Experiments to Measure Consumption and Attitudes

Rand Corporation, Santa Monica CA

Investigators

Abstract

A central issue in public policy is how consumption levels and patterns change after retirement, whether such changes are expected or not, and how they are related to financial planning in pre-retirement years. The empirical analysis of these questions requires reliable data on households' consumption expenditure and attitudes. The broad aim of this project to use experimental survey techniques to assess the reliability of data obtained from established survey questions, and to explore alternative formulations. Existing largescale field surveys such as the Health and Retirement Study (HRS) as well as the upcoming Survey of Health, Aging and Retirement in Europe (SHARE) cannot provide the space to collect reliable consumption data using a detailed list of expenditure items. The first specific aim of this study is to use two alternative designs of questions on households' consumption expenditure that require only a small number of items, and to compare response behavior and quality of data obtained from these designs. The first design follows common practice, based on enumeration of expenditures for a small list of items. The second procedure is new; it uses data on financial flows and the income identity to determine the flow of non-durable consumption as a residual. In addition to reliable consumption data, information on households' attitudes is required to estimate some recent economic models that provide more flexible and realistic alternatives to the established life-cycle theory of household behavior, particularly with respect to individual heterogeneity. Such data is typically obtained using hypothetical-choice questions, but these questions might be subject to bias related to the cognitive processes that govern survey response behavior. The second specific aim of the proposed project is therefore to test response behavior in new formulations of established hypothetical choice questions that are designed to avoid some of these potential biases, and to compare response behavior in these new versions with traditional designs.