Policy Research Working Paper 8880 Impact of the West African Ebola Epidemic on Agricultural Production and Rural Welfare Evidence from Liberia Alejandro de la Fuente Hanan G. Jacoby Kotchikpa Gabriel Lawin Poverty and Equity Global Practice June 2019 Policy Research Working Paper 8880 Abstract The 2014–15 Ebola epidemic took a devastating human and finds that higher Ebola prevalence (as proxied by local mor- economic toll on three West African countries, of which tality) led to greater disruption of group labor mobilization Liberia was perhaps the hardest hit. The pathways through for planting and harvest, thereby reducing rice area planted which the crisis affected economic activity in these largely as well as rice yields. Household welfare, measured by per agrarian societies remain poorly understood. To study these capita expenditures spanning two points before and after mechanisms in the context of rural Liberia, this paper links the crisis, fell by more in Ebola prevalent areas with more a geographically disaggregated indicator of Ebola disease intensive rice farming, precisely those areas that were more mortality to nationally representative household survey data adversely affected by agricultural labor shortages. on agricultural production and consumption. The paper This paper is a product of the Poverty and Equity Global Practice. 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 adelafuente@worldbank.org. 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 Impact of the West African Ebola Epidemic on Agricultural Production and Rural Welfare: Evidence from Liberia1 Alejandro de la Fuente2 Hanan G. Jacoby3 Kotchikpa Gabriel Lawin4 Keywords: Ebola, Liberia, Agriculture, Welfare. JEL Codes: Q12, I31, N37 1 The authors are grateful to Andrew Dabalen and Kristen Himelein for their comments on an earlier version. We like to thank Siobhan Murray for her assistance on geo-referencing and mapping; as well as Min. Tolbert Nyenswah, Luke Bawo and Nelson Dunbar from the Ministry of Health in Liberia for providing the Ebola administrative data and graciously helping us understand how they were collected. The findings, interpretations, and conclusions of this paper are those of the authors and should not be attributed to the World Bank Group or to its member countries. 2 World Bank (adelafuente@worldbank.org) 3 World Bank (hjacoby@worldbank.org) 4 Consultant (kotchigab@gmail.com)      1. Introduction Between 2014 and 2015, the Ebola Virus Disease (EVD) wreaked havoc across three West African countries. The epidemic killed more than 11,300 people and infected some 28,600 as it swept through Guinea, Sierra Leone and Liberia.  Of these countries, Liberia was reported by many organizations as the most severely affected. The first case was identified in Lofa County in March 2014. During the next year and a half, more than 10,670 would become infected and over 4,800 would die from the disease. Although Liberia was finally declared free of Ebola in July 2016, concerns about the epidemic’s long-term impacts, as well as its reemergence, remain. In addition to Ebola’s morbidity and mortality toll, there were indirect, yet widespread, economic effects resulting from fear of contagion and from domestic and international disease containment measures. International trade and travel were severely disrupted due to restrictions on individuals from Ebola-affected areas entering many countries. Domestic travel restrictions, including limitations on the movement of people and goods and full quarantines of certain areas, were in place for long periods. Schools and markets were closed. Non-Ebola health services, including maternal health care, declined, as medical facilities were requisitioned to serve as Ebola treatment centers and patients avoided hospitals and clinics for fear of encountering infected persons. Given that Liberia is among the poorest countries in the world, with poverty especially pervasive in rural areas (72% of the rural population is officially classified as poor, compared to 32% in urban areas), limited means were available to cope with a shock of this magnitude.5                                                              5 After an initial delay, hundreds of millions of dollars poured into Liberia as part of the international response effort, widely considered integral to halting the spread of Ebola. 2        In this paper, we seek to unpack the economic effects of EVD, and epidemics more broadly, in agrarian societies. We focus on the availability of agricultural labor for critical and time-sensitive tasks such as planting and harvesting, which normally require a degree of worker coordination and mobilization. Liberia, for example, has a labor recruitment system known as kuu for assembling agricultural workers for seasonal tasks, especially in the production of rice, the principal crop. We consider the hypothesis that fear of congregating in groups disrupted worker mobilization and led to severe labor shortages, thus depressing rice production and, ultimately, rural welfare. To explore this mechanism, we link a geographically disaggregated indicator of Ebola disease mortality with nationally representative household survey data on agricultural production and household welfare. To preview, we find that higher Ebola prevalence (as measured by local mortality) led to greater disruption of group-labor mobilization for planting and harvest and reduced rice area planted as well as rice yields. Consistent with this evidence, household welfare, measured by