Policy Research Working Paper 10726 Who Did Covid-19 Hurt the Most in Sub-Saharan Africa? Feraud Tchuisseu Seuyong Ifeanyi Edochie David Newhouse Ani Rudra Silwal Poverty and Equity Global Practice & A verified reproducibility package for this paper is Development Data Group available at http://reproducibility.worldbank.org, March 2024 click here for direct access. Policy Research Working Paper 10726 Abstract How did the economic crisis caused by the Covid-19 pan- insecurity and distressed asset sales among these households demic impact poor households in Sub-Saharan Africa? This during the second half of 2020. With respect to education, paper tackles this question by combining 73 High-Frequency children in the bottom quintile were 15 percentage points Phone Surveys collected by national governments in 14 less likely to engage in learning activities than those in the countries with older nationally representative surveys top quintile in the immediate aftermath of the crisis, and containing information on household consumption. In the engagement gap between the bottom 40 and top 60 particular, it examines how outcomes differed according widened in the summer before narrowing in the fall due to predicted per capita consumption quintiles in the first to large declines in engagement among the top 60. Poorer wave of the survey, and in subsequent waves by house- households were slightly more likely to report receiving holds’ predicted per capita consumption. The initial shock public assistance immediately following the shock, and this affected households throughout the predicted welfare dis- difference changed little over the course of 2020. The results tribution. Households in the bottom 40 percent responded highlight the widespread impacts of the crisis both on wel- by sharply increasing farming activities between May and fare and children’s educational engagement, the importance July of 2020 and gradually increasing ownership of non- of agriculture and household non-farm enterprises as safety farm enterprises starting in August. This coincided with an nets for the poor, and the substantial recovery made by the improvement in welfare, as measured by a decline in food poorest households in the year following the crisis. This paper is a product of the the Poverty and Equity Global Practice and the Development Data Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/prwp. The authors may be contacted at dnewhouse@worldbank.org. A verified reproducibility package for this paper is available at http://reproducibility.worldbank.org, click here for direct access. RESEA CY LI R CH PO TRANSPARENT ANALYSIS S W R R E O KI P NG PA The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Who did Covid-19 hurt the most in Sub-Saharan Africa? Feraud Tchuisseu Seuyong Ifeanyi Edochie David Newhouse Ani Rudra Silwal JEL codes: O12, I31 Keywords: Covid-19, Sub-Saharan Africa, Distributional impacts Introduction How did the economic impact of the crisis caused by Covid-19 impacted poor households in Sub-Saharan Africa? While statistics on infections and deaths suggest that Sub-Saharan Africa (SSA) may have escaped the worst of Covid-19, the economic cost was major. Poverty in the region was projected to increase by 2 percentage points or more in 2020 (Montes et al. 2020 and Lakner et al. 2021). While the full extent of the economic damage is still unknown, data from phone surveys collected since the pandemic provide an indication of the adverse impact on livelihoods in the region. 1 However, little is known about how impacts differed for poor and wealthy households within countries. This note aims to help fill this gap in order to better understand the distributional impacts of the crisis and inform recovery efforts. We combine data collected through 73 phone surveys in SSA since the early stages of the global pandemic with richer multi-topic household surveys collected prior to the pandemic to examine how the impact of Covid-19 differed for poorer and wealthier households, as measured by predicted household per capita consumption. There are six main findings: 1. The initial impact of Covid-19 on the prevalence of income declines and the ability to work as usual was widespread throughout the predicted consumption distribution. 2. During the summer of 2020, there were large (20 pp) increases in the share of households in the bottom 40 percent that engaged in farming activities, and a more gradual increase 10 pp increase in the share of these households that owned or partially owned a non-farm enterprise. 3. During the latter part of 2020 and early 2021, food insecurity and the incidence of distressed asset sales declined faster for the bottom 40 than the top 60 percent. Conversely, the prevalence of income declines and the inability to work as usual declined faster for the top 60 percent, who were better able to cope with the medium-term impacts of the crisis. 4. Children in the bottom 40 percent were less likely to engage in educational activities in the immediate aftermath of the crisis. This disadvantaged narrowed greatly over the course of December 2020, as the share of children engaging in learning or educational activities fell about 10 percentage points for the bottom 40 and about 17 pp for the top 60 both between June and December 2020. 