Measurement error in long-term retrospective recall surveys of earnings

Abstract

Several recent studies in labour and population economics use retrospective surveys to substitute for the high cost and limited availability of longitudinal survey data. Although a single interview can obtain a lifetime history, inaccurate long-term recall could make such retrospective surveys a poor substitute for longitudinal surveys, especially if it induces non-classical error that makes conventional statistical corrections less effective. In this paper, we use the unique Panel Study of Income Dynamics Validation Study to assess the accuracy of long-term recall data. We find underreporting of transitory events. This recall error creates a non-classical measurement error problem. A limited cost-benefit analysis is also conducted, showing how savings from using a cheaper retrospective recall survey might be compared with the cost of applying the less accurate recall data to a specific policy objective such as designing transfers to reduce chronic poverty.