Report No. 40328-GH Ghana Job Creation and Skills Development (In Two Volumes) Volume II: Background Papers May 29, 2009 Human Development, AFTH2 Country Department AFCW1 Africa Region Document of the World Bank Acknowledgements This report was prepared by a team consisting of Setareh Razmara (AFTH2, task team leader), Arvil Van Adams (Consultant), Quentin Wodon (AFTPM), Pieter Serneels (PREM), Harold Coulombe (Consultant), and Moukim Temourov (AFTH2). Moses Anwonoor, Alonso Sanchez and Kalpana Mehra (Consultants) helped with early data analysis. Montserrat Pallares-Miralles (HDNSP) also provided valuable inputs on pension reforms in Ghana. The report draws from two background papers that provided substantial inputs: (i) The Role of Employment and Earnings for Shared Growth: the Case of Ghana (Pieter Serneels, PREM); and (ii) Education, Skills and Labor Market Outcomes in Ghana: Planning reforms for skills development (Arvil V. Adams, Consultant). In addition, as part of this work, Francis Teal (Center for the Study of African Economies, University of Oxford), prepared two background papers on: (i) Apprenticeship in Ghana; and (ii) Formal and Informal Employment in Ghana: Job Creation and Skills. Significant suggestions and comments were provided by Ishac Diwan, Zeljko Bogetic, Pierella Paci, Harold Alderman, Peter Darvas, Katherine Bain, and Daniel Kwabena Boakye. Peer reviewers were Louise Fox, Gordon Betcherman and Zafiris Tzannatos. A major part of this report on the labor market outcomes and education and skills was incorporated in the background papers of the Country Economic Memorandum (CEM). We would like to thank Mats Karlsson who was the Country Director at the time the study was launched. Kathryn Bach provided valuable editorial inputs. Administrative support was provided by Ruth Mulahi, Marietou Touré and Nicole Hamon (AFTH2). The preparation of this study also benefited from the cooperation of the PREM and Human Development/Social Protection Anchors based on the BNPP Project on Labor Market Indicators. The Volume 2 comprises seven background papers authored by the following team members: Arvil Van Adams, Harold Coulombe and Quentin Wodon authors of "Education, Employment and Earnings in Ghana" (Annex 1); Courtney Monk, Justin Sandefur and Francis Teal authors of "Apprenticeship in Ghana" (Annex 2); Quentin Wodon and Harold Coulombe authors of: "Domestic or productive work? Changes in patterns of time use in Ghana" (Annex 4), "Estimating the Potential Cost for Firms of Enforcing the Minimum Wage" (Annex 5) and, with contribution of Moukim Temourov, "Ghana's National Youth Employment Program and Poverty Reduction" (Annex 7); and Harold Coulombe, George Joseph, Caglar Ozden, Yoko Niimi, and Quentin Wodon authors of "International Migration from Ghana: Patterns, Brain Drain and Policy Implications" (Annex 6). Valuable support from the counterpart team in the Ghanaian ministries of Manpower, Youth and Employment (or new Ministries of Employment and Social Welfare and Youth and Sports), Education, Science and Sports (or new Ministry of Education and Sciences), Economic Development and Finance, and Ghana Statistical Services are gratefully acknowledged. The team also benefited from discussions with, the Development Partners (particularly DFID, JICA, and ILO), Ghana Labor Unions, Employers Associations and representatives of the private sector. To engage the new government, the findings of the report were presented in the context of the Country Management Unit's "development dialogue" series on competitiveness on April 15 and 17, 2009, and were endorsed by the authorities. ii GHANA Job Creation and Skills Development Table of Contents Annex 1: Education, Employment and Earnings in Ghana.................................................................1 Annex 2: Apprenticeship in Ghana.....................................................................................................19 Annex 3: Variable Definitions Applied to the Household Data .........................................................34 Annex 4: Domestic or Productive Work? Changes in Patterns of Time Use in Ghana ..................35 Annex 5: Estimating the Potential Cost for Firms in Ghana of Enforcing the Minimum Wage....57 Annex 6: International Migration from Ghana:.................................................................................74 Annex 7: Ghana's National Youth Employment Program and Poverty Reduction........................97 List of Tables in Annex 1: Table 1: Summary Statistics for Key Variables of Interest, Ghana 1991-2006 (National Level)......... 12 Table 2: Multinomial Logits for Tupe of Employment, Ghana 1991-2006 (Urban Areas)............... 13 Table 3: Multinomial Logits for Type of Employment, Ghana 1991-2006 (Rural Areas)................. 14 Table 4: Wage Regressions with Sample Selection, Ghana 1991-2006 (Urban Areas).................... 15 Table 5: Wage Regressions with Sample Selction, Ghana 1991-2006 (Rural Areas)...................... 16 Table 6: Wage Regressions with Sample Selection, Ghana 1991-2006 (Urban Areas)..................... 17 Table 7: Wage Regrssions without Sample Selection, Ghana 1991-2006 (Rural Areas).................. 18 List of Tables in Annex 2: Table 1: Training in Ghana in 2006............................................................................... 20 Table 2: Types of Training.......................................................................................... 21 Table 3: Educational Background................................................................................. 21 Table 4: Occupational Outcomes in 2006........................................................................ 22 Table 5: Summary Statistics.......................................................................................... 24 Table 6: Earnings Equations with Years of Schooling.......................................................... 27 Table 7: Earnings Equations with Schooling Dummies......................................................... 28 Appendix Table...................................................................................................... 33 List of Figures in Annex 2: Figure 1: Probability of Entering Occupational Sector for Apprentices.................................... 30 Figure 2: Probability of Entering Occupational Sector for Teachers/Nurses................................. 31 List of Tables in Annex 4: Table 1a: Average Number of Weekly Hours Spent on Various Activities, by Year, Sex and Locality, 41 Ghana, Age Group 7-24.......................................................................................... Table 1b: Average Number of Weekly Hours Spent on Various Activities, by Year, Sex and Locality, 42 Ghana, Age Group 25-64.......................................................................................... Table 1c: Average Number of Weekly Hours Spent on Various Activities, by Year, Sex and Locality, 43 Ghana, Age Group 65+........................................................................................... Table 2: Time Poverty Incidence and Poverty Gap, Ghana 1991/92 to 2005/06........................... 46 Table 3a: Determinants of Time Worked, by Locality and Sex, 2005/06, 7-24 Year Old.................. 48 Table 3b: Determinants of Time Worked, by Locality and Sex, 2005/06, 25-64 Year49 Old....................... Table 3c: Determinants of Time Worked, by Locality and Sex, 2005/06,-65+ Year Old.................. 50 (iii) GHANA Job Creation and Skills Development List of Figures in Annex 4: Figure 1: Distribution of Total Time Worked, Individuals Ages 25-64, Ghana............................ 39 Box in Annex 4 Box 1: Distribution of Total Time Worked, Individuals Ages 25-64,45 Ghana........................................ List of Tables in Appendix to Annex 4: Table a: Determinants for Logarithm of Total Time Worked, by Locality and Sex, 2005/05 54 For 7-24 Year Old................................................................................................. Table b: Determinants for Logarithm of Total Time Worked, by Locality and Sex, 2005/05 55 For 25-64 Year Old ............................................................................................... Table c: Determinants for Logarithm of Total Time Worked, by Locality and Sex, 2005/05 56 For 65+ Year Old................................................................................................... List ot Tables in Annex 5: Table 1: Basic Statistics on Number of Individuals in Sample, Ghana 1991-2006........................ 66 Table 2: Basic Statistics on Wages(Actual and Imputed) in Sample, Ghana 1991-2006.................. 67 Table 3: Headcount Index, Wage Gap, and Squared Wage Gap Measures, Ghana 1991-2006........... 68 Table 4: Headcount Indes, Wage Gap, and Squared Wage Gap Measures for Current Workers 69 ..According to Formal Versus Informal Employment, Ghana 1991-2006.................................... Table 5: Wage Gap Ratio, Ghana 1991 -2006................................................................... 70 List of Figures in Annex 5: Figure 1: Density Function for Wages, Age 18-30,1991/92................................................... 63 Figure 2: Density Function for Wages, Age 18-30, 2005/06................................................... 63 Figure 3: Density Function for Wages, Age 25-64, 1991/92................................................... 64 Figure 4: Density Function for Wages, Age 25-64, 1998/99................................................... 64 Figure 5: Density Function for Wages, Age 25-64, 2005/06................................................... 65 List of Tables in Appendix to Annex 5: Table A1: Linear Regressions Used for Imputed Wages for the Unemployed............................... 72 List of Tables in Annex 6: Table 1: Characteristics of Households with at Least One External Migrant................................ 84 Table 2: Number and Characterics of International Migrants, by Demographics and Employment 86 Ghana 2005/06...................................................................................................... Table 3: Number and Characterics of International Migrants, by Education, Ghana 2005/06............. 87 Table 4: National Education Expenditure by Level of Education............................................. 94 Table 5: Estimats of Total Education Investment in Ghanaian International Migrants.................... 95 List of Figures in Annex 6: Figure 1: Share of Migrants Abroad Relative to Home Population.......................................... 76 Figure 2: Major Destinations (OECD and Non-OECD Countries)........................................... 76 Figure 3: Major Destinations of College Education Migrants................................................. 77 Figure 4: Share of Colleage Educated Migrants Abroad to College Educated Home Population........ 78 Figure 5: Share of Migrants by Education Attainment: 2000................................................. 78 Figure 6: Share of Labor Force by Education Attainment: 2000............................................. 79 Figure 7A: School Enrollment Rate, Tertiary: Panel A. 1991................................................ 80 (iv) GHANA Job Creation and Skills Development Figure 7B: School Enrollment Rate, Tertiary: Panel B. 2001................................................. 80 Figure 8: Age Profile of Migrants in the US on Arrival........................................................ 81 Figure 9A: Share of Migrants by Education Attainment: Panel A: Migrants Arrived in the US at Age 82 ...of 22 or more...................................................................................................... Figure 9B: Share of Migrants by Education Attainment: Panel B: Migrants Arrived in the US at Age 83 ...of less than 22 but currently 22 or more....................................................................... Figure 10: Destination of Migrants in GLSS Data............................................................... 83 Figure 11: Selcted Characteristics of External Migrants, 2005-06............................................ 88 Figure 12A: Labor Market Performance of College Educated Migrants: Panel A: Migrants Arrived 90 ...in the US at Age of 22 or more................................................................................. Figure 12B: Labor Market Performance of College Educated Migrants: Panel B: Migrants Arrived 90 ...in the US at Age of Less Than 22 but Currently 22 or More................................................ Figure 13: Occupation Before and After Migration............................................................. 91 List of Tables in Annex 7: Table 1: The NYEP Youth Employment Registry Data...................................................... 101 Table 2: The NYEP Beneficiary Data, 2006-07............................................................... 102 Table 3: NYEP Annual Expenditures, 2006-07................................................................ 103 Table 4: Potential Beneficiaries of the NYEP Among Individuals Ages 18-35, National 2005-2006 106 Table 5: Potential Beneficiaries of NYEP, Individuals Ages 18-35 with Junior Secondary Education 107 Completed, 2005-2006........................................................................................... Table 6: Potential Leakage Effects of the NYEP for Poverty Reduction, by Region for Junior 114 Secondary Education Completed and High Wages, 2005-2006............................................. Table 7: Potential Leakage Effects of the NYEP for Poverty Reduction, by Region Without 115 Education Criteria and Low Wages, 2005-2006............................................................ Table 8: Potential Impact on Poverty of the NYEP and Public Works, National, 2005-2006......... 116 List of Figures in Annex 7: Figure 1: Distribution of Potential Beneficiaries of Public Works, National, No Education Criteria 108 And Low Wages................................................................................................... Figure 2: Distribution of Potential Beneficiaries of Public Works, Urban, No Education Criteria 108 And Low Wages.................................................................................................... Figure 3: Distribution of Potential Beneficiaries of Public Works, Rural, No Education Criteria 109 And Low Wages.................................................................................................... Figure 4: Distribution of Potential Beneficiaries of NYEP, National, Junior Secondary Education 109 ...Completed and High Wages..................................................................................... Figure 5: Distribution of Potential Beneficiaries of NYEP, Urban, Junior Secondary Education 110 ...Completed and High Wages...................................................................................... Figure 6: Distribution of Potential Beneficiaries of NYEP, Rural, Junior Secondary Education 110 ...Completed and High Wages...................................................................................... (v) ANNEX 1 EDUCATION, EMPLOYMENT AND EARNINGS IN GHANA 1 Substantial work has been done in Ghana and other developing countries analyzing the determinants of type of employment and earnings. Three features of this paper allow us to extend this literature. First, we rely on comparable cross-sections of household survey data collected at three points in time over a 15-year period for our analysis looking at changes in occupation and earnings patterns. Such data are rarely available in Sub-Saharan Africa. Second, the literature on the returns to education tends to focus on the benefits to wage workers of formal education. Yet, in countries like Ghana, technical and vocational education and training (TVET) and traditional apprenticeships play an important role in preparing workers for wage and self- employment. The data from Ghana's household surveys enable us to examine the impact of various types of education and training on the employment held, wage versus self-employment in agriculture and non-agriculture, and on the earnings in each type of employment. Third, to look at the returns to various forms of education, taking into account the different types of employment that workers may have, we estimate wage and employment regressions simultaneously. This is done using a two-step procedure, by first analyzing type of employment using a multinomial logit procedure, and then analyzing the determinants of wages taking self-selection into account. Our results suggest that the impact of formal and TVET education on type of employment and earnings is large, while the evidence on the positive effects of apprenticeships is much weaker. Our results also suggest that completing at least junior secondary education is needed to improve earnings as well as employment type, so that having a primary education is not enough to get a better job. 1. Introduction 1.1 In a world marked by rapid change and globalization, education and skills are becoming more important, not only for individuals to obtain good jobs and earn decent wages, but also for countries to compete. Skills may be acquired in many different settings, ranging from the classroom to the workshop and on-the-job, and have become crucial to improve the prospects for employment and earnings of the labor force, and especially youth as they move from school to work (Adams, 2007). In Ghana, despite a strong record of growth and poverty reduction over the last 15 years (Coulombe and Wodon, 2007), there are concerns as to whether skill deficits have been or may become a constraint to development (Ministry of Education Science and Sports, 2006). 1.2 In Ghana, basic education consists of primary and junior secondary education accounting for nine years of schooling. Recognizing that this education provides limited opportunities to develop the skills needed by the labor force, Ghana's authorities prepared a White paper in 2004 that calls for increasing the emphasis on TVET (technical and vocational education and training) and apprenticeship.2 Yet today, there is widespread popular belief in Ghana that TVET does not lead to strong qualifications and earnings for the individuals pursuing this type of education. This is probably one of the main reasons why few students who complete their junior secondary education choose TVET, as opposed to the more standard senior secondary schooling track. Using a series of 1This background paper was prepared by: Arvil Van Adams, Harold Coulombe, Quentin Wodon, and Setareh Razmara, December 2007 2Ministry of Education, Science and Sports (2004) 1 household surveys with detailed information on labor outcomes for Ghana, our objective in this paper is to measure the impact of education on employment and earnings, with specific attention to TVET and apprenticeships. 1.3 Our work contributes to an extensive literature. There have been a large number of studies on the returns to various types of education (see among others Card 1999, Psacharopoulos 1973, 1981, 1985, 1994; Psacharopoulos and Patrinos, 2004; and Willis (1986).3 It is well known that education influences wages and earnings directly by raising the productivity of the worker and indirectly by promoting entry into more lucrative forms of employment. In Ghana, returns to schooling have been estimated by Glewwe (1996, 1999) with each additional year of schooling found to increase wage earnings by 8.5 percent. Education has also been found to increase wages in the manufacturing sector (Teal, 2000), income in non-farm self-employment (Vijverberg, 1995) and profit from farm and non-farm self-employment (Joliffe, 2004). Blunch and Pörtner (2005) and Blunch (2006) also find substantial returns to formal education, as well as to other forms of skills acquisition, including literacy programs. 1.4 Beyond the findings of previous work, three features of this paper enable us to bring additional value to the debate on the impact of education, including TVET and apprenticeships, on employment and earnings. First, we rely on household survey data collected at three points in time over a 15-year period to conduct our empirical work. Among 26 countries in West and Central Africa, Ghana is the only country that offers repeated and comparable household surveys with sufficiently detailed information on labor outcomes for a sustained period of time permitting analysis of the impact of education on employment and earnings. Using the Ghana Living Standards Surveys for 1991/92, 1998/99, and 2005/2006, we have constructed comparable sets of data to measure trends in the impact of education during a period of massive growth and poverty reduction in the country. 1.5 Second, the literature on returns of education has tended to focus on the benefits of formal education on the earnings of wage earners, although there are a number of notable exceptions dealing with apprenticeships and/or TVET explicitly (see among others CINTERFOR/ILO, 2000, Gill et al., 2000, Atchoarena and Delluc, 2001, OECD, 2000, Johanson and Adams, 2004; Adams, 2007). In countries such as Ghana, technical and vocational education and training, as well as apprenticeships, play an important role4, and many workers are self-employed. Data from the last three rounds of the GLSS enable us to look at the impact of TVET and apprenticeships on both employment and earnings for different types of workers (wage workers, as well as various categories of the self- employed). 1.6 Third, we conduct the analysis recognizing the fact that if one is to look at the returns to various forms of education, taking into account the different types of employment that workers have, we need to estimate wage and employment regressions simultaneously. This is done using a two- step procedure, by first analyzing type of employment using a multinomial logit procedure, and then analyzing the determinants of wages taking self-selection into account. This is also the approach used by Blunch (2006), but using only the 1998/1999 GLSS, while we rely here on data from the last three rounds of the GLSS, including the survey implemented in 2005/2006. 3 For example Psacharopoulos and Patrinos (2004) review 133 studies for 98 countries and estimate that each additional year of education may increase earnings by about 10 percent. 4On apprenticeships in Ghana, see Haan and Serriere (2002), Atchoarena and Delluc (2001), and Monk, Sandefur and Teal (2007). On other training provided in firms, see Rosholm, Nielsen, and Dabalen (2007). On TVET, see UNESCO (2003). 2 1.7 The paper is structured as follows. Section 2 provides a brief overview of the education system and opportunities for skills development. In Section 3, we outline our methodology and basic statistics and in Section 4 we provide the empirical results. Section 5 contains our conclusions. 2. An Overview of Education 1.8 Ghana has made progress in increasing young people's access to primary education as part of its efforts to meet the Millennium Development Goals for education. As a consequence, net enrolment rates of youth 6-11 years of age increased nationally from 59 percent in 2004/05 to nearly 79 percent in 2006/07, with net admission rates for 6 year olds more than doubling to 69.3 percent from the 2004/05 figure of 26.2 percent. Gross enrolment rates, including over-age and under-age youth, rose to nearly 91 percent in 2006/07. The increase in the net enrolment rate for primary education to 78.6 percent pushed Ghana above the Sub-Saharan Africa average of 66.3 percent.5 1.9 Similar progress is evident in junior secondary education which together with primary education comprises nine years of basic education. Net admission rates for 12 year old first-year junior secondary students in 2004/05 surged from 12.2 to 44.4 percent in 2006/07. Although enrolments in senior secondary schools have doubled since 1999, net enrolment rates remain low at 10.6 percent. Nine out of every ten youth 15-17 years of age are not enrolled in a secondary school. Within senior secondary schools, students must choose between a TVET and a general secondary curriculum with slightly less than nine out of ten choosing the latter. The absolute number of youth choosing a TVET education has actually declined from about 21,400 in 2004 to 18,000 in 2006. 1.10 Those not continuing in a senior secondary school join others who have already entered the world of work or who are out of the labor force. The skills of those working are acquired through non-formal training centers like the vocational training institutes operated by the Ministry of Manpower, Youth and Employment and centers operated by for-profit and not-for-profit institutions. Employers also play an important role in skills development through providing skills learned on-the- job, through short-term training, and traditional apprenticeships offered mainly by the informal sector. Monk, Sadefur, and Teal (2007) report that 55 percent of those working were a current or past apprentice compared with 17 percent who had a vocational training background from a school or training center. Another 25 percent acquired their skills on-the-job.6 3. Methodology and Basic Statistics Methodology 1.11 This section explores the impact of this education and training on the employment and earnings of the workforce. It begins by explaining our methodology for estimating the returns to various types of education in Ghana, including apprenticeships and TVET, and the impact that education has on employment choice. Before describing our empirical model, a few comments are necessary. First, we must recognize that returns to various types of education may depend on the type of job that a person gets, but the type of job is itself influenced by the education of the individual. This means that instead of treating employment status as an exogenous independent variable in wage regressions, it is better to consider type of employment as endogenous. In what follows, we will make the distinction between four types of employment: not working or unpaid employment, self-employed workers in agriculture, self-employed workers in other sectors, and wage 5Ministry of Education, Science and Sports (2007) 6Monk, Sadefur, and Teal (2007) 3 employees. For example, we would expect that the marginal impact of education on earnings would be larger for salaried workers than for the self-employed in agriculture that tend to have fewer opportunities to use their education. 1.12 Given the need to consider type of employment as endogenous, we proceed with a two-step empirical procedure. First, we analyze the determinants of the employment status j of individual i using a multinomial logit model as follows: O = oj + oj Ei +oj Xi + ij (1) ij where Ei is a vector of educational attainment variable according to the highest level of education (some primary education, primary education completed, junior secondary education completed, senior secondary completed, TVET completed, and tertiary education), augmented with a set of dummy variables denoting whether the individual has benefited from training as an apprentice, and X is a vector of additional variables which may have a direct impact on wages, controlling for education. 1.13 The variables in X are the age of the individual and its square, whether the individual is married or not, the sex of the individual, the number of infants and children of the individual, the geographic location variables for the capital area of Accra, as well as the Forest and Coastal areas (the excluded region being the Savannah areas, located in the north of the country), and finally information on the education level of the father of the individual. 1.14 The second stage regression for the determinants of wages in each of the employment categories (the excluded category is that of individuals not working or working without pay) is estimated taking into account sample selection from the multinomial logit using the Durbin- McFadden (1984) procedure (see also Bourguignon, Foumier, and Gurgand, 2004). The second stage regressions are: W = wj + wj Ei + wj Zi + wj wij + ij (2) ij where Wij denotes the natural logarithm of the earnings of individual i with occupational status j, and the variables are the estimated inverse Mills ratios obtained from the first-stage multinomial logit, and is a normally distributed error term. Standard errors are estimated using bootstrapping. 1.15 There is however a debate in the literature as to whether the two steps procedure outlined above is necessarily the best way to proceed. Puhani (2000) gives a good survey of this literature on the appropriateness of the different sample-selection correction procedures based in large part on the work of Heckman (1976). Amongst the different causes leading to a potentially harmful correction procedure, the problem of collinearity between the error terms of the multinomial/probit equation and the wage equation is a serious one. This collinearity problem mainly occurs if the former equation is poorly identified. In our case, finding good identifying variables that are statistically significant is not easy. In that case, Puhani (2000) argues that a Heckman-type procedure can often do more harm than good and recommends using OLS or an adapted version of standard regressions instead. 1.16 Given this debate, we also estimate simple linear regressions for the determinants of the logarithm of earnings without taking into account sample selection from the multinomial logit. In this alternative approach, we estimate and report simply: W = 'wj + 'wj Ei + 'wj Zi + 'ij (3) ij 1.17 The variables includes in Z are the following: the assumed number of years of experience of the individual (computed as the age minus six years minus the number of years of education completed), the number of years of experience squared, whether the individual is married or not, the sex of the individual, whether the individual has a formal sector job (this applies only to the wage 4 regression for wage earners), whether the individual works in a firm or organization with a union, whether the individual is employed in the public sector, and finally the geographic location variables. Therefore, some of the variables in X are not in Z and vice versa. This is our baseline specification for which we report results in section 4. 1.18 We also estimate additional regressions where we include variables on the education of the father of the individual, since better educated parents may be able to help their children obtain better wages in their jobs, but the results are not shown here as the coefficients from these variables were not found to be statistically significant. All the regressions are estimated for a sample of individuals aged 25 to 64, because in the 1998/99 survey, the information available on labor outcomes for individuals aged 15 to 24 is not reliable (this is because for that year labor questions are asked only to youth not enrolled in school, which introduces a bias in the sample). Basic Statistics 1.19 Basic statistics for all variables used in the regressions are provided in Table 1 at the national level. As shown in this table, the three paid employment categories exhibit different earnings patterns, with wage employees earning significantly more than the self employed, and with those self-employed in non-farm activities earning more than the self-employed in agriculture. These differences are unadjusted for the education and training profile of workers in the different employment categories. 1.20 Wage workers tend to have higher levels of education than the self-employed. For example, 20.8 percent of wage workers in 2005/06 have post-secondary education, versus less than one percent among the self-employed in agriculture and only two to three percent among the self-employed in non-agriculture sectors and those not working. Apprenticeship plays an important role in skills attainment for workers in all three types of employment. For workers with low (primary) to medium (junior and senior secondary) education, 20 to 30 percent of workers have completed an apprenticeship, while the proportion is much lower for those with high (tertiary) levels of education. 1.21 The mean age of the workforce is about 40 years in all sectors, although the estimated experience is higher in agriculture, simply because most workers in agriculture have low levels of education, and this is the variable which together with age is used to define experience. About three- fourths of workers are married, and a majority of wage earners and self-employed workers in agricultural activities are males, while the reverse is true for the self-employed outside of agriculture, who are more likely to be female. On average, workers have one infant and one child. Most wage earners are located in urban areas and most also have formal jobs. By contrast most individuals self- employed in agriculture live in rural areas, where formal jobs in this group are rare. A majority of the self-employed outside of agriculture are in urban areas, but almost none has a formal job. 1.22 Changes have occurred in employment over the 15-year period reducing the share of employment in union and public sector jobs. In 2005/06, slightly less than half of the wage workers were employed in firms with a union, and this proportion was approximately thirteen percentage points higher fifteen years earlier (on unions in Ghana, see Anyemedu, 2000). In 2005/06, only slightly more than one-third of wage workers were employed in the public sector, while the proportion was almost twice as high 15 years ago. Table 1 also gives data on the geographic location of various types of workers as well as on the type of employment held by parents of the workers. 5 4. Regression Results Type of employment 1.23 Table 2 provides results from the multinomial logit regressions in urban areas. As a reminder, all regressions are based on a sample of workers aged 25 to 64. The estimated coefficients are odds ratios, hence a value above one implies that the likelihood of having a certain job increases, while a value below one implies a lower likelihood of having a certain job. T-statistics are provided in parenthesis to assess whether the coefficients are different from one. The key variables are those for the education of workers, namely some primary education, primary completed, lower secondary completed, TVET certificate completed, senior secondary complete, and finally some tertiary education. The apprenticeship variable is interacted with the level of education in order to assess whether the fact of having gone through an apprenticeship has a different effect depending on the level of education of the workers (one would expect apprenticeships to be more important for workers with low levels of education, i.e. primary and under). These results are important for assessing the impact of education on well-being since education influences earnings indirectly through the type of job held. 1.24 The regression coefficients reveal a strong and statistically significant influence of education on the type of job held. An education (at the level of lower secondary education or higher) increases the chances of becoming a wage worker or a non-agriculture self-employed worker. A primary education also raises the chance for non-agriculture self-employment. Most of the coefficients are relatively stable over time, but one can see a decrease in the impact of TVET on the likelihood of becoming a wage earner. The impact is still present and large in 2005/06, but less so than in 1991/92 and 1998/99. While the impact of having completed an apprenticeship is at times not statistically significant, there is strong evidence in 2005/06 that having done an apprenticeship increases the probability of being self-employed outside of agriculture, and reduces the likelihood of employment in other sectors. 1.25 The likelihood of employment in all three types of employment increases with age, but the rate of increase declines as one grows older. Married individuals are also more likely to hold the various types of employment. Men are more likely to be a wage earner or a self-employed person in agriculture than women, but women are more likely than men to be self-employed outside of agriculture. Living in Accra or the Forest area increase the likelihood of being a wage earner, and living in the Forest area also increases the likelihood of being self-employed outside of agriculture. 1.26 Table 3 provides the same results for rural areas. Many of the results are qualitatively similar, but there are some exceptions. For example, completing a primary education affects type of employment in rural areas, whereas this was the case in urban areas only with a lower secondary education. In addition, the impact of TVET on formal employment increases in 2005/06 in rural areas, while the reverse was observed in urban areas. For those with limited or medium levels of education, having an apprenticeship in rural areas increases significantly the probability of being a wage worker or of being self-employed in the non-agricultural sector. Earnings 1.27 Education also influences earnings directly, after controlling for the type of employment of the household. The estimates from the wage regressions are provided in Tables 4 and 5 for urban and rural areas respectively when taking sample selection into account, and in Tables 6 and 7 without accounting for sample selection. Given the semi-log specification, the parameters can roughly be interpreted as representing percentage increases in wages associated with various characteristics. 6 1.28 Consider first the estimates provided with sample selection. In virtually all cases partial or completed primary education is not enough to increase earnings as compared to having no education at all (there is only one exception for the self-employed in agriculture in 1991/92 in urban areas). In urban areas, the returns to education start to be statistically significant as of the junior secondary level, although even then, this is not systematically the case. In 2005/06, the year for which those coefficients are statistically significant, the returns appear to be different from zero for wage earners and the self-employed in agriculture, but not the self-employed outside of agriculture. 