Wage Data: The wage data for tables 2, 3, and A1 come from the Outgoing Rotation Group (ORG) files of the Current Population Survey (CPS) for 1979-97. The sample includes all wage and salary workers, ages 18 to 64, with positive hourly wages between $0.50 and $100 in 1989 dollars. For hourly paid workers, the reported hourly wage is used; for weekly workers, the hourly wage is constructed by dividing usual weekly earnings by usual weekly hours. Top-coded weekly earnings were replaced with the estimated value of the mean weekly salary above the top code, using the assumption that the upper "tail" of the distribution follows a Pareto format. Quantile estimates, such as those shown in figure 6, use a smoothing procedure to accommodate "clumps" in the reported distribution of earnings. The construction of this wage series is discussed in greater detail in Webster (1999).
Table A1, Oaxaca decomposition: The wage data for this table also come from the CPS ORG, as described above. We use the following equation to decompose the changes in characteristics (Xs) and returns (Bs):
where p is the change in the probability of low-wage work (in our case, the change in the likelihood of earning less than $7.90 in 1997 dollars), B bar is the average of the returns between the two time periods, X bar the average of characteristics between the two time periods, a the intercept term, and I indexes the independent variables, 1 through k. Variables in the regression include education, potential experience (age-education-6), industry, occupation, race, region, and marital status. The regressions use the CPS ORG population weights, and separate equations were estimated for men and women.
Thus, the first term represents that part of the change attributable to changing characteristics, the second term represents that part of the change attributable to shifts in returns, and the third term captures the change in the intercept.27