“Earnings Inequality Trends in the United States: Nationally Representative Estimates from Longitudinally Linked Employer-Employee Data”, John Abowd (Cornell University and U.S. Census Bureau), Kevin McKinney (U.S. Census Bureau), Nellie Zhao (Cornell University)
We track sources of earnings inequality using the statistical technique introduced to the labor economics literature in 1999 (Abowd, Kramarz and Margolis, Econometrica 1999). When this technique has been used in Europe (Card, Heining and Kline QJE 2013 for Germany, in particular), the biggest contributor to the increase in earnings inequality appears to be increased employer-level heterogeneity (called the firm effect in AKM). Using the Census Bureau’s Longitudinal Employer-Household Dynamics Infrastructure data for 1990-2013, we show that with respect to the U.S. data, the CHK result does not hold. There has been very little change in employer-level earnings heterogeneity in the U.S. when one compares wage measures similar to the ones used to analyze the European data. European administrative databases allow one to construct something akin to a wage rate (usually, the amount that would be earned if an individual worked full-time full-year). The American data does not directly allow that. We develop a statistical approximation to the full-year full-time wage rate, using integrated Current Population Survey, Census 2000, and American Community Survey data. Using that measure, the earnings inequality trends in the U.S. look more similar to the European analyses.
But, for the purposes of studying earnings inequality, considering only the wage rate, and not the amount of time a person actually works, is seriously incomplete—especially in the U.S. where there is very little statutory employment security except in the public sector. The most important determinant of increased earnings inequality in our analyses is changes in labor force attachment (weeks worked in the year, hours worked per week).
In attempting to estimate how important the labor-force attachment component is, we reconstruct the work-eligible population (18-70) for each year from 1990-2013. The administrative records database developed at the Census Bureau uses an encrypted SSN to track individuals. The researcher can tell if the number that was encrypted is a valid SSN, and can also access the demographic details and employment history associated with the underlying SSN. In our model, there are two kinds of SSNs that are suspect: ones that are not valid (this means that the employer reported earnings in a state’s UI system for an SSN that was never issued) and ones associated with demographic characteristics that mean it is unlikely that the owner of the SSN used it (leading case: the SSN was issued to a person who was less than 10 years old in the year during which the SSN was used to report UI eligible earnings). Our working hypotheses are: (1) the use of an invalid SSN reflects the work of a single undocumented immigrant, so we add that person to both the eligible population and the working population and (2) the use of a valid SSN issued to someone who appears to be too young (or too old) to work legally represents one person in the population (not working, not immigrant; i.e., eligible to get an SSN by virtue of birth in the U.S.) and at least one other person both working and in the work-eligible population, who is an undocumented immigrant.
Getting the non-working work-eligible population as accurate as possible is important because, especially during the Great Recession, many persons had no income from work for a full calendar year. We have no trouble finding these people for properly documented native-born and immigrant subpopulations, but we have to estimate how many work-eligible non-documented immigrants are still in the U.S. looking for work in any given year.
We also link data from the 1992-2012 Economic Censuses. These data are used to construct a measure of surplus per worker (revenue minus factor opportunity costs) for every private establishment in the censuses. These data show similar results for the population of working persons employed in the private sector. In particular, they show that there has not been an increase in overall earnings variability for this population.
“Total Variability Measures for Selected Quarterly Workforce Indicators and LEHD Origin Destination Employment Statistics in OnTheMap”, Kevin McKinney (U.S. Census Bureau), Lars Vilhuber (Cornell University and U.S. Census Bureau), John Abowd (Cornell University and U.S. Census Bureau), Andrew Green (Cornell University)
We report results from the first comprehensive total quality evaluation of three major indicators in the U.S. Census Bureau’s Longitudinal Employer-Household Dynamics (LEHD) Program Quarterly Workforce Indicators (QWI): beginning-of-quarter employment, full-quarter employment, and average monthly earnings of full-quarter employees. Beginning-of-quarter employment is also the main tabulation variable in the LEHD Origin-Destination Employment Statistics workplace reports as displayed in OnTheMap (OTM). The evaluation is conducted using the multiple threads generated by the edit and imputation models used in the LEHD Infrastructure File System. These threads conform to the Rubin (1987) multiple imputation model. Each implicate is the output of formal probability models that address coverage, edit and imputation errors. Design-based sampling variability and finite population corrections are also included in the evaluation. We derive special formulas for the Rubin total variability and its components that are consistent with the disclosure avoidance system used for QWI and LODES/OTM workplace reports. These formulas allow us to publish the complete set of detailed total quality measures for QWI and LODES. The analysis reveals that the three publication variables under study are estimated very accurately for tabulations involving at least 10 jobs. Tabulations involving three to nine jobs have acceptable quality. Tabulations involving one or two jobs, which are generally suppressed in the QWI, have substantial total variability but their publication in LODES allows the formation of larger custom aggregations, which will in general have the accuracy estimated for tabulations in the QWI of similar magnitude.
“Formal Privacy Protection for Data Products Combining Individual and Employer Frames”, Ashwin Machanavajjhala (Duke University), Samuel Haney (Duke University), Matthew Graham (U.S. Census Bureau), Mark Kutzbach (U.S. Census Bureau), Lars Vilhuber (Cornell University and U.S. Census Bureau), John Abowd (Cornell University and U.S. Census Bureau)