Is there persistence in the impact of emergency food aid? Evidence on consumption, food security, and assets in rural Ethiopia.

Natural disasters, financial crises, and other economic shocks can have significant negative consequences for uninsured households (Skoufias 2003; Block et al. 2004). When the resulting destruction of assets and changes in economic activity are sufficient to prevent recovery, these shocks lead to poverty traps with lasting effects on household welfare (Barrett and Maxwell 2005). In this setting, food aid or other assistance given in the aftermath of an economic shock may insure households from deleterious shock effects. Emergency food aid intended primarily to sustain short-term food and nutrition security may also serve as a safety net, protecting welfare in the long run and possibly reducing the need for further assistance in the future.

There is a small body of research that assesses the impact of food aid programs on household food security and welfare and, to a more limited extent, nutrition (Barrett 2002). A common finding is that food aid programs such as general food distribution or food-for-work have at most a small impact on food consumption or nutrition and often only a short-run effect on aggregate consumption (see Yamano, Alderman, and Christiaensen 2005; and Quisumbing 2003, for evidence from Ethiopia; see Stifel and Alderman 2003, for evidence from Peru). However, there is little rigorous evidence about whether timely food aid distribution in response to a shock may play an important safety net role by reducing vulnerability and protecting assets (Barrett, Holden, and Clay 2004, is an exception). By preserving stocks of productive assets or savings during a crisis, emergency food aid may have a positive impact on future asset holdings and a persistent effect on welfare. A major challenge of identifying food aid impacts that has been ignored in much of the literature is to account for selection into the programs; failing to do so makes it impossible to attribute causation of welfare gains to food aid.

This article examines this issue in the context of Ethiopia's experiences following the 2002 drought. While initial accounts suggested that poor rainfall was of concern primarily in northeast pastoral areas, rains started late in parts of Amhara region and most crop-dependent areas received below-normal rainfall in August and September. By December 2002, it was estimated that 11.3 million Ethiopians would face severe food shortages in 2003 with an additional 3 million people at risk of significant shortages, double the estimate only four months earlier. Cereal production was estimated to have fallen by 25 % (FEWS NET 2002-03). The worst affected areas included much of the pastoral areas of Afar, parts of eastern Tigray, eastern Oromiya, parts of Amhara, and SNNPR. In response, the government expanded its two major food aid programs, an emergency food-for-work program called the Employment Generation Schemes (EGS) and free food distribution also known as "Gratuitous Relief" (FFD). (1)

This article uses rural longitudinal household survey data collected in 1999 and 2004 to measure the effect of these programs on consumption levels, food security, and asset holdings eighteen months after the peak of the drought, when food aid transfers had substantially or entirely ceased in most program villages. The data, the setting, and the methodology used in this analysis all provide the conditions for a rigorous evaluation. First, the timing of the data collection makes it possible to control for predrought household and farm characteristics and to observe key outcomes roughly two years after the onset of the drought. Second, several features of these data improve the quality and extent of our knowledge of food aid's effects. The household questionnaire used in 2004 included retrospective questions on the effects of the drought and on the timing and size of food aid participation and receipts. The questions on drought effects include information on perceived changes in famine risk, a useful summary measure of changes in household food security. Also, detailed information on livestock holdings provides useful measures of asset holdings in a country where livestock dominate all other assets as a form of investment. Finally, we measure the average treatment effect of the food aid program using a difference-in-differences matching technique based on propensity score matching. Heckman, Ichimura, and Todd (1997, 1998) and Heckman et al. (1998) show that under certain conditions on the data--all of which are satisfied in this study--propensity score matching estimators provide reliable estimates of program impact.

We find a large, significant effect of EGS participation on the growth in consumption and food consumption (in per adult equivalents) of recipients one-and-a-half years after the 2002 drought. EGS beneficiaries experienced a reduction in famine risk relative to five years ago, while a comparison group of non-beneficiaries reported an increase in famine risk over the same period. We find a significant average impact of FFD participation on growth in food consumption, but, surprisingly, a negative impact on food security. After disaggregating impact estimates by pre-drought household consumption tertiles, we find significant impacts of public works participation on food consumption and food security for some households in the middle to upper tail of the expenditure distribution. The better-targeted FFD program showed greater benefits for the poor.

The article is organized as follows. The next section presents the ERHS data and summarizes the effects of the drought and food aid receipts by sample households. We then present the methodology for measuring longer-term food aid impacts and provide the impact estimates. The final section discusses the implications of the impact estimates for food aid policy.

Evidence of Drought Effects and Food Aid Use

Our data come from the Ethiopia Rural Household Survey (ERHS), a longitudinal household data set collected in six survey rounds from 1994 to 2004 in fifteen rural Ethiopian villages. (2) The sampling frame was stratified on the main agro-ecological zones (excluding pastoral and urban areas) and village sample sizes were chosen to generate an approximate self-weighting sample in terms of farming system. Given the relatively small number of sampled villages, extrapolation of results to rural Ethiopia as a whole must be done with care. (3)

We use data from the 1999 and 2004 rounds of the ERHS to estimate food aid impacts after the drought. (4) The 2004 round captures a variety of information about the incidence of the 2002 drought among sample households, about the breadth and depth of drought effects, and about receipt of food aid through the EGS or FFD. Pilot testing suggested that the eighteen-month time gap between the peak of the drought and the 2004 survey enumeration was too long to capture immediate drought effects on yields, consumption or assets. Instead, qualitative questions about the incidence and effects of the drought were asked in a detailed shocks module and in a separate drought module. Detailed information about the timing and size of transfers from each program were captured in survey modules on food aid, off-farm income and food consumption. These data show that most food aid transfers were made in the first twelve months after the drought. Although food aid resumed in some villages in the period captured in the 2004 round of the survey, with the exception of one village food aid transfers at that time were too small to account for the observed growth in consumption. (5)

Table 1 presents summary information on drought incidence and the food aid response for the fifteen ERHS villages. Column 2 lists mean consumption per adult equivalent in 1999 as an indication of predrought welfare levels. In four villages, less than 3% of respondents reported a drought (self-defined) in 2002. This figure was less than 15% in two other villages (column 3). Drought incidence ranged from 30% to 85% in the remaining nine villages, which were the ones that received food aid from September 2002 to March 2004 (columns 4 and 5). The self-reports of drought are consistent with rainfall data from nearby weather stations: household self-reported drought incidence in 2002 was fairly closely correlated to deviations of 2002 rainfall during the main growing season (August-December) from long-run averages (p = 0.27). Table 1 also shows that the incidence of the drought was spread broadly across the expenditure distribution of villages in the sample. Indeed, receipt of food aid is more closely correlated with the share of drought-affected households in the village (Spearman correlation coefficient =0.35) than with the wealth of the village in quintiles (Spearman correlation coefficient =-0.16). This pattern reflects the disaster relief motivation of food aid during this period.