4.2 Marginal Incidence:
-
Average participation rates are a useful first approximation of the distribution of program benefits. However, they do not necessarily provide a reliable guide to the possible effects of a change in aggregate spending on the program. For instance, it is possible that as the size of a program increases, the likelihood of detecting leakage to richer households may decline, with the result that the rich may now be more likely to participate and thus gain a disproportionately large share of the marginal benefits, even though their share of average benefits is low. Alternatively, if public spending tends to reach the rich before the poor, and there is some level of saturation in the transfers appropriated by the rich, then the poor may benefit more from an increase in spending beyond existing levels (Lanjouw and Ravallion, 1999, Wodon and Ajwad, 2000).
Table 6. Marginal Odds of Participation
Quintile
|
VGD
|
VGF
|
FFE
| |
Poorest
|
1.83
|
1.51
|
1.95
|
|
(17.42)
|
(10.54)
|
(19.38)
|
2nd
|
1.73
|
1.86
|
1.37
|
|
(17.30)
|
(13.17)
|
(12.61)
|
3rd
|
0.71
|
1.17
|
0.31
|
|
(7.57)
|
(9.37)
|
(2.64)
|
4th
|
0.58
|
0.54
|
0.55
|
|
(6.19)
|
(4.46)
|
(4.70)
|
Richest
|
0.55
|
0.41
|
0.30
|
|
(6.76)
|
(3.63)
|
(2.29)
|
|
|
|
|
Note: The table reports the coefficient estimates from quintile and program-specific linear probability models that relate the probability of household participation in the program to the average participation rate in the region. The numbers in parentheses are t-ratios.
-
Marginal incidence analysis asks how changes in spending on a program are likely to be distributed across different groups. In order to assess marginal incidence, we examine the effect of expanding the overall size of the program on the probability of participation by households in different quintiles of the expenditure distribution. In particular, for each program and expenditure quintile, we estimate a linear probability model that relates probability of participation in the program to the average participation rate in the region (including all quintiles). Regression coefficients therefore estimate the effect of a change in overall program size in the region on the probability of participation for individuals in different quintiles; these marginal odds of participation (MOP) are reported in Table 6.15
-
The MOP estimates suggest that at the margin, an expansion of all three programs would be decidedly pro-poor. Equivalently, scaling back the programs would reduce participation by the poor more than the rich. In addition, the MOP coefficients broadly confirm the conclusion from the average odds of participation (Table 2) that the food assistance programs tend to reach the poor, and that the FFE program performs best at reaching the poorest. The MOP for the poorest quintile is highest for FFE, while the VGD has the highest MOP for the richest quintile. Notably, for the FFE, the marginal odds of participation fall more steeply as one moves from the poorest to the richest quintile than do the average odds. Thus the average odds under-estimate how ‘pro-poor’ an increase in average spending on the FFE would be. However, as with the average odds calculation, it is important to note that the pro-poor distribution of marginal benefits for the FFE derives in part from the fact that the poor tend to have more children of primary school age.
5.Estimates of Leakage From the System
-
Most estimates in Bangladesh of the total number of beneficiaries of the various food assistance programs have relied primarily on macro data on total disbursements in conjunction with entitlements per beneficiary under program guidelines to infer the total number of households benefiting from them. The 2000 Household Income and Expenditure Survey (HIES) provides an invaluable opportunity to estimate the number of beneficiaries of the various programs using data from two sources: (i) the community questionnaire that was administered in each of the rural PSUs where household interviews were carried out, and (ii) the household questionnaire that was administered to 7,440 households throughout the country. Data from both these surveys can be used to “blow-up” the number of households reported as participating in the programs, thus arriving at a rough estimate of the total number of program beneficiaries throughout the country.