1.29 The gain in earnings associated with education becomes larger in urban areas at the senior secondary level as well as for TVET, with estimates for TVET matching those from senior secondary education for wage earners. For other categories, the returns from TVET are not statistically significant. In rural areas, the situation is a bit different. In most cases, even a senior secondary or TVET education does not seem to bring in much gain. The exception in 2005/06 is for the self- employed in agriculture, where returns from senior secondary education are significant. 1.30 The highest returns to education are obtained from post-secondary education, as expected, especially for wage earners in urban areas. In general, the coefficient estimates for the returns to education are more often statistically significant for wage earners than for the other two groups. This may reflect the limited opportunities for using one's education in the informal self-employment sector in both urban and rural areas. In urban areas, some coefficients for high levels of education are statistically significant for the self-employed outside of agriculture, while this is only observed for the self-employed in agriculture in rural areas. 1.31 The impact of apprenticeship is less favorable than that of TVET. In 2005/06 for example, in most cases in urban and rural areas the coefficients are not statistically significant, and in some cases apprenticeships are associated with a decrease in wages for workers with medium or high levels of education. Thus, overall, apprenticeship does not appear to be a positive factor influencing earnings, at least when sample selection is taken into account in the regressions (we come back to the results for apprenticeship below in discussing standard linear regressions). We note, however, its importance to the type of job held. 1.32 There are also interesting results in Tables 4 and 5 regarding the impact of additional variables on earnings. First, having a formal job brings in an increase in earnings of about 20 percent among wage earners in urban areas in both 1998-99 and 2005-06. This is significant, but still below the premium that formal sector workers enjoyed in the early 1990s. In rural areas, there is no similar premium to formal employment. Second, working for a firm or agency where there is a union also increases earnings significantly, by about 14 to 30 percent depending on the year, with stronger statistical significance for wage earners (there is a curious negative impact of unions in rural areas for the self-employed outside of agriculture, but this is probably a data issue as this is not a sector that is unionized). 1.33 There is also some evidence that the premiums enjoyed by workers from unions is decreasing over time, in both urban and rural areas, at least between 1998/99 and 2005/06. The coefficients were not statistically significant in 1991/92. Third, controlling for formal employment and the presence of a union, there is no statistically significant premium for public sector workers. Still, because public sector workers work in the formal sector and have unions, they benefit from the two other premiums. Fourth, experience brings in a gain in earnings for wage workers while this is less the case for other categories. 7 1.34 As expected, there are also important differences in wages among regions. In Accra, earnings are substantially higher than in the Savannah area (the reference excluded region in the regression), with a gain of about 50 percent for wage earners in urban areas. There is also evidence that for wage workers (and in one case in 2005/06 for the self-employed outside of agriculture in urban areas) being in the Forest or Coastal area brings a gain in earnings as compared to living in the Savannah and poorer part of the country. Finally, there is evidence that men are better paid than women. In the case of wage earners in urban areas, the premium was at about 44 percent in 2005-06, although in other sectors this was not significant. 1.35 Consider now the estimates provided without sample selection in Tables 6 and 7. There are five main differences in these estimates versus the ones obtained with sample selection. First, we find there are positive returns to junior secondary education in both urban and rural areas. Second, in addition to what we had obtained with sample selection, we now find statistically significant returns for TVET in rural areas for wage earners, and in urban areas for the self-employed outside of agriculture. Third, the premium associated with being a wage earner is generally lower in terms of the differences in the estimate of the returns to education between sectors, and in some cases the returns are lower without sample selection. Fourth, we find a positive impact of experience not only for wage workers, but also for the self-employed outside of agriculture. Fifth, we find a positive impact of taking up an apprenticeship in rural areas for wage workers with limited education. 1.36 This last result is similar to the results of Monk, Sadefur, and Teal (2007) who also allow the level of education to interact with participation in an apprenticeship. These authors find that those with the lowest levels of education (junior secondary or less) and an apprenticeship have a 42 percent higher income than those with the same level of education and no participation in an apprenticeship (our own estimate for rural areas is very close, at 46 percent). Those with higher levels of education, however, fail to realize any gain in earnings from an apprenticeship, as is the case for us whether we use sample selection or not in estimating the wage regressions. Overall our finding on apprenticeships are consistent with the fact that fewer individuals with a senior secondary education or higher participate in an apprenticeship. Also, the incentives for an apprenticeship are found for those with lower levels of education, perhaps in terms of wages but also in terms of getting certain jobs.7 5. Conclusion 1.37 This paper has provided a detailed analysis of the determinants of employment, type of employment, and earnings in Ghana over the last 15 years, with an emphasis on the impact of TVET and apprenticeships, apart from other forms of education. Several findings stand out. First, the impact of primary education on employment and earnings is not statistically significant in most cases. Second, when estimations are based on a model taking sample selectivity into account, even the impact of junior secondary education is often low on earnings, although there is an impact on type of employment. Third, the returns and the impact on type of employment start to be large as of senior secondary education, and the gains from TVET are similar to those from senior secondary schooling. Fourth, apprenticeships tend to generate very limited gains, if any, although they do matter for the type of employment held. These findings should be of use for Ghana's authorities as they aim to implement their strategy to improve the education and skills of the country's labor force. 7Monk, Sadefur and Teal (2007) assert that the reasons for undertaking an apprenticeship can be found, in part, in the effect of an apprenticeship on the probability of informal employment relative to having no job. 8 References Adams, Arvil V. (2007). The Role of Youth Skills Development in the Transition from School to Work: A Global Review, HDNCY Discussion Paper No. 5, Washington, D.C.: World Bank Anyemedu K. (2000). "Trade Union Responses to Globalization: Case Study of Ghana", University of Ghana. Atchoarena, David and Andre Marcel Delluc (2001). Revisiting Technical and Vocational Education in Sub-Saharan Africa: An Update on Trends, Innovations, and Challenges, IIEP/Prg.DA/1,320. Paris: International Institute for Educational Planning. Batra, Geeta (2000). "Foreign Direct Investment and Skills Upgrading in Developing Countries," Foreign Investment Advisory Service. Washington, D.C.: World Bank reference available at http://www.oecd.org/dataoecd/11/12/2731690.pdf. Blunch, Niels-Hugo (2006) Skills, Schooling and Household Well-Being in Ghana, Unpublished PhD Dissertation, The George Washington University, Washington, DC. Blunch, Niels-Hugo and Claus Pörtner (2005) "Literacy, Skills and Welfare: Effects of Participation in an Adult literacy Program," Working Paper UWEC-2005-23, Department of Economics, University of Washington, Seattle. Bowers, Norman, Anne Sonnet, and Laura Bardone (1999). "Giving Young People a Good Start: the Experience of OECD Countries," in Preparing Youth for the 21st Century: the Transition from Education to the Labor Market. Paris: OECD: 7-86 Card, David (1999) "The Causal Effect of Education on Earnings," in O. Ashenfelter and D. Card (eds) Handbook of Labor Economics, Volume 3, Elsevier Science B. V. Coulombe, H. and Q. Wodon (2007). Poverty, Livelihoods, and Access to Basic Services in Ghana. Ghana CEM: Meeting The Challenge of Accelerated and Shared Growth. Durbin, J.A. and D. McFadden (1984) "An Econometric Analysis of Residential Electric Appliance Holdings and Consumption," Econometrica, 52(2): 345-362. Glewwe, Paul (1996) "The Relevance of Standard Estimates of Rates of Return to Schooling for Education Policy: A Critical Assessment," Journal of Development Economics, 51: 267-290. Glewwe, Paul (1999) "The Impact of Cognitive Skills on Wages," in Paul Glewwe (ed) The Economics of School Quality Investments in Developing Countries: An Empirical Study of Ghana, London: Macmillan. Gill, Indermit, Fred Fluitman, Amit Dar (2000). Vocational Education and Training Reform: Matching Skills to Markets and Budgets. World Bank and International Labor Organization: Washington, D.C. and Geneva Government of Ghana (2006). Youth Employment Implementation Guidelines. Government of Ghana, ILO and UNDP (2004). "An Employment Framework for Poverty reduction in Ghana". Haan, Hans Christiaan and Nicolas Serriere (2002). "Training for Work in the Informal Sector: Fresh Evidence from West and Central Africa," Occasional Papers of the International Training Centre of the ILO, Turin: ILO Heckman, J. J. (1976) The Common Structure of Statistical Models of Truncation, Sample Selection and Limited Dependent Variables and a Simple Estimator for Such Models Annals of Economic and Social Measurement, 5:475-92 Johanson, Richard and Arvil V. Adams (2004) Skills Development in Sub-Saharan Africa, Washington, D.C.: World Bank Joliffe, Dean (2004) "The Impact of Education in Rural Ghana: Examining Household Labor Allocation and Returns On and Off the Farm," Journal of Development Economics, 73: 287-314. 9 Ministry of Education, Youth and Sports (2003). "Education Strategic Plan I and II," Accra ______ (2004). White Paper on the Report of Education Reform Review Committee, Ministry of Education, Youth and Sports, Accra ______ (2006). "Preliminary Education Sector Performance Report 2006," Accra ______ and Ministry of Manpower, Youth and Employment (2004). "Executive Summary: TVET Policy for Ghana," Accra Ministry of Manpower, Youth and Employment (2006). The National Employment Policy, Draft. Ministry of Manpower, Youth and Employment (2006). The National Social Protection Strategy: Investing in People, Draft. Ministry of Manpower, Youth. and Employment (2006). The National Employment Policy - draft. Monk, Courtney, Justin Sandefur, and Francis Teal. 2007. "Apprenticeship in Ghana," Centre for the Study of African Economies, Department of Economics, University of Oxford, (processed). National Education Reform Implementation Committee (2006). "Report of the TVET Sub-Committee," Accra Rosholm, M., H. S. Nielsen and A. Dabalen 2007. "Evaluation of Training in African Enterprises," Journal of Development Economics, vol. 84(1), pages 310-329. Palmer, Robert (2007). Skills Development, the Enabling Environment and Informal Micro-Enterprise in Ghana, doctoral thesis (mimeo), Edinburgh: University of Edinburgh Psacharopoulos, George (1973) Returns to Education: An International Comparison, Joessey Bass, Elsevier. Psacharopoulos, George (1981) "Returns to Education: An Updated International Comparison," Comparative Education, 17: 321-341. Psacharopoulos, George (1985) "Returns to Education: A Further International Update and Implications," Journal of Human Resources, 20(4): 583-611. Psacharopoulos, George (1994) Returns to Investment in Education: A Global Update, World Development, 22(9): 1325-1343. Psacharopoulos, George and Harry Anthony Patrinos (2004) Returns to Investment in Education: A Further Update, Education Economics, 12(2): 111-134. Puhani, P. A. (2000) "The Heckman Correction for Sample Selection and its Critique," Journal of Economic Surveys, 14: 53-68 Sandefur, J. (2006). Explaining the Trend toward Informal Employment in Africa: Evidence from Ghanaian Manufacturing. mimeo, The World Bank. Sandefur, J., P. Serneels, et al. (2007). Poverty and Earnings Mobility in Three African Countries. Employment and Shared Growth. Rethinking the Role of Labor Mobility for Development. P. Paci and P. Serneels, The World Bank. Soderbom, M. and F. Teal (2001). "Firm Size and human capital as determinants of productivity and earnings." CSAE Working paper 2001-09. Soderbom, M., F. Teal, et al. (2001). "Does firm size really affect earnings?" CSAE Working paper 2002- 08. Spence, Michael A. (1973) "Job Market Signaling," Quarterly Journal of Economics, 87(3): 55-74. Teal, Francis (2000) "Real Wages and the Demand for Skilled and Unskilled Male Labor in Ghana's Manufacturing Sector: 1991-1995," Journal of Development Economics, 61: 447-461. Tan, Hong (2005). "In-service Skills Upgrading and Training Policy: Global and Regional Perspective," Paper presented at the MNA Job Creation and Skills Development Conference, December 2005, Cairo 10 Tan, Hong and Pei Savchenko (2003). "In-service Training and Productivity: Results from Investment Climate Surveys," Washington, D.C.: World Bank (processed). Teal F. (2005). "Ghana Investment Climate Assessment", Draft, University of Oxford. UNESCO (2003). "Synthesis of Main Findings from Two Case Studies Carried Out in Ghana and Zambia on Private Technical and Vocational Education and Training (TVET)," Paris: UNESCO Unevoc (2006). "Participation in Formal Technical and Vocational Education and Training Programmes Worldwide," UNESCO International Centre for Technical and Vocational Education and Training: Bonn Vijverberg, Wim P.M. (1995) "Returns to Schooling in Non-Farm Self-Employment: An Econometric Case Study of Ghana," World Development, 23(7): 1215-1227. Vijverberg, Wim P.M. (1999) "The Impact of Schooling and Cognitive Skills on Income from Non-Farm Self-Employment," in Paul Glewwe (ed) The Economics of School Quality Investments in Developing Countries: An Empirical Study of Ghana, London: Macmillan. Willis, Robert J. (1986) "Wage Determinants: A Survey and Reinterpretation of Human Capital Earnings Functions," in O. Ashenfelter and R. Layard (eds) Handbook of Labor Economics, North- Holland: Elsevier. 11 ng ton orki w .. 0.094 0.107 0.233 0.023 0.048 0.029 0.093 0.100 0.010 39.479 30.466 0.744 0.310 1.158 1.221 0.353 0.647 0.022 0.014 0.013 0.118 0.370 0.129 0.702 0.030 0.033 0.159 0.068 groa 06 f els non 7.843 0.124 0.138 0.377 0.044 0.051 0.022 0.131 0.214 0.026 39.193 28.544 0.740 0.274 0.854 1.027 0.612 0.388 0.006 0.001 0.000 0.209 0.419 0.192 0.538 0.055 0.041 0.260 0.087 2005/ )levella f els groa 7.371 0.125 0.135 0.264 0.006 0.020 0.004 0.109 0.092 0.006 42.955 34.405 0.817 0.617 1.097 1.285 0.124 0.876 0.001 0.000 0.000 0.001 0.483 0.187 0.852 0.010 0.034 0.135 0.021 r e oni rnea Wag E 8.270 0.044 0.076 0.382 0.082 0.138 0.208 0.077 0.195 0.040 39.293 24.725 0.721 0.772 0.665 0.783 0.727 0.273 0.692 0.424 0.357 0.342 0.371 0.184 0.439 0.062 0.044 0.290 0.146 atn( 6 002 ng 1- 991 to orki N w .. 0.100 0.109 0.267 0.014 0.050 0.033 0.092 0.113 0.015 39.245 29.953 0.740 0.306 1.093 1.235 0.332 0.668 0.049 0.005 0.010 0.114 0.366 0.178 0.663 0.043 0.039 0.166 0.066 a anh G,tseretni lan 99 groa f tioa els non 6.291 0.124 0.140 0.349 0.028 0.053 0.026 0.129 0.185 0.025 38.476 28.149 0.765 0.303 0.938 1.252 0.529 0.471 0.107 0.003 0.000 0.207 0.381 0.234 0.507 0.083 0.045 0.192 0.089 N 1998/ f of els groa 5.521 0.108 0.132 0.271 0.005 0.020 0.011 0.136 0.101 0.006 42.910 34.224 0.777 0.586 1.148 1.343 0.115 0.885 0.026 0.000 0.000 0.008 0.510 0.186 0.850 0.015 0.037 0.094 0.023 selbai 12 r e rnea arv Wag E 6.699 0.043 0.070 0.358 0.028 0.161 0.253 0.083 0.146 0.042 40.528 25.862 0.799 0.781 0.788 1.127 0.589 0.411 0.808 0.524 0.516 0.272 0.419 0.205 0.451 0.079 0.065 0.278 0.126 yek orf scitsi ng ton orki w .. 0.068 0.096 0.213 0.008 0.036 0.020 0.099 0.070 0.006 39.147 30.834 0.754 0.247 1.373 1.473 0.308 0.692 0.026 0.007 0.027 0.077 0.362 0.203 0.732 0.031 0.040 0.143 0.042 atts y ).siseh ar groa nt 92 f re m els non 4.523 0.089 0.103 0.317 0.014 0.028 0.005 0.129 0.151 0.006 38.740 29.694 0.793 0.236 1.251 1.423 0.558 0.442 0.104 0.001 0.000 0.155 0.324 0.284 0.574 0.085 0.045 0.177 0.044 pa muS 1991/ ni 1: f el elS gro A 3.931 0.091 0.114 0.238 0.005 0.010 0.001 0.141 0.087 0.000 43.071 35.060 0.805 0.619 1.301 1.328 0.091 0.909 0.016 0.000 0.000 0.000 0.463 0.217 0.866 0.013 0.036 0.066 0.010 scitsitat -s ab (t T r at e rnea da Wag E 4.867 0.048 0.054 0.430 0.035 0.153 0.117 0.104 0.185 0.035 39.766 26.585 0.803 0.755 1.000 1.114 0.644 0.356 0.835 0.555 0.645 0.231 0.374 0.256 0.531 0.049 0.051 0.229 0.091 SS L G ngi r)e r) duce duce duce e d.e hg us ghe og)l ow ry m 0-6 7-14 e ddli n rya (l (hi wol- ­ hi- rya mt hors (i m ry ry eci eci eci e d d de m ut nce gea gea gro ddli ra ng pri obj e rya nt nt nt T ri de dsi dsi n l objla on ci ts r_a r_t r_pri r_m r_pos A:e rnia m E lats E omS riP ondaceS V T ondaceS ondaceSt osP ppre ppre ppre ge xpe rria ela K K rba A A A A E M M # # U Rura ormF ni U ublP racc A oreF Coa hetaF hetaF hetaF hetaF hetaF ourcS Table 2: Multinomial Logits for Type of employment, Ghana 1991-2006 (urban areas) Urban ­ age group 25-64 1991/92 1998-99 2005/06 Wage Self Self Wage Self Self Wage Self Self Earner Agro Non Agro Earner Agro Non Agro Earner Agro Non Agro Some Primary 1.247 1.718* 1.266 1.201 1.009 1.244 1.398 1.425 1.426** (0.78) (1.83) (0.85) (0.57) (0.03) (1.05) (1.52) (1.54) (2.17) Primary 0.991 1.115 0.738 1.760** 1.195 1.139 1.321 1.006 1.322* (0.03) (0.37) (1.47) (2.02) (0.75) (0.68) (1.28) (0.03) (1.92) Secondary (lower) 2.878*** 0.815 0.808 4.439*** 0.689 1.080 2.706*** 0.684* 1.133 (4.33) (0.74) (1.08) (5.87) (1.48) (0.46) (5.63) (1.84) (0.92) TVET 14.85*** 3.489 1.412 7.856*** 0.000 1.360 3.763*** 0.166*** 0.981 (4.21) (1.54) (0.47) (6.15) (117.58) (1.18) (5.57) (3.61) (0.10) Secondary (higher) 5.544*** 0.575 0.503* 7.382*** 0.272*** 0.903 2.813*** 0.365*** 0.575*** (5.26) (0.93) (1.85) (6.59) (2.74) (0.40) (4.62) (2.87) (2.85) Post Secondary 5.601*** .. 0.175*** 11.64*** 0.195*** 0.676 9.072*** 0.091*** 0.524*** (5.41) (3.57) (7.69) (3.22) (1.59) (8.73) (4.24) (2.63) Apprentice - Low educ 1.819*** 0.749 1.383* 1.691** 0.657 1.329* 1.054 0.746 1.283* (2.62) (1.16) (1.89) (2.29) (1.59) (1.88) (0.29) (1.50) (1.71) Apprentice - Med educ 0.909 0.671 2.058*** 0.737 0.703 1.927*** 0.743** 0.656* 1.573*** (0.50) (1.38) (4.00) (1.53) (1.12) (4.22) (2.24) (1.77) (4.09) Apprentice - High educ 1.617 .. 1.401 0.966 1.110 3.349*** 1.099 2.429* 4.678*** (1.09) (0.65) (0.09) (0.13) (3.37) (0.31) (1.75) (6.25) # kids aged 0-6 1.208 1.169 1.247* 0.605** 1.188 0.700*** 0.796* 1.074 1.059 (1.29) (0.72) (1.78) (2.52) (0.85) (3.07) (1.73) (0.36) (0.51) # kids 0-6 squared 0.976 0.969 0.948 1.085 0.953 1.092** 1.065 0.946 0.990 (0.57) (0.52) (1.51) (1.06) (0.61) (2.16) (1.15) (0.80) (0.23) # kids aged 7-14 0.854 0.767** 0.896 1.120 0.976 1.074 0.835** 0.916 0.944 (1.40) (1.97) (1.50) (0.76) (0.20) (0.78) (2.25) (0.57) (0.72) # kids 7-14 squared 1.005 1.033 1.028** 0.941 0.981 0.991 1.017 1.003 1.003 (0.20) (1.44) (2.24) (1.48) (0.51) (0.36) (0.81) (0.07) (0.12) Age 1.436*** 1.277*** 1.288*** 1.385 1.245*** 1.291*** 1.416*** 1.472*** 1.462*** (7.02) (3.47) (5.03) (6.57)*** (4.37) (7.11) (9.20) (6.63) (11.77) Age squared 0.996*** 0.997*** 0.997*** 0.996*** 0.998*** 0.997*** 0.996*** 0.996*** 0.996*** (6.52) (3.12) (4.94) (6.22) (3.77) (7.12) (9.26) (6.15) (11.62) Married 1.853*** 1.573** 1.570*** 1.867*** 0.828 1.567*** 1.257** 1.368* 1.201** (4.16) (1.98) (3.69) (3.88) (1.23) (4.05) (2.16) (1.85) (2.06) Male 3.642*** 4.024*** 0.564*** 3.566*** 3.872*** 0.616*** 4.435*** 4.400*** 0.850* (7.36) (6.01) (4.39) (8.76) (6.59) (3.65) (15.05) (10.57) (1.85) Accra 1.421 0.024*** 1.811** 1.583* 0.241** 1.749** 2.168*** 0.043*** 1.387 (1.10) (3.68) (2.04) (1.85) (2.30) (2.19) (3.36) (5.96) (1.50) Forest 1.122 1.195 1.512 1.346 2.080** 1.530* 1.615** 1.610 1.513** (0.39) (0.39) (1.48) (1.23) (1.99) (1.70) (2.11) (1.49) (2.03) Coastal 1.085 1.103 1.549 0.974 1.208 1.254 1.496 1.160 1.353 (0.28) (0.18) (1.50) (0.10) (0.40) (0.80) (1.50) (0.38) (1.26) Father_primary 0.714 0.718 0.734 1.332 0.667 0.984 0.967 1.150 0.803 (1.08) (0.76) (1.17) (1.05) (0.92) (0.07) (0.17) (0.58) (1.17) Father_middle school 0.988 0.437*** 0.766* 0.974 0.448*** 0.790** 1.009 0.703* 1.092 (0.07) (2.89) (1.68) (0.19) (3.47) (2.23) (0.09) (1.94) (0.92) Father_post middle 0.971 0.097** 0.838 0.928 0.494* 0.880 0.785 0.346*** 0.911 (0.13) (2.29) (0.74) (0.32) (1.77) (0.75) (1.70) (3.25) (0.70) # of observations 2247 3104 5006 Source: Authors using GLSS data (t-statistics in parenthesis). 13 Table 3: Multinomial Logits for Type of employment, Ghana 1991-2006 (rural areas) Rural ­ age group 25-64 1991/92 1998-99 2005/06 Wage Self Self Wage Self Self Wage Self Self Earner Agro Non Agro Earner Agro Non Agro Earner Agro Non Agro Some Primary 2.275*** 1.446** 1.648*** 0.556* 0.905 1.310* 1.851*** 1.230* 2.261*** (3.29) (2.31) (2.78) (1.67) (0.79) (1.95) (2.74) (1.84) (5.65) Primary 1.437 1.382** 1.413* 1.021 1.086 1.447** 2.188*** 1.150 1.794*** (1.19) (2.12) (1.75) (0.06) (0.62) (2.40) (3.23) (1.05) (3.91) Secondary (lower) 4.055*** 1.262 1.735*** 3.278*** 0.940 1.497** 5.280*** 1.142 2.418*** (4.74) (1.42) (2.85) (3.38) (0.46) (2.56) (6.91) (1.00) (5.58) TVET 3.971* 0.525 1.486 2.201 0.588 0.638 20.12*** 0.830 10.54*** (1.94) (1.35) (0.72) (1.11) (1.48) (0.68) (6.67) (0.48) (6.10) Secondary (higher) 4.866*** 0.426* 1.101 6.354*** 0.636* 1.165 13.61*** 0.698 2.092* (2.83) (1.78) (0.17) (4.42) (1.71) (0.43) (8.30) (1.35) (1.94) Post Secondary 12.50*** 0.082*** 0.294 25.14*** 0.356*** 0.594 68.60*** 0.194*** 1.884 (6.83) (3.88) (1.54) (8.34) (3.19) (1.28) (11.71) (3.91) (1.50) Apprentice - Low educ 1.047 0.780 1.277 1.864** 1.069 1.706*** 4.061*** 1.165 2.142*** (0.17) (1.63) (1.50) (2.11) (0.49) (3.64) (6.63) (1.30) (5.31) Apprentice - Med educ 0.984 0.838 1.647** 0.743 0.887 1.567*** 1.425* 0.822 1.901*** (0.07) (0.92) (2.43) (1.55) (0.78) (3.04) (1.81) (1.32) (3.33) Apprentice - High educ 0.346 0.227 2.375 0.673 1.190 2.284* 1.241 1.952 2.948 (1.31) (1.38) (0.92) (0.91) (0.38) (1.83) (0.37) (1.04) (1.47) # kids aged 0-6 0.758* 1.258*** 0.908 0.842 1.052 1.087 0.823 0.972 0.941 (1.90) (2.64) (1.02) (1.30) (0.64) (0.96) (1.49) (0.46) (0.72) # kids 0-6 squared 1.031 0.938*** 1.019 1.053 1.005 0.968 1.031 0.999 0.982 (0.97) (3.52) (0.92) (1.41) (0.24) (1.17) (0.93) (0.07) (0.87) # kids aged 7-14 0.859 0.975 0.918 1.035 1.073 0.968 0.769*** 0.927 0.869* (1.15) (0.33) (0.93) (0.32) (0.97) (0.45) (2.64) (1.24) (1.81) # kids 7-14 squared 1.013 0.979 1.004 0.986 0.982 1.002 1.044* 1.016 1.027 (0.51) (1.46) (0.19) (0.58) (1.09) (0.14) (1.92) (1.12) (1.55) Age 1.485*** 1.175*** 1.127** 1.429*** 1.231*** 1.204*** 1.319*** 1.183*** 1.149*** (6.56) (4.35) (2.50) (5.76) (7.36) (4.92) (5.17) (6.74) (3.98) Age squared 0.996*** 0.998*** 0.999** 0.996*** 0.998*** 0.998*** 0.997*** 0.998*** 0.998*** (6.24) (3.60) (2.46) (5.58) (6.49) (5.02) (5.16) (6.18) (4.17) Married 1.189 0.947 1.136 1.281 0.835** 0.970 0.778* 1.167* 0.981 (0.79) (0.39) (0.74) (1.23) (2.06) (0.27) (1.67) (1.69) (0.20) Male 14.76*** 7.560*** 0.992 6.723*** 3.974*** 0.944 7.284*** 4.393*** 0.481*** (10.88) (13.19) (0.05) (9.12) (12.62) (0.44) (12.70) (13.97) (6.24) Accra .. .. .. .. .. .. .. .. .. Forest 2.687*** 1.930*** 0.859 4.214*** 2.316*** 1.441* 2.896*** 2.739*** 1.494** (4.03) (3.89) (0.70) (4.55) (4.12) (1.66) (4.42) (5.68) (2.21) Coastal 3.203*** 1.792*** 1.667** 7.479*** 1.554* 2.978*** 7.127*** 3.925*** 3.438*** (3.88) (2.74) (2.11) (6.61) (1.87) (4.37) (6.39) (6.22) (5.65) Father_primary 0.683 0.752 1.152 1.325 0.887 0.921 1.321 1.042 1.277 (1.11) (1.17) (0.54) (0.90) (0.61) (0.34) (0.98) (0.21) (1.33) Father_middle school 0.903 0.547*** 1.058 1.222 0.715*** 0.927 0.985 1.113 1.251* (0.40) (3.35) (0.33) (1.12) (2.73) (0.57) (0.09) (0.99) (1.68) Father_post middle 1.167 0.578 0.624 1.492 0.718 1.663** 1.050 0.850 0.756 (0.38) (1.63) (1.03) (1.54) (1.39) (2.32) (0.16) (0.81) (1.10) # of observations 4201 5567 7973 Source: Authors using GLSS data (t-statistics in parenthesis). 14 Table 4: Wage Regressions with Sample Selection, Ghana 1991-2006 (urban areas) Urban ­ age group 25-64 1991/92 1998-99 2005/06 Wage Self Self Wage Self Self Wage Self Self Non Non Non Earner Agro Agro Earner Agro Agro Earner Agro Agro Some Primary -0.240 -0.222 -0.125 0.027 0.122 -0.238 -0.004 0.183 0.077 (0.70) (0.59) (0.68) (0.08) (0.39) (1.70) (0.02) (0.76) (0.62) Primary 0.047 0.740 0.161 0.597 0.564 0.025 0.128 0.161 0.018 (0.16) (1.64) (0.72) (1.63) (1.56) (0.17) (0.80) (0.60) (0.13) Secondary Lower 0.480 0.143 -0.014 0.537 0.833 0.143 0.329* 0.955** 0.221 (1.57) (0.25) (0.04) (1.65) (1.44) (0.73) (1.93) (2.39) (1.47) TVET 0.810 -0.199 -0.568 0.000 .. 0.000 0.848*** 1.494 0.344 (1.57) (0.19) (0.74) (0.00) (0.00) (3.73) (1.48) (1.53) Secondary Higher 0.916* 0.200 0.307 1.036** 2.275** 0.829** 0.839*** 0.572 0.448* (1.93) (0.16) (0.42) (2.53) (2.11) (2.52) (3.57) (0.80) (1.88) Post Second. 1.457* .. 1.650** 1.182** 2.210 0.907** 1.904*** 0.971 0.771* (1.81) (2.23) (2.40) (1.54) (2.00) (4.46) (0.59) (1.81) - Apprentice - low 1.028* educ -0.134 * -0.184 -0.239 -0.638* -0.068 -0.090 0.238 0.141 (0.49) (2.38) (0.98) (0.77) (1.90) (0.53) (0.58) (0.87) (1.27) Apprentice - med educ -0.283 -0.626 0.016 -0.246 -0.881 -0.064 -0.082 -0.536 -0.089 (1.08) (1.06) (0.06) (1.19) (1.63) (0.40) (0.63) (1.26) (0.71) - Apprentice - high 0.643** educ -0.021 .. * -0.351 -1.437 -0.081 -0.459** 0.419 -0.244 (0.12) (3.40) (1.16) (1.08) (0.33) (2.11) (0.48) (0.83) 0.070* Experience * -0.089 -0.004 0.037* -0.008 0.057*** 0.041*** 0.038 0.028 (2.05) (1.38) (0.12) (1.85) (0.13) (2.61) (2.73) (0.69) (1.46) Experience squared -0.001 0.002 0.000 0.000 0.000 -0.001*** -0.001** -0.001 0.000 (1.50) (1.74)* (0.11) (1.30) (0.50) (2.75) (2.39) (0.97) (1.38) Married 0.185 -0.181 -0.058 0.030 0.017 0.096 -0.010 0.259 -0.015 (0.98) (0.60) (0.34) (0.29) (0.06) (0.96) (0.15) (1.39) (0.23) Male 0.186 1.258 0.034 0.353 1.290* 0.518** 0.414* 0.663 0.155 (0.48) (1.38) (0.07) (1.25) (1.64) (2.00) (1.78) (0.92) (0.63) 0.420* Formal job * -0.265 -0.654** 0.217* 0.211 0.063 0.180*** .. -0.421 (2.36) (0.52) (2.36) (1.75) (0.41) (0.44) (2.81) (1.16) Union -0.003 .. 0.081 0.317*** .. 0.039 0.144*** .. -0.183 (0.03) (0.32) (4.51) (0.05) (2.86) (0.18) Public 0.097 .. .. 0.078 .. .. 0.095 .. .. (0.85) (1.12) (1.68) Accra 0.435 -1.617 0.391 0.462*** -0.421 0.218 0.499*** -0.390 -0.001 (1.38) (1.25) (1.29) (3.16) (0.61) (1.41) (3.57) (0.30) (0.00) Forest 0.116 -0.308 0.054 0.236* 0.130 -0.129 0.183 -0.138 0.193* (0.65) (0.91) (0.32) (1.87) (0.39) (1.01) (1.84) (0.60) (1.80) Coastal -0.017 -0.075 0.261 0.154 -0.392 -0.310** 0.253** -0.140 0.116 (0.09) (0.21) (1.55) (1.30) (1.48) (2.36) (2.32) (0.53) (0.96) _m1 -0.149 2.292 -0.645 0.129 1.152 -0.999 -0.204 3.158 -1.090 (0.30) (0.95) (0.40) (0.37) (0.49) (0.85) (0.74) (1.55) (1.20) - _m2 -0.499 0.423 -0.852 0.234 -0.018 -1.308 1.413*** 0.313 -0.520 (0.57) (0.47) (0.61) (0.33) (0.02) (1.30) (3.41) (0.39) (0.81) _m3 -0.619 -1.246 0.034 -0.474 -3.522 -0.945** -1.450* 1.404 -0.180 (0.41) (0.34) (0.05) (0.55) (1.32) (2.12) (1.87) (0.58) (0.46) 5.347* - _m4 -0.996 * 0.299 -0.237 -0.358 -2.334** 1.462*** -0.426 0.149 (0.82) (2.17) (0.20) (0.33) (0.18) (2.44) (2.86) (0.25) (0.22) 5.930* 4.310** Constant 1.915 * * 4.149*** 3.002 4.605 5.281*** 7.067*** 7.007*** (1.45) (1.96) (3.78) (4.42) (1.25) (6.21) (6.62) (3.40) (11.81) # of observations 631 186 817 640 315 1209 1413 408 1745 R2 0.289 0.067 0.071 0.246 0.090 0.089 0.319 0.064 0.077 Source: Authors using GLSS data (t-statistics in parenthesis). 15 Table 5: Wage Regressions with Sample Selection, Ghana 1991-2006 (rural areas) Rural ­ age group 25-64 1991/92 1998-99 2005/06 Wage Self Self Wage Self Self Wage Self Self Non Non Earner Agro Non Agro Earner Agro Agro Earner Agro Agro Some primary 0.075 0.137 0.283 0.526** 0.225** -0.017 -0.110 -0.044 0.309 (0.42) (1.01) (1.14) (2.22) (2.35) (0.11) (0.47) (0.44) (1.26) Primary 0.128 0.253** 0.382 0.308 0.071 0.118 -0.181 0.097 0.142 (0.67) (2.02) (1.43) (1.26) (0.66) (0.76) (0.82) (0.96) (0.69) 0.529* Secondary Lower * 0.654*** 0.525 0.399 0.177 0.363* 0.341 0.288** 0.323 (2.18) (3.42) (1.55) (1.48) (1.20) (1.80) (1.06) (2.25) (1.18) TVET 1.075 1.356** 0.691 0.612 0.031 0.234 0.321 0.371 0.772 (1.28) (2.17) (1.05) (1.00) (0.07) (0.53) (0.55) (0.73) (0.92) Secondary Higher 0.834* 1.870*** 0.175 0.482 0.061 0.880* 0.628 0.728** 0.450 (1.89) (3.73) (0.21) (1.08) (0.17) (1.74) (1.21) (2.11) (0.91) Post Second. 1.262* 2.214** 0.199 0.352 0.326 1.102* 1.770** 1.959** 1.226 (1.66) (2.18) (0.17) (0.47) (0.59) (1.82) (2.09) (2.28) (1.49) Apprentice - low educ 0.029 -0.038 0.255 0.210 0.083 -0.118 0.349 -0.053 0.230 (0.17) (0.35) (1.20) (0.93) (0.76) (0.87) (1.28) (0.39) (0.82) Apprentice - med educ -0.123 -0.075 0.177 -0.204 -0.007 -0.013 -0.401* -0.057 0.046 (0.65) (0.53) (0.69) (1.02) (0.06) (0.07) (1.90) (0.32) (0.14) Apprentice - high educ -0.441 -2.009** 0.780 0.058 -0.062 0.271 -0.295 -0.468 0.167 (0.86) (2.00) (0.77) (0.21) (0.15) (0.51) (1.06) (0.61) (0.35) Experience 0.043* 0.074*** 0.044 0.034 -0.032 0.080*** 0.020 0.027 0.002 (1.72) (3.73) (1.23) (1.41) (1.61)* (2.76) (1.05) (1.26) (0.06) - - Experience squared -0.001* 0.001*** 0.000 0.000 0.000 0.001*** 0.000 0.000 0.000 (1.76) (3.54) (1.01) (1.01) (1.92)* (3.21) (0.57) (1.18) (0.45) Married 0.112 -0.040 0.167 0.027 0.114 0.175* 0.134 0.184** -0.052 (0.95) (0.52) (1.25) (0.19) (1.64)* (1.81) (1.06) (2.06) (0.36) Male 0.297 1.076*** 0.144 0.247 -0.193 0.272 0.697 0.573 -0.426 (0.65) (3.40) (0.26) (0.76) (0.93) (0.75) (1.62) (1.35) (0.47) Formal job -0.063 0.085 0.039 0.125 -0.195 -0.022 -0.141 0.398 -0.647 (0.33) (0.31) (0.27) (0.87) (0.94) (0.21) (1.18) (0.49) (1.40) Union 0.169 .. .. 0.294** .. 1.893** 0.179* .. -0.758* (1.56) (2.20) (2.19) (1.68) (1.91) Public 0.077 .. .. -0.167 .. .. 0.158 .. .. (0.66) (1.38) (1.40) Accra .. .. .. .. .. .. .. .. .. Forest 0.171 0.488*** 0.385* 0.072 0.152 0.051 0.375** 0.232 -0.165 (0.87) (3.87) (1.75) (0.32) (1.19) (0.25) (1.97) (0.97) (0.44) 0.763** Coastal 0.312 0.171 * -0.196 -0.428** 0.091 0.476* 0.047 -0.475 (1.51) (1.14) (2.88) (0.59) (2.23) (0.39) (1.81) (0.18) (1.04) _m1 0.297 1.482 0.932 -0.050 -1.097 1.871 -0.331 0.509 -0.577 (1.11) (1.57) (0.68) (0.13) (1.17) (1.44) (0.87) (0.56) (0.51) _m2 0.402 -0.253 0.568 1.677* -0.556 -0.598 -1.361 -0.634 -3.164* (0.40) (0.76) (0.44) (1.90) (1.53) (0.50) (1.58) (1.62) (1.65) _m3 0.320 -1.350 0.710 1.096 0.138 0.102 -2.449* -1.303 -0.248 (0.18) (1.19) (1.23) (0.91) (0.15) (0.24) (1.79) (1.45) (0.44) _m4 -0.102 -1.179 -0.987 -0.655 1.251 0.426 -2.189** -1.186 -0.148 (0.09) (1.03) (0.51) (0.63) (1.46) (0.38) (2.37) (1.12) (0.11) 3.043* 6.128** 6.589** 5.022** 5.847** Constant * 1.199 1.912 * * 4.519*** * * 6.922*** (1.97) (1.35) (0.99) (4.13) (8.15) (4.15) (3.70) (3.85) (5.21) # of observations 350 1838 633 394 2353 968 488 2976 1060 R2 0.200 0.088 0.042 0.217 0.107 0.077 0.215 0.056 0.053 Source: Authors using GLSS data (t-statistics in parenthesis). 16 Table 6: Wage Regressions without Sample Selection, Ghana 1991-2006 (urban areas) Urban ­ age group 25-64 1991/92 1998/99 2005/06 Wage Self Self Wage Self Self Wage Self Self Non Earner Agro Agro Earner Agro Non Agro Earner Agro Non Agro Some Primary -0.250 0.088 -0.109 0.038 0.208 -0.236** -0.028 0.066 0.100 (0.80) (0.50) (0.74) (0.12) (0.83) (2.02) (0.17) (0.35) (0.87) Primary 0.072 0.350 0.181 0.548 0.383 -0.003 0.148 0.074 0.047 (0.34) (1.39) (1.17) (1.46) (1.37) (0.02) (0.95) (0.29) (0.35) Secondary Lower 0.415*** -0.145 0.095 0.417** 0.424 0.057 0.188 0.556** 0.340*** (2.70) (0.41) (0.61) (2.00) (1.46) (0.48) (1.59) (2.23) (3.02) TVET 0.539*** -0.164 -0.213 0.525** 0.137 0.584*** 1.037* 0.515*** (3.27) (0.55) (0.85) (2.26) (0.78) (4.53) (1.83) (3.60) Secondary Higher 0.725*** -0.120 0.499* 0.838*** 1.516** 0.460** 0.585*** 0.006 0.581*** (4.39) (0.18) (1.86) (3.86) (2.53) (2.38) (4.40) (0.01) (3.71) Post Second. 1.183*** 1.719*** 0.911*** 0.939 0.231 1.302*** 0.421 1.001*** (7.12) (5.44) (4.42) (1.51) (1.03) (10.01) (0.54) (5.28) Apprentice - low - educ -0.165 0.756*** -0.186 -0.247 -0.474* -0.088 -0.046 0.206 0.127 (0.83) (3.26) (1.38) (0.79) (1.91) (0.77) (0.34) (1.10) (1.13) Apprentice - med - educ 0.277*** -0.061 -0.053 -0.148 -0.318 -0.089 0.031 -0.396* -0.138* (3.27) (0.22) (0.44) (1.43) (0.89) (0.88) (0.46) (1.85) (1.74) Apprentice - high educ -0.183 0.237 -0.241** -0.568 0.041 -0.284*** 0.457 -0.221 (1.34) (0.79) (2.03) (0.69) (0.26) (3.13) (0.72) (1.29) Experience 0.047*** -0.003 0.010 0.029** 0.003 0.035*** 0.030*** -0.013 0.047*** (3.15) (0.05) (0.59) (2.11) (0.12) (2.76) (2.90) (0.33) (4.06) Experience squared -0.001* 0.000 0.000 0.000 0.000 -0.001*** 0.000** 0.000 -0.001*** (1.91) (0.47) (0.42) (1.58) (0.14) (2.91) (2.17) (0.18) (3.74) Married 0.076 0.235 0.016 -0.005 0.180 0.013 -0.011 0.206 0.007 (0.83) (1.26) (0.15) (0.06) (1.02) (0.17) (0.18) (1.36) (0.12) Male 0.090 0.596** 0.335*** 0.143** 0.368** 0.389*** 0.147*** 0.201 0.406*** (1.18) (2.38) (3.51) (2.26) (2.40) (4.91) (3.05) (1.15) (7.22) Formal job 0.427*** -0.188 -0.601** 0.218* 0.163 0.074 0.201*** -0.362 (3.44) (0.55) (2.19) (1.83) (0.30) (0.44) (2.94) (1.19) Union -0.029 0.149 0.317*** 0.046 0.159*** -0.254 (0.38) (0.45) (4.55) (0.06) (3.09) (0.26) Public 0.120 0.106 0.091 (0.95) (1.28) (1.52) - Accra 0.228** 1.722*** 0.126 0.434*** 0.012 0.061 -0.056 -0.957* -0.240* (2.48) (6.06) (0.74) (3.90) (0.03) (0.48) (0.54) (1.77) (1.83) Forest 0.072 -0.030 0.002 0.203* 0.168 -0.170 0.125 -0.372 0.244* (0.61) (0.08) (0.01) (1.69) (0.43) (1.20) (1.20) (1.21) (1.94) Coastal -0.044 0.143 0.225 0.186 -0.313 -0.319* 0.209* -0.303 0.174 (0.41) (0.39) (1.30) (1.60) (1.02) (1.92) (1.75) (0.87) (1.16) Constant 6.364*** 6.547*** 7.473*** 6.430*** 6.335*** 7.374*** 7.099*** 7.761*** 6.958*** (28.27) (8.03) (22.63) (22.71) (11.58) (30.98) (39.44) (10.76) (31.90) # of observations 1413 408 1745 R2 0.319 0.064 0.077 Source: Authors using GLSS data (t-statistics in parenthesis). 17 Table 7: Wage Regressions without Sample Selection, Ghana 1991-2006 (rural areas) Rural ­ age group 25-64 1991/92 1998/99 2005/06 Wage Self Self Wage Self Self Wage Self Self Non Earner Agro Agro Earner Agro Non Agro Earner Agro Non Agro Some primary -0.017 0.059 0.056 0.480** 0.134 0.003 -0.016 -0.015 0.364** (0.11) (0.68) (0.32) (2.35) (1.36) (0.02) (0.08) (0.18) (2.50) Primary 0.104 0.182 0.203 0.267 0.126 0.144 -0.120 0.091 0.275* (0.60) (2.00) (0.97) (1.25) (1.16) (1.08) (0.78) (1.01) (1.94) Secondary Lower 0.367** 0.336 0.185 0.432*** 0.329** 0.214 0.432** 0.200** 0.460*** (2.37) (3.24) (0.98) (2.78) (2.50) (1.47) (2.31) (2.05) (2.76) TVET 0.937*** 0.928 0.411 0.776 0.030 0.125 0.472** -0.035 0.452 (3.03) (1.78) (0.82) (1.71) (0.08) (0.33) (2.26) (0.15) (1.57) Secondary Higher 0.624*** 1.060 -0.148 0.770*** 0.275 0.324 0.637*** 0.326* 0.391 (3.53) (3.66) (0.26) (3.90) (1.12) (1.02) (3.16) (1.87) (1.20) Post Second. 1.023*** 0.846 0.244 1.086*** 0.675*** 0.148 1.230*** 0.745* 0.221 (7.30) (1.54) (0.41) (5.92) (2.65) (0.82) (6.26) (1.76) (0.70) Apprentice - low educ 0.020 0.001 0.176 0.145 0.106 -0.164 0.455*** -0.155* 0.208 (0.14) (0.01) (1.03) (0.63) (1.21) (1.55) (3.22) (1.88) (1.50) Apprentice - med educ -0.131 -0.059 -0.020 -0.253* -0.088 -0.006 -0.256 -0.108 -0.152 (0.90) (0.58) (0.15) (1.67) (0.98) (0.05) (1.84) (0.97) (1.11) Apprentice - high educ -0.302 -1.293 0.605 -0.137 -0.065 0.533 -0.200 -0.298 0.171 (1.14) (4.46) (0.81) (0.74) (0.23) (1.50) (1.09) (0.43) (0.54) Experience 0.029* 0.038 0.017 0.019 0.014 0.083*** 0.018 0.025** 0.036** (1.95) (2.86) (0.73) (1.36) (1.17) (5.07) (1.14) (2.04) (1.97) Experience squared 0.000* 0.000 0.000 0.000 0.000 -0.001*** 0.000 0.000 -0.001* (1.79) (2.58) (0.42) (0.95) (0.29) (4.58) (0.87) (1.45) (1.94) Married 0.136 -0.012 0.166 0.064 0.084 0.129 0.128 0.209*** 0.029 (1.26) (0.16) (1.31) (0.52) (1.23) (1.37) (1.21) (2.73) (0.26) Male 0.004 0.470 0.031 0.073 0.386*** 0.329*** 0.165 0.436*** 0.220** (0.04) (6.50) (0.26) (0.67) (5.53) (3.35) (1.48) (6.15) (2.19) Formal job -0.047 0.108 0.046 0.145 -0.193 -0.010 -0.149 0.510 -0.559 (0.25) (0.41) (0.20) (0.86) (0.89) (0.08) (1.27) (0.62) (1.21) Union 0.169 0.288** 1.945*** 0.207** -0.805*** (1.59) (1.96) (3.15) (1.99) (5.30) Public 0.076 -0.184 0.142 (0.65) (1.10) (1.19) Accra Forest 0.020 0.195 0.287* -0.092 0.427*** 0.025 0.069 0.167 0.250* (0.21) (1.87) (1.67) (0.62) (2.86) (0.16) (0.54) (1.43) (1.65) Coastal 0.076 -0.123 0.348* -0.275* -0.302 -0.159 0.084 -0.102 -0.060 (0.63) (0.93) (1.66) (1.76) (1.61) (1.01) (0.58) (0.78) (0.36) Constant 7.099*** 6.031*** 6.935*** 7.129*** 6.072*** 6.163*** 7.276*** 6.467*** 6.984*** (22.29) (24.14) (16.24) (25.83) (23.82) (18.93) (28.55) (28.83) (21.23) # of observations 488 2976 1060 R2 0.215 0.056 0.053 Source: Authors using GLSS data (t-statistics in parenthesis). 