Printer Friendly

CPS contemporaneous and retrospective unemployment compared.

The Current Population Survey (CPS),(1) the official source of labor force information in the United States, measures unemployment in two different ways. In each monthly survey respondents provide a series of answers on current labor market activity, which are translated into a "contemporaneous" measure of unemployment. In addition, the March survey contains a supplement, the Work Experience Survey, that asks questions on labor market activity in the preceding calendar year, and thus can provide a "retrospective" measure of unemployment.

Although, in principle, both sources of data provide similar information, aggregate unemployment rates calculated from the contemporaneous survey are typically higher than those computed from the retrospective survey.(2) Furthermore, the disparity differs dramatically among demographic groups and has a strong cyclical component. For example, among teens and adult women, the contemporaneous rates tend to be higher than the retrospective rates. Also, as the unemployment rate rises, the ratio of the contemporaneous rate to the retrospective rate declines.

In this report, microdata from the CPS are used to examine these discrepancies. The study covers the period 1970-88, with special emphasis on the later years. The longitudinal nature of the CPS, which allows respondents' answers to be matched between one year and the next, permits selection of a sample of workers counted as unemployed on the regular monthly survey date, and examination of data on their retrospectively reported labor market behavior for that same date, as collected in the following year's March work experience survey. The results of this study suggest that roughly 40 percent of all contemporaneous unemployment spells go unreported retrospectively. Individuals whose main activity in the regular monthly survey was keeping house" or "in school," and those who initiated their unemployment spell in search of temporary work are significantly less likely to report unemployment retrospectively. Although these results are due, in part, to structural differences between the contemporaneous and retrospective parts of the survey - as discussed below - they seem to suggest that workers with relatively weak labor force attachment are more likely to be counted as unemployed in the contemporaneous measure.

Structure of the CPS

If labor market activity were measured the same way in the contemporaneous and retrospective parts of the survey, the difference between the two unemployment rates would have to be attributed to reporting errors by respondents. However, the structures of the two parts of the survey are quite different. Because differences in the questions asked may partially explain differences in measured unemployment, it is useful to review the format of the CPS questionnaire and the determination of an individual's labor market status.(3)

To be declared unemployed in the contemporaneous part of the survey, a respondent must provide certain answers to a series of questions. First, the respondent must report not working "most of last week." Then the worker must report not having done any work last week, and that he or she was not temporarily absent from a job due to, say, illness or a vacation. However, temporary absence from a job will lead to the classification of unemployed if the worker reports being absent due to "temporary" or "indefinite" layoff. Otherwise, a worker is classified as unemployed only if he or she then reports having looked for work within the past 4 weeks, and reports at least one method of job search used (such as answering newspaper advertisements or contacting employers directly). Still one more test is required of respondents who otherwise would not be considered unemployed: only those who report that they could have taken a job last week or that they have a temporary illness are officially classified as unemployed.

The contemporaneous unemployment rate is the ratio of the total number of workers unemployed to the total number of workers in the labor force (those employed or unemployed), weighted by CPS sampling weights.(4) The annual contemporaneous unemployment rate is the weighted average of the monthly unemployment rates over the year.

An individual's labor market status in the preceding year is determined in an entirely different manner in the retrospective March supplement.(5) In this part of the survey, a different set of questions is asked of workers, depending on whether they report any work experience in the preceding calendar year. Those who report that they did work last year are asked how many weeks they worked. Respondents who report fewer than 50 weeks are then asked how many of the remaining weeks (52 minus weeks worked) were spent looking for work or on layoff from a job. The response to this question then provides an estimate of weeks unemployed and - in combination with information on the number of weeks worked - the number of weeks in the labor force.

