The objective of the research is to estimate the causal effect of unionization on the distribution of wages. A fundamental obstacle to measuring the effect of unionization on wages is selection bias: wages within unionized plants may differ for reasons other than the fact that workers there are represented by a union. Differences in either worker or plant characteristics could give rise to an observed difference between the distribution of wages inside and outside of the unionized sector, confounding the true effect of unionization. This study seeks to overcome these selection issues by using a regression discontinuity (RD) research design. The critical feature of the design in this context is that in order for a union to represent workers at a plant, a majority of the workers must vote for it in a certification election. If plants and workers where the union barely won the election are comparable to plants and workers where the union barely lost, then the effect of unionization on the distribution of wages can be estimated. The research implements a newly developed methodology to estimate distributional effects using the RD design. The methodology allows for unionization to have heterogeneous effects, and allows for selection effects in union status. The estimates produced by this methodology are consistent for the effect of unionization on the distribution of wages for plants that were close to the election share cutoff of 50 percent. In order to measure the effect of unionization on the distribution of workers' wages, the research uses individual-level wage data from the U.S. Census Bureau's Longitudinal Employer-Household Dynamics (LEHD) program. The Employment History File (EHF) contains quarterly records of individuals' UI-covered earnings. This project uses the LEHD's Unit-to-Worker imputation to merge individual wage data with plant-level NLRB union certification election data.
The role of unions in the labor market and their effect on the distribution of wages have long been an important topic in labor economics, but especially over the past three decades, as the decline in unionization rates may be a contributing factor to the increasing inequality observed in the US income distribution. Research to date has tended to focus on average effects because of available econometric techniques. A number of recent studies (e.g., DiNardo and Lee, 2004) have introduced quasi-experimental methods, including RD to estimate the causal effect of unions on economic outcomes. This is a promising approach to overcome the classical selection bias. Many studies have also considered the effect of unions on inequality, but none to date have combined quasi-experimental methods with a focus on distributional effects. The intellectual merit of this study lies in the application of new econometric techniques to estimate using a quasi-experimental design, the distributional effect of unionization, and in the use of large administrative datasets that are likely to provide better accuracy and statistical power than survey data used in many previous studies.
Broader Impacts: The causal effect of unionization on the distribution of wages is important to to discussions of inequality, both in policy and academic circles. The broader impact of the research includes informing policy debates around the role of unions in the labor market (for example, the Card Check Bill recently introduced in Congress), as well as the academic debate about the relative importance of institutional change versus skill-biased technological change. Another broader impact is the creation of a single dataset with merged administrative data on both individual workers and plants which will of value to future researchers.
