Targeting mechanisms for cash transfers using regional aggregates

Abstract

We propose an empirical method for improving food assistance scoring and targeting, which minimizes under-coverage and leakage of food and cash assistance programs. The empirical strategy relies on a joint econometric estimation of food insecurity and economic vulnerability indicators at the household level, using data-driven instead of predetermined quantiles. We applied the method to recent micro data on Syrian refugees in Lebanon, to explore how regional and community-based aggregates can improve the targeting effectiveness of aid programs, notably food aid by the World Food Program in Lebanon. Our results confirm that using regional aggregates are useful for augmenting the Balanced Poverty Accuracy Criterion, and our method performs much better than the current policy in terms of targeting effectiveness and accuracy for economically vulnerable households.

Appendices

Appendix 1: Computation of the Food Insecurity Indicator

As part of the overall WFP vulnerability score, the food security sector score is constructed from three variables: food consumption score, food expenditure share and coping strategies index. The resulting score is converted into ordinal classes (categories) according to a formula developed by WFP VAM.³ This score has been derived through an iterative process, and is based on several endogenous variables, which would be problematic in predictive models of food insecurity. We reviewed the VASyR dataset and considered all food security related variables in the dataset, to be used as potential food security outcomes (y) in a targeting formula.

The following indicators were considered, and were constructed according to standard WFP methods (World Food Program 2009):

• Food Consumption Score (FCS) – a measure of quality of food utilization at household level; widely used to establish prevalence of food insecurity

• Child Diet Diversity Score – access to food quality by the most vulnerable

• Coping Strategies Index (CSI) – a measure of household economic access to food, food quality and food quantity; used in targeting food assistance in various contexts

• Reduced Coping Strategies Index (rCSI)- cross culturally validated measure of access to food

• Food Expenditure Share – a measure of household economic access to food

It has been highlighted that reliance on a single measure which captures one dimension of food insecurity can misclassify the food insecure, and that combining indicators can improve the measurement of food insecurity (see Maxwell et al. 2013; Jones et al. 2013).

FCS and dietary diversity tend to capture elements of diet quality and diversity, whereas CSI and rCSI reflect quantity or sufficiency. Of these, child dietary diversity was not further explored as this would have reduced the sample of the dataset to households with children under the age of 2 years only. The Coping Strategies Index (CSI) asks a series of questions about how households manage to cope with a shortfall in food for consumption and consists of a numerical score. It was not possible to construct the CSI according to standard methods as the VASyr 2015 posed the coping strategies questions in a way that does not allow the computation of the full index. The reduced Coping Strategies Index is a subset of the CSI that focuses on five food-related coping strategies and results in a cross-culturally validated tool to assess access to food. As the rCSI has been shown to reflect food insecurity as well as the full CSI, the rCSI was considered instead. Conceptually, we considered food expenditure as an economic determinant of food insecurity and therefore used it to validate the food security measure rather than as a component of the measure itself.

We therefore used FCS and rCSI as proxies of food quality and quantity, and used an empirical approach to derive cut-offs for relative vulnerability to food insecurity within this population, rather than international cut-offs developed for use in acute emergency settings.

Using FCS and rCSI as continuous variables, we derived both empirical and bootstrapped quantiles for each of the variables. As both of these approaches yielded similar results, we used the bootstrapped data in order not to impose restrictions on quantiles.

The simplest approach to combine the two variables was to cross classify these quantiles in the derivation of a food insecurity gradient, as has been done by others elsewhere (Maxwell et al. 2013). This cross classification yields a gradient of vulnerability to food insecurity. Considering rCSI Q1 to be the quantile with lowest coping, and FCS Q1 to be that with highest food consumption score, cases falling in the top left cell in Table 9 therefore have the lowest vulnerability to food insecurity. Conversely, cases falling in the bottom right cell (rCSI Q5 and FCS Q5) have the highest vulnerability to food insecurity.

Table 9 Cross classification of bootstrapped quantiles of FCS and rCSI, % of households falling into each category

In order not to impose arbitrary cut-off lines in classifying vulnerability to food insecurity, we tested the food insecurity gradient against economic variables conceptualized as determinants of vulnerability to food insecurity; food expenditure, total expenditure, extreme poverty (below SMEB) & overall poverty (below MEB).

Assuming a food insecurity gradient across quantiles of rCSI and FCS, leads to 9 levels of vulnerability to food insecurity (along diagonals of Table 9). Table 10 displays average food and total monthly household expenditures (in USD), proportion of households categorized as poor and extreme poor by food insecurity gradient in the sample. Data show that, as the food insecurity gradient increases, mean monthly food expenditures and total expenditures decrease, while poverty, extreme poverty and percentage share of food expenditure increase.

Table 10 Economic characteristics of households at different levels of the food insecurity gradient

Based on an analysis of differences in mean monthly food and total expenditures across gradients, thresholds of vulnerability to food insecurity were derived. In brief, where there were significant differences in expenditures across gradients, a threshold line was drawn, yielding five categories of vulnerability to food insecurity. Table 11 displays the cross classification of the bootstrapped quantiles, with thresholds drawn between gradients 2 and 3, 3 and 4, 4 and 5, and 6 and 7.

Table 11 Cross classification of bootstrapped quantiles of FCS and rCSI, according to gradient thresholds

Appendix 2

Table 12 Description of variables