per capita expenditures, fell by more in Ebola prevalent areas with more intensive rice-farming, i.e., in precisely those areas differentially affected by agricultural labor shortages. Other research, produced during and in the immediate aftermath of the EVD crisis in Liberia, also suggests that serious agricultural labor shortages occurred due to contagion fears. However, these findings are based largely on impressionistic or subjective evidence from non-representative samples.6 Using a simulation model, FAO (2015) estimates that,                                                              6 Among respondents to a World Bank cellphone survey who had already harvested, over 80 percent reported lower production in 2014 than in the previous year, the main reason being the labor disruption due to Ebola (World Bank 2015). This was also the main reason cited by those with crops still in the field as to why the harvest had not been completed. UNDP (2014) finds a decline in the supply of labor due to deaths, the migration of workers to escape the disease, and the unwillingness to engage in collective activities such as land preparation and harvesting. 3        as a result of labor shortages, rice production declined by 12 percent at the national level, with more precipitous declines (up to 20 percent) in the most affected counties such as Lofa and Margibi; cassava production, which is less labor and input intensive than rice, supposedly experienced a 5 percent drop at the national level.7 Our study stands apart from this work by providing the first econometric evidence on the Ebola epidemic’s impact using nationally representative household survey data combined with administrative records on EVD mortality at the district level. More broadly, past literature on the economic impact of infectious disease focuses on long-term consequences of early life shocks operating through the mechanism of human capital formation and, ultimately, through the labor market (Almond 2006; Bleakley 2007, 2010; Baird et al., 2016). There is also a small economics literature focusing on aggregate labor supply effects of epidemic disease (e.g., Young 2005; Voigtländer and Voth 2012). To our knowledge, the mechanism explored in this paper -- fear of contagion -- is novel, albeit relevant only to virulent epidemics, where the (perceived) likelihood of contagion from proximity to an infected individual is significant. The remainder of the paper is organized as follows: Section 2 describes the data and the context of the EVD outbreak in Liberia. Section 3 details the empirical strategy used to assess the Ebola crisis’ impacts on labor, agricultural production, and consumption. Section 4 presents the main findings. Section 5 briefly summarizes the paper.                                                              7 Using a market-chain analysis approach, FAO (2016) shows that the Ebola crisis disrupted the collection and transportation of agricultural production to consumption areas, which in turn reduced farmers’ incomes (higher costs of inputs and lower negotiating power with collectors whose numbers decreased and producer were forced to sell at lower prices than expected).   4        2. Data and Context 2.1 Household survey data Our main data source is the multi-topic Household Income and Expenditure Survey (HIES) collected by the Liberia Institute of Statistics and Geo-Information Services (LISGIS). We use the latest two rounds, 2014 and 2016. Both rounds are representative at the national level and, for rural areas – i.e., localities with population less than 2,000 -- representative at the county level (Liberia is divided into 15 counties). Both surveys also use the same Enumeration Areas (EAs), but fresh households were selected from each EA in each round. The 2014 round of the HIES was planned for the full calendar year, but only administered from January through July due to the outbreak of EVD in the country. When fieldwork was halted in August 2014, approximately half of the data had been collected. Fortunately, the HIES sample covered during the first two quarters of 2014 was designed to be nationally representative. The HIES 2016 started in January of that year --- after which time only a few cases of Ebola were reported --- and continued through December. Both the 2014 and 2016 HIES allow us to construct a household consumption aggregate; we use per capita expenditures to proxy household welfare. Consumption expenditures comprise food and non-food components and are reported in Liberian dollars (see LISGIS 2017 for details). To sum up, our analysis of household welfare impacts will effectively use an EA-level panel between 2014 and 2016, focusing, for purposes of comparability, on households surveyed in the first semester (January-July) of each round. HIES 2016 provides extensive and detailed information on farm labor and agricultural production based on recall of the last completed farming season. The agriculture module also includes questions about the perceived impact of the Ebola crisis 5        on rice farming. Since HIES 2014 does not include a comparable agricultural module, we cannot construct a 2014-16 panel of EAs for the agricultural production data along the lines of our consumption expenditure data. Instead, our analysis of the impact of the Ebola outbreak on farm labor and production outcomes is purely cross-sectional. 