18 ANNEX 2: Apprenticeship in Ghana8 In Ghana there is a highly developed apprenticeship system where young men and women undertake sector specific training which is paid for usually by those responsible for the apprentice and which yields skills used primarily in the informal sector. In this annex we use a recent urban based household survey with detailed questions on the background, training and earnings of workers in both wage and self-employment to ask how apprenticeship compares with other forms of training in terms of pay and employment outcomes. We show that apprenticeship is by far the most important institution providing training and is undertaken primarily by those with junior secondary school or lower levels of education. In comparing those with some form of training to those with none, training to be a nurse or teacher gives by far the highest return, a three fold increase in earnings. In contrast those who have done an apprenticeship earn significantly less than those with no training. Once an allowance is made for the level at which the apprentice enters the system there is evidence that for those with the lowest level of education apprenticeship does lead to a substantial increase in earnings, some 41 per cent. For those with higher levels of education it does not lead to any increase at all. In fact the point estimates imply a decrease. We argue that the reasons for undertaking an apprenticeship can be found, in part, in the powerful effect undertaking an apprenticeship has in increasing the probability of informal employment relative to having no job. In contrast to these outcomes for apprentices they show that training to be a nurse or teacher not only pays off in earnings but substantially shifts the probability of being in formal relative to informal employment. In summary these forms of training, one undertaken in the private sector and the other in the public sector, are associated with radically different outcomes for those receiving the training. 1 Introduction 2.1 Skills training in Ghana occur in both the private and public sectors. By far the most important institution which provides such training in the private sector is the apprenticeship system. Apprentices are young men and women who undertake highly sector specific training. Some of these apprentices then go on to form their own businesses; others go on to work in the firm in which they were apprentices as masters, some move to other firms or occupations. It matters where apprentices go as incomes differ substantially across these different outcomes, Sandefur, Serneels and Teal (2007). While much is known about the institution in terms of its structures and forms we know much less about how well apprenticeship pays relative to other forms of training and relative to more academic education. Frazer (2006) explores the institution of apprenticeship in Ghana and argues that it is a form of skill acquisition which pays off in self-employment if the apprentice acquires 8. This background paper was prepared by: Courtney Monk, Justin Sandefur and Francis Teal, from Centre for the Study of African Economies, Department of Economics, University of Oxford, August 2007. The data used in this annex were collected by the Centre for the Study of African Economies, Oxford, in collaboration with the Ghana Statistical Office (GSO) in 2006. The survey was funded in part by the Department for International Development (DfID) of the UK as part of its work on assessing the outcomes of education and the Economic and Social Research Council (ESRC) as part of the Global Poverty Research Group. We have been assisted by numerous collaborators in enabling us to collect this data. We are also greatly indebted to Moses Awoonor-Williams, Geeta Kingdon and Andrew Zeitlin for their assistance in the design and implementation of the survey. 19 sufficient capital to start their own business. It is clear from his analysis, and the work that has been carried out on enterprises in Ghana, that apprenticeship does not pay-off in the wage sector. Thus to establish the effect of apprenticeship it is essential to be able to observe individuals in both the wage and self-employment sectors. 2.2 Over the period 2004 to 2006 the CSAE in conjunction with the Ghana Statistical Office has carried out an urban based labor market survey which in its most recent round had very detailed questions concerning the skills and training that the individuals had received. The survey also sought to measure the incomes of the self-employed with as much accuracy as possible in a manner that would allow incomes to be compared across the formal and informal sectors. This survey is a follow-up to two earlier surveys carried out in 2004 and 2005. In this note the data on apprentices from the 2006 round of this survey is used to address a range of questions about the background of apprentices and the outcomes of their training. We leave to later work linking this data to the earlier rounds of the survey. The questions we address are: 1. How important is apprenticeship as a form of training? 2. What is the educational background and occupational outcomes of the apprentices? 3. Does being an apprentice pay? 4. Do the occupational outcomes from apprenticeship differ from that of other forms of training? 2.3 These four questions provide the framework for the paper and will be addressed in the next four sections. While we address questions specific to Ghanaian apprenticeship the analysis links to the long history in Ghana, and elsewhere, of the relative value of academic relative to vocational education, Foster (1965a,b) and to the policy debate as to how public provision compares with the private provision of training, Middleton, Ziderman, and van Adams (1993). 2 How important is apprenticeship as a form of training? 2.4 We begin by asking how important any form of training among the urban population is. As Table 1 shows within our sample of people aged 15 to 65, including those both in and outside the labor force, 33 per cent have had some form of training either as an apprentice or attending some vocational or technical school. Table 1: Training in Ghana in 2006 Number of % of total observations No formal training 1099 67 Any apprentice/vocational/technical training, past or current 544 33 Total individuals, excluding children and the elderly 1643 2.5 In the analysis that follows we are going to identify four kinds of training which occur outside the main academic educational stream. These are firstly attending a vocational or technical school, secondly undertaking an apprenticeship, thirdly being trained on-the-job and finally being trained as a teacher or nurse. The reason for separately identifying training as a teacher or nurse will be apparent from the results that we present below. Many 20 individuals do more than one form of training so in Table 2 we present the number of training events in the data, i.e. one of the training activities identified in the survey. Table 2: Types of Training Number % of total Current apprentice 122 15 Past apprentice 317 40 Current vocational trainee 16 2 Past vocational trainee 112 14 Current on-the-job trainee 40 5 Past on-the-job trainee 158 20 Trained teacher/nurse 25 3 Total number of training events 790 100 2.6 In Table 2 we identify both current and past training. It is clear that apprenticeship is by far the most common form of training, 16 per cent of the training events in the survey are current apprenticeship while 41 per cent are past apprenticeships. The second most important form of training in that classified as on-the-job, followed by vocation training, excluding that for teachers and nurses, who constitute 3 per cent of the training events. 3 Educational background and occupational outcomes of the apprentices? 2.7 In Table 3 we present the education background of the individuals in the sample and for apprentices in order to asses how their educational patterns differ. Table 4 presents a similar breakdown for occupational outcomes. Table 3: Educational Background Complete sample Number % of total No education (years<6) 226 14 Primary (years between 6 and 9) 218 13 Middle (9 or 10 years -- jss or middle) 896 55 Secondary 283 17 Post secondary (strictly academic) 13 1 Polytechnic 7 0 Total 1643 Individuals who did an apprenticeship in the Number % of group past No education (years<6) 29 9 Primary (years between 6 and 9) 32 10 Middle (9 or 10 years -- jss or middle) 233 74 Secondary 23 7 Post secondary (strictly academic) 0 0 Polytechnic 0 0 Total 317 21 2.8 It is clear from Table 3 that by far the most common pattern for apprentices is to enter it after the end of junior high school, which under the old education system was the end of middle school. Of those individuals in the sample who had done an apprenticeship in the past 74 per cent entered at the junior high school level. However it will prove to be of importance for the results which will be shown below to note that while this pattern is by far the most common there are different paths. Some 9 per cent had done an apprenticeship with no education. A comparison of the educational background of those who did an apprenticeship with the whole sample shows that it is those with junior secondary or below who are more likely to be apprentices than those with higher levels of education. Indeed there is nobody in the sample who undertook an apprenticeship who completed a post secondary qualification. Table 4: Occupational Outcomes in 2006 Complete sample Number % of total Self-employed 549 33 Small firm 248 15 Large firm 169 10 Public sector 64 4 No earned income 613 37 Total 1643 Individuals who did an apprenticeship in the Number % of total past Self-employed 181 57 Small firm 52 16 Large firm 30 9 Public sector 8 3 No earned income 46 15 Total 317 Note: this includes those who also did, in addition, other types of training. 2.9 It is clear from Table 4 that the most common pattern for apprentices is self- employment. However it will prove to be of importance for the results which will be shown below to note that some 15 per cent of the sample, who had done an apprenticeship, currently had no income. A comparison of the educational background of those who did an apprenticeship with the whole sample shows that it is those with junior secondary or below who are more likely to be apprentices than those with higher levels of education. 4 Does being an apprentice pay? 2.10 In order to answer the question does apprenticeship pay it is clearly crucial to be able to measure self-employment income. Our data is taken from a longitudinal labor market survey conducted by the Centre for the Study of African Economies (CSAE) at Oxford University, under the direction of the authors and in collaboration with the Ghana Statistical Office (GSO). The urban panel survey (UPS) collects information on incomes, education and labor market experience, household characteristics and various other modules for labor force participants (ages 15 to 60) in urban areas. For Ghana these areas span the four largest urban centers in the country: Accra (and neighboring Tema), Kumasi, Takoradi and Cape Coast. The samples were based on a stratified random sample of urban households from the 2000 22 census in Ghana.9 While the initial sample was household based, interviews were conducted on an individual basis, and the unit of analysis in what follows will be at the individual level. A total of 830 were interviewed in the first round of the survey in Ghana, which was conducted between October 2003 and June 2004. 2.11 Collecting income data on the self-employed in low-income countries is a controversial endeavor. Field guides for the World Bank's Living Standards Measurement Surveys (LSMS), which serve as the international standard for household surveys in development economics, recommend survey managers not collect this information. The stated rationale is that self-employed business people in the informal sector rarely keep written accounts and their self-reported income data may be too noisy to be of use. For household based enterprises, the distinction between business and personal expenditures may be completely alien to respondents. We acknowledge the validity of these concerns. 2.12 However, because the non-agricultural self-employed constitute a majority of the urban working population in Ghana, we feel measuring such incomes are essential to our current objective of understanding the impact of apprenticeship on welfare. Our income measure for the self-employed is based on self-reported profits. Profits are net of routine operating expenses and gross of fixed capital expenditure, if any. The concepts of ``revenue'', ``business costs'', and ``profits'' are explained to respondents by enumerators with experience in conducting firm and household surveys. As the surveys are entered directly onto handheld computers, a simple mechanical check forces enumerators to go over the numbers again if revenue, cost and profit figures are inconsistent. Enumerators have reported few conceptual difficulties with this portion of the questionnaire. 2.13 In Table 5 we report the descriptive statistics on which our analysis will be based where we make a distinction between wage employees and the self-employed. 2.14 Using the data summarized in Table 5 we carry out a series of tests to ask of the data the following question: which of the four forms of training we have identified - vocational, apprenticeship, on-the-job and teaching and nurse training - pays the most? The answer to that question will clearly depend on how much we control for in any equation. In Table 6 we control for gender, age, as a measure of general work experience, education measured in year and a raven's test which is intended as an indicator of reasoning ability similar to that originally used in Knight and Sabot (1990). We will also investigate how controls for occupation influence our measures of the returns to training. We identify four occupation classes - the self-employed, those working in a small firm, defined as one with less than 10 employees, those working in large firms, defined as those with more than ten employees and those working in the public sector. Workers employed in pubic sector firms are not separately identified; they will be included in the large firm category so in our analysis the public sector is essentially civil servants. 9For the narrow group of readers familiar with the CSAE's survey work in Ghana, we should note that the analysis in this paper does not incorporate data from the Ghana Manufacturing Enterprise Survey (GMES). The UPS and the GMES are conducted in parallel with a common survey instrument. However, we restrict ourselves in this paper to the population based sample of the UPS, excluding the firm-based sample of manufacturing employees interviewed through the GMES. 23 Table 5: Summary Statistics Sample (excluding students, including people with no earned income): N = 1356 Mean Standard Deviation Male (=1 if male) 0.43 0.50 Age (years) 36.62 11.59 Raven's Score (out of 20) 4.68 4.75 Education (years) 8.34 3.81 Past apprentice (=1 if past apprentice) 0.23 0.42 Past vocational (=1 if past vocational trainee, excluding teachers and nurses) 0.08 0.27 Past on-the-job training (=1 if past on-the- job trainee) 0.11 0.31 Teacher/nursing training (=1 if past teacher or nursing trainee) 0.01 0.12 Monthly earnings in cedis 108,589 109,679 Monthly earnings in dollars 88.48 89.37 Log of monthly earnings in dollars n/a n/a Sample with earned income (excludes students): N = 932 Mean Standard Deviation Male (=1 if male) 0.44 0.50 Age (years) 35.38 10.59 Raven's Score (out of 20) 4.52 4.81 Education (years) 8.21 3.99 Past apprentice (=1 if past apprentice) 0.29 0.45 Past vocational (=1 if past vocational trainee, excluding teachers and nurses) 0.09 0.29 Past on-the-job training (=1 if past on-the- job trainee) 0.14 0.34 Teacher/nursing training (=1 if past teacher or nursing trainee) 0.02 0.13 Monthly earnings in cedis 111,627 110,379 Monthly earnings in dollars 90.95 89.94 Log of monthly earnings in dollars 4.11 0.95 24 Table 5 (continued): Summary Statistics Self-employed (positive income earners): N = 541 Mean Standard Deviation Male (=1 if male) 0.28 0.45 Age (years) 36.97 10.45 Raven's Score (out of 20) 3.32 4.16 Education (years) 7.05 4.23 Past apprentice (=1 if past apprentice) 0.33 0.47 Past vocational (=1 if past vocational trainee, excluding teachers and nurses) 0.09 0.28 Past on-the-job training (=1 if past on-the- job trainee) 0.11 0.32 Teacher/nursing training (=1 if past teacher or nursing trainee) 0.002 0.04 Monthly earnings in cedis 99,064 88,335 Monthly earnings in dollars 80.72 71.97 Log of monthly earnings in dollars 3.96 1.01 Wage employees (positive income earners in SMEs, large firms, and the public sector): N = 391 Mean Standard Deviation Male (=1 if male) 0.66 0.47 Age (years) 33.16 10.40 Raven's Score (out of 20) 6.19 5.14 Education (years) 9.80 2.96 Past apprentice (=1 if past apprentice) 0.22 0.42 Past vocational (=1 if past vocational trainee, excluding teachers and nurses) 0.10 0.30 Past on-the-job training (=1 if past on-the- job trainee) 0.16 0.37 Teacher/nursing training (=1 if past teacher or nursing trainee) 0.04 0.19 Monthly earnings in cedis 129,011 133,259 Monthly earnings in dollars 105.12 108.58 Log of monthly earnings in dollars 4.30 0.83 2.15 In Table 7 we extend the set of controls in two dimensions. Firstly, instead of controlling for education by a continuous measure we use dummies for the highest level completed. Secondly, we allow the return from forms of training to differ depending on where the student enters the training system. Such a distinction has been found to be important in understanding the returns to training in Tanzania, Kahyarara and Teal (2006). 2.16 We begin by asking the simplest descriptive question: how do the earnings of those who received at least one of these four forms of training compare with those who received none? This question is answered in Table 6 column [1]. Our data imply that the returns from different forms of training differ radically. Those who undertake training to be either a nurse or teacher receive incomes some three times higher than those with no training. In contrast those with some on-the-job training receive only 27 per cent more income those with no training [obtained by exp (0.24)-1]. In even greater contrast those who have undertaken an 25 apprenticeship receive incomes 17 per cent lower than those with no training. It is important to recognize that his does not imply that undertaking an apprenticeship lowers earnings. It implies that simply as a descriptive fact apprenticeship is associated with a range of circumstances which lead to lower levels of income on average than those with no training. What those circumstances might be we now investigate by including controls in the equation. 2.17 In Table 6 Columns [2] and [3] we present two basic earning functions to establish a basis for how the effects of training and occupation may impact on earnings. In Column [2] we only control for gender, age and education, in Column [3] we include our control for reasoning ability, the Raven score. While this measure of ability decreases the return to education a little the impact is not large. This is consistent with a very wide range of evidence that any positive upwards bias on the OLS estimates of the return to education through any correlation between ability and education are small, see Card (2001) for a review. 2.18 Our first test as to whether training is linked to increases in income, once we control for human capital, is in Table 6 Column [4] where we include the training measures as well as our controls for gender and human capital. The effect of these controls is to remove any significant effect of training on earnings for all except those going to teacher or nursing school. It remains true that, accepting the point estimates in the equation, that this last form of training is by far the highest doubling incomes. It is also true that the point estimate on apprenticeship remains negative but, as already noted, it is no longer significantly different from zero. In the final Column of Table 6 we include in addition controls for occupational outcomes. The effect here is to roughly half the point estimate on the teacher and nursing parameter, suggesting that about half of the return to this type of training occurs through access to the public sector. The occupational dummies suggest a hierarchy of earnings by which those in the public sector earn about 80 per cent more than those in the small firm sector, with those in self-employment earnings 17 per cent more than those in small firms and those in large firms earning 37 per cent more than those in self-employment. It is important to remember these sectoral differences control for human capital so do not reflect the full extent of differences across sectors. Further as there are clearly many factors which induce sorting across sectors that we do not observe these sectoral cannot be given any causal interpretation. There is however a common finding across all the regressions that apprenticeship never has a positive effect on earnings. Which raises an obvious question: why do so many do it? As we have already shown it is by far the most common form of training in Ghana. 26 6 e bla 27 T 28 7 e bla T 2.19 In Table 7 we begin to address that question by asking if the effects of apprenticeship on earnings depend on when on the education path the apprenticeship is undertaken. In order to do that we move from measuring education by the number of years and instead use a series of dummy variables for the highest level of education reached. For completeness we repeat in Table 7 Columns [1] - [4] the same comparisons as we have already reported in Table 6. It is clear that changing this way of modeling education does not alter our result so far, to be found in Table 7 Columns [3] and [4], that undertaking an apprenticeship has no positive impact on earnings. In the final column of Table 7, we interact the apprenticeship and vocational training variables with the educational background. So (App x Primary) in the Table means that the apprenticeship was undertaken after primary education was completed and a similar definition holds for the other interaction terms. Although the result is only significantly different from zero at the 10 per cent level we now find a substantial positive impact on earnings for those who undertook an apprenticeship but have less than primary completed education, this includes those who did not start and those who did not complete primary. The point estimate in Table 7 Column [5] implies that those with the lowest levels of education achievement have a 42 per cent higher income from undertaking an apprenticeship than those, with these low levels of education, who do not do one. The point estimates imply that for those with any higher level of education the effect is negative. Indeed we know from Table 3 above that nearly three quarters of those doing an apprenticeship undertake it as the end of junior secondary school. For those the point estimates in Table 7 Column [5] imply that their incomes are 8 per cent lower as a result of undertaking the apprenticeship and at the ten per cent significance level this is different from zero. Our big puzzle remains. There is no evidence for the vast majority who do an apprenticeship that it pays. Is it possible doing an apprenticeship operates on access to employment? To that question we now turn. 5 Do the occupational outcomes from apprenticeship differ from that of other forms of training? 2.20 In section 3 above we set out how the educational background and occupational outcomes for apprentices differ from those for our whole sample. In this section we address more formally the question as to how far undertaking an apprenticeship compares with other forms of training in affecting the probability of different occupational outcomes. In Table 6 and 7 we have identified four occupational outcomes - self-employment, employment in small and large firms and the public sector. These outcomes are conditional on have a job so we are faced with seeking to model five possible occupational outcomes, those used in Table 6 and 7 and the no income category. Such a model would be complex to interpret and our sample is too small to obtain other than very imprecise estimates. So in this section we reduce the possible options to three- no income (which includes those outside the labor force and those in the labor force without a job), informal employment (which includes both the self-employed and those in small firms) and the formal sector (which includes both those in large firms and those in the public sector). We report in the appendix a multinomial logit which can be given an interpretation as modeling the determinants of occupational choice as a function of education and training. As the coefficients on such models are hard to interpret we concentrate on using the predicted probabilities from the model to ask two questions. Firstly, how is the probability of working in either the formal or informal sector affected by having an apprenticeship? Secondly, how does the effect of having an apprenticeship compare with that from training to be a teacher or nurse which the previous section showed to be by far the training with the highest return. 2.21 We answer the first of these question in Figure 1 below which shows how the probability of moving between having no income and the two types of employment we identify, informal and formal, as a result of doing an apprenticeship varies across the education background of the 29 apprentice (the data which underlies the Figure is given in the appendix). The same pattern is apparent for all education backgrounds. The effect of having an apprenticeship is to substantially increase the probability of having an informal sector job. For middle/junior school completers the effect of apprenticeship is to increase the probability of an informal sector job by 22 percentage points from 51 to 73 per cent. Most of the shift to informal employment comes from the no income category where for middle/junior school completers the probability of having no income falls from 32 to 18 per cent. The major effect of undertaking an apprenticeship is a shift from no-income to informal employment, the shift within employment from formal to informal is much less important. Figure 1 Probability of Entering Occupational Sector for Apprentices Predicted probabilities evaluated at the means .8 ytil .6 biabo .4 pr .2 0 No app Past app No app Past app No app Past app No app Past app Less than primary Primary JSS/middle Secondary Predicted prob. of informal Predicted prob. of formal Predited prob. of no income Note: Estimation sample excludes students. Categories reported exclude those without past apprentices. 2.22 As can be seen from the multinomial logit in the appendix the effect of apprenticeship on increasing the probability of informal employment is highly significant. Why do we observe such a highly significant effect for the occupational outcomes but such weak evidence of any positive earnings effect once in employment? We outline her two possible elements of an answer to that question. The first is that the role of apprenticeship is to enhance the skills of the worker to a point where their earnings exceed their reservation wage. Thus apprenticeship makes possible employment in the informal sector in the sense that the costs of working are now lower than the benefits. The second part of a possible answer is that apprenticeship is undertaken by those with relative low ability and is an option forced on them by their exclusion from undertaking further formal education. If this is the case the negative point estimate on the apprenticeship dummy reflects not the lack of a positive impact but the failure of the regression to control fully for such unobserved low ability. That this may well be part of the explanation is suggested by the control for ability we do have in the equation, the raven score, which if dropped causes the point estimate on the apprentice dummy to become more negative and more significant (results not reported). 2.23 We turn now to our second question: how does the effect of having an apprenticeship compare with that from training to be a teacher or nurse? We answer that in Figure 2 which presents a similar breakdown to that of Figure 1 and show the probability of being in one of the occupational categories as a result of undertaking this form of training. The contrast with the effects of undertaking an apprenticeship is striking. With this form of training there is a very substantial shift 30 from informal to formal employment. For the secondary completers the probability of a formal sector job increases from 24 to 87 per cent and reduces the probability of an informal sector job from 44 to 3 per cent. In other works while apprenticeship acts to shift the probability of employment between informal and none, this form of training acts to shift workers between the informal and formal sectors. Figure 2 Probability of Entering Occupational Sector for Teachers/Nurses Predicted probabilities evaluated at the means .8 ytili .6 ab obrp .4 .2 0 No t/n t/n No t/n t/n No t/n t/n Primary JSS/middle Secondary Predicted prob. of informal Predicted prob. of formal Predited prob. of no income Note: Estimation sample excludes students. Categories reported exclude those without t/n (teacher/nursing) trainees. 6 Summary and Conclusion 2.24 We began this note with four questions. Firstly, how important is apprenticeship as a form of training? Secondly, what is the educational background and occupational outcomes of the apprentices? Thirdly, does being an apprentice pay? Fourthly, do the occupational outcomes from apprenticeship differ from that of other forms of training? We now briefly summarize our answers to those questions. 2.25 Apprenticeship is, on the basis of our survey, by far the most important form of training in urban Ghana. Of the training events our survey identified over half were either current or past apprenticeships. The vast majority of apprenticeships are undertaken by those with junior secondary school or less. While nearly 60 per cent of those who did an apprenticeship in the past are self- employed some 9 per cent work in small firms and some 15 per cent have no job. Our earnings data suggest that those who undertook an apprenticeship earn significantly less than those with no training. and the majority work in the self-employed sector. Once an allowance is made for the level at which the apprentice enters the system there is evidence that, for those with the lowest level of education, apprenticeship does lead to a substantial increase in earnings, some 41 per cent. For those with higher levels of education it does not lead to any increase at all. In fact the point estimates continue to imply a decrease. We have argued that part of the explanation for this may be due to our limited controls for ability. Apprenticeship is an outcome forced on many individuals who cannot proceed further through the academic educational system. We have also shown that apprenticeship has a powerful effect in increasing the probability of being in informal employment relative to having 31 no job. Therefore, one possible role for apprenticeship is to increase the supply of skills so that the benefits of working exceed the costs. 2.26 In contrast to these outcomes for undertaking apprenticeships are the very high earning increase that accrue from being trained as a nurse or teacher. Their earnings are twice those of apprentices who enter after junior secondary school, with a full set of controls for education and occupation. Further such training substantially shifts the likelihood of obtaining a formal relative to informal sector job. In summary these forms of training, one undertaken in the private sector and the other in the public sector, are associated with radically different outcomes for those receiving the training. References Card, D. (2001) "Estimating the return to schooling: progress on some persistent econometric problems", Econometrica, Vol. 69, pp. 1127-1160. Foster, P.J. (1965a) "The vocational school fallacy in development planning", in: C.A. Anderson and M.J. Bowman (eds.) Education and Economic Development, Chicago, Illinois, Aldine. Foster, P.J. (1965b) Education and Social Change in Ghana, London, Routeledge and Kegan Paul. Frazer, G. (2006) "Learning the master's trade: Apprenticeship and human capital in Ghana", Journal of Development Economics, 81: 259-298. Kahyarara, G. and F. Teal (2006) "To train or to educate? Evidence from Tanzania", GPRG Working Paper, GPRG-WPS-051. http://www.gprg.org/pubs/workingpapers/pdfs/gprg-wps-051.pdf Knight, J.B. and R.H. Sabot (1990) Education, Productivity and Inequality: The East African Natural Experiment, Oxford University Press. Middleton, J.A., Ziderman, A. and van Adams, A. (1993) Skills for Productivity: Vocational Education and Training in Developing Countries. Oxford University press, New York. Sandefur, J, P. Serneels, and F. Teal (2007) "Poverty and Earnings Mobility in Three African Countries", Chapter 5 in P. Paci and P. Serneels (eds.) Employment and Shared Growth: Rethinking the role of labor Mobility for Development, The World Bank, Washington, D.C. 32 Appendix Table Education No Apprenticeship Apprenticeship Less than primary Informal: 0.739 Informal: 0.879 Formal: 0.036 Formal: 0.015 No income: 0.225 No income: 0.105 Primary Informal: 0.534 Informal: 0.750 Formal: 0.112 Formal: 0.055 No income: 0.354 No income: 0.194 JSS/Middle Informal: 0.511 Informal: 0.734 Formal: 0.167 Formal: 0.084 No income: 0.322 No income: 0.182 Secondary Informal: 0.374 Informal: 0.614 Formal: 0.278 Formal: 0.161 No income: 0.348 No income: 0.224 Education No t/n t/n Primary Informal: 0.597 Informal: 0.073 Formal: 0.093 Formal: 0.705 No income: 0.310 No income: 0.223 JSS/Middle Informal: 0.576 Informal: 0.053 Formal: 0.139 Formal: 0.794 No income: 0.285 No income: 0.154 Secondary Informal: 0.438 Informal: 0.025 Formal: 0.242 Formal: 0.866 No income: 0.320 No income: 0.108 33 ANNEX 3 VARIABLE DEFINITIONS APPLIED TO THE HOUSEHOLD DATA 10 Variable Definition Labor force Working age (25-64) individual employed or unemployed in the last seven days Employed Did any work for pay in the lst 7 days Unemployed Not working but actively looking for a job Unemployment rate Percentage of the labor force not working and looking for one Working age population Population aged 25-64 Child labor Children aged 7-14 reporting employment Youth labor Individuals aged 15-24 reporting employment Elderly labor Individuals aged 65 and more reporting employment Inactive Individuals not working not unemployed in the last 7 days Labor market status Wage Public sector Wage earner in the public sector Wage Private sector Formal Wage earner in private sector, formal sector Wage Private sector Informal Wage earner in private sector, informal sector Self-employed Agro ­ Paid Self employed in the agriculture sector and declaring "earnings" Self-employed Agro - Unpaid Self employed in the agriculture sector and not declaring "earnings" Self-employed Non Agro Self employed in the non agriculture sector Level of education No education Never attended school Some Primary Incomplete primary only Primary Completed primary level Secondary (lower) Completed JSS or middle school Secondary (higher) Completed SSS or secondary school TVET Attended some technical or vocational training Post Secondary Attended post-secondary education Earning Total annual earning divided by hours worked Locality/ecological zone Accra Individual resides in the Greater Accra Metropolitan area Coastal Individual resides in the Coastal ecological zone (south of Ghana) Forest Individual resides in the Forest ecological zone Savannah Individual resides in the Savannah ecological zone (north of Ghana) Region Administrative regions (10) Sector Industry sectors using the first-digit classification 10This Annex was prepared by: Harold Coulombe, January 2008 34 ANNEX 4 DOMESTIC OR PRODUCTIVE WORK? CHANGES IN PATTERNS OF TIME USE IN GHANA FROM 1991 TO 200611 Ghana's economy has evolved markedly over the last 15 years, with high levels of growth generating many new jobs and contributing to substantial poverty reduction. Yet it is not clear whether the patterns of time use of households have evolved as much. Given that labor force participation and wage data show a substantial and persistent gender gap, it is worth investigating whether the division of labor between productive work (mostly carried on by men) and reproductive work (mostly carried out by women) is changing. Ghana is special in this respect since it is one of few developing countries where detailed data on time use is available over a long period of time thanks to repeated, nationally representative and comparable cross-section household surveys with time use modules. Using the 1991/92, 1998/99 and 2005/06 Ghana Living Standard Surveys, this paper provides basic statistics on the time allocated to domestic and productive work by various household members according to characteristics such as gender, age, urban/rural location status, access to infrastructure and quintiles of consumption. The concept of time poverty is then used to compute measures of the share of the population working above what could be considered as a maximum threshold for time worked. Women are found to work more than men, especially on domestic tasks, and thereby are more likely to be time poor, and less likely to obtain earnings from productive work. 1. Introduction 4.1 Work can broadly be understood as labor that produces something of value for oneself or other people (Gacitua-Mario et al., 2002). As noted among others by Gerstel and Gross (1987), Hartman (1987), Mann (1990), work includes both productive work (for the production of goods and services for the market) and domestic or reproductive work (production of goods and services for household consumption). In most societies gender is a key means of determining the division of labor and responsibilities among household members. Women's domestic work ­particularly that related to biological reproduction- increases disproportionately in relation to other adults in the household as the number of infants and children grows. Mothers of large families spend more time pregnant, breast-feeding, and caring and cooking for children and other family members. While being male or female is a biological fact, becoming a man or a woman is a cultural and social process which affects labor outcomes and leads to gender disparities in labor markets (e.g., Correia, 1999). Gender analysis is thus crucial because it looks at the social scripts people are exposed to, the type of choices they see as viable and legitimate, and the cost and benefits associated with those choices (Staudt, 1994).12 4.2 Time use data are an important tool for gender analysis in order to measure and understand patterns of domestic and productive work. There is ample evidence that women allocate substantial time to domestic chores as well as labor in Sub-Saharan Africa, and that this burden imposes a burden on them. The constraints on time use imposed on women not only by domestic work but also by work in the fields were already recognized in the 1960s. Data from that period in two villages from the Central African Republic showed that men worked 5.5 hours/day, versus 8 hours/day for 11This background paper was prepared by: Harold Coulombe and Quentin Wodon, January 2008 12This paragraph has been adapted with minor changes from Gacitua-Mario et al. (2001). 