For those who report no weeks worked in the previous calendar year, a different series of questions is asked. Respondents first are asked to reaffirm that they did not work at all last year, not even at part-time or seasonal jobs. If a respondent reports some work in response to this additional prompt, his or her labor market status is determined by the above method. Otherwise, respondents are asked if they spent any time looking for work or on layoff in the preceding calendar year and, if so, in how many different weeks. For individuals following this stream of questioning, the response to this question provides estimates of both weeks unemployed and weeks in the labor force. The retrospective unemployment rate is then calculated as the ratio of total weeks unemployed in the previous calendar year to total weeks in the labor force in that year.(6)

The longitudinal nature of the CPS provides a way to test how a given individual responds to the contemporaneous and retrospective questions in reference to the same period.(7) Respondents to the CPS are surveyed for 4 consecutive months, and then again for 4 consecutive months after an 8-month absence. As a result, for a maximum of half the sample, an individual's responses can be matched between one year and the next.(8) In year t, the first year the individual is in the survey, he or she answers questions for the current year and, in year t + 1, provides retrospective infortnation for year t. If an individual is unemployed contemporaneously in March of year t, then in year t + 1, he or she should report some unemployment retrospectively for year t.(9) This characteristic of the data has not been exploited in the literature and will be used extensively here.

Comparison of the rates

Before proceeding with the analysis of the microdata, a review and update of differences in the aggregate measures will be useful. Evidence regarding the disparity is presented in tables 1 and 2.(10) Table I shows that, in virtually every year, the contemporaneous rate is higher than the retrospective rate, and a strong cyclical component exists in their difference.(11) During the periods of recession (expansion), the contemporaneous and retrospective rates converge (diverge). For example, 1982 marked the depth of a downturn, and the contemporaneous rate was only 8 percent greater than the retrospective rate. By comparison, the contemporaneous rate was 24 percent higher during 1987, which was a year of economic expansion. One striking exception to this pattern is 1983. Although the unemployment rate for that year was the second highest over the study,period according to both measures, the contemporaneous rate is 19 percent higher than the retrospective rate - a difference more typical of a boom.


Analysis of table 2 indicates that the two rates also differ systematically across demographic groups. In particular, contemporaneous rates are substantially higher relative to the retrospective rates for teens, sometimes by more than 50 percent, regardless of race or sex. Prime-age women also have consistently higher relative contemporaneous rates, although to a lesser extent than do teens. The observed cyclical pattern for all workers appears to be generated mainly by these two groups, as prime-age white men exhibit retrospective rates almost identical to the contemporaneous in every year.(12)


Analysis of the microdata

The first step in the analysis involves comparing the percentage of workers unemployed in March of year t who in year t + 1 do not report any retrospective unemployment in year t. The results of this exercise indicate that a large number of contemporaneous spells go unreported retrospectively. Over the period 1978 to 1987, between 35 and 50 percent of those reporting contemporaneous spells in March of one year report no retrospective spells in the next March survey.(13) (See table 3.) The rates differ by demographic group. For example, only about 32 percent of prime-age men (between 20 and 59 years old) fail to report unemployment retrospectively, whereas 58 percent of female teens do so. The patterns over time and across demographic groups closely follow those that might be inferred from the above discussion of contemporaneous and retrospective unemployment rates. During periods of recession, the rate of under-reporting falls, while during good times it rises. Teens and women report retrospective spells less often. Furthermore, shorter in-progress spells of unemployment are less often reported retrospectively.


To provide some indication of the types of workers providing inconsistent answers regarding their unemployment status, the probability that retrospective reporting of unemployment varies with the individual's major activity in the survey week or with the reason the unemployment spell began was considered.(14) Tables 4 and 5 show the results of this analysis for different demographic groups. For those who report "looking for work" as their main activity, a significantly lower percentage fail to report their spell retrospectively, compared to those who report "keeping house" or being "in school." While the fraction reporting no unemployment retrospectively is about one-third for those "looking for work," it is roughly 50 percent for those who were "keeping house" and 60 percent for those who report being "at school." Workers whose spell began because they lost a job or were on indefinite layoff were more likely to retrospectively report their unemployment spell compared to those whose spell began because they wanted temporary work. Roughly one-quarter of workers who report "indefinite layoff" and about one-third of those who report "lost job" fail to report unemployment retrospectively, compared to about 60 percent of those who report "wanted temporary work."


These results suggest that, compared to the retrospective measure, the contemporaneous measure of unemployment is more likely to include workers with relatively weak labor force attachment, if one accepts the premise that being in school, keeping house, or becoming unemployed because one wants temporary work are indicators of such weaker attachment.(15)

This report further documents the discrepancies in unemployment rates computed from contemporaneous and retrospective CPS data. A major factor contributing to this difference is that workers counted as unemployed contemporaneously are not retrospectively counted as unemployed for the same period. The data suggest that a disproportionately large share of these workers have relatively weak labor force attachment.