-
HIES Community Questionnaire: The 2000 HIES community survey was administered in a total of 252 rural communities throughout Bangladesh. The survey collected information on infrastructure, agricultural and employment practices, prices, etc. in rural areas in each of the villages where the household survey was administered. Information was also collected on whether the village concerned was covered under the Food-for-Work (FFW), Food-for-Education (FFE), Vulnerable Group Feeding (VGF), and Vulnerable Group Development (VGD) programs during the past 12 months. In addition, in villages where the program had been active during the past year, data was also collected on the total number of households in the village that participated in the program.
Table 7. Coverage Rates: 2000 HIES Community Questionnaire Findings
Program
|
% households living
in a village where program was offered
|
Average participation rates(program villages)
|
Estimated number
of beneficiary households
|
% of rural households covered
|
|
(1)
|
(2)
|
(3)
|
(4)
|
|
|
|
|
|
FFW
|
67.7%
|
--
|
--
|
|
VGD
|
56.8%
|
0.051
|
562,344
|
2.9%
|
VGF
|
63.3%
|
0.127
|
1,560,597
|
8.0%
|
FFE
|
28.3%
|
0.267
|
1,466,834
|
7.5%
|
|
|
|
|
|
Source: 2000 Household Income and Expenditure Survey (HIES) Community Questionnaire
Note: Column 3 and 4 derived based on the 2001 Population Census’s estimate of 19.43 million households living
in rural areas in Bangladesh.
-
Data from the HIES community survey show coverage rates to be quite high at the village level, ranging from 67.7 percent for the FFW to 28.3 percent for the FFE (Table 7, column 1). Data on number of participating households in each village where the program was active, in conjunction with the total number of households residing in the village, was then used to infer the average probability of being a beneficiary in program villages (column 2). These two estimates – namely the coverage rate and average participation rates within a covered village – were then used to estimate the total number of beneficiary households covered by the program.16 The community survey data thus suggest that about 7.5 percent of all rural households participate in the FFE, an estimate that is not too far off from the official program estimate of approximately 10 percent coverage. Similarly, the 562,000 beneficiary estimate from the HIES community survey for the VGD is very close to the official estimates of 550,000 women covered in each 18-month VGD cycle.
-
HIES Household Questionnaire: The HIES household survey also collected information on receipts of food grains from the various programs during the past 12 months. While the community questionnaire was administered to a fairly large and representative group of village residents, it is likely that the information on number of beneficiaries was taken from official records maintained at the mauza or thana level. Thus, comparing the community questionnaire estimates with the information collected at the household-level provides a useful means to cross-check their validity and accuracy. On comparing these estimates, the HIES household survey’s findings with regard to the proportion of households receiving food grains from the various programs were found to be considerably lower than that reported by the community survey (Table 8). For instance, the proportion of households reporting having received food grains from the FFE was only about one-third that reported by the community survey (2.5% vs. 7.5%). Similarly, the proportion of households reporting receipts from the VGF was likewise also considerably lower than that indicated by the community questionnaire (3.2% vs. 8.0%).17
-
The HIES estimates of total number of households benefiting from the VGD program warrant further explanation as, at first sight, they appear to be higher even than the official program estimates. Amongst the possible reasons why this might be so is that in some cases (i) card-sharing (i.e. division of benefits amongst more than one household) is sometimes encouraged by UP officials in order to cover a larger number of poor women in their constituency, and (ii) some of the VGD allocations for each union are distributed to destitute women who come to the distribution center (IFADEP Journal, 1999). In addition, another reason the HIES may overstate the number of VGD beneficiaries compared to the program estimates is that the HIES survey, which took place between February 2000 and January 2001, could have picked up participants from two separate VGD cycles (see related discussion below).
