Earnings of workers change from year to year for various reasons. This earnings volatility has important links to poverty, rising income inequality, declining economic mobility, and the use of social welfare programs.  Much of what is known about volatility and inequality comes from survey data, which generally offers a broad collection of variables. However, survey data suffers from data quality issues such as item nonresponse (refusing to answer survey questions about earnings) and measurement error (failing to report earnings accurately).  These data quality issues create obstacles in drawing conclusions about earnings from survey data.  Administrative data on earnings and other related topics, on the other hand, avoid some of the measurement pitfalls of surveys.  However, administrative data alone do not include important variables such as education, race, and family structure necessary to fully investigate the reasons and trends of earnings volatility.  This project seeks to reconcile the diverging results from survey and administrative data by linking a large survey data that is widely used to understand U.S. poverty rate from income and earnings to administrative data on worker earnings. By linking these two datasets, this project explores important questions on earnings volatility trends through time, underlying demographic elements that cause earnings volatility, the effects of economic shocks on earnings and how compensation structure may lead to shifts in earnings. Overall, these questions advance our understanding of how earnings change, which is crucial in explaining rising inequality and designing social welfare programs.

The project consists of four studies that utilize restricted-access survey and administrative from the Current Population Survey (CPS) Annual Social Economic Supplement (ASEC) and Social Security Administration?s Detailed Earnings Records (DER). The ability to observe both multiple reports of earnings (administrative and survey) combined with the short panel structure of the CPS and the full earnings history available in the DER, allows identification of permanent income, as well as measurement error structure.  The availability of DER earnings for those who are non-respondents in the CPS allows identification of the distribution of income for non-respondents. Combining these two aspects allows the investigation of earnings levels and volatility in ways that neither survey nor administrative data alone could accomplish. The first project specifies a finite mixture model of earnings response to examine differences between continuous survey responders to both continuous nonresponders and switchers from response to nonresponse or vice versa. The second project examines whether there are differences in levels and trends in volatility, adjusting for panel attrition, nonlinkage between the ASEC and DER, and measurement error. The third project provides new estimates of permanent and transitory shocks to earnings. Finally, fourth project is on variance decomposition of volatility to isolate how much of the level and trend differences are driven by differences in annual hours of work, hourly wages, or the covariance of hours and wages.  

The key findings are as follows. First, measurement error exists in both survey data and administrative records, suggesting that the administrative data are no panacea for missing survey responses. However, we also find evidence of so-called good reporters (respond to the survey and provide accurate amounts) and bad reporters (fail to respond to the survey, but when they do, provide inaccurate amounts), suggesting that it may be possible to isolate the primary source of measurement error in longitudinal survey data by restricting attention to continuous reporters. We also find that while item non-response is rising over time, the quality of the remaining respondents data has improved.  Second, once data are restricted to continuous survey respondents we find that the level and trends in volatility of earnings among both men and women obtained from survey data align quite closely to those obtained in administrative data. For the two decades from 1995-2016 we find significant cyclical fluctuation in male earnings volatility, but no overall trend, while female earnings exhibit little fluctuation over the business cycle but a downward trend, with levels of male and female earnings volatility converging. Third, the stability of male earnings over the past two decades stems from offsetting rising permanent shocks and declining temporary shocks. These offsetting forces are most pronounced among Black men without a college education. The rising permanent shocks of Black men point to a worsening of long-term relative economic status across race. 

To date, two papers have been published in the peer-reviewed Journal of Business and Economic Statistics, and a third has received an invitation to revise from the Journal of Labor Economics.

Last Modified: 07/06/2023
Modified by: James P Ziliak