2.2 Mortality during the Ebola Outbreak On 30 March 2014, Liberia confirmed its first two cases of Ebola virus disease in Foya, Lofa County. By June the first deaths from Ebola occurred in Monrovia when seven patients died from the disease. Although, at the time, only 16 cases were reported in Liberia in total, with the urbanization of the disease the outbreak quickly spiraled out of control, peaking in the first quarter of 2015. By June 2016, when Liberia was (provisionally) declared Ebola-free, 10,678 people had been infected with Ebola, and over 4,800 had died. Ebola-related burial data obtained from the Ministry of Health up to December 2015 indicate 4,621 total casualties, corresponding reasonably closely to the official figures. Of these deaths, 2,221 were matched to rural areas.8 In rural Liberia, adult males were the main group of mortal victims from Ebola, but close to a third of the fatalities were under-fives (8%) and adults over 65 (24%) (See table A1.2 in Appendix 1).                                                              8  The rural classification was done for georeferenced burials (4257/4621) at the locality level using the official cutoff population of < 2,000 as used in the HIES.    6        Figure 1. Ebola-Related Deaths in Rural Areas, March 2014 – Dec. 2015   300 2500 250 2000 200 <‐‐ HIES 2014        <‐‐ HIES 2016     fieldwork ‐‐> fieldwork ‐‐> 1500 150 1000 100 50 500 0 0 Apr‐14 May‐14 Jul‐14 Oct‐14 Nov‐14 Apr‐15 May‐15 Jul‐15 Oct‐15 Nov‐15 Apr‐16 May‐16 jul‐16 Mar‐14 Jan‐14 Feb‐14 Jun‐14 Aug‐14 Dec‐14 Mar‐15 Mar‐16 Sep‐14 Jan‐15 Dec‐15 Feb‐15 Jun‐15 Aug‐15 Sep‐15 Jan‐16 Feb‐16 Jun‐16 Aug‐16 New deaths (left) Cumulative deaths (right)   Source: Authors with data from Ministry of Health and Social Welfare. We use total deaths (burials) suspected to be Ebola-related in a district per 1,000 district population as a measure of Ebola prevalence in the district.9 Liberia is divided into 130 districts, for which population figures are derived from the (latest available) 2008 Population Census. Based on this measure, Ebola is more prevalent in northern and western Liberia relative to the south and east. FAO (2004) classifies 7 of 15 counties as “severely affected” by Ebola (Bomi, Grand Cape Mount, Lofa and Margibi, Bong, Grand Bassa and Nimba). While, according to our fatality measure (see Figure 2 and Table A1.1 in Appendix 1), Ebola prevalence is not especially high in Nimba nor in Grand Bassa, we will retain the FAO categorization for the purposes of our empirical analysis. Both measures agree that Grand Gedeh and Maryland counties, on the southeast border with Côte d’Ivoire, were by far the least affected by the EVD crisis in relative terms.                                                              9  In a large subset of these cases, health officials obtained mouth swab tests to confirm Ebola virus. We do not restrict deaths to confirmed Ebola cases because the selection of individuals to test was clearly not random (e.g., testing was probably less likely in more remote areas). 7        Figure 2: Ebola-related burials in rural areas Source: Authors with data from Ministry of Health and Social Welfare. Note: The rural classification was done for georeferenced burials (4257/4621) at locality level. Following the official definition, any locality with a population equal or greater than 2,000 and all county capitals were classified as urban. The remainder of the localities were classified as rural. 2.3 Ebola and Agriculture Agriculture constitutes more than a third of Liberia’s economy in terms of GDP (38.8%), employing more than 70% of the population. Liberia’s major crops are natural rubber, rice, cassava, bananas and palm oil. Rice, alongside cassava, is the main staple crop, grown by about three-quarters of rural households. Given its importance for food security and rural living standards more broadly, this study focuses on rice production. Ebola struck Liberia at a critical point during the main rice harvest of the agricultural season 2014-15. As figures 1 and 3 show, the epidemic plateaued well into the land preparation and planting activities for the following agricultural season 2015-16, precisely the season recalled by most respondents to the 2016 HIES. 8        Figure 3. Agricultural Calendar (Rice) and HIES Timeline Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec 15 15 15 15 15 15 15 15 15 15 15 15         Land preparation Main   & planting harvest           Lean         Season       Rainy season Rainy season     South/central South/central                 Rainy season     Northeast  HIES 2016 – Agriculture Survey Module Recall Period Source: The seasonal calendar was adapted from FAO (2016). Rice production in Liberia uses a labor-intensive, rain-fed system involving team- work at certain crucial production stages, namely planting, maintenance and harvest. The kuu system of labor recruitment assembles such work teams, offering members food or labor on their own farms in exchange for work (FAO, 2016). Fear of infection would be expected to reduce farmer willingness to join kuu teams and can explain the impact on rice production when and where Ebola raged. Two channels of impact of the Ebola epidemic on Liberian agriculture are thus as follows: First, there is a direct effect of the Ebola outbreak on the rural labor force through sickness or death, temporarily or permanently removing agricultural workers and thereby disrupting planting and harvesting in agriculture or, more generally, raising farm wages to the extent that a casual labor market exists.10 However, given the magnitude of Ebola morbidity and mortality relative to the size of the rural population (around two fatalities                                                              10  This is the mechanism that economists have explored in the context of HIV in Africa (Young 2005) and the Black Death in Europe (Voigtländer and Voth 2012).  