5. Public government assistance slightly favored poorer households in the spring and summer of 2020, a pattern that changed little as the crisis evolved. Overall, the results indicate that the initial impacts of the Covid-19 were widespread throughout the predicted welfare distribution, and that the agricultural sector and establishment of non-farm enterprises acted as an important safety net. The movement back to agriculture may have helped poorer households reduce food insecurity and distressed asset sales, but may have delayed their ability to stem declines in income. Households faced serious challenges maintaining their children’s engagement with educational activities during the summer, and households in the bottom 40 were slightly more likely than those in the top 60 to benefit from public assistance throughout 2020. 1 See Bundervoet et al, 2021, Khamis et al, 2021, Dang et al, 2020, and Gupta et al, 2020. 2 Data and methodology High Frequency Phone Surveys and Global Monitoring Database This section describes the data sources and the methodology used to examine the impact of Covid-19 across the welfare distribution in 14 countries in Sub-Saharan Africa. We rely on two different databases for our analysis: High Frequency Phone Surveys (HFPS) implemented since the pandemic began and the Global Monitoring Database (GMD), a database of household surveys implemented prior to the pandemic, compiled and managed by the World Bank. We rely on survey-to-survey imputation, which entails developing a model of household consumption in the GMD for each country, and using the model to impute the welfare status of households in each wave of that country’s HFPS. Since Spring 2020, the World Bank has supported national statistical offices in developing countries to design and implement phone surveys to examine various aspects of Covid-19 on lives and livelihoods. 2 The rapid pace at which the Covid-19 pandemic has affected countries has made access to timely knowledge on the impacts of Covid-19 critical to effectively design, target and evaluate policy interventions. However, implementing face-to-face interviews was severely restricted in the context of widespread government-imposed social distancing practices. Phone surveys filled this gap by eliciting information from individuals and households rapidly and at low cost. In all 14 of the countries we consider, phone surveys were implemented as follow-up surveys from previous nationally representative surveys. These previous surveys were used to reweight the phone survey to make it more representative of the national population of households. However, the phone surveys disproportionately interview household heads, meaning that individual level characteristics such as the ability to work at usual were not representative of the underlying population of workers. Nonetheless, phone surveys can be helpful in tracking the responses to and impacts of the pandemic, particularly if they are combined with traditional household surveys. Table 1 lists the countries and waves of HFPS that were used for this analysis. Table 1: Sample size of HFPS waves and the corresponding multi-topic survey from GMD GMD Country High Frequency Phone Survey survey Wave 1 2 3 4 5 6 7 8 9 10 11 Survey 1,968 2,210 2,125 2,104 2,071 2,032 2,011 1,998 Burkina Faso 2018 EHCVM Ethiopia 3,249 3,107 3,058 2,878 2,770 2,753 2,536 2,222 2,077 2,178 1,982 2015 HICES Gabon 1,656 2017 EGEP Ghana 3,265 2016 GLSS-VII Gambia 1,436 1,385 1,335 1,332 1,297 2018 EHCVM Mali 1,766 1,935 1,901 1,901 1,766 1,884 1,760 1,694 2018 EHCVM Mozambique 1,097 978 960 841 901 2014 IOF Malawi 1,729 1,646 1,624 1,616 1,589 1,591 3,112 3,098 3,090 2016 IHS-IV Niger 1,264 2018 EHCVM Nigeria 1,950 1,820 1,790 1,789 1,773 1,762 1,726 1,723 1,706 1,699 1,680 2018 LSS Senegal 1,220 2018 EHCVM 2 https://www.worldbank.org/en/topic/poverty/brief/high-frequency-monitoring-surveys 3 Sierra Leone 6,571 4,885 2018 SLIHS Uganda 2,226 2,199 2,145 2,135 2,122 2,100 2016 UNHS Zimbabwe 1,747 1,639 1,451 2017 PICES A secondary source of data is the Global Monitoring Database (GMD), which is a collection of globally harmonized household survey data compiled by the World Bank. The GMD includes most recent household surveys used by national statistical agencies around the world, and subsequently, by the World Bank, to compute the official poverty statistics. In addition to the welfare aggregate, surveys in GMD include variables on demographics, assets, labor, education, and other topics that have been harmonized to a common set of specifications, making the GMD well-suited for cross-country analysis. This analysis covers 14 countries in SSA (Table 1). Although more countries are included in both the GMD and HFPS, we were limited by three factors. First, the analysis could only be conducted in countries where the HFPS was conducted and which also has a survey available in GMD that was conducted prior to the pandemic. Second, we limited our analysis only to surveys in GMD that were conducted in