Two hypotheses that could explain these results come to mind. First, workers may simply forget about previous unemployment spells. In fact, George A. Akerlof and Janet L. Yellen argue that this form of "recall bias" leads to the disparity in the two aggregate unemployment rate series.(16) From this point of view, workers with weak labor force attachment would be more likely to forget about their unemployment spells. Second, the difference could result from the structure of the survey itself. Because different questions are asked in the two parts of the survey, the contemporaneous measure may count workers with weak labor force attachment as unemployed, while the retrospective measure would not. Additional research is needed to determine whether either of these, or any other, hypotheses can explain the results presented here.


(1) The Current Population Survey is a survey of about 60,000 households conducted monthly by the U.S. Bureau of the Census for the Bureau of Labor Statistics to collect demographic, social, and economic information on the working age population. (2) See, for example, Richard D. Morgenstern and Nancy S. Barrett, "The Retrospective Bias in Unemployment Reporting by Sex, Race and Age," Journal of the American Statistical Association, June 1974, pp. 355-57; Francis W. Horvath, "Forgotten unemployment: recall bias in retrospective data," Monthly Labor Review, March 1982, pp. 40-43; and George A. Akerlof and Janet L. Yellen, "Unemployment Through the Filter of Memory," Quarterly Journal of Economics, August 1985, pp. 747-73. (3) For more detail, see National Commission on Employment and Unemployment Statistics, "How the Government Measures Unemployment," Counting the Labor Force: Readings in Labor Force Statistics Volume III (Washington, U.S. Government Printing Office,, 1979), appendix. (4) However, when the Bureau of Labor Statistics reports the national unemployment rate, four adjustments are made. (See The Current Population Survey: Design and Methodology, Technical Paper No. 40 (Bureau of the Census, January 1978). These adjustments account for differences in personal characteristics between the sample and the total population, for seasonal differences in labor force patterns, and for loss in coverage due to nonresponse. In addition, an adjustment is made, taking into account that three-quarters of the sample overlap between one month and the next (see below), to improve the reliability of the final estimate. (5) The process described here refers to the supplement after it was changed in 1980. While the format was slightly different prior to 1980, for all practical purposes, the changes are irrelevant. (6) If the definition of a week of unemployment is identical between the two surveys and the labor market is in a steady state, the two measures should be identical. (7) Use of the longitudinal structure of the CPS for labor force analysis is not new. For example, in "Keeping Time: An Analysis of Errors in the Measurement of Unemployment Duration," Journal of Business and Economic Statistics, April 1984, pp. 140-49, Norman Bowers and Francis W. Horvath conducted an analysis matching consecutive monthly CPS data to test for mismeasurement in the duration of unemployment. In "Job search methods of the unemployed, 1991," Monthly Labor Review, Steven M. Bortnick and Michelle Harrison Ports also match consecutive monthly surveys to examine the outcomes of different job search techniques. (8) Because the CPS is a household-based survey, there are no person-specific identifiers, making the matching process somewhat imprecise. The basic framework of matching people involves matching household identifiers and then matching individuals within the household according to age, race, and sex. Because of the household basis of the survey, a respondent cannot be matched if he or she moves. On the other hand, false matches may be made if a new person moves into a household, and has the same characteristics as a person who has moved out. Furthermore, the matched sample is no longer nationally representative, because the nonrandom sample of people who move cannot be matched. On average, roughly two-thirds of those eligible to be matched actually are matched (including those matched incorrectly). The 1985 and 1986 surveys cannot be matched due to changes in the CPS sampling structure. See Kathleen P. Creighton and Robert Wilkinson, "Redesign of the Sample for the Current Population Survey," Employment and Earnings, May 1984, pp. 7-10. (9) Another interesting exercise would consider those workers who, in year t + 1, report retrospective unemployment in year t and estimate the fraction who do not report any contemporaneous unemployment in year t. Unfortunately, because a worker is in the sample for, at most, 4 months of the calendar year, this exercise is not possible with these data. (10) Data on weeks unemployed and weeks in the labor force in the retrospective survey for years prior to 1976 are reported in categories. To estimate the retrospective unemployment rate for these years, the total number of weeks spent unemployed and in the labor force need to be obtained by taking the product of the number of individuals recorded in each category and the mean number of weeks among these individuals. The mean number of weeks for the two variables are estimated from the 1977 through 1982 CPS.

