Fewer Questions, More Answers

The assignment of social programmes to their target population, known as targeting, is key to effective policy implementation. Proxy means testing is a widely used targeting approach where means testing is infeasible due to economic informality. This paper proposes a novel, practically feasible assessment approach that aims to reduce average proxy means test data collection costs, or allow more extensive data collection within a given resource envelope. Combining variable selection and prediction intervals, it develops a household-level truncated early stopping algorithm, which can reduce average interview length while maintaining predictive accuracy close to a standard proxy means test baseline. Applying the approach to Indonesian data, simulation of a 40 percent population coverage programme shows that targeting questionnaires could be shortened by 61 percent while maintaining PMT-level accuracy. A case study of a large health insurance programme in an urban area suggests that the share of intended beneficiaries who are among the targeted population can potentially be increased from 65.6 percent to 78.3 percent if enumerators conducted more of the shorter surveys that the truncated early stopping algorithm generates.