Food Assistance Programs In Bangladesh

4.1 Average Participation Rates, Targeting Accuracy, and Incidence:

31. 
    Table 2 reports quintile-specific average participation rates and average odds of participation, by program, for the rural sample. Quintiles have been defined over the distribution of per capita expenditures (corrected for differences in cost-of-living across regions) within rural areas, with equal numbers of individuals in each. The average participation rate is the percentage of the population in the quintile who benefit from the program. The average odds of participation is given by the ratio of the quintile-specific average participation rate to the overall average.
    

32. 
    Overall participation rates in each of the three programs were very low, ranging between 3 – 5 percent of the rural population. Given the low coverage, while the programs may be extremely important for the beneficiaries themselves, it is unlikely that any one of these could be expected to have a significant impact on overall poverty reduction. Table 2 also reports the joint coverage of the three programs. While each program serves only a small part of the poor population, joint coverage rates are higher. Approximately 10 percent of the population received assistance from at least one of the three programs in the 12 month period. The limited overlap in coverage by the three programs (only 7 percent of the beneficiaries received transfers from more than one program) conforms with program targeting criteria for the VGD and FFE programs that restrict participation to those households that are not already assisted by other food-assistance programs.
    

33. 
    All three programs are targeted towards the lower quintiles, with participation rates declining as expenditure per person increases. Overall, the odds of the poorest quintile participating in at least one of the three programs is 1.81, versus 0.40 for the richest quintile. In other words, the poor are nearly 5 times more likely to participate in the food assistance programs than people in the richest quintile. FFE is slightly better targeted to the poor than the VGD and VGF: the odds of the poorest and richest quintiles participating in FFE are 1.92 and 0.4, respectively.
    

34. 
    While participation rates for the entire rural population are indicative of the extent to which the safety net program can reduce rural poverty, population-wide coverage rates present a distorted picture of the targeting effectiveness of a targeted program. All three of the programs are targeted towards specific populations to attain different objectives. Below, we assess how successful the VGD and FFE programs are at reaching their intended beneficiaries.
    

35. 
    The VGD and FFE programs use a two-step targeting mechanism whereby resources are concentrated within rural thanas to selected economically backward unions, and then targeted to households that satisfy the beneficiary targeting criteria (as described in Section 2) within the selected unions. We use the same criteria to define the sample of eligible households within PSUs that have the program. In the case of FFE, which by design is limited to households with children enrolled in primary school, we further restrict the target group sample to households that have at least one child currently enrolled in primary school.
    

36. 
    Identifying the target population for the VGF program is more difficult. VGF transfers are allocated to areas that are affected by disasters (such as floods), and as such are not necessarily restricted to rural areas. Among those households affected by disaster, the program attempts to reach the poor, as identified by their income, assets, and principal occupation of the household head. Therefore, the target population of the program is the group of poor households that were affected by disasters for which VGF resources were mobilized during the past year. Since the HIES survey does not allow us to identify this specific group of poor disaster-affected households, we are unable to examine how effective the VGF program is in reaching its intended beneficiaries.
    

37. 
    Table 3 reports various measures of targeting accuracy for the VGD and FFE programs. Under-coverage is defined as the share of the eligible population that is not covered by the program. Another way of looking at under-coverage is to compute exclusion (or type I) errors – the share of the population that is eligible, but does not receive benefits from the program (i.e. as a share of the total population). Complementary to the concepts of under-coverage and exclusion errors are leakage and inclusion errors respectively. Leakage is defined as the share of participants who are ineligible for the program. Inclusion (or type II) error is the share of the population that is ineligible for the program but still participates.
    