35 women (Berio, 1983). Studies based on data from the 1980s and 1990s have confirmed large differences in time burdens according to gender (Blackden and Bhanu, 1999; Ilahi, 2000). For example, women have been shown to spend about three times more time in transport activities than men in Ghana, Tanzania, and Zambia (Malmberg-Calvo 1994), and in Uganda, time savings from better access to water and wood were estimated at 900 hours/year, mostly to the benefit of women (Barwell, 1996). More recent work using new data on Benin, Ghana, Madagascar, Mauritius and South Africa (Charmes, 2006), as well as on Guinea (Bardasi and Wodon, 2006a) and Malawi (Wodon and Beegle, 2006), have provided additional evidence that women have to work more than men in Sub-Saharan Africa (see also Ilahi and Grimard, 2001, for Pakistan, and World Bank, 2001, for a broader discussion of related gender issues). 4.3 As discussed by Blackden and Wodon (2006), existing patterns of time use have potentially important consequences for households. One key issue is that the "household time overhead" (a concept introduced by Harvey and Taylor, 2000) or the number of hours that household members, and especially women, must allocate to basic chores is high. Taking care of children and possibly the elderly, preparing meals, washing clothes, cleaning the dwelling, fetching water and wood may together represent a full-time occupation for several household members. Especially when households do not have access to basic infrastructure services such as electricity, piped water and sanitation facilities, the time necessary for performing domestic chores is typically much higher than when access is available. In turn, because the time spent on domestic chores is not easily dispensable, and because domestic chores are performed mainly by women, many women have limited opportunities to engage in productive activities. This does not mean that they do not actually work outside of the house, but simply that the type of work that they do is constrained by their responsibility towards domestic chores as well as cultural factors. This in turn means that there are constraints to their income and decision power within the household. Scarcity of time also means that women have limited opportunities to further their education and training. Overall, it could thus be argued that "time poverty", especially among women, is one of the determinants of consumption poverty. 4.4 To make the argument clearer, assume that one would estimate the labor market value of the time available to various household members, or the value of the time savings that could be obtained from policies such as those facilitating access to infrastructure services. The value of these time savings could then be taken into account to assess how additional labor market earnings generated through additional time allocated to work in the labor market could help in reducing monetary or consumption-based poverty. This has been done for example by Bardasi and Wodon (2006b) using Guinea data, with the authors finding that if all household members were indeed to work a certain given amount of time, monetary poverty could be reduced substantially. From a policy point of view, this could then imply that investments aiming to reduce household time overhead through access to better infrastructure services could be critical for poverty reduction. 4.5 There are many steps and implicit assumptions in the above argument, and these will not be tested in this paper. Our objective here is more limited. It is to provide a descriptive analysis of domestic and productive work time in Ghana. This should still be interesting however, because the time use data available in the third (1991/92), fourth (1998/99) and fifth (2005/06) rounds of the Ghana Living Standard Surveys are rich. The fact that we have three surveys at our disposal also makes it feasible to provide at least some comparisons over time in terms of changes in pattern of time use, even though there are some issues of comparability between the surveys. In fact, within Africa, Ghana is one of the very few countries meeting data requirements for the analysis of time use patterns over time. 36 4.6 The paper is structured as follows. In Section 2, we use all three rounds of the GLSS to provide basic statistics on the time allocated to domestic work as well as labor according to various individual characteristics such as gender, age, and urban/rural location. In section 3, following the methodology proposed by Bardasi and Wodon (2006a), we provide measures of time poverty. The empirical results obtained in the paper confirm conventional wisdom. For example, women are found to work more than men, and the time spent on domestic chores is much higher especially for women when households do not have access to basic infrastructure. In section 4, we analyze the determinants of time worked. A brief conclusion follows. 2. Basic Statistics Comparability issues between surveys 4.7 Before presenting basic statistics on time use, a word of caution on the data is required. Although the time use questionnaires from the three GLSS are very detailed and permit to measure the time worked by each individual interviewed, and although the questions asked are almost identical, a few changes make comparability of total time use over time difficult. The main problem concerns the measurement of time spent on domestic chores. In the GLSS 3, domestic chores were split into three different categories (fetching water, fetching wood and "other chores") while the GLSS 4 questionnaire further breaks down "other chores" into ironing, caring of children, washing vehicles, sweeping, disposing garbage, cooking, shopping, washing dishes and other housekeeping. Further changes were seen in GLSS 5 questionnaires when compared to GLSS 4, namely the dropping "washing cars, "sweeping" and "disposing of garbage" but adding "washing clothes", "cleaning" and "taking care of elderly and the sick". Expanding the list of "other chores" in the GLSS 4 and 5 and adjustment to the list of chores between GLSS 4 and GLSS 5 creates problems of comparability. This is important as time-saving infrastructures are more likely to help reduce the time spent on domestic chores than in labor market activities, and as domestic chores take up a large part of total time worked, particularly among women. 4.8 The more detailed list of domestic activities in the GLSS 4 and GLSS 5 lists has surely also created a problem of parallel counting of activities, especially regarding the time spent taking care of children, elderly and the sick, activities that is often combined with other types of domestic work or even employment. In the estimates provided here, we will exclude the time spent taking care of children/elderly/sick for that reason (many respondents claim, rightly, up to 24 hours a day plus their other activities. As an illustration of the problem, consider the fact that more than 700 respondents (out of 20,000) claimed more than 112 hours of activities in 1998/99, while only 9 did so in 1991/92 (out of 15,500). This discrepancy was almost exclusively due to double counting of the time spent taking care of children. At around 10 cases per year, removing caretaking from our estimates of time lead to very few cases where individuals claims more than 112 hours a week (16 hours/day). In our computations, we censored those few cases at 112 hours. We also made some further minor adjustments to make the lists of housekeeping chores from GLSS 4 and GLSS 5 as comparable as possible. 4.9 It should be noted that when estimating domestic time use, one of the most time-consuming tasks usually undertaken by women was not taken into account. As discussed previously, taking care of children was only measured in 1998/99 and 2005/06 surveys although the results were clearly flawed since many respondents claimed taking up to 24 hours a day for taking care of their children. Even if this may happen in some cases, taking care of children is then obviously done simultaneously with other activities, whether on the labor market or at home on other domestic chores. Including child rearing as measured by GLSS surveys would have strongly biased upward the time use figures, 37 so we did not do it. That issue raises the question as to whether the existing GLSS questionnaire design is the most appropriate way to measure time use. A better approach could be to ask respondents to fill up a time sheet where they have to say which activities the person is doing at different time of the day. For example, such survey was performed in Benin in the 1990s with respondents asked to record their activities at each 15-minute interval over a single chosen day (24 hours). The respondents filled up the time sheet questionnaire from a list of around 60 different activities, and simultaneous activities were allowed. Such approach would probably be better suited to precisely measure time use in general and activities as child rearing in particular. Charmes (2006) provides a review of the different approaches used to measure time use, including the experience of Benin and Ghana. 4.10 These comparability issues between the time use data from the two surveys imply that we must be careful in drawing conclusions regarding trends in time use over time. At the same time, most individual categories of time use are clearly comparable, and as for total time use, given that the list of activities in the GLSS 4/5 is more complete than in the GLSS 3, the hypothesis is that we would rather underestimate time savings for domestic chores over time, rather than overestimating them. Thus, if we find a decrease in time spent on domestic chores, this probably reflects a true decrease rather than be a statistical oddity due to comparability issues. Empirical results 4.11 As in Bardasi and Wodon (2006a), the individual-level indicator that we use to examine time use patterns (and in the next section assess who is time poor) is the total amount of time spent by individuals working, whether in the formal and informal labor market, in domestic chores or in collecting water and wood. The three panels of Figure 1 show the distribution of the total individual working hours per week for adult individuals aged between 25 and 64, for all three years of data. Only a small share of the adult population does not work at all (between 2 and 3%), and in both years, the decreasing part of the distribution starts after about 50 hours per week, with a small share of the population putting in more than 100 hours each week. While the distribution of working time is a bit flatter in 1998/99 than in 1991/92, one cannot say that major changes in the distribution are visible at first sight. If one were to draw the same figures separately for men and women, as well as for urban and rural areas, we would see that women work a higher number of hours than men, and rural individuals also tend to work more than their urban counterparts. 4.12 Tables13 2a, 2b and 2c provide data on the average number of working hours for the main uses of working time by gender and by age group. For example, the mean working time at the national level is 47.0 hours for the adult population (between 25 and 64 years of age) in 1991/92, 46.2 hours in 1998/99 and 48.4 hours in 2005/06 (see Table 2b). There are substantial differences between adult men and women in all three years. In 1991/92, the average workload reaches 51.9 hours for women, and 40.8 hours for men. This gender gap of around 11 working hours is also observed in 1998/99 (workload of 51.1 hours per week for women versus 40.4 hours for men). However, this gender gap had shrunk by 2005/06 to 5.4 hours. The diminishing gender gap was due to male being more occupied by 5 hours a week. However, a closer look at the results shows a rather more complex and interesting stories. If we compare 1998/99 with 2005/06, the additional hours by 13Unlike GLSS 3 and 5, the different questions on employment activities in GLSS 4 questionnaire were only asked to individuals not going to school. Since some individuals are simultaneously going to school AND working, this glitch in GLSS 4 questionnaire would necessarily bias downward any time use statistics. To counter this bias we first define three age groups where the first cover all school age individuals (7-24 year old), the second one cover prime age adults (25-64), and a last one cover elderly (65+). Then we omitted any statistics based on the younger group surveyed in 1998/99 (GLSS 4). 38 male was essentially spent on the labor market, being a wage earner or self-employed. In the case of female, their time spent on domestic chores was reduced by 4.8 hours (from 25.9 hours in 1998/99 to 21.1 hours in 2005/06) but that time free of domestic chores was fulfilled by an almost equivalent number of hours (4.6) on the labor market. This labor market activity seems to be due to increasing earnings and job opportunities in both urban and rural areas. That large increase in labor force participation rates for both male and female is further examine in Coulombe and Wodon (2007). While men spend a few more hours on average in the labor market than women, the amount of time spent by women in domestic chores is much higher than for men, and this explains the large differences in total working time. Figure 1: Distribution of total time worked, individuals aged 25-64, Ghana a) 1991/92 3 .0 2 .0 ytis en D 1 .0 0 0 50 100 Total individual time (hours/week) 1991/92 b) 1998/99 3 .0 2 .0 ytis en D 1 .0 0 0 50 100 Total individual time (hours/week) 1998/99 c) 2005/06 2 .0 ytis en 015. D 1 .0 50 .0 0 0 50 100 Total individual time (hours/week) 2005/06 Source: Authors' calculations from GLSS data. 39 4.13 Table 2a, focusing on the 7-24 age group, shows that girls also work longer hours than boys due to a higher burden from domestic work, but the total amount of work remains reasonable (about 24.3 hours per week on average in 2005/06). The reduction in working hours observed for children and youth aged 7 to 24 over time may be related in part to an increase in school enrollment rates during the 1990s and early 2000s in Ghana. As for comparisons across urban and rural areas, the results suggest differences in total working time, with on average rural individuals working five hours more per week than their urban counterparts. The main difference by location is the higher burden of domestic chores for women as well as children (boys and girls) in rural areas, especially due to the need for rural households to fetch water and wood. However it is worth noting that fetching water and wood has taken half as much time in 2005/06 than in early 90s, which may be due in part to better access to infrastructure services. 4.14 The gender gap is present as well for the adult age group (Tables 2b). Women spend more than 21 hours a week in 2005/06 on domestic chores while men only spend around 7 hours. In terms of the total number of hours worked, women are at 50.9 hours per week in 2005/06, as compared to 45.5 hours for men. The total time spent working has decreased slightly for women over time (from 51.9 hours in 1991/92 to 50.9 hours in 2006/06), while it has increased more substantially for men (from 40.8 hours in 1991/92 to 45.5 hours in 2005/06). The increase for men is largely due to an increase in productive work (from 35.5 hours to 38.6 hours), while the decrease for women is due to a decrease in domestic work hours (from 24.7 hours to 21.1 hours). Estimates are also provided in table 2c for the elderly population, where the hours worked are significantly lower than for the working age population aged 25-64 represented in table 2b. 40 2 4 0 7 3 6 0 6 1 4 2 6 2 3 5 1 7 0 7 5 3 2 5 0 All 0. 1. 9. 3. 3. 7. 1. 2. 8. 1. 9. 0. 2. 9. 6. 2. 8. 10. 17. 10. 13. 23. 12. 21. 3 6 2 1 8 2 0 1 5 1 6 2 8 1 9 1 0 5 7 2 2 9 2 3 ale 0. 1. 3. 3. 7. 1. 3. 7. 1. 8. 1. 2. 6. 1. 8. 42-7 06/ 11. 13. 20. 13. 18. 27. 12. 16. 24. Fem 0052 uporg 2 2 6 0 8 1 9 9 7 8 1 5 4 5 9 5 5 6 9 0 4 4 9 8 ale 0. 1. 6. 8. 2. 4. 6. 0. 1. 7. 9. 8. 1. 9. 0. 1. 6. 9. 6. 2. 8. eg 14. 19. 17. M A, naah G,y .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. All litaco L .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ale nda 99/ x Fem Se 9981 .sl r,a .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. dua vi Ye ale yb, M ndii gea iesti hoolcs 41 8 9 1 8 2 3 5 1 3 1 0 4 8 6 3 7 8 3 1 2 8 8 7 8 Activ All 0. 2. 4. 1. 5. 2. 5. 8. 9. 0. 1. 4. 9. 7. 0. 8. for 11. 14. 20. 15. 10. 25. 15. 23. us desa rio bi 9 2 6 7 3 0 4 8 0 0 9 9 7 4 1 9 2 0 9 1 1 6 7 7 Va rea ale 0. 3. 5. 1. 6. 3. 6. 9. 0. 2. 5. 8. 0. 8. no 92/ 13. 17. 23. 10. 19. 10. 29. 11. 19. 27. y het Fem sa 9911 de Spent ntes 7 6 5 8 0 5 6 2 7 2 4 4 9 7 6 0 4 7 4 6 6 0 6 0 urso ale 0. 2. 8. 3. 1. 4. 1. 4. 5. 9. 0. 1. 3. 6. 7. 1. 8. pre 11. 16. 11. 10. 22. 11. 20. H M not .5 rea & s ekly 4,3 S gure Wefo fi LS G 99 no ber 1998/ m 3) 3) 3) he desab n Nu ) ) ) ) ) T tio ega odo reta se 2++1( +6 +7 +6 (4 odo reta se 2++1( +6 +7 (4 odo reta se 2++1( ) d. +7 (4 udel erv w w se e w w se e w w se e A:a1 saer ng ng hor boral (5rob (5rob (5rob nci la mti s ng ng hor boral la mti ng ng hor boral la mti rea timase'sro A het gea ea het gea het gea thu e n hicteF hicteF orhcr Cll O A robalfleS orhcr orhcr talo talo W T T bla bar 1 2 3 4 5 6 7 8 Arlar hicteF hicteF Cll O A robalfleS talo talo W T T 1 2 3 4 5 6 7 8 lanotia hicteF hicteF Cll O A robalfleS talo talo rose W T T 1 2 3 4 5 6 7 8 Z:e A:ec T U Ru N Not Sour 3 8 2 3 6 3 9 2 6 8 3 7 2 7 9 5 1 4 0 5 5 4 9 4 All 0. 0. 1. 1. 3. 1. 1. 8. 10. 11. 21. 15. 36. 48. 13. 16. 28. 31. 48. 12. 14. 25. 33. 48. 4 2 6 1 7 8 5 6 5 8 2 4 1 0 0 4 6 1 3 1 5 3 8 9 ale 0. 1. 6. 2. 2. 1. 1. 2. 3. 46-52 06/ 14. 16. 25. 32. 48. 19. 24. 27. 28. 52. 17. 21. 26. 29. 50. Fem 0052 uporg 2 5 4 1 2 5 7 8 7 6 2 4 5 9 4 9 5 5 9 9 3 3 6 5 eg ale 0. 0. 5. 6. 0. 0. 6. 7. 6. 0. 0. 5. 6. 17. 24. 41. 47. 29. 36. 43. 24. 14. 38. 45. M A, naah G,y 4 9 0 3 8 6 4 6 6 8 3 7 3 5 8 5 2 5 5 1 1 0 1 2 All 0. 0. 1. 1. 3. 1. 1. 6. 13. 14. 22. 10. 33. 47. 15. 18. 23. 26. 45. 14. 17. 23. 29. 46. litaco L 5 4 2 0 0 1 1 1 5 9 1 6 4 2 5 1 8 4 8 9 0 2 2 1 ale 0. 1. 4. 2. 2. 1. 1. 2. 2. nda 99/ 19. 21. 26. 30. 51. 23. 28. 21. 22. 51. 21. 25. 23. 25. 51. x Fem Se 9981 r,a .sl 3 4 7 4 0 2 2 5 5 5 8 8 6 3 9 7 4 4 8 6 2 5 8 4 dua Ye vi ale 0. 0. 5. 6. 0. 0. 5. 6. 6. 0. 0. 5. 6. yb, 19. 18. 37. 43. 25. 31. 38. 23. 10. 33. 40. M ndii gea iesti hoolcs 42 8 8 7 3 0 0 0 1 6 6 5 6 6 7 2 9 0 0 3 2 0 9 9 0 Activ All 0. 1. 2. 3. 3. 2. 3. 6. for 12. 15. 21. 13. 34. 49. 10. 16. 25. 29. 45. 11. 16. 24. 30. 47. us desa rio bi Va 3 8 1 0 1 3 3 2 0 6 1 6 7 0 6 2 0 6 1 7 5 8 3 9 rea ale 1. 2. 6. 4. 5. 1. 3. 4. 2. no 92/ 19. 23. 24. 30. 53. 16. 25. 24. 25. 51. 17. 24. 24. 27. 51. y het Fem sa 9911 de Spent ntes urso 3 6 8 6 1 3 5 1 8 0 4 2 8 1 9 1 6 9 9 3 4 1 5 8 ale 0. 0. 4. 5. 0. 1. 3. 5. 7. 0. 0. 3. 5. pre 17. 21. 38. 44. 26. 33. 39. 23. 12. 35. 40. H M not .5 rea & ekly s 4,3 S Wefo gure fi LS G 99 no ber 1998/ m 3) 3) 3) he desab Nu ) ) ) T n ega ) ) ) tio odo reta d. +6 +7 +6 odo +7 +6 +7 erv se 2++1( (4 reta 2++1( (4 odo reta 2++1( (4 udel w w se se e w w se se e w w se e A b:1 saer ng ng hor boral (5rob (5rob (5rob nci la mti s ng ng hor boral la mti ng ng hor boral la mti rea timase'sro A het gea ea het gea het gea thu leba n hicteF hicteF orhcr Cll O A robalfleS orhcr orhcr talo talo W T T bar 1 2 3 4 5 6 7 8 Arlar hicteF hicteF Cll O A robalfleS talo talo W T T 1 2 3 4 5 6 7 8 lanotia hicteF hicteF Cll O A robalfleS talo talo rose W T T 1 2 3 4 5 6 7 8 Z:e A:ec T U Ru N Not Sour 2 2 0 4 1 7 8 2 7 4 8 0 7 7 3 3 6 4 2 2 9 3 2 4 All 0. 0. 6. 6. 2. 0. 0. 7. 9. 0. 0. 0. 7. 8. 1. 11. 13. 20. 16. 17. 26. 14. 16. 24. 3 3 4 0 3 5 7 7 1 6 6 3 8 2 0 2 8 5 5 8 6 3 9 7 ale 0. 0. 7. 8. 0. 1. 0. 0. 0. 0. 9. 0. 06/ 10. 10. 18. 10. 12. 13. 14. 26. 10. 12. 12. 23. +56 Fem 0052 uporg 2 1 1 5 1 6 8 3 3 2 6 2 0 3 2 4 3 2 5 0 6 6 2 1 ale 0. 0. 4. 4. 5. 0. 0. 4. 5. 1. 0. 0. 4. 5. 2. eg 12. 17. 22. 20. 21. 26. 17. 20. 25. M A, naah G,ytliac 3 3 0 6 3 7 9 5 8 3 6 7 4 5 0 7 6 3 4 4 8 2 0 3 All 0. 0. 7. 7. 2. 0. 0. 7. 8. 0. 0. 0. 7. 8. 1. 13. 15. 23. 15. 16. 24. 14. 16. 24. Lo 4 5 2 0 6 6 2 3 2 5 6 3 9 1 9 2 0 5 5 9 5 2 7 6 ale 0. 0. 0. 1. 0. 0. 1. 0. 0. nda 99/ 10. 11. 12. 13. 24. 10. 12. 13. 13. 26. 10. 11. 13. 13. 25. Fem x Se 9981 .sl r,a 2 2 2 6 1 0 1 7 2 1 8 1 5 2 6 7 2 1 6 0 4 4 8 7 dua vi Ye ale 0. 0. 3. 3. 5. 0. 0. 3. 4. 1. 0. 0. 3. 4. 2. 14. 19. 22. 17. 18. 22. 16. 18. 22. M ndii by, gea ies it iv hoolcs 43 2 7 7 4 5 8 3 7 9 2 9 9 7 0 7 6 7 0 1 8 9 7 6 4 Act All 0. 0. 6. 7. 3. 0. 1. 5. 7. 1. 0. 1. 6. 7. 1. for 12. 16. 23. 19. 20. 28. 17. 19. 27. uso desa bi ri 3 1 4 6 5 5 0 6 4 7 1 2 4 2 6 7 1 5 1 8 6 3 9 6 rea Va ale 0. 1. 9. 0. 1. 1. 9. 0. 1. 1. 9. 0. no 92/ 10. 12. 13. 23. 12. 15. 15. 27. 11. 14. 14. 26. y het Fem sa nt 9911 de Spe ntes rsuo 0 1 1 3 5 0 6 8 3 6 5 5 3 8 1 6 2 5 7 4 6 2 9 3 ale 0. 0. 3. 3. 8. 0. 0. 2. 3. 1. 0. 0. 2. 3. 3. pre 12. 20. 23. 24. 26. 29. 21. 24. 28. H M not .5 y rea & s 4,3 S gure Weeklfo fi LS G 99 no erb 1998/ 3) 3) 3) he desab n Num ) ) ) ) ) T tio eg odo reta se 2++1( +6 +7 +6 (4 odo reta se 2++1( +6 +7 (4 odo reta se 2++1( ) d. +7 (4 udel erav w w se e w w se e w w se e A c:1 saer ng ng hor boral (5rob (5rob (5rob nci la mti s ng ng orhcr hor boral la mti ng ng orhcr hor boral la mti rea timase'sro A het gea ea het gea het gea thu e n hicteF hicteF orhcr Cll O A robalfleS talo talo W T T bl bar 1 2 3 4 5 6 7 8 Arlar hicteF hicteF Cll O A robalfleS talo talo W T T 1 2 3 4 5 6 7 8 lanotia hicteF hicteF Cll O A robalfleS talo talo rose W T T 1 2 3 4 5 6 7 8 Z:e A:ec Ta U Ru N Not Sour 2. Time Poverty 4.15 This section provides estimates of time poverty among the population. Our framework to measure time poverty is straightforward as we simply apply the traditional concepts and techniques used for the analysis of income or consumption poverty to time poverty. Table 3 presents figures on time poverty for all three rounds of the GLSS. The results are broken down by age group, gender and location. Two age group-specific poverty lines are computed. In the absence of a well- established procedure to compute time poverty lines, we follow Bardasi and Wodon (2006) by using arbitrary multiples of median numbers of hours spend on work. After having computed the median time use of the 7-24, 25-64 and over 64 age groups, a first line was set at 1.5 times the median. A second time poverty line was at twice those medians. The lower poverty line was set at 25.5 hours for the 7 to 24 year old, 72 hours for the adults (25-64) and at 37.5 for the elderly. The higher time poverty lines were set at 34, 96 and 50 hours respectively. 4.16 Focusing first on the lower line, results from Table 3 shows 17 percent of adults aged 25-64 were time poor in 2005/06. This represents an increase of almost 4 percentage points over 1991/92, and a much lower increase of 0.6 percentage point over 1998/99. We also find that a much higher percentage of women (21.1 percent) are time poor than men (12.3 percent). Although the gender gap in time poverty seems to have declined slightly for this population, it remains large, with still almost twice as many women who were time poor in 2005/06 than men. The data also suggest that time poverty is higher for adults aged 25-64 in rural than in urban areas. Similarly, for young persons aged 7 to 24, time poverty is higher in rural than in urban areas, although the threshold of relative time poverty is much lower for this age group, so it is less likely to be an issue. In addition, while the younger age group has higher time poverty rates at more than 30 percent in 2005/06, this still represents a decline of about 5 percentage points since the early 1990s. On the other hand, girls have a higher incidence of time poverty than boys (35.1 percent versus 26.5 percent). For completeness, time use measures are also provided for the elderly population, although as for younger individuals, the time poverty lines are much lower than for the adult population aged 25-64. The results obtained with the higher time poverty line are broadly similar to those obtained with the lower poverty line, especially in terms of gender gaps. 44 Box: Measuring Time Poverty In most empirical research on poverty, poverty measures of the so-called FGT class (Foster, Greer, and Thorbecke 1984) are used. The first three measures of this class are the headcount index of poverty, the poverty gap, and the squared poverty gap. As noted by Bardasi and Wodon (2006a), in a time poverty framework the headcount index is the share of the population that is time poor, i.e. the proportion of the population that works a number of hours y that is above a time poverty line z. If we have a population of size n in which q individuals are time poor. The headcount index of time poverty is defined as: (1) H = q n The time poverty gap represents the mean distance separating the population from the time poverty line, with the non-time poor being given a distance of zero. This measures the time deficit of the entire population, i.e. the amount of time that would be needed to shift all individuals who are time poor below a given time poverty line through perfectly targeted "time transfers". Such transfers are actually provided to some households in some developed countries. Mathematically the time poverty gap is defined as: q (2) PG = 1 n i=1 yi - z z where yi is the total working hours of individual i, and the sum is taken over individuals who are time poor. The time poverty gap can be written as being equal to the product of the headcount index of time poverty by the time gap ratio I, i.e. PG = H * I, with I itself defined as: (3) I = yq - z where yq = 1 q yi is the mean working hours of the time poor. z q i=1 While the time poverty gap takes into account the distance separating the time poor from the time poverty line, the squared time poverty gap takes the square of that distance into account, so that more weight is given to those who have extra long working hours (this factors in inequality among the time poor): q (4) SPG = 1 n i=1 yi - z 2 z The headcount, poverty gap, and squared poverty gap are part of the FGT class of poverty measures with the parameter taking a value of zero, one, and two in the following expression: (5) P = 1 q n i=1 yi - z z Contrary to what happens with monetary poverty measures, the (normalized) time poverty gap need not always be smaller than the time headcount index, and the squared time poverty gap need not be smaller than the time poverty gap. When using (z-y)/z as the household level indicator for consumption or income poverty, the normalization of (z-y) by z implies that we always have values that are between zero and one. For time poverty by contrast, because the definition in (5) relies instead on the value of (y-z)/z, we may have relatively large values for y-z, so that some values at the individual level may be larger than one, and the poverty gap may itself have a higher value than the headcount index in the aggregate (in case of confusion due to this, it would suffice to use an alternative normalization, such as (y-z)/168 when using weekly hours as the benchmark in order to ensure that time poverty measures are between zero and one.) Source: Bardasi and Wodon (2006a). 45 Table 2: Time Poverty Incidence and Poverty Gap, Ghana 1991/92 to 2005/06 1991/92 1998/99 2005/06 Urban Rural All Urban Rural All Urban Rural All Age Sex Time Poverty Incidence, time poverty line=1.5x median 7-24 Male 20.5 33.4 29.1 .. .. .. 19.9 30.1 26.5 Female 33.2 47.9 42.5 .. .. .. 26.8 40.6 35.1 All 27.0 40.3 35.6 .. .. .. 23.5 35.1 30.7 25-64 Male 11.1 5.8 7.6 15.5 6.4 9.6 17.5 8.5 12.3 Female 22.6 15.0 17.6 25.6 20.1 22.0 20.9 21.3 21.1 All 17.5 10.9 13.2 20.9 13.9 16.4 19.2 15.5 17.0 65+ Male 30.6 36.5 35.1 28.3 27.8 27.9 26.8 30.4 29.3 Female 30.6 34.6 33.6 27.5 35.2 32.9 21.1 30.8 27.6 All 30.6 35.5 34.3 27.9 32.0 30.7 23.5 30.6 28.4 Time Poverty Gap, time poverty line=1.5x median 7-24 Male 14.1 24.8 21.2 .. .. .. 21.3 25.7 24.1 Female 30.5 44.3 39.2 .. .. .. 30.0 42.7 37.6 All 22.4 34.1 30.0 ... .. .. 25.9 33.8 30.8 25-64 Male 1.8 0.8 1.2 3.2 1.2 1.9 3.0 1.4 2.1 Female 4.4 2.6 3.3 5.8 4.0 4.6 4.2 4.3 4.3 All 3.3 1.8 2.3 4.6 2.7 3.4 3.6 3.0 3.2 65+ Male 14.5 11.8 12.4 18.0 9.7 12.3 15.6 14.3 14.7 Female 17.4 18.2 18.0 18.3 18.1 18.2 12.3 17.4 15.8 All 16.1 15.1 15.3 18.1 14.4 15.6 13.8 16.0 15.3 Time Poverty Incidence, time poverty line=2x median 7-24 Male 12.7 23.5 19.9 .. .. .. 15.5 23.2 20.5 Female 23.6 35.0 30.8 .. .. .. 21.3 31.7 27.5 All 18.2 28.9 25.2 .. .. .. 18.5 27.2 23.9 25-64 Male 0.8 0.3 0.5 3.0 1.0 1.7 2.0 1.1 1.5 Female 4.6 2.2 3.0 6.7 3.7 4.8 4.1 4.5 4.3 All 2.9 1.4 1.9 5.0 2.5 3.4 3.1 2.9 3.0 65+ Male 15.3 12.2 12.9 17.8 10.8 13.1 16.2 14.1 14.7 Female 20.7 20.3 20.4 18.8 20.1 19.7 14.3 18.9 17.4 All 18.4 16.3 16.8 18.4 16.0 16.7 15.1 16.6 16.2 Time Poverty Gap, time poverty line=2x median 7-24 Male 6.6 11.4 9.8 .. .. .. 11.6 12.5 12.2 Female 15.8 22.8 20.3 .. .. .. 16.5 23.0 20.4 All 11.3 16.8 14.9 .. .. .. 14.2 17.5 16.2 25-64 Male 0.1 0.0 0.0 0.2 0.1 0.1 0.1 0.1 0.1 Female 0.3 0.2 0.2 0.5 0.3 0.4 0.3 0.3 0.3 All 0.2 0.1 0.1 0.4 0.2 0.3 0.2 0.2 0.2 65+ Male 5.2 2.9 3.4 7.6 2.5 4.1 6.4 5.1 5.5 Female 6.7 6.7 6.7 7.8 6.7 7.0 4.7 6.8 6.1 All 6.0 4.8 5.1 7.7 4.9 5.7 5.5 6.0 5.8 Source: Authors' estimation based on GLSS3 (1991/92), GLSS4 (1998/99) and GLSS5 (2005/06). Note: medians=17, 48 and 25, yielding lines of 25.5, 72 and 37.5 hours by age group when using 1.5x median, and yielding lines of 34, 96 and 50 hours when using 2x median. 46 3. Regression analysis for the determinants of time worked 4.17 The basic statistics on time use presented according to gender and age group in the previous section are useful, but they do not provide a precise idea of the correlates or determinants of the number of hours worked by individuals. For example, In order to assess the links between individual and households characteristics and total hours worked while controlling for the potential effect of other characteristics, regression analysis is needed. In table 5, regressions for the correlates of total hours worked are presented separately for urban men, urban women, rural men, and rural women, and for each of the three age groups. The dependent variable is the individual's total hours of work per week (as this variable appears to be roughly normally distributed, so that it is not needed to use a logarithmic transformation). 4.18 The independent variables include household demographic variables, as well as household access to water and electricity, and household location according to Ghana's agro-ecological zones (Savannah, Forest, and Coastal areas, plus the capital city of Accra). At the individual level, we control for age, education level, marital status, occupation status, and relationship to the household head. We also estimated regressions with as additional independent variables the quintiles of per capita consumption of households, but these variables in general were not statistically significant, and could also be considered as being endogenous (since household equivalent consumption itself depends on household income, and therefore on the number of hours worked of household members). The regressions presented in table 5 are in levels; regressions in log are in an appendix. 4.19 Although we provide the regression results for the three age groups, we focus first in the discussion on the adult population aged 25-64, since this is the segment for which the issue of time poverty is most important. There is no relationship between age and total hours worked for men, while for women, the relationship follows an inverted-U pattern: total hours worked increase first as the women gets older, and then decrease. In both urban and rural areas, the peak in total hours worked is reached between 35 and 40 years of age. Education is not related to total hours of work in a statistically significant way, except for individuals with post-secondary education in urban areas who tend to work four hours (for women) to five hours (for men) less than otherwise comparable individuals with lower levels of education. There is also a lack of relationship between marital status and total hours worked in most cases. As to the relationship with the head of households, for urban women, an interesting finding is that not being a family member leads to a large increase in working hours, probably because of the fact that many individuals in this position are likely to be domestic workers. For men, the impact is opposite. 4.20 The most important variables driving the total number of hours worked are the occupational status dummies. Any person working in the labor market is clearly spending a much larger amount of total time at work, and the impacts are large (at about 40 hours of work additional per week). In most samples, the highest additional workload is observed among private formal and informal wage workers, followed by the self-employed in non-agricultural occupations. The impact of being a public sector worker is slightly lower, and again a bit lower for the self employed in agriculture, probably in part due to seasonality issues in farm work. 47 Table 3: Determinants of Time Worked, by Locality and Sex, 2005/06 a) 7-24 year old Urban Urban Rural Rural Male Female Male Female Intercept 0.249 6.054 3.971 -12.160 *** Age 1.445 *** 2.290 *** 0.871 *** 3.036 *** Age squared -0.032 ** -0.048 *** -0.005 -0.063 *** Education level No education Primary -0.937 0.769 -0.863 -1.129 Secondary -0.553 -0.665 -2.214 ** -3.708 *** secondary2 -1.822 -0.422 -3.834 * -3.071 postsec -7.107 *** -3.094 -9.193 ** -2.774 Marital Status Single Married -5.357 -3.269 -1.872 1.563 Divorced 20.074 ** -6.443 * -1.362 7.157 ** Widowed -21.370 17.524 *** 4.739 Occupation Status No work Wage earner (public) 45.153 *** 42.578 *** 32.404 *** 23.652 *** Wage earner (private formal) 45.508 *** 41.054 *** 42.036 *** 23.585 *** Wage earner (private informal) 40.969 *** 50.989 *** 35.503 *** 28.954 *** Self-employed (agro) 23.070 *** 18.661 *** 27.027 *** 29.457 *** Self-employed (non-agro) 47.529 *** 42.803 *** 31.725 *** 32.020 *** Relationship to household head Head Spouse -4.460 -3.246 10.623 8.435 ** Son/daughter -0.307 -5.113 * -0.449 4.672 Other Family member -0.890 -4.177 -0.184 6.621 * Non Family member 5.577 * -1.502 -2.745 9.999 ** Household Demographic Composition Infants 0.714 1.693 0.332 -1.628 ** Infants squared 0.069 -0.495 -0.119 0.441 ** Children 0.009 0.212 0.075 0.436 Children squared -0.008 -0.070 -0.055 -0.055 Adults -1.792 ** -1.636 * -0.906 ** -0.613 Adults squared 0.148 * 0.113 0.047 -0.007 Seniors 4.123 6.191 0.690 -1.460 Seniors squared -3.185 * -2.668 -0.615 0.000 Geographical Location Accra -5.819 *** -7.815 *** Coastal -3.269 -4.594 2.072 -0.992 Forest -4.471 ** -4.802 * -1.482 -1.786 Savannah Infrastructure Water to proximity 3.383 *** -0.676 -1.407 -1.080 Electricity -0.511 -3.124 ** -0.923 -0.837 R2 Source: Authors' calculation based on GLSS5 (2005/06) Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Robust OLS Regressions 48 b) 25-64 year old Urban Urban Rural Rural Male Female Male. Female Intercept 14.934 * 12.158 12.167 ** 23.550 *** Age 0.010 0.793 ** 0.203 0.653 ** Age squared -0.001 -0.011 ** -0.004 -0.011 *** Education level No education Primary -0.819 -2.133 -0.069 -0.267 Secondary 0.292 -2.329 ** -1.443 0.486 secondary2 -2.045 1.235 -2.235 1.940 Postsec -5.017 *** -4.286 ** -1.956 0.890 Marital Status Single Married 0.691 0.175 -3.641 ** -2.521 Divorced 2.721 -0.011 -0.772 -3.135 Widowed -4.919 1.413 -0.943 -4.552 Occupation Status No work Wage earner (public) 47.289 *** 38.948 *** 41.808 *** 31.251 *** Wage earner (private formal) 52.094 *** 40.907 *** 54.936 *** 39.932 *** Wage earner (private informal) 48.072 *** 42.954 *** 47.874 *** 35.845 *** Self-employed (agro) 37.051 *** 35.927 *** 40.763 *** 39.149 *** Self-employed (non-agro) 49.154 *** 45.992 *** 44.567 *** 37.625 *** Relationship to household head Head Spouse 4.164 2.481 * 5.413 2.877 ** Son/daughter 2.777 0.092 -1.405 0.130 Other Family member 4.553 -1.159 -2.877 -5.601 *** Non Family member -19.640 ** 11.116 *** 0.715 -10.194 * Household Demographic Composition Infants -0.375 0.622 -2.079 *** -1.421 * Infants squared 0.157 -0.087 0.609 *** 0.441 ** Children -1.377 * -0.539 -0.579 -0.415 Children squared 0.476 *** 0.129 0.034 0.035 Adults -3.619 *** -3.113 *** -1.030 -1.916 *** Adults squared 0.433 *** 0.328 *** 0.050 0.077 * Seniors -1.924 -1.762 0.247 -0.608 Seniors squared 0.677 0.538 -0.491 0.264 Geographical Location Accra 2.630 -2.829 Coastal 0.563 -3.312 -1.800 -4.578 ** Forest -2.027 -4.672 ** -5.103 *** -8.298 *** Savannah Infrastructure Water to proximity 0.544 0.738 0.808 -2.787 ** Electricity -0.493 -0.883 1.063 0.801 R2 Source: Authors' calculation based on GLSS5 (2005/06) Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Robust OLS Regressions 49 c) 65+ year old Urban Urban Rural Rural Male Female Male. Female Intercept -186.07 * 13.914 54.733 116.197 ** Age 4.801 * 0.030 -1.045 -2.118 * Age squared -0.034 ** -0.002 0.006 0.012 * Education level No education Primary -0.363 -6.032 ** 0.117 -2.118 Secondary -5.726 ** 1.203 -2.711 -1.630 secondary2 -3.820 -6.654 15.807 .. postsec -4.792 * 0.777 -4.733 1.291 Marital Status Single Married 13.122 -4.790 -8.088 * -10.868 ** Divorced 16.486 -5.356 3.503 -17.525 *** Widowed 18.689 -6.340 -5.071 -13.898 *** Occupation Status No work Wage earner (public) 58.260 *** 34.867 *** 56.061 *** Wage earner (private formal) 51.041 *** *** 39.887 *** Wage earner (private informal) 44.404 *** 58.850 *** 66.072 *** 56.536 *** Self-employed (agro) 27.481 *** 31.082 *** 33.904 *** 33.635 *** Self-employed (non-agro) 44.347 *** 44.360 *** 30.292 *** 40.013 *** Relationship to household head Head Spouse -6.429 0.438 2.565 1.925 Son/daughter 5.229 Other Family member 2.926 -3.410 * -3.881 * -1.853 Non Family member -4.859 -11.741 Household Demographic Composition Infants 10.306 *** -6.124 -1.002 -0.999 Infants squared -3.679 *** 3.955 * 0.223 -0.098 Children -3.498 -2.980 1.124 0.249 Children squared 0.543 0.856 -0.046 0.025 Adults -1.674 3.484 ** -1.873 ** -2.857 ** Adults squared 0.129 -0.654 ** 0.154 ** 0.152 Seniors 16.310 8.881 2.476 -0.724 Seniors squared -4.610 -2.826 -1.221 -0.065 Geographical Location Accra -2.856 -3.523 Coastal 2.195 -2.114 0.329 -0.994 Forest -6.614 * -4.263 ** -1.893 -4.760 ** Savannah Infrastructure Water to proximity 1.521 4.680 ** 0.076 0.561 Electricity 4.221 -0.430 -0.555 4.114 ** R2 Source: Authors' calculation based on GLSS5 (2005/06) Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Robust OLS Regressions 50 4.21 Other variables have a limited impact on the total number of hours worked. For example, the impact of a higher number of children or infants is not clear-cut, although a higher number of adults in the household tends to reduce the work load of both men and women. Somewhat surprisingly, access to water and electricity are not strongly correlated in many cases to total work time, and where there is a statistical impact, it is as often positive as negative. This is a somewhat surprising finding, because in other countries, access to basic infrastructure services was found to decrease total hours of work, although the impact is mostly significant for domestic work, and there may well be a substitution effect between domestic and productive work here (as domestic work is reduced, individuals do more productive work, so that there is no clear relationship in the regression between infrastructure and total time work). Finally, individual in the forest area tend to work less than those living elsewhere. 