 31.1 Other 13.0 3.6 0.0 16.6 Total 63.1 5.8 31.1 100.0 Source: Calculations based on EDAM4-IS. Notes: Includes 3,139 individuals that answered to all corresponding questions. Private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. CHAPTER 2 56 Outside of the public sector, the services account for the Regarding the size of the firm that the employed individuals majority of the jobs available. In a country with tropical work, we find that firms in the public and formal private desert on the coast and in the north, and semidesert in sectors are large, many employing more than 20 the south-central highlands, the agricultural sector plays individuals (Table 2.5). On the other hand, firms in the a minor role in employing people. The services sector informal private sector are predominantly one-person takes the lion’s share of employment both formally and firms, pointing to largely self-employed individuals working informally. About a quarter of the employed population by themselves. The numbers for women are even higher, has an informal job in the services sector. In this group, with nearly 47 percent of those employed working by the most common occupation is in sales (beignets, fruits, themselves. Thus, most of the private sector in Djibouti is vegetables, cereals, and other fresh food), accounting for in the informal sector, while this sector is more dominant 29 percent of those employed. Half of employed women for women. are working informally in the services sector. Among them, 59 percent sell similar kinds of products. TABLE 2.5 Distribution of Employed Individuals by Firm Size and Sector of Employment (percent) Private Public informal Private formal administration Total Firm size Overall population 1 20.8 0.6 1.1 22.5 2–3 8.7 0.5 2.2 11.4 4–5 4.8 1.1 2.7 8.6 6–10 3.0 1.0 3.3 7.3 11–20 3.3 1.1 5.1 9.6 More than 20 6.6 6.1 27.9 40.6 Total 47.3 10.4 42.3 100 Firm size Women’s population 1 46.8 0.7 0.5 48.0 2–3 8.8 0.3 2.4 11.5 4–5 3.8 0.9 2.8 7.5 6–10 1.7 0.1 2.5 4.3 11–20 0.7 0.3 4.1 5.1 More than 20 3.6 3.6 16.4 23.5 Total 65.4 5.9 28.7 100.0 Source: Calculations based on EDAM4-IS. Notes: Sample includes 2,806 individuals with complete information. Private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. 57 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI The public sector has an enormous appeal to highly informal private sector, which employs the largest share qualified individuals, with important implications for the of those employed, seem to be the lowest. There is also a pool of private sector workers (Table 2.6). The distribution small segment of the population that works in the formal of educational attainment among public sector workers private sector, drawing high wages. To understand the is varied: 26 percent don’t have any formal education, return to being employed in the public sector as well as 25 percent have only primary, and close to half of the to education, we use a regression-based framework, public employees (45 percent) have at least a secondary accounting for sector of work, number of years of work education. Moreover, the public administration hires 7 out experience, and frequency of payment of wages (Table of every 10 Djiboutian workers with at least secondary 2.8). We find that each year of formal education is education. Thus, the pool of available talent in the private associated with an increase in wages by 7.8 percent. sector is constituted of a population in which individuals Controlling for public sector work provides similar results. It with low skills are overrepresented. also seems that working in the public sector is associated with higher wages, as it carries a premium of 18 percent, The strong pull from the public sector appears to be even after controlling for other worker’s characteristics. a result of its higher wages (Table 2.7). It seems that The result also holds when the premia is estimated using salaried individuals working in the public administration categorical variables for educational attainment. get paid more than in the private sector. Earnings in the TABLE 2.6 Distribution of Employed individuals by Educational Attainment and Sector of Employment (percent) Public Private informal Private formal administration Total No education/own education 30.8 2.5 11.5 44.8 Less than primary 2.5 0.4 1.3 4.2 Primary but less than secondary 9.1 3.2 11.0 23.3 Secondary and above 3.9 3.9 19.9 27.6 Total 46.3 9.9 43.8 100 Source: Calculations based on EDAM4-IS. Notes: Sample includes 3,132 individuals with complete information. Private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector.Cells show the percentage of the working population. CHAPTER 2 58 TABLE 2.7 Wages Earned in Private and Public Sector by Status of Employment Public Private informal Private formal administration Employer or self-employed Sample size 822 58 402 Mean monthly wages (DF) 89,829 232,452 101,171 Mean wage per hour (DF) 1,698 1,843 2,226 Tacheron (contractor) Sample size 186 - - Mean monthly wages (DF) 77,123 - - Mean wage per hour (DF) 2,646 - - Salaried Sample size 367 233 892 Mean monthly wages (DF) 78,170 98,632 106,187 Mean wage per hour (DF) 892 1,283 1,396 All Sample size 1492 300 1313 Mean monthly wages (DF) 84,221 118,442 104,161 Mean wage per hour (DF) 1,599 1,349 1,698 Source: Calculations based on EDAM4-IS. Note: The private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. A group of 117 workers were identified in an “Other” status of employment in the informal private sector only and are not shown due to the small sample size and lack of comparator groups. Djibouti’s high economic growth rate has not translated unappealing because of lower salaries than in the public into the creation of a dynamic formal private sector. The sector, low job security, and the absence of job protection most dominant sector of employment in Djibouti is the and social security. For the country as a whole, the informal private sector, employing 46 percent. This is informal sector is unlikely to drive development, as informal closely followed by the public sector, with 43 percent jobs are largely low-skill jobs with low productivity. The of those employed. Informal private jobs are more workers engaged in informal jobs, however, may represent concentrated among women and tend to be largely an untapped mass of potential entrepreneurs facing one-person firms. Informal jobs may be considered obstacles such as high government regulation. 59 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 2.8 Returns to Education Dependent variable: log of wages [1] [2] [3] [4] Years of education 0.078*** 0.073*** (0.001) (0.001) Works in public sector 0.180*** 0.184*** (0.006) (0.006) Work experience 0.018*** 0.018*** 0.019*** 0.019*** (0.001) (0.001) (0.001) (0.001) Less than primary education 0.158*** 0.152*** (0.014) (0.014) Primary but less than secondary education 0.386*** 0.354*** (0.008) (0.008) Secondary education and above 1.064*** 0.997*** (0.009) (0.009) Constant 13.008*** 13.025*** 13.086*** 13.098*** (0.015) (0.015) (0.015) (0.015) R2 0.214 0.225 0.214 0.225 Source: Calculations based on EDAM4-IS. Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Other variables in the regression include square of work experience and frequency of payment of wages. Public sector includes those employed declaring public administration as the main branch of activity. Private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). The literature also emphasizes the importance of a healthy Djibouti. Evidently, poverty rates are six times higher when formal sector. One of the key arguments is that formal the household head is employed in the informal private firms are the driver of economic development, while sector as compared to the formal private sector (Table informal firms are a by-product of poverty (Rauch 1991; 2.9). Individuals living in households where the head is La Porta and Schleifer 2014). In this literature, as an employed in the informal private sector also spend more economy modernizes, formal firms grow, and informality on health-related expenses, pointing to their vulnerability to becomes less prevalent. Countries with an abundance of health shocks. In addition, Table 2.10 displays the share of informal firms suffer from low aggregate productivity, too. each type of income in the total household income for the As countries develop, more workers transition from the population as a whole and the bottom 20 percent. About informal to the formal sector. Thus, efforts encouraging the 31 percent of the income for the bottom 20 percent comes growth of the formal sector in Djibouti may be important from formal and public sector work, while this share stands to stimulate job-driven growth. While the expansion of a at 51 percent for an average Djiboutian. This contrast formal sector is good for development, informality is found shows that the poor depend heavily on informal sector to be correlated to the poverty status of individuals in work, more so than an average Djiboutian. CHAPTER 2 60 TABLE 2.9 Characteristics by Sector of Employment of Household Heads Private Public informal Private formal administration Average monthly wages of head (DF) 87,081 121,576 101,077 Extreme poverty rate 18% 3% 11% Extreme poverty rate when poverty line is 5% higher 20% 6% 12% Extreme poverty rate when poverty line is 10% higher 22% 6% 13% Percentage of household experienced a health problem 28% 24% 27% Health expenses as a percentage of total: average 3% 2.5% 1.9% Health expenses as a percentage of total: 90th percentile 7.2% 6.7% 4.9% Source: Calculations based on EDAM4-IS. Notes: Wage information is available for 2,415 individuals and 1,531 heads. The rest of the sample of heads includes 1,996 individuals with complete information. Private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases, and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. Cells show the percentage of the working population. TABLE 2.10 Share of Various Income Sources in Total Income Overall Bottom 20% Share of informal private labor income 37% 40% Share of formal private labor income 8% 2.5% Share of public labor income 41% 21.5% Share of transfers 10.5% 20.1% Share of revenue from agricultural economic activity 3.8% 14.3% Source: Calculations based on EDAM4-IS. Notes: Only households that reported income and have complete information are included (2,426 and 532 households, respectively). Bottom 20 percent is calculated based on the full distribution of consumption per adult equivalent. Private sector comprises all branches of activity except public administration—agriculture, manufacturing, services (including private administration), and others (international organizations, military bases and unspecified categories). Within the private sector, employers, contractors and independent workers are considered working in the formal sector when they have accounting records, a trading license (commonly referred to as “patente”) and registration in the chamber of commerce. Within the private sector, salaried individuals, family helpers and apprentices are considered working in the formal sector if the worker has a contract and is registered in social security. All remaining individuals working in the private sector are considered as working in the informal sector. 61 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI There are also obstacles in the demand side of the and suppliers is also higher in Djibouti (71.6 percent) than labor market that prevent the continuous development other countries in the MENA region (50) and other lower of a dynamic formal private sector. Some insights can middle-income counties (64.8 percent). Meanwhile, the be directly obtained from the perceptions of the firms percentage of full-time female workers is very low: 26.6 conducting business in the country. According to the percent, even lower than the average of lower middle- enterprise survey conducted in 2013, high tax rates 24 income countries (32.4 percent). were found to be a hindrance for 15 percent of the small Access to credit did not seem to be an issue, as 30.5 and medium firms. The low educated workforce was percent of all firms have a bank loan or line of credit. The a constraint declared by 13 percent of medium-size rate is 23.6 percent among small firms, 35.9 percent firms and 28 percent of large firms. About 35 percent of among large firms, and highest among the medium firms the large firms also reported the paucity and quality of at 44.7 percent. The average rate is much higher than electricity as a constraint. Yet another barrier to running a that found in MENA: 19.8 percent (but slightly lower than business smoothly was corruption, which was declared in lower middle-income economies’ 34.7 percent). Not by 14 percent of small firms and 15 percent of large firms. surprisingly, 91.6 percent of firms have a checking and/ About 43 percent of firms believe they needed to give gifts or savings account, including 87.7 percent of small firms to secure a government contract, and this percentage was reporting so. higher among larger firms at 66 percent. Djibouti has made tremendous strides in the “doing Other findings from the survey include 50 percent reporting business” (DB) arena in the past few years. As per the electricity being the main problem. Firms were found to DB 2014 report (World Bank 2014), Djibouti was in be suffering from about 1.3 power outages in a typical 160th place of all countries evaluated. Several areas of month, leading to losses of about 2.8 percent of sales, improvement were identified: starting a business (a rank of with (slightly) higher losses being suffered by medium-size 127), getting electricity (144), registering property (133) and firms (20–99). The average number of days needed to get getting credit (180), among others. Of particular attention an electric connection was found to be 34.1 (small firms to starting a business were the costs associated with took 25 days; medium-size firms took about 57.6 days, startup (184.7 percent of income per capita between all and large firms took about 28 days). At the same time, the the permits needed) and those to get electricity: 180 days average number of days needed to get a water connection at a cost of 7,487 percent of income per capita. was 16.1. Finally, just less than half of firms (47.1 percent) believe that the court system in Djibouti is fair, impartial, In the DB 2019 report (World Bank 2019), Djibouti made and uncorrupted. significant strides, jumping 55 places and placing itself on 99th place (as opposed to 154th a year before). The Firm usage of Internet is relatively high (40.7 percent) with largest gains came from changes in protecting minority respect to the MENA region (36.2 percent) and other investors, registering property and getting electricity. Other lower middle-income countries (36.9 percent)—although improvements were seen in insolvency and starting a differences exist between small firms (1–19 workers) at business. Dealing with construction permits was the only 35 percent and large firms (100 or more workers) at 74.9 index in which a decrease was observed (ranking fell from percent. The use of email to communicate with clients 84 to 101). 24 Source: World Bank Enterprise Survey 2013. The sample comprised 266 formal nonagricultural private firms. Public utilities, government services, health care, and financial services sectors are not included in the sample.” CHAPTER 2 62 Regarding the indicator related to starting a business, higher than MENA (479.9) and OECD high-income (64.2) there were improvements in the number of procedures averages. Costs are accrued from paying the estimate and and time required to get the entire process done. It obtaining external works from Electricité de Djibouti, signing is notable, however, that the actual cost of opening a the supply contract, and obtaining the final connection. It is business increased when compared to DB 2018: it was notable that the reliability of supply and transparency score estimated at 35 percent of income per capita, whereas of 0 has not moved and is much lower than 4.2 in MENA the latest indicator is 41.9 percent. These costs reflect and 7.5 in OECD. Some improvements have been found typically spending DF 90,000 on a lawyer to draft the in the ability to conduct business over time in Djibouti, but company’s articles, DF 29,000 needed for registration many concerns persist and make it difficult for the formal fees, name registration, service fee of the Guichet unique private sector to boom, in line with the high economic (one stop shop), registration of the articles of association growth of the past few years. (including DF 1,000 per page for the stamp duty), as well Overall, even though the youth have higher educational as DF 7,000–10,000 needed to create the company seal, attainment and women are catching up to men, these letterhead, and books. It is true that this cost has come population groups are still the least attached to the labor down substantially: it was estimated to be 184 percent of market. The most dominant sectors of employment in income per capita in the DB 2014, but it is still higher than Djibouti are the informal private sector and public sector, the MENA average (22.6 percent) and that of Organisation which together employ 90 percent of the population. of Economic Co-ordination and Development (OECD) Salaries are higher in the public and formal sector as high-income countries (3.1 percent). compared to the informal private sector too. For the The price of electricity seems to have slightly decreased country as a whole, the informal sector is unlikely to drive in the past few years. There are four procedures required development, as the jobs are largely low-skill jobs with to get electricity now (lower than the MENA average of low productivity. The workers engaged in informal jobs 4.7), and there has been a substantial improvement in the may represent potential entrepreneurs who face obstacles number of days required to secure electricity: down from such as high government regulation. Thus, to develop 125 days to 52 (also lower than 72.4 MENA average). In the labor market and absorb the increasingly educated addition, there has been an improvement in the relative cost population, obstacles in the labor demand side should also of setting up the electricity service: 941.8 percent of income be addressed and policies to boost the formal sector must per capita (from 5,974 percent a year ago), although this is be promoted. 63 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI REFERENCES Ashenfelter, O., and A. Krueger. 1994. Estimates of the Heckman, J. J., L. J. Lochner, and P. E. Todd. 2006. Economic Return to Schooling from a New Sample of “Earnings Functions, Rates of Return and Treatment Twins. The American Economic Review. Vol. 84, No. 5 Effects: The Mincer Equation and Beyond.” Handbook of (Dec., 1994), pp. 1157-1173 the Economics of Education 1: 307–458. Azevedo, J. P., G. Inchauste, S. Olivieri, J. Saavedra, and La Porta, R., and A. Shleifer. 2014. “Informality and H. Winkler. 2013. Is Labor Income Responsible for Poverty Development.” Journal of Economic Perspectives 28 (3): Reduction? A Decomposition Approach. Washington, DC: 109–26. World Bank. López-Calva, L. F., and C. Rodríguez-Castelán, 2016. Barros, R. P. D., M. D. Carvalho, S. Franco, and R. Pro-Growth Equity: A Policy Framework for the Twin Goals. Mendonça, R. 2006. Uma análise das principais causas Washington, DC: World Bank. da queda recente na desigualdade de renda brasileira. Ministere de l’Education Nationale et de la Formation No 1203, Discussion Papers, Instituto de Pesquisa Professionnelle. 