Forgotten unemployment : recall bias in retrospective data

Abstract

It is a well documented fact that the ordering and specific wording of a survey questionnaire can produce different readings of the same underlying event. Differences can also arise when the same general question is asked of individuals at different times. One area in which such discrepancies have been found is in comparisons of unemployment data collected on a monthly basis with those obtained from a onceyearly survey . The source of the monthly unemployment figures is the Current Population Survey (cps), conducted by the Census Bureau for the Bureau of Labor Statistics. Each month, representatives of approximately 60,000 households are asked questions about their labor force activity and that of other household members during a given reference week . Annual unemployment estimates are constructed by averaging monthly data for the year. Annual unemployment data are also obtained, on a retrospective basis, from the Work Experience Supplement to the Current Population Survey in March of each year . Persons responding to the supplement questions are asked to recall events which occurred from January through December of the previous year . These respondents are asked to aggregate the year's labor force activities into summary figures. That is, they are asked questions such as : How many weeks was (household member's name) looking for work? and How many weeks was (household member's name) working? However, no attempt is made to ascertain the months in which these events took place. The supplement data reflect the number of persons with unemployment at any time during the previous year ; a person experiencing two or more spells of unemployment is counted only once . Recently, this estimate has been about 18 to 21 million persons a year . In con-