The Pay Equity Model 1997-2015

Pay Equity Model *

UC Irvine’s annual faculty pay equity study applies the model recommended by the American Association of University Professors (AAUP) to identify women and minority faculty members who appear to be underpaid. It should be noted that “minority” is consistent with federal definitions for employment affirmative action, and includes Black/African-American, Asian/Pacific Islander, Native American, and Hispanic regardless of citizenship or country of origin.

UCI’s pay equity model methodology has been reviewed and endorsed several times by campuswide ad hoc faculty committees, with the understanding that the model is an appropriate screening device — although not the only screening device — for an assessment of equity in faculty compensation. This view is consistent with AAUP recommendations. The purpose of the model is to flag individuals and academic units with salaries that may require closer scrutiny. As it does not include any subjective measures of quality or merit, it is expected that some faculty members will have results that are not explained well by the model, which relies exclusively on quantifiable objective measures.

UCI Pay Equity Study

Each academic year’s pay equity study is conducted for ladder rank faculty based on the October Payroll/Personnel data combined with data from the Office of Academic Personnel. The data include only ladder rank faculty members who were on active pay status as of the end of October for each academic year. The analysis excludes faculty administrators such as Deans, the Chancellor, Vice Chancellors, Provosts, and Vice Provosts. General campus faculty whose salaries are paid on a fiscal year basis are not excluded, but for purposes of the study we only consider their academic year ladder rank salary, or its equivalent.

Variables and definitions are reviewed each year. Every effort is made to assure accuracy in the input data. Although minor adjustments in the data can impact year to year comparability, overall trends remain unaffected. Consistency is important to the analysis of multiple years of study results, but we remain open to suggestions from faculty members who would like to test the model for validity and reliability.

Regression Analysis and the Estimation Equation

The model uses data for white men to calculate a predicted salary. Two factors that provide objective measures of experience are used to define the faculty member’s background: the year of the highest degree (usually a Ph.D.) and the year of hire into the ladder ranks at UCI. A third factor, the degree indicator, describes the level of the highest degree (M.S., Ph.D., etc.) in units where there are Acting Assistant Professors who have not completed their Ph.D. or faculty members with a variety of terminal degrees.

In technical terms, the model is based on a multivariate linear regression, with salary as the dependent variable and year of degree, year of hire into the ladder ranks at UCI and level of degree (the latter only in some units) as the independent variables. For most academic units, these background factors do a very good job predicting the salary of white men (R2 of .50 to .80).

The regression analysis is conducted separately for each academic unit, resulting in different values in the equation to calculate estimated salaries. At UCI, our revised model uses predictors or independent variables defined as follows:

  • degree year: four-digit year the highest degree was awarded (e.g., 1987)
  • UCI ladder ranks hire year: four-digit year the faculty member was hired into the ladder ranks at UCI (e.g., 1990)
  • degree indicator: an indicator of the highest degree
    (0 = Baccalaureate or lower; 1 = Master’s; 2 = Doctorate)

The variable degree indicator does not contribute to the regression equation in academic units where all faculty members have the same degree level.

Standard Error

The standard error is an index indicating how large the potential error is in predicting one’s salary from one’s degree indicator, degree year, and UCI ladder ranks hire year.

Calculating the Residual

After determining the estimation equation for each academic unit, the next step in applying the AAUP model is to calculate the residual for women and minority faculty members. The residual is the difference between actual salary and salary predicted by the regression equation. A negative residual indicates that the salary is lower than the amount predicted for white men in the same academic unit who have the same attributes. A positive residual indicates that the salary is higher than the predicted value.

To evaluate the residual, one needs to consider the standard error for the school for that year (i.e. the standard deviation of the residuals of the white men). If the residual falls within the standard error, the residual falls within the range expected for approximately two-thirds of the white men. The standard error is noted at the top of each graph.

This study identifies large negative residuals (falling outside of the standard error) which are brought to the appropriate dean’s attention for further analysis. The study does not address why someone’s residual might be negative or positive. Factors that might affect the salary of any individual faculty member include but are not limited to the rank and step at initial hire, different rates of advancement through ranks and steps such as accelerations or no-actions, off-scale salaries due to retention offers, and above-scale salaries.

*Revisions to the Methodology

2014-2015 Study Year: We included the School of Law, Program in Public Health, and the Department of Pharmaceutical Sciences in the study (they were previously excluded due to the small population of their ladder rank faculty).

2003-2004 Study Year: Responding to questions and feedback over the years, we decided to drop year of birth as an input variable, as of 2003. Of all the variables that measure years of experience, year of birth consistently contributes the least to the regression equation. For 2003, we compared results with and without this variable and found the differences overall to be insignificant. Acknowledging that any adjustments to the model can impact year-to-year comparability, we reran data as far back as 1998 using the revised model.

Comments are closed