4.22 We now briefly turn to the results for the other age groups. For the 7-24 age group, workload increase with age and decreases with the education level of the child, at least at higher levels of schooling. There is some evidence that time work increases (for urban boys and rural girls) when the individual is divorced (which is likely to happen only for the older group of youths in the regression, and rarely so). When the child or youth is working, time work increases very significantly. When the child is not related to the household head, time worked also increases, at least for urban boys/males and rural girls/females. This may perhaps reflect the impact of orphanhood, whereby orphans who are welcomed by other families must work more than other children (see for example Siaens et al., 2006). Most of the variables related to demographic structure and geographic location are not statistically significant, except for the fact that individuals located in the Forest area appear to work less. Surprisingly, access to water in the proximity of the household's dwelling increases time work for urban boys/males. Many of the coefficients that are statistically significant for the 7-24 age group are also statistically significant for the elderly above 64 years of age. 4.23 The insights provided by the regressions are thus limited, and it would be interesting in future work to look more in details at the joint patterns of domestic and productive work, for example to assess to what extent access to infrastructure, while not having an impact on total time work, helps reduces domestic time worked to the benefit of productive work. 4. Conclusion 4.24 Who works the most in Ghana? To a large extent, the results provided in this paper confirm conventional wisdom. Women are found to work much more than men, especially on domestic tasks and in rural areas. The workload of a rural adult female individual reaches more to 50 hours per week, a level that would be considered as a full-time occupation in many countries. By contrast, adult male individuals work on average more than five hours less. The differences in workload between men and women are much larger in rural areas (8.5 hours differential) than in urban areas (difference of less than one hour in time worked). Thus, while it is often argued on the basis of labor statistics that women participate less than men in the labor force of developing countries such as Ghana, especially in urban areas, this reflects more the impact of traditional definition of what constitutes work in household survey questionnaires than any real differences between genders, as women tend to work at least as hard as men. 4.25 A second finding is that for many children, the burden of domestic work is high as well, reaching more than 20 hours per week on average in some cases. A third and surprising finding was that access to basic infrastructure services (water and electricity) did not affect in a statistically significant way total time worked, but this may be because better access to infrastructure reduces the 51 amount of time spent on domestic work, with this time then used by workers to increase time for productive work. References Bardasi, E. and Q. Wodon, 2006a, Measuring Time Poverty and Analyzing Its Determinants: Concepts and Applications to Guinea, in Blackden, C. M., and Q. Wodon, editors, Gender, Time Use and Poverty in Sub-Saharan Africa, World Bank Working Paper No. 73, Washington, DC. Bardasi, E. and Q. Wodon, 2006b, Poverty Reduction from Full Employment: A Time Use Approach, in Blackden, C. M., and Q. Wodon, editors, Gender, Time Use and Poverty in Sub-Saharan Africa, World Bank Working Paper No. 73, Washington, DC. Barwell, I., 1996, Transport and the Village: Findings from African Village-Level Travel and Transport Surveys and Related Studies, World Bank Discussion Paper No. 344, Africa Region Series, Washington, D.C. Berio, A. J., 1983, Time Allocation Surveys, Paper presented at the 11th International Congress of Anthropology Sciences, Vancouver, Canada. Blackden, C. M., and C. Bhanu, 1999, Gender, Growth, and Poverty Reduction, Special Program of Assistance for Africa 1998 Status Report on Poverty, World Bank Technical Paper No. 428, Washington D.C. Blackden, C. M., and Q. Wodon, editors, 2006, Gender, Time Use and Poverty in Sub-Saharan Africa, World Bank Working Paper No. 73, Washington, DC. Charmes, J. 2006, A Review of Empirical Evidence on Time Use in Africa from UN-sponsored Surveys, in Blackden, C. M., and Q. Wodon, editors, Gender, Time Use and Poverty in Sub-Saharan Africa, World Bank Working Paper No. 73, Washington, DC. Correia, M., 1999, Las Relaciones de Género en la Argentina, World Bank, Buenos Aires. Gerstel, N., and H. Gross, 1987, Families and Work (Women in the Political Economy), Temple University Press, Philadelphia. Hartman, H., 1987, The Family as The Locus of Gender, Class, and Political Struggle. The example of Housework, in S. Harding, ed., Feminism and Methodology, Indiana University Press & Open University Press, Bloomington, IN. Harvey, A. S., and M. E. Taylor, 2002, Time Use, in M. Grosh and P. Glewwe, Editors, Designing Household Survey Questionnaires for Developing Countries, Lessons from 15 Years of the Living Standards Measurement Survey, World Bank, Washington D.C. Ilahi, N., 2000, The Intra-household Allocation of Time and Tasks: What Have We Learnt from the Empirical Literature?, Policy Research Report on Gender and Development, Working Paper Series No. 13., World Bank, Washington, DC. Ilahi, N., and F. Grimard, 2001, Public Infrastructure and Private Costs: Water Supply and Time 52 Allocation of Women in Rural Pakistan, Economic Development and Cultural Change 49: 45-75. Malmberg-Calvo, C., 1994, Case Study on the Role of Women in Rural Transport: Access of Women to Domestic Facilities, SSATP Working Paper No. 11, Technical Department, Africa Region, World Bank. Mann, Susan Archer. 1990, The Civilization of Nature, in Agrarian Capitalism in Theory and Practice, The University of North Carolina Press, Chapel Hill, NC Staudt, K., 1994, Technical Assistance and Women: From Mainstreaming Toward Institutional Accountability. UN Technical Paper, Division for the Advancement of Women. Siaens, C., K. Subbarao and Q. Wodon, 2006, Assessing the Welfare of Orphans in Rwanda: Poverty, Work, Schooling, and Health, in C. M. Blackden and Q. Wodon, editors, Gender, Time Use and Poverty in Sub-Saharan Africa, World Bank Working Paper No. 73, Washington, DC, 135-152. Wodon, Q., and K. Beegle, 2006, Labor Shortages Despite Underemployment? Seasonality in Time Use in Malawi, in Blackden, C. M., and Q. Wodon, editors, Gender, Time Use and Poverty in Sub-Saharan Africa, World Bank Working Paper No. 73, Washington, DC. World Bank, 2001, Engendering Development: Through Gender Equality in Rights, Resources, and Voice, World Bank Policy Research Report, Washington, D.C. 53 Appendix to Annex 4: Determinants of Logarithm of Total Time Worked, by Locality and Sex, 2005/06 a) 7-24 year old Urban Urban Rural Rural Male Female Male Female Intercept 0.590 0.337 1.013 *** 0.452 ** Age 0.203 *** 0.296 *** 0.133 *** 0.245 *** Age squared -0.005 *** -0.007 *** -0.003 *** -0.006 *** Education level No education Primary -0.027 0.060 0.015 -0.006 Secondary -0.101 -0.098 -0.119 * -0.126 ** secondary2 -0.063 -0.055 -0.097 -0.018 postsec -0.366 0.111 -0.444 -0.448 Marital Status Single Married -0.012 -0.206 -0.034 0.027 Divorced 0.445 ** -0.291 * 0.081 0.138 Widowed -0.933 ** 0.323 *** 0.298 * Occupation Status No work Wage earner (public) 2.490 *** 1.206 * 1.724 *** 1.387 *** Wage earner (private formal) 2.269 *** 1.661 *** 1.838 *** 0.909 *** Wage earner (private informal) 2.129 *** 1.745 *** 1.788 *** 0.988 *** Self-employed (agro) 1.591 *** 0.884 *** 1.578 *** 1.169 *** Self-employed (non-agro) 2.313 *** 1.550 *** 1.621 *** 1.161 *** Relationship to household head Head Spouse -0.454 0.190 0.441 0.209 ** Son/daughter 0.129 -0.007 0.062 0.152 Other Family member 0.133 0.063 0.066 0.206 * Non Family member 0.517 *** 0.192 0.010 0.476 *** Household Demographic Composition Infants 0.079 0.083 0.017 -0.063 * Infants squared 0.019 -0.011 -0.004 0.020 ** Children -0.071 0.014 -0.017 0.012 Children squared 0.006 -0.012 ** -0.002 0.000 Adults -0.148 ** -0.119 * -0.082 *** -0.081 *** Adults squared 0.009 0.007 0.006 ** 0.004 Seniors 0.255 0.453 ** -0.072 -0.043 Seniors squared -0.159 * -0.250 0.008 -0.019 Geographical Location Accra -0.695 *** -0.671 *** Coastal -0.301 -0.349 ** 0.163 0.045 Forest -0.490 *** -0.331 ** -0.064 -0.030 Savannah Infrastructure Water to proximity 0.122 -0.113 -0.125 ** -0.131 ** Electricity -0.169 ** -0.259 *** -0.022 -0.036 R2 0.39 0.43 0.45 0.41 Source: Authors' calculation based on GLSS5 (2005/06) Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Robust OLS Regressions 54 b) 25-64 year old Rural Urban Male Urban Female Rural Male. Female Intercept 1.775 *** 2.332 *** 1.636 *** 2.788 *** Age -0.012 0.031 ** 0.013 0.019 ** Age squared 0.000 0.000 *** 0.000 * 0.000 *** Education level No education Primary -0.064 -0.050 0.022 -0.017 Secondary 0.000 -0.062 * -0.040 -0.008 secondary2 -0.060 0.029 -0.027 0.038 Postsec -0.085 -0.183 ** -0.198 * 0.073 Marital Status Single Married -0.018 -0.007 -0.115 ** -0.002 Divorced 0.097 0.026 -0.078 -0.001 Widowed 0.001 0.051 0.001 -0.084 Occupation Status No work Wage earner (public) 2.563 *** 1.466 *** 2.252 *** 1.053 *** Wage earner (private formal) 2.626 *** 1.502 *** 2.411 *** 1.214 *** Wage earner (private informal) 2.541 *** 1.488 *** 2.242 *** 1.134 *** Self-employed (agro) 2.346 *** 1.339 *** 2.157 *** 1.201 *** Self-employed (non-agro) 2.586 *** 1.532 *** 2.175 *** 1.146 *** Relationship to household head Head Spouse 0.236 0.127 *** 0.059 0.059 Son/daughter 0.166 * -0.032 0.004 -0.054 Other Family member 0.154 -0.139 -0.136 -0.172 *** Non Family member -1.035 ** 0.349 *** 0.086 -0.146 Household Demographic Composition Infants -0.016 0.002 -0.058 ** -0.025 Infants squared 0.009 0.001 0.016 ** 0.006 Children -0.020 -0.015 -0.009 -0.015 Children squared 0.011 0.004 0.000 0.001 Adults -0.095 * -0.093 *** -0.023 -0.045 *** Adults squared 0.012 ** 0.008 * 0.000 0.003 ** Seniors 0.072 -0.030 -0.097 -0.027 Seniors squared -0.081 0.012 0.047 0.019 Geographical Location Accra 0.062 -0.182 ** Coastal 0.059 -0.156 * -0.044 -0.107 ** Forest -0.085 -0.133 ** -0.147 *** -0.152 *** Savannah Infrastructure Water to proximity 0.002 -0.010 0.014 -0.080 *** Electricity -0.002 -0.053 0.020 0.027 R2 0.53 0.50 0.40 0.35 Source: Authors' calculation based on GLSS5 (2005/06) Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Robust OLS Regressions 55 c) 65+ year old Urban Urban Rural Rural Male Female Male. Female Intercept -4.945 -6.415 1.452 2.844 Age 0.202 0.273 0.019 0.014 Age squared -0.002 -0.002 0.000 0.000 Education level No education Primary 0.253 0.002 0.144 -0.098 Secondary -0.123 0.353 -0.145 -0.022 secondary2 0.040 0.937 *** 0.357 postsec -0.249 0.411 -0.365 0.106 Marital Status Single Married -0.132 -0.556 -0.372 ** -0.945 *** Divorced 0.226 -0.560 -0.268 -1.116 *** Widowed 0.245 -0.489 -0.310 -1.011 *** Occupation Status No work Wage earner (public) 2.977 *** 0.669 2.375 *** Wage earner (private formal) 2.563 *** 2.198 *** Wage earner (private informal) 2.668 *** 2.281 *** 2.580 *** 2.140 *** Self-employed (agro) 1.839 *** 1.663 *** 2.023 *** 1.628 *** Self-employed (non-agro) 2.324 *** 1.915 *** 1.704 *** 1.651 *** Relationship to household head Head Spouse -0.972 ** 0.550 0.394 0.155 Son/daughter 1.539 *** Other Family member 0.297 -0.539 ** -0.455 ** -0.045 Non Family member -0.078 Household Demographic Composition Infants 0.186 *** -0.465 -0.046 0.012 Infants squared -0.029 *** 0.289 * -0.007 -0.035 Children 0.004 -0.147 0.040 -0.039 Children squared 0.016 0.056 0.001 0.001 Adults -0.076 0.033 -0.063 -0.160 *** Adults squared 0.014 -0.018 0.005 0.016 * Seniors -0.147 -0.024 0.270 Seniors squared 0.131 * 0.027 -0.023 -0.075 Geographical Location Accra -0.649 ** -1.257 *** Coastal 0.015 -0.463 * 0.039 0.088 Forest -0.395 * -0.404 ** -0.168 ** -0.054 Savannah Infrastructure Water to proximity 0.000 0.092 -0.028 -0.073 Electricity -0.018 0.070 -0.030 0.179 ** R2 0.63 0.58 0.47 0.54 Source: Authors' calculation based on GLSS5 (2005/06) Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Robust OLS Regressions 56 ANNEX 5 ESTIMATING THE POTENTIAL COST FOR FIRMS IN GHANA OF ENFORCING THE MINIMUM WAGE 14 This paper provides a simple method inspired from the poverty measurement literature to assess what could be the potential impact of increasing the minimum wage over time and enforcing this minimum wage on the cost for firms of employing workers. We illustrate the approach using data from Ghana, a country in which the minimum wage has been rising much faster than inflation over the last 15 years. Two key parameters affect the potential cost for firms of enforcing the minimum wage. The first is the proportion of wage workers who have wages below the minimum wage. The second is the average distance separating the wages of workers from the minimum wage for those workers who have a wage below the minimum wage. The product of both parameters gives an estimate of the potential cost for firms of enforcing the minimum wage as a proportion of the minimum wage. For the unemployed, the method is applied using both an imputed wage obtained from regression analysis and information on the worker's reservation wage available in the data. In Ghana the potential cost for firms that would arise from an enforcement of the minimum wage has increased sharply over the last 15 years for both employed wage workers, and for the unemployed looking for work. 1. Introduction 5.1 For workers, there is a clear trade-off between the wage and employment effects of an increase in the minimum wage. Increasing the minimum wage may lead to higher earnings for low- skill workers, but it may also reduce employment opportunities if workers become too expensive, so that firms may not hire anymore or may hire less. This means that the overall impact on workers in unclear ŕ priori15. In developed countries, the impact of increasing the minimum wage on the population as a whole is likely to be limited because only a small share of workers are likely to be affected. In developing countries, if there were stronger enforcement of the minimum wage, the impact on both wages and employment could be much larger, because the minimum wage is relatively high in comparison to the earnings of workers. 5.2 From the point of view of firms, the impact of an increase in the minimum wage is in principle easier to predict, since it is likely to result in an increase in production costs, and thus in a reduction in profits and perhaps in the size of small firms, especially in the service sector in developed economies. Yet, while there is a large literature on the impact of the minimum wage on labor outcomes for workers, less has been written on the impact on firms. Research on the UK suggests that while the minimum wage did not have large employment effects, it did contributed to higher wages for low-skills workers, a decrease in firm profitability and higher prices for selected consumer services, but not to an increase in firm exists, suggesting that the main effect of the minimum wage was to redistribute quasi-rents from firms and consumers to low-skill workers (Draco et al., 2006; Metcalf, 2006). In Ireland, the introduction of the minimum wage may also have led to a reduction in the size of a small number of firms that were relying the most on low-skills workers, but the impact was modest (O'Neil et al., 2006). 5.3 The analysis of the impact of the minimum wage on firms in developing countries is even more limited. Harrison and Corse (2004) use data from Indonesia to assess the impact on firms and 14This background paper was prepared by: Harold Coulombe and Quentin Wodon, January 2008. 15See among many others Brown, Gilroy and Kohen (1982); Burkhauser and Finegan (1989); Mincy (1991); Card and Krueger (1995); Bell (1997); Addison and Blackburn (1999); Saget (2001); Pereira (2003); Yuen (2003); and Angel-Urdinola and Wodon (2004, 2005). 57 workers of a doubling of the minimum wage as well as pressures from human rights and anti- sweatshop activists to improve workers' conditions. Wages for workers at targeted plants increased dramatically, but this led some firms to leave Indonesia and relocate elsewhere. Using survey data from small-scale manufacturing enterprises in Ghana's Central Region, Mensah et al. (2007) find that most industry groups would become unprofitable if the legal minimum wage were used to value the time spent on their firms by the firm owners. 5.4 In this paper our objective in this paper is to propose a very simple method inspired from the poverty measurement literature to assess what could be the potential impact of increasing the minimum wage over time and enforcing this minimum wage on the cost for firms of employing workers. The method can also be used to assess the wage gap that separate workers from what they could earn if the minimum wage were perfectly enforced. We illustrate the approach using data from Ghana, a country in which the minimum wage has been rising much faster than inflation over the last 15 years. As stipulated in the Labor Act, the minimum wage is set by a National Tripartite Committee with representatives of workers unions, employers associations, and the Government. In 2005/06, the year of the latest Ghana Living Standard Measurement survey, the minimum wage was at 13500 cedis per day (US$ 1.5), versus the equivalent in real terms (adjusted for inflation) of 8242 cedis per day in 1998/99, and 5985 cedis per day in 1991/92. Thus, in real terms, the minimum wage has more than doubled in the last 15 years. 5.5 Our methodology works as follows. We aim to assess how much more firms would have to pay for wage earners if the workers had to be paid at least the minimum wage (assuming for simplicity no employment effects, so that firms are not reducing the number of workers on their payroll after an increase in minimum wage). Two key variables then affect the potential cost for firms of enforcing the minimum wage. The first parameter is the proportion of workers who have wages below the minimum wage. This is equivalent to the headcount index in poverty measurement. The second parameter is the average distance separating the wages of workers from the minimum wage when workers have a wage below the minimum wage. This is the income gap in poverty measurement. The product of the share of workers below the minimum wage and the average distance separating these workers from the minimum wage gives us our estimate of the potential cost for firms of enforcing the minimum wage. 5.6 We compute the potential cost of enforcing the minimum wage for firms for both for current workers and the unemployed. In the case of the unemployed, we compare their imputed wage (using wage regressions) as well as their reservation wage to the minimum wage to make a determination as to whether the minimum wage would raise their earnings if it were enforced. In the analysis, we also distinguish young and older workers, as we expect that changes in the minimum wages, if they were enforced, would have a stronger impact on younger workers who are often located at the lower tail of the distribution of wages (e.g., Abowd et al., 2000; Angel-Urdinola and Wodon, 2005). Overall, we show that for both current wage earners and the unemployed, the potential cost for firms of enforcing the minimum wage has been rising rapidly in Ghana due to the progressive rise in real terms of the minimum wage over time. 5.7 We wish not to imply that in the specific case of Ghana, the higher level of the minimum wage over time has actually led to an increase in costs for firms, as well as to other potential effects. Indeed, the minimum wage is not strictly enforced in Ghana, so that its impact is likely to be limited. In this respect, it is worth noting that data from the Doing Business indicators published annually by the World Bank suggest that Ghana's rigidity of employment index, at 37, is not very high (the index takes on a value between zero and 100). This index is itself the sum of three sub-indices, one of which is the hiring index that takes into account the flexibility of contracts and the ratio of minimum wage to the value-added per worker in the economy. Given that Ghana's hiring index is only 22, the 58 country compares favorably to many other developing countries. At the same time, one should also acknowledge that previous research on Ghana by Jones (1988) has suggested that in the 1970s and 1980s, minimum wage policy did alter employment patterns, with higher minimum wages leading to displacement of workers from the formal to the informal sector. When the minimum wage is high, as is starting to be the case in Ghana, enforcement could well have displacement effects for workers out of the formal sector. 5.8 The paper is structured as follows. In section 2, we present our methodology. Section 3 provides our empirical results. A brief conclusion follows. 2. Methodology 5.9 Our methodology is very simple, as it consists of using measures inspired from the poverty literature to assess the potential cost for firms of an enforcement of the minimum wage (for a review of poverty measurement, see Coudouel et al., 2002). We use the equivalent of the first two poverty measures of the FGT class (Foster, Greer, and Thorbecke 1984), namely the headcount and the poverty gap. In our context, the headcount is the share of wage earners with wages y below the minimum wage mw. Suppose we have a population of n wage earners in which q workers are earning less than the minimum wage. The headcount index is defined as: H = q (1) n 5.10 The wage gap is defined as the mean distance separating the worker's wages from the minimum wage, with workers earning more than the minimum wage being given a distance of zero. The wage gap is a measure of the wage deficit of the entire population of wage earners, where the notion of "wage deficit" captures the additional resources (as a proportion of the minimum wage) that firms would need to allocate to workers in lift all wage earners up to the minimum wage. It is defined as: WG = 1 q (2) n i =1mw - wi mw where wi is the wage of individual i, and the sum is taken on individuals who are earning less than the minimum wage. The wage gap can be written as being equal to the product of the wage gap ratio and the headcount index defined above, where the wage gap ratio is defined as: WGR = mw - wq 1 q where wq = wi is the average wage below the mimimum wage. (3) mw qi=1 5.11 It must be emphasized that the wage gap ratio WGR in itself is not a good measure of the potential cost for firms of increasing (or reducing) the minimum wage. Assume that the minimum wage is reduced, and that some workers who were below the minimum wage are now above that minimum. The wage gap ratio may very well increase if the mean distance separating wage earners below the new minimum wage increases (this may happen because some of those who were closest to the minimum wage have now a wage above the minimum) ­ so that those workers who remain below the minimum are on average further away from the minimum wage. This would suggest an increase in the potential cost for firms of an enforcement of the minimum wage, while costs have clearly decreased since the minimum wage has decreased. Although in this example the wage gap ratio will increase, the wage gap itself will decrease, because the headcount index will decrease, suggesting a reduction in the potential cost for firms of enforcing the minimum wage. The problem with the wage gap ratio is that it is defined only on the workers below the minimum wage, while the wage gap is defined over all workers. 59 5.12 Finally, the squared wage gap can be described as a measure of the severity of the difference between the minimum wage and the wage paid to workers. While the wage gap takes into account the distance separating workers from the minimum wage, the squared wage gap takes the square of that distance into account. When using the squared wage gap, the wage gap is weighted by itself, so as to give more weight to those far away from the minimum wage. From a firm point of view, this places in the estimation of the cost of the minimum wage more emphasis on those workers whose productivity may be lower (assuming the wages paid to them reflect this productivity), but it is not fully clear that this would be a relevant parameter for a firm. For workers by contrast, this is a relevant parameter as it measures how much workers loose out in wages when the minimum wage is not enforced. The squared wage gap is: W 2 = 1 q (4) n i =1mw - wi 2 mw 5.13 The headcount, the wage gap, and the squared wage gap are equivalent to the first three poverty measures of the Foster-Greer-Thorbecke class. The general formula for this class of measures depends on a parameter which takes a value of zero for the headcount, one for the wage gap, and two for the squared wage gap as follows: W = 1 q (5) n i=1mw - yi mw 3. Empirical Results 5.14 To implement the above methodology, we use repeated rounds of the Ghana Living Standards Surveys for 1991/92, 1998/99 and 2005/06. The surveys include a detailed labor module, with information on employment and wages, including reservation wages for the unemployed. We restrict our analysis to wage earners, and to the unemployed (since they could become wage earners, and unemployment is higher in urban areas where opportunities to become a wage earner are also higher). We do not include in our analysis workers who work without pay, or are self-employed in agriculture or outside of agriculture. Data are provided separately for workers aged 18 to 30, and for workers aged 25 to 64. There is a bit of overlap between the two groups, but we defined young workers as being between 18 and 30 years of age in order to increase the sample size for that group, and also because youth are defined in Ghana as a rather large group (for example in the context of the National Youth Employment Program recently put in place by the authorities, youth are defined as those between 18 and 35 years of age; see Coulombe et al., 2008). 5.15 Tables 1 and 2 provide basic statistics on the number of workers in various categories and their mean wage earnings. Data are provided on actual earnings and imputed earnings for wage earners, and on imputed wages and reservation wages for the unemployed. Imputed wages are based on estimates of the likely earnings that the unemployed would command using regression techniques (see the Annex). All statistics are presented in terms of monthly wages. We could have expressed the statistics in hourly wages using data on the time spent working available in the time use module of the survey, and this is done in appendix, but since the minimum wage is actually set per day, we feel that the statistics based on monthly wage data are probably a better estimate of what people earn in a typical day of full work than the statistics based on hourly wages. All the results are qualitatively similar in terms of trend over time in the potential cost of enforcing the minimum wage whether we use monthly of hourly wage data, although in terms of the level of the estimates, the potential cost for firms of enforcing the minimum wage is higher when estimated using monthly as opposed to hourly wage data. We present here the estimates based on monthly wages only as we deem them to be more realistic and representative of the situation on the ground in the country. 60 5.16 The basic statistics from tables 1 and 2 are complemented by Figures 1 to 5, which graph the distribution function of current wage earners, as well as the expected wage and the reservation wage of the unemployed. The vertical line on the figures represent the values of the minimum wage in nominal terms for each specific year. The data are provided for all wage workers as well as unemployed workers aged 18 to 24 in 1991/92 and 2005/06, and for wage workers and the unemployed between 25 and 64 years of age for all three survey years, including 1998/99. In 1998/99, we do not provide statistics for workers aged 18 to 24 because for that year, the survey did not ask to youth that were studying whether they were also working and earning wages, so that estimates would not be comparable with the other two surveys. 5.17 The messages that emerge from tables 1 and 2 and from Figures 1 to 5 are very clear. The minimum wage has increased rapidly, and while in 1991/92, very few wage workers were earning less than the minimum wage, this was not the case anymore in 2005/06. The fact that so many wage workers are earning less than the minimum wage in 2005/06 also shows that enforcement is weak. Finally, there is an interesting discrepancy between the imputed wages for the unemployed and their reservation wage, in that many workers report a reservation wage higher than what they could expect on the market given their characteristics. 5.18 Table 3 provides our estimates of the potential cost for firms of enforcing the minimum wage. The main parameter of interest is the wage gap, since this is expressed in quasi monetary terms, as a proportion of the minimum wage. For workers employed and between the ages of 18 and 30, the wage gap has increased at the national level from 3.7 percent in 1991/92 to 15.1 percent in 2005/06. The increase is even larger for the unemployed according to their imputed wages, from 0.3 percent in 1991/92 to 20.4 percent in 2005/06. For the unemployed, using the reservation wage, we find a smaller but still significant increase in the wage gap from 1.1 percent in 1991/92 to 5.6 percent in 2005/06. For workers between the ages of 25 and 64, the wage gap has increased as well, although the level is smaller than for younger workers. As expected, the wage gaps are larger in rural than in urban areas, but they are still substantial even in urban areas. In 2005/06 for example, for current wage earners between the age of 18 and 30, the wage gap in urban areas is at 13.9 percent, while it is at 6.6 percent for the older age group. 5.19 The increase in the wage gap is due to both an increase in the headcount index of workers below the minimum wage, and an increase in the wage gap ratio over time. Data on the headcount indices are provided in table 3 as well. For example, in urban areas, 32.1 percent of actual wage earners today are below the minimum wage among the 18 to 30 years age group, and the proportion is 17.4 percent for older workers. For the unemployed, the proportion is a bit lower on the basis of the worker's declared reservation wage, but it is significantly higher on the basis of the imputed wage. 5.20 In table 4, the exercise is repeated among current workers by distinguishing between those in the formal and informal sectors (this can be done only for those who already work, since we do not know for the unemployed whether they would join the formal or informal sector). A worker is considered as belonging to the formal sector if one or more of the following is observed: the worker has a written contract, paid holidays, social security, sick or maternity leave, pension benefits, subsidized free medical care, or works in an organization with a trade union. The results suggest that headcount index of workers below the minimum wage as well as the other measures (wage gap and squared wage gap) are substantially higher for informal than for formal sector workers, as expected. But for both groups of workers, there is a large increase in these measures over time, and the potential cost for firms of enforcing the minimum wage is higher for youth as opposed to older workers. 61 5.21 The values of the wage gap ratios are provided in table 5, and they have increased substantially as well between 1991/92 and 2005/06, although the increase has been smaller than that observed for the headcount index, wage gap, and squared wage gap. Said differently, the largest contribution to the increase in these measures comes from the fact that a larger share of workers have actual or imputed earnings before the minimum wage, rather than form the fact that there has also been an increase in the distance separating those workers from the minimum wage (even though this increase took place as well for most groups). 4. Conclusion 5.22 We have provided in this paper a simple method to assess the potential cost for firms of enforcing the minimum wage, with an example for Ghana, a country where the minimum wage has increased substantially. The method is based on the poverty literature, and the results can also be interpreted as a measure of the losses in wages for workers due to the lack of enforcement of the minimum wage. In the case of Ghana, between 1991/92 and 2005/06, there has been a very large increase in the share of workers (both current workers and the unemployed) with actual or predicted wages below the minimum wage, and the distance separating the wages received by these workers and the minimum wage has also increased substantially. This combined increase implies that if the minimum wage were to be enforced, the cost for firms could today be potentially large, especially for youth. This was not the case fifteen years ago. 5.23 As explained in the introduction, we do not wish to imply from our work that the higher level of the minimum wage over time has actually led to an increase in costs for firms. In fact, the minimum wage in Ghana is poorly enforced, as is the case in other developing countries, simply because most of the economy is informal, even in urban areas. Yet at the same time, previous research has shown that in Ghana, high minimum wages have led in the past to the informalization of part of the labor force. While we have not tested here whether this has happened in Ghana over the last fifteen years, our work does suggest that the pressure for firms of going informal, at least with respect to the minimum wage, may have increased over time. 62 Figure 1: Density Function for Wages, Age 18-30, 1991/92 5 1. yit 1 ens D nel er K .5 0 2 4 6 8 10 Monthly Wage in constant January 2006 cedis (in '000, in log) Minimum wage Imputed wages - Unemployed Actual wages Reservation wage Source: Authors' computation based on GLSS3 Figure 2: Density Function for Wages, Age 18-30, 2005/06 1 .8 yti .6 ens D nel er .4 K .2 0 2 4 6 8 10 Monthly Wage in constant January 2006 cedis (in '000, in log) Minimum wage Imputed wages - Unemployed Actual wages Reservation wage Source: Authors' computation based on GLSS5 63 Figure 3: Density Function for Wages, Age 25-64, 1991/92 5 1. yit 1 ens D nel er K .5 0 2 4 6 8 10 Monthly Wage in constant January 2006 cedis (in '000, in log) Minimum wage Imputed wages - Unemployed Actual wages Reservation wage Source: Authors' computation based on GLSS3 Figure 4: Density Function for Wages, Age 25-64, 1998/99 1 .8 yt sine .6 Dlenr .4 Ke .2 0 2 4 6 8 10 12 Monthly Wage in constant January 2006 cedis (in '000, in log) Minimum wage Imputed wages - Unemployed Actual wages Reservation wage Source: Authors' computation based on GLSS4 64 Figure 5: Density Function for Wages, Age 25-64, 2005/06 1 .8 yit .6 Dens nel er .4 K .2 0 2 4 6 8 10 Monthly Wage in constant January 2006 cedis (in '000, in log) Minimum wage Imputed wages - Unemployed Actual wages Reservation wage Source: Authors' computation based on GLSS5 65 Table 1: Basic Statistics on Number of Individuals in Sample, Ghana 1991-2006 18-30 years old 25-64 years old 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 Number of wage earners Urban 142,316 .. 317,842 470,854 449,790 878,264 Rural 80,242 .. 139,950 264,193 275,655 302,914 Total 222,558 .. 457,792 735,047 725,445 1,181,178 Number of unemployed Urban 124,905 .. 269,326 123,391 169,200 273,680 Rural 49,205 .. 174,782 60,560 112,800 240,714 Total 174,110 .. 444,108 183,951 282,000 514,394 Number of wage earners below minimum wage (actual wages) Urban 9,841 .. 101,386 24,224 26,790 133,108 Rural 9,841 .. 47,272 14,383 40,185 72,152 Total 19,682 .. 148,658 38,607 66,975 205,260 Number of wage earners below minimum wage (imputed wages) Urban - .. 93,922 - - 19,282 Rural - .. 47,894 - 2,115 23,636 Total - .. 141,816 - 2,115 42,918 Number of unemployed below minimum wage (imputed wages) Urban 2,271 .. 146,170 - 705 44,162 Rural 3,028 .. 154,878 - 11,280 143,682 Total 5,299 .. 301,048 - 11,985 187,844 Number of unemployed below minimum wage (reservation wages) Urban 2,271 .. 32,966 757 4,230 34,832 Rural 3,785 .. 42,918 4,542 15,510 82,726 Total 6,056 .. 75,884 5,299 19,740 117,558 Source: Authors' estimates using GLSS data. 66 Table 2: Basic Statistics on Wages (Actual and Imputed) in Sample, Ghana 1991-2006 18-30 years old 25-64 years old 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 Mean monthly wage, wage earners (actual wages) Urban 564.5 .. 660.6 751.4 906.2 1076.3 Rural 439.8 .. 561.2 615.7 640.9 767.5 Total 519.5 .. 