and each of seven organizational units. The base group for the regional comparison was the National Office, and the base group for the organizational comparison was the Employment Standards Administration. After examining the results using individual dummies for each region and organizational unit, the dummy variables were collapsed into composite dummy variables, five regional ones and three organizational ones. A twofold criteria guided the collapsing process: (1) that tne individual dummies have a similar affect on salary and (2) that they represent organizations of comparable size. Race and sex. Dummy variables also were used to study the affect of race and sex on salary. Two different approaches for representing race and sex were utilized. First, race dummies for black and nonblack minority with white as the base group were used together with a sex dummy for female with a male base group. Second, the base group was established as white male and individual dummies were used for each of the following groups: white female, black male, black female, nonblack minority male, and non black minority female. C. The Theoretical Models. Various different models were developed. The basic model for all, however, was the following: annual salary 1 f(race, sex, education, age, government service, geographic Several We believed that annual salary was a function of the factors identified in the equation above, and we believed that they would all be statistically significant. initial computer runs were made to determine the best variables to represent age and experience, to examine for comparable affects among the various geographic and organizational dummies prior to collapsing, and to observe the affect of the educational degree dummies. These runs showed that, as in the Rees and Schultz Chicago labor market study, the best variables to use for age and experience were the natural logarithm of years of government service and years of age plus age-squared. However, the natural logarithm of age alone produced nearly as good a fit as Nevertheless, it was the Rees and Schultz variables. decided to use the logarithmic-quadratic combination because it kept both age and years of government service in the equation; it was felt that when models were developed individually for the different racial-sexual groups, the importance of age or government service as a determiner of salary might vary. For example, service might be more important in predicting the salary of whites, while age might be more important in predicting blacks' salaries, or vice versa. The initial runs also showed that using the educational degree dummies together with the number of years of completed education significantly improved the prediction over that obtained by using just years of education. This would indicate a relationship between education and salary in which salary increases at a uniform average rate for each additional year of education, but then jumps a certain amount over the uniform average increase at those educational years-completed levels where degrees are normally earned. (i.e., roughly at 12, 16, 18, and 20 years). Finally, as mentioned earlier, these initial runs provided the basis for collapsing the eight regional dummies into five and the seven organizational dummies into three. Using the results of these initial runs, we can further specify the basic model given at the beginning of this section. We now hypothesize that salary is a linear function of the following variables: annual salary = f(race and sex, years of education, educational degrees, age plus age squared, natural log of years of The hyphenated regional and organizational variables in the equation show how the regions and A&O's were grouped in the collapsing process. As discussed in the "variables" section above, the educational dummies used were degree, advanced degree, two-year certificates, and no high school diploma; the reference group was employees with a high school diploma. This basic model was applied in individual computer runs on the different racial-sexual groups. The only variants were the choices of race and sex dummies. Using the basic model, then, salary determination models were developed for various groupings as follows, and with the form of race-sex dummies as indicated. The correlation was performed on the National Institutes of Health Computer System using a regression program made available by the Bureau of Labor Statistics. The program, called Step Supra, provides for stepwise multiple linear regression analysis. Program options were selected to yield the following output: a listing of the imput data; a table of means, standard deviations, and sum of squares; a correlation matrix; a table of beta coefficients and elasticities at the means; and a plot of |