38. 
    There are two points of note. First, even among the eligible population, under-coverage is extremely high, with 85 percent and 91 percent of the target populations eligible for the FFE and VGD not included in the programs. Relatively high rates of exclusion errors show that, with the current targeting criteria, a large share of the population qualifies as being eligible. It would be worth investigating if finer targeting criteria that do not entail excessive screening costs, could be developed. Second, leakage is fairly low for the VGD program. However, 30 percent of the FFE beneficiaries do not meet any of the targeting criteria incorporated in the program design. Minimizing leakage must of course be balanced against the additional administrative and political costs of achieving better targeting. Yet, 30 percent leakage appears to be quite significant, especially in light of the vast share of the eligible population that is currently not benefiting from the program.¹² Moreover, it also does not account for leakage that arises from misappropriation of food resources; we return to this second source of leakage in the next section.
    

39. 
    While poverty reduction is not the principal objective of either the VGD or the FFE program, it is worth checking how well the targeting criteria work in channeling resources towards the poor. While the chosen targeting criteria (such as landlessness, day laborer occupational class etc) are correlated with low incomes, the correlation between the set of indicators and poverty is far from exact. This is confirmed in Table 4 which shows that while over 50 percent of the eligible population for both programs fall in the bottom two quintiles, approximately one-fourth of the population that meets the eligibility criteria actually falls in the richest two quintiles.¹³ Some of the non-poor can be expected to be program beneficiaries simply because the targeting criteria are an imperfect proxy for living standards. However, as Table 4 shows, even amongst the eligible population, the odds of participation in the bottom quintile are much higher than that for persons in the higher quintiles. Since participation rates are higher for the poor than the rich, the distribution of benefits (i.e. incidence) of these programs is likely to be correspondingly pro-poor as well.
    

40. 
    To summarize, several factors underlie the pro-poor distribution of benefits. First, the targeting criteria narrow down the eligible population such that more than half the beneficiaries that would satisfy program requirements are from the bottom two-fifths of the population. Second, even amongst eligible beneficiaries, it appears that the local program administrators use additional means beyond the stated targeting criteria to identify the poor from amongst the eligible populations. Thus, even amongst the group of eligible beneficiaries, a person from the lowest quintile is about 2.5 times as likely to be selected for the program as an individual from the richest quintile. Finally, in the case of the FFE as was mentioned earlier, part of the reason the distribution is pro-poor is simply because poor households on average have a higher share of children of primary school-going age.
    

41. 
    The assumption of constant subsidy rates that we make to infer a pro-poor distribution of benefits based on average participation rates across the welfare distribution is possibly a strong one. It pre-supposes that all program participants receive the same quantity of food transfers, and derive the same benefit per kilogram of rice or wheat received. It also assumes that other benefits (e.g., training received) and opportunity costs of program participation are similar for rich and poor beneficiaries. If the poor tend to receive transfers that are of higher value to them, or if their opportunity costs of participation are lower than those incurred by the rich, the actual incidence of benefits may be even more progressive than shown here. On the other hand, it could be that poor participants receive lower quantities than the rich on average, in which case the actual incidence of benefits would be less pro-poor than that shown earlier.
    

42. 
    Using data on the quantity of food received by each household in the past 12 months, we can check if the amount of transfers received vary across the expenditure distribution. Because of data limitations, we are unable to measure all costs and benefits of program participation. A priori, transfers should be distributional neutral, as each beneficiary is entitled to a fixed allocation per month. In practice, evidence from other studies suggests that actual transfers may be different from entitlements. For example, in the case of the FFE, Ahmed and del Ninno (2001) find that 71 percent of FFE beneficiaries report receiving less than their entitlements, as private dealers responsible for distributing food were diverting grains to the black market. In other instances, the same study reports that dealers were withholding food grains from beneficiaries as payment for prior loans made to beneficiaries at high interest rates.
    

43. 
    Table 5 reports average transfers of wheat and rice received by program beneficiaries. While transfers do indeed vary across expenditure quintiles, there does not appear to be any clear discernable pattern that would suggest that rich receive more than the poor.¹⁴ Since participation rates are decidedly pro-poor, accounting for differences in the quantity of transfers received still yields a pro-poor distribution of benefits. However, the table does suggest that average transfers received by beneficiaries in both the VGD and FFE programs are considerably smaller than their entitlements. We return to the implications of these findings for the two programs in Section 5 when estimating the extent of leakage from the system.