2018. “Cartographie des ecoles de base.” Econômica Aplicada- IPEA. http://qgiscloud.com/Destini/projet_ETS_djibouti/ Card, D., and A. B. Krueger. 1992. “Does School Quality ?bl=mapnik&st=&l=ecole-djibouti%2Cregion-djibouti&t Matter? Returns to Education and the Characteristics of =projet_ETS_djibouti&e=4790662%2C1291018%2C4810 Public Schools in the United States.” Journal of political 982%2C1300946. Economy 100 (1): 1–40. Narayan, Ambar, Roy Van der Weide, Alexandru Cojocaru, DISED (Direction de la Statistiques et des Etudes Christoph Lakner, Silvia Redaelli, Daniel Gerszon Démographiques). 2012a. Annuaire Statistique 2012. Mahler, Rakesh Gupta N. Ramasubbaiah, and Stefan Thewissen. 2018. Fair Progress?: Economic Mobility DISED. 2012b. EDAM 3-IS. Djibouti: DISED. Across Generations Around the World. Washington, DC: DISED. 2017a. Annuaire Statistique 2017. Djibouti: DISED. World Bank. https://openknowledge.worldbank.org/ handle/10986/28428. DISED. 2017b. EDAM4-IS. Djibouti: DISED. Nunez, J. I., and L. Miranda, L. 2010. “Intergenerational Duflo, E. 2001. “Schooling and Labor Market Income Mobility in a Less-Developed, High-Inequality Consequences of School Construction in Indonesia: Context: The Case of Chile.” B. E. Journal of Economic Evidence from an Unusual Policy Experiment.” American Analysis and Policy 10 (1) Economic Review 91 (4): 795–813. Piraino, P. 2015. “Intergenerational Earnings Mobility Gong, H., A. Leigh, and X. Meng. 2012. “Intergenerational and Equality of Opportunity in South Africa.” World Income Mobility in Urban China.” Review of Income and Development 67: 396–405. Wealth 58 (3): 397–592. CHAPTER 2 64 Psacharopoulos, G., and H. A. Patrinos. 2018. “Returns to World Bank. n.d. “Tables, Charts, and Technical Notes.” Investment in Education: A Decennial Review of the Global Global Consumption Database. http://datatopics. Literature.” Education Economics 26(5): 445-458. worldbank.org/consumption/detail. Rauch, J. E. 1991. “Modelling the Informal Sector World Bank. 2018a. Economic Transformation in Djibouti: Formally.” Journal of development Economics 35 (1): Systematic Country Diagnostic. Washington, DC: World 33–47. Bank. WDI (World Development Indicators). Database. World Bank. 2018b. Poverty and Shared Prosperity 2018: Washington, DC: World Bank. https://datacatalog. Piecing Together the Poverty Puzzle. Washington, DC: worldbank.org/dataset/world-development-indicators. World Bank. World Bank. 1997. Djibouti. Crossroads of the Horn World Bank. 2018c. World Development Report 2018: of Africa. Poverty Assessment Report 16543-DJI. Learning to Realize Education’s Promise. Overview Washington, DC: World Bank. booklet. Washington, DC: World Bank. World Bank. 2014. Doing Business. Washington, DC: World Bank and OECD (Organization for Economic Co- World Bank. operation and Development). 2018. “GDP Growth (Annual %).” World Bank Open Data. https://data.worldbank.org/ World Bank. 2019. Doing Business. Washington, DC: indicator/NY.GDP.MKTP.KD.ZG. World Bank. UNESCO Institute for Statistics. 2018a. “Literacy Rate, Adult Total (% of People Ages 15 and Above).” World Bank Open Data. https://data.worldbank.org/indicator/SE.ADT. LITR.ZS. UNESCO Institute for Statistics. 2018b. “Literacy Rate, Youth Total (% of People Ages 15–24).” World Bank Open Data. https://data.worldbank.org/indicator/SE.ADT.1524. LT.ZS. World Bank. 1997. Djibouti: Crossroads of the Horn of Africa. Poverty Assessment Report 16543-DJI. Washington, DC: World Bank. http://documents. worldbank.org/curated/en/557421468748480282/ Djibouti-Crossroads-of-the-Horn-of-Africa-poverty- assessment 65 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI CHAPTER 2 66 SELECTED TOPICS TO INFORM PUBLIC POLICY CHAPTER 3 IN DJIBOUTI Chapter 3 provides a compilation of short notes that the targeting tools to help support the most vulnerable in exploit recent data to address current policy questions Djibouti. Proxy means test (PMT) approaches have been relevant to the Djiboutian context. The EDAM4 provides used in the country for several years, with the previous a very recent snapshot of the population (as of 2017) model being defined in 2013. However, with the availability that allows us to run simulations of the potential impact of new data, coupled with an updated view of the welfare of certain policy actions and at the same time better landscape in the country and the recognized needs understand the current state of certain populations in the of social programs, we are able to identify and better country and identify their future needs. serve the poorest among the Djiboutian population. The section presents the development of a PMT model using The first section, “Introducing competition to the telecom EDAM4 data and discusses its performance. The section sector,” presents an overview of the telecom sector in concludes with a simulation of the potential benefits for Djibouti. It describes the current state of affairs, challenges poverty reduction by the expansion of a social program in access and coverage, and the take-up of telecom that uses PMT as its primary targeting mechanism. services among the population. Based on international experiences, the section runs a hypothetical scenario in Finally, in the third section, “Nomads and pastoralists,” we which the sector is opened to new competitors. Using discuss nomadism in Djibouti. According to the Djibouti a World Bank tool, we present the potential implications census of 2009, about 20 percent of the population was for welfare of the price changes that are expected to found to be nomadic. Since this is a significant percentage accompany higher competition. of the Djiboutian population, it is important to know more about the well-being of this population and implications The second section, “Improved targeting of social they may have for public policy. programs,” discusses the use of recent data as one of INTRODUCING COMPETITION TO THE TELECOM SECTOR The rise of digital technologies and the digital economy models. Digital technologies are expanding access to offers a once-in-a-generation opportunity to unlock new global markets, changing business models, and delivering pathways for rapid economic growth, economic mobility, enormous productivity gains. Digitization is expanding innovation, job creation, and access to quality services access to basic needs and services. In 2016, the global that would have been unimaginable even a decade digital economy was worth $11.5 trillion, or 15.5 percent ago. The accelerating pace of technology diffusion, the of global gross domestic product (GDP). It is expected to convergence of multiple technologies, and the emergence reach 25 percent in less than a decade, far outpacing the of global platforms are disrupting traditional development growth of the ‘traditional’ economy. 67 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI The information communications technology (ICT) sector the sector has been driven by the economic strategy of holds enormous potential for welfare improvements Djibouti Telecom (DT)—the single national fixed and mobile by creating and enabling more jobs and economic operator to provide ICT services.25 DT has a monopoly in opportunities, improving labor productivity, and generating all telecommunications markets, including those that are more consumer surplus (World Bank 2016). Through these typically open to competition such as mobile and data channels, continuous development of ICT services holds services. It represents a significant portion of the economy, enormous potential for economic growth and poverty with revenues accounting for up to 6 percent of GDP alleviation. While the Internet could be an effective force of (World Bank 2015). development, it also poses potential risks. Some of the key DT’s strategy is outward oriented, selling much of its risks highlighted in the report are the risk of concentration capacity on the international market, mainly to Ethiopia. of market power as well as greater inequality among The operator has banked on the country’s geostrategic the nation’s population, manifested by the digital divide. position between the Middle East and Africa to Further, issues such as privacy, cybersecurity, and Internet successfully become a reliable offeror of connectivity governance need to be kept in mind as approaches to between the two regions for global communications regulate the Internet are considered. Djibouti is particularly carriers. Unfortunately, this approach has not been equally well placed to take advantage of the digital economy. matched by efforts to expand broadband access and use The country has a regional comparative advantage as the in the domestic market. DT’s relatively cheap bandwidth landing site of the undersea fiber optic ICT cables. The is not widely offered to consumers and businesses locally, full potential of this geostrategic position does not seem leaving the domestic market in shortage. to have been exploited, however. The development of FIGURE 3.1 Selected ICT World Rankings for MENA and SSA countries Djibouti ITU, ICT Development Index Ranking 2016 Djibouti UN, e-government ranking 2016 Source: ITU 2017; UNDESA 2017. Notes: MENA countries in blue, SSA countries in orange. Djibouti’s 2016 rank shown in green. 25 The only two other countries in the world with a single operator are Ethiopia and Cuba. North Korea introduced a second mobile operator in 2015. CHAPTER 3 68 A lack of competition to drive service delivery, productivity, several products and services available digitally. From the and innovation could be one reason for the poor standpoint of both the government and the private sector, performance in the sector. In 2016, Djibouti ranked the Internet can bring major benefits in the provision and 161st out of 175 countries in the ICT Development delivery of services. Digital identification can also improve Index published by the United Nations International participation of and help for the disadvantaged groups Telecommunications Union (ITU 2017), the last place in the country to become integrated. The advantages among lower middle-income countries (Figure 3.1). More of a well-developed ICT sector are manifold and worryingly, the country has fallen in the rankings over the transformational. past five years. Another ranking, provided by the United Nations Department of Economic and Social Affairs PROVISION OF ICT SERVICES IN (UNDESA), provides a similar conclusion when evaluating DJIBOUTI the country’s e-government: Djibouti was ranked 187th out The sluggish development of the ICT sector and limited of 193 countries in 2016, and its ranking has dropped since availability of digital services have differential effects across 2010.26 Finally, GSMA (Global System Mobile Association) the income distribution. The subgroup of the population estimates that the unique mobile subscription penetration that is most affected by the sectoral environment is represents less than 30 percent of the total population in expected to be the (mostly poor) rural population of 2018—lagging the Sub-Saharan Africa (SSA) average of 45 Djibouti. Despite having one of the highest proportions percent and the MENA average of 64 percent. of urban population among the lower middle-income countries in MENA and SSA (with about 85 percent of Unfortunately, the (under)performance of the ICT sector the population), mobile broadband coverage in Djibouti has significant macroeconomic effects on the economy in 2018 remains below the regional average (at about 76 and poverty reduction (World Bank 2016). For example, percent) and far lower than many MENA or SSA countries in rural Peru, the expansion of mobile network coverage that have a smaller share of urban population (GSMA boosted household real consumption by 11 percent 2018). This low national coverage rate implies that the (Beuermann, McKelvey, and Vakis 2012). The ICT sector vast majority of the rural population (which account for 23 has a great capability to enable job creation and innovation percent of the total population) have no mobile broadband in the private sector, with the corresponding increase in coverage in Djibouti. economic opportunities for the population. Moreover, the Internet is also an enabler of higher labor productivity Further evidence of the disparities of access can be found by means of lowering information and search costs. For in the recent EDAM4, conducted in 2017. Results from example, the introduction of mobile phones in the grain the survey show the extent of the digital gap between the markets of Niger led to farmers obtaining grain price richest population and the poorest population. Eighty- information over the phone, thereby reducing search costs nine percent of the households in the richest quintile (the by 50 percent (Aker 2010) and reduced dispersion of 20 percent of households with the highest consumption grain prices across markets by 10 percent (Aker and Mbiti levels) own a least one mobile phone, a figure similar to 2010). People’s perceptions also reveal that the Internet the households in the third quintile, where 81 percent own has led to an increase in consumer welfare by making 26 The recent launch of the e-government site by the Agence Nationale des Systèmes d’Information de l’État is not likely to be captured in this ranking. 69 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI at least one mobile phone. However, the figure drops to least one mobile phone, and the rural population, where 70 percent for the second quintile, and 41 percent for only 25 percent of the households own a mobile phone poorest quintile (the bottom 20 percent). The discrepancy (Figure 3.3). is higher in relative terms for computer ownership: whereas The lack of mobile phone infrastructure (and associated 27 percent of the households in the top quintile own a coverage) in rural areas—combined with less than optimal computer, there are virtually no households in the bottom transport infrastructure and low penetration of telephone 20 percent that own one (Figure 3.2). As the richest service—helps explain why a large share of the rural households are mostly located in urban areas and the population remains offline. Only one-third of the rural poorest households are mostly located in rural areas, population is less than 10 minutes away from a location there is a geographical digital divide between the urban where a phone call can be made, compared to two- population, where 70 percent of the households own at thirds of the urban population. Even worse, a third of the FIGURE 3.2 Digital gap between the richest and poorest households by quintile National average 10% 75% Richest quintile 27% 89% 4th quintile 9% 86% 3rd quintile 3% 81% 2nd quintile 2% 70% Poorest quintile 0% 41% Computer Cell phone Source: Calculations based on EDAM4-IS. FIGURE 3.3 Digital gap between urban and rural households National average 10% 75% Djibouti City 13% 89% Urban (except capital) 7% 70% Rural 0% 25% Computer Cellphone Source: Calculations based on EDAM4-IS. CHAPTER 3 70 rural population are more than an hour away from such whole income of the poorest 60 percent of the population. infrastructure, compared to less than 5 percent for the Currently, the cost of a basic asymmetric digital subscriber urban population (Figure 3.4). line (ADSL) package (3 megabits per second, or Mbps) in Djibouti is about DF 90,000–100,000 per year (Table Affordability also appears to play an important role in 3.1). This is higher than the estimated average annual explaining the low coverage of ICT services. Gelvanovska, consumption per capita of the bottom 20 percent of the Rogy, and Rossotto (2014) estimated that the price of population, and roughly equal to the estimated yearly per fixed broadband in Djibouti would absorb roughly the FIGURE 3.4 Travel Distance from Telephone Service Infrastructure National average Djibouti City Urban (except capital) Rural 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1-10min 10-30min 30-60min 60+min Source: Calculations based on EDAM4-IS. capita consumption of the third decile of the population consumption level of an individual belonging to the average (Table 3.2). A 4 Mbps Internet package would instead household in the ninth decile.27 It is notable that the prices be equivalent to the yearly consumption for an individual from 2018 are substantially lower than the costs before; belonging to the fifth decile of the population. The cost for according to ITU (2018), DT had cut the price of an entry- the fastest ADSL service (6 Mbps) is roughly equal to the level plan by 70 percent between 2016 and 2017.28 TABLE 3.1 Djibouti Telecom Costs, by Service (Djiboutian francs, DF) ADSL for private use* Available plans 3 Mbps 4 Mbps 5 Mbps 6 Mbps Monthly plan 9,000 14,000 18,000 25,000 Yearly plan (15% discount) 91,800 142,800 183,600 255,000 Source: Djibouti Telecom, http://www.djiboutitelecom.dj/particulier-internet-adsl-particulier.html, accessed November 22, 2018. * Internet for particuliers (private individual). 27 To put this estimate in context, in 2018, the Broadband Commission for Sustainable Development agreed on a target for 2025 for entry-level broadband services at less than 2 percent of monthly gross national income (GNI) per capita. The current estimate for Djibouti for a fixed-broadband basket is 9.8 percent of GNI per capita (ITU 2018). 28 The price drop was accompanied by a reduction in the cap (from 50 gigabits to 30 gigabits per month) and the speed (from 3 to 1 Megabytes/s). 71 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 3.2 Average per Capita Consumption (DF by Quintile) Quintile Average per capita consumption (annual) 1 56,182 2 108,506 3 156,757 4 224,548 5 496,690 National 208,224 Source: Calculations based on EDAM4-IS. POTENTIAL IMPACT OF A NEW Findings suggest that new entrants usually do charge less ENTRANT ON RETAIL PRICES IN than incumbent operators,29 but it is not clear the extent DJIBOUTI to which potential reductions in quality of service (such as a lower national mobile coverage or a lower quality of While there is a broad consensus that regulatory reform customer care) affect the overall benefits of consumers. in telecommunications is beneficial for businesses and individual consumers, it is very difficult to assess ex More recent studies have focused on the effect of market ante the evolution of the performances of the sector concentration on retail prices. For example, Genakos, with expected changes in firm ownership, regulatory Valletti, and Verboven (2015) focused on the link between framework, and market structure (especially with the measures of market concentration and prices paid by end introduction of a second operator in a monopolistic users in mature OECD markets. They conclude that one market). Boylaud and Nicoletti (2001) conducted a additional competitor is associated to a price reduction of study on Organization for Economic Co-operation and approximately 8.6 percent (ranging from 7.9 percent to 15.9 Development (OECD) countries and concluded that the percent, depending on numerous factors). At the same time, gains in the reduction of mobile prices were linked to the authors caution that impacts from entry and mergers of the number of competitors (proxied by the share of new different firm sizes could not be reliably established. A study entrants or by the number of competitors), the market published by the British regulator Ofcom in 2016 contains shares of the competitors, the prospect of competition a cross-country econometric analysis of the effects of the (as proxied by the number of years remaining before number of mobile networks on prices, and the effects of liberalization), the specificities of the national markets, and “disruptive” entry in mature OECD markets (Ofcom 2016).30 other factors (effect of ownership, economic structure, It finds that an increase in the number of mobile networks technology, and price rebalancing). reduces prices by 7.3 to 9.2 percent. It also finds that where a disruptive player is present, prices are lower by 10.7 to In 2002, the European Commission released a report on 12.4 percent. This brings the total effect of disruptive new the dynamics of the telecommunications sector in the entry to 17.2 to 20.5 percent. European Union after market liberalization (EC 2002). 