Design effects in survey-based measures of household consumption

An important issue in designing survey questions on household consumption is the level of aggregation of expenditure categories. In this paper, I present experimental evidence on biases that are related to the choice of expenditure categories. Comparing responses to a 'one shot' question with responses to a more disaggregated design that comprises 35 expenditure categories reveals significant underreporting in the one-shot question. The same holds true when responses to one of these still relatively broad categories (food and beverages) are compared with measures from a budget survey that is based on a very detailed expenditure diary. There is also some evidence that these biases affect subsequent substantive economic analysis such as the estimation of Engel curves.