Welfare Analysis with a Proxy Consumption Measure Evidence from a Repeated Experiment in Indonesia

Every three years, Indonesia fields simultaneously two nationwide surveys which collect consumption data. One collects consumption using 23 questions, the other using 320 questions. Based on a repeated experiment in which the two questionnaires were randomly assigned across households, I examine the consequences of using a higher level of aggregation in questioning. A mapping of distribution functions reveals the combined effect of systematic differences in measurement and measurement error. Using a pseudo crosssection approach, I eliminate the effect of measurement error and find that using a high level of aggregation yields a lower consumption measure, and that the fraction of underestimation increases as consumption rises. A one percent increase in average consumption increases the fraction by which consumption is underestimated by about .4 percent point. Next, I examine the consequences of using the short consumption questionnaire in welfare analysis. Higher relative measurement error in the consumption measure derived from the short questionnaire results in higher poverty estimates even if the poverty line is adjusted to take account of the systematic underestimation. Small differences are found for analysis that is based on the rank the individual holds in the consumption distribution. In gradient analysis, it seems impossible to devise a simple correction factor for the higher consumption elasticities that follow when the short questionnaire is used