Beneficiary Views on Cash and In-Kind Payments: Evidence from Ethiopia's Productive Safety Net Programme

Abstract

1. Introduction

Social protection programs that seek to reduce poverty are now widespread (ILO 2014). In Africa alone, the number of social protection programs has tripled in the last 15 years (Cirillo and Tebaldi 2016) to the extent that today all African countries operate at least one such program (Beegle, Honorati, and Monsalve 2018). Alongside their introduction have been discussions regarding technical aspects of their implementation with issues such as targeting, the level and frequency of payments, the duration of these programs, whether receipt of benefits should be conditioned on specified behaviors, and so on A key design issue relates to their payment modality: cash or in-kind. While Bastagli et al. (2016) report that approximately 130 low- and middle-income countries have at least one cash transfer program, Alderman, Gentilini, and Yemtsov (2017) note that the use of in-kind payments such as food in social programs remains widespread.

From the perspective of program designers, cash transfers have advantages with respect to timeliness of delivery and reduced delivery costs (Gentilini 2007; Margolies and Hoddinott 2015). Despite these advantages, Aker (2017) shows that in-kind transfers such as food remain widespread for reasons including paternalism (Currie and Gahvari 2008; Cunha 2014), targeting (Nichols and Zeckhauser 1982; Moffitt 1989; Bearse, Glomm, and Janeba 2000; Pirttilä and Tuomala 2002; Currie and Gahvari 2008), the need to supply goods not available in local markets (Aker 2017), and the sense that these are more politically acceptable than cash transfers (Epple and Romano 1996). Additionally, cash transfers are seen to provide beneficiaries with choice. Building on the theoretical work by Southworth (1945), transfers are infra-marginal for some recipients: the amount of the in-kind transfer is less than what the household spends on the transferred good after the transfer. These recipients will be indifferent between cash and in-kind transfers. For others, in-kind transfers are extra-marginal: they distort consumption toward the transferred goods – unless the in-kind transfers can be converted into cash. Based on these micro-economic insights, economists typically default to the assumption that program beneficiaries would rather receive cash than an in-kind transfer because the cash transfers do not constrain their choices (Currie and Gahvari 2008). Thurow (1974) goes as far as stating that “the general economic case for cash transfers is strong enough that the burden of proof should always lie on those who advocate restricted transfer (p. 195).”

Yet, while there is a growing empirical literature on the impacts of cash and food transfers in developing countries – see Gentilini (2016) and GAO (2016) for summaries – the evidence base supporting the assertion that beneficiaries would rather receive cash to food or other in-kind transfers is surprisingly thin.¹ Based on focus group discussions held in eight countries, Berg, Mattinen, and Pattugalan's (2013, p. 52) report “an overwhelming preference for cash over in-kind assistance.”² But quantitative studies describe more varied responses. Ethiopian data from the late 1980s showed that across three rural localities, 19, 83, and 90 percent would have liked to have received food rather than cash payments (Webb and Kumar 1995). Predilections for in-kind transfers are also reported in surveys conducted in India (Khera 2013), Niger (Hoddinott, Sandström, and Upton 2013), and, to a lesser extent, Afghanistan (UNHCR 2015).

Why, if given a choice, would many program beneficiaries rather receive an in-kind transfer instead of cash? Southworth (1945), along with more modern restatements such as Currie and Gahvari (2008) and Cunha (2014) make clear that their models are predicated on an assumption that cash and in-kind transfers are equal valued. But in practice, this may not hold as equivalent value can be affected by: the purchasing power of the cash transfer (e.g., given prices of purchased goods, how much food can be purchased with a cash transfer); the potential resale value of in-kind goods (if the in-kind transfer could be sold, how much cash would be received); the transactions costs associated with receiving a cash or in-kind payment; and the transactions costs associated with using a cash or in-kind payment (on this, see Doocy and Tappis 2017). Beneficiaries might also be concerned about price volatility. Suppose a household can chose between a food transfer and a cash transfer. Food prices might be either “high” or “low”; averaging over these, the food and cash transfer are of equivalent value. But in terms of food consumption, for risk-averse beneficiaries, the expected utility of the certainty of food transfer is higher than the expected utility of the cash transfer. There may also be behavioral or psychological factors at work. Beneficiaries might wish to receive payments in cash if in-kind transfers are perceived to be stigmatizing (Moffitt 1983). Receiving in-kind payments can be seen as a commitment device. As part of the debate regarding replacing India's food transfer system (the Public Distribution System, PDS) with cash payments, Jean Dreze has noted PDS participants “are worried that cash can be more easily misused than food, because food can only be consumed in small quantities over time while cash is easily spent in one go.”³ Lastly, if questions regarding beneficiaries’ views about payment modalities are asked in the context of an ongoing program, responses might reflect a concern that giving the “wrong” response might lead to the transfer program being terminated (Berg, Mattinen, and Pattugalan 2013).