recent years, using a cutoff of 2014. Third, we limited our analysis to countries that both had a minimum set of variables that could be used to model household per capita consumption. Survey-to-survey imputation Survey-to-survey imputation (S2S) predicts household per capita consumption in a household survey that collects information only on selected covariates of consumption by using information from another survey that collected both the per capita consumption and its covariates. This literature builds largely on the poverty mapping literature (Elbers, Lanjouw, and Lanjouw, 2003) and has been widely implemented (Dang 2020, Newhouse et al. 2014, Stifel and Christiaensen 2007). This literature is in contrast to the multiple imputation literature inspired by Rubin (1987), which largely focuses on filling missing data within a survey rather than using information from one survey to make predictions in another one. There are generally two types of S2S studies in the literature: one that conducts the imputation over time and another that conducts the imputation within the same period but across surveys. The first type includes studies such as those by Newhouse and Vyas (2022) and Stifel and Christiaensen (2007). Studies that impute across surveys within the same time frame include Beltram et al. (2020), Dang (2019), Lucchetti et al. (2018), Dang et al. (2018), and Cuesta and Ibarra (2017). Dang (2020) and Dang et al. (2019) provide guidance on the appropriate methodology and caveats to using S2S. Implementation of S2S in this study first involved identifying a set of variables that are available both in HFPS and GMD. This yielded the following variables: household size, urban/rural location, subnational administrative area, share of children, and share of elderly residents in the household. This list of variables was selected by carefully comparing the data dictionaries of HFPS and GMD. Both the HFPS and the GMD are both designed to be nationally representative, except for Mozambique, where the HFPS only covers urban areas. A major difference between the HFPS and the GMD is that while the GMD sample covers both households that own phones and those that do not, the HFPS (by design) only covers households that own a mobile phone. Since households that own mobile phones are likely to be a non-random sample of the population, we select a comparable sub-sample of mobile phone owners in the GMD for the purposes of S2S. Figure 1 compares these samples for the 14 countries in the sample. Appendix Figure 1 presents these differences for individual countries in our analysis sample. 4 As mentioned above, the HFPS does not randomly sample respondents within the household, and we therefore mainly considered household-level characteristics for imputation. 3 In particular, respondents in the HFPS are more likely to be household heads, older, male, and urban. This implies that modeling consumption using individual variables may yield a consumption distribution (in HFPS) that is significantly different than the national distribution in GMD. Therefore, we use survey-to-survey models specified at the household level. Figure 1: Comparison of variables between GMD and HFPS Wave 1 Note: Table shows the mean or median of the differences across the 14 countries in the sample between the full GMD, the GMD with phone owners only, and the HFPS. Appendix Figure 2 presents numbers for all countries. The next step in the S2S procedure involved building a model of household per capita consumption as a function of the variables (and their interaction terms) that are common between the GMD and the HFPS. The Least Absolute Shrinkage and Selection Operator (LASSO) technique was implemented to select a subset of all the candidate variables, to avoid overfitting the model to the specific GMD sample. 4 The coefficients of the (post-)lasso model were then used to predict per capita household consumption of all households in the HFPS across all waves. The S2S methodology implemented with the adjusted R-squared estimated for each country are described in more details in the appendix. The section also compares the observed and predicted distributions of per capita household in the GMD with the predicted consumption in the HFPS. We then generate quintiles of the predicted per capita household consumption in the HFPS to tabulate various outcomes across the quantiles. 5 Once the quintiles are generated, we perform two sets of analysis: The first looks at differences across quintiles in the initial wave of the survey, while the second looks at the average differences between the bottom 40 and top 60 by month. The next section describes the main findings of this analysis. Tracking the evolution of the crisis 3 The only individual level characteristic considered is the ability to work as usual. 4 When implementing LASSO, we selected the penalty parameter lambda that minimize the Bayesian Information Criterion (Zhang, Li, and Tsai 2010). We ensured that if an interaction term was selected by LASSO, the appropriate level terms were also included in the final model, following Chen, Li, and Wang (2020). A Stata ado file that implements this is available from the authors upon request. 