29 For national calls, the study finds that operators charged up to 56 percent less in the United Kingdom, 46 percent less in France, and 35 percent less in Germany. 30 A disruptive player is defined by Ofcom as a player that presents the following behavior: (i) introduction of new innovative services which supersede others, (ii) introduction of new production technologies that increase efficiency for existing services, and (iii) aggressive behavior where the player competes vigorously and prioritizes gaining market share above other considerations such as profits or cost recovery in the short or even medium term. CHAPTER 3 72 Even though the introduction of a new mobile entrant will in 2003. The third operator is Saudi and obtained his surely have an impact on the performance of the sector license in 2009 and was fully operational in 2011. and potentially decrease retail prices significantly, it is Additional licenses generated approximately US$33 difficult to quantify this effect ex ante. Based on cross- million (approximately BD 83 million in local currency) country empirical analysis, the decrease in retail prices and approximately US$500,000 in annual spectrum induced by a new entrant ranges from 10 to 20 percent (in fees. The introduction of the third operator boosted mature markets), and up to 35 to 56 percent (in developing competition and increased subscriber penetration markets, that is, the mobile market situation in the from 65 percent in 2010 to 91 percent in 2009. European Union in the early 2000s). 31 Brunei—Population 393,162 and two telecom The following cases from international experience can operators: DST-Group and B-mobile. The second provide hints at the potential effects in the Djiboutian operator (B-mobile) was a joint venture between the context. The examples listed come from countries with local fixed operator TelBru and a local conglomerate a comparable population, and that have been able to (QAF Comserve) and launched its operation in 2005. develop a controlled sector by introducing a second The amount of the second license is not disclosed, mobile operator alongside the incumbent: but the introduction of competition boosted single subscriber penetration from 47 percent in 2005 to 67 Cabo Verde—Population 500,000 and two percent in 2007. telecom operators: Cabo Verde Telecom and T+. Following an unsuccessful bidding process in 2004 Mauritius—Population 1.3 million and three for the second mobile license, the government began telecom operators: Emtel, Cellplus, and MTML. direct negotiations with a foreign investor (Alexander The first two mobile operators launched their GSM Group Telecommunications), which launched the (Global System of Mobile) services in 1996 (Cellplus) T+ operator in 2005 on a limited scale. The treasury and 1999 (Emtel), so that competition in Mauritius was benefited from the sale of the second license, which already effective in the early 2000s. The third operator was purchased for the equivalent of US$ 2 million obtained its license in 2005. Because competition (about CVEsc 180 million in local currency). In 2007, between the first two operators was already fierce, the second operator finally launched its mobile services with strong mobile growth and lower prices, the on the island of Santiago, which is home to about half introduction of the third operator had a minor impact of the population, and the single penetration of mobile on the market. subscribers (percentage of the population with at least To further explore the potential effects of changes to the one mobile line) doubled in two years, from 21 percent structure of the ICT sector, we conduct the following in 2007 to 40 percent in 2009. thought exercise. In the presence of more competition, Bahrain—Population 1.2 million and three what could be the potential effects on prices in telecom operators: Bahrain Telecommunication telecommunication services in Djibouti? And, what would Company, Zain Bahrain and Viva. The second be the potential effects in terms of increased welfare for operator, Zain Bahrain, is 55 percent owned by the households? The results of this simulation are presented in Kuwaiti operator Zain Group, which obtained a license the next section. 31 Another example is the entry of AT&T in Mexico, where a decrease of about 13 percent in the price of mobile services was experienced. 73 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI SIMULATING INCREASED of telecommunication services among the Djiboutian COMPETITION IN THE ICT SECTOR IN population partly explains why only about 22 percent of DJIBOUTI the population is found to be a telecom user, with this share being driven mostly by purchase of cards of credit We use the EDAM4 data from 2017 to explore the ICT for mobile phone minutes. Moreover, there are marked sector in Djibouti and conduct a simulation on the potential differences in usage rates across the consumption benefits of market forces in decreasing the prices of distribution. Among the poorest decile, less than 5 percent ICT services. We present basic summary statistics on of the population declared having expenditures on telecom expenditures related to the sector. Next, we discuss the services. This rate increases, with the level of consumption methodology and the tool used for simulations, before reaching 23 percent around the middle of the distribution presenting the findings. and to about 35 percent among the top 10 percent (Table We identify as “consumer of telecom services” the 3.3). Expenditures on telecom services also appear to be households that responded having spent money during correlated with the level of consumption, with users in the the past month on fees on phone services (fixed or top 10 percent of the distribution spending more than 10 mobile), bought mobile “credit,” used a phone booth, times the level of expenditures of those in the bottom 10 and/or paid a monthly contract for phone (fixed or percent. It is notable that users devote similar shares of mobile) or Internet services. The relatively low penetration their total expenditures to telecom services. TABLE 3.3 Summary Statistics on Telecom Expenditures and Users expenditure (%) Consumers (%) consumers (%) expenditure of Consumption Consumption consumption consumption among users consumers of telecom per capita per capita Telecom Telecom services Annual Annual among Decile 1 4.1% 166 4,009 0.0% 1.2% 40,491 50,457 2 13.1% 1,004 7,668 0.2% 1.3% 71,982 74,627 3 16.6% 1,570 9,468 0.2% 1.1% 97,836 99,662 4 22.2% 2,401 10,800 0.2% 1.1% 119,223 120,107 5 22.9% 6,946 30,367 0.4% 1.9% 143,328 143,115 6 22.8% 4,557 19,967 0.4% 1.6% 170,249 169,540 7 27.2% 5,034 18,504 0.4% 1.4% 203,287 202,935 8 25.3% 7,525 29,757 0.4% 1.7% 245,532 247,397 9 29.2% 17,411 59,569 0.7% 2.4% 318,732 324,248 10 34.6% 21,704 62,762 0.6% 1.8% 672,770 699,677 Overall 21.8% 6,822 31,299 0.4% 1.6% 208,224 266,367 Source: Calculations based on EDAM4-IS. Note: Deciles based on the distribution of consumption per capita. The coefficient of variation of telecom usage was 20 percent or higher in deciles 1 and 2 (grayed out) and thus may be interpreted with caution. CHAPTER 3 74 TABLE 3.4 WELCOM Simulations Elasticity = -1.5 Elasticity = -2.5 Price change 22.2 % 13.3% Welfare gains (DF per year) per user Quintile 1 n.a. n.a. Quintile 2 828 276 Quintile 3 1,792 597 Quintile 4 2,290 763 Quintile 5 7,347 2,449 Source: Calculations based on EDAM4-IS. Note: Quintiles based on the distribution of consumption per capita. Results for the bottom quintile were 594 and 198, respectively, but are not shown as it is considered that there is too little information (i.e. sample sizea coefficient of variation of 18 percent) to draw reliable estimates. To conduct the simulation, we apply the World Bank– Table 3.4 presents the main results from the simulations. developed tool on welfare and competition (WELCOM). As expected, both simulations show a decrease in price The tool allows the simulation of the distributional effects after a new entrant establishes itself in the telecom sector. of changes in market competition through its impact on In the scenario where individuals are assumed to be prices. In short, the tool (i) estimates the expected change less responsive to price changes, the price of telecom in prices resulting from the increased competition, by service experiences a drop of 22 percent, whereas when assuming that after new entrants come into the sector individuals are more responsive, the price drops by 13 competition pushes prices toward the marginal cost (the percent. In terms of welfare gains, they follow the patterns result in perfectly competitive markets); (ii) identifies all of consumption observed above, with richer individuals users of telecom services; and (iii) applies the estimated spending more on telecom services than poorer individuals. price decrease to households that are currently users Thus, richer households reap more of the benefits of a of telecom services as an estimate of their gain in decrease in prices. Under the lower elasticity scenario, an welfare. Given the lack of guidance in the literature on an 32 individual in a household that uses telecom services in the appropriate price elasticity for telecom services, we run the second quintile is estimated to experience a welfare gain of simulations assuming a price elasticity of -1.5 and -2.5 to DF 828. In the same scenario, an individual in the top 20 provide a range of results. To note is that in both cases the percent would gain DF 7,347. The welfare gains are lower simulations assume that the demand for communications across the distribution corresponding to the lower price services is elastic, since in monopolistic markets the firm change in the scenario of an elasticity of -2.5. optimizes its output in the elastic section of the demand The welfare gains are low with respect to the overall curve. Thus, we present the results for the simulation level of consumption in each quintile, and thus are not where the opening of the sector leads to a drop in the conducive to tangible changes in poverty or inequality. market share of the incumbent firm to half of the market. It is important to highlight that the results presented are That is, we simulate the expected effects in the short or limited, as they can simulate the potential welfare gains medium term after the opening of the sector. 32  For a more detail description of the methodology, see Araar et al. (2018). 75 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI only among current telecom users. In a context such as Second, we simulate the increase in the probability of Djibouti, an equally important question would be, what using telecom services by assuming that all nonusers would happen to the individuals that are not currently would (potentially) benefit from the average welfare gains consuming these items (due to several reasons, including just described based on the quintile of consumption that affordability or coverage)? That is, given an expected drop they currently belong to. Table 3.5 illustrates that there in price and an associated welfare gain, should we expect are limited potential gains in the extensive margin (that more people to adopt and use telecom services? is, take up of telecom services). Using the low elasticity as benchmark, households in the fifth quintile would be We attempt to answer this question by conducting the 0.0014 percentage points more likely to use telecom following thought exercise. First, we run a probit model services (assuming all other covariates constant). The to look at the predictors of telecom usage among the probability of take up is correlated with households’ Djiboutian population. We use as covariates a series of consumption, but differences are not significant. This result household characteristics, household head characteristics, is likely due to the small gains in the short run. Higher dwelling characteristics, location variables, the household’s effects could be expected as competition leads to larger consumption level, as well as the usage of telecom drops in prices in the long run. Besides the further drop in services among households in the same primary sampling prices due to more competition, other indirect effects such units (PSUs). We estimate the probit via a stepwise as having neighbors who take up telecom services could regression and find that consumption level and access also have an impact on individual take up. within the PSU are highly correlated with telecom usage. TABLE 3.5 Estimated increase in probability of using telecom services due to welfare gains Elasticity = -1.5 Elasticity = -2.5 Quintile 1 n.a. n.a. Quintile 2 0.0003 0.0001 Quintile 3 0.0006 0.0002 Quintile 4 0.0008 0.0003 Quintile 5 0.0014 0.0005 Source: Calculations based on EDAM4-IS. Note: Results shown in percentage points. Quintiles based on the distribution of consumption per capita. Results for the bottom quintile are not shown as it is considered that there is too little information (i.e sample size) to draw reliable estimates. CONCLUSION narrow angle of welfare gains to households, the simulations presented here hint at important gains due to the expected In this section we bring some insights as to the hypothetical price decreases following increased competition in the gains to the population that an opening of the telecom sector. More work is needed to fully understand the sector could bring about in Djibouti. The ICT sector is implications of changes to the telecom sector in Djibouti. currently believed to be below its potential, thus limiting The huge potential the sector holds for contributing to the the access to such services and, somewhat indirectly, development of the country should be enough incentive to preventing more spaces for the private sector in Djibouti to pursue this important agenda.33 become bigger players in the economy. From a relatively 33  The EDTIC survey conducted in 2018 by DISED is expected to provide additional insights on the sector. CHAPTER 3 76 IMPROVED TARGETING OF SOCIAL PROGRAMS Djibouti’s approach to expanding social safety nets (SSN) USAID, and the Norwegian Refugee Council), which were has been increasingly aligned to poverty reduction and mainly focused on providing food to vulnerable populations. building of human capital. This focus is in contrast to other At present, the scale and funding of an integrated SSN MENA countries in the region whose primary motivation for program remains inadequate to protect most poor and reform has been driven by efforts to rebalance generalized vulnerable groups. subsidies on commodities (mainly energy products) with Nonetheless, expenditure on SSNs in Djibouti remains low. more targeted transfers to households.34 Djibouti has While the MENA as a region spends, on average, 1 percent made efforts in recent years to invest in adaptive SSNs and of GDP on SSNs, Djibouti spends only 0.2 percent of GDP incentivize households to invest in human development. (as of 2014). Within the region, only two other countries Until recently, most SSN programs had been established in (Jordan and Egypt) spend less as a percentage of GDP. the wake of drought shocks and were largely donor-driven Meanwhile Iraq, the West Bank and Gaza, and Syria all initiatives (such as World Food Program, UNICEF, Food spend above the global average (Figure 3.5). and Agriculture Organization, Islamic Development Bank, FIGURE 3.5 Social Assistance Expenditure as a Percentage of GDP 3.0% 2.5% Percentage of GDP 2.0% 1.5% 1.0% 0.5% 0.0% Ba a, 2 , za C C 15 A ld A an –13 4 ba , 20 10 A 0 Tu me –13 5 SA 0 5 20 4 a, 201 01 EA EN n, 3-1 LA SS EC -1 –1 or 1 Sy Ga ib 14– 0 20 an 012 Ye 09 09 2 W rd 011 M 1 d n, oc in, 0 Eg uti, 20 ,2 tB ,2 ri 2 a hr o t, an o k es raq yp c no si Dj I ni Jo or Le M W Source: Report calculations and ASPIRE database. Note: Calculated individual programs expenditure as a percentage of that year’s GDP (real GPD) and aggregated all programs by country. Data are administrative data collected from country counterparts. MENA stands for Middle East and North Africa, ECA stands for Europe and Central Asia, SSA stands for Sub-saharan Africa, LAC stands for Latin America and Carribean, EAC stands for East Asia and SA stands for South Asia. 34 In the MENA region, generalized price subsidies have been a major part of the social contract between governments and the people. Large-scale subsidies are still pervasive, including in foods and fuels. Yet generalized subsidies are both costly and usually badly targeted. This is especially the case with energy products. Only 7 percent of fuel subsidy spending in poor countries benefits the poorest quintile of households, while 43 percent benefit the richest quintile. For example, in 2008, the poorest 40 percent of the population in Egypt received only 3 percent of gasoline subsidies (Sdralevich et al. 2014). Globally, the IEA estimates that only 8 percent of subsidies accrue to the poorest fifth of the population (IEA 2012). 