Measuring absolute and relative poverty: The sensitivity of estimated household consumption to survey design

Abstract

This paper illustrates that questionnaire design significantly affects estimates of household consumption and absolute poverty. In a between-groups designed experiment in El Salvador, longer, more detailed questions on consumption result in an estimate of mean, household consumption that is 31 percent greater than the estimate derived from a condensed version of the questionnaire. The distribution of household consumption from the long questionnaire first-order stochastically dominates the distribution from the short questionnaire over 96 percent of the range of the distribution. This difference in estimated consumption results in a measure of absolute, severe poverty from the short questionnaire that is 46 percent greater than the estimate derived from the long questionnaire. In contrast, the level of relative poverty is unaffected by the changes in questionnaire design. An implication of this paper is that modifications over time to questionnaires will result in spurious estimates of change in consumption and absolute poverty levels.