9        out of every thousand), this direct effect is likely to be negligible. Second, agricultural labor force participation could fall indirectly due to measures to contain the virus, such as travel restrictions, and quarantines, but primarily due to the increased fear of congregating in work teams as discussed in the previous paragraph. Lower availability of labor, especially for land preparation and planting, would reduce area devoted to rice and potentially lower yields on rice land that is planted. Subjective recall data from the 2016 HIES on how Ebola affected rice cultivation in the last completed season lends support to the importance of farm labor availability. A year after the onset of the Ebola outbreak (during the agricultural season 2015/16) many farmers in Liberia were unable to grow or sell their rice. The primary constraint was lack of farm labor for clearing, planting and harvesting. Specifically, 9 percent of rural farming households report not planting rice due to the presence of Ebola in their community. Another 14 percent planted rice, but encountered difficulties during cultivation because of EVD. Of the 48 percent of rural farming households planting rice during the course of the epidemic, 22 percent reported that land clearing and planting were affected, whereas 1 percent did not harvest any of the planted rice and 3 percent did not complete their harvest because of Ebola sickness in their community. While these recall data provide some insights, they are, after all, subjective impressions of farmers. The remainder of this study, therefore, focuses on objective recall data, which are less prone to ex-post rationalization and related biases. One limitation of the HIES questionnaire is that it does not distinguish the use of kuu labor from that of (conventional) hired labor, although we suspect that the former institution is far more widespread than the latter in Liberia’s traditional rice farming 10        communities. At any rate, the use of hired/kuu labor is highest in Lofa, Nimba and Bong counties (Table A1.1), known as Liberia’s 'breadbasket'. Some of these counties in northern and western Liberia were also Ebola hotspots, implying a positive spatial correlation between kuu labor and Ebola in the country overall. Indeed, Table A1.3 shows higher shares of hired/kuu labor in counties designated by FAO as severely affected by Ebola. 3. Empirical Strategy 3.1 Labor Use and Rice Production Qualitative and other evidence cited above suggests that, at its height, Ebola caused significant disruption to Liberia’s rice production system. The main channel of impact was the shortage of labor for land clearing and harvesting. The agricultural calendar and timeline (fig. 2) show that the Ebola outbreak had not subsided, or at least continued to be disruptive, at the time of the 2015/16 rice season, the earliest for which recall data are available. Our regression model for rice production outcome of household in district is β ϵ , (1) where is a continuous measure of district-level EVD prevalence based on the total number of Ebola-related burials per 1,000 district population over the course of the outbreak, is a vector of household demographic, enumeration area (EA), and district level characteristics., and ϵ is a random disturbance term. Standard errors are clustered at the district level. 11        As noted, agricultural production, land and labor use are covered in the 2016 HIES, in a one-year recall format. To ensure that farmers are referring to the 2015 agricultural season, we restrict the sample to agricultural households (rural households engaged in farming) interviewed between January-April 2016; i.e., before the 2016 planting season was underway. Our empirical analysis of agricultural production is by necessity cross- sectional, comparing post-Ebola production outcomes between districts differentially affected by the crisis. There are two threats to validity of a causal interpretation of our estimates: (1) Omitted variables correlated with both agricultural production outcomes and the incidence of EVD; and (2) Simultaneity between kuu labor use and Ebola. In other words, where farmers rely more heavily on kuu labor, Ebola could have spread more quickly. Simultaneity bias would lead to a positive association between kuu labor and Ebola, as suggested by the higher kuu share in severely affected counties noted above. To address these issues, we control for whether the household is in a “severely affected” county as defined by FAO, thus focusing on variation within highly affected areas as well as within less affected areas. We also control for distance to major roads from the center point of each Enumeration Area (so all the household records within an EA should have the same value), as Ebola may have spread along road networks that are more likely to be placed where agriculture productivity is particularly high. Finally, the production regressions also control for rainfall patterns11 (total rainfall at Enumeration Area level in 2015) that may have influenced rice planting choices and yields.                                                              11  The rainfall source data were Climate Prediction Center (CPC) Africa Rainfall Climatology Version 2.0 (ARC2).    