77 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Accompanying this relatively low spending on SSNs, build on the capacity development from recent projects, Djibouti’s programs are also fragmented, and donor the development of the social registry, and a management driven. The low level of spending in SSNs is not channeled information system. To that end, creating the fiscal space efficiently. Djibouti has nine different safety net programs and investing in the institutional and capacity building (Table 3.6). Most of these programs are cash transfer to establish a country-owned SSN system in Djibouti is programs (either conditional or not). The high fragmentation becoming a key priority for GoD. of these programs leaves significant scope for improvement The GoD plans to reinforce its social protection system by through consolidation and better use of the resources. The expanding the PNSF by 5,000 households. In particular, most prominent SSN program in Djibouti is the Programme the expansion is expected to target households outside National Solidarite Familiale (PNSF), a conditional cash the capital and in subprefectures selected due to their transfer program that makes up more than one-third of the high level of need. Furthermore, PNSF includes soft available SSN budget. Besides regular cash assistance, conditionalities related to participation in community-level beneficiaries of the PNSF are also eligible for a heavily behavioral change communication sessions. Thus, besides subsidized medical insurance under the Programme the direct support for increased consumption, the program d’Assistance Sociale de Santé. will aim to nudge beneficiaries toward building their human Djibouti has begun laying the groundwork for a social capital. Finally, the expansion of PNSF will be supported by protection system through the national expansion of the incorporating elements of targeting based on proxy means PNSF and the establishment of a social registry. PNSF testing. Here we present the results of an analytical exercise currently serves approximately 3,000 households. The using the EDAM4-IS data to build a prediction model that program has, however, the potential to be the cornerstone can be used (i) to estimate the consumption level of a of a country-owned and adaptive social protection system. households and (ii) as input for the determination of whether The emphasis is now on the scaling up of PNSF to cover a household is eligible for certain social programs. The a greater share of the poor population. The scale-up will following sections describe the findings of this exercise. TABLE 3.6 Social Safety Net Programs in Djibouti Program name Program type Expenditure (US$) Zakat UCT .. Food distribution Program Food, in-kind assistance, and near-cash transfers .. Scholarships for education Other Social Assistance Programs .. School feeding program Food, in-kind assistance, and near-cash transfers .. Food voucher (July–September) UCT 500,000 University students (canteen and transport subsidy) CCT 929,600 Social Assistance Pilot Program on Labor and Public works 1,381,623 Human Capital Peri-urban voucher program Food, in-kind assistance, and near-cash transfers .. Programme National Solidarite Familiale Rural and urban cash transfer program 1,521,000 Source: Compiled from GoD data. Note: UCT = unconditional cash transfer; CCT = conditional cash transfer. CHAPTER 3 78 A PROXY MEANS TEST (PMT) consumption per capita on the set of explanatory variables. APPLICATION USING EDAM4-IS Formally, the model is specified as In 2013, the first PMT formula was used by the Secreatariat (1) d’Etat des Affaires Sociales to help in its targeting efforts and identify the most vulnerable population groups in where Wі is the log of consumption per capita per Djibouti. Given the strong economic growth observed in household (welfare aggregate), Xіj is a vector of the country in recent years, it is thus important to explore characteristics of j variables (poverty predictors) for i whether an updating of the formula is required to better households, βj are the coefficients (weights) of the poverty reflect the current conditions of poor households in Djibouti. proxies, α is the constant, and εі the error term. It is thus fortunate that in 2017, the DISED conducted The goal of the exercise is to maximize the R-squared— EDAM4 to update the profile of welfare and poverty in the explanatory power of the model—and to minimize the the country. Such recent data can provide a wealth of inclusion (leakage) and exclusion (undercoverage) errors. information for such an exercise. Table 3.7 presents the different possible performance While PMT models are known to have limitations, there scenarios. Households will be successfully targeted (S1 and are contexts in which the application of this approach is S2) when poor and nonpoor households are beneficiaries warranted. Given Djibouti’s high level of informality and 35 and nonbeneficiaries, respectively. On the other hand, weakly integrated information systems, a PMT model when poor households are nonbeneficiaries because can be an informative tool to help achieve the objective predicted as nonpoor and conversely, nonpoor household of finding the most vulnerable population. In this section, beneficiaries, these will incur exclusion (E1) and inclusion we describe the estimation of a predictive model that can (E2) errors, respectively. From this set of possibilities, the help identify vulnerable households by collecting data on a following concepts are defined. Coverage rate is the ratio of short list of observable characteristics. We first describe the beneficiaries to total population (M1/N*100). Leakage rate is methodology of the PMT, followed by the estimated model. the ratio of inclusion error to total beneficiaries (E2/M1*100). We conclude by describing the performance of the model. Undercoverage rate is the ratio of exclusion error to total poor (E1/N1*100). Performance of the models is tested by METHODOLOGY comparing undercoverage and leakage rates between PMT. Consumption per capita is the welfare aggregate used to measure and predict poverty. An ordinary least squares From a practical point of view, other considerations need to estimator is used to predict the natural logarithm of be taken into account in choosing a particular PMT model. One objective, for instance, could be to raise the welfare of TABLE 3.7 Inclusion and Exclusion Errors Poor Nonpoor Total Beneficiary Successful targeting (S1) Inclusion error (E2) M1 Nonbeneficiary Exclusion error (E1) Successful targeting (S2) M2 Total N1 N2 N 35  See the discussion in Brown, Ravallion, and Van de Walle (2018) on PMT for nine African countries. 79 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI the poor, in which case minimizing the exclusion error can services like electricity, improved sanitation, and water; be give more weight in the decision. On the other hand, characteristics of the head of household, such as officials may be interested in keeping costs of the program education level and employment status; location, such as contained and thus may choose to focus on minimizing region of residence; and ownership of durable goods, such the errors of inclusion. Certainly, the administrative costs of as televisions, refrigerators, radios, and so forth. Most implementing one (or many) PMT models in an effective way of these indicators are difficult to falsify or unlikely to be should also be taken into account. falsified. Unannounced household visits by social workers also limit the ability to falsify information. THE MODEL We find that ownership of durable goods, such as cell A first attempt at defining the PMT model was done by phones, televisions, refrigerators, bicycles, mattresses, applying the 2013 PMT specification to the EDAM4- and radios, is associated with higher predicted welfare. IS data. However, such an approach results in higher Household demographic indicators such as size and undercoverage (exclusion error) and lower national dependency ratio—the ratio of children and elders to coverage today than in 2013. This is not surprising, as the adults—are negatively correlated with welfare. Large context and environment in Djibouti has changed since households with a high prevalence of financially dependent EDAM3-IS, resulting in a weaker performance of the 2013 children and elders are not surprisingly predicted poorer. PMT. A revision of the model, which was previously very The poorer regions of Dikhil and Tadjourah are associated accurate, is necessary to account for the progress and with higher poverty likelihood. Educated heads of change Djibouti has experienced since 2013. The re- household and those working in the public sector are less estimation of the 2017 PMT tested the old covariates, as likely to be poor. Households living in villas or apartment well as new correlates (that aid as proxies) of consumption buildings are less likely to be poor compared to those living (and in turn of poverty). The simplest approach to the re- in tents, makeshift accommodation, ordinary housing (no estimation would be to repopulate the 2013 PMT with new camps, makeshift dwellings etc.), or collective housing. data in order to update the beta coefficients. However, this would limit the potential of the probability model to PERFORMANCE accurately predict current levels of welfare. Therefore, a The myriad possible combinations of the groups of new model is estimated with EDAM4-IS data. variables mentioned demand that we set up an approach The revised 2017 PMT models include a range of to select the most accurate and robust PMT model. The poverty predictors from five categories of household-level common default is to estimate a national PMT model using indicators. The choice of predictors is determined by two a nationally representative household survey. However, factors: (i) the correlation with consumption per capita, given the strong heterogeneity between urban and rural which is important for accurate predictions, and (ii) the communities, estimating two distinct models for each verifiability of the predictor by social workers who conduct environment may be desirable. PMT predictors, such as home visits, which is important to determine the accuracy dwelling characteristics (for example, materials used to of the information imputed in the PMT formula. The build roof, walls, and floors), are not easily comparable different categories of predictors assessed to determine across regions. Thus, following consultations with partners probability of poverty are the following: household and government representatives, we present here the characteristics, such as household size and dependency results from a single national model and a regional model ratio; dwelling characteristics, such as materials used that combines two separate PMT formulae for rural and to construct walls and roofs, as well as access to basic urban households.36 36 A third model for three distinct regions was also estimated, resulting in City of Djibouti PMT, Other Urban Areas PMT (including only regional capital cities), and a Rural PMT. However, for logistical reasons, such an approach was not favored. CHAPTER 3 80 Conventionally, the performance of a PMT model can Table 3.8 compares the performance indicators of two be evaluated by measuring predictions errors. Two 37 models. One is a national model that applies one formula indicators are of key interest. First, the exclusion error or to the entire country. The second model allows the undercoverage occurs when households are flagged as formula to vary between urban and rural areas and is nonbeneficiaries by the PMT formula, but in reality a more labeled “flexible.” The results on performance indicators detailed observation of their consumption level would are presented at the aggregate (national) level for ease of have put them below the eligibility threshold (which many comparison. There are no significant differences between times implies that they are poor). Second, the inclusion the exclusion and inclusion errors of the two models, nor error, or leakage rate, refers to the households that are does one appear to be clearly preferable to the other. As classified as beneficiaries to the program due to a low a rule of thumb, the PMT models are assessed based on PMT score, when in fact they may not be poor or their their performance at the 30th percentile. It is, however, also real consumption levels are above what a given program interesting to use as guidance of performance the 35th considers to be the eligibility threshold. These rates percentile. The reason is that, in the case of Djibouti, this were simulated at different levels of the consumption per level is qualitatively similar to poverty in per capita terms, capita distribution (deciles), as well as the 25th and 35th the upper official poverty rate, and provides relatively percentile, as these are close to the national extreme and more desirable levels for the performance indicators. global poverty rates estimated by DISED (21 percent and Assuming that the population of interest in concentrated 36 percent, respectively). 38 in the bottom 30 or 35 percent of the distribution, the TABLE 3.8 Performance of National and Flexible Model National Flexible Percentile Coverage Exclusion Inclusion Coverage Exclusion Inclusion 10 8.5% 41.7% 31.5% 8.4% 42.3% 31.1% 20 17.1% 35.0% 23.9% 16.6% 35.5% 22.2% 25 23.8% 29.1% 25.4% 22.9% 30.9% 24.5% 30 27.3% 28.1% 21.1% 27.8% 27.5% 21.8% 35 33.3% 24.8% 20.6% 32.6% 25.1% 19.6% 40 39.3% 21.6% 20.1% 38.8% 21.6% 19.0% 50 50.4% 16.4% 17.0% 49.7% 17.1% 16.4% 60 61.4% 12.4% 14.4% 61.7% 12.4% 14.8% 70 72.4% 9.0% 12.1% 72.6% 8.8% 12.0% 80 83.3% 4.4% 8.2% 82.7% 4.7% 7.8% 90 93.2% 1.3% 4.6% 93.1% 1.3% 4.6% Source: Calculations based on EDAM4-IS. Notes: Flexible refers to countrywide indicators obtained by putting together predictions of an urban and a rural model. 37 Each specification was obtained using stepwise regression in STATA. Multicollinearity among regressors on the final specifications was formally tested using the variance inflation factor (VIF) ratio. No multicollinearity was detected, given that all indicators had a ratio below 5. Mean VIF for the three models (national, urban, and rural, respectively) stood at 1.75, 1.35, and 1.64. 38 Poverty rates from DISED are based on an adult equivalent scale. In results not shown, comparable models have been produced to estimate performance based on an adult equivalent scale. The GoD will have to take into consideration potential differences of these models in defining the targeting criteria of programs that will use the PMT. 81 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI national model appears to be slightly better in terms of 24.8 percent (25.1 percent) and a leakage of 20.6 percent exclusion, while the flexible model performs better in terms (19.6 percent). These as well as those at the rates for the of inclusion. A targeting at the 35th percentile would lead 30th percentile are—empirically—low rates of error when in the national (flexible) model to an undercoverage rate of compared to other settings (Figure 3.6). FIGURE 3.6 Errors of Inclusion and Exclusion for PMT Model in Djibouti Compared to Other Countries (cutoff point at 30th percentile) 60 50 40 30 20 10 0 Sri Lanka Mongolia Bolivia Jamaica Peru Djibouti Djibouti (urban) (urban) (urban) national exible Exclusion rate Inclusion rate Source: Calculation based on EDAM4-IS (Djibouti); and compilation in Atamanov et al. 2015 from Grosh and Baker 1995, Narayan and Yoshida 2005, and Araujo and Cararro n.d.. ADDITIONAL PERFORMANCE INDICATORS OF THE of the population.39 The focus is on three groups: small PMT MODEL households defined as household sizes below or equal the Performance in population subgroups 10th percentile (that is, three members), large households as those above or equal to the 90th percentile (that is, ten Overall, homogeneity is greater among households in rural members), and small and elderly households (defined as settings than households in urban regions—performance those with at least 1one elderly member—60 or more years among rural communities is therefore expected to be more old—in a two-person household at most). These households accurate than among urban families. A common concern account for 13 percent, 13 percent, and 1 percent of the by partners on the ground is the failure of PMT models to population sample, respectively. Poverty incidence among predict poverty accurately for small and large households. these groups is, respectively, 16 percent, 29 percent, and 21 This concern is tested by assessing the performance of percent (the national poverty rate being 21 percent). the specifications described above for different subsets 39 Additional estimations were done following the flexible model PMT, in which the household size was included as a series of dummy variables instead of using it in continuous form. The idea was to allow the prediction model to nonparametrically estimate differences in consumption across households of different sizes and that differences could vary between smaller and larger households. This econometric specification worsened exclusion and inclusion errors. CHAPTER 3 82 As a sensitivity analysis, we present performance indicators errors by using the poorer segment of population to of the flexible PMT specification for these three subgroups run the predictive PMT model. The argument is that by (Table 3.9). There is no clear evidence suggesting that the 40 focusing on the poorest half of the population as the basis PMT would significantly underperform in these population for the formula (and assuming that richer individuals and subgroups. For instance, using these results, we find households are unlikely to apply to a social program), that rates of exclusion are lower for all three populations the approach can lead to much lower exclusion errors. of concern than among the overall population. At the Certainly, the assumption that no one from above the 35th percentile, undercoverage is 23.2 percent for small program cutoff would apply for benefits will not apply households, 15.1 percent for large households, and 16.5 cannot be guaranteed. percent for small-elderly households, whereas it is 25.1 The national and flexible models were re-estimated by percent at the population level. Notably, leakage rates are fitting only the bottom 55 percent of the population (a higher, especially in the case of large households (30.2 truncated model). This specification included 35 variables percent), compared to the national average of 19.6 percent. (instead of the more than 50 in the previous models). Table Performance in the bottom of the distribution 3.10 shows the results obtained from (i) the reestimation of both national and flexible models to only the bottom Following Grosh and Baker (1995), we further test the 55 percent of the population, (ii) using the typical cutoff desirability of an alternative to reduce measured exclusion of 35th percentile, and (iii) showing the results of whether TABLE 3.9 Performance of Flexible Model for Subsets of Population Small households Large households Small, elderly Households Under- Under- Under- Percentile coverage Leakage coverage Leakage coverage Leakage 10 50.1% 39.1% 29.6% 43.6% 61.9% 58.6% 20 34.2% 33.1% 19.2% 26.0% 14.2% 36.1% 25 30.6% 28.5% 18.2% 35.1% 18.4% 29.9% 30 25.5% 25.1% 13.8% 32.4% 18.2% 23.0% 35 23.2% 23.0% 15.1% 30.2% 16.5% 25.4% 40 19.2% 21.1% 6.7% 25.0% 9.8% 24.4% Source: Calculations based on EDAM4-IS. TABLE 3.10 Errors of Exclusion and Inclusion for Truncated Model with Predictions for the Whole and for the Restricted Sample Anyone applies to the program Only bottom 55% applies to the program Exclusion Inclusion Exclusion Inclusion National model 11% 34% 11% 23% Flexible model 11% 34% 11% 23% Source: Calculations based on EDAM4-IS. 40  Results for the national model are qualitatively the same. 83 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI the assumption on richer households being able to apply SIMULATING AN EXPANSION OF PNSF to (and be considered for receiving benefits from) a given Using information from the PMT that allows a differentiation social program using the PMT formula. Restricting the 41 between the urban and rural settings (model 2 without sample to the bottom 55 percent of the population and truncation), we run the following hypothetical scenario of using the truncated model generates errors of exclusion an expansion of the PNSF to illustrate the potential of the of 11 percent regardless of our assumption on who is able application of a PMT model. In particular, we assume the to apply to the program. This implies that the program following: would miss about 1 actual poor person from 10. Errors of u The PNSF is rolled out nationally using the PMT inclusion are between 23–34 percent. model from above to identify and rank (based on If we need to compare between the models presented households’ needs) potential beneficiaries. earlier and the truncated model, it is necessary to make u We use the EDAM2017 data to obtain a PMT sure we select a cut-off point producing approximately score for all households (that is, we assume that all similar program participation rates (and thus the identical households apply to the PNSF program and get a budget). Aiming for a total participation of 32 percent, the score). Households with the most needs (the lowest relevant cutoff for the national/flexible models is the 35th scores) are then ranked first. percentile, whereas for the truncated model it is the 25th percentile. Thus, if we predict errors using these cut-offs, u The PNSF is assumed to expand to 5,000 households we find that exclusion (inclusion) rates under the national nationally. and flexible models are about 25 percent (20–21 percent). Meanwhile, the truncated model (and a cut-off of a 25th u The PNSF provides about US$56 per household per percentile) would lead to an exclusion rate of 16–18 month to the new identified beneficiaries.42 percent and an inclusion rate of 33 percent. Therefore, the u The program is rolled out and effective as of truncated model performs better in terms of targeting poor September 2018. people but has higher leakage. u There are no behavior responses (changes to the The performance indicators associated with the PMT labor decisions or consumption patterns for instance) models presented here indicate that there is no clear as a result of receiving the program. “winner” in terms of providing the least errors across all dimensions. It is therefore crucial for implementing agencies Under these assumptions, new beneficiaries find to incorporate additional considerations based on their themselves with a higher level of consumption per adult mandate to further scrutinize the model on the ground in the equivalent than under a situation before the program context of Djibouti and to decide on the approach to follow expanded. This then leads to an increase in welfare and or whether additional filters are need as “inclusion criteria” decrease in poverty. Table 3.11 presents the results. or “categorical targeting” (for example, female-headed Overall, there is a 0.6 percentage point decrease in households, widowed-headed households, and so forth). poverty in the country as a whole, but with substantial 41 Tests using the VIF ratio were also performed in each mode for multicollinearity. The ratios were 1.62, 1.27, and 1.53 for the national, urban, and rural models, respectively. 