Table 8. Coverage Rates: 2000 HIES Household Questionnaire Findings
Program
|
% households in rural areas reporting receipt of food grains from program in past 12 months
|
Estimated number
of beneficiary households
|
|
(1)
|
(2)
|
|
|
|
VGD
|
5.2%
|
1,043,645
|
VGF
|
3.2%
|
685,742
|
FFE
|
2.5%
|
506,947
|
|
|
|
Source: 2000 Household Income and Expenditure Survey (HIES) Household Questionnaire
-
Given that the HIES household questionnaire collected data not only on whether the household received any benefits from the program during the past 12 months, but also on how much food grain was received, these data can be used to estimate the total amount of food grains received through these programs by all households in the country. Estimates of aggregate household transfers for the VGD, VGF, and FFE obtained by “blowing-up” the HIES data on household grain receipts are reported in Table 9.
Table 9. Program Outlays vs. Survey Estimates
Program
|
2000 HIES-based
survey estimates
(metric tons)
|
95% Confidence Interval for estimate
|
Program Off-take
for FY 1999-2000
(metric tons)
|
Survey estimate as
% of aggregate program allocation (confidence
(intervals)
|
|
(1)
|
(2)
|
(3)
|
(4)
|
|
|
|
|
|
VGD
|
99,978
|
[72,894, 127,061]
|
216,675
|
[34%–59%]
|
VGF
|
70,760
|
[44,251, 97,267]
|
149,138
|
[30%–65%]
|
FFE
|
49,951
|
[27,192, 72,710]
|
285,973
|
[10%–25%]
|
|
|
|
|
|
Source: 2000 Household Income and Expenditure Survey (HIES) Household Questionnaire
Aggregate Program Off-take: Bangladesh Food Grain Digest, World Food Program, Dhaka, Bangladesh
-
Why are the survey estimates of aggregate transfers from the three programs so much lower than aggregate program off-take statistics? Part of the reason may be because the reference period for the survey (12 months preceding the date of the interview) does not exactly match the aggregate off-takes comparison year (1999-2000). However, this is unlikely to be an important factor in explaining the large discrepancy between the two sets of estimates, as program off-takes have been fairly constant over the period spanned by the survey recall period (February 1999 to January 2001).
-
In the case of the VGD, the main reason for the discrepancy is because the total amount reported by each recipient household in the survey is considerably lower (around 96 kg per annum) than what might be expected (around 180 kg per annum) given the program guidelines.18 By contrast, in the case of the VGF and FFE, the main reason for the discrepancy was because a much lower percentage of households reported benefiting from the program than might be expected given total program outlays. Each VGF household covered in the HIES reported receiving, on average, 103 kg through the program during the past 12 months preceding the survey. This figure does not appear to be too far off from what typical VGF entitlements are likely to be, given that the program is meant to provide only temporary relief during times of need. Similarly, each FFE household reported receiving, on average, 98.5 kg of food grains during the past year, not too far off from the expected 120 – 150 kg of food grain transfers.19
-
The survey-based estimates of leakage should be interpreted with caution, as other reasons could potentially explain why the survey reports such low coverage rates and receipts.20 However, studies of other transfer programs (e.g., Alderman 2001) find a closer correspondence between what households reported and what was authorized using the same methodology; therefore the finding of leakage is not built into the methodology. Any conclusion, overall, as to the “pro-poor” nature of spending on these programs based on the incidence analysis presented earlier must be balanced against these findings that suggest that a large share of the total resources allocated to the programs fail to reach their intended beneficiaries.
-
Problems of leakage are not confined to Bangladesh alone, nor is leakage an inevitable outcome of such food distribution programs. Using a similar approach in the Indian context, Ahluwalia (1993) found that roughly a third of the food grains that were supplied in the Public Distribution System in India did not reach beneficiaries. Alderman (1988) found that approximately 69 percent of subsidized wheat released by the government in Pakistan to ration shops did not reach consumers. By contrast, however, Jayne et. al. (2000) found little evidence of leakage of food grains from the distribution system in Ethiopia. What is particularly worrisome in the Bangladeshi context is that similar calculations for the FFE program using the 1995-96 HES indicate substantially lower shortfall, indicating that problems of leakage have deteriorated with time. 21
Share with your friends: |