12        3.2 Consumption Expenditures As noted, the consumption expenditure recall period for the 2014 HIES covers the half- year before the onset of the Ebola epidemic in June 2014, whereas the consumption expenditure recall period for the 2016 HIES (based on the restricted sample of households interviewed in the first semester) covers the half-year starting from January 2016, during which time Liberia was virtually Ebola-free (see figure 1). While the data represent repeated cross-sections of households, for our welfare measure (per capita consumption), we have variation over time, i.e., just before and just after the epidemic. Presumably, household consumption in the first half of 2016 reflects any adverse income shock due to disrupted rice production in 2015. We estimate an EA fixed effect regression of the form: β u , (2) where is household welfare (varying over two points in time t), is an EA fixed effect, is a dummy for 2016 that multiplies all the time invariant covariates: the Ebola prevalence variable introduced in equation (1), the interaction of Ebola prevalence with a measure of community resilience – i.e., the share of agricultural land in the district planted to rice – and a vector of household demographic characteristics including a constant term. 4. Results 4.1. Labor Use and Rice Production Table 1 shows that within the counties that were severely affected by Ebola, use of hired/kuu labor was lower in districts with a particularly high Ebola incidence. While the 13        negative effect is not statistically significant for overall labor use, it is significant for narrower categories of farming activities that are especially hired/kuu labor intensive, namely land clearing and planting. In quantitative terms, multiplying the number of burials per 1,000 population by 10 reduces the intensity of kuu labor in overall labor used for planting by 2.7 percentage points. This disruption in land clearing and planting labor is reflected in a significantly lower share of cultivated land devoted to rice in high Ebola prevalence areas (see Table 2, column 1). Rice yields were also significantly lower in these areas (column 2); a tenfold increase in the number of fatalities per 1,000 population reduces rice yields by 6%.12 As a reality check, we run the same specifications in Tables 1 and 2 using the full sample of farming households, which includes households interviewed later in 2016 and who are thus likely to be reporting on 2016 rice production activities. By 2016, the epidemic in rural Liberia had largely run its course. Thus, as expected, the estimated effects of Ebola mortality on the use of hired/kuu labor and on rice cultivation are substantially attenuated (see Appendix Tables A2.1 and A2.2).                                                                  12  A rapid assessment made by Mercy Cops in November 2014 reported an even larger reduction of rice yields, on the order of 10-25% due to the disruption of the kuu system.    14        Table 1. Ebola Mortality and Agricultural Labor Hired/Kuu labor per total labor for: (1) (2) (3) (4) (5) Hired/Kuu labor per total labor Clearing Planting Management Harvesting Log all burials per 1000 population -0.010 -0.017 -0.027 -0.003 -0.016 (0.010) (0.009)** (0.010)*** (0.012) (0.011) Severely affected counties (dummy) 0.140 0.164 0.152 0.124 0.131 (0.024)*** (0.028)*** (0.026)*** (0.026)*** (0.027)*** Distance weighted to nearest major road (km) 0.000 -0.000 0.000 -0.001 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) Head of household is female 0.033 0.127 -0.010 0.019 -0.022 (0.030) (0.033)*** (0.031) (0.024) (0.032) Household head age -0.000 -0.001 -0.000 0.000 -0.001 (0.001) (0.001) (0.001) (0.001) (0.001) Head education level dummies (ref= no education) Household head has primary education -0.071 -0.050 -0.087 -0.039 -0.043 (0.031)** (0.038) (0.031)*** (0.018)** (0.032) Household head has secondary education -0.034 0.021 -0.049 -0.021 -0.049 (0.027) (0.031) (0.029)* (0.023) (0.028)* Household head has tertiary education 0.080 0.117 0.123 0.129 -0.121 (0.069) (0.066)* (0.081) (0.131) (0.094) Constant 0.328 0.401 0.351 0.119 0.281 (0.039)*** (0.051)*** (0.040)*** (0.035)*** (0.041)*** Mean of dependent variable 0.389 0.508 0.411 0.195 0.331 Observations 3122 3122 3122 3122 3122 Fstat 7.182 11.900 6.969 3.397 4.987 Prob > Fstat 0.000 0.000 0.000 0.001 0.000 Adjusted R-squared 0.061 0.078 0.057 0.039 0.038 15        Source: Authors based on HIES 2016. Note: Standard errors in parentheses clustered at the district level. *** p<0.01, ** p<0.05, * p<0.1. The sample consists of rural farming households interviewed from January to April 2016.     16        Table 2. Ebola Mortality and Rice Production (1) (2) Share of land area allocated to rice Log rice yield [kg/ha] Log all burials per 1000 population -0.014 -0.060 (0.007)** (0.024)** Severely affected counties (dummy) 0.036 -0.177 (0.025) (0.074)** Distance weighted to nearest major road (km) 0.002 0.001 (0.001)** (0.002) Head of household is female -0.008 -0.202 (0.016) (0.065)*** Household head age -0.001 0.003 (0.000)* (0.002) Head education level dummies (ref= no education) Household head has primary education -0.034 -0.056 (0.018)* (0.087) Household head has secondary education -0.041 0.002 (0.016)** (0.065) Household head has tertiary education -0.092 0.069 (0.040)** (0.290) Total Rainfall in 2015 0.000 (0.000)** Constant 0.217 6.409 (0.029)*** (0.172)*** Mean of dependent variable 0.212 6.676 Observations 3122 2387 Fstat 4.122 3.535 Prob > Fstat 0.000 0.001 Adjusted R-squared 0.027 0.027 Source: Authors based on HIES 2016. 