5 Observations were weighted using household survey weights when constructing the quintiles. 5 With a measure of household predicted welfare in hand, it is straightforward to report averages of outcomes from the first survey wave by predicted welfare quintile, to get a sense of how the initial impacts of the crisis varied over the predicted per capita consumption distribution. Analyzing how these differences evolved over time, however, is more complicated because countries enter and exit the sample, forming an unbalanced sample. To analyze the evolution of the impact of Covid-19, we therefore pooled data from 73 phone surveys from the 14 countries with imputed welfare. The goal of the analysis is to investigate how trends in income loss, employment, food insecurity, and social protection evolved for households in the bottom two quintiles as opposed to the top three. To examine trends, we estimate a logit model using the following model specified at the household level: ℎ = 1 () + 2 () ∗ 40ℎ + + ℎ ℎ() is the outcome of interest observed in household h, region (i.e. state/province) r, country c, and month t. In this case it is in this case is one of eight indicators: “Household received decreased total income”, “Respondent could work as usual”, ”Household with children engaged in any learning/education activities since school closures”, “Households with adult member skipped a meal in the last month due to lack of money or other resources”, “Household engaged in farming activities since the beginning of 2020”, “Households owning a non-farm enterprise”, “Households receiving government assistance” , and “Households that sold assets to cope with the pandemic”. 40ℎ is a dummy variable taking on one if household h is in the bottom 40 percent. is a country fixed effect, and ℎ is a dummy variable indicating that household h resides in an urban area. The term 1 () is the cubic polynomial in the month of the survey, measured since March 2020. Thus, the 1 () term reflects the average trend across months for the top three quintiles and the 1 () + 2 () ∗ 40ℎ term represents the average trend for households in the bottom 40 percent. Finally, ℎ is the error term, clustered at the level of the survey wave. After estimating the model, we plot the average predictive margins for the bottom 40 and top 60 percent by month, evaluated at the means of the other covariates and treating all dummy variables as balanced. Results This section contains the main findings of this analysis based on a S2S imputation of per capita household computation in HFPS based on a model of consumption in GMD. We examine the impact of Covid-19 on changes in household income and labor, food security, children’s education, and coping strategies implemented by households across the consumption distribution. Household Income and labor The initial impact of Covid-19 on the prevalence of reported total income declines was widespread, with minor differences across the distribution. Figure 2a plots the share of households that experienced a decrease in total income immediately after Covid-19 struck. Across the 14 countries analyzed in this sample, the average country reported that more than 76% of households experienced a decrease in total income. This share is about 5 percentage points higher for the poorest quintile (Q1). Appendix Figure 3 show a similar pattern for farm income. Declines in wage income, in the initial wave of the HFPS, were most prevalent in the third and fourth quintile. Figure 2c describes how the prevalence of total income declines has evolved since the beginning of the pandemic, after controlling for country and urban location. 6 Through the first six months, households in the top 60 percent of the welfare distribution were equally likely to report total income declines since the previous survey wave than the bottom 40 percent. Around November 2020, the bottom 40 percent became 5 to 10 percentage points more likely to report income declines, with a statistically significant difference in February 2021. Figure 2: The impact of Covid-19 on income and labor across the welfare distribution Figure 2a: Households that experienced a decrease Figure 2b: Share of Individuals that stopped in total income (wave 1) working (wave 1) Figure 2c: Evolution of probability of household Figure 2d: Evolution of the inability to work as income decline usual .3 .35 .4 .45 .5 .55 .6 .65 .7 .75 .8 Share of households Apr May Jun Jul Aug Sep Nov Dec Jan Feb Apr Top 60% Bottom 40% Note: % of HHs that experienced decrease in total income. Standard errors clustered on survey wave Sources: HFPS and GMD: Q1 is poorest, Q5 is wealthiest. Figure 2b suggests that approximately a quarter of the respondents in the average country stopped working since the beginning of the pandemic. Respondents are mainly household heads, however, and are therefore not representative of the full population of workers. Although we see a small increase in the share of the population in higher quintiles, a large share of respondents reported stopping work throughout the welfare distribution. Figure 2d compares the evolution of the impact of Covid-19 on the ability to work as usual for the bottom 40 and top 60 percent of the population. A larger share of the top 60 percent of the population were unable to work as usual in the early stages of the pandemic. The 7 magnitude of the difference ranged from five to ten percentage points during the spring and summer of 2020, and was statistically significant in the first five months. 