42 Figures from 2016 suggest that each PNSF beneficiary household received approximately DF 80,302 per year (US$37.7 per month). There were 3,362 households receiving the program in 2018. There are few households in the EDAM4 that declare currently receiving PNSF benefits. In the simulation, we assume they keep their current benefits. The simulated benefits for new beneficiaries are about DF 120,000 per year per household. CHAPTER 3 84 TABLE 3.11 Simulation Results on Poverty Rate after Hypothetical Expansion of PNSF Ex ante After expansion Djibouti city 13.6% 13.5% Ali-Sabieh 27.2% 25.5% Dikhil 52.9% 49.3% Tadjourah 65.4% 60.9% Obock 40.4% 39.9% Arta 31.6% 31.1% Total 21.1% 20.5% Source: Calculations based on EDAM4-IS. variation across regions. In particular, poverty in Tadjourah u We assume there are no behavior responses (changes decreases by 4.5 percentage points. Poverty in the capital to the labor decisions or consumption patterns for city decreases only slightly. This result may directly follow instance) as a result of receiving the program. from the fact that higher levels of deprivation are found in u The PNSF is randomly assigned among rural regions that are heavily rural and thus under the simulation households based on the percentage of poor get much higher priority under the new PNSF. population living in either rural areas or in the city of We also simulate an alternative scenario assessing the Djibouti. This implies, for instance, that if 20 percent of impact of a hypothetical expansion of the PNSF under a poor households live in Dikhil, then 1,000 (20 percent mixed implementing strategy. In this second simulation of 5,000) households in Dikhil will randomly receive eligibility is based on using both the PMT and geographical the PNSF expansion. However, the PNSF expansion targeting. We simulate this scenario as the logistics for the city of Djibouti will be based on the percentage and administrative costs of running a PMT nationally of poor households living there but selected through may be too high. In light of pervasive poverty in rural the PMT scores generated by the model, where areas, a randomly assigned assistance package in rural the lowest scores are ranked first and assuming all communities coupled with PMT-targeting in the capital city households apply to the program. could potentially reduce costs. Meanwhile, applying a PMT Table 3.12 presents the results from the second simulation. targeting in a zone with high population density such as Under the assumptions and conditions just outlined, the capital, may be cost-effective. beneficiaries of the PNSF expansion have a higher level Using the same PMT model (model 2 without truncation), of consumption per adult equivalent than before the we run the following hypothetical scenario: expansion. This leads to an overall decrease in poverty nationally and regionally. Poverty is reduced by 0.7 u The PNSF is assumed to expand to 5,000 percentage points in the country as a whole. Similar to the households. first simulation, there are substantial differences across u The PNSF provides about US$56 per household per regions. Tadjourah experiences the largest percentage month to the identified beneficiaries. point decrease in poverty (4.1 percentage points) and Obock the highest percentage decrease (9 percent as the result of a 3.6 percentage point drop in poverty). 85 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE 3.12 Simulation Results on Poverty Rate after Alternative Hypothetical Expansion of PNSF Ex ante After expansion Djibouti city 13.6% 13.4% Ali-Sabieh 27.2% 26.1% Dikhil 52.9% 50.5% Tadjourah 65.4% 61.3% Obock 40.4% 36.8% Arta 31.6% 29.1% Total 21.1% 20.4% Source: Calculations based on EDAM4-IS. Compared to the results from a PMT-exclusive targeting, The potential of the PMT models and the results presented the national poverty rate is reduced slightly less under here to inform the design of policies is still high. Despite a mixed strategy, but this difference is not statistically the shortcomings known of PMT models, there is great significant. From a regional perspective, poverty in Djibouti potential for the use of such models to improve targeting city, Ali-Sabieh, Dikhil, and Tadjourah is not reduced in Djibouti, especially as the GoD seeks to expand its as much under the second simulation. In light of these SSN and increases the focus of SSN towards the most findings, and with consideration to costs, we observe that vulnerable.43 Additional analyses could help complement a mix of PMT and geographical targeting shows promising the results presented here. In particular, the next analytical results. However, it is up to the Secrétariat d’Etat chargé steps could be expanded in at least two ways. First, des Affaires Sociales to evaluate the most efficient it would be helpful to conduct a comparison between approach to distributing benefits in rural areas given the the population of current beneficiaries of PNSF and the known high levels of poverty. population that was used in estimating the PMT models. Should the demographic and economic characteristics CONCLUSION be significantly different between the survey population This section note presents a preliminary analysis of the and, say, the pool of households that will be recertified to application of PMT models to the Djiboutian case. The a program, caution should be used in the application of a models use recent data collected in 2017 from the EDAM4 new PMT model to “out-of-sample” populations. Second, and thus help to capture the livelihoods of Djiboutians more analysis should be done to explore the implications as accurately as possible. The results presented here of official poverty rates being expressed in adult equivalent show that PMT models appear to perform well in line with terms and PMT models in per capita terms. The eligibility inclusion and exclusion errors that have been found in definition and criteria should follow the most informative other settings. However, the different applications of the strategy for the government and should be taken into PMT methodology did not allow to find a specification that account for the classifications of vulnerability in the was superior to other tests across all indicators. analytical exercises. 43 A back of the envelope calculation suggests that PNSF, the largest SSN program, reaches only 3 percent of households (in a context where about a fifth of household are considered extreme poor). CHAPTER 3 86 NOMADS AND PASTORALISTS About 20 percent of the Djiboutian population was found under threat. In this section, we explore the current status to be nomadic in 2009, with about 56 percent living in the of nomadism in Djibouti and the vulnerability that nomads regions of Dikhil and Tadjourah, according to the Djibouti may face. We refer to Lara Ibarra, Weiser, and Martinez- census of 2009. Given its large share and the fact that Cruz (2018) for a review of nomadism, climate risk, and the Djibouti has experienced persistent droughts over the links between them. past decade, it becomes important to study the welfare Nomadic people (or nomads) are persons without a fixed of this population, as nomads’ livelihoods may have come place of usual residence who move from one site to another MAP 3.1 Major Livestock Migration Routes in Djibouti Source: FAO 2011. 87 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI in certain seasons or periods, usually according to well- potential of the pastoralists to sustain their activities. It is established patterns of geographical mobility (UN 1998). also argued that climate change may have positive effects at They also usually return to their original location at different local levels, while having negative effects in others (Byg and times in a year. Nomads can be classified into three types: Salick 2009; Martinez-Cruz, Juárez-Torres, and Guerrero hunter-gatherers, whose livelihood depend on wild plants; 2017). Nevertheless, in summary, nomads are more likely to pastoralists, who raise herds of animals and move with be affected negatively from climate change, for instance, in them; and peripatetic nomads, who subsist by practicing a Sub-Saharan Africa (Schlenker and Lobell 2010). trade from place to place. Nomadism is not to be confused Nomadism in Djibouti takes the form of pastoralism, with migration, as the latter entails a long-term or permanent in which individuals raise herds of animals and travel move from one place to another (Tanabe 2015). with them. Map 3.1 highlights the key routes taken by Estimates from a report by the United Nations (UN 2009) nomads in Djibouti circa 2011. The season of Heys/ place the nomadic population in the world at 200 million, Dada rains runs normally from October to February and found in different parts of the world and across continents. the season of Karan/Karma rains runs normally from July Among other reasons, nomadic populations have become to September. As in other parts of the world, nomads in a vulnerable group due to the dramatic changes in weather Djibouti are no exception at being increasingly vulnerable conditions and exacerbated risk from climate change. For to climate change, due their direct dependence on instance, Tao et al. (2015) argue that in recent decades weather conditions to earn income. Figure 3.7 shows climate change has had significant impacts on the lakes in that temperatures in Djibouti are becoming more extreme the Mongolian Plateau, which are the source of livelihood over time. Temperatures in January have been dropping for the Mongolian nomadic population. As is observed in and those in July have been increasing. With the series Djibouti, Sheik-Mohamed and Velema (1999) argue that of droughts in Djibouti in the past decade along with annual rainfall in the Sahara Desert or semiarid areas like temperature changes, the livelihoods of pastoralists have the Sahil and northeastern Africa is expected to become come under a serious threat. even more varied. This is likely to have an effect on the FIGURE 3.7 Average Temperature in Djibouti (Celsius) 50 45 40 35 30 25 20 15 2005 2007 2009 2011 2013 2015 2017 January July Linear (January) Linear (July) Source: World Bank Climate Change Knowledge Portal. http://sdwebx.worldbank.org/climateportal/index.cfm?page=country_histori- cal_climate, accessed October 19, 2018. CHAPTER 3 88 The pastoralist population was covered in the latest survey in the country. Both Tadjourah and Obock have about 3 of EDAM4-IS for the first time in the EDAM history. With percent of the population as pastoralists. Overall, outside this survey information, a simple yet informative profile of Djibouti city, about 6 percent of the population is found of this population can be provided. Figure 3.8 shows a to practice transhumance. snapshot of the population in Djibouti engaged in animal Part of the explanation of this low percentage of rearing and in pastoralism. We find that ownership and pastoralists observed, as compared to the census of rearing of livestock is highly prevalent in Tadjourah and Djibouti in 2009, could be due to successive droughts that Obock. About 55 percent of the population owns and/or Djibouti has been facing over the past decade (see Box rears animals, mostly goats, in these two regions. About 1.1 for more detail). The resilience strategies developed 15 percent own camels, while sheep are more commonly by the nomadic populations against the negative effects owned and/or reared in Ali Sabieh and Arta. Of those of the drought have been (i) to migrate to regions, that own any of the four most common animals (goats, sometimes beyond the country’s borders, or (ii) to settle camels, sheep, or cattle), questions on the practice of around villages or other points, abandoning traditional transhumance were asked in EDAM4-IS. A subset of transhumance in favor of nearby pastures. In fact, in 2009, those who practice transhumance in neighboring countries the GoD adopted the National Strategy for Food and or in further distant areas within Djibouti are defined as Nutrition Security (Décret n°2009-0113/PRE). In volume pastoralists. Figure 3.9 shows that 15 percent of the 1 of this report, it is mentioned that, due to drought, population of Obock practices transhumance, the highest FIGURE 3.8 Percentage of Population Owning and/or Raising Animals 57% 55% 42% 31% 25% 2% Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta Any animal Goats Camels Sheep Cattle Source: Calculations based on EDAM4-IS 89 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI FIGURE 3.9 Percentage of Population Practicing Transhumance and Is Pastoralist 3% 2% 12% 2% 2% 3% 6% 4% 4% 3% 3% 0% 0% 1% Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta Other regions Practices transhumance Pastoralist Source: Calculations based on EDAM4-IS. Notes: The percentage of population practicing transhumance out of all surveyed refers to those practicing transhumance to rear goats, camels, sheep, or cattle in the neighborhood. Pastoralist are defined as those practicing transhumance of goats, camels, sheep, or cattle either in other countries or within the country but outside of the neighborhoods. “nomadic pastoralists are led to move their animals . . . are pastoralists are higher, especially in Ali Sabieh and in the interior of the country, especially to high-altitude Dikhil. When we compare characteristics of the households pastures and to other countries in the region. . . . More that practice transhumance to those that do not in the substantial and more regular humanitarian aid to Ethiopian regions, interesting findings emerge. We find that while pastoralists has helped set up pastoralists in Ethiopia, they live in households of similar size, the dependency whose home lands were traditionally in Djibouti. Therefore, ratio is higher in households that practice transhumance. there is a double hemorrhage of the rural population.” This Table 3.13 shows that the demographic characteristics phenomenon led to the substantial downward revision of households’ heads are similar, but educational of DISED’s estimation of the nomadic population when outcomes of household members vary significantly. Only estimating the framework of EDAM4 (DISED 2018). 7 percent of individuals living in households that practice transhumance are literate as compared to 31 percent in Estimates from consumption expenditure show that other households in regions. The former also have only 0.5 the welfare of this subgroup is much worse than the years of schooling on an average compared to 2.8 years. population as a whole (Figure 3.10). The poverty rates These low levels of education and literacy would make among those who practice transhumance and those who CHAPTER 3 90 FIGURE 3.10 Extreme (Monetary) Poverty Rate among Various Population Groups 93% 73% 65% 67% 58% 50% 53% 49% 43% 38% 30% 26% 14% 0% Djibouti city Ali Sabieh Dikhil Tadjourah Obock Arta Other regions Not practising tranhumance Practice tranhumance Source: Calculations based on EDAM4-IS. Notes: The percentage of population practicing transhumance out of all surveyed refers to those practicing transhumance to rear goats, camels, sheep, or cattle. TABLE 3.13 Characteristics of Households in the Regions That Practice Transhumance and Those That Do Not Households that practice Households that do not practice transhumance transhumance Household size 5.2 5.4 Dependency ratio (%)* 137.1 120.9 Household head characteristics Age** 46.8 45.9 Married 80.8% 82.7% Female 76.0% 74.9% Individual characteristics Literate (15 years and older) 7.1% 30.8% Years of education (15 years and older) 0.55 2.8 Source: Calculations based on EDAM4-IS. Notes: The percentage of population practicing transhumance out of all surveyed refers to those practicing transhumance to rear goats, camels, sheep, or cattle. The sample consists of households in the regions only, out of which 181 households practice transhumance and 2,258 do not. * Only households with at least one working age (15–64) individual are included. ** The sample comprises 4,359 household heads. For the remaining households, the age of the household head is unknown. 