17        Note: Standard errors in parentheses clustered at the district level. *** p<0.01, ** p<0.05, * p<0.1. The sample consists of rural farming households interviewed from January to April 2016. 18        4.2. Impacts on Welfare Results reported in columns 1-3 of Table 3 indicate that per capita food consumption and, to a somewhat lesser extent, per capita total consumption fell by more between 2014 and 2016 in areas with higher EVD prevalence, indicating that households did not have the means to completely self-insure against this (income) shock. Recall that our estimation sample consists of two repeated cross-sections of households, which form a panel at the level of the enumeration area. Thus, fixed household characteristics can be included in the EA fixed effect regression. Moreover, the burials variable is assumed to be zero in 2014 (prior to the Ebola outbreak), which is equivalent to interacting burials with a dummy for 2016. Next, we investigate whether households in districts more involved in rice cultivation, and hence more subject to on-farm labor disruptions, might have experienced larger consumption declines. We do so by adding an interaction term between the share of district land under rice (in 2014) and Ebola mortality (columns 4-6). This term attracts a statistically significant negative coefficient, at least in the case of food consumption, suggesting that rice cultivating areas were indeed more vulnerable to the Ebola outbreak. 5. Conclusions Consumption growth in rural Liberia was flat between the 2014 and 2016 waves of the HIES (Appendix 1 Table A1.2). Moreover, rural poverty increased from 70% in the first half of 2014 (just before the Ebola crisis) to 82% in the first half of 2016 following the Ebola crisis (LISGIS, 2017).  The evidence presented in this paper is consistent with the Ebola outbreak being partially responsible for higher poverty in rural Liberia. The 19        mechanism is novel: fear of congregating in groups disrupted worker mobilization and led to severe labor shortages, thus depressing rice production and, ultimately, rural welfare. 20        Table 3. Ebola Mortality and Per Capita Consumption   Log annual per capita consumption (1) (2) (3) (4) (5) (6) Food Non-food Total Food Non-food Total Log (all burials per 1000 population) # Survey year dummy (1 if 2016) -0.035 -0.025 -0.029 0.036 -0.033 0.016 (0.019)* (0.027) (0.016)* (0.049) (0.064) (0.045) Log (all burials per 1000 population) # Share of District Land under rice # Survey year dummy (1 if 2016) -0.322 0.040 -0.203 (0.184)* (0.249) (0.165) Survey year dummy (1 if 2016) -0.056 -0.080 -0.059 -0.076 -0.077 -0.072 (0.028)* (0.060) (0.034)* (0.028)*** (0.063) (0.035)** Head of household is male 0.039 0.168 0.075 0.038 0.168 0.074 (0.020)* (0.046)*** (0.023)*** (0.020)* (0.046)*** (0.023)*** Age of Household head -0.000 -0.001 -0.000 -0.000 -0.001 -0.000 (0.000) (0.000)*** (0.000)*** (0.000) (0.000)*** (0.000)*** Head education level dummies (ref= no education) Primary 0.083 0.196 0.105 0.084 0.196 0.106 (0.025)*** (0.047)*** (0.027)*** (0.025)*** (0.047)*** (0.027)*** Secondary 0.150 0.352 0.204 0.149 0.352 0.204 (0.025)*** (0.038)*** (0.024)*** (0.025)*** (0.039)*** (0.024)*** College degree 0.112 0.823 0.359 0.109 0.823 0.357 (0.069) (0.079)*** (0.070)*** (0.068) (0.080)*** (0.069)*** Log household size -0.294 -0.316 -0.320 -0.293 -0.316 -0.319 (0.020)*** (0.031)*** (0.018)*** (0.020)*** (0.031)*** (0.018)*** Dependency ratio (# hh members aged < 5 / hh size) -0.247 -0.470 -0.313 -0.245 -0.470 -0.312 (0.034)*** (0.097)*** (0.035)*** (0.034)*** (0.097)*** (0.035)*** Share of elderly (# hh members aged>64/ hh size) -0.031 -0.391 -0.128 -0.031 -0.391 -0.128 (0.053) (0.085)*** (0.054)** (0.053) (0.085)*** (0.054)** 21        Survey year dummy (1 if 2016) # Counties severely affected by EVD (dummy) 0.077 -0.053 0.043 0.092 -0.055 0.053 (0.042)* (0.078) (0.044) (0.041)** (0.079) (0.043) Constant 10.609 9.422 10.948 10.609 9.422 10.949 (0.028)*** (0.066)*** (0.030)*** (0.028)*** (0.066)*** (0.030)*** Mean of dependent variable 5.160 0.426 0.550 0.448 0.230 0.366 Observations 5365 5365 5365 5365 5365 5365 Number of clusters 123 123 123 123 123 123 Fstat 39.942 51.174 90.213 36.386 47.367 82.968 Prob > Fstat 0.000 0.000 0.000 0.000 0.000 0.000 Adjusted R-squared 0.183 0.155 0.231 0.185 0.155 0.232 Source: Authors based on HIES 2014 and 2016. Note: Standard errors in parentheses clustered at the district level. *** p<0.01, ** p<0.05, * p<0.1. All models control for Enumeration Area fixed effects and observations are weighted to be representative of the population. 22        References Almond, D. (2006) Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 US population..Journal of Political Economy, 114(4), 672-712. Baird, S., J. Hicks, M. H. Kremer, and E. Miguel. (2016) Worms at work: Long-run impacts of a child health investment. Quarterly Journal of Economics, 131(4), 1637-1680. Bleakley, H. (2007) Disease and development: evidence from hookworm eradication in the American South. Quarterly Journal of Economics, 122(1), 73-117. Bleakley, H. (2010) Malaria eradication in the Americas: A retrospective analysis of childhood exposure. American Economic Journal: Applied Economics, 2(2), 1-45. Bowles, J., J. Hjort, T. Melvin, and E. Werker. (2015) Ebola, jobs and economic activity in Liberia. Journal of Epidemiol Community Health, 70: 271-277   Food and Agriculture Organization of the United Nations (FAO). (2016) Impact of the Ebola virus disease outbreak on market chains and trade of agriculture products in West Africa. Food and Agriculture Organization of the United Nations, Dakar, 2016. Food and Agriculture Organization of the United Nations - FAO. (2015) FAO/WFP Crop and Food Crop Security Assessment – Liberia, Sierra Leone and Guinea. Food and Agriculture Organization of the United Nations - FAO. (2014) FAO/WFP Crop and Food Security Assessment – Liberia. Famine Early Warning Systems Network - FEWS NET. (2017) Ebola and Livelihoods. The effect of the Ebola virus on household livelihoods in Guinea, Liberia, and Sierra Leone. Washington, DC: FEWS NET. Govindaraj, R., Ch. H. Herbst, and J. P. Clark, eds. (2018) Strengthening Post-Ebola Health Systems: From Response to Resilience in Guinea, Liberia, and Sierra Leone. Washington, DC: World Bank. Liberia Institute of Statistics and Geo-Information Services (LISGIS). 2017. “Household Income and Expenditure Survey. Statistical Abstract” August 2017. Mercy Corps. (2014) Economic impact of the Ebola crisis on select Liberian markets: Focus on Monrovia, Lofa, and Nimba Counties. November 4th, 2014. Michler, J., K. Baylis, M. Arends-Keunning, and K. Mazvimavi. (2016) “Conservation agriculture and climate resilience.” Accessed on 9 April 2018 from https://goo.gl/35dv55.   23        UNDP. (2014) Assessing the socio-economic impacts of Ebola Virus Disease in Guinea, Liberia and Sierra Leone – The Road to Recovery. United Nations Development Group (UNDG) – Western and Central Africa. Voigtländer, N. and H. J. Voth. (2012). The three horsemen of riches: Plague, war, and urbanization in early modern Europe. Review of Economic Studies, 80(2), 774-811. World Bank. (2014) The Socio-Economic Impacts of Ebola in Liberia, Results from a High Frequency Cell Phone Survey, November 19, 2014. World Bank, Washington DC. World Bank. (2015a) The Socio-Economic Impacts of Ebola in Liberia, Results from a High Frequency Cell Phone Survey Round 3, January 12, 2015. World Bank, Washington DC. World Bank. (2015b) The Socio-Economic Impacts of Ebola in Liberia, Results from a High Frequency Cell Phone Survey Round 4, February 24, 2015. World Bank, Washington DC. World Bank (2015c) The Socio-Economic Impacts of Ebola in Liberia, Results from a High Frequency Cell Phone Survey Round 5, April 13, 2015. World Bank, Washington DC. World Bank. (2018) Republic of Liberia - Systematic Country Diagnostic. From Growth to Development: Priorities for Sustainable Poverty Reduction and Achieving Middle Income Status by 2030. Report No. 113720-LR.   Young, A. (2005). The gift of the dying: The tragedy of AIDS and the welfare of future African generations. Quarterly Journal of Economics, 120(2), 423-466. 24        Appendix 1. Descriptive Statistics Table A1.1. Descriptive Statistics, by Counties County Grand Cape Mount Grand Gedeh Montserrado Grand Bassa River Cess Grand Kru River Gee Maryland Gbarpolu Margibi Nimba Sinoe Bomi Bong Lofa Ebola Fatalities 230 298 130 449 20 35 231 261 23 180 106 26 59 96 77 Fatalities per every 1000/district inhabitants 4.42 1.77 0.84 4.86 0.20 1.13 1.72 3.65 0.38 1.73 0.54 1.13 1.17 0.77 1.57 Farm Labor Hired/Kuu labor per total labor 0.30 0.49 0.30 0.37 0.36 0.27 0.58 0.35 0.31 0.20 0.44 0.27 0.31 0.25 0.44 Hired/Kuu labor per total labor for clearing 0.44 0.62 0.46 0.46 0.50 0.34 0.65 0.48 0.43 0.28 0.59 0.43 0.39 0.38 0.54 Hired/Kuu labor per total labor for Planting 0.30 0.52 0.35 0.36 0.39 0.30 0.59 0.32 0.34 0.17 0.49 0.31 0.36 0.28 0.43 Hired/Kuu labor per total labor for farm mgmt 0.08 0.29 0.11 0.19 0.09 0.14 0.43 0.16 0.16 0.08 0.20 0.05 0.12 0.10 0.11 Hired/Kuu labor per total labor for harvesting 0.15 0.46 0.25 0.32 0.33 0.24 0.54 0.26 0.20 0.16 0.37 0.21 0.25 0.17 0.43 Rice production Share of area devote to rice 0.12 0.30 0.17 0.19 0.33 0.21 0.24 0.18 0.15 0.02 0.24 0.17 0.22 0.29 0.29 Rice yield (kg/ha) 649 1063 1305 865 2222 1229 1218 1187 1439 536 1471 1731 1552 2058 1298 Source: Authors with data from Ministry of Health and Social Welfare for Ebola statistics; and with data from Agriculture Recall Survey from HIES 2016 for labor and production statistics. 25        Table A1.2. Descriptive Statistics for Rural Samples Rural 2014 2016 (SE) (SE) Dependent variables Annual per capita total consumption (LD) 46352 46682 (29282) (69672) Annual per capita total food consumption (LDs) 32487 33057 (17442) (18365) Annual per capita total nonfood consumption (LD) 13865 13625 (17469) (61052) Farm Labor Hired/Kuu labor per total labor 0.39 (0.29) Hired/Kuu labor per total labor for clearing 0.51 (0.34) Hired/Kuu labor per total labor for Planting 0.41 (0.34) Hired/Kuu labor per total labor for farm management 0.20 (0.31) Hired/Kuu labor per total labor for harvesting 0.33 (0.33) Rice production and price Share of area devote to rice 0.21 (0.23) Rice yield (kg/ha) 1198 (1414) Ebola Total Ebola-related fatalities 2,221 Age 0-4 174 5-15 170 16-64 1,330 65 and above 538 Sex Female 975 Male 1,245 Ebola-related fatalities per every 1000/ district inhabitants 1.86 Household demographics Highest education level attained by household head None 0.48 0.50 (0.50) (0.50) Primary 0.17 0.18 (0.38) (0.38) 26        Rural 2014 2016 (SE) (SE) Secondary 0.32 0.31 (0.47) (0.46) College degree 0.03 0.01 (0.17) (0.12) Age of household head (years) 49 44 (73) (27) Household head is male 0.76 0.75 (0.43) (0.43) Community/Location Characteristics Distance weighted to nearest major road (km) 14 14 (15) (15) Distance to border with Guinea (km) 121 121 (83) (83) Distance to border with Ivory Coast (km) 143 143 (81) (81) Distance to border with Sierra Leone (km) 172 172 (117) (117) Source: Authors based on HIES 2014 and 2016. Note: LD = Liberian Dollars.       