Few studies provide systematic empirical evidence on why households might rather receive in-kind rather than cash transfers.⁴ This lacuna motivates this study, which uses unique data well suited to analyzing this issue; data drawn from evaluations of Ethiopia's Productive Safety Net Programme (PSNP). The PSNP is the second-largest safety net program in Africa reaching about eight million chronically food-insecure people. A feature of the program is that its beneficiaries receive their payments either in cash or food – or in some combination of both. Further, the share of cash and food payments in total payments have varied over the years since the program began in 2005. Household and locality panel data were collected biannually over five survey rounds fielded between 2006 and 2014. In each round, beneficiary households were asked whether they would rather receive payments in cash or in-kind (food). These data make it possible to describe predilections for cash and for food, show how these change over time, and allow an assessment of associations between them and household and locality characteristics. Ethiopia is an especially appropriate country to situate such a study as, since the beginning of the PSNP, there has been an active debate on whether the payments should be made in cash or in kind. Thus, while speaking to the international literature on this topic, this study also contributes to this local policy debate in Africa's second-most-populous country.

2. Context and Data

This section provides contextual information. It describes the data used in the paper including information on sampling, household and market characteristics, PSNP payments, market prices, and beneficiary views on cash and food payments.

Ethiopia's Productive Safety Net Programme

The PSNP began in 2005 in response to continual appeals for emergency food assistance, widespread food insecurity, and concerns that repeated droughts were leading to asset depletion (Slater and McCord 2013; Devereux et al. 2014). Starting in the highland regions of Ethiopia (Amhara, Oromia, Southern Nations, Nationalities and Peoples' Region, and Tigray), the program was expanded (after initial piloting in some districts) to the agro-pastoral regions of Afar and Somali in 2010.⁵ The PSNP provides transfers to food-insecure households in chronically food-insecure localities while also attempting to prevent asset depletion at the household level and creating assets at the community level (Government of Ethiopia 2004, 2010). Unlike the annual emergency appeals that it largely replaced, the PSNP was conceived as a multiyear program designed to provide recipients with predictable and reliable transfers. Most beneficiary households receive payments for undertaking public works. A small proportion of beneficiaries (largely households with elderly or disabled members) receive unconditional payments called direct support, and a few receive both. Public works payments typically account for 80–85 percent of all PSNP payments (Berhane et al. 2015). Most PSNP activities and transfers are made during the nonagricultural season, the months Tir to Sene in the Ethiopian calendar (approximately mid-January to mid-July). The PSNP uses a mix of geographic- and community-based targeting to identify chronically food-insecure households in chronically food-insecure woredas (districts). Multiple evaluations of the program have shown that it is well-targeted (Coll-Black et al. 2012) and has reduced household food insecurity and distress sales of assets (Berhane et al. 2013, 2015).

Payments are made to beneficiaries in the form of food or cash. The PSNP's Program Implementation Manual (Government of Ethiopia 2004, 2010, 2014) provides details of how cash and/or food allocations are determined. Planning for a given year's operation begins with the development of a woreda food-security plan. This includes the number of beneficiaries, the amount of cash and food requested, and the timing of payments. An initial goal of the PSNP was to shift the financing of the program from food aid to cash (Government of Ethiopia 2004, 2014), motivated in part by the hope that cash transfers would create positive spill-over effects to small-holder farmers, stimulating local agricultural production and food markets.⁶Woredas are encouraged to request payments in cash if there are local food markets with food available and when they have the capacity (e.g., finance officers, cashiers, safes) to disburse cash (Government of Ethiopia 2004, 2014).⁷ These requests are first passed to region officials for review and then to the Federal Food Security Coordination Directorate (FCSD) for further review and approval. The FCSD decides on the cash-food split for each region based on region requests and the predicted availability of food and cash; this availability reflects the level of multiyear (or in some instances, annual) commitments made by the international donors who provide cash (such as the World Bank and the UK's Department for International Development) to the program. It also produces a “food flow requirements” analysis indicating when food will be made available to woredas for payment (a similar plan is made for cash payments). Once these plans are finalized, the decision to provide food and/or cash is effectively locked in for the upcoming year.⁸ Food payments consist of three kilograms of cereals (usually wheat or maize) for each day worked (Government of Ethiopia 2010). Cash payments are intended to be approximately equal to the value of the food payments. When the PSNP began operations, cash payments were set at a uniform six birr per day, but by 2015 they had risen to 14–18 birr per day in nominal terms (Berhane et al. 2015).

