Measuring Poverty Rapidly using Within-Survey Imputations

Poverty is an indicator of paramount importance for gauging the socioeconomic well-being of a population. Especially during or after a shock, poverty estimates are invaluable for assessing the severity of the impact and for identifying which parts of the population were most affected. The measurement of consumption-based monetary poverty, however, has traditionally been very time consuming. A household consumption questionnaire usually includes more than 200 items, including food and nonfood items, often requiring more than two hours to administer. This paper proposes a new methodology that combines an innovative questionnaire design with standard imputation techniques. It substantially shortens the time required to administer a household consumption questionnaire to less than 60 minutes by imputing deliberately absent consumption values for items that are not explicitly asked. The proposed methodology makes it possible to derive poverty estimates without compromising the credibility of the resulting estimate, and it performs considerably better than alternative approaches based on reduced consumption aggregates and cross-survey imputations. This new methodology is particularly useful in fragile states given the significant risks associated with lengthy interviews, as well as to rapidly assess the impact of a shock or of a project. It can also be useful to reduce enumerator and respondent fatigue, or to mitigate the problem of high nonresponse rates