The rates presented here are slightly different from those presented by Akerlof and Yellen, in "Unemployment," pp. 747-73; and Lawrence H. Summers, "Why Is the Unemployment Rate So Very High Near Full Employment?" Brookings Papers on Economic Activity, no. 2, 1986, pp. 339-83. This is because my rates are estimated from the microdata directly. The other sets of rates use published data that report weeks unemployed and weeks in the labor force categorically. In all years, these authors pick the midpoint of the reported range to be the mean. (11) Prior to 1964, only heads of households and/ or their spouses were allowed to answer questions for the entire household in the CPS. (Since then, any household member is allowed to respond to the survey. See Jodie T. Allen, "A Guide to the 1960-71 Current Population Survey Files," Annals of Economic and Social Measurement, April 1973, pp. 189-99.) Because these respondents seem less likely to report that another family member, such as a teenage child, is unemployed, this would tend to lower the contemporaneous unemployment rate and make it more comparable to the retrospective rate as observed in the data. (12) These findings are consistent with those presented by Akerlof and Yellen, "Unemployment," pp. 747-73; and Kevin M. Murphy and Robert H. Topel, "The Evolution of Unemployment in the United States: 1968-1985," in National Bureau of Economic Research, NBER Macroeconomics Annual, 1987 (Cambridge, MA, MIT Press, 1987). However, there is a slight difference in the age groupings used by Akerlof and Yellen, compared to those used by Murphy and Topel and myself. Therefore, a direct comparison of my estimates by demographic group with Akerlof and Yellen's estimates is not possible. (13) The phenomenon of rotation group bias could potentially be explained by those with weaker labor force attachment reporting their unemployment in the first month of the survey, but not in others. To test this, the fraction of contemporaneous spells that were not reported retrospectively for rotation groups one through four separately was estimated, but no difference between groups was found. For more information, see Gary Solon, "Effects of Rotation Group Bias on Estimation of Unemployment," Journal of Business and Economic Statistics, January 1986, pp. 105-09. (14) To determine the reason that a spell began, responses to fl= questions need to be merged. First respondents are asked if they "have a job or a business from which they were temporarily absent or on layoff last week." If yes, then they are asked why. If the response is temporary or indefinite layoff, the respondents are considered unemployed and it is assumed for study purposes that this is the reason that their spells were initiated. If the response to the first question is no, and the respondents are eventually classified as unemployed, they are asked why they started looking for work. For respondents following this path, their answer to this question was taken to be the reason they became unemployed. (15) Additional evidence may be obtained from interpreting the results of Nancy A. Mathiowetz and Greg J. Duncan, "Out of Work, Out of Mind: Response Errors in Retrospective Reports of Unemployment," Journal of Business and Economic Statistics, April 1988, pp. 221-29. These authors conducted a validation study on retrospectively reported unemployment for the Panel Study of Income Dynamics (PSID) by comparing the employment records of a large manufacturing firm to the retrospective unemployment information reported by its employees. Because most of the firms's employees were older, with more tenure, than a nationally representative sample, their attachment to the labor force should be very strong. Consistent with the hypothesis presented in this report, they find that underreporting exists, although it is much less frequent than indicated here. (16) Akerlof and Yellen, "Unemployment," pp. 747-73.
COPYRIGHT 1993 U.S. Bureau of Labor Statistics
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1993 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Current Population Survey
Author:Levine, Phillip B.
Publication:Monthly Labor Review
Date:Aug 1, 1993
Previous Article:Training among young adults: who, what kind, and for how long?
Next Article:Working wives' contributions to family incomes.

Related Articles
Revisions of state and local area labor force statistics.
Strong employment growth highlights first half of 1987.
Labor force data in the next century.
Labor force trends of persons with and without disabilities.
Using gross flows to explore movements in the labor force.
Utilization of labor resources in Japan and the United States: on the basis of U.S. definitions, Japan's unemployment rate surpassed the U.S. rate...
Improving estimation and benchmarking of State labor force statistics.
Lower unemployment in 2005: CPS data show that the labor market continued to improve in 2005, as employment grew substantially and unemployment...
Household survey data show labor market improvements in 2006: unemployment declined, employment increased, and earnings were about in line with...
Household survey indicators weaken in 2007.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters