Which Form of Safety Net Transfer is Most Beneficial Which Form of Safety Net Transfer is Most Beneficial? Impacts on Income, Food Security, and Child Nutrition Akhter Ahmed, John Hoddinott, Wahid Quabili, Shalini Roy, Fiona Shaba, and Esha Sraboni International Food Policy Research Institute Stakeholder Workshop 3 December 2013, Dhaka
TMRI Objectives The overall objective of the Transfer Modality Research Initiative is to provide evidence that can be used to streamline the social safety net system in Bangladesh. The research will inform policymakers which type of program can best improve the income status and food and nutrition security of the poor and thus be a valuable tool to the government as it prepares its social protection strategy. The research has the following specific objectives: Measure the impact and cost-effectiveness of transfer methods on these key outcomes: household income household food security child nutrition Evaluate the process of delivering benefits (that is, transfers and nutrition knowledge) at the operational level and solicit feedback from program participants
Evaluating Impacts IFPRI designed a rigorous impact evaluation of the Transfer Modality Research Initiative (TMRI) in the north and the south: Only cash (north & south) Only food (north & south) Food + cash (north & south) Nutrition behavior change communication (BCC) + cash (north) Nutrition BCC + food (south) We developed a randomized controlled trial (RCT) design to evaluate the impact of the 5 transfer modalities Randomization is often termed as the “gold standard” for impact evaluation because it is the most powerful way to construct a valid counterfactual of what might have happened without the program We used RCT with “before-and-after” and “with-and-without” differences for estimating the impact of transfers We used the analysis of covariance (ANCOVA) method of estimating impact
RCT impact estimate with difference-in-differences Outcome Baseline (Before) Follow-up (After) PA CA Program Control Impact = (PA - CA) - (PB - CB) PB = CB
RCT impact estimate using Analysis of Covariance (ANCOVA) regression The ANCOVA regression model that we used to estimate impact is the following (example for the north): 𝒀 𝒕 = ∝ 𝟎 𝑵 + 𝜸 𝟏 𝑵 𝒀 𝒕−𝟏 + 𝜷 𝟏 𝑵 𝑭𝒐𝒐𝒅 𝒕 + 𝜷 𝟐 𝑵 𝑪𝒂𝒔𝒉 𝒕 + 𝜷 𝟑 𝑵 𝑭𝒐𝒐𝒅 & 𝑪𝒂𝒔𝒉 𝒕 + 𝜷 𝟒 𝑵 𝑭𝒐𝒐𝒅 & 𝑩𝑪𝑪 𝒕 + 𝜺 𝒕 𝑵 With difference-in-differences: Impact = ((Yttreat – Yt-1treat) - (Ytcontrol – Yt-1control)) With ANCOVA regression: Impact = ((Yttreat – αYt-1treat) - (Ytcontrol – αYt-1control)) ANCOVA is “more flexible” in the sense that the ANCOVA estimate is equivalent to the diff-in-diff estimate if α = 1. But ANCOVA allows estimating the autocorrelation rather than imposing it to be 1 ANCOVA estimates are preferred over diff-in-diff estimates, given the high variability and low autocorrelation of the data at baseline and follow-up (McKenzie 2012, Journal of Development Economics)
Household surveys for impact evaluation The required quantitative data for impact evaluation come from three household surveys The first household survey, carried out in April 2012 (just before the start of transfers), provides the information needed for the baseline study A first follow-up survey was conducted in June 2013, just after 12 months of transfer distributions were completed A second follow-up or endline survey will be conducted in June 2014, after 24 months of transfer distribution The surveys include TMRI participants and non-participant control households
Baseline per capita monthly expenditures (proxy for income): 19% higher average income in the south
Impact of transfers on per capita monthly expenditure (proxy for income): Absolute change (taka)
Impact of transfers on per capita monthly expenditure (proxy for income): Percentage change
Impact of transfers on per capita monthly food expenditure: Percentage change
Impact of transfers on per capita monthly non-food expenditure: Percentage change
Impact of transfers on per capita daily food energy (calorie) acquisition: Absolute change (kcal)
Impact of transfer on food poverty: Percentage points reduction in prevalence of <2,122 kcal/person/day
Impact of transfer on hard-core food poverty: Percentage points reduction in prevalence of <1,805 kcal/person/day
Aggregate food groups and weights to calculate the Food Consumption Score (Source: WFP) Food items Food group Weight 1 Rice and other cereals Staples 2 Beans, lentils, peas and nuts Pulses 3 Vegetables and fruits 4 Beef, goat, poultry, eggs, and fish Meat, eggs and fish 5 Milk, yogurt, and other dairies Milk 6 Sugar, sugar products, and honey Sugar 0.5 7 Oils, fats, and butter Oil
Impact of transfer modalities on diet quality: Absolute change in Food consumption score
