The presence of unobserved heterogeneity between union and nonunion workers in the same skill category introduces another potentially important question: how to distinguish the true union wage effect Aw(c) from the heterogeneity component 6(c). A natural solution to this problem is to use longitudinal data on union status changers, as in Card (1996) or Lemieux (1992). Both of these papers use longitudinal samples stratified into observable skill groups on the basis of predicted wages in the nonunion sector. My paper considers five “skill quintiles” whereas Lemieux considers three “skill tiers”. The empirical results in these papers suggest two important conclusions. First, the true union wage effect is higher for less-skilled workers and lower for high-skilled workers. For example, the results in Card (1996) for U.S. men suggest that the union wage effect ranges from about 30 percent for men in the bottom 20 percent of the skill distribution to about 10 percent for men in the top 20 percent of the skill distribution. Second, union workers with lower observed skills are positively selected, whereas those with higher observed skills are negatively selected. In other words, 6(c) > 0 for low skill groups and 0(c) < 0 for high skill groups. In the analysis below I use these findings to make rough adjustments to the observed union wage gaps for workers in different deciles of the predicted wage distribution.


This paper uses Current Population Survey (CPS) data on wages from the May 1973 and 1974 surveys, and from the 12 monthly surveys in 1993. The May 1973 sample is the first CPS that contains both union status information and wage data for individuals’ current jobs. Thus, this sample forms the earliest benchmark against which to compare current levels of unionization and wage inequality. In view of the relatively small sample size of this survey, I elected to pool the May 1973 and May 1974 data. The 1993 CPS is the last survey prior to the introduction of a new computer-assisted survey instrument that substantially changed the nature of the earnings questions. I therefore use this sample to measure recent patterns of unionism and wage inequality.

Tags: , ,