628.8 702.7 797.7 991.6 Mean monthly wage, wage earners (imputed wages) Urban 447.7 .. 485.0 608.6 711.1 798.4 Rural 365.5 .. 409.8 537.3 466.9 593.4 Total 418.0 .. 461.0 583.0 611.1 742.2 Mean monthly wage, unemployed (imputed wages) Urban 365.8 .. 338.0 477.2 513.9 512.6 Rural 255.4 .. 228.4 388.9 262.1 302.0 Total 334.6 .. 300.9 448.1 408.2 437.2 Mean monthly wage, unemployed(reservation wages) Urban 803.6 .. 730.4 833.0 872.0 925.3 Rural 428.7 .. 568.4 407.7 464.3 598.0 Total 697.6 .. 675.5 693.0 700.8 808.1 Mean monthly wage, wage earners (actual wages) Urban 74.3 .. 162.9 74.4 109.9 168.5 Rural 78.2 .. 147.6 86.9 100.3 169.7 Total 76.3 .. 157.6 79.1 103.7 168.9 Mean monthly wage, wage earners (imputed wages) Urban - .. 212.8 - - 262.2 Rural - .. 232.1 - 167.3 262.5 Total - .. 219.3 - 167.3 262.4 Mean monthly wage, unemployed (imputed wages) Urban 125.5 .. 209.7 - 176.0 267.4 Rural 114.2 .. 196.8 - 165.5 237.5 Total 119.0 .. 203.9 - 166.5 246.7 Mean monthly wage, unemployed(reservation wages) Urban 108.0 .. 190.0 100.0 123.6 191.0 Rural 78.7 .. 196.2 71.7 141.8 178.8 Total 89.7 .. 193.1 75.8 139.9 184.0 Source: Authors' estimates using GLSS data. 67 Table 3: Headcount index, wage gap, and squared wage gap measures, Ghana 1991-2006 P0 P1 P2 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 Youth (18-30) Imputed wage ­ unemployed Urban 0.018 .. 0.541 0.001 .. 0.159 0.000 .. 0.063 Rural 0.062 .. 0.861 0.008 .. 0.291 0.001 .. 0.125 Total 0.030 .. 0.650 0.003 .. 0.204 0.000 .. 0.084 Reservation wage ­ unemployed Urban 0.018 .. 0.123 0.003 .. 0.044 0.001 .. 0.020 Rural 0.077 .. 0.237 0.031 .. 0.080 0.019 .. 0.040 Total 0.035 .. 0.161 0.011 .. 0.056 0.006 .. 0.027 Actual wage ­ wage earners Urban 0.069 .. 0.309 0.030 .. 0.139 0.019 .. 0.090 Rural 0.123 .. 0.348 0.050 .. 0.175 0.027 .. 0.118 Total 0.088 .. 0.321 0.037 .. 0.151 0.022 .. 0.099 Older workers (25-64) Imputed wage ­ unemployed Urban 0.000 0.011 0.146 0.000 0.000 0.015 0.000 0.000 0.002 Rural 0.000 0.129 0.594 0.000 0.011 0.119 0.000 0.001 0.028 Total 0.000 0.060 0.306 0.000 0.005 0.052 0.000 0.001 0.011 Reservation wage ­ unemployed Urban 0.006 0.019 0.127 0.001 0.006 0.045 0.000 0.003 0.024 Rural 0.075 0.224 0.307 0.034 0.049 0.122 0.022 0.016 0.066 Total 0.029 0.105 0.191 0.012 0.024 0.073 0.007 0.009 0.039 Actual wage ­ wage earners Urban 0.051 0.057 0.152 0.022 0.022 0.066 0.013 0.013 0.039 Rural 0.054 0.147 0.231 0.019 0.066 0.099 0.008 0.038 0.058 Total 0.053 0.094 0.174 0.021 0.040 0.075 0.011 0.023 0.044 Source: Authors' estimates using GLSS data. 68 Table 4: Headcount index, wage gap, and squared wage gap measures for current workers according to formal versus informal employment, Ghana 1991-2006 P0 P1 P2 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 Youth (18-30) Actual wage ­ wage earners ­ Formal Urban 0.042 .. 0.242 0.012 .. 0.103 0.065 .. 0.032 Rural 0.095 .. 0.272 0.034 .. 0.134 0.095 .. 0.031 Total 0.060 .. 0.250 0.019 .. 0.111 0.073 .. 0.032 Actual wage ­ wage earners ­ Informal Urban 0.152 .. 0.438 0.087 .. 0.209 0.140 .. 0.059 Rural 0.188 .. 0.426 0.086 .. 0.218 0.142 .. 0.138 Total 0.167 .. 0.433 0.087 .. 0.213 0.141 .. 0.095 Older workers (25-64) Actual wage ­ wage earners ­ Formal Urban 0.026 0.136 0.094 0.011 0.059 0.039 0.022 0.006 0.037 Rural 0.040 0.348 0.165 0.014 0.136 0.064 0.036 0.007 0.076 Total 0.031 0.249 0.111 0.012 0.100 0.045 0.026 0.006 0.058 Actual wage ­ wage earners ­ Informal Urban 0.198 0.044 0.345 0.089 0.017 0.155 0.092 0.054 0.009 Rural 0.149 0.089 0.351 0.049 0.046 0.164 0.098 0.019 0.028 Total 0.181 0.062 0.347 0.076 0.028 0.158 0.094 0.042 0.016 Source: Authors' estimates using GLSS data. 69 Table 5: Wage gap ratio, Ghana 1991-2006 18-30 years old 25-64 years old 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 Imputed wage ­ unemployed Urban 0.047 .. 0.294 - 0.030 0.100 Rural 0.133 .. 0.337 - 0.087 0.200 Total 0.096 .. 0.313 - 0.082 0.169 Reservation wage ­ unemployed Urban 0.180 .. 0.360 0.241 0.318 0.357 Rural 0.402 .. 0.339 0.455 0.218 0.398 Total 0.314 .. 0.350 0.425 0.228 0.380 Actual wage ­ employed Urban 0.436 .. 0.451 0.435 0.394 0.433 Rural 0.406 .. 0.503 0.340 0.447 0.429 Total 0.421 .. 0.469 0.399 0.428 0.431 Actual wage ­ employed ­ Formal Urban 0.275 .. 0.426 0.413 0.430 0.413 Rural 0.360 .. 0.491 0.348 0.392 0.385 Total 0.321 .. 0.445 0.383 0.401 0.403 Actual wage ­ employed - Informal Urban 0.573 .. 0.478 0.452 0.376 0.450 Rural 0.460 .. 0.511 0.326 0.509 0.467 Total 0.521 .. 0.491 0.417 0.450 0.456 Source: Authors' estimates using GLSS data. 70 Appendix to Annex 5 Estimating Imputed Wages for Ghana's Unemployed This appendix explains our methodology for estimating the imputed wages of the unemployed in Ghana. The standard procedures for doing this consists generally in estimating a treatment model, with a first regression modeling the decision to participate in the labor market and earn a wage, and a second regression for the determinants of the logarithm of wages. In a country like Ghana with segmented labor markets, it makes sense to use a multinomial logit in the first stage, in order to consider various groups of workers, such as for example adult individuals not working but willing to be employed as well as individuals working but not being paid for their work, wage earners (which are those on whom we focus here), self-employed workers with earnings from agriculture, and self-employed workers with outside of agriculture. Next, in the second stage, a wage regression can then be estimated for one or more of the groups for whom we observe earnings. The second stage regression for the determinants of wages in each of the employment categories with earnings must then take into account sample selection from the multinomial logit, for example using the Durbin-McFadden (1984) procedure (see also Bourguignon, Foumier, and Gurgand, 2004). There is a debate in the literature as to whether the two steps procedure outlined above is necessarily the best way to proceed. Puhani (2000) gives a good survey of the literature on the appropriateness of the different sample-selection correction procedures. Amongst the different causes leading to a potentially harmful correction procedure, the problem of collinearity between the error terms of the multinomial/probit equation and the wage equation is a serious one. This collinearity problem mainly occurs if the former equation is poorly identified. In that case Puhani (2000) argues that a Heckman-type procedure can often do more harm than good and recommends using OLS or an adapted version of standard regressions instead. In this paper, we use the wage regressions only for estimating the imputed wages of the unemployed, on the basis that these unemployed workers could be potential wage earners, and thereby could be affected by the minimum wage. We tried three different approaches for estimating the wage regressions: a first stage multinomial logit followed by the wage regression with sample selectivity, a first stage probit followed by the wage regression with sample selectivity, and no first stage regression at all before running a linear regression on the logarithm of wages. We compared the distribution of predicted wages obtained under the three alternatives for workers with wage earnings to the actual distribution of wages among those workers. By far, the estimation without correction for self-selection performed better. In addition, the estimation of wages for the unemployed with the simple linear regression seemed much more reasonable than the estimation with either of the two procedures to take sample selection into account (that is, with the two stage procedures, the imputed wages for the unemployed were very high compared to the wages of the working population, which was deemed unreasonable). Thus, because of the better predictive power of the simple linear regression without treatment effect, we used this procedure here for estimating the imputed wages for the unemployed, even though we would have preferred to use a treatment effect regression in theory. The coefficient estimates for the final regression implemented on wage workers and used to estimate the imputed wages for the unemployed are given in table A1 (standard errors take into account cluster effects). 71 Table A1: Linear regressions used for imputed wages for the unemployed 18-30 years old 25-64 years old 1991/92 1998/99 2005/06 1991/92 1998/99 2005/06 urban 0.142 .. 0.010 -0.037 0.224*** 0.148*** (1.39) (0.12) (0.73) (3.97) (3.05) Some Primary 0.314 .. 0.063 0.025 0.123 -0.082 (1.26) (0.36) (0.21) (0.80) (0.71) Primary 0.295 .. 0.302 0.217* 0.193 0.126 (0.95) (1.61) (1.84) (1.52) (1.23) Secondary (lower) 0.809*** .. 0.633*** 0.394*** 0.502*** 0.332*** (2.83) (3.64) (4.78) (5.19) (4.11) TVET 1.082*** .. 1.004*** 0.616*** 0.621*** 0.597*** (3.09) (4.53) (4.43) (3.94) (5.95) Secondary (higher) 1.003*** .. 1.015*** 0.734*** 0.923*** 0.652*** (3.25) (5.43) (7.37) (8.38) (7.01) Post Secondary 1.555*** .. 2.015*** 0.938*** 1.044*** 1.354*** (4.48) (9.65) (9.25) (9.96) (15.77) Apprenticeship -0.089 .. -0.049 -0.199*** -0.106* -0.060 (0.86) (0.61) (4.04) (1.87) (1.31) Experience 0.018 .. 0.198*** 0.045*** 0.035*** 0.027*** (0.40) (6.74) (4.90) (3.43) (3.22) Experience squared 0.001 .. -0.005*** -0.001*** 0.000** 0.000** (0.75) (4.56) (3.60) (2.30) (2.03) Married 0.202** .. 0.078 0.261*** 0.263*** 0.104** (2.02) (1.02) (4.67) (4.06) (2.24) Male 0.022 .. 0.514*** 0.161*** 0.233*** 0.382*** (0.21) (6.73) (3.06) (4.02) (8.36) Accra 0.392** .. 0.376*** 0.469*** 0.463*** 0.332*** (2.41) (3.19) (5.92) (5.20) (5.03) Forest 0.303** .. 0.102 0.176** 0.194** 0.164*** (2.14) (0.95) (2.55) (2.32) (2.64) Coastal 0.310** .. 0.175 0.193*** 0.009 0.156** (2.00) (1.42) (2.62) (0.10) (2.22) Constant 4.268*** .. 3.175*** 4.676*** 4.383*** 4.809*** (10.98) (12.26) (29.93) (24.74) (35.32) N 294 .. 736 971 1029 1899 R2 0.177 .. 0.281 0.228 0.268 0.279 Source: Authors' estimates using GLSS data. References Abowd, J. M., F. Kramarz, T. Lemieux, and D. M. Margolis, 2000, Minimum Wages and Youth Employment in France and the United States, in Blanchflower, D. G., and R. B. Freeman, eds., Youth employment and joblessness in advanced countries, NBER Comparative Labor Market Series, Chicago and London: University of Chicago Press. Addison, John and Blackburn, McKinley (1999), Minimum Wages and Poverty, Industrial and Labor Relations Review, Vol. 53, n2: 393-409. Angel-Urdinola, D., and Q. Wodon, 2004, The Impact on Inequality of Raising the Minimum Wage: Gap- Narrowing and Reranking Effects, Labor 18: 317-27. 72 Angel-Urdinola, D., and Q. Wodon, 2005, Do Changes in the Minimum Wage Affect Younger and Older Workers Differently? Evidence for Paraguay Using Semi-Parametric Methods, Estudios de Economía, 32(1): 25-38. Bell, L., 1997, The Impact of Minimum Wages in Mexico and Colombia, Journal of Labor Economics 15: S102-35. Brown, C., C. Gilroy, and A. Kohen, 1982, The Effect of the Minimum Wage on Employment and Unemployment, Journal of Economic Theory, 20: 487-528. Burkhauser, R., and Finegan A., 1989, The Minimum Wage and the Poor: The End of a Relationship. Journal of Political Analysis and Management, 8(1): 53-71. Card, D., and Krueger, A., 1995, Myth and Measurement: The New Economics of the Minimum Wage. Princeton University Press. Princeton, N.J. Coudouel, A., J. Hentschel, and Q. Wodon, 2002, Poverty Measurement and Analysis, in J. Klugman, editor, A Sourcebook for Poverty Reduction Strategies, Volume 1: Core Techniques and Cross- Cutting Issues, World Bank, Washington, DC. Draca, M., S. Machin, and J. Van Reenen, 2006, Minimum Wages and Firm Profitability, CEPR Discussion Paper No. 5456, London School of Economics. Foster, J. E., J. Greer, E. Thorbecke, 1984, A Class of Decomposable Poverty Indices, Econometrica, 52: 761-766. Harrison, A. and J. Scorse, 2004, Moving Up or Moving Out? Anti-Sweatshop Activists and Labor Market Outcomes, NBER Working Paper No. 10492, National Bureau of Economic Research. Jones, P., 1998, The impact of minimum wage legislation in developing countries where coverage is incomplete, Centre for the Study of African Economies Working Paper Series, Paper No. 66, Oxford. Mensah, J. V., M. Tribe, and J. Weiss, 2007, The Small-Scale Manufacturing Sector in Ghana: A Source of Dynamism or of Subsistence Income?, Journal of International Development, 19(2): 253-73. Metcalf, D., 2006, Why Has the British National Minimum Wage Had Little or No Impact on Employment?, Centre for Economic Performance, LSE, CEP Discussion Papers. Mincy R., 1991, Raising the Minimum Wage: Effects on Family Poverty, Monthly Labor Review, 113: 18-25. O'Neill, Donal; Nolan, Brian; Williams, James, 2006, Evaluating the Introduction of a National Minimum Wage: Evidence from a New Survey of Firms in Ireland, Labor, 20(1): 63-90. Pereira, S. C., 2003, The Impact of Minimum Wages on Youth Employment in Portugal, European Economic Review 47: 229-44. Saget, C., 2001, Poverty reduction and decent work in developing countries: Do minimum wages help?, International Labor Review, 140(3): 237-269. Yuen, T., 2003, The Effect of Minimum Wages on Youth Employment in Canada: A Panel Study, Journal of Human Resources 38: 647-72. 73 ANNEX 6 INTERNATIONAL MIGRATION FROM GHANA: PATTERNS, BRAIN DRAIN AND POLICY IMPLICATIONS 16 This background paper sheds light into the observed patterns of international migration from Ghana, and their implications for labor markets and education policies using various sources of data including a special migration module in the fifth Ghana Living Standards Survey (GLSS5) implemented in 2005-06. Very often information on migrants are absent from household surveys since the migrants themselves are absent from the household at the time of the survey. The migration module in GLSS5 contains individual level information on the migrants if their family is still present in Ghana. The survey provides data on their demographic and education profiles, labor market participation patterns and can be a rich source of understanding regarding the dynamics of international migration from Ghana. This paper also presents a quantitative estimate of the loss to the Ghanaian economy represented by migration abroad in terms of the cost of educating the migrants before they migrate to other countries. 1. Introduction 6.1 Migration operates as an important equilibrating mechanism by allocating labor to regions and sectors where it is in most in demand. As such, migration, whether internal and international, can generate significant welfare gains for the migrants, their families and both the destination and the source economies. Since the majority of the gains accrue to the migrants through higher wages, and to their families through the remittances sent back, migration may become an important tool for economic development and poverty reduction. 6.2 However, international migration data in countries such as Ghana reveal that a significant share of the migrants, especially to the OECD countries, has tertiary education. Given the importance of human capital in economic development and growth, brain drain, especially in sub-Saharan Africa, is likely to have important development and labor market implications. The negative spillovers created by the brain drain to the rest of the economy might undermine the private benefits enjoyed by the migrants and their families. 6.3 Migration of both unskilled and skilled workers to international destinations also generates important challenges for the migrants and both the host and home countries. For example, the absence of highly skilled migrants might lead to relatively high prices for key services or products (such as in engineering and finance) or even shortages in the provision of certain public services (such as health and education). This might generate serious bottlenecks in the development process and have long term negative effects. Furthermore, if the educational demands in the labor market of the host country are different than the norms in the native country, highly educated migrants might end up in occupations below their level of education. This would on the one hand mean waste of scarce educational resources of the home country and on the other negative social effects on the migrants themselves. 6.4 In many cases, appropriately designed domestic labor, financial market and education policies can enhance benefits while decreasing the costs of migration (for a discussion of some of these programs, see Angel-Urdinola et al., 2008). There are numerous distributional issues related to migration that may need to be addressed. For example, education of many highly skilled people is 16This background paper was prepared by George Joseph, Yoko Niimi, Caglar Ozden, and Quentin Wodon. 74 publicly financed but the benefits from migration are privately enjoyed. Also, loss of educated labor force due to migration also means a fall in potential tax revenues to the government. In short, migration issues are complex and need to be an integral part of development agenda and labor market analysis in every developing country. 6.5 The objective of this paper is to shed some light into the observed patterns of international migration from Ghana, and their implications for labor markets and education policies using various sources of data including a special migration module in the fifth Ghana Living Standards Survey (GLSS5) implemented in 2005-06. Very often information on migrants are absent from household surveys since the migrants themselves are absent from the household at the time of the survey. The migration module in GLSS5 contains individual level information on the migrants like their demographic and education profiles and labor market participation patterns. It provides a rich source of understanding regarding the dynamics and characteristics of international migration from Ghana. This paper also presents a quantitative estimate of the loss to the Ghanaian economy represented by migration abroad in terms of the cost of educating the migrants before they migrate to other countries. 6.6 The paper is structured as follows. In section 2, we present data on international migration abroad using the Docquier & Marfouk (2006) dataset. This data set is especially useful to compare external migration in Ghana to that observed in other African countries. Next, in section 3, we use the data from a special module of the last Ghana Living Standards Survey providing a more detailed profile of international migrants. In section 4, we discuss the issue of brain waste, and in section 5, we estimate the loss in education expenditure or investments associated with international migration. A brief conclusion follows. 2. Aggregate External Migration Profile 6.7 According to the Docquier & Marfouk (2006) dataset on bilateral measures of migration, the number of international emigrants to developed OECD countries from Ghana has more than doubled over the 1990s. Based on the population censuses of destination OECD countries, about 162,000 people born in Ghana were found to be in the labor force in one of these countries in 2000,17 as compared to about 80,000 in 199018. Ghana is not the only country in Africa exhibiting such a trend. As shown in Figure 1, apart from Benin and Niger, the number of migrants in the OECD labor force has significantly increased in several other countries included in the figure, although for Africa as a whole, the increase in external migration to OECD countries has been smaller than it has been in Ghana. Among all of the West African countries included in Figure 1, Nigeria is the only country that has a higher number of total migrants than Ghana, but this is a reflection of the relative sizes of the two countries. In terms of share of migrants abroad to the home population,19 Ghana has the highest proportion of migrants both in 1990 and 2000 in the West African countries listed in Figure 1: this ratio is 2.2% for Ghana, compared to the average ratio of 0.9% for all Sub-Saharan African countries, and 0.6% for Nigeria. 17Note that the Docquier & Marfouk dataset is based on the censuses of the destination OECD countries, not on the information from sending countries. It excludes children and other migrants who are not in the labor force. For more detailed information on the database, see Docquier and Marfouk (2006). 18Preliminary evidence indicates that an even larger number of Ghanaians migrate within Africa. However, the intra-African migration data is not reliable and this issue will be addressed in more detail later using the GLSS data. 19It was drawn only for those who are in the labor force either in one of the OECD countries or in origin countries. 75 Figure 1: Share of Migrants Abroad Relative to Home Population 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Africa Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone 1990 2000 Source: Docquier & Marfouk (2006). 6.8 The Global Migrant Origin Database20 highlights that a significant share of migrants from the African countries migrates to another sub-Saharan countries. In the case of Ghana, two-thirds of migrants are in one of the sub-Saharan countries. The most popular destinations for Ghanaian migrants are Cote d'Ivoire (32%), Nigeria (13.1%), Burkina Faso (9.8%) and Guinea (8.7%), which are all Ghana's neighboring countries. The United States, the most popular destination among OECD countries, comes in the fifth place (7.3%) followed by the United Kingdom (6.0%). This clearly reflects the ease of migration to the neighboring countries (Figure 2). Since there are no detailed data on the migrants to other African countries in the Global Migrant Origin Database, this issue will be further addressed using the GLSS5 data in section 3. Figure 2: Major Destinations (OECD and Non-OECD countries) 100% 80% 60% 40% 20% 0% Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone North America EU Rest of OECD Sub-Saharan Africa Source: Parsons et al. (2007). Note: North America includes the United States, Canada, Australia and New Zealand, while EU includes its main 15 member countries. 20Global Migrant Origin Database constructed by the Development Research Centre on Migration, Globalization and Poverty at the University of Sussex, contains the international bilateral migration stock data that were collected for 226 by 226 countries and dependent territories. The detailed description of the migration database can be found in Parsons et al. (2007). 76 6.9 A key feature of Ghanaian migration experience is the emigration of the highly skilled and educated labor force to the OECD countries. For those who move out of the region, the European Union countries represent the main destination for migrants without college degree: in 2000 about 60% of Ghanaian migrants without college degree moved to the EU countries, about 30% to North America and less than 10% to other OECD countries. The share of Ghanaian migrants working in North America has increased from 13% in 1990 to about 28% in 2000. Particularly the United States seems to be a more popular destination among the educated migrants: in 2000 as many as 56% of educated Ghanaian international migrants were in North America (Figure 3). This seems to reflect the country specific immigration policy of destination countries and also the fact that educated migrants are more capable of meeting higher costs of migrating to countries that are further away. The share of migrants to native labor force is significantly higher among the college educated ones. As high as about 47% of at least college educated people who were born in Ghana are working abroad as of 2000 (Figure 4). This certainly sheds light on the growing issue of the migration of highly-skilled and educated people, the so-called brain drain, which discussed in more detail in the subsequent sections. Figure 3: Major Destinations of College Educated Migrants 1990 100% 80% 60% 40% 20% 0% Africa Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone North America EU Rest of OECD 2000 100% 80% 60% 40% 20% 0% Africa Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone North America EU Rest of OECD Source: Docquier & Marfouk (2006). Note: North America includes the United States, Canada, Australia and New Zealand, while EU includes its main 15 member countries. 77 Figure 4: Share of College Educated Migrants Abroad to College Educated Home Population 60% 50% 40% 30% 20% 10% 0% Africa Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone 1990 2000 Source: Docquier & Marfouk (2006). 6.10 It is evident that Ghana, as well as her neighboring countries, face one of the most debated issues related to migration, the so-called brain drain. The rest of this section focuses on the education profile of the migrants. According to available information, among the migrants from Ghana, about 75% of them have either secondary or tertiary education (about 44% of them have at least a college degree). Among the neighboring countries, the migrants from Nigeria and Sierra Leone, in particular, are found to be drawn from the upper end of the education distribution. Figure 5: Share of Migrants by Education Attainment: 2000 100% 80% 60% 40% 20% 0% Africa Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone secondary tertiary Source: Docquier & Marfouk (2006). 78 Figure 6: Share of Labor Force by Education Attainment: 2000 30% 25% 20% 15% 10% 5% 0% Africa Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone secondary tertiary Source: Docquier & Marfouk (2006). 6.11 The portion of people with secondary education in the labor force is not small in Ghana when compared with the ratios for Africa or the neighboring countries. However, the share of college educated people in the labor force is only 1.1% in Ghana, whereas the same number for Africa is 2.8%. This is not surprising if we take into account the relatively low rate of enrollment for tertiary education in Ghana (3%), compared to Cote d'Ivoire (5.9%), Togo (3.8%), and Nigeria (10.5%). The relative scarcity of human capital in Ghana is illustrated in Figure 7. Although there was a slight increase in the enrollment rate for tertiary education in Ghana and some other neighboring countries between 1991 and 2001, it was only 3% in Ghana in 2001. Given the relatively high proportion of college educated people among the migrants abroad, the issue of brain drain is probably more alarming for Ghana than for other countries. 6.12 It should be pointed out that if these migrants had migrated as children and obtained their college degree in their destination country, it is debatable whether they would constitute the brain drain of their home countries since there is no guarantee that they would have received the same level of education had they not migrated. Even if they had stayed home and received a college degree, it could be argued that they would have just replaced somebody else and would not have caused the overall level of education to increase. In order to examine this issue, we will look at the age profile of the migrants. However, according to the 2000 US Census data, more than 75% of Ghanaian migrants were at the age of 22 or above when they entered the United States (Figure 8). This implies that Ghanaian migrants in the United States are most likely to have obtained their college degree in Ghana. Despite the variation, this seems to apply to the migrants from all the neighboring countries, confirming the importance of the brain drain issue for the region. 79 Figure 7: School Enrollment Rate, Tertiary Panel A. 1991 4% 3% 2% 1% 0% Ghana Togo Benin Burkina Faso Niger Sierra Leone Guinea Panel B. 2001 11% 8% 6% 4% 2% 0% Ghana Cote Togo Benin Burkina Nigeria Niger Sierra Guinea d'Ivoire Faso Leone Source: WDI (2000). Note: For Panel A, there was not data for Cote d'Ivoire and Nigeria. For Panel B, the graphs are based on the data for the year 2001 except for Cote d'Ivoire (1999), Nigeria (2003), Niger (2003) and Guinea (2003). 80 Figure 8: Age Profile of Migrants in the US on Arrival 70% 60% 50% 40% 30% 20% 10% 0% Ghana Cote Togo Benin Nigeria Niger Sierra Leone Guinea d'Ivoire 0-5 6-12 13-17 18-21 22-30 Source: US 2000 Census. 6.13 Even though a relatively small portion of migrants came to the United States at an age lower than 22, it would be interesting to observe how likely it is for those young migrants obtain a college degree. Figure 9 presents the share of migrants by education attainment. Panel A shows the figure for those migrants who entered the US when they were 22 years or older and Panel B reports it for the migrants who were younger than 22 at time of migration, but currently are 22 or older. In other words, Panel A illustrates the education distribution of the migrants who are likely to have completed their education in their home countries, whereas Panel B shows that of the migrants who completed in the United States. We drew both figures for those who arrived in the 1980s and also for those arrived in the 1990s. Note that given the limited sample size for some of the countries, we will only look at Ghana, Nigeria and Sierra Leone from hereafter as well as the overall figures for all African countries. 6.14 Panel A shows the relatively high education level of migrants from the region for migrants who came to the United States at the age of 22 or older. However, there is a decline in the share of college educated migrants in the 1990s when compared with the 1980s arrivals. This may reflect, among other things, some changes in the US immigration policy and/or some reduction in the migration costs. In the case of Ghana, about 45% of migrants who entered the US in the 1980s had at least a college degree and only 23% had the degree among the 1990s arrivals. Moreover, Ghanaian migrants have a relatively lower education level, particularly those who came to the United States in the 1990s, when compared with the overall figure for Africa. Turning to the education distribution of those who are currently 22 years or older but entered the United States when they were younger than 22, although the share of migrants who completed their high school education is higher when compared with Panel A, we find the share of college educated migrants is actually lower among the migrants who came to the country before the age of 22. In addition, we again observe a decline in the portion of college educated migrants among the migrants who arrived in the 1990s when compared with those who arrived in the 1980s. 81 Figure 9: Share of Migrants by Education Attainment Panel A. Migrants arrived in the US at age of 22 or more 100% 80% 60% 40% 20% 0% 1980s 1990s 1980s 1990s 1980s 1990s 1980s 1990s Africa Ghana Nigeria Sierra Leone College+ High school Panel B. Migrants arrived in the US at age of less than 22 but currently 22 or more 100% 80% 60% 40% 20% 0% 1980s 1990s 1980s 1990s 1980s 1990s 1980s 1990s Africa Ghana Nigeria Sierra Leone College+ High school Source: US 2000 Census. 3. External Migration Profile based on GLSS5 data 6.15 The Docquier-Marfouk dataset provides valuable aggregate statistics and historical patterns. It is especially useful in comparing migration patterns and profiles from neighboring countries that have similar historical and cultural backgrounds. However, aggregate data fail to provide important information on the dynamics and local patterns of migration. Fortunately, the GLSS5 includes a detailed migration module designed by the World Bank Development Research Group and provides much more detailed information on the migrants, their families and the overall impact of migration on Ghana. 82 6.16 According to preliminary data from GLSS5, a total of 422,436 individuals or 1.91% of the population is comprised of international migrants. This is a lower estimate than that reported in section 2, where it was suggested that 2.2% of Ghana's population has migrated to OECD countries. The reason for this lower estimate is probably due to the fact that with the GLSS5, we identify only migrants who emigrated alone or in small groups, leaving their families back at home. When a whole family migrates, there is nobody who can be interviewed back in Ghana to provide information on whom migrated. This is especially likely to be true for high-skilled migrants who are able to migrate with their entire families. Still, the order of magnitude of the two estimates is comparable. Figure 10: Destination of migrants in GLSS5 data Individual Migrant Destinations 5% 23% 27% 16% 29% US UK Rest of OECD Africa Other Source: Authors, using GLSS5 data 6.17 As shown in Figure 10 and table 1 from GLSS5, the most favored destinations for external migrants are the US, the UK and other OECD countries followed by other African countries. The most popular destinations within Africa are Cote d'Ivoire, Nigeria, Burkina Faso and Guinea, which are all neighboring countries. In table 1, data are provided on the geographical location of the families of origin of external migrants and the economic status of their households, as measured by the quintile of consumption per equivalent adult to which they belong. A total of more than 288,000 households have an international migrant, which represents 5.3% of the total number of households. The main destinations of migrants are the US (19%), the UK (16%), other OECD countries (32%) and African countries (26%). The vast majority of households with migrants to the OECD countries (including the US and the UK) are from urban areas and belong to the fourth and fifth consumption quintiles. On the other hand, families with migrants to African countries tend to be poorer and rural. This suggests selectivity in migration decisions to the OECD which favors the wealthier, urbane and more educated sections of the population. In terms of regional distribution, migrants to OECD countries are predominantly from Ashanti and Greater Accra whereas there is significant number of migrants to other African countries in the Volta, Eastern and Brong Ahafo regions. 83 ds e hole erh 9 900 207 39 0 0 747 23 0 0 uso with stnarg mi 0,2 936,8 2,1 ,21 875,1 511,2 752,8 486,7 094,1 271,1 283,1 1,1 490,4 1, H elsew ds to hole icar 0 7 360 449 911 30 493 370 885 5 Af 42 742 0 71 uso with tsnar ig e 4,7 6,2 7,4 ,56 820,9 0,2 3,1 4,2 174,3 988,3 318,6 9, 675,9 9,2 823,9 1, 38 H m th ds to fot hole esr CD 262 535 727 16 963 253 625 0 613 324 7 0 0 uso with tsnar ig e OE 2,9 0,6 1,3 ,82 056,9 0,1 4,1 4,5 380,5 641,3 689,1 677,1 0,6 9,1 38 H m th ds to K tnargi 6 hole U 862 454 408 0 290 581 0 462 e 02 0 0 uso with tsnar ig th 4,4 6,2 8,1 880,6 049,5 0,1 2,2 375,1 597,2 088,5 942,4 6,2 769,2 1, H m m alnre ds xte hole with stnarg US 074 280 0 0 472 0 609 3 0 0 e uso eht mi 6,5 7,4 947,8 147,2 5,1 887,73 083,1 047,3 224,9 612,3 2,3 861,5 58 to ont H 84 asel fo ds at e hti ar hole aln % % % % 0% 48% us with ter stnarg 56 48 15 65 w Sh ex 0%3.5 5%2.7 3%8.3 .51 4%2.3 7%1.4 3%3.5 %04.8 1%9.1 6%7.2 6%8.2 2. 5%5.2 6.1 3%1.8 0. 1. 0. mi sd ho olhesuoh for ds hole aln 0 5 0 456, 409, 047, 362 660 301 934 157 42 218 173, 917 3 26 30 mbeu us with ter stnarg ex mi 882 691 191 585,01 6,2 225,24 1,6 626,741 1,1 4,1 5,2 9, 0,2 611 0,4 2, 2, 77 of N ho scitsiret ofr ds tal be hole 435, 492, 942, 179, 835, 791, 301, 016, 614 061, 744, 062, 819 808 624 ac To us arh num ho 454,5 363,2 091,3 976,307 148 870,210,1 571,1 420,857,1 8, 2, 9, 8, 905 415 798 36 947 779 035 47 19 11 C 1: el ab sdlohes nso .at gie da 5SS T L ou Hfor tilesniu Rlac G ngi Q n tileniu elitniu tile tile Q inu elitniu hipa of Q inu ark ha tsa tse us, mbeu tiop Qt d Q A nr G g Er Wr hors ut Nlato nabr laru musno rsiF onceS irdh T htruoF Q ogre fthiF nrets n-igir lartne Aretaer at ol nretsa itnahs onr hetro pep pep A:e We C G V E A B N U U T U R C O ourcS 6.18 The data are also informative with respect to the characteristics of international migrants themselves. As mentioned earlier, many international migrants, especially the highly skilled ones who migrated to the OECD countries, manage to bring their families with them. As a result, their immediate families will be absent from the underlying sample frame used in the GLSS. This would indicate that the data from the GLSS might underestimate international migrants, especially the ones in the OECD, when compared to the Docquier-Marfouk database which is based on the censuses of the destination countries. First, it is interesting to see that the order of magnitude of external migration as observed in the GLSS5 is still similar to that reported in the Docquier-Marfouk data set despite the problems mentioned above due to differences in the underlying samples. According to this data set, there were approximately 162,000 Ghanaian workers in OECD in 2000, and this had represented a doubling versus the number of workers in 1990. If it was assume that, the same rate of increase in the number of Ghanaian workers abroad has persisted during the five years between 2000 and 2005/2006, then roughly 250,000 Ghanaians would be working in OECD countries at the time of the survey. The estimate from the GLSS survey (see the row in Table 2 on the employment status identifying individuals who are working) is 279,247 for the US, the UK, and the other OECD countries combined 6.19 Individual level data also reveal important patterns in terms of human capital and labor market characteristics of international migrants. A majority of migrants are males, reflecting perhaps psychological costs to migration due to separation from families and possibility of increased remittances to families at home Migrants to OECD (including the US and the UK) countries come predominantly from urban areas when compared to migrants to other African countries, as was already observed in table 1 based on data at the household level. 6.20 In terms of the age distribution, 71% of migrants are between the age of 25 and 44, pointing to a reduction from the productive workforce within the country. Also it should be noted that this age group would have normally completed most of their education in Ghana about which we discuss later. The main reason to migrate is to work. Marriage is the second main reason, followed by family reunion (joining other relatives) and education. 