Data: Households

Since 2006, Ethiopia's Central Statistical Agency has administered a large household panel survey in localities in which the PSNP operates. The initial sample was based on a stratified sampling. Districts (woredas) were randomly selected proportional to their size from a list of chronically food-insecure woredas stratified by region where the PSNP was operating in 2006. Within each woreda, enumeration areas (EAs) were randomly selected from subdistricts (kebeles) where the PSNP was operating. Within each EA, 15 beneficiary and 10 nonbeneficiary households were sampled from separate lists for each group, yielding 25 households per EA. This generated a sample of 146 EAs and, because a few sampled households were not interviewed, 3,688 households (Gilligan, Hoddinott, and Taffesse 2009). These same households were followed biannually until 2014. In each survey round, the interviews took place during the lean season (roughly late May to early August). The annual attrition rate across the five survey rounds about 1.9 percent per year. Much of this attrition is due to kebeles being dropped where the PSNP ceased operating. Work investigating whether potential differences in attrition rates can be attributed to differences in baseline characteristics shows that being a program beneficiary was not correlated with the probability of attrition. Older and smaller households were slightly more likely to attrite than other household types, but the correlation between attrition and these characteristics was small (Berhane et al. 2013).

Program retargeting was carried out periodically, and as a result many households that were nonbeneficiaries became beneficiaries by the time later survey rounds were fielded. In addition, new households were periodically added to the sample to reflect the geographic expansion of the program but also because of the decreasing share of nonbeneficiary households in the original sample. In this paper, the sample is restricted to households that benefitted from the program at the time of the interview – or at any time in the past. This maximizes the sample size while ensuring that the sample is formed of households that have a direct experience with the PSNP. After dropping households with missing observations, an unbalanced panel dataset remains, consisting of 11,989 observations from 3,649 households.⁹

In addition, surveys were conducted at the kebele level using structured focus group discussions with kebele leaders and other people who are knowledgeable about the locality. The information collected in these discussions includes kebele characteristics such as infrastructure and population size, and aspects of PSNP implementation. As a part of this survey, retrospective price data were collected on monthly prices on grains consumed by households in that locality. The informants were also asked to estimate the distance to the nearest food market and bank; distributions of these are reported in figs. S1.2 and S1.3 in the online appendix.

This population was poor and food insecure. Mean monthly per capita consumption in 2006 was 133 birr per capita or, approximately |${\$}$|0.49 US per person per day. Food accounted for 82 percent of consumption. The Government of Ethiopia measures food security using a construct called the “food gap.” This was measured by asking survey participants to report the number of months, out of the preceding 12 months, when they had “problems satisfying the food needs of the household” with a month where the household had “problems satisfying food needs” being defined as one where the household experienced hunger for five or more days. Table 1 shows that only 37 percent of households in the sample report being fully food secure (i.e., food gap is zero) at any point across the five survey rounds. The mean food gap is 2.4 months. More than three-quarters of household heads (77.9 percent) have never attended school, and just over a quarter (28.7 percent) are female headed. They are physically isolated. Only about half of the sample live within three kilometers of a food market, and just 5 percent live within three kilometers of a bank. While the sample is evenly spread across the four regions (Tigray, Amhara, Oromia, SNNP), there is a slightly lower proportion of observations in earlier rounds (2006, 2008) compared to the later rounds (2012, 2014).

