UNH ADVANCE Salary Equity Analysis Report February 17, 2015

Transcription

1 UNH ADVANCE Salary Equity Analysis Report February 17, 2015 Prepared by the UNH ADVANCE Salary Equity Team: Ben Brewer, PhD Student, PAUL Martha Gleason, Compensation Manager, UNH HR Mary Malone, Associate Professor, COLA Carolyn Mebert, Associate Professor, COLA Leigh Anne Melanson, Associate Provost for Academic Administration Christine Shea, Professor, PAUL, UNH ADVANCE Co-PI, Committee Chair Neil Vroman, Associate Dean, CHHS Craig Wood, Associate Professor, PAUL Introduction In May of 2014, the ADVANCE Salary Equity Committee convened to analyze the equity of salaries for UNH tenured and tenure-track faculty. The committee focused on gender equity, and designed a study to determine if there were significant differences between the salaries of female and male faculty, after controlling for all relevant factors typically related to salary (e.g., rank, years of service, etc.). This report summarizes the committee s findings. Most importantly, the committee found that gender did not have a direct impact on salary. Controlling for all other factors, the analysis indicated that female faculty did not earn significantly lower salaries than male faculty. However, gender did have an indirect impact on salary. Gender was a significant and powerful predictor of rank, and rank in turn significantly predicted salary. Female faculty were 70% less likely to be full professors compared to their male counterparts, and this difference was statistically significant. In this report, the committee elaborates on these findings and discusses the methodology utilized in this analysis. It concludes with recommendations for improving salary equity at UNH, primarily by focusing on the linkage between gender and rank. Methodology The Provost s Office at the University of New Hampshire provided data for this ADVANCE salary report. Based upon these data, we constructed a dataset at the level of the individual faculty at UNH. We included all full-time and part-time tenured and tenure-track faculty as of May 1, 2014 in this report. The analysis excluded any former administrators who returned to the faculty. Table 1 lists the variables included in this analysis, as well as their measurements. To select these variables, we relied on the most recent AAUP report, Paychecks: A Guide to Conducting Salary- Equity Studies for Higher Education Faculty (Second Edition). The Paychecks report provided a valuable guide for the design and methodology of our study. 1

2 Table 1: Variables in Salary Equity Analysis Variables Measurement Faculty Salary Dollar amount of tenured and tenure-track faculty salaries as of May 1, 2014 Gender Women = 1; Men = 0 Employment Months Length of appointment year (9 months or 12 months) Full Time Equivalence Standardization of salaries; all reported salaries adjusted to be nine month, full-time salaries Degree Highest degree earned; dichotomized into doctorate (Ph.D., J.D., Ed.D) and non-doctorate (all MA and BA degrees) Year Highest Degree Earned Year faculty earned their highest degree Year Hired at UNH Year faculty hired at UNH Current Service Years 2014 year hired at UNH Current Rank Current rank (full professor, associate professor, assistant professor) Years in Current Rank Number of years faculty occupied current rank as of May 1, 2014 Years Pre-Hire and Post-Degree Years between highest degree earned and hire at UNH Modified Service Years Current service years - Years in current rank College Dichotomous variables measuring faculty college (COLA, COLSA, CEPS, Paul, CHHS, Library, UNHM) Past Administrative Appointment Faculty held prior administrative appointment = 1; faculty had not held prior administrative appointment = 0. To prepare the data for analysis, we first adjusted all reported salaries to be nine month, full time salaries. This involved adjusting twelve month salaries to their nine month equivalents, and adjusting part-time faculty to their full-time equivalent. We created a simplified highest-degree variable based on the degree specified in the Provost s Office data. We coded degrees of Ph.D., J.D., and Ed.D as doctorate, any master s degree (e.g., MBA, M.Ed., or MFA) as masters, and bachelor s degrees as bachelors. Due to the very low number of faculty with bachelor s degrees (N=3), we ultimately combined the master s and bachelor s categories. This resulted in a dichotomous variable to measure degree status: doctorate and non-doctorate. We calculated the years employed at UNH as the number of years a faculty member worked at UNH as of May 1, 2014 (years employed = 2014 year hired at UNH). We also used the year hired at UNH to calculate a variable to measure the amount of time between faculty s completion of highest degree and employment at UNH: years pre-hire and post-degree. This variable was then used to create a variable describing the time between completion of faculty s highest degree and being hired at UNH (year of hire year of highest degree). This variable is negative for the faculty who completed their highest degree after being hired at UNH and positive for those employed outside UNH after earning their highest degree. 2