6 As was the case for total income declines, however, this distributional gap narrowed and had disappeared by October 2020. Figure 3a shows the evolution of the difference between the bottom 40 and top 60 in those reporting that they had engaged in farming since the start of 2020. In the initial stages of the crisis, poor households were more likely to enter agriculture than wealthier households. Wealthier households also increased greatly farming activities later, in December and January, narrowing the gap back to pre-crisis levels. The shrinking differences in early 2021 are imprecisely estimated, however. Meanwhile, the share of households in the bottom 40 owning, even partially, a non-farm enterprise increased between July 2020 and March 2021 (Figure 3b). Like farming, wealthier households saw a large increase in non-farm enterprise ownership in early 2021. Overall, the results indicate that poor households were more likely to establish household enterprises in response to the crisis during the summer and fall of 2020. Figure 3a: Evolution of share of households Figure 3b: Evolution of share of households engaged in farming activities owning a non-farm enterprise Food security A major concern during Covid-19 has been its potential impact on food security of households in SSA. While the HFPS data contain several measures of food insecurity, we focus on the share of households in which an adult went without eating for a full day due to lack of money, because it is both objectively defined and an indicator of severe financial distress. As expected, poorer households report greater food insecurity than non-poor households. The difference is fairly modest, however, as about 17 percent of the bottom and second quintile report not eating for a fall day, while 11 percent of the top quintile does. This is consistent with the notion that the initial impacts of the crisis were widespread throughout the welfare distribution. Figure 4b presents how the share of households in the bottom 40 percent in which an adult went without eating for a day changed over the course of the pandemic, relative to households in the top 60 percent. Food insecurity increased significantly for the bottom 40 percent between April and June 6 Results of the tests for statistical significance are not shown 8 2020, before falling dramatically between July and April 2021, suggesting a strong recovery for the poorest households in the second half of 2020. Figure 4a: Share of households in which an adult Figure 4b: Evolution of share of households in went without eating for a whole day in the last which an adult went without eating for a whole month due to lack of resources (wave 1) day in the last month due to lack of resources Children’s education: Perhaps the population that was most affected by Covid-19 is children, whose learning was adversely affected by school closures and the lack of alternative arrangements when the pandemic began. Figure 5a plots the share of households with children 6 to 18 that reported engaging in educational activities since school closures by predicted welfare quintile. There is a clear pattern of households in the bottom 3 consumption quintiles being about 15 percentage points less likely to report children engaged in educational activities than those in the top quintile. Figures 5b and 5c corroborate the notion that educational engagement was more difficult for poor households than wealthier households when it came to completing assignments provided by the teacher or watching education TV programs. Finally, Figure 5d plots how the share of children engaged in education activities since the beginning of the pandemic for the Bottom 40 and the Top 60 percent of the population. Engagement increased slightly for the top 60 between April and June, before falling dramatically between June and January. Engagement declined more gradually for the bottom 40, about 10 percentage points between April 2020 and January 2021. The difference between the bottom 40 and top 60 was statistically significant until November. This points to the widespread challenges faced by both wealthier and poorer households in continuing educational activities during such a prolonged crisis. Figure 5: The impact of Covid-19 on children’s educational engagement across the welfare distribution 9 Figure 5a: Households with children engaged in Figure 5b: Households with children that any learning/education activities since school completed teacher assignment since school closures (wave 1) closures (wave 1) Figure 5c: Households with children that watched Figure 5d: Evolution of children’s engagement in educational TV programs since school closure any learning/educational activities (Bottom 40 vs (wave 1) Top 60) Sources: HFPS and GMD