NC State Faculty Salary Equity Study
As the result of a competitive bid process, the North Carolina State University (NCSU) Office for Equal Opportunity (now the Office for Institutional Equity & Diversity) retained Haignere, Inc., to conduct university-wide data analyses so as to diagnose whether or not systemic gender and race faculty salary differences exist. NCSU has conducted similar annul faculty salary equity studies since 1982.
The database for this study includes the 1581 full-time faculty members at NCSU in the fall of 2000. This population differs from the populations studies in previous studies because of the addition of two groups: faculty members with an administrative title below department head, and distinguished, named and titled faculty members. University Planning and Analysis compiled the study's database with assistance from the Office of the Provost.
Assessing the Potential for Variables to Mask or Suppress Salary Inequities
Even a cursory review of the methodological literature concerning the assessment of gender bias in faculty salaries reveals substantial discussion of what variables should and should not be included. This discussion revolves around "tainted variables." Tainted variables are those that are likely to have discrimination embedded in them and, thus, mask or suppress gender effects. For example, if height were included in a salary disparity analysis where gender bias exists, the shortness of female faculty relative to male faculty could explain much of the gender differences in salaries.
We estimated whether or not the variables Rank, Tenure, Administrative Title, and Rank Modifiers may act to suppress findings of salary bias using frequency tables displaying the representation of white men relative to female and minorities. The results cannot be interpreted to demonstrate bias because frequency tables do not control for other variables. For example, low representation of women in the full professor rank could indicate a glass ceiling at the full professor level, or it could merely reflect "time in the pipeline." The objective of the frequency table analyses is to establish whether or not it is necessary to systematically vary a variable's inclusion in the analyses so as to estimate whether or not it is functioning as a suppressor variable. To the degree that the university ca address the under-representation of women and minorities in the categories examined some of the complexity of diagnosing systemic gender and race salaries differences can be minimized.
The frequency tables indicate that women, including minority women, are disproportionately visiting and less likely to be in research positions. Women do not hold distinguished professor rank modifiers in the proportions that men do. Women are less likely to be in tenure-track lines than are men even when controlled for degree level. Minorities are less likely to hold below department head administrative positions.
Concerning rank, minorities and women are less likely to have made it into the full professor rank. White women are less likely than minority men to be full professors and minority women much less likely to be full professors than any other race/gender category. Even though women and minorities predominate in the visiting ranks, only one (1.4%) white woman holds a senior rank visiting appointment. By comparison, ten white males (17%) hold senior rank visiting appointments. Over two-thirds of the women and minorities in the visiting ranks are lecturers compared to half of the white males.
The results of the frequency-distribution analyses indicate that proportional representation exists in the rewarding of rank, non-tenure track positions, and rank modifiers. Thus, it is feasible that these variable mask gender and/or race disparity when included in the regression analyses of salaries. The classic dilemma regarding potential confounding variables is that excluding them may overestimate disparity while including them may underestimate disparity. We address this dilemma by systematically excluding each potentially tainted variable with the exception of rank. Rank is included in all analyses. Even if there is considerable evidence bias in current rank, we recommend a conservative approach of including Rank in the analyses. Having done so, however, it is important to remember that the results probably underestimate the amount of disparity that exists in salaries.
Diagnosis - Do systemic race and gender disparities exist?
The university wide analyses indicate that there is reason to be concerned about both gender and race salary disparities. When we subset the NCSU faculty population so as to eliminate all potential suppressor variable effects by studying only tenure track faculty who do not have rank modifiers, the results indicate roughly $1000 annual salary disparity between women faculty and comparable white males. For minority males, there is a disparity in the neighborhood of $2000 between them and comparable white males. These amounts are roughly equal to the midrange of the disparities indicated when we systematically vary the potentially tainted variables included in the regression analyses for the whole NCSU faculty population. In our groups, these are substantial salary disparities that need to be addressed. We suggest a group/systemic approach to remedy based on the greater consistency of this approach with the multiple regression statistical methods, ease of application and greater fairness to both high and low performing women and minorities.
NCSU has an impressive history of doing salary equity studies annually. These studies have emphasized college level analyses and used the white-male equation approach. Little attention has been paid in the past to the university level analyses. It remains to be seen whether the university level analyses will be used differently this year. If it is determined that salary adjustments will be made based on the university level analyses, it may be important to focus further on the variations in the results between the three different regression models.
Conducting college level analyses should pose few problems at the four largest NCSU colleges: Agriculture and Life Sciences, Humanities and Social Sciences, Engineering, and Physical and Mathematical Sciences. At the College of Veterinary Medicine and the remaining five of the NCSU colleges, the small number of faculty may lead to methodological complexities. The general rule of five cases (faculty members) per independent/predictor variable should be respected. At the smaller colleges, respecting this limit can mean combining or eliminating some variables. White-male analyses may be particularly problematic for these smaller colleges. Not only are there fewer faculty members in white-male analyses but calculating the average residuals for the women and minorities requires excluding any women and minorities for whom there is no white-male match.
Submitted by Haignere, Inc., July 2001
For additional information on this study or the history of the study at NCSU, please contact the Office for Institutional Equity & Diversity at 919-515-3148.