Out of Work, Out of Mind: Response Errors in Retrospective Reports of Unemployment

This article examines several aspects of the validity of retrospective reports of unemployment, obtained by a comparison of respondent reports with company records for a sample of workers from a large manufacturing firm. We find that reports of the total amount of unemployment in the calendar year prior to the interview are reasonably accurate, but reports of the timing of spells of unemployment within the year are much less accurate. Estimates of a regression model relating response error to measures of respondent characteristics, the length of the recall period, the difficulty of the reporting task, and the likely salience of the events show that the salience measures are clearly the most important and that time and demographic factors are relatively unimportant predictors of response error.