Data on receipt of PSNP payments were recorded during the survey. Enumerators recorded payment data directly from “client cards,” a document held by PSNP beneficiaries that lists their monthly payments. When these cards were not available, respondents were asked to recall their payments by month. Payment information includes whether the payment is made as cash or as food – and if food, what type of food – as well as the total amount of payment received. Comparisons of payment data based on client cards and from self-reports show that levels, distributions, and timing of payments are comparable. Further, limited access to PSNP payroll data confirms that the self-reported data correlate with the payroll data (Berhane et al. 2013). By converting the quantities of food payments received by beneficiaries using the market price data, it is possible to calculate the share of payments made in food and in cash. In 2006, cash payments accounted for about 60 percent of all payments.¹⁰ This increased to 71 percent in 2008, fell in 2010, but rose thereafter with cash accounting for 73 percent of the value of all payments in 2014. At the household level, 85 percent of the households in the sample report having received at least one food payment, and 89 percent report receiving at least one cash payment during the study period. Consequently, 75 percent of the households have direct experience with both cash and food payments. Reselling of food transfers is rare. In 2016, 93 percent of the households that received food transfers reported that they never sold any of them, and most of the remaining 7 percent did so rarely (Hoddinott et al. 2018).¹¹

Starting in 2010, the household surveys began collecting more detailed information on beneficiaries’ experiences associated with obtaining their payments. In order to receive their payments, beneficiaries are instructed – typically with 1–2 days notice – to go to a “pay point,” nearly always a roadside location. These are supposed to be sited in such a way that beneficiaries do not have to travel more than three hours (one way) from their home. In practice, however, logistical considerations – most notably road conditions, which determine where and how far vehicles carrying payments can travel – play a role in the siting of these pay points. Table 2 describes the transactions costs associated with receiving food and cash payments by year.¹² While there is some variability in these data from round to round, the following features emerge. First, generally pay points for food are farther away than those for cash, reflecting the difficulties that heavy goods vehicles face when carrying several tons of grains to remote areas. Second, irrespective of transfer modality, virtually all beneficiaries walk to the pay points. Third, because it takes longer to get to a pay point distributing food, and because it takes even longer to get home again, a significant proportion of food beneficiaries are forced to sleep over at the pay point. Consequently, fourth, beneficiaries reporting that they received food payments are more likely to report having spent money on food and accommodation. But the magnitudes of these costs are small. In 2014, the average food recipient spent 17.6 birr (or 4 percent of the average monthly transfer)¹³ on transaction costs while the cash recipient spent 10.4 birr (or 2.3 percent of the average monthly transfer). Collectively, these four points suggest that there are higher transaction costs associated with receipt of food payments, but these are time, not monetary, costs. Once at the pay point, food and cash beneficiaries report similar experiences in terms of how they were treated and their payments and relatively few reported being harassed or robbed when travelling home after receiving payment.

Data: Markets and Prices

Hoddinott et al. (2018) show that most PSNP kebeles in the highlands are served by periodic markets; only 13 percent have a food market operating daily. The markets are typically situated on good (all-weather) roads and have access to electricity and cell phone coverage. Grains are available in all markets, most offer some fruit and vegetables, but animal-source foods are not always found. These markets are large, with 72 percent having 50 traders or more. But they are relatively distant from urban centers; on average, the nearest city with a population of 20,000 people or more is 46 km away (median: 40 km).

Previous research on PSNP finds that in years of high inflation the real value of the unindexed cash payments lost value relative to food payments (Sabates-Wheeler and Devereux 2010). Given the lower real value of the cash payments, high food inflation should shift infra-marginal households towards wishing that they could receive food payments. The challenge is to construct a meaningful measure of inflation that is specific to the PSNP localities where the data were collected. To address this, note that each survey round includes 12 months recall data on the prices paid by consumers for staples. This information can be used to construct a cereal price index, specifically a weighted average of prices of the six main cereals (maize, teff, barley, wheat, sorghum, and millet) in each community in each survey round. The weights are based on the consumption shares of each cereal type in the community and derived from the data collected in the food consumption module of the household questionnaire. Applying weights from the first (2006) and last (2014) survey rounds, a Fisher Ideal price index is constructed (see Deaton and Tarozzi 2000) that captures both the temporal and spatial differences in price levels. In the supplementary online appendix, appendix S2, “Price Index,” provides more information on this cereal price index; fig. S2.1 shows how the index varied over the survey rounds in each region.

This cereal food price index can be used to adjust the nominal daily public work wages to assess their purchasing power over time. Figure 1 shows how the real daily wage rates (expressed in 2014 birr) have fluctuated over time. The most dramatic is the reduction in the purchasing power of the PSNP Public Works wage between 2006 and 2008, a consequence of the massive food price spike that occurred across Ethiopia (and indeed, around the world) in 2008. While the PSNP wage had been increased from 6 to 8 birr per day, this was nowhere near sufficient to offset the increase in cereal prices.

Figure 1.