Kernel density functions of FCS: Examples South: Endline Food vs. Food+BCC North: Baseline vs. Endline Cash+BCC
Impact of transfer modalities on diet quality: Change in dietary diversity (number of food consumed out of 12 food groups)
Impact of transfer on child nutritional status: Percentage points reduction in prevalence of stunting (children 6-59 months <-2 height-for-age Z-score)
Summary and conclusions Our estimation strategy relies on the randomized design, which eliminates systematic differences between participants and non-participants and minimizes the risk of “selection bias” As a result, average differences in outcomes across the groups after the intervention can be interpreted as being truly caused by, rather than simply correlated with, the receipt of transfers and transfers with nutrition education Moreover, we take advantage of the baseline survey and estimate the treatment effect using Analysis of Covariance (ANCOVA) regression, which is our preferred method over difference-in-difference estimates
Summary and conclusions Differences in the size of impact as revealed from the F-tests: Income in the north: “Only cash” has statistically significant higher impact than cash+food. There are no statistically significant differences between “only cash” and “only food”. “BCC+cash” has significantly higher impact than those of the other 3 treatment arms. Income in the south: “BCC+food” has significantly larger impact than those of the other 3 treatments. No statistically significant difference between other treatment arms.
Summary and conclusions Differences in the size of impact as revealed from the F-tests: Calories in the north: “BCC+cash” has significantly larger impact than those of the other 3 treatments. No statistically significant difference between other treatment arms. Diet quality (FCS) in the north and the south: “Only food” has significantly higher impact than “only cash” and “cash+food”. “BCC+cash” and “BCC+food” have significantly larger impact than those of the other 3 treatment arms. Stunting in the north: “BCC+cash” has significantly larger impact than those of the other 3 treatments. No statistically significant difference between other treatment arms.
Summary and conclusions In the north, the poorest region, we found statistically significant positive impacts of all 4 modalities on (1) income, (2) food expenditure, (3) non-food expenditure, (4) calorie acquisition, (2) food poverty, (5) diet quality, and (6) child stunting, with cash+BCC having the biggest size of impacts on all 6 indicators. However, in the south, which is a disaster prone, but higher income region than the north, “cash only” has statistically significant impact only on diet quality. “Food only” has significant impact on income and food and non-food expenditures, and diet quality. “Cash+food” has significant impacts on income, food expenditures and diet quality. “Food+BCC” has significant impacts on income, food and non-food expenditure, calories, and diet quality, but not on stunting.
Summary and conclusions It is intriguing to find that food and cash transfers have by far the leading impact when they are combined with nutrition BCC. Why does BCC have the largest impact even though the BCC training curriculum does not include non-nutrition livelihoods attributes? Does participation in BCC activities raise women’s status/empower them? We will probe into this question in early 2014 through an in-depth qualitative study. Why do patterns in the north and the south differ? In the south, participants of “cash only”, “food only” and “cash+food” improved their diet quality rather than quantity. Only “food+BCC” group shows improvements in both diet quantity and quality. Our survey results indicate that, the greater the risk of disaster, the less likely a household is to immediately “consume” a transfer – and, for example, more likely to use it for precautionary savings given the risk of future bad shocks, or to use it to repair/improve houses that were damaged in a bad shock. But BCC may result in overcoming some of that.
Interim policy options Integrate nutrition into social safety nets Increase the size of transfers of safety nets to generate sizable impacts The size of transfer relative to household income is tremendously important when trying to achieve sustainable improvements in the food security and livelihoods of the poor There are numerous safety net programs currently operating in Bangladesh. However, most of these programs have limited coverage, are uncoordinated, and are not adequately funded. Consolidate and simplify programs and phase out high-cost, ineffective programs. Improve the targeting performance of existing safety nets