Safety nets and asset sales One silver lining of the global pandemic has been that many governments were, at least initially, able to target assistance to poorer households. Figure 6a presents the share of households that reported receiving any form of government assistance in wave 1. Although only about 12 percent of households in 10 the lower two quintiles reported receiving assistance, the households in lower quintiles were significantly more likely to receive assistance compared to households in higher quintiles, with a difference of about 4 pp. However, figure 6c suggests that this progressive targeting was less apparent when considering data from all of waves, with a difference of about 1 percentage point. This difference changed little between April 2020 and January 2021. Figures 6b and 6d examine the prevalence of asset sales in response to the crisis. Figure 5b indicates that poor households were substantially more likely to sell assets than wealthier ones in the immediate aftermath of the crisis, although less than 7 percent of respondents in the first and second quintile reported asset sales. Figure 6d suggests that over the course of the pandemic, poorer households initially became more likely to sell assets, after which asset sales declined for the poor during the summer of 2020. This is consistent with the decline in food insecurity for the poor reported above. Figure 6a: Households that received government Figure 6b: Households that sold assets to pay assistance when experiencing job/income loss (wave for basic living expenses (wave 1) 1) Figure 6c: Evolution of share of households that Figure 6d: Evolution of share of households received any form of government assistance that sold assets to pay for basic living expenses 11 Conclusion Understanding the impact of Covid-19 on households in different parts of the welfare distribution was an important factor when considering appropriate policy responses in response to the pandemic and will provide lessons for similar events in the future. We combined the High-Frequency Phone Surveys collected by national governments since the beginning of the pandemic with older surveys with information on household consumption to shed light on this question. We find that the initial impact of Covid-19 has been fairly widespread throughout the consumption distribution. In the initial aftermath of the crisis, poorer households were equally likely to report income declines but less likely to report the inability to work as usual. While government assistance slightly favored poorer households initially, this progressive targeting disappeared quickly and the share of households receiving government assistance was similar for the bottom 40 and top 60 percent for the year after the crisis. In the summer and fall of 2020, there was a large increase for households in the bottom 40 in farming activities and non-farm household enterprises. This coincided with economic recovery, as seen by large reductions in food insecurity and distressed asset sales among the bottom 40 during this time, as well as a reduction in the prevalence of income declines. However, the engagement of children in educational activities remains a concern. By the end of 2020, only about one third of households with children reported that children aged 6 to 18 were engaged in educational activities, a decline of about 15 percentage points. To sum up, the results suggest that the initial impact of the shock was severe throughout the welfare distribution, and that there was a marked recovery for poor households associated with an increase in farming and non-farm household enterprises. However, poor households remained more likely than non- poor households to report income declines and not working as usual in the medium term. The impact on educational engagement was severe and felt throughout the distribution. The results highlight the importance of policies to ensure productivity in the agricultural sector, as well as non-agricultural household enterprises, in response to the crisis. The analysis in this note could be extended in several directions. We do not closely examine variation in the impact of the global pandemic across the 14 countries in our analysis sample, or differences between urban and rural areas, which could provide further insights. We also do not examine the differences across various socio-economic characteristics such as age, gender, and labor market characteristics of individuals, although this is challenging in phone surveys because of the non-representative nature of the sample. As more face-to-face surveys are implemented and analyzed, a clearer picture will emerge of the distributional impacts of this prolonged pandemic in Sub-Saharan Africa. 12 References Beltram, Theresa, et al. "Estimating Poverty among Refugee Populations: A Cross-Survey Imputation Exercise for Chad." World Bank Policy Research Working Paper 9222 (2020). Bundervoet, Tom; Davalos, Maria E.; Garcia, Natalia. 