Real Value of PSNP Public Works Wages, by Round and Region

Source: Productive Safety Net Programme evaluation surveys 2006–2014.

Note: Public works wages have been deflated using a food price index. This is a weighted average of prices of the 6 main cereals (maize, teff, barley, wheat, sorghum, and millet) in each community in a given year capturing both the temporal and spatial differences in price levels (see section 2 for more information).

Open in new tabDownload slide

Data: Beneficiary Views on Cash and Food Payments

Lastly, and crucially for this analysis, in each round PSNP beneficiary households were asked the following question:¹⁴ “Given a choice, what proportion of your payment would you like to receive in cash and what proportion would you like to receive in kind?” The response options were: “All cash”; “75% cash; 25% food”; “50% cash; 50% food”; “25 percent cash; 75 percent food”; and “All food.” Even though most PSNP payments are paid in cash, and even though the (temporal) transaction costs associated with food payments are higher than payments received as cash, the overwhelming majority of the beneficiary households indicated that they wanted to receive payments either only or partly in food (table 3). On average, across rounds, only 12 percent of the households wanted cash-only payments. More than half would have liked that 75 percent or more of their payments be made in food (table 3; also see first row of table 1), the dependent variable used in the econometric analysis below. Table 3 also suggests that these stated preferences are not stable over time. For example, in 2008 stated preferences shifted toward food-type payments. As noted above, with the purchasing power of the wage falling by 50 percent or so (see fig. 1), it is perhaps not surprising that in 2008 there is a sharp shift in household preferences away from cash toward food payments (table 3). These two observations – that few PSNP beneficiaries state that they would prefer only cash payments and that preferred shares of payments in food or cash are not stable over time – form the first two findings of this paper.

3. Econometric Approach

4. Results

This section provides results. It begins with basic results before considering robustness checks, interaction terms, and household fixed effects. It also discusses the idea of food payments as commitment devices.

Basic Findings

Basic results are found in table 4. Column (1) focuses primarily on household characteristics. Controlling for time and region fixed effects, food-secure households are 3.1 percentage points less likely to state that they prefer receiving payments in the form of food. Stated preferences for food rise with household size (but at a diminishing rate) and with the age of the household head. The associations between preferences for food and the schooling levels or the sex of the household head are not statistically significant. The coefficient on the year dummy for 2008 equals 0.212, indicating that relative to the omitted category (2006, the year of the first survey) preferences for food jumped by more than 20 percentage points during the height of the food price crisis. Considering that 52.4 percent of the households in the sample (table 1) prefer food payments, this translates into a 40 percent change in payment modality preferences.

Column (2) adds variables denoting households’ access to food markets and financial services (banks). Living within three km of a food market reduces preferences for food by 2.8 percentage points. Living within three km of a bank has a similar association, reducing preferences for food by 3.6 percentage points.

Column (3) adds the food price index. The coefficient is positive and statistically significant. A one-unit increase in food price index is associated with a 2.3 percentage point increase in the likelihood of stating a preference for food payments. While this magnitude may seem small, the cumulative effect of food prices, food and financial services access and food security is large. Consider a food-secure household living within 3 km of both a food market and a bank where food prices are 1 standard deviations¹⁵ lower than elsewhere. Ceteris paribus, these joint characteristics shift preferences towards cash by 12.3 percentage points. Considering that 52.4 percent of the households in the sample (table 1) prefer food payments, this translates into a 23 percent shift in preferences toward cash.

Finally, column (4) shows that the main findings do not change if variables capturing the transfers received by the households are excluded from different forms of the model.

Robustness: Modifying the Outcome Variable

The next step is to explore whether these findings are robust to alternative econometric techniques. First, using a logit model instead of a linear probability model yields nearly identical point estimates and marginal effects; see table S1.1 in the supplementary online appendix.

Second, as the selection of the cut-off point in the outcome variable may seem arbitrary, table 4 is re-estimated, replacing the outcome variable with a binary variable that obtains a value 1 if the household prefers 100 percent food; 75 percent food or 50 percent both, and 0 otherwise. Table S1.2 in the supplementary online appendix shows that the results are robust to defining the outcome variable in this way.

Third, the decision to collapse the categorical outcome variable into a binary one discards potentially useful information. This is addressed by using ordered logit models that are more suited to analyzing ordered outcome variables. Table S1.3 in the supplementary online appendix reports results based on an ordered logit model. The estimated marginal effects are in line with what is reported in column (3) of table 4. Food-secure households are less likely to state that they prefer food and more likely to prefer cash. Access to food markets and banks is associated with a shift in preferences away from food to cash payments. Finally, increases in food prices are associated with a shift in preferences away from cash to food payments.

