The Employment Number's Understatement

The official unemployment rate reported every month by the U.S. Bureau of Labor Statistics is one of our most important and widely watched economic statistics. The data used to calculate it are collected by the Bureau of the Census in its Current Population Survey (CPS), a sample of households nationwide.

Studies going back to the 1970s have shown that the official unemployment rate is significantly understated in periods of high unemployment, like now.

The data on unemploymnent collected by the Census Bureau each month are subject to nonsampling as well as sampling error. Errors in the CPS arise, for example, when interviewers misrecord answers or when respondents misinterpret questions or proxy respondents don't answer correctly.

Determining who is employed is fairly straightforward in the CPS and respondent errors are relatively small - the householder either has or does not have a job. Determining who is unemployed, however, is more complex and prone to error.

To be classified as unemployed, a person without a job must have actively looked for work in the previous four weeks and be currently available for work. Active jobseeking includes such specific activities as contacting an employer, going to a job interview, asking friends or relatives about a job, contacting a school employment center, sending out a resume, filling out a job application, answering or placing an advertisement for a job, or checking a job register. Less direct activities, such as reading about job openings in the newspaper or online don't count, and passive jobseekers are classified as not in the labor force.

Suppose, for example, in an interview there was only one person at home to answer questions in a household where more than one person of working-age resides. The rules permit a proxy respondent to answer questions about members of the household who are not present.

If an absent person is classified as not employed, the interviewer asks about job search, which will determine if that person is recorded as unemployed or out of the labor force. The proxy respondent, however, may not know that the absent person actively looked for a job in the past four weeks and so says no to the job search query, causing the person to be erroneously classified as not in labor force instead of unemployed.

Studies have found that response errors in unemployment are cyclical - they are relatively small when unemployment is low but increase as unemployment rises. The official unemployment rate has been shown to be significantly understated when the jobless rate reaches recession levels.

My own earlier research based on data collected in Census Bureau reinterviews (where more highly skilled interviewers repeat questions about unemployment and employment for a subsample of the CPS) found that when the average reported unemployment rate exceeded 7 percent in 1974 - 1975, it was understated by 0.6 percentage point.

Professors James M. Poterba (MIT) and Lawrence H. Summers (Harvard) also analyzed CPS reinterview data and found that when the reported unemployment rate was over 7 percent in mid-1976, "ten percent of the truly unemployed had been classified as not in the labor force." This was also equal to a 0.6 point understatement in the official unemployment rate.

Subsequent research in 1998 by Joseph L. Gastwirth (George Washington University) and Michael D. Sinclair (Response Analysis) that analyzed unemployment misclassification estimated the understatement to be slightly less. When the officially reported unemployment rate was in the 9 percent range in 1982-83, they found the understatement in the reported unemployment rate averaged 0.53.

Unfortunately, Census Bureau data on the response error in unemployment are not available for the 2008 - 2010 period of high unemployment.

However, a 2010 paper by economists Shuaizhang Feng (Princeton) and Yingyao Hu (Johns Hopkins) analyzed misclassification error in the estimation of unemployment rates using panels formed by matching CPS monthly data sets, an alternative approach that did not rely on reinterview information. They found that for 1996 to 2009 "the corrected monthly unemployment rates are 1 to 4.6 percentage points...higher than the official rates, and are more sensitive to changes in business cycles."

If the research results of the above four studies were used to recalculate the latest official 9.0 percent unemployment rate, the adjusted rate would range from 9.6 percent to well into double digit territory.