91 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI assimilation in the labor force difficult for households that transition to alternative livelihoods, desert tourism was practice transhumance. All these findings point further to promoted among the Tuareg people in Algeria and the the vulnerability of this subgroup of the population and Bedouin people in Jordan. This economic activity did not lend support to the recommendation that this subgroup bring the expected economic gains but has disturbed the requires attention. livelihood of these groups (Chatelard 2005; Hobbs and Tsunemi 2007). It, thus, becomes imperative to protect The report by the United Nations (UN 2009) argues that this vulnerable group, especially in light of the ever- pastoralism has proven to be a very effective livelihood persistent threat of climate change. Equally important, in drylands. The strategies employed by pastoralists to the statistical systems in Djibouti must redouble efforts subsist in these parts of the world have been shown to to observe this population. The upcoming population be sustainable for the environment and been found to census (expected to be conducted in 2020) is an be economically feasible. Pastoralist livelihoods have opportunity that should not be missed to improve our also promoted biological and floral diversity in regions understanding of this important subgroup. It would also where the practitioners travel. In some cases, efforts be worthwhile to investigate resilience strategies that have been made by national agencies and international could be employed by this population as well as the organizations to improve the livelihoods of nomads and policy actions that may facilitate the promotion of these have met with failure. For example, in an effort to promote strategies and alternative livelihoods. CHAPTER 3 92 REFERENCES Aker, J. C. 2010. “Information from Markets Near and Chatelard, G. 2005. Desert Tourism as a Substitute for Far: Mobile Phones and Agricultural Markets in Niger.” Pastoralism? Tuareg in Algeria and Bedouin in Jordan. In American Economic Journal: Applied Economics 2 (3): Nomadic societies in the Middle East and North Africa: 46–59. entering the 21st century. Chatty, D. (Ed.), Brill Academic Publishers, Netherlands, 710-736. Aker, J. C., and I. M. Mbiti. 2010. “Mobile Phones and Economic Development in Africa.” Journal of Economic DISED. 2018. Résultats de la Quatrième Enquête Perspectives 24 (3): 207–32. Djiboutienne Auprès de Ménages pour les Indicateurs Sociaux (EDAM4-IS). Djibouti city: DISED. http://www. Araujo, C. and Carraro, L. n.d. A proxy-Means Test dised.dj/Rapport1_resultats_EDAM4.pdf. Exercise for the Selection of Beneficiaries of Poverty Targeted Programs in Mongolia. Washington, DC: World EC (European Commission) 2002. Eighth Report Bank. from the Commission on the Implementation of the Telecommunications Regulatory Package, 3/12/2002 Araar, A., E. Malasquez, S. Olivieri, and C. Rodriguez- COM(2002) 695 final, SEC(2002) 1329. Brussels: EC. Castelan. 2018. WELCOM User Manual. Washington, DC: World Bank. FAO (UN Food and Agriculture Organization). 2011. Major Livestock Migration Routes in Djibouti. Map. Rome: Italy. Atamanov, A., C. Castaneda, D. Jolliffe, D. Krayem, and H. Sayed. 2015. Updating the Proxy Means Testing Formula Gelvanovska, N., M. Rogy, and C. M. Rossotto. 2014. in Lebanon. Technical report. Washington, DC: World Broadband Networks in the Middle East and North Africa: Bank. Accelerating High-Speed Internet Access. Washington, DC: World Bank. Beuermann, D. W., C. McKelvey, and R. Vakis. 2012. “Mobile Phones and Economic Development in Rural Genakos, Valletti, Verboven. 2015. Evaluating Market Peru.” Journal of Development Studies 48 (11): 1617–28. Consolidation in Mobile Communications. Brussels: Centre on Regulation in Europe (CERRE). <> “Regulation, Market Structure, and Performance in Telecommunications.” OECD Economic Studies, 2001(1), Grosh, M., and J. Baker. 1995. “Proxy Means Tests for 99-142. Targeting Social Programs: Simulations and Speculation.” Living Standards Measurement Study Working Paper 118. Brown, C., M. Ravallion, and D. Van de Walle. 2018. A World Bank: Washington, DC. Poor Means Test? Econometric Targeting in Africa. Journal of Development Economics 134 (September): 109–24. GSMA (Global System Mobile Association). 2018. The Mobile Economy Middle East & North Africa. Byg, A., and J. Salick. 2009. “Local Perspectives on a Global Phenomenon: Climate Change in Eastern Tibetan Villages.” Global Environmental Change 19 (2): 156–66. 93 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Hobbs, J. J., and F. Tsunemi. 2007. “Soft Sedentarization: Sdralevich, M. C. A., M. R. Sab, M. Y. Zouhar, and G. Bedouin Tourist Stations as a Response to Drought in Albertin. 2014. Subsidy Reform in the Middle East and Egypt’s Eastern Desert.” Human Ecology 35 (2): 209–22. North Africa: Recent Progress and Challenges Ahead. Washington, DC: International Monetary Fund. IEA (International Energy Agency). 2012. World Energy Outlook. Paris: IEA. Sheik-Mohamed, A., and J. P. Velema. 1999. “Where Health Care Has no Access: The Nomadic Populations of ITU (United Nations International Telecommunications Sub-Saharan Africa.” Tropical Medicine and International Union). 2017. “ICT Development Index 2017.” Health 4 (10): 695–707. Geneva: ITU. http://www.itu.int/net4/ITU-D/ idi/2016/#idi2016countrycard-tab&DJI. Tanabe, Jennifer. 2015. “Nomad.” New World Encyclopedia. St. Paul, MN: Paragon House. ITU. 2018. Measuring the Information Society Report. Vol. http://www.newworldencyclopedia.org/p/index. I. Geneva: ITU. php?title=Nomad&oldid=989416. Lara Ibarra, Gabriel, Christina Weiser, and Adan Martinez- Tao, S., J. Fang, X. Zhao, S. Zhao, H. Shen, H. Hu, Z. Cruz. 2018 “Estimating the Value of Housing Services Tang, Z. Wang, and Q. Guo. 2015. “Rapid Loss of Lakes of Nomadic Populations.” Photocopy. Washington, DC: on the Mongolian Plateau.” Proceedings of the National World Bank. Academy of Sciences 112 (7): 2281–2286. Martínez-Cruz, A. L., M. Juárez-Torres, and S. Guerrero. UN (United Nations). 1998. “Recommendations on 2017. “Assessing Impacts from Climate Change on local Statistics of International Migration,” revision 1. Statistical Social-Ecological Systems in Contexts Where Information Papers Series M, 58. New York: United Nations. Is Lacking: An Expert Elicitation in the Bolivian Altiplano.” https://unstats.un.org/unsd/publication/SeriesM/ Ecological Economics (137): 70–82. SeriesM_58rev1e.pdf. Narayan, A., and N. Yoshida. 2005. “Proxy Means Test for UN. 2009. State of the World’s Indigenous Peoples. Targeting Welfare Benefits in Sri Lanka.” South Asia Region New York: UN. https://www.un.org/development/desa/ PREM Working Paper Series SASPR-7. World Bank: indigenouspeoples/publications/state-of-the-worlds- Washington, DC. indigenous-peoples.html. Ofcom. 2016. A Cross-Country Econometric Analysis of UNDESA (United Nations Department of Economic and the Effect of Disruptive Firms on Mobile Pricing. London: Social Affairs). 2016. 2016 E-Government for Sustainable Ofcom. Development. New York: UN. Schlenker, W., and D. B. Lobell. 2010. “Robust Negative World Bank. 2015. Telecommunications Sector in Djibouti: Impacts of Climate Change on African Agriculture.” Finding the Path to Growth. Washington, DC: World Bank. Environmental Research Letters 5 (1): 014010. World Bank. 2016. World Development Report ‘Digital Dividends’. Washington, DC: World Bank. CHAPTER 3 94 NOTE ON EDAM4-IS APPENDIX A 2017 AND SAMPLING This appendix presents a more detailed description of ended before the Ramadan period. The second phase the Fourth Djibouti Household Survey for the 2017 Social of the data collection was implemented in November- Indicators. The EDAM4-IS survey was deployed in two 44 December 2017. phases. The first was conducted in April-May 2017 and THE DJIBOUTI HOUSEHOLD SURVEY (EDAM4-IS) The 2017 EDAM questionnaire covered the following topics: The EDAM4-IS questionnaire is the result of an adaptation of the traditional EDAM questionnaires with the aim of 1. Characteristics of household members (demographics deepening the measurement and analysis of poverty and and nationality) well-being. Compared to earlier surveys, the revisions 2. Education that were incorporated into the EDAM4-IS questionnaire 3. Health include the following: 4. Employment 5. Characteristics of housing 1. The collection of a list of 100 food items that account 6. Possession of assets and durables for the majority of the products in the Djibouti 7. Food expenses (consumed at home and meals taken household consumption basket outside) 2. Information on transfers from public and private 8. Nonfood expenditure sources in more detail 9. Sources of revenue (private and public transfers) 3. Information on the purchase and present value of durable 10. Shocks and survival mechanisms goods collected to account for the flow of services 11. Perceptions of poverty 4. Specific modules for income from economic activities 12. Governance 5. Information on nationality and migration (following 13. Access to services recommendations from International Organization for 14. Income from farming activities Migration) 6. Information on education, health, and housing expenditures better placed in the questionnaire to improve collection of data (a comprehensive module on water and sanitation services was integrated following the direct request of the National Office of Water and Sanitation of Djibouti, or ONEAD).  he information presented here draws extensively from DISED (2018). Additional information is available in DISED’s microdata 44  T catalog. http://www.dised.dj/djibnada/index.php/catalog/12. 95 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI By comparing the collection of data on food consumption, To better capture the welfare derived from durable goods, and using lessons learned from international experience, information was collected on assets owned by the a recall period of consumer goods expenditure of seven household, their values at the time of purchase, the length days was applied. Also, expenses on purchases and food of time have been owned, and the estimation of their donations consumed outside the home were captured for current value. all household members age five years and older. Specific Last, unlike the previous EDAM and EBC surveys, the questions on the consumption of tobacco and other similar coverage of EDAM4-IS 2017 was extended to nomadic products were also included in the questionnaire. households. SAMPLING AND EXTRAPOLATION COEFFICIENTS OF THE EDAM4-IS The main database and maps for the compilation of A stratified two-stage sampling model was used for the EDAM4 frame are those of the General Census of EDAM4. In the first step, EA samples were systematically Population and Housing of Djibouti 2009. The census selected with size-proportional probabilities within each enumeration areas (EAs) or primary sampling units (PSUs) stratum. Fifteen survey strata have been defined: the five sampling frame is stratified by region, urban area, and rural districts of Djibouti city, and each of the five regions of the area for most national surveys. Although the urban and interior of the country divided in urban and rural areas. rural areas of each region are individual sampling strata to Before proceeding to draw sample households at the improve the efficiency of the sample design, the total results second stage, an update of the household listings of the for other urban and rural areas is limited at the national level. sample EAs was carried out in April 2017. At the second TABLE A.1 Enumeration Areas (EAs) Selected in the Sampling Total Urban Rural Region Total EAs Total EAs Selected EAs Total EAs Selected EAs Djibouti city District 1 74 74 42 - - District 2 120 120 42 - - District 3 38 38 38 - - District 4 156 156 42 - - District 5 142 142 42 - - Ali Sabieh region 46 27 16 19 18 Dikhil region 64 20 9 44 25 Tadjourah region 57 12 4 45 30 Obock region 45 8 8 37 26 Arta region 51 10 8 41 26 Totals 793 607 251 186 125 APPENDIX A 96 stage, a systematic random sample of 12 households The remaining 122 sample EAs for Djibouti city were was selected in each EA sample (or cluster) for Djibouti surveyed during the second phase, from October to city and 15 households per EA for other regions. In the November 2017, to provide reliable district-level survey case of rural samples, the sample of 15 households was results from the combined data for both phases. to include both sedentary and nomadic households. Given The final size of the sample is 4,474 households the large percentage of nomadic households in rural areas distributed as given in table A.2. according to the published results of the 2009 census (more than 50 percent), this sampling procedure had to SAMPLING WEIGHTS provide a sufficient sample of nomadic households for analysis at the national level. In total, the sample of EDAM4 For sample estimates from the 2017 EDAM4 data to be includes 376 EAs (251 in urban areas, 125 in rural areas). representative of the population, it will be necessary to The table A.1 shows EDAM4’s EAs distribution. multiply the data by a sampling weight or an expansion factor. The base weight for each sample household is DATA COLLECTION equal to the inverse of its selection probability (calculated by multiplying the probabilities at each sampling stage). Due to the limited time available for data collection prior to Ramadan, which began on May 27, 2017, it was decided Based on the two-stage stratified sample design, the to cover the sample for Djibouti city in two phases. overall probability of selection for the 2017 EDAM sample households can be expressed as follows: For the first phase, before Ramadan, a subsample of 84 of 206 EAs for Djibouti city was selected to obtain reliable results for this area. For the other regions, all sample groups were surveyed in the first phase. The 84 EA where: samples in Djibouti city for the first phase were allocated Phi =  overall sampling probability for households selected to the districts in proportion to the number of households. for EDAM4 in the i-th group of samples in stratum h Data collection for the first phase ran from April 27 to May 26, 2017. nh =  number of sample clusters selected in stratum h for EDAM4 TABLE A.2 Sample Distribution by Region Region Households interviewed Djibouti city 2,035 Ali Sabieh 495 Dikhil 496 Tadjourah 493 Obock 475 Arta 480 Djibouti country 4,474 Source: Calculations based on EDAM4-IS 97 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI Mhi =  total number of households in the i-th sample group DOES THE EDAM REPRESENT THE in stratum h of the sampling frame based on the ACTUAL DJIBOUTIAN POPULATION? 2009 Djibouti census (partially updated) Given the implementation of the EDAM4, an important Mh =  total number of households in stratum h of the implication is that there may exist a difference between frame the population that is represented in the survey and the overall population in the country. Current estimates of mhi =  12 or 15 = number of sample households selected the Djiboutian population are about 1 million but include for EDAM4 in the i-th group of samples in stratum h subgroups of the population such as homeless and refugees that are not part of the survey. Moreover, they M'hi =  total number of households in the new list for the are based on extrapolations of the distribution of the i-th group of samples in stratum h population from the 2009 Census that may overestimate The base weight for sample households of EDAM4 is the evolution of the nomadic population. This difference the inverse of this probability of selection, expressed as may raise concerns on the ability of the survey to be follows: representative of the country’s population. Table A.3 presents a selection of indicators that may help Whi =  base weight for sample households EDAM4 in the allay these concerns. The survey results are compared i-th group of samples in stratum h with results using other sources, specifically administrative data on schooling from the Ministry of National Education After the collection of EDAM4 data, the base weights were and Professional Training in Djibouti (MENFOP) and adjusted to account for noninterviews, as follows: Enquete Djiboutienne sur l’emploi, le secteur informel et la consummation, or EDESIC 2015 (DISED 2015). The indicators on the percentage of students enrolled in various Where: years of primary school is strikingly similar in EDAM4-IS and official data from MENFOP in 2016–17. Using indicators W'hi =  adjusted weight for sample households EDAM4 in on education and access to services, we see that survey the i-th group of samples in stratum h results from EDAM4-IS are similar to those obtained from m'hi =  number of sampled households completing EDESIC 2015, published by DISED. This validation exercise interviews in the i-th group of samples in stratum h provides confidence in terms of the sampling methodology and the representation of the survey. APPENDIX A 98 TABLE A.3 Comparison of Selected Indicators between EDAM4 and DISED Published Data EDAM4-IS Other sources Percentage of students enrolled in: 1st year of school 20 19 2nd year of school 18 19 3rd year of school 18 19 4th year of school 19 18 5th year of school 25 25 Percentage of students in public primary school who are girls 46 47 Access to water 90 86 Access to electricity 60 58 Open defecation 12 14 Gross enrolment ratio, primary (children ages 6–10 years) 96 94 Literacy rate among men (15 years and older) 69 67 Literacy rate among women (15 years and older) 49 53 Median age 22 20 Source: Calculations based on EDAM4-IS, and DISED 2017. Note: Access to electricity reflects the use of electricity as the main source of lighting. Access to water reflects the availability of water in running water (ONEAD indoor connection), direct connection from a borehole, ONEAD outdoor connection by pipe, public fountain and borehole (with a pump). 99 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI REFERENCES DISED (Direction de la Statistiques et des Etudes Démographiques). 2015. Enquete Djiboutienne sur l’emploi, le secteur informel et la consummation, 2015. Djibouti: DISED. DISED. 2017. Annuaire Statistique 2017. Djibouti: DISED. DISED. 2018. Résultats de la Quatrième Enquête Djiboutienne Auprès de Ménages pour les Indicateurs Sociaux (EDAM4-IS). Djibouti city: DISED. http://www. dised.dj/Rapport1_resultats_EDAM4.pdf. APPENDIX A 100 NOTE ON POVERTY MEASUREMENT APPENDIX B METHODOLOGY The calculation of poverty rate and other associated representing the minimum welfare level below which a measures entails the construction of a consumption person is deemed to be poor. The methodology used for aggregate as a measure of welfare and a poverty line each of these is discussed in turn. THE AGGREGATE OF CONSUMPTION AS A MEASURE OF WELL-BEING The measure of standard of living used depends either on to obtain the value of consumption that will be used as the the consumption or income of a household. The choice basis for determining poverty rates are presented. between consumption and income depends on many factors, such as the availability of data, the design of the EXPENDITURES ON FOOD PRODUCTS survey, as well as the context of the country. For example, The value of consumption expenditure incurred for each in the case of a country like Djibouti, the recording of food in the preceding seven days has been reported by income is difficult because of the underreporting as well as households in the EDAM4, 2017 and is directly used in the presence of an informal sector of employment. Thus, the consumption aggregate. The survey includes a very consumption is considered as a better approximation of detailed list of 100 foods divided into 12 groups. Food well-being. Subsequently, temporal adjustments must be consumption data include the value and quantity of food made to household consumption to ensure that poverty purchased as well as the amount of food produced at measures are comparable across different data collection home and foods that may have been acquired through periods. The size and composition of the household also nonmonetary transactions such as donations, barter or play an important role and need to be adjusted. trade, and so forth. These three modes of acquisition are used to construct a measure of household food In defining the consumption aggregate, its four groups consumption expenditure. of components are considered (Deaton and Zaidi 2002): (i) expenditures on food products, (ii) nonfood Another section of the questionnaire focused on food expenditures, (iii) consumer durables, and (iv) rent and consumed outside the home by each household member housing. In the sections that follow, estimation of each over the age of five in the preceding seven days. This of these components is discussed. Aggregating these includes all meals taken outside the house, such as at four components will determine nominal household restaurants. This section also recorded the value of the expenditures. Subsequently, decisions on price deflators, meals received as a gift. The aggregation of data from the time deflator, and household composition adjustments these two components provides us with a measure of 101 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI total food consumption expenditure. This expense is then spaghetti and sugar. The average expenditures on each of converted into an annual expense. these 10 food products is also presented. Not surprisingly, average expenditure on mutton is the highest among Figure B.1 shows that the top 10 foods that are consumed these products. Table B.1 disaggregates food expenditure by the Djiboutian population. About 90 percent of the by food-item group and by quintiles of the per capita population consumes imported rice, followed closely by FIGURE B.1 Percentage of Population Consuming and Mean Annual Expenditure (DF) by Food Product 45,000 100% 87% 90% 40,000 84% 87% 90% 77% 80% 35,000 70% 73% 30,000 70% 54% 60% 25,000 50% 20,000 31% 40% 15,000 28% 30% 10,000 20% 5,000 10% - 0% Wheat Couscou Mutton Potatoes Onions Tomatoes Macaroni Sugar Spaghetti Imported our- meat rice tapioca/ gari Mean annual expenditure (DF) Percentage consumed Source: Calculations based on EDAM4-IS. TABLE B.1 Shares of Food Group to Total Food Consumption Expenditure by Quintile Poorest Richest 1 2 3 4 5 Total Cereals 30% 18% 14% 12% 10% 17% Pasta and bread 18% 20% 19% 17% 12% 17% Meat 5% 11% 14% 16% 16% 12% Fish 1% 2% 3% 3% 6% 3% Milk products and eggs 3% 6% 7% 8% 11% 7% Oil and butter 8% 6% 6% 5% 5% 6% Fruits 1% 3% 3% 4% 6% 3% Vegetables 11% 14% 14% 13% 12% 13% Canned vegetables 2% 2% 2% 2% 2% 2% Tubers and plantains 3% 4% 4% 4% 3% 4% Sugar and condiments 8% 6% 6% 5% 5% 6% Others 12% 10% 9% 10% 12% 11% Total 100% 100% 100% 100% 100% 100% Source: Calculations based on EDAM4-IS. APPENDIX B 102 distribution of food expenditure. As expected, the share of nutrient-rich food such as meat, milk products, and fruits. staple foods (bread and cereals) to total food expenditure In table B.2, various characteristics of the food basket of declines the higher the overall value of food consumption, the Djiboutian population are presented. and it is matched by an opposite increase in relatively more TABLE B.2 Djibouti’s Food Basket Daily expenditure Percentage Daily expenditure per adult consuming (DF) equivalent (DF) Unit cost (DF) Cereals 7% 57 13 n.a. Pasta and bread 17% 87 18 n.a. Meat 7% 157 36 n.a. Fish 6% 103 25 n.a. Milk products 9% 150 36 n.a. Eggs 7% 53 13 n.a. Oil and butter 16% 64 15 n.a. Fruits 12% 48 11 n.a. Vegetables 20% 28 7 n.a. Canned vegetables 10% 69 17 n.a. Tubers and plantains 0% 67 21 n.a. Sugar and condiments 4% 58 14 n.a. Others 21% 29 7 n.a. Imported rice 90% 70 16 161 Couscous 31% 56 12 166 Wheat flour: tapioca 28% 69 17 130 Macaroni 84% 59 13 191 Spaghetti 87% 58 13 188 Mutton meat 54% 194 45 1095 Onions 73% 67 15 176 Tomatoes 77% 69 15 233 Potatoes 70% 56 12 130 Sugar 87% 58 13 162 Source: Calculations based on EDAM4-IS. Notes: n.a. = not applicable. 103 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI EXPENDITURE ON NONFOOD expenditures relative to total expenditures. The higher PRODUCTS the elasticity, the higher the case for its inclusion in the consumption aggregate. Total consumption expenditures from nonfood items are calculated by aggregating expenditures for various Using the information from the health section, the following goods and services. These expenses are recorded in regression is performed: different sections of the questionnaire; the reference (1) period, therefore, differs. Major groups of goods and services include fuel, vehicle use, transportation and where healthexp includes all per capita household communication, games, books and recreational services, expenses related to health and hhexp includes all per personal care products and services, home repairs, capita household expenses except health. The estimated β appliances, clothing and footwear, textiles, glassware, is 1.46. Thus, it is considered appropriate to include health tools and nondurable goods, music equipment, video and expenditures in the aggregate of well-being. computer (excluding durable goods), jewelry, watches and personal items, tourism, insurance and other services, Deaton and Zaidi’s (2002) recommendation on education energy expenditures and public services, education spending is also noted: education spending could be expenditures, and health expenditures. Each of these considered an investment and not necessarily an increase expenses is annualized using a factor corresponding to the in well-being. An estimation similar to (1) for education recall period. For example, the previous month’s expenses expenditures is done. The estimate found is 1.34, and it is are multiplied by 12. concluded that these expenditures should also be included in the calculation of the consumption aggregate.45 Finally, expenditures for tobacco products, drugs, and alcohol consumed in the preceding seven days for each According to good practices in measuring poverty, household member older than 12 years are also added to household consumption expenditure excludes irregular the nonfood expenditure aggregate. These products include expenditures. These include spending on celebrations cigarettes, cigars, shisha, chewing tobacco products, such as Eid, weddings, and other ceremonies. These alcoholic beverages, and other drugs. The quantity of expenses are incurred a few times or once in a lifetime; these products received in donations, exchanges, and so therefore, including them in the year in which they were forth is also captured in the questionnaire and included in made would lead to underestimation of poverty and the consumption aggregate after obtaining an estimate overestimation of inequality. Charitable donations and of the unit value from the information on individuals’ assistance (cash or in-kind) to family and friends were also purchases. These expenses are then converted into annual excluded because they do not contribute directly to the expenditures using a factor of 52 (weeks). well-being of the household. Health expenditures are often irregular expenditures, DURABLE GOODS and a choice must be made about their inclusion in the The aggregate of well-being used in the previous EDAMs aggregate of well-being. On the one hand, some types of (2002, 2012, and 2013) included the cost of acquisition health expenditures improve the well-being of individuals of durable goods. In other words, the total purchase value and are discretionary. However, other types of health of durable goods acquired in the preceding year has been expenditure (for example, for the sick) could be a necessity included in the consumption aggregate. However, it can and necessarily linked to a loss of well-being. Deaton and be argued that such an expenditure is an investment Zaidi (2002) recommend calculating the elasticity of health and not a consumption, as households are likely to 45 If we exclude health expenditures from the welfare aggregate, the poverty rates remain largely unaffected. (The extreme poverty rate is 21 percent and the overall poverty rate is 35.7 percent). APPENDIX B 104 benefit from these purchases over a long period of time. The standard approach is to impute the cost of use as Therefore, including them as consumption in the reference follows: period can lead to an overestimation of well-being and ) / (1- (2) bias the inequality indices obtained from the distribution of consumption. In line with current good practices, where information on the purchase value, the present value, and the number of years of possession of a durable goods is = purchase price of durable good, used to include the “flow of service” of durable goods as a = nominal interest rate, component of household welfare. = inflation rate, and The flow of service, or “use value,” is the value of services provided by durable goods to households over a period of = depreciation rate. time. A standard approach involves estimating the “user A detailed inventory of 45 durable goods was obtained cost” of durable goods. The purchase price reflects the value from the asset possession section of the questionnaire. of the durable good over its entire life, but it is preferred to The final list of durable goods included in the welfare capture the flow of services provided by the durable good aggregate came from those in which at least 60 during the reference period. The user cost has potentially households provided information on the purchase value, two components: (i) the opportunity cost of the locked-in the present value, and the number of years the household funds in the durable good (captured by the interest rate) and had owned the good. These criteria allowed enough (ii) the temporal depreciation (in rare cases, appreciation) variation to obtain a reliable depreciation rate for each of the good itself. The depreciation rate is based on the asset. Table B.3 shows the durable goods included in date and cost of acquisition of the good, combined with the final calculation and the estimated corresponding assumptions about the current value of the item. TABLE B.3 Durable Goods Included in the Consumption Aggregate Name of article Depreciation rate Satellite dish/ digital -8.2% Wardrobe/ dresser -5.9% Air conditioner -6.8% Iron -10.9% Bed (bed + mattress) or mattress -8.3% Washing machine -5.9% Mixer -7.6% PC/ laptop* -15.0% Radio -19.4% Gas stove (kerosene) -10.5% Refrigerator / Freezer -5.9% Living room furniture (armchairs and coffee table) -8.3% Mobile phone* -20.0% Television -6.8% Source: Calculations based on EDAM4-IS. Notes: * The depreciation rates for these items were taken from United Policy Holders website. 105 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI depreciation rate. It is worth noting that for most goods, his or her personal attachment to it. Thus, a hedonistic the information provided by households has closely semilogarithmic model to estimate a rental value for owners followed the depreciation rates used by some insurance and in-kind occupants was used. The approach used to companies. There are two items for which households estimate the hedonic model is as follows. tended to underestimate the depreciation rate: computers Information on rents paid by a subset of the population, and mobile phones. This is understandable, given the namely, tenants as well as self-reported rent by owners relatively recent availability of these items. Thus, the rates and in-kind dwellers, is used. Following Freeman (1993), of these goods for this exercise have been replaced by it is argued that the rent paid by a tenant (ln (rent)) can be those found on the United Policy Holders website. expressed in terms of its characteristics (Xh), neighborhood To estimate equation (2), data from the Red Sea Bank of characteristics (Nh), and location (Lh). The following formula Africa site was obtained, in which the nominal annual return is adopted: for a savings account was 1.59 percent. Finally, the inflation (3) used for the calculation was -0.58 percent. This is based on a deflation rate calculated between 2016 and 2017. The variable on the left is the log of the rent paid by households living in rented dwellings, the self-reported RENT AND HOUSING SERVICES rent by occupants of the service dwellings and the Among the components of measures of well-being, housing owners occupying their homes. This analysis is performed services are a key element. When consumption is used as separately for Djibouti city and other regions. The housing a measure of well-being, as is the case in most developing characteristics, such as type of housing, type (or materials) countries, housing services should appropriately capture the of floor, type of wall, type of ceiling, source of lighting, utility generated by the consumption of housing amenities water source, type of toilet, number of rooms, services to (Deaton and Grosh 2000). This estimate of the flow of dispose of wastewater, garbage system, and a sewage services received by households must be comparable disposal system are controlled for. Finally, some household between households. If two households with the same characteristics such as size (and square), employment number of members and the same age structure reside in a status, type of job, and type of business where the head house with similar characteristics, both should be measured of household is employed; and ε is the residual term are as having the same flow of housing services. also included in the regression. The estimated coefficients are then applied to owner-occupied and in-kind dwellers’ A natural choice to capture the value of the service market housing characteristics to produce a predicted value of their of any home could be the rent paid by its tenant. For implicit rent:46 homeowners, or households who received their housing for free (as an in-kind benefit) the rent is unknown. In EDAM4- (4) IS, households living in owner-occupied dwellings are encouraged to self-report the rental price of the dwelling if To obtain the final imputed rent, a Smearing correction the household were to rent it. However, this self-declared (Duan 1983) based on the average residue of the rent may be an imperfect indicator of the real rental value regression is performed. As a sensitivity analysis, the of housing. Owners or in-kind occupants may not be fully rent at the national level using the fixed effects of the aware of the rental market conditions where they reside. region is estimated. The imputed final values did not differ Their responses may also reflect what is known as “owner qualitatively (extreme poverty rate changed to 21.3 percent pride,” the tendency to overstate the rental value of the and overall poverty rate changes to 35.9 percent). The residence that the owner currently occupies because of results of the hedonic model are provided in table B.4. 46  Rent was found to be missing for 27 households and thus was imputed as the mean rent of the PSU. APPENDIX B 106 TABLE B.4 Results of the Estimation of Hedonic Model (1) (2) Variable Djibouti city Other regions Occupancy status—Owner 0.134*** -0.039 Occupancy status—Free housing -0.072 -0.506*** -0.08 -0.082 Type of accommodation—apartment in a building 0.815*** 0.713*** -0.108 -0.259 Type of accommodation—Single villa 0.278 -0.429* -0.23 -0.222 Type of accommodation—Villa with floor (duplex) -0.407** -0.373*** -0.176 -0.109 Household size -0.037* 0.017* -0.022 -0.01 More than two rooms 0.234*** 0.557*** -0.042 -0.062 Building materials—Wood/board -0.241* -0.323*** -0.139 -0.099 Roof materials—Wood -0.233** -0.101 Roof materials—Beton 0.341*** -0.429* -0.118 -0.253 Roof materials—Straw 0.283** -0.114 -0.134 -0.125 Lighting source—Lamp oil (kerosene) -0.086 -0.400*** -0.062 -0.087 Lighting source—Generator -0.533*** -0.375*** -0.064 -0.064 Source of water—Direct connection from a borehole -0.328*** -0.109 Source of water—ONEAD external hose connector -0.306*** -0.437*** -0.043 -0.074 Source of water—Public fountain -0.385*** -0.480*** -0.14 -0.075 Source of water—Drilling (then with pump) -0.583*** -0.096 Source of water—Wells without pump -0.394 -0.875*** -0.258 -0.102 Source of water—Traditional wells -0.963 -0.377** -0.73 -0.156 Toilet—Toilet without flush -0.101** -0.065 -0.05 -0.075 Toilet—Single latrine (single pit with concrete slab or local materials and hole) -0.343*** 0.272*** -0.052 -0.075 107 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI (1) (2) Variable Djibouti city Other regions Toilet—Hole in the ground with rudimentary fence -1.374*** 0.291*** -0.075 -0.084 Toilet—open -1.282*** -0.148 Garbage disposal—Dropped in a special place -0.067 Garbage disposal—Burn -0.260*** -0.099 -0.062 -0.069 Wastewater disposal—Covered wells or pit -0.332*** -0.156 -0.06 -0.124 Wastewater disposal—On the street or in nature -0.276** -0.118 Region dummy—Ali Sabbieh Region dummy—Dikhil 0.363*** -0.069 Region dummy—Tadjourah 0.251** -0.101 Region dummy—Obock 0.361*** -0.081 Region dummy—Arta 0.379*** -0.082 Employment of the head—Salaried senior executive, engineer and related 0.456*** 0.548** -0.152 -0.232 Employment of the head—Medium executive officer, officer 0.271** -0.109 Employment of head—(Employee) other employee/(employee) worker 0.178*** 0.103 -0.046 -0.064 Employment of head— (Employee) laborer -0.158 0.178 -0.109 -0.142 Unemployment of head of household 0.085 0.051 -0.054 -0.063 Chef’s employment sector—NGO/International Organization/collectives/ others 0.229*** 0.168* -0.079 -0.095 Constant 3.602*** 2.517*** -0.082 -0.12 Observations 1,764 1,081 R-squared 0.57 0.675 Source: Calculations using EDAM4-IS. Notes: Standard deviation shown in parentheses. *** p <0.01, ** p <0.05, * p <0.1. The results are displayed for the selected variables. Other variables included in the regression are the type of dwelling, the source of water, the em- ployment of the head of household, the sector of employment of the head, the garbage disposal, the square of the household size. APPENDIX B 108 There is a consensus that the details of the approach used 8 percent), creating a distribution of rents that is based in estimating a hedonic model depends on the context of on a few observed characteristics. Thus, this approach the country and the data available. The hedonic regression is not followed. A second alternative is to estimate the will vary depending on the groups of observations above hedonic model using only self-reported rent (renters, owners, and so forth) that end up included in information for owners and in-kind dwellers (and exclude the estimation. Thus, due to the reasons mentioned, rent rent reported by tenants). Sensitivity analysis based on declared by tenants and self-reported rent by owners may this approach is conducted and did not find any significant yield very different answers due to owner’s pride and other changes to measures of well-being or poverty rates (the reasons. Three alternatives to explore the robustness of extreme poverty rate changed to 19.7 percent and the our chosen approach (and its implications for welfare and overall poverty rate changed to 34.4 percent). In the third poverty measurement) are considered. approach, the estimated coefficients from equation 3 are applied to not only owner-occupied and in-kind dwellers’ First, one could estimate the hedonic model from equations but also tenant’s housing characteristics to produce a 3 and 4 using reported rent information for tenants only. predicted value of rent. Using this approach, no significant However, given the low number of tenants in Djibouti (only changes to measures of well-being or poverty rates are 19 percent of all dwellings are rented nationally, and the found (the extreme poverty rate changed to 20.6 percent numbers are even lower in the regions at an average of and the overall poverty rate change to 35.6 percent). ANALYSIS OF THE CONSUMPTION AGGREGATE The consumption aggregate is finally obtained by the total tends to decline for quintiles higher up in the aggregating food expenditure and nonfood expenditures, total expenditure distribution. The share of expenditure together with user cost of consumer durables and housing. on housing is high across all quintiles of per capita expenditure. As it is further discussed in the section on The relative importance of each of these classes as it defining the poverty line, variations in the composition of varies by quintiles of the per capita total expenditure the consumption aggregate along the distribution of total distribution is show in table B.5 In general, as one would expenditure imply that the composition of the consumption expect from Engel’s law,47 the share of food items to TABLE B.5 Share of Consumption Aggregate Components, by Quintile Quintile Food Nonfood Housing Durables Poorest 1 53.9% 19.7% 26.3% 0.2% 2 46.1% 25.1% 28.3% 0.4% 3 41.1% 27.1% 31.1% 0.7% 4 39.6% 28.4% 31.1% 0.8% Richest 5 38.1% 30.8% 29.9% 1.1% Source: Calculations based on EDAM4-IS. Note: Quintiles are calculated based on per capita consumption 47 Engel’s law is an empirical observation in economics that states as income rises, the proportion of income spent on food falls, even if the level of expenditure on food rises. 