2021. The Short-Term Impacts of COVID-19 on Households in Developing Countries: An Overview Based on a Harmonized Data Set of High-Frequency Surveys. Policy Research Working Paper;No. 9582. World Bank, Washington, DC. © World Bank. Chen, Kedong, William Li, and Sijian Wang. "An Easy-to-Implement Hierarchical Standardization for Variable Selection under Strong Heredity Constraint." Journal of statistical theory and practice 14 (2020): 1-32. Cuesta, Jose, and Gabriel Lara Ibarra. "Comparing cross-survey micro imputation and macro projection techniques: Poverty in post revolution Tunisia." Journal of Income Distribution 25.1 (2017): 1-30. Dang, Hai‐Anh H. "To impute or not to impute, and how? A review of poverty‐estimation methods in the absence of consumption data." Development Policy Review (2020). Dang, Hai‐Anh, Dean Jolliffe, and Calogero Carletto. "Data gaps, data incomparability, and data imputation: A review of poverty measurement methods for data‐scarce environments." Journal of Economic Surveys 33.3 (2019): 757-797. Dang, Hai-Anh H., Peter F. Lanjouw, and Umar Serajuddin. "Filling Gaps when Poverty Data are Missing: Updating Poverty Estimates Frequently with Different Data Sources in Jordan." Contemporary Issues in Development Economics. Palgrave Macmillan, London, 2016. 90-96. Dang, Hai-Anh, Toan Luu Duc Huynh, and Manh-Hung Nguyen. "Does the Covid-19 pandemic disproportionately affect the poor? Evidence from a six-country survey." (2020). Dang, Hai-Anh, and Paolo Verme. "Estimating Poverty for Refugee Populations: Can Cross-Survey Imputation Methods Substitute for Data Scarcity?." World Bank Policy research Working Paper 9076 (2019). Gupta, Anubhab, Heng Zhu, Miki Khanh Doan, Aleksandr Michuda, and Binoy Majumder. "Economic Impacts of the COVID− 19 Lockdown in a Remittance‐dependent Region." American Journal of Agricultural Economics (2020). Khamis, Melanie; Prinz, Daniel; Newhouse, David; Palacios-Lopez, Amparo; Pape, Utz; Weber, Michael. 2021. The Early Labor Market Impacts of COVID-19 in Developing Countries : Evidence from High- Frequency Phone Surveys. Jobs Working Paper;No. 58. World Bank, Washington, DC. © World Bank. Lakner, Christoph; Yonzan, Nishant; Mahler, Daniel Gerszon; Castaneda, Andres; Wu, Haoyu. Updated estimates of the impact of COVID-19 on global poverty: Looking back at 2020 and the outlook for 2021. Blog post accessible at https://blogs.worldbank.org/opendata/updated-estimates-impact-covid-19- global-poverty-looking-back-2020-and-outlook-2021 (2021). Lucchetti,Leonardo Ramiro; Corral Rodas,Paul Andres; Ham, Andres; Garriga,Santiago.2018. Lassoing Welfare Dynamics with Cross-Sectional Data (English). Policy Research working paper;no. WPS 8545 Washington, D.C. : World Bank Group. 13 Montes, Jose; Silwal, Ani; Newhouse, David; Chen, Frances; Swindle, Rachel; Tian, Siwei. 2020. How Much Will Poverty Rise in Sub-Saharan Africa in 2020?. World Bank, Washington, DC. © World Bank. Newhouse, David Locke, and Pallavi Vyas, 2022, "Estimating Poverty in India without Expenditure Data: A Survey-to-Survey Imputation Approach." Economic and Political Weekly, Vol 57 no. 48 Stifel, David, and Luc Christiaensen. "Tracking poverty over time in the absence of comparable consumption data." The World Bank Economic Review 21.2 (2007): 317-341. Zhang, Yiyun, Runze Li, and Chih-Ling Tsai. "Regularization parameter selections via generalized information criterion." Journal of the American Statistical Association 105.489 (2010): 312-323. 14 Appendix Appendix: S2S estimation model Step 1: identify a set of variables that are available both in HFPS and GMD. We define the vector of regressors X1 as the variables in Appendix table1 plus the interaction terms. X1 is common to all the countries. Appendix table 1: Level variables available in both GMD and HFPS Variables Label hsize Household size hhsizeSq Household size squared children_share Share of children residents in the household elderly_share Share of elderly residents in the household urban Area of residence (urban/rural) departement subnational administrative area phone Phone Ownership (household level) For each country, � 1 X1 Step 2: In the GMD dataset, we estimate log() = = β GMD + ε and use the “Lasso” GMD GMD method to select a set of regressors, X 2 ⊂ X1 , we ensured that if an interaction term was selected by LASSO, the appropriate level terms were also included in the final model. For instance, if the interaction c.hsize#i.urban is selected by the “Lasso”, the variables hsize and urban should be in 2 . � 2 X2 Step 3: We estimate the model log() = = β GMD GMD + ε, (X 2 being allowed to differ across the countries) Step 4: Predict welfare within the HFPS ̂2 2 � = , using the coefficients of step 3. Appendix table 2: Adjusted R-squared of the S2S estimation model S2S estimation model Country Adj. R-squared Burkina Faso 0.54 Ethiopia 0.36 Gabon 0.36 Ghana 0.43 Gambia 0.43 Mali 0.43 Mozambique 0.43 Malawi 0.53 Niger 0.46 Nigeria 0.55 Senegal 0.55 Sierra Leone 0.56 15 Uganda 0.36 Zimbabwe 0.50 Appendix Table 3: Comparison of distribution of actual and predicted per capita household consumption 16 17 Appendix Figure 1: Comparison of variables between GMD and HFPS (wave 1) by country 18 Appendix Figure 2: Comparison of individual-level variables between GMD and HFPS (wave 1) by country Appendix Figure 3: 19