6.21 As noted earlier, the migration of educated workers can have negative effects on their home countries as discussed earlier. Among these are (i) the loss of potential positive externalities generated by the educated workers ­ whether they are doctors, engineers or entrepreneurs; (ii) the loss of potential tax revenue on the incomes of these people; and (iii) the loss of public funds spent on their education. The negative effect of the brain drain has also been much argued in the growth literature (e.g. Miyagiwa, 1991; Haque and Kim, 1995). Most developing countries already suffer from the scarcity and low level of human capital, which has been increasingly identified as one of the key determinants of economic growth and development. An important underlying implication of this is that a "brain drain" may leave a developing country in a poverty trap. 6.22 Table 3 provides data on the education background of migrants. On average, migrants constitute a better educated and more skilled section of the Ghanaian labor force, and education affects the place of destination. Some 14 percent of OECD migrants have tertiary education and another 75 percent have secondary education. This is rather different than the averages for migrants to other African countries where 57 percent have secondary education and only 1 percent has post- secondary education. 85 er stnarg 4 8 441 234 207 0 115 739 0 0 077 36 0 Oth mi 2,2 0,1 2,1 123,2 301,7 516,7 483,5 9,1 263,3 319 283,1 7,1 424,2 9,1 3, 87.3 .231 05.8 6 /05002 stn cair 7 1 3 3 2 4 rag 597, 158 439 073 855 157 425 Af 08 572 025 035 840 07 75 314, 28 0 Mi to 131 0,3 3,8 6,1 2,4 4,3 9,1 1, 2,7 1,4 5,2 222,1 0,7 4, 766,8 3, 101 3, 91.6 .911 .481 na ha G stn t D res C 1 7 6 rag eht E 519, 868 651 457 892 750 279 14 393 126 890 O ent 181 5,7 2,4 0,1 6,3 9,4 0,2 1, 5,9 3,2 0,1 610,4 234 0 8,9 172,1 11 573, 94 0 4, 2, 82.4 98.7 67.9 my Mi to of 151 plo stn UK 7 0 8 em rag eht 475 239 236 320 598 152 93 601 874 226 112 5 9 575 90 0 nda 9,6 6,4 3,2 694,3 8,1 0,3 5,1 1, 9,3 9,2 2,1 343,7 6,4 82 110,2 96 7,6 1, 25.7 40.7 .061 Mi to hicsp stn US 7 4 4 rago rag eht 404 509 894 8,9 8,7 9,1 339,2 154,44 101,72 898 31 282 121 2,2 1, 4,7 4,2 374,9 471,1 593 0 9,7 220,6 20 099 30 0 2, 6,9 2, 64.5 12.8 6 .801 Mi to m de yb, fo in n e aln ar ter tsnar tio % tal nts to 37 ra Sh ex ig m laupop 8%8.1 5%8.2 9%2.1 7%2.0 %78.4 %12.6 0%2.3 0. 3%7.2 6%0.1 1%1.4 ig 86 mlan n tio 45 61 58 73 25 iot laup anah su aln G pl ter stnarg 7 9 3 98 275, 491, Po in ex mi 2,2 554,8 036, 153, 4,1 3,1 784,760,3 023,504,2 0 0 6 751, 776, 8, 006,2 48 022, 469, 1, 1,1 1,1 449,9 terna info n s tio ict laup anah 1 2 8 55 185, 03 36 G 069, eris Po in 2,2 142,8 855, 118, 3,1 3,1 631,819,2 260,652,2 9 7 3 740, 09 76 78 390, 3, 175,2 48 721, 347, 1, 0,1 1,1 365,9 ctar r hac aln 2 6 nda mbeu of ter stnarg 436, 009, 427, 245 N ex mi 224 412 811 5,3 153,941 752,941 101,38 48 963, 473, 519 092 518, 89 368 042 638, 5, 003 211 8,5 5,1 123 4, 0,2 1,1 084 ador ber .at m ab no sr da Nu:2 43 44 06 43 44 06 5SS L G e bla n dna dna dna atirgi s 06 sevitale kro w own eayfo s dna dna dna ngi T tiou 52 52 53 54 Rr orf K erb 52 53 54 us, n n ib hat onitac Statut t en gn n'o mu hors istr ween ween ween easor kro nteraP heto n my ngikro N ut ween ween ween lato nabr laru D sse nahtretaer mrof el egairra du okio e L Bet Bet Bet G ela M E W oiJ nioJ reht D/o ega A:e O W L N T U R Ag M maeF lop ain er Bet Bet Bet M Em Av ourcS er stnarg 441 234 207 0 0 308 900 0 00 00 98 00 00 00. Oth mi 2,2 0,1 2,1 9,1 748 283,1 319 0,2 0. 0. 7. 7. 2. 12 stn cair rag 597, 158 439 114 363 897 0 591 35. 89. 97 14 00 00 Af Mi to 131 0,3 3,8 6,3 1,1 6,5 715,6 300,1 4,6 595,6 10 12 8. 5. 5. 0. stn rag eht fots CD 519, 868 651 876 550 049, 00 90 60 18 08 48 to re OE Mi 181 5,7 2,4 358,1 216,5 7,7 6,1 979,6 533,7 091 282,1 0. 8. 6. 5. 9. 9. stn rag eht 475 239 236 0 0 412 217 631 to UK 9,6 6,4 3,2 6,3 2,1 5,1 152,5 647 4,6 856,3 00 53 39 10 68. 0. 9. 6. 8. 17 Mi 60 stn rag eht 404 509 894 US 005/2 to 8,9 8,7 9,1 048,1 556,3 430 197 419 6,4 1,2 6,1 557,6 539 0,9 600,6 00 89 44 52 87 79 4. 4. 8. 7. 5. 5. Mi ana fo Gh,noitacude e aln ar ter stnarg n l taot tio Sh ex mi in laupop 8%8.1 5%8.2 9%2.1 3%5.0 1%8.0 0%8.5 0%7.6 8%5.8 0%0.3 n yb,st tio laup anah su aln G pl ter stnarg 7 9 87 98 275, 45 046, 081, 540, 189, 812, 695, 491, 036, Po in ex mi an 2,2 554,8 4,1 534,7 375,2 840,4 578 824 746 gri n mlanoitanretnifo tio laup anah 1 2 6 55 185, 03 518, 43 684, 321, 1, 780, 407, 440, G 069, Po in 2,2 142,8 855, 3,1 134,7 21 5, 165,2 478,3 997 414 546 r aln mbeu of ter stnarg 436, 009, 427, 753 640 922, 409 405 254 725, 532 scitsiretcarahc N ex mi 224 412 811 9,3 0,2 362 7,5 1,4 0,2 493 7,1 ador n .at io ab da atr sr dna ig 5SS L reb yra yra M yr G yr da yra ngi mu yra da otroi anah eayfo yra erb us, Pr G mu yra N:3 onces n hors io /e mi yr T n N / elba lato nabr laru catud on N pr-erP ndocesro ndocesro oncest E io an yra ndocesro T ut mairP E niuJ nieS osP V T catud ha edistu ega neo G O er N mirp-erP mirP niuJ orineS ndocestsoP V A:e T T T U R E E Av ourcS 6.23 Another important indicator of the loss of educated workforce is the share of educated workers at home and abroad. As shown in Figure 11, only 22 percent and 2.15 percent of the total population have secondary, and post secondary education within Ghana while around 70 percent of migrants have secondary education and 10 percent have post secondary education. This suggests further that the present migration trends in Ghana are leading to brain drain and have the potential of further aggravating the problem to low human capital at home. 6.24 We have already noted that a significant proportion of migrants from Ghana belong to the age group between 25 and 44. As shown in Table 3, a vast majority of international migrants (82.8 percent) have completed their education in Ghana. This means that a significant proportion of the expenditure on education in Ghana has gone to those who later migrated to other countries. We will come back to this question in section 4. Figure 11: Selected Characteristics of external migrants, 2005-06 Migrant Destination by Level of Education Education Levels of Migrants to OECD 100% 1% 80% 7% 3% 14% 60% 40% 20% 18% 0% 57% one ary N Prim Junior secondary Senior secondary condary ET TV Post se None Primary Junior secondary Senior secondary Post secondary TVET OECD Africa Other Share of Educated at Home and Abroad Education Levels of Migrants to Africa 60% 1% 0% 50% 40% 6% 32% 30% 20% 10% 51% 0% 10% one ary N Prim None Primary Junior secondary Junior secondary Senior secondary Post secondary ET TV Senior secondary Post secondary TVET National Migrants Africa Other OECD Source: Authors, using GLSS5 data. 88 4. Brain Waste 6.25 What happens to migrants where they migrate? The brain waste argument is based on the educational attainment at home and the labor market performance of migrants in the host countries. If the migrants fail to find jobs in the host country commensurate with their educational qualification and labor market status at home country, we can safely say that there is a brain waste due to migration. From a global perspective, brain waste further adds to the losses due to brain drain and may lead to a reduction in global welfare. How immigrants perform in the labor market of destination countries has been one of the fundamental questions in the migration literature (Borjas, 1994). Following Mattoo, Neagu and Ozden (forthcoming), we measure the labor market performance in terms of occupational outcomes. Given the limited availability of data, we again rely on the US 2000 Census data for our discussion first, and then use the GLSS5. 6.26 Figure 12 presents the labor market performance of the college educated migrants in the United States. Again while Panel A shows the figure for those migrants who entered the United States when they were 22 years or older, Panel B reports for the migrants who were younger than 22 when they migrated to the United States but are currently 22 years or older. As far as Panel A is concerned, as in the education distribution, the figure illustrates the worse performance in the US labor market among those migrants who arrived in the 1990s when compared with the performance of the migrants who arrived in the United States in the 1980s, especially for the migrants from Ghana and Sierra Leone,. As a result, for those migrants who arrived in the 1990s, Ghanaian college educated migrants perform worse in the US labor market relatively to the performance of the migrants from the other countries in Africa. For instance, about 41% of Ghanaian college educated migrants have highly-skilled occupations in the United States, whereas about 49% of African college educated migrants have such jobs. The observed variation, though relatively small, may be partly due to the underlying quality of education in origin countries and partly due to the selection effect, i.e., variation in the abilities of migrants because they are drawn from different sections of the skill distribution of their home countries (Mattoo, Neagu and Ozden, 2005). In order to examine some of these issues, we have drawn the same figure for the college educated migrants who came to the United States under the age of 22, but who are currently 22 years or older (Panel B). 6.27 The comparison between Panels A and B illustrates that the migrants from Ghana seem to perform better if they had completed their degree in the United States. The same story applies to most of the cases. Moreover, the variation in the performance of the migrants in the US labor market becomes smaller when they obtained their college degree in the United States. 89 Figure 12: Labor Market Performance of College Educated Migrants Panel A. Migrants arrived in the US at age of 22 or more 100% 80% 60% 40% 20% 0% 1980s 1990s 1980s 1990s 1980s 1990s 1980s 1990s Africa Ghana Nigeria Sierra Leone High skilled Medium skilled Panel B. Migrants arrived in the US at age of less than 22 but currently 22 or more 100% 80% 60% 40% 20% 0% 1980s 1990s 1980s 1990s 1980s 1990s 1980s 1990s Africa Ghana Nigeria Sierra Leone High skilled Medium skilled Source: US 2000 Census. 6.28 Using the GLSS5, we can also briefly examine the labor market performance of migrants to the OECD and African countries. Some 10.40 percent of OECD migrants were employed as professionals and another 15.76 percent were crafts and related workers prior to migration. In addition, around 13.46 percent were full time students. This implies that a large portion of migrants left Ghana right after completing their education which is another indication of the extent of brain drain. Some 33.19 percent of migrants to other African countries were agricultural workers prior to migration, 10.15 percent were looking for work and 13.46 percent were students. 90 6.29 While 10.4 percent of the migrants were employed as professionals at home, only 3.68 percent could gain a similar occupation in the OECD countries upon migration. The same is true for technicians and associate professionals (4.02 percent to 1.15 percent) and craft and related workers (15.76 percent to 3.46 percent). However, significant numbers of migrants to the OECD are reported as engaged in "other activity" (over 69.51 percent) which means that either their household members in Ghana did not know what the migrant was exactly doing or perhaps the migrant was in an unskilled category. This would confirm the "brain waste" hypothesis where the migrants are employed in areas below their education level. 6.30 These observed variations though relatively small, may be partly due to the underlying quality of education in origin countries and partly due to the selection effect, i.e., variation in the abilities of migrants because they are drawn from different sections of the skill distribution of their home countries (Mattoo, Neagu and Ozden, 2008). These observations pose some important questions for the education policy of the source countries as well as for the immigration policy of destination countries. For example, what skills should the potential migrants obtain to perform well in the labor market of destination countries? Is it an economically efficient policy to provide free tertiary education if the migrant is likely to end up in an unskilled job in the destination? How the countries can maximize the link between the educated migrants abroad and the population at home, i.e., the positive externalities of the migration of educated people? Figure 13: Occupation before and after migration Occupation before and After Migration to OECD 80% 70% 60% 50% 40% 30% 20% 10% 0% rs ls s n ge s rs mana nals rkers rators sio Clerk yworker ations catio ope r work cup edu ficials and ofes professiona oc gfo in Pr rviceworkefisher hine time Other activity Se relatedwo ll look and mac fu of Senior nicians, associate lturaland and ementary El icu Craft ant Pl ch edagr Te ill Sk Prior to Migration After Migration 91 Occupation before and After Migration to Africa 35% 30% 25% 20% 15% 10% 5% 0% and managers sionals sionals erks rkers rkers rkers s tion rk Cl erators cation ofes ery wo op cupa edu ing for wo her activity officials Pr sociateprofes Servicewond fish and related wod machine time Ot full look an Senior Technicians, as ltural a aft Elementary oc Cr ricu ant Pl Skilled ag Prior to Migration After Migration Source: Authors, using GLSS5 data. 5. Losses in Education Expenditure due to External Migration 6.31 We have already outlined the profile of migrants form Ghana, and their preferred destinations. We found that the more educated prefer to migrate to OECD countries despite the strong possibility of not finding an occupation commensurate with their educational qualifications. Also we have noted that the vast majority of migrants are between the age of 25 and 44 and most are educated in Ghana. In this short section, we would like to estimate the actual loss to the economy due to migration from the perspective of educational expenditures on migrants relative to the population at home. 6.32 We now propose to combine household survey data with detailed information on international migrants and their education level with administrative data on public spending for education in order to compute the share of total education investment made in a country that is lost due to migration. The migration module of GLSS5 provides information on the highest level of education achieved by each migrant in Ghana before they migrated. We can calculate the cumulative cost of education for each migrant by summing over the educational cost of each level of education the person have obtained. 6.33 Table 4 shows total educational expenditure or rather investments for Ghana at 2004 unit costs. The unit costs are derived from data available in the Preliminary Education Sector Performance Report of the Ministry of Education and Sports (2005) in Ghana. Due to lack of data in trends in unit costs over long periods of time, we estimate the investment made in education for all individuals, even older ones, using the 2004 unit costs. The total cumulative expenditure on 92 education for the country as a whole is the sum of the education expenditure for all Ghanaians who have remained in their country as well as the expenditure for those who have migrated (whose numbers are provided in table 4), and it comes up to 49,116 billion cedis (US$ 5 billion). 6.34 Migration module of GLSS5 reports information on the highest level of education achieved by each migrant before migration. We can thus calculate the cumulative cost of education for each migrant by summing over the cost of each level of education the person has obtained. Among the population between the age of 25 and 60, a total of 381,709 individuals or 4.96 percent of the population are international migrants. However, as expected, the rate of migration is higher for individuals who are better educated. Between 10 and 13 percent of those with a secondary of post- secondary education migrate. Note that among those with at least some education, a vast majority of international migrants have completed their education in Ghana; for simplicity, for those that did not complete all of their prior education in Ghana, we do not know which part of their education was completed in Ghana, and which part was completed abroad, so that for simplicity we assume that all their education was completed in Ghana. We may thus slightly overestimate brain drain for that reason, but at the same time, our estimate of brain drain can be considered as a lower bound on the loss of education expenditure due to migration to the extent that we use unit costs from public education, which is typically cheaper than private education. 6.35 In table 5, we compute the total investment in education allocated to individuals who have migrated, which comes up to 4,312 billion cedis. In other words, about 8.07 percent of the total investment in education is lost to migration. Some of this may be recuperated if some migrants come back in their country, but the evidence from the survey is that the number of return migrants is very small. 93 Table 4: National Education Expenditure by level of education Total Cumulative Number Actual unit cost Education Levels Number Number of years (Cedis Total cost Pre-school 11,037 5,228,591 1 469,523 2,454,943,888,444.16 P1 58,653 5,217,555 1 698,077 3,642,254,955,348.44 P2 131,479 5,158,902 1 698,077 3,601,310,889,394.39 P3 194,538 5,027,423 1 698,077 3,509,528,144,280.59 P4 144,136 4,832,885 1 698,077 3,373,725,780,469.99 P5 218,026 4,688,749 1 698,077 3,273,107,893,613.39 P6 353,526 4,470,724 1 698,077 3,120,909,306,649.89 JSS1 97,082 1,982,324 1 1,043,523 2,068,601,177,907.81 JSS2 93,829 1,885,242 1 1,043,523 1,967,293,846,716.12 JSS3 674,698 1,791,414 1 1,043,523 1,869,381,377,594.64 M1 166,771 147,325 1 698,077 102,844,054,409.60 M2 103,044 97,252 1 698,077 67,889,447,230.93 M3 91,873 97,604 1 698,077 68,135,163,354.16 M4 1,550,075 1,462,936 1 698,077 1,021,241,974,072.00 SS1 41,093 1,116,715 1 2,612,625 2,917,558,780,935.00 SS2 27,783 1,075,622 1 2,612,625 2,810,197,580,906.25 SSS3 213,550 587,146 1 2,612,625 1,533,993,598,436.25 S1 9,695 169,786 1 1,043,523 177,176,020,791.86 S2 16,032 160,092 1 1,043,523 167,059,546,370.96 S3 9,689 144,060 1 2,612,625 376,374,229,749.75 S4 9,726 134,371 1 2,612,625 351,060,323,241.00 S5 124,645 124,645 1 2,612,625 325,649,598,075.00 L6 9,947 9,947 1 2,612,625 25,987,456,909.50 U6 43,377 43,377 1 1,043,523 45,264,771,948.24 Teacher Training Col. 98,588 98,588 2 8,552,318 1,686,311,340,828.92 Nursing 24,499 24,499 2 2,935,256 143,819,266,578.08 Poly technical 103,916 103,916 2 2,935,256 610,037,776,787.20 University 98,109 98,109 4 10,421,595 4,089,820,727,606.40 Other tertiary 48,486 48,486 2 10,421,595 1,010,593,530,904.50 TVET 460,693 460,693 2 2,935,256 2,704,503,197,764.80 Total investment 49,116,575,647,319.80 Source: Authors, using GLSS5 data. 94 nte mts 68. 81 15 00 9. 97 06 25. 74 76. 56 06. 9. teda veni stna 955, 13 7. 9. 549. 29 57 111, 2. 69 927, 8. 13 808, 22 3, tim es noita 0, % gr 625, 6,84 7, 2, 8, 62 07 05 579, 6,13 96 mi 5, 86 669, 390, 4, 9, 07 8. tal To ucde ni 237, 82 66 23 6, 044, 0,81 875, 070, 895, 372, 766, 2, oft 204,1 00 7,3 7, 429, 1, 15 732 2,8 182 0,7 450,1 31 4, 6002, 3,5 osc ntsargi 625 mlan tsoc nosr 3 3 5 077, 52 562, 00 2612 52 5. 183, 562, 59 it tioa per )side 3, 3, 421, Un (C 986 04 pe 1, 359,2 62 dna 04 1, 6122 235, 525,8 359,2 0,1 intern nai 430,1 na ha G drad r be sra n of of io 2+ fo s ni ye of catu cleyc 6 3 2 sra SSJ 3 sra 2 2 4 SSS Stan num of ed erp ye ye SSJ 3 3 eary ent mt 95 e es nvi ativlu for 199, 116 208 943 463 797 987 084 n mu mbeu stnarg 1, mi 393 103,6 32 0,5 4,3 7,1 2,1 1,1 5,2 tioa C N r educlat tal 818 To mbeu of stnarg 588 208 463 987 084 tofo N mi 7,1 2, 103,6 16 0,5 943,43 7,1 797,21 1,1 5,2 s )4 tea m tis elbat E:5 dna 3 e bla ET) elbat tsocla .at T e da egll TV(l m ott t of 5SS Co G elsv gni naoiss onitac osctna er L has ngi us, Le ainr gri sat n T io yr l l s er feorPlaci due m (frotsoclatot osc hors ut catu T E vee vee deta deta nta A:e Ed mairP SSJ V L L hnc each yraitre mits mit gri T O SSS A T Te T E Es M ourcS 6. Conclusion 6.36 This paper is intended to shed some light into the observed patterns of international migration from Ghana and to provide an estimate of the loss in educational expenditures due to migration. We found, similar to the previous works on Ghana, that there is an overwhelming preference among the more educated towards migrating to OECD countries, providing further evidence to the brain drain from Ghana. In terms of the labor market performance, the dominant hypothesis of brain waste is validated in the case of Ghanaian migrants to OECD countries where they find it hard to find occupations commensurate with their educational qualifications. Though international migrants comprise of only 1.91 percent of the total population, the loss in education expenditure due to migration can be estimated to be 8.07 percent of total expenditure on education on the current population. In the light of these findings, it is important to design appropriate policies for the retention of educated human capital in Ghana. References Angel-Urdinola, D., T. Takeno, and Q. Wodon, 2008 (forthcoming), Student Migration to the United States and Brain Circulation: Issues, Empirical Results, and Programmes in Latin America, in A. Solimano, editor, The International Mobility of Talent: Types, Causes, and Development Impact, Oxford University Press, Oxford and New York. Borjas, G. (1994) "The Economics of Immigration", Journal of Economic Literature, Vol.32, No.4, pp.1667-1717. Docquier, F. and A. Marfouk (2006) "International Migration by Education Attainment, 1990-2000", in Ozden, C. and M. Schiff (eds) International Migration, Remittances and the Brain Drain, Washington, D.C.: World Bank and Palgrave Macmillan. Haque, N.U. and S.J. Kim (1995) "Human Capital Flight: Impact of Migration on Income and Growth", IMF Staff Papers, Vol.42, No.3, pp.577-607. Mattoo, A., I. C. Neagu and C. Ozden (forthcoming) "Brain Waste? Educated Migrants in the US Labor Market", Journal of Development Economics (older version available as World Bank Policy Research Working Paper, No.3581, Washington, D.C.: World Bank). Miyagiwa, K. (1991) "Scale Economies in Education and the Brain Drain Problem", International Economic Review, Vol.32, No.2, pp.743-759. Parsons, C. R., R. Skeldon, T. L. Walmsley, and L. A. Winters (2007) "Quantifying International Migration: A Database of Bilateral Migrant Stocks", in Ozden, C. and M. Schiff (eds) International Migration, Economic Development and Policy, Washington, D.C.: World Bank and Palgrave Macmillan. World Bank (2005) Global Economic Prospects: Economic Implications of Remittances and Migration, Washington, D.C.: World Bank. 96 ANNEX 7 GHANA'S NATIONAL YOUTH EMPLOYMENT PROGRAM AND POVERTY REDUCTION 21 To deal with youth unemployment, Ghana introduced in October 2006 the National Youth Employment Program (NYEP) which aims to employ some 500,000 young people between 2006 and 2009. The wages paid by the NYEP appear to be high in comparison to market wages which makes the program costly. In addition, several components of the program target youth with a junior secondary education completed, which excludes many among the poor. One may therefore wonder whether the high budgetary cost of the program is justified from the point of view of poverty reduction. To provide an assessment of the potential impact of the NYEP on poverty, and compare it to a traditional rural public works program, we use Ghana's 2005/06 Living Standards Survey and simulation techniques. We identify the individuals who might be interested in participating in rural public works as well as the NYEP and consider two parameters that affect the impact of a program on the poor: the targeting performance of the program and the substitution effect of the program, whereby only part of the wages paid to potential beneficiaries generate additional income since some beneficiaries would have done other work if they had not participated in the program. The results suggest that while the substitution effect may not be too large, targeting performance is likely to be limited in the case of the NYEP as compared to a public works program. For poverty reduction, rural public works are likely to be four or five times more cost-effective than the NYEP. 1. Introduction 7.1 Youth unemployment and underemployment is a major issue in sub-Saharan Africa as in many other areas of the developing world (World Bank, 2007a). In many African countries, children and youth represent up to 40 percent of the population. Thanks to programs such as the Education for All initiative, school enrolment rates are rapidly increasing, but many youth remain out of school, and are often without work or with work that do not build their skills22. 7.2 In Ghana, as discussed among others in World Bank (2008) and MOESS (2006), the net primary school enrolment rate increased from 59 percent in 2004/05 to 79 percent in 2006/07. The admission rates for 12 year old students in the first year of junior secondary education, which accounts for the first three years of secondary education, have also increased substantially, from 12 percent in 2004/05 to 44 percent in 2006/07. Yet enrolment in senior secondary schools (the last three years of secondary education), while also increasing, still reached only 11 percent of the appropriate age group in 2005/06. This means that today, despite progress in expanding access to education, nine out of every ten Ghanaian 21This background paper was prepared by: Harold Coulombe, Moukim Temourov, and Quentin Wodon, January 2008 22On skills training in Africa and Ghana, see Adams (2007), Blunch (2006), Glewwe (1999), Haan and Serriere (2002), Johanson and Adams (2004), Monk, Sandefur, and Teal (2007), Palmer (2007), and Rosholm, Nielsen and Dabalen (2007). 97 youths between 15 and 17 years of age are not enrolled in senior secondary schools. Some of these children may be enrolled in lower grades (due to repetition in past grades), and others are working. But most remain unemployed or under-employed, and many lack the skills that would facilitate their transition to the labor market and lead to a productive life with decent earnings enabling them not to live in poverty. 7.3 According to the 2005/06 Ghana Living Standard Survey (GLSS5), the unemployment rate among youth ages 15-24 was about twice as high as the national unemployment rate (6 percent compared to 3 percent for the working population as a whole), but in addition, many youth appear to be underemployed, and many also declare working (hence are not considered unemployed) although they do not get any pay because they are trapped in subsistence activities. There is also a perception that although national poverty measures have been steadily decreasing in the country as a whole for the last 15 years, poverty in urban areas is increasing, especially in the capital area of Greater Accra, so that initiatives to help youth find jobs could contribute to increasing urban standards of living. This is however more a perception than a fact, as careful analysis of the available data suggests that over time, poverty in urban areas as a whole has decreased, even though there may have been a limited increase in the Greater Accra area (Coulombe and Wodon, 2008). 7.4 In order to deal with youth unemployment, various strategies can be used. One strategy consists in providing the right skills set to youth, so that they are better prepared to join the labor force. For this, traditional apprenticeships can prove to be a cost-effective alternative for especially lower skilled youth. Throughout West Africa, and especially in Ghana, traditional apprenticeships between a master craftsman and apprentice are a popular source of skills. The main strengths of traditional apprenticeship are its practical orientation, self-regulation, and self-financing. These apprenticeships also cater to individuals who lack the educational requirements for formal training, but at the same time, evidence from Ghana from Adams et al. (2008) suggest that the impact from apprenticeships on occupational choice and wages may be limited. 7.5 Another approach used in Ghana consists in providing combined employment and training opportunities. Recently, Ghana introduced in October 2006 a new program, the National Youth Employment Program. This is a large program which aims to employ some 500,000 young people between 2006 and 2009. Ghana has a number of other employment- related programs, such as the Special Presidential Initiatives, the Rural Enterprise Development Program, the National Board of Small Scale Enterprises, and other small programs run by a number of sectoral ministries and agencies. But the new National Youth Employment Program (NYEP) is by far the largest. Apart from providing temporary employment, the NYEP aims to train youths in various trades and occupations (Ministry of Manpower, Youth and Employment, 2006). 7.6 The launch of the NYEP may appear to be a sound idea in order to help youth find employment and improve their skills. Indeed, according to lessons from a Youth Employment Inventory of 289 programs and interventions from 84 countries recently carried out by the World Bank (2007), public works and training program are more suitable than formal sector wage subsidy programs for youth in developing countries, since wage subsidies do not go far in developing countries due to the small size of the formal wage sector and also do not reach the poor. Public works and training programs are also more likely to succeed than targeted youth entrepreneurship schemes. This is because while these schemes may 98 improve opportunities for young entrepreneurs in low-income countries where job growth in the formal economy tends to be rare, the evidence indicates that not all youth will be well suited for self-employment and that failures rates for young entrepreneurs can be high. 7.7 However, careful targeting and screening for these programs is important to success and cost-effectiveness, and it may well be that training programs are substantially more expensive than public works program, especially if the training programs target relatively better educated workers and pay a high wage for the period of training. Training programs are also more successful when they involve the private sector in providing practical work experience and in identifying the kind of skills required. Engagement of the private sector in training is an effective tool to mitigate the risk of high-cost training disconnected from market demand and to increase on-the-job training. This is something that is also attempted in the NYEP, but it is unclear whether it is actually working. 7.8 While the NYEP has a number of attractive characteristics at least in principle and on paper, its actual implementation may not yet be an example of best practice, as will be discussed in the next section. Moreover, and more importantly for the purpose of the present paper, while the program focuses primarily on job creation, rather than on reducing poverty, it is nevertheless useful to assess what might be the likely impact on poverty of the program. Indeed, in a country such as Ghana where public resources remain scarce, difficult trade-offs must be arbitrated to ensure that public spending is allocated in order to improve the well- being of the population. 7.9 The NYEP appears to be an expensive program, because wages paid are high. The program also targets to some extent urban areas, probably because that's in part where the issue of youth unemployment is most visible, and that's also where it is easier to provide meaningful training to youths participating in the program thanks to the network of firms and non-profit organizations that can employ youths there. To the extent that poverty is much more severe in rural areas, one may wonder whether the high budgetary cost of the program is justified from the point of view of the objectives set forth in Ghana's growth and poverty reduction strategy. 7.10 Assessing the impact of the NYEP on poverty is a complicated matter, because the program is supposed not only to provide temporary employment, but also to build skills which may lead to a stream of higher future incomes for participants. Furthermore, the NYEP started to be implemented right after the last nationally representative survey with data on income and consumption was carried out in 2005-2006, so that it is not possible to assess the impact of the program on poverty by looking using impact evaluation techniques such as matching procedures. At the same time, given that Ghana implements national consumption and income surveys only once every seven years on average, we cannot wait for the next survey to begin to try to assess what the potential impact of the program might be. 7.11 To provide a preliminary assessment of the potential impact of the NYEP on poverty, we rely in this paper on simulation techniques rather than on impact evaluation techniques. The approach is very simple. We assess who may be potentially interested in participating in the NYEP by identifying working individuals without pay, as well as for every level of proposed wage in the NYEP, those individuals who work but now earn less than the NYEP wage, since all these individuals may indeed be interested in participating in the program to increase their earnings. We also consider as potential beneficiaries the unemployed whose 99 reservation wage is below the proposed NYEP wage. Next, we randomly select among the pool of potential beneficiaries of the program a number of participants so as to match the distribution of NYEP beneficiaries by region that is available from administrative records of the program. Finally, we estimate for the assumed participants to the program two key parameters which affect the potential impact of the program on the poor: the targeting performance of the program, and the substitution effect of the program, whereby only part of the wages paid to beneficiaries generate additional income, because at least some of the beneficiaries would probably have done other work if they had not participated in the program. 7.12 In conducting this assessment, we also compare the potential impact of the NYEP to that of a more standard rural public works program where wages are likely to be substantially lower (on the impact of public works on poverty in developing countries, see among others Ravallion, 1999). While for the NYEP the findings suggest that targeting performance is likely to be limited, because the program targets youth with junior secondary education completed (as opposed to enabling all potentially interested youth to participate), targeting performance is better for rural public works program, so that the overall impact of the public works on poverty is also likely to be four or five times larger than that of the NYEP for the same cost. 7.13 The paper is structured as follows. In Section 2, we describe briefly the NYEP. In Section 3, , using recent household survey data we provide data on the potential demand for programs such as the NYEP and rural public works by looking at the number of youths who are either not working but willing to work (the unemployed), or are working but with a level of pay that is below what the program provides. In section 4, we simulate the potential impact on poverty of the two types of programs through the payment of wages to the participating youths (we deliberately do not consider the additional impact which may come from the training component of the program since we do not have data to estimate its impact). The simulations take into account the likely targeting performance of the two types of programs, as well as the likely substitution effects. A conclusion follows, 2. Description of Ghana's NYEP 7.14 The NYEP addresses job creation for the youth, defined as young people between the ages of 18 to 35. Launched in October 2006, the program aims to "empower the youth to be able to contribute more productively towards the socio-economic and sustainable development of the nation" according to the Youth Employment Implementation Guidelines (Government of Ghana, 2006). The NYEP program is built on the experience of the Skills Training and Employment Placement Program (STEP), which focused mainly on vocational training, including apprenticeship for graduates of junior and senior secondary schools, agricultural training for rural areas, and the teaching of entrepreneurship skills to college graduates. According to the available information from the Ministry of Manpower, Youth and Employment, between 2002 and 2004, a total 18,000 beneficiaries (9,384 men and 8,928 women) were trained by the STEP program at a total cost over three years of about 19 million cedis (about US$190,000). 7.15 The target number of 500,000 jobs to be created comes from a national youth employment survey/registry carried out prior to the program. This survey identified and registered approximately 175,000 young people willing to work, with only about 50 percent 100 of them employed at the time of the survey. The survey also revealed large regional disparities in youth unemployment, with the highest unemployment rates in large urban areas, particularly Ashanti and Greater Accra regions. At the same time, while youth unemployment may be higher in these areas, these are also the richest areas in the country, which suggests that the impact on poverty of the program may not be large, an issue to which we will come back below. Table 1: The NYEP Youth Employment Registry Data No. of youth Actual No. of youth Share of youth employed Region registered employed (%) Ashanti 24,322 7,537 31.0 Brong Ahafo 19,868 7,932 39.9 Central 13,016 7,697 59.1 Eastern 19,100 8,600 45.0 Greater Accra 22,363 7,922 35.4 Northern 21,959 16,528 75.3 Upper East 13,271 9,530 71.8 Upper West 12,590 9,688 76.9 Volta 18,094 8,674 47.9 Western 10,087 7,967 79.0 Total 174,670 92,075 52.7 Source: Ministry of Manpower, Youth and Employment 2005. 7.16 The NYEP is a broad-based program, involving a number of national ministries and agencies, district assemblies, community-level groups, as well as NGOs and the private sector. The youth employment program targets a wide range of activities in different sectors, such as education, health, water and sanitation, agriculture, and others, and given its national coverage, the program operates in all 10 regions of the country. To reach its objectives, the program interacts on regular basis with a number of governmental structures at the national and regional levels and it also contracts out some of its activities to NGOs and the private sector. 