Interaction Terms

Do the associations documented above differ between food-secure and food-insecure households? This is explored by interacting household's food security status with their access to markets, banks, and food price index. Many of these interaction terms appear with insignificant coefficients (table S1.4 in the supplementary online appendix). The exception is the interaction on the food price index; the coefficient on the interaction term is negative, and the joint significance test cannot reject the null that the coefficient is equal – in absolute terms – to the coefficient on the un-interacted price index variable (Wald test; p = 0.115). This implies that, in contrast to food-insecure households, the stated preferences of food-secure households are less influenced by higher (or lower) food prices.

Household Fixed Effect Models

Table 5 presents the results when specification (2) is estimated. The coefficient on the price index is nearly identical what was obtained in table 4 and thus not sensitive to the inclusion of household fixed effects.

Food Payments as Commitment Devices

Some households might prefer food payments rather than cash because – absent the sale of the transfer – food pre-commits the households’ consumption of the transfer.¹⁶ Although the data in the 2006–2014 household panel survey do not allow this to be tested formally, the household survey fielded as part of the evaluation of the current (PSNP IV) phase of the program includes a question asking why households prefer food (cash). Their responses are shown in table 6.

Consistent with the results shown in table 5, relative prices are the dominant reasons, 59.0 and 51.1 percent, why respondents prefer at least some of their payment in food (cash). Respondents who stated that they preferred cash payments also noted the ability to exercise choice that cash provides; 22.3 percent noted that cash allows them buy foods that are not part of the PSNP payment package, and 10.1 percent stated that cash payments mean that they can buy nonfood items or save the money for future investments. Transactions costs (money, time) to collect payments are relatively unimportant. Consistent with the idea that food transfers are a form of commitment device, 21.8 percent of respondents who stated that they preferred food payments, did so because they were concerned that cash would be more easily misused.

6. Conclusions

Economists often default to the assumption that cash is always preferable to an in-kind transfer. Do beneficiaries feel the same way? This paper uses longitudinal household data from Ethiopia where a large-scale social safety net intervention – the PSNP operates. Even though most PSNP payments are paid in cash, and even though the (temporal) transaction costs associated with food payments are higher than payments received as cash, the overwhelming majority of the beneficiary households stated that they prefer their payments only or partly in food. However, these preferences are neither homogeneous nor stable. Higher food prices induce shifts in stated preferences towards in-kind transfers but more food-secure households − those closer to food markets and to financial services − are more likely to prefer cash. Though shifts occur, the stated preference for food is dominant: In every surveyed year, the median household prefers no more than 50 percent cash, and in no year do more than 17 percent of households prefer all cash. There is suggestive evidence that stated preferences for food are also driven by self-control concerns.

These results present a conundrum for governments designing social protection interventions and for the donors that support them. As the examples in footnote 1 illustrate, it is an article of faith amongst organizations such as the World Bank that beneficiary preferences need to be taken into account when interventions are designed. But in examples such as the PSNP, such preferences skew towards food while, the Government of Ethiopia and its development partners, would prefer that cash be provided, not least because as Margolies and Hoddinott (2015) show, the implementation costs associated with cash transfers are considerably lower than the cost of providing food. That said, these results also suggest such preferences are not immutable. Ensuring that the purchasing power of cash payments keeps pace with changes in food prices and ensuring that beneficiaries have access to food markets and financial services all make it more likely that the preferences of beneficiaries and program designers (and their funders) will align.

Footnotes

Notes

Kalle Hirvonen is a Senior Research Fellow at the International Food Policy Research Institute, based in Addis Ababa, Ethiopia; his email address is [email protected]. John Hoddinott (corresponding author) is the H.E. Babcock Professor of Food & Nutrition Economics and Policy at Cornell University, Ithaca, NY; his email address is [email protected]. Funding supporting this work was received from the European Union, the United States Agency for International Development, and the CGIAR Research Program on Policies, Institutions, and Markets (PIM). Data collection was conducted by the Central Statistical Agency (CSA) of Ethiopia and funded by the PSNP Donor Working Group. The authors thank Katrina Kosec and the conference participants at the Ethiopian Economic Association Conference for useful comments. The authors are responsible for all opinions and errors. A supplementary online appendix is available with this article at The World Bank Economic Review website.

References