109 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI bundle defining basic needs should be estimated on a easy to communicate and understand. The disadvantage subsample of “relatively poor” households. is that such a correction may be inappropriate because the preferences of household members are not homogeneous HOUSEHOLD COMPOSITION and there are economies of scale associated with certain For each household, food, nonfood, services of durable goods and services. goods, and housing flows are aggregated to arrive at annual In the report “Measuring Poverty and Inequality” (DISED consumption expenditures. Since consumption varies within 2014), for the purposes of adjustment of household the household according to demographic composition, composition, an equivalent scale is used. The construction it is important to define a measure of well-being for an of an equivalence scale is theoretically the most appropriate individual household member. One of the simplest ways to method for moving from family welfare to individual well- do this is to divide consumption expenditure by household being. The Food and Agriculture Organization equivalence size and express the aggregate well-being in per capita scale (also used by Afristat) is applied, which assigns weight terms. The main advantage of this approach is that it is to each household member by age group and sex (table B.6). TABLE B.6 Scales for an Equivalent Adult Males Females 0–1 years 0.27 0.27 1–3 years 0.45 0.45 4–6 years 0.61 0.61 7– 9 years 0.73 0.73 10–12 years 0.86 0.73 13–15 years 0.96 0.83 16–19 years 1.02 0.77 20–50 years 1.00 0.77 51 years and more 0.86 0.79 Source: DISED 2014. TEMPORAL AND SPATIAL (corresponding to deflation rate observed during this period ADJUSTMENT of -1.02 percent and -0.71 percent, respectively). Intertemporal price adjustments account for differences in Another factor to consider is the difference in the cost of the prices of goods that form the aggregate of well-being living across space. One way to account for difference in over time. As usual, for the intertemporal adjustment of the cost of living is to perform a spatial price adjustment. prices, inflation rates constructed from the consumer Deaton and Zaidi (2002) recommend using the Paasche price index are used. More precisely, the values obtained price index at the household level for spatial price from the second phase in November 2018 and in adjustment. Using this approach, it is found that the poverty December 2017 are adjusted with a factor of 0.9898 and rate in Djibouti city declines from 13.6 percent to 11.7 0.9992, respectively, to make the real values comparable percent, causing the poverty for Djibouti as a country to to the values captured in the first phase in May 2017 decline from 21.1 percent to 19.8 percent. APPENDIX B 110 ESTIMATION OF POVERTY LINE The poverty line is estimated to represent, according to ESTIMATION OF THE FOOD the norms of a given society, the cost for a household POVERTY LINE to achieve a level of well-being considered the minimum The estimation of absolute poverty thresholds is typically necessary and which allows it to satisfy its food and based primarily on the assessment of a cost of food nonfood needs. The data from EDAM4 were used to energy requirements. The food threshold is defined as the determine—from the level of consumption (per adult product of the minimum caloric intake that a basket of food equivalent) considered the minimum necessary and whether consumption must guarantee by the unit cost of acquiring a a household’s consumption expenditures are below this kilocalorie. It is, thus, given by the following equation: threshold—if the household can be deemed poor. Although a similar exercise was conducted in 2014 by DISED and (1) African Development Bank, the improvements incorporated into the EDAM4 questionnaire change the methodology where C is the minimum daily input required for the dietary for calculating household consumption. The availability of needs of an active adult (2,115 kilocalories), COST is the a recent consumption basket of Djiboutian households average unit cost of a kilocalorie. The steps to calculate contributed to the decision to define a new poverty line each of these is discussed in turn. for 2017. The threshold is a new baseline for poverty Step 1: Set the minimum caloric intake. The minimum monitoring in Djibouti and will be used as a reference for caloric intake of this basket of consumption must allow the future EDAM analyses. daily activities of an average individual to be carried out. To construct poverty lines, the absolute concept to DISED has estimated 2,115 kilocalories a day is a good calculate the poverty line is used. The cost-of-basic- 48 reference for the caloric needs of the population. needs approach is used,49 which consists of estimating a Step 2: Calculate the unit cost of acquiring a food component and a nonfood component of the poverty kilocalorie. To calculate the unit cost of acquiring a line. The food component is based on the calculation kilocalorie, one needs to identify the food calories and of a food threshold that guarantees a minimum level of the cost or price associated with each product covered energy intake of 2,115 kilocalories per day.50 For the in EDAM4-IS. In addition, a reference population must be construction of the nonfood component of the poverty line, identified for whom the unit cost of a kilocalorie will be the Ravallion (1998) method was applied to construct two calculated. nonfood poverty lines, which are explained in turn. 48 Poverty lines typically follow one of two main approaches. A first approach considers that the poverty line is an absolute concept, indicating the level of consumption needed to meet just minimum food and nonfood needs. The second approach interprets the poverty line as a relative concept, so poverty is understood as a situation of relative deprivation. The actual values of the relative poverty lines increase with the economic (and social) conditions of the country. 49  The cost-of-basic-needs approach has been used in DISED (2014). 50 Although lower than the thresholds used in other African countries, this level was considered appropriate in the current Djiboutian context. 111 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI The choice of a reference population must ensure that its the current exercise also took the entire population as a consumption basket is closest to the population living near reference, where average caloric consumption per adult the poverty line, but at the same time, that such basket of equivalent stands at 1,791 calories (the distribution is consumption can reach a sufficient nutritional level. shown in figure B.2). It is also found that consumption of the second and fourth deciles are low: this population Using EDAM-BC 2013, DISED (2014) determined the consumes only 15 food products on average, resulting in reference population as follows: instead of using the 1,444 kilocalories per day per equivalent adult. This result population between the second and fourth deciles, suggests that calculating the unit cost of a kilocalorie which is a common method, the entire population was using the consumption of the population from the second chosen as the reference. According to DISED’s estimates, to the fourth decile would lead to an average value of the the population between the second and fourth decile cost that would underrepresent the real cost of a basket “consumes a basket of food products whose nutrient that provides a desirable minimum caloric intake. The intake, mainly protein, is low. Therefore, their consumption consumption basket of the reference population (entire does not reflect a minimum caloric intake required but population) identified is used and this forms the basis for rather the low income that does not allow them to buy the calculation of food poverty line. food of reasonable quality.” The approach followed for FIGURE B.2 Distribution of per Adult Equivalent Daily Intake of Calories Density 0 2000 4000 6000 8000 10000 12000 14000 16000 Per adult equivalent daily intake of calories Source: Calculations based on EDAM4-IS. APPENDIX B 112 It is necessary to obtain the caloric composition of each EXTREME POVERTY THRESHOLD food product covered in EDAM4 in Djibouti to calculate the The extreme poverty line includes a food component and unit cost of a kilocalorie. Since food caloric data are not a nonfood component. The nonfood component of the available specifically for Djibouti, a table was constructed extreme poverty line is calculated by looking at households for Djibouti from the Food and Agriculture Organization’s whose total consumption expenditure is equal to the food table prepared for the West Africa (FAO 2012) and for poverty line. Households for which these two are equal North African countries (Tunisian National Table). This has choose to divide their consumption between food and been done keeping in mind the food products that are nonfood items. It follows that these households consider mainly consumed in Djibouti.51 that spending on nonfood items will boost their level of The next step to calculate the unit cost of acquisition of satisfaction. The expenditure corresponding to these a kilocalorie is to establish the price of each product. Out nonfood products is therefore considered indispensable. of 100 food products in EDAM4-IS, all the necessary To estimate this extreme poverty line, Ravallion’s (1998) information could be gathered for 44 products. The price method of estimating a Engel function is done. It shows the paid (as a proxy for the unit cost) based on the quantity relationship between the budgetary part of food expenditure purchased and the amount paid that was entered in the ( ), consumption per adult equivalent ( ) normalized by survey is then calculated. In addition, this price can be the food poverty line ( ), and the deviation of household allocated by kilogram or liter.52 Djiboutian households spend size ( ) from the average household size, as follows: 81 percent of food expenditure on these 44 products. The remaining 56 products were not included because they did not satisfy one of the following criteria: (i) there (2) is a credible mapping between the product and calorie intake (for example, in food group codes such as “cookies, The quantile regression approach is used to estimate the pastries and croissants”); (ii) consumption was reported coefficient in equation (2) for the entire country in order to mainly in nonstandard units and could not be transformed calculate the unique food and nonfood poverty lines for the into kilograms or liters; and (iii) the number of households entire country. Table B.7 shows the estimated value of the reporting consumption of the food product was low. parameters of equation (2). FOOD POVERTY THRESHOLD The budget coefficient is found to be 0.59 or 59.4 percent With these elements, the level of annual expenditure (per for the country. The income coefficient of effect describes equivalent adult) needed to meet the minimum nutritional income elasticity—this elasticity is less than 1. That means needs of 2,115 kilocalories using the cost-of-basic- that spending on food is a necessity expense. needs approach is determined. The median unit cost of the calorie is used to determine the food component of When the household consumption per adult equivalent is the poverty line. The average unit cost of a kilocalorie is exactly equal to the food poverty line and the household estimated at DF 0.103 in 2017 and the estimated food size is equal to the average household size, the part of the poverty line is DF 79,480. household budget devoted to food consumption is given 51  The reliability of the data source has been verified in the poverty calculations performed in previous years by DISED. 52 All the information needed to transform the different acquisition units was collected and analyzed. Using detailed product data and the most common prices collected by DISED, a map of most units could be made in kilograms or liters. This is the key to being able to map the quantities of calories consumed by households. 113 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI TABLE B.7 Estimation of the Engel Curve Coefficients in 2017 Coefficient Standard deviation t P>t Income effect -0.101 0.007 -15.33 0 Income effect squared -0.009 0.003 -2.98 0.003 Household size -0.020 0.002 -13.43 0 Constant 0.594 0.005 120.28 0 Source: Calculations based on EDAM4-IS. by α (according to equation (2)). Thus, the nonfood poverty The overall nonfood poverty line can be calculated using line can be deduced as follows: equation (3) for y = and a food budget share equal to / : (3) (5) Thus, a relationship between the food poverty line ( ) and the extreme poverty line ( ) is given as follows: Finally, this overall nonfood poverty line is added to the (4) food poverty line to calculate the overall poverty line. Using the 2017 EDAM4 survey and the baseline consumer Finally, this nonfood poverty line is added to the food basket that guarantees 2,115 kilocalories per day per poverty line to calculate the extreme poverty line. The adult, the extreme poverty line of 2017 equals DF 306 per extreme poverty line reflects the bare minimum necessary day and the overall poverty line is estimated at DF 415 per to have sufficient nutritional intake to survive. day. Using the extreme poverty line, table B.8 presents results at different levels of aggregation. OVERALL POVERTY THRESHOLD The overall poverty line also includes a food component Other alternatives of poverty lines to explore the (as previously defined) and a nonfood component that robustness of our chosen approach (and its implications are calculated by determining the total consumption of for welfare and poverty measurement) are considered. households whose food consumption expenditure is equal First, the poverty line used for 2013 is used and adjusted to the food poverty line. This ensures that households can with inflation experienced between May 2013 and May achieve the caloric intake equal to the minimum as defined 2017 (3.84 percent). The consequent poverty line gives by the food poverty line without forgoing consumption to poverty rates that do not differ qualitatively from the official nonfood products. results (extreme poverty rate of 18.3 percent and overall poverty rate of 36.5 percent). Second, the poverty line APPENDIX B 114 of 2013 is used and adjusted with inflation experienced collected in two phases, in May 2017 and November- between May 2013 and May 2017 for food expenditures December 2017, some concern was expressed regarding and nonfood expenditures separately (11.3 percent the role of seasonality. To test this, restricting to phase one inflation for the food part and 0.54 percent deflation for the only (conducted in May 2017), it is found that the results nonfood part). Using the resulting poverty thresholds, the on poverty do not differ from the official ones qualitatively extreme poverty rate stands at 19.4 percent and overall (resulting in an extreme poverty rate of 23.4 percent and poverty rate of 37 percent. Finally, since the data were overall poverty rate of 39.4 percent). TABLE B.8 Extreme Poverty Rate and Corresponding Standard Errors and Confidence Intervals Estimate Standard error 95% confidence interval National 21.1% 0.90 19.40 22.90 Regions 45.0% 1.30 42.50 47.50 Djibouti city 13.6% 1.10 11.40 15.80 Rural 62.6% 1.30 59.90 65.20 Other urban 14.8% 2.00 10.80 18.80 Ali Sabieh 27.2% 2.30 22.60 31.70 Dikhil 52.9% 2.70 47.60 58.10 Tadjourah 65.4% 2.60 60.30 70.60 Obock 40.4% 2.90 34.80 46.10 Arta 31.6% 2.70 26.40 36.80 1st district 4.7% 2.20 0.40 9.00 2nd district 9.8% 1.80 6.20 13.30 3rd district 8.5% 1.80 5.00 12.00 4th district 18.3% 2.30 13.70 22.90 5th district 15.8% 2.30 11.40 20.20 Source: Calculations based on EDAM4-IS. 115 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI REFERENCES Deaton, Angus, and Margaret Grosh. 1998. Consumption. In Designing Household Survey Questionnaires for Developing Countries: Lessons from Ten Years of LSMS Experience Edited by: Grosh, M. and Glewwe, P. mimeo, Ch.17 Deaton, A., and S. Zaidi. 2002. Guidelines for Constructing Consumption Aggregates for Welfare Analysis. Washington, DC: World Bank. DISED (Direction de la Statistiques et des Etudes Démographiques). 2014. “Mesures de la pauvreté et des inégalités.” Photocopy. Djibouti: DISED. Duan, Naihua. 1983. “Smearing Estimate: A Nonparametric Retransformation Method.” Journal of the American Statistical Association 78 (383): 605–10. FAO (UN Food and Agriculture Organization). 2012. West African Food Composition Table. Rome: FAO. http://www. fao.org/3/i2698b/i2698b00.pdf. FAO. 2007. Tunisian National Table. Rome: FAO. Freeman, A. Myrick. 1993. The measurement of Environmental and Resource Values Resources for the Future.” Washington, DC: World Bank. Ravallion, M. 1998. “Poverty Lines in Theory and Practice.” LSMS Working Paper 133. World Bank: Washington DC. United Policy Holders. Depreciation Schedule – residential personal property. https://www.uphelp.org/library/ resource/depreciation-schedule-residential-personal- property APPENDIX B 116 117 CHALLENGES TO INCLUSIVE GROWTH A POVERTY AND EQUITY ASSESSMENT OF DJIBOUTI