7.17 Most of the beneficiaries are employed to provide basic social services in the public sector. Since October 2006, the NYEP has provided employment to 92,075 young people, with about 42,000 of them working as teaching assistants and health and sanitation workers. Agrobusiness (16,383) is another important employment module that promotes farm and non- farm income-generating activities in rural areas. The internships module (5,041) targets mainly the educated youth in urban areas seeking employment with the private and public sectors. The employment modules for trades and vocation and ICT are still being developed. Overall the largest share of NYEP employment is in the public sector, with teaching and nurse assistants financed by the NYEP filling in for absent teachers and medical staff, especially in remote areas of the country. While the program is providing important and necessarily social services, the youth hired by the program often lack proper training and do not have all necessary qualifications to carry out their tasks. 7.18 The program is costly, with an annual budget several times larger than that of the Ministry of labor itself. Originally the Government of Ghana has planned an earmarked annual allocation of 1,300 million cedis (US$120 million) to finance NYEP activities. Based on the available data, the NYEP budget allocations for 2007 are estimated at 677,000 million 101 cedis, which corresponds to about five times the total budget of the Ministry of Manpower, Youth, and Employment (estimated at 103,000 million cedis for 2007). Since the program's launch in October 2006, the government has spent about 445,000 millions cedis (US$42 million) to benefit 92,075 young people, or about US$450 per beneficiary. Table 2: The NYEP Beneficiary Data, 2006-07 Employment modules Beneficiaries Community education teaching assistants 23,021 Agro-business 16,383 Health extension workers 14,000 Internship 5,200 Waste and sanitation 5,041 Community protection 1,300 Trades and vocation - ICT - Others 26,760 Program staff 370 Total 92,075 Source: Ministry of Manpower, Youth and Employment 7.19 Various sources of financing are used to finance the program. The program is financed from four main sources: (i) specialized funds and national programs, such as Poverty Alleviation Fund, HIPC, Road Fund, Ghana Education Trust (GET) Fund, National Health Insurance System (NHIS), Women Development Fund, Food and Agriculture Budget Support Funds; (ii) cost-sharing schemes and collaborative funding by district assemblies common funds (DACF), government agencies, civil society organizations, etc.; (iii) funds recovered from the program participants; and (iv) other state sources. An earmarked amount from each specialized fund is used annually to financing various NYEP employment modules. However, the budget execution is well below the planned allocation: in 2006, the NYEP received 270 billion cedis (US$25 million) out of 552 billions cedis planned for the program; and in the first half of 2007, about 22 percent of the budget allocated to the program was actually executed. 7.20 The main concern is the long-term sustainability of the NYEP interventions and their impact. The program focuses mainly on short-term goals and tasks that project beneficiaries would not be able to carry out independently of program subsidies. Since most of the NYEP interventions are creating temporary jobs in the public sector, this approach may create unrealistic expectations about future employment and earning prospects among the youth and could affect long-term employability of the beneficiaries. 102 Table 3: NYEP Annual Expenditures, 2006-2007 Annual budget (million cedis) Source of financing 2006 2007 Plan Actual Plan Actual* DACF 180,600 121,918 244,005 72,450 HIPC 113,850 67,116 80,000 GET FUND 138,690 0 177,520 56,838 NHIS 151,366 61,000 182,390 44,809 ROAD FUND 106,900 20,000 111,380 0 Other 144,767 552,716 270,178 795,345 174,097 Source: Ministry of Manpower, Youth and Employment, as of May 22, 2007. (*) Estimates for first half of 2007. 7.21 International experience also suggests that better longer-term results could be achieved if employment programs are combined with other education-related interventions. An inventory of youth employment interventions carried out by the World Bank provides some insights on how other countries deal with the issue of youth unemployment. Among the most important lessons learned is the fact that youth interventions should involve the private sector from the beginning to ensure an appropriate match between the skills provided by youth programs and employers' needs. 7.22 Finally, the NYEP's monitoring and evaluation system is weak and there is no information on the project's impact on youth employment. Given the high complexity of the program and its huge costs, it is crucial to put in place a strong monitoring and evaluation system to monitor results on the ground and ensure efficient use of public resources. 3. Potential Demand for employment programs 7.23 Using the 2005/06 Ghana Living Standards Survey, we provide in this section estimates of the number of youths aged 18 to 35 who could be interested by a national youth employment program such as the NYEP or a public works program. We distinguish the NYEP from a public works program in two ways. First, we assume that the NYEP targets youth who have at least a junior secondary education completed, while the public works program is open to all. Second, we assume the NYEP pays much higher wages than a public works program would. While we have no detailed data on the wages paid under the NYEP, anecdotal evidence suggests that these wages are high. Those with a Master degree may receive on a monthly basis up to 2 million cedis, while wages are of the order of 1.5 million for a Bachelor degree, one million for the Higher National Diploma offered at the polytechnic level, 8,000 cedis for Senior Secondary School graduates and 500,000 cedis for Junior Secondary School graduates. The majority of program participants are likely to be Higher National Diploma and Senior Secondary School graduates23. These wages are well above the level of the minimum wage which was at the time of the GLSS5 survey at 13,500 cedis per day. 23We are grateful to Daniel Kwabena Boakye for this information on wage levels. 103 7.24 By contrast, it is likely that a public works program would pay substantially lower wages, probably even well below the minimum wage, given that the minimum wage itself is high in comparison to the earnings of youth (Coulombe and Wodon, 2008). Specifically, for the simulations for the NYEP, we will consider wages ranging from 500,000 cedis to 2 million cedis per month, while for the public works simulations we will consider wages ranging from 50,000 to 200,000 cedis per month. These values for the wages to be paid are indicative only and somewhat arbitrary. Yet we are using enough different values to be able to assess the targeting performance and potential impact on poverty of both types of programs, we can use any combination of wages for public works and the NYEP to compare the two programs, as well as test for the sensitivity of the comparisons to the choice of wages to be paid. 7.25 Tables 4 and 5 provide data on the distribution of earnings of individuals who are already working, as well as on the distribution of the reservation wage declared by individuals who are unemployed and looking for work. The groups of individuals are presented in the first column of the table in terms of their monthly wages in thousand cedis. The difference between the two tables resides in the sample that is considered. In table 4, all youth are included, which corresponds to our simulations for the potential impact of public works. In table 5, only youth who have completed their junior secondary education are included, which corresponds to our minimum wage simulations. Thus table 4 enables us assess the potential population that could be interested in a job in a public works program without eligibility condition in terms of education, while in the other table, an eligibility condition is imposed. In the actual NYEP program documentation, it is not entirely clear whether a strict education eligibility condition is imposed, but due to the training component of the program and the types of jobs proposed to participants, it seems that the program targets youth with at least junior secondary education completed as opposed to youth with lower levels of education. 7.26 Consider first the statistics provided in table 4. We see for example that there is a very large group of youth who are working but are not paid (41.8 percent of the youths who are working at the national level). These individuals are likely to be interested in public works. Clearly, some may not apply for such a program due to various constraints (they may not be paid, but still doing important work that has to be done for their household, and hence they may not be able to participate in the program). Also, depending on the wage paid by public works, additional individuals could be interested in participating in the program if their current wage is below that proposed by the program. We cannot identify those who would actually be interested and those who would not. But for the purpose of the simulations in the next section, all the individuals unpaid for their work, as well as all individuals who earn less than the proposed wage are potential beneficiaries of the program, and we can randomly chose some of these individuals as participants in public works for each proposed wage level in order to simulate the impact of the program on poverty. Finally, among the unemployed, those who have a reservation wage below the proposed wage would also be potential beneficiaries. 7.27 The estimates in table 4 therefore give us an upper bound for the potential number of youths that might be interested in a public works program, depending on the wage provided in the program, and without any eligibility condition as proxied by the education level of the individual. Figures 1 to 3 summarize the data on the potential number of participants by quintiles of per capita consumption of the households to whom the individuals who are 104 potential beneficiaries belong. This is done for four potential wage levels, from 150,000 cedis per month to 300,000 cedis per month, which is about the level of the minimum wage at the time of the survey. Two findings stand out. First, the number of individuals potentially interested in the program appears to be very large, especially because many youth are working without pay and might thereby be interested in getting cash income through public works. Second, the targeting performance or likely benefit incidence of the program depends fundamentally on whether the program is implemented mostly in urban or rural areas. Clearly, in urban areas, the program would probably be regressive, since most of the potential beneficiaries belong to the better off quintiles of the population (this is because urban households tend to have much higher levels of consumption than rural households). By contrast, the programs could be well targeted to individuals belonging to households who tend to be poor if the focus is placed on providing employment in rural areas. 105 Table 4: Potential beneficiaries of the NYEP among individuals aged 18-35, National 2005-2006 Wage of workers Unemployed reservation wage % #people Monthly Weekly % % #people Monthly Weekly % Group Group Wage Hours Poor Group Group Wage Hours Poor Total 0 37.0 1,503,374 0.0 33.5 42.8 1-50 7.2 290,474 30.1 33.1 30.2 0.7 3,732 39.0 77.0 51-100 6.9 278,656 78.7 36.5 23.9 1.7 8,708 82.9 52.1 101-200 9.3 379,420 156.3 38.8 18.6 7.5 39,186 184.5 26.3 201-300 8.6 348,320 254.6 40.0 19.5 11.9 62,200 272.1 38.8 301-400 5.4 220,188 357.8 40.1 17.3 13.0 68,420 384.8 36.1 401-500 6.0 243,202 455.9 41.9 12.7 22.8 119,424 489.0 32.3 501-600 3.2 130,620 556.8 42.3 7.4 7.5 39,186 596.5 24.8 601-700 2.7 110,716 659.7 41.7 10.6 4.3 22,392 685.1 34.7 701-1000 5.5 223,298 852.9 44.8 8.5 19.2 100,764 920.2 9.1 1001-2000 5.5 223,920 1403.9 48.2 3.6 9.7 51,004 1539.1 14.5 2001+ 2.7 109,472 3855.1 46.8 5.1 1.8 9,330 3859.2 7.1 Urban 0 16.9 236,982 0.0 37.1 15.7 1-50 5.0 70,908 30.7 38.1 18.6 51-100 5.5 77,750 78.5 42.7 13.0 1.4 4,354 80.4 12.4 101-200 9.9 139,328 158.4 44.2 6.4 5.9 18,660 179.6 13.6 201-300 10.6 148,658 260.0 46.8 6.0 9.0 28,612 272.6 21.9 301-400 7.9 111,338 363.9 44.7 8.8 11.1 35,454 384.2 14.1 401-500 8.3 116,936 463.7 44.7 4.3 21.1 67,176 493.7 23.4 501-600 5.8 80,860 559.6 45.7 5.4 9.2 29,234 598.6 17.6 601-700 4.5 62,822 662.1 44.6 7.4 3.7 11,818 697.0 9.7 701-1000 9.9 138,706 861.4 48.5 3.5 24.4 77,750 912.7 6.6 1001-2000 10.6 149,280 1417.8 52.3 1.2 12.1 38,564 1516.9 12.0 2001+ 5.1 71,530 3976.9 50.4 1.6 2.1 6,842 4198.5 0.0 Rural 0 47.7 1,266,392 0.0 32.7 49.3 1-50 8.3 219,566 29.9 31.3 34.2 1.8 3,732 39.0 77.0 51-100 7.6 200,906 78.7 33.9 28.5 2.1 4,354 85.8 100.0 101-200 9.0 240,092 155.0 35.5 26.0 10.0 20,526 190.8 42.5 201-300 7.5 199,662 250.8 35.3 29.0 16.3 33,588 271.5 61.2 301-400 4.1 108,850 352.2 35.9 25.0 16.0 32,966 385.5 61.9 401-500 4.8 126,266 448.9 39.3 20.2 25.4 52,248 480.7 48.2 501-600 1.9 49,760 552.9 37.6 10.0 4.8 9,952 588.7 51.1 601-700 1.8 47,894 656.6 38.0 14.7 5.1 10,574 668.9 68.9 701-1000 3.2 84,592 839.0 39.0 16.5 11.2 23,014 947.2 18.2 1001-2000 2.8 74,640 1376.2 40.0 8.2 6.0 12,440 1635.1 25.4 2001+ 1.4 37,942 3623.4 40.0 11.7 1.2 2,488 3091.9 23.1 Source: Authors' estimation using GLSS5 data. 106 Table 5: Potential beneficiaries of NYEP, individuals aged 18-35 with junior secondary education completed, 2005-2006 Wage of workers Unemployed reservation wage % #people Monthly Weekly % % #people Monthly Weekly % Group Group Wage Hours Poor Group Group Wage Hours Poor Total 0 23.8 399,946 0.0 34.0 22.3 1-50 5.5 92,056 31.4 35.4 15.4 0.2 622 50.0 0.0 51-100 5.7 95,166 79.0 39.1 10.8 1.1 3,110 78.1 23.6 101-200 10.2 171,672 157.4 42.1 13.1 6.2 17,416 192.7 16.0 201-300 9.5 159,854 260.3 42.3 11.4 7.1 19,904 281.8 27.9 301-400 6.6 110,094 363.3 42.8 10.2 9.3 26,124 386.8 26.7 401-500 7.4 124,400 459.1 43.4 3.3 22.3 62,822 494.5 19.3 501-600 4.8 81,482 557.7 44.1 3.7 9.3 26,124 598.3 14.2 601-700 4.0 67,798 662.0 44.4 7.0 4.0 11,196 693.9 18.1 701-1000 8.5 143,060 862.8 46.8 5.0 25.4 71,530 923.9 6.6 1001-2000 9.7 162,342 1420.3 50.3 1.9 12.4 34,832 1526.3 8.7 2001+ 4.3 72,774 3749.0 50.9 0.8 2.7 7,464 4081.5 8.6 Urban 0 14.9 138,706 0.0 39.6 7.1 1-50 3.9 36,076 32.5 41.6 9.8 51-100 4.3 39,808 79.2 44.8 5.1 1.4 3,110 78.1 23.6 101-200 9.4 87,702 159.1 46.8 4.9 4.7 10,574 188.7 9.8 201-300 9.8 91,434 264.1 47.7 2.5 5.0 11,196 280.5 28.7 301-400 7.6 70,908 366.4 44.7 6.9 8.1 18,038 387.3 9.2 401-500 8.5 78,994 464.3 45.7 1.3 22.9 51,004 495.8 21.6 501-600 6.2 57,846 561.7 46.8 3.1 10.3 23,014 598.1 16.0 601-700 4.7 44,162 663.5 47.1 3.9 4.2 9,330 697.6 8.8 701-1000 11.3 105,118 866.6 49.3 2.5 26.5 59,090 916.7 6.4 1001-2000 13.3 123,778 1419.7 53.1 1.0 14.0 31,100 1518.2 6.7 2001+ 6.3 58,468 3880.1 51.8 0.0 2.8 6,220 4189.6 0.0 Rural 0 34.9 261,240 0.0 30.9 30.8 1-50 7.5 55,980 30.6 31.3 19.1 1.1 622 50.0 0.0 51-100 7.4 55,358 78.9 34.7 15.2 101-200 11.2 83,970 155.6 37.1 21.8 11.7 6,842 200.0 27.6 201-300 9.2 68,420 255.9 35.9 21.7 14.9 8,708 283.6 26.8 301-400 5.2 39,186 358.4 39.6 15.3 13.8 8,086 386.0 56.7 401-500 6.1 45,406 450.1 39.5 6.7 20.2 11,818 489.7 10.1 501-600 3.2 23,636 549.6 38.4 5.0 5.3 3,110 600.0 0.0 601-700 3.2 23,636 659.1 39.4 13.0 3.2 1,866 678.2 57.5 701-1000 5.1 37,942 853.3 40.9 11.1 21.3 12,440 952.6 7.7 1001-2000 5.2 38,564 1422.3 41.7 4.7 6.4 3,732 1615.5 30.4 2001+ 1.9 14,306 3277.6 47.7 3.6 2.1 1,244 3640.4 43.6 Source: Authors' estimation using GLSS5 data. 107 Figure 1: Distribution of potential beneficiaries of public works, National, no education criteria and low wages 000 1, 0 000' 80 ni, esi arici 600 ef en B 0 of 40 ber um N 200 0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 E150 E200 E250 E300 Source: Authors' estimation using GLSS5 data. Figure 2: Distribution of potential beneficiaries of public works, Urban, no education criteria and low wages 250 000' 0 ni,sei 20 aricif 0 15 ene B of 100 ber mu N 50 0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 E150 E200 E250 E300 Source: Authors' estimation using GLSS5 data. 108 Figure 3: Distribution of potential beneficiaries of public works, Rural, no education criteria and low wages 800 000' ni, 600 esi arici enef B 400 of ber um N 200 0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 E150 E200 E250 E300 Source: Authors' estimation using GLSS5 data. Figure 4: Distribution of potential beneficiaries of NYEP, National, junior secondary education completed and high wages 800 000' ni,sei 600 aricif 0 ene 40 B of ber mu 0 N 20 0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 E750 E1000 E1250 E1500 Source: Authors' estimation using GLSS5 data. 109 Figure 5: Distribution of potential beneficiaries of NYEP, Urban, junior secondary education completed and high wages 500 000' 0 ni, 40 esi arici 0 30 enef B of 200 ber um N 100 0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 E750 E1000 E1250 E1500 Source: Authors' estimation using GLSS5 data. Figure 6: Distribution of potential beneficiaries of NYEP, Rural, junior secondary education completed and high wages 200 000' ni,sei 150 aricif 0 ene 10 B of ber mu N 50 0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 E750 E1000 E1250 E1500 Source: Authors' estimation using GLSS5 data. 110 7.28 In table 5, the same procedure is repeated but we restrict the sample of eligible individuals to those with junior secondary education completed. For Figures 4 to 6 we also change the wages paid, as we are now simulating the potential beneficiaries of the NYEP. Clearly, restricting participation according to education reduces substantially the sample of potential beneficiaries, even though the wages proposed are much higher. Most of the drop in potential participants takes place in rural areas, where education levels are lower. In addition, the profile of beneficiaries according to the quintile of per capita consumption of households also changes, especially at the national level. While the potential program participants for public works are more likely to belong to the poorest quintiles at the national level when no education criteria are imposed, the reverse is observed in most cases for the NYEP when a junior secondary education is required and wages paid are higher. Within rural areas, the profile of potential participants remains pro-poor even when a junior secondary education is required, but it is less so than when no education criteria are imposed. In urban areas, the profile is in favor of better off households whether education criteria are imposed or not, but the bias against the poor is stronger with education criteria as expected. 4. Potential poverty impact of public works and the NYEP 7.29 Even with a junior secondary education criteria, the number of potential beneficiaries of the NYEP as estimated from the survey is much larger than the number of persons employed today by the NYEP or the number of youths registered in the program. For public works when no education criteria are imposed, the potential number of participants is even much larger. This is shown in tables 6 and 7, which also provide data on leakage rates and substitution effects. 7.30 In order to assess the potential impact of the NYEP as it is operating now on the basis of the numbers of jobs actually created, we randomly select among all potential beneficiaries of the program (the number of which depends on the wage provided) a number of participants so as to match the distribution of the actual program participants provided in table 1 by region (this is done using 25 random replications, and average values obtained with the 25 replication). This is done for each of the wages assumed to be provided. The same procedure is used to simulate the potential impact of public works and for comparison purposes with the NYEP we select here again the same number of participants as the numbers observed in table 1, but without geographic quotas although we restrict participants to rural areas. The results of this procedure and the related statistics on targeting performance are provided in tables 6 and 7. 7.31 Consider first table 6 which provides data for NYEP potential participants. The first column provides the estimate of the total number of potential beneficiaries of the program depending on the wage level, as estimated from table 4 at the national level. For example, at a wage level of one million cedis per month, 156,122 individuals in the Western region might be potential beneficiaries of public works according to our method for identifying such potential beneficiaries. The second column provides the share of those individuals living in households who are poor. For example, at a wage of one million cedis, 9.6 percent of the potential beneficiaries in the Western region live in a household in poverty according to the definition of poverty used by the Ghana Statistical Service (see Coulombe and Wodon, 2007). The third column provides the additional wage to be obtained by each individual, on average, depending on the wage proposed for the program. At a wage of one million cedis, 111 out of that amount, on average 765,100 cedis represents additional income for potential participants to the program. 7.32 The next column provides the leakage rate, which is computed as the product of the poverty rate times the additional wage divided by the reference wage of the program. In the Western region, out of each payment of one million cedis, 691,000 cedis would not result in additional income for the poor first because there is a substitution effect (slightly less than one fourth of the wage is "lost" due to the fact that some of the participants would have to give up other earnings in order to participate in the program), but mostly because most of the participants are not in poverty. The last two columns provide data on the ratios of the number of individuals registered in the NYEP or participating in the program as compared to the number of potential beneficiaries of the NYEP as we estimate them by region. With a wage rate of one million cedis per month, the registration rate among potential participants is only 10.4 percent and jobs are provided to only 5.5 percent of potential participants. We are probably overestimating the potential number of participants since we do not know about specific constraints that some individuals may have in participating in the program, but this helps to show that at the high wages now provided under the NYEP, there could be a large demand for jobs that would be difficult to supply at an affordable cost. 7.33 The results in table 6 suggest that while the substitution effects is not negligible, most of the wages obtained by NYEP participants are likely to be additional wages for them and their household. This is in part because some of the individuals concerned are now working without pay and are considered as potential beneficiaries of the program. By contrast, losses for poverty reduction due to the fact that many potential participants are not poor are much larger. This results in a high leakage rate from the point of view of poverty reduction of 64.1 percent at the national level for a wage of 750,000 cedis, and this increases further as wages paid are higher. 7.34 The poor performance in terms of targeting of the NYEP is observed despite efforts to target the program, at least to some extent, to poorer areas. Indeed, because in poorer areas there are few individuals with junior secondary education completed, the number of program participants as a share of the number of potential beneficiaries is higher in poorer areas. Remember that we simulate the NYEP impact using data on how the program is geographically targeted. For example, the ratio of participants to potential beneficiaries is lowest in Ashanti and Greater Accra. In terms of registered youth as compared to potential beneficiaries, the ratio is 6.6 percent in Greater Accra (for a wage of one million cedis), as compared to a rate of 31.5 percent in the Northern area. Clearly, the simulated public works program has pro-poor regional bias when we take into account the eligibility criteria, but this does not help much for its potential impact on poverty (although to the extent that some program participants are involved in the education and health sectors, there may be positive impacts from their work later on through higher human capital accumulation, especially for children). 7.35 The procedure is repeated in table 7 when no education condition is imposed and wages are lower for public works. In this case, we focus the program on rural areas, but without quotas per region. Hence, we do not replicate the geographic allocation of NYEP participants from table 1, but on the other had we impose as an eligibility condition the fact that the program is rural only. In addition, as already mentioned, we create a number of jobs equivalent to the number of jobs created by the NYEP. The leakage rates decrease 112 substantially. For example, at the national level, the leakage rate decreases from 66.2 percent with education criteria and a wage of one million cedis for the NYEP to 34.6 percent when no education is required and wages of 200,000 cedis per month are paid. The decrease in the leakage rates is due to the fact that individuals in rural areas willing to work at low wages tend to be in majority poor. 113 Table 6: Potential leakage effects of the NYEP for poverty reduction, by region for junior secondary education completed and high wages, 2005-2006 Region #of Poverty Additional Leakage #registered/ #job/ people Headcount wage Rate #people #people in % in '000 monthly in % in % in % E750 Western 139,950 10.9 557.0 66.2 7.2 5.7 Central 105,118 14.6 578.3 65.9 12.4 7.3 Greater Accra 263,728 8.9 488.7 59.4 8.5 3.0 Volta 118,802 19.9 573.9 61.3 15.2 7.3 Eastern 187,222 12.1 583.4 68.4 10.2 4.6 Ashanti 357,650 11.1 580.7 68.8 6.8 2.1 Brong Ahafo 135,596 18.8 581.3 62.9 14.7 5.8 Northern 62,822 22.5 542.9 56.1 35.0 26.3 Upper East 51,004 46.6 603.3 42.9 26.0 18.7 Upper West 57,846 62.7 672.5 33.5 21.8 16.7 Total 1,479,738 14.2 560.5 64.1 11.8 6.2 E1000 Western 156,122 9.6 765.1 69.1 6.5 5.1 Central 121,912 14.5 795.3 68.0 10.7 6.3 Greater Accra 338,990 7.8 679.1 62.6 6.6 2.3 Volta 125,022 19.3 803.6 64.9 14.5 6.9 Eastern 209,614 11.0 768.6 68.4 9.1 4.1 Ashanti 396,214 10.4 788.7 70.7 6.1 1.9 Brong Ahafo 144,926 18.4 797.4 65.1 13.7 5.5 Northern 69,664 21.4 725.3 57.0 31.5 23.7 Upper East 57,846 46.0 857.6 46.3 22.9 16.5 Upper West 60,956 61.8 896.9 34.2 20.7 15.9 Total 1,681,266 13.1 761.3 66.2 10.4 5.5 E1250 Western 162,964 9.3 984.0 71.4 6.2 4.9 Central 126,266 13.9 1010.5 69.6 10.3 6.1 Greater Accra 362,626 7.4 891.9 66.1 6.2 2.2 Volta 129,376 18.9 1034.5 67.1 14.0 6.7 Eastern 213,968 11.1 1003.1 71.4 8.9 4.0 Ashanti 414,874 10.1 1009.1 72.6 5.9 1.8 Brong Ahafo 148,658 17.8 1022.0 67.2 13.4 5.3 Northern 70,286 21.2 969.4 61.1 31.2 23.5 Upper East 58,468 46.6 1098.1 46.9 22.7 16.3 Upper West 62,822 62.3 1130.6 34.1 20.0 15.4 Total 1,750,308 12.7 983.4 68.7 10.0 5.3 E1500 Western 166,074 9.6 1215.1 73.2 6.1 4.8 Central 129,376 13.6 1235.3 71.1 10.1 5.9 Greater Accra 392,482 6.8 1093.8 67.9 5.7 2.0 Volta 133,108 18.4 1252.6 68.2 13.6 6.5 Eastern 218,322 10.8 1225.1 72.9 8.7 3.9 Ashanti 427,936 9.7 1225.5 73.8 5.7 1.8 Brong Ahafo 153,012 17.3 1240.3 68.3 13.0 5.2 Northern 71,530 20.7 1196.6 63.2 30.7 23.1 Upper East 61,578 45.2 1313.8 48.0 21.6 15.5 Upper West 62,822 62.3 1380.6 34.7 20.0 15.4 Total 1,816,240 12.3 1199.3 70.1 9.6 5.1 Source: Authors' estimation using GLSS5 data. 114 Table 7: Potential leakage effects of public works for poverty reduction, by region without education criteria and low wages, 2005-2006 Region #of Poverty Additional Leakage people Headcount Wage Rate in% In'000monthly in% E150 Western 161,098 21.4 125.9 66.0 Central 111,338 19.0 114.6 61.9 Greater Accra 120,046 7.7 100.7 61.9 Volta 163,586 37.8 119.1 49.4 Eastern 190,954 20.7 119.0 62.9 Ashanti 326,550 24.7 124.1 62.3 Brong Ahafo 238,848 34.4 129.0 56.4 Northern 368,224 57.4 127.1 36.1 Upper East 246,312 69.6 126.4 25.6 Upper West 335,258 83.8 146.4 15.8 Total 2,262,214 36.4 123.5 52.4 E200 Western 178,514 20.2 162.5 64.8 Central 125,022 18.4 152.7 62.3 Greater Accra 142,438 7.6 134.5 62.2 Volta 188,466 36.2 151.6 48.4 Eastern 219,566 19.5 152.3 61.3 Ashanti 375,688 23.0 157.5 60.7 Brong Ahafo 255,020 33.7 170.5 56.5 Northern 391,860 55.2 168.3 37.7 Upper East 264,350 69.7 168.9 25.6 Upper West 340,856 83.7 195.6 15.9 Total 2,481,780 34.6 160.8 52.6 E250 Western 193,442 20.0 199.1 63.7 Central 137,462 17.2 190.8 63.2 Greater Accra 155,500 7.0 172.7 64.2 Volta 200,906 35.8 190.2 48.8 Eastern 241,958 18.5 187.8 61.2 Ashanti 409,276 22.0 195.9 61.2 Brong Ahafo 271,192 33.6 210.9 56.0 Northern 434,156 54.1 200.7 36.9 Upper East 277,412 69.7 214.6 26.0 Upper West 348,320 83.9 244.5 15.7 Total 2,669,624 33.7 198.7 52.7 E300 Western 213,346 18.8 229.3 62.1 Central 154,256 17.9 223.2 61.1 Greater Accra 190,332 8.2 194.2 59.4 Volta 222,676 34.7 223.8 48.8 Eastern 266,216 17.9 220.4 60.3 Ashanti 457,792 21.5 228.2 59.7 Brong Ahafo 283,632 32.7 251.6 56.4 Northern 460,280 52.3 236.3 37.6 Upper East 297,316 70.5 257.6 25.4 Upper West 355,162 84.1 293.0 15.5 Total 2,901,008 32.6 232.4 52.2 Source: Authors' estimation using GLSS5 data. 115 7.36 The estimated potential impact of the program on poverty is given in table 11. The estimates are obtained in a very simple way. For the participants in the program (as simulated by us on the basis of the distribution of workers in the NYEP by region or of the distribution of participants in public works in rural areas) who belong to households living in poverty, we add to the consumption aggregate of the household the gains in earnings obtained by the participants, and we recomputed poverty using the same poverty lines (for a discussion of poverty measurement in Ghana, see Coulombe and Wodon, 2008). In other words, we assume that the full amount of the earnings gains for program participants translate into additional consumption for their households. For higher wages, the impact is higher, since the additional earnings obtained by participants are higher, but note that the number of participants is kept unchanged since it is based on the data provided in table 1 for the NYEP, and a similar number of participants in public works for the comparison with that type of program. 7.37 With education criteria, at the current wage of the NYEP, which is probably of the order of one million cedis per month or higher, one can conjecture that the program is reducing the headcount index of poverty (which is simply the share of the households with consumption per equivalent adult below the poverty line) with its current size at the national level by 0.059 percentage point, which is very small. If no education eligibility criteria are imposed and the wages paid are lower through public works in rural areas, say at about 250,000 cedis per month, the reduction is only slightly lower, at 0.047 percentage point. Both impacts are very small, but in terms of cost efficiency, public works are much better since the cost is about four times lower for public works at a wage of 250,000 cedis than for the NYEP at a wage of one million cedis per month. A similar story emerges with the poverty gap, which takes into account the distance separating the poor from the poverty line (with a zero distance given to the non-poor) and for which the much less costly rural public works seem to reduce poverty more than the NYEP. Table 8: Potential impact on poverty of the NYEP and public works, National, 2005-2006 Within target group Whole population of potential beneficiaries Headcount Poverty Gap Headcount Poverty Gap National ­ All individuals ­ Public works E150 0.307 0.169 0.032 0.017 E200 0.350 0.189 0.039 0.021 E250 0.384 0.207 0.047 0.025 E300 0.397 0.215 0.052 0.028 National ­ JSS completed ­ NYEP E750 0.845 0.276 0.057 0.019 E1000 0.778 0.245 0.059 0.019 E1250 0.784 0.241 0.062 0.019 E1500 0.757 0.232 0.062 0.019 Source: Authors' estimation using GLSS5 data. 5. Conclusion 7.38 As in many other countries, youth unemployment is a major issue in Ghana. In October 2006, the National Youth Employment Program (NYEP) was created by Ghana's government in order to provide temporary employment as well as training to up to 500,000 people between the ages of 18 and 35 over the period from 2006 to 2009. Even though 116 efforts are made to serve the northern regions, the program still appears to have a partial bias in favor of urban and relatively wealthier areas (at least in terms of number of registered youth), probably because that's in part where the issue of youth unemployment is most visible, and that's also where it is easier to provide meaningful training to youths participating in the program thanks to the network of firms and non-profit organizations that can employ youths there. More importantly, the program targets youth with a junior secondary education completed or higher, and this is likely to mean that the impact of the program on poverty is likely to be limited. Wages also appear to be high, so that the program is costly, and perhaps leading to unrealistic expectations in terms of future wages for the youth enrolled in the program. 7.39 Using Ghana's 2005/06 Living Standards Survey and simple simulation techniques, we have estimated in this paper the likely impact on poverty of the program. Our estimates suggest that at the current wage of the NYEP of about one million cedis per month, the program is reducing the headcount of poverty at the national level by a very small amount. This is in part because the leakage rate (with considers both wage substitution effects and leakage of the program to the non-poor) is high, at 64 percent to 70 percent depending on the simulations. The main reason for this high leakage rate is not the fact that there is a substitution effect of the program in terms of wages, whereby only part of the wages paid to beneficiaries generate additional income since some beneficiaries would have done other work if they had not participated in the program. Rather, the limited impact and poor targeting performance of the program is due to the education criteria used to screen participants and the fact that the program does not focus on poor areas. When compared to a rural public works program, the NYEP appears to be four to five times more expensive for reducing poverty than public works. References Adams, Arvil V. (2007). The Role of Youth Skills Development in the Transition from School to Work: A Global Review, HDNCY Discussion Paper No. 5, Washington, D.C.: World Bank Blunch, Niels-Hugo (2006) Skills, Schooling and Household Well-Being in Ghana, Unpublished PhD Dissertation, The George Washington University, Washington, DC. Coulombe, H. and Q. Wodon (2008). Poverty, Livelihoods, and Access to Basic Services in Ghana. Ghana CEM: Meeting The Challenge of Accelerated and Shared Growth. Glewwe, Paul (1999) "The Impact of Cognitive Skills on Wages," in Paul Glewwe (ed) The Economics of School Quality Investments in Developing Countries: An Empirical Study of Ghana, London: Macmillan. Government of Ghana, ILO and UNDP (2004). "An Employment Framework for Poverty reduction in Ghana", mimeo. Haan, Hans Christiaan and Nicolas Serriere (2002). "Training for Work in the Informal Sector: Fresh Evidence from West and Central Africa," Occasional Papers of the International Training Centre of the ILO, Turin: ILO Johanson, Richard and Arvil V. Adams (2004). Skills Development in Sub-Saharan Africa, Washington, D.C.: World Bank 117 Ministry of Education, Youth and Sports (2006). "Preliminary Education Sector Performance Report 2006," Accra Ministry of Manpower, Youth and Employment (2006). "Youth Employment Implementation Guidelines (Ghana Youth Job Corps Program)", Accra. Monk, Courtney, Justin Sandefur, and Francis Teal (2007). "Apprenticeship in Ghana," Centre for the Study of African Economies, Department of Economics, University of Oxford, (processed). Palmer, Robert (2007). Skills Development, the Enabling Environment and Informal Micro- Enterprise in Ghana, doctoral thesis (mimeo), Edinburgh: University of Edinburgh Ravallion, M., 1999, Appraising Workfare, World Bank Research Observer, 14: 31-48. Rosholm, M., H. S. Nielsen and A. Dabalen (2007). "Evaluation of Training in African Enterprises," Journal of Development Economics, vol. 84(1), pages 310-329. Teal, Francis (2000) "Real Wages and the Demand for Skilled and Unskilled Male Labor in Ghana's Manufacturing Sector: 1991-1995," Journal of Development Economics, 61: 447- 461. World Bank (2007a). World Development Report 2007: Development and the Next Generation, World Bank, Washington, DC. World Bank (2007b). Global Inventory of Interventions to Support Young Workers, World Bank, Washington, DC. World Bank (2008). Ghana: Job Creation and Skills Development, Report No. 40328 ­ GH, World Bank, Washington, DC. 118 MAP SECTION IBRD 33411 2°W 0° 2°E To Tenkodogo BURKINA FASO To Hamale Navrongo U P P E R E A S T Bobo- Diolasso WalewaleTumu Bolgatanga GHANA Nakpanduri U P P E R W E S T Walewale To Dapaong 10°N Black Volta Kolpawn Wa Wa 10°N Gushiegu White Volta To Djougou N O R T H E R N Yendi To Tamale Ferkéssédougou BENIN Sawla Fufulsu Bole Daka To Djougou CÔTE Nakpayili Oti TOGO D'IVOIRE To BlackVolta Salaga Bouna Makongo Yeji Kintampo Dambai 8°N 8°N Jema V O L T A B R O N G - A H A F O Atebubu Kwadwokurom Tain Techiman Pru Berekum To Sokodé Sunyani Mount Afadjato (880 m) K w Bia a h Afram u Lake Kpandu nges AgbovilleoT P Volta To l Agogo Ra Abomey Goaso a t e a u Kumasi Bibiani E A S T E R N -Togo Ho Krokosue A S H A N T I Anum im To Porto- no Obuasi Diaso Birim p Novo a aT Kade Koforidua kw Volta Aflao 6°N Oda W E S T E R N A 6°N Dunkwa Enchi GREATER 2°E ACCRA AbidjanoT Tema Twifo Praso ACCRA Prestea C E N T R A L GHANA Ankobra Pra Tarkwa Winneba SELECTED CITIES AND TOWNS Cape Coast REGION CAPITALS Newtown NATIONAL CAPITAL This map was produced by Sekondi the Map Design Unit of The Axim Takoradi RIVERS World Bank. The boundaries, Gulf of G uinea colors, denominations and any other information shown MAIN ROADS on this map do not imply, on the part of The World Bank 0 20 40 60 80 Kilometers RAILROADS Group, any judgment on the legal status of any territory, REGION BOUNDARIES or any endorsement or 0 20 40 60 Miles a c c e p t a n c e o f s u c h boundaries. 2°W 0° INTERNATIONAL BOUNDARIES SEPTEMBER 2004