27        Table A1.3. Descriptive Statistics, Severely Affected and Less Affected Counties from Ebola Severely Affected Less Affected 2014 2016 2014 2016 (SE) (SE) (SE) (SE) Dependent variables Annual per capita total consumption (LD) 42,662 44,937 52,989 50,659 (23,795) (81,158) (36,244) (29,796) Annual per capita total food consumption (LD) 30,169 31,450 36,656 36,720 (14,940) (16,944) (20,571) (20,804) Annual per capita total nonfood consumption (LD) 12,493 13,487 16,333 13,939 (13,899) (72,565) (22,304) (14,992) Farm Labor Hired/Kuu labor per total labor 0.43 0.29 (0.29) (0.28) Hired/Kuu labor per total labor for clearing 0.55 0.40 (0.32) (0.34) Hired/Kuu labor per total labor for planting 0.45 0.31 (0.34) (0.33) Hired/Kuu labor per total labor for farm management 0,23 0.10 (0.33) (0.23) Hired/Kuu labor per total labor for harvesting 0.37 0.24 (0.34) (0.31) Rice production and price Share of area devote to rice 0.22 0.19 (0.24) (0.22) Rice yield (kg/ha) 1,087 1,571 (1,291) (1,716) Imported rice price (LD/kg) 62 68 73 75 (12) (15) (12) (19) Local rice price (LD/kg) 61 67 67 64 (25) (33) (10) (13) Ebola Total Ebola-related fatalities 1,704 516 Age 0-4 146 28 5-15 138 32 16-64 1015 315 65 and above 404 134 Sex Female 753 222 Male 951 294 Ebola-related fatalities per every 1000/district inhabitants 2.15 1.19 28        Severely Affected Less Affected 2014 2016 2014 2016 (SE) (SE) (SE) (SE) (2.05) (1.84) Household demographics Highest education level attained by household head None 0.51 0.52 0.44 0.46 (0.50) (0.50) (0.50) (0.50) Primary 0.16 0.18 0.18 0.16 (0.37) (0.39) (0.39) (0.37) Secondary 0.31 0.28 0.33 0.37 (0.46) (0.45) (0.47) (0.48) College degree 0.02 0.01 0.04 0.01 (0.15) (0.12) (0.20) (0.11) Age of household head (years) 51 44 45 44 (83) (26) (48) (29) Household head is male 0.76 0.74 0.75 0.75 (0.43) (0.44) (0.43) (0.43) Community/Location Characteristics Distance weighted to nearest major road (km) 13 13 17 17 (12) (12) (19) (19) Distance to border with Guinea (km) 88 88 196 196 (63) (63) (72) (72) Distance to border with Ivory Coast (km) 149 149 130 130 (81) (81) (78) (78) Distance to border with Sierra Leone (km) 131 131 261 261 (72) (72) (146) (146) Source: Authors based on HIES 2014 and 2016. Note: LD = Liberian Dollars               29      Appendix 2: Ebola-related Burials and Selected Agriculture Labor Outcomes – Full HIES 2016 sample   Table A2.1: Ebola-Related Burials and Selected Agriculture Labor Outcomes – Full HIES 2016 sample Hired/Kuu labor per total labor for: (1) (2) (3) (4) (5) Hired/Kuu labor per Clearing Planting Management Harvesting total labor Log all burials per 1000 population -0.006 -0.009 -0.017 -0.007 -0.007 (0.009) (0.009) (0.010)* (0.011) (0.011) Severely affected counties (dummy) 0.130 0.146 0.133 0.116 0.111 (0.023)*** (0.028)*** (0.025)*** (0.023)*** (0.026)*** Distance weighted to nearest major road (km) 0.000 0.000 0.001 -0.001 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) Head of household is female 0.034 0.119 -0.000 0.033 -0.014 (0.028) (0.032)*** (0.029) (0.022) (0.030) Household head age -0.000 -0.001 -0.001 0.000 -0.001 (0.001) (0.001) (0.001) (0.001) (0.001) Head education level dummies (ref= no education) Household head has primary education -0.057 -0.034 -0.071 -0.033 -0.031 (0.029)** (0.036) (0.029)** (0.017)** (0.031) Household head has secondary education -0.036 0.018 -0.044 -0.017 -0.056 (0.024) (0.028) (0.027)* (0.021) (0.024)** Household head has tertiary education 0.103 0.149 0.190 0.094 -0.146 (0.059)* (0.058)** (0.081)** (0.122) (0.084)* Constant 0.317 0.393 0.348 0.107 0.274 (0.039)*** (0.052)*** (0.041)*** (0.029)*** (0.041)*** Mean of dependent variable 0.383 0.496 0.404 0.189 0.323 29        Observations 3389 3389 3389 3389 3389 Fstat 7.227 9.871 6.629 3.854 4.382 Prob > Fstat 0.000 0.000 0.000 0.000 0.000 Adjusted R-squared 0.051 0.060 0.042 0.035 0.028 Source: Authors based on HIES 2016. Note: Standard errors in parentheses clustered at the district level. *** p<0.01, ** p<0.05, * p<0.1. The sample consists of rural farming households interviewed from January to December 2016. 30        Table A2.2: Ebola-Related Burials and Selected Agriculture Production Outcomes – Full HIES 2016 sample (1) (2) Share of land area allocated to rice Log rice yield [kg/ha] Log all burials per 1000 population -0.011 -0.025 (0.007) (0.028) Severely affected counties (dummy) 0.031 -0.284 (0.023) (0.078)*** Distance weighted to nearest major road (km) 0.002 0.001 (0.001)** (0.003) Head of household is female -0.013 -0.140 (0.017) (0.059)** Household head age -0.001 0.003 (0.000)* (0.003) Head education level dummies (ref= no education) Household head has primary education -0.037 -0.005 (0.019)* (0.073) Household head has secondary education -0.049 -0.000 (0.014)*** (0.063) Household head has tertiary education 0.027 -0.161 (0.116) (0.436) Total Rainfall in 2015 0.000 (0.000) Constant 0.221 6.698 (0.029)*** (0.225)*** Mean of dependent variable 0.212 6.591 Observations 3389 2560 Fstat 3.080 2.753 Prob > Fstat 0.003 0.006 Adjusted R-squared 0.024 0.022 Source: Authors based on HIES 2016. 30        Note: Standard errors in parentheses clustered at the district level. *** p<0.01, ** p<0.05, * p<0.1. The sample consists of rural farming households interviewed from January to December 2016. 30