Finding the Best Indicators to Identify the Poor

Abstract

Proxy-means testing (PMT) is a method used to assess household or individual welfare level based on a set of observable indicators. The accuracy, and therefore usefulness of PMT relies on the selection of indicators that produce accurate predictions of household welfare. In this paper I propose a method to identify indicators that are robustly and strongly correlated with household welfare, measured by per capita consumption. From an initial set of 340 candidate variables drawn from the Indonesian Family Life Survey, I identify the variables that contribute most significantly to model predictive performance and that are therefore desirable to be included in a PMT formula. These variables span the categories of household private asset holdings, access to basic domestic energy, education level, sanitation and housing. A comparison of the predictive performance of PMT formulas including 10, 20 and 30 of the best predictors of welfare shows that leads to recommending formulas with 20 predictors. Such parsimonious models have similar predictive performance as the PMT formulas currently used in Indonesia, although these latter are based on models of 32 variables on average.

Finding the Poor vs. Measuring their Poverty: Exploring the Drivers of Targeting Effectiveness in Indonesia

“…For instance, one cost-effective alternative may be to shorten the targeting questionnaire to allow a larger number of households to be surveyed at a lower cost, which might not necessarily come at the expense of targeting accuracy. Indeed, Bah (2013) shows that going from 10 to 30 indicators included in a PMT formula does not significantly increase the accuracy of predicted household poverty levels; nor does it reduce targeting errors. 26 Finally, the targeting accuracy results in this paper are based on the assumption of perfect (or strong) correspondence between the beneficiary lists from the UDB registry and the households who actually end up receiving social program benefits.…”

“…The Indonesian PMT models have a predictive performance that appears similar to PMT regressions in other countries. On average, the PMT models used to rank households in the UDB have an R-squared of 0.5, a rank correlation between actual and predicted consumption of 0.67, and predicted model targeting error rates at the 40 th percentile cutoff of about 30 percent (Bah, 2013). Below, we use information outside the UDB to examine the potential targeting improvements associated with increasing the predictive accuracy of the PMT.…”

Finding the Poor vs. Measuring Their Poverty: Exploring the Drivers of Targeting Effectiveness in Indonesia

“…For instance, one cost-effective alternative may be to shorten the targeting questionnaire to allow a larger number of households to be surveyed at a lower cost, which might not necessarily come at the expense of targeting accuracy. Indeed, Bah (2013) showed that increasing the number of indicators included in a PMT formula from 10 to 30 does not significantly increase the accuracy of predicted household poverty levels nor reduce targeting errors. 38 Furthermore, the targeting accuracy results in this paper are based on the assumption of perfect (or strong) correspondence between the beneficiary lists from the UDB registry and the households who actually end up receiving social programme benefits.…”

“…However, in general, the evidence suggests that community targeting is best for identifying the very poorest households. Other studies focusing on the design of optimal PMT formulas (e.g., Sumarto et al 2007, Muller and Bibi 2010, and Bah 2013 show that targeting errors are unavoidable when using simple indicators, although the degree of error can be minimised with more careful selection of the proxies for This paper provides the first attempt to assess the relative contribution of the household registration and ranking processes to the overall accuracy of a centralised targeting registry. We do so using Indonesia's newly developed household targeting registry and aim to identify priority actions for improving targeting effectiveness.…”

Essays in Empirical Development and Education Economics

“…Overall, the model fit we obtain based on the most sophisticated model with 46 parameters (37 variables, eight interaction terms, and a constant) still compares favorably to other results in the literature. For instance, a recent study that investigated models created from 340 candidate variables in the Indonesian Family Life Survey finds that even the best models based on 40 variables do not attain an R 2 higher than 60 percent [Bah, 2013].…”