Rectifying Shortcomings in Measured Employer-to-Employer Flows in the U.S. Labor Market

Economists pay close attention to the rate at which workers in the U.S. labor market move from one job to another without a break in employment. This employer-to-employer (EE) reallocation implies that workers are undertaking job searches while working, and employers are competing for these workers, promoting seamless employment transitions. The direct reallocation of workers between firms has been shown to be an important source of U.S. productivity growth.¹ Moreover, jumping directly from one employer to another can provide an important boost to workers’ lifetime earnings — as opposed to finding a new job after experiencing unemployment, in which case workers tend to end up with a job that pays significantly less than their previous job.² Total EE flows in the U.S. labor market are large — about the same volume as the sum of worker flows moving from unemployment into jobs and from out of the labor force into jobs.

In their paper, “Measuring Employer-to-Employer Reallocation,” Shigeru Fujita of the Philadelphia Fed, Giuseppe Moscarini of Yale University and the Philadelphia Fed, and Fabien Postel-Vinay of University College London improve a measure of EE flows derived from the monthly Current Population Survey (CPS) — an important source of real-time information on the U.S. labor market — by accounting for missing answers to a pertinent survey question.

The CPS, which surveys about 60,000 households, was redesigned in the mid-1990s to include a new question: Was a person’s employer the same in the current month as in the previous month? The authors call this question “SAMEMP.” In an earlier paper, Federal Reserve economists Bruce Fallick and Charles Fleischman used responses to SAMEMP to create a monthly EE series and an average EE monthly transition probability (defined as the ratio of the total number of persons that answer “no” to SAMEMP to the total number employed in the previous month). Their data series has become a standard in the field of labor economics.³

In 2008, the U.S. Census Bureau instituted a new interviewing policy for the CPS that directly impacted the SAMEMP response rate: As of 2008, respondents could opt out of answering the question for privacy reasons. This revised policy reflects the fact that the CPS covers household units over consecutive months, and each month only one person answers the survey for all household members. Without the privacy protection, a household member (who is not necessarily a family member) may learn about job-related information that another household member provided in an earlier month. When Fujita, Moscarini, and Postel-Vinay found that a significant number of respondents chose not to answer the SAMEMP question following the implementation of the new interviewing policy, they decided to explore how to improve the measurement of EE flows.

The authors began by reconstructing EE transition data for the period 1997–2020 using Fallick and Fleishman’s method. This raw data revealed a dramatic drop in EE flows in the U.S. labor market starting in 2007. Moreover, EE flows followed a long-term (secular) downward trend without a clear recovery long after the Great Recession ended. They also observed that a large and abrupt increase in missing answers to the SAMEMP question occurred from early 2007 through 2009 and that the number of missing answers continued to increase gradually through 2015. The authors argue that the secular decline in the EE transition rate since 2007 was largely due to the increasing share of missing answers.

The authors then developed a selection model that accounts for both observed and unobserved worker characteristics. Using this model, they impute missing answers to the SAMEMP question and create a new aggregate EE time series.

Fujita, Moscarini, and Postel-Vinay find that observed worker characteristics such as demographics are relatively unimportant for imputing missing answers. (That is, the series that incorporates only the observed characteristics in the imputation procedure is very similar to the one that simply ignores missing answers.) However, respondents with certain unobserved  characteristics were more likely than other respondents to decline to answer SAMEMP. By taking advantage of the timing of sudden jumps in missing answers and changes in EE rates among valid answers around those jumps, they were able to account for the impacts of unobserved worker characteristics. Their results indicate that the CPS’s new survey design created a large downward bias in measured EE transitions. The authors’ revised EE series, by adjusting for missing answers, shows a reduction in the secular decline of EE flows by half as compared to the unadjusted data over the last 14 years. Their EE series reached a full recovery in 2016, followed by a mild decline before dropping sharply during the early days of the COVID-19 pandemic. This is the first time researchers have documented the behavior of EE flows during the COVID-19 recession.

The authors show that their adjusted EE rate, like the unadjusted rate, is generally procyclical, which means that transitions from one job to another occur less frequently during an economic downturn. However, compared to the unadjusted data, their adjusted EE series is less volatile, drops substantially less, and is in closer congruence with the U.S. business cycle, including during the past two recessions.

In summary, the authors’ new aggregate adjusted employer-to-employer (EE) time series shows no downward trend since 2000. Thus, it eliminates the appearance of a decline in the dynamism of EE transitions in the U.S. labor market, a decline that some researchers had previously argued was happening. The authors’ findings have important implications for labor economists and policymakers who want to understand the nature of the labor market, including the cyclical reallocation of labor between firms, industries, and occupations.