3 We relied upon the years in rank and current service-years to create a new variable: modified service years (current service years years in rank). This allowed us to measure the amount of time faculty had spent at UNH prior to promotion to their current rank. The modified service years variable allows us to make a distinction between years of service and years in rank, thus mitigating problems with multicollinearity. Overview of Data Tables 2 and 3 provide summary statistics for the key variables utilized in our study. Table 2 reports means and standard deviations for all numerical variables. Table 3 provides frequencies and associated percentages for all categorical variables. All (n = 605) Table 2. Summary of Numeric Variables by Gender Variable Mean Std. Dev. 9 Month FTE Salary Female Years pre-hire/post-degree Years in Rank Service Years Modified Service Years* Administrative Appointment Female (n = 234) 9 Month FTE Salary Years pre-hire/post-degree Years in Rank Service Years Modified Service Years* Administrative Appointment Male (n = 371) 9 Month FTE Salary Years pre-hire/post-degree Years in Rank Service Years Modified Service Years* Administrative Appointment *Modified Service Years are years of service prior to current rank. Note that there is one less observation for Mod. Service Years due to a missing response. 3

4 Table 3. Faculty Distribution by College, Degree, and Rank College Simplified Degree Current Rank Level Freq. Percent Level Freq. Percent Level Freq. Percent All (n = 605) COLSA Bachelors Asst. Prof COLA Doctorate Assoc. Prof CEPS Masters Professor WSBE CHHS LIBR UNHM Total Total Total Female (n = 234) COLSA Bachelors Asst. Prof COLA Doctorate Assoc. Prof CEPS Masters Professor WSBE CHHS LIBR UNHM Total Total Total Male (n = 371) COLSA Bachelors Asst. Prof COLA Doctorate Assoc. Prof CEPS Masters Professor WSBE CHHS LIBR UNHM Total Total Total As shown in Table 2, on average, women are paid $90, and men are paid $108,305 per 9- month appointment. This difference, however, does not account for the many other factors that differ between male and female faculty, and could explain variations in salary. For example, on average women at UNH have spent less time between achieving their highest degree and their employment at UNH compared to men (3.52 years compared to 4.85). Women also exhibit fewer years in their current rank (7.54 years compared to 11.65), fewer years of service (14.77 years compared to 20.89), and fewer modified years of service (7.23 years compared to 9.24). Women are less likely to have served in an administrative role. In our later analysis, we control for all factors that could potentially determine salary levels, allowing us to isolate the impact of gender on faculty compensation. These multivariate models provide more accurate assessments of salary differences between male and female faculty. 4

5 Table 3 reports the overall distribution of faculty across colleges, degrees, and ranks. As Table 3 indicates, women comprise 39% of the faculty at UNH. The College of Liberal Arts (COLA) is the largest college with 213 faculty, 102 of which are female and 111 of which are male. The STEM colleges, The College of Life Science and Agriculture (COLSA) and the College of Engineering Physical Science (CEPS) are smaller with 94 and 132 faculty, respectively. In these colleges, women represent a smaller portion of the faculty with 22 female faculty in COLSA and 28 in CEPS. The remaining colleges are smaller. The Paul College of Business and Economics (Paul), The College of Health and Human Services (CHHS), and UNH-Manchester (UNHM) all exhibit roughly equal proportions of male and female faculty. The Library (LIBR) is predominantly female. Women and men hold doctorate and non-doctorate degrees in roughly the same proportions with about 90% doctorate and 10% master s or bachelor s. Our analysis excludes the UNH Law School, resulting in the loss of 13 observations. The UNH Law School is a very recent addition to UNH, officially joining the university in Currently, law school faculty are not members of the same collective bargaining agreement as faculty in the other colleges. In addition, the UNH Law School salaries exhibit a difficult salary structure as they are on a full-year basis. Due to these differences, we deemed it prudent to exclude the law school from the analysis, although it is important to note that the results of our multivariate analysis are robust, and do not vary according to the inclusion or exclusion of the UNH Law School. These summary statistics highlight important characteristics of the faculty at UNH. However, the differences we detect in these tables are not necessarily causal. Most importantly, the differences we see in male and female salaries may not be attributable to gender. In order to isolate the impact of gender on faculty salaries, we first need to control for all other factors that could determine levels of compensation. We turn now to discuss our multivariate analysis of faculty salaries. Analysis Our analysis is not able to identify a particular cause (be it bias, discrimination, or otherwise) for the differences in male and female salaries. What we are able to identify is the average salary difference between men and women at UNH conditional on (or controlling for) as many observable factors as we can. By controlling for all possible factors that could determine salary levels, we are able to isolate the impact (if any) of gender on faculty compensation. The difference we observe after controlling for these factors then represents the upper limit of the size of difference in salary attributable to gender. Before we turn to our multivariate analysis of salaries, however, we first must address the fact that some of our control variables could themselves be influenced by gender. Indeed, the AAUP s Paychecks report cautions analysts to be wary of the impact of biased variables in their models. One variable widely regarded as particularly problematic is rank. It might be the case that women earn roughly the same salary as men within a particular rank, but that women advance to higher ranks less frequently. While not a causal analysis, the descriptive summary of gender and rank in Table 3 suggests this could be the case. Roughly half of women are associate professors while 5

6 22.2% are full professors. In contrast, 36% of male faculty are associate professors and 50% are full professors. While rank (assistant, associate, or full professor) may be an important determinant of salary, rank itself is potentially subject to discrimination or bias. Women may be disproportionately denied tenure or denied advancement to higher ranks compared to men. If this is true, then controlling for rank in a salary regression will mask the full size of any potential discrimination or bias associated with being a female faculty member at UNH. The data in Table 3 underscore the differences observed between men and women at the various ranks, as a greater portion of women are concentrated at the associate rank and fewer occupy the rank of full professor. In contrast, almost half of male faculty at UNH are full professors. However, just as there are many factors that affect faculty s salaries, there are also many factors that determine faculty s rank. To control for these factors and isolate the impact of gender on rank, we estimate a multinomial logit model to assess the impact of gender alongside other variables that could determine rank: modified service years and years pre-hire post-degree. Table 4 reports the results of this multinomial logit model for rank. Table 4: Rank Regressions Rank Relative Risk Ratio Standard Error Assistant Professor Female Years pre-hire/post-degree 0.81*** 0.04 Modified Service Years 0.56*** 0.03 Associate Professor (base outcome) Professor Female 0.30*** 0.08 Years pre-hire/post-degree 1.24*** 0.03 Modified Service Years 1.40*** 0.04 N 604 Pseudo R-sq ¹Additional regressions were run to test for non-linearity but the results strongly suggested linear effects, which are reported here. * p<0.05, ** p<0.01, *** p<0.001 In Table 4, we report the statistical results of our multinomial logit model for rank; the dichotomous variable, female, is of primary interest. In a multinomial logit model, all parameters are interpreted in relation to the base outcome which, in this case, is associate professor. For interpretability, we report the relative risk ratio (RRR) and associated standard errors. This model controls for modified years of service and the amount of time between faculty s completion of highest degree and employment at UNH. After controlling for these variables, we find that women are significantly less likely to be full professors than men. For ease of interpretation, we report the substantive interpretation of our results in Table 5. Table 5 explains how gender, years 6

7 pre-hire and post-degree, and modified service years predict the likelihood that a faculty member will be an assistant or full professor (as compared to an associate professor). The rank of associate professor is the reference category. Table 5: Substantive Interpretations of Multinomial Logit Regression for Rank To Predict the Likelihood of Being an Assistant Professor Female This estimate indicates that being female is associated with a 4% increase in the probability of being an assistant professor as opposed to an associate professor after controlling for the influence of the other variables included in the model. This difference is not statistically significant. Years Pre-Hire This estimate indicates that an additional year between the completion of Post-Degree faculty s highest degree and when they were hired at UNH is associated with a 19% decrease in the probability that they are an assistant professor as opposed to an associate professor after controlling for the influence of the other variables included in the model. This difference is statistically significant. Modified Service Years This estimate indicates that an additional year of service prior to achieving faculty s current rank is associated with a 44% decrease in the probability that they are an assistant professor as opposed to an associate professor after controlling for the influence of the other variables in the model. This difference is statistically significant. To Predict the Likelihood of Being a Full Professor Female This estimate indicates that being female is associated with a 70% decrease in the probability of being a full professor as opposed to an associate professor after controlling for the influence of the other variables included in the model. This difference is statistically significant. Years Pre-Hire This estimate indicates that an additional year between the completion of Post-Degree faculty s highest degree and when they were hired at UNH is associated with a 24% increase in the probability that they are a full professor as opposed to an associate professor after controlling for the influence of the other variables included in the model. This difference is statistically significant. Modified Service Years This estimate indicates that an additional year of service prior to achieving one s current rank is associated with a 40% increase in the probability that they are a full professor as opposed to an associate professor after controlling for the influence of the other variables in the model. This difference is statistically significant. The statistical results in Table 4 suggest that women disproportionately occupy associate professor positions compared to full professor positions, and that men disproportionately occupy full professor positions compared to associate professor positions, even after controlling for differences in time between hire at UNH and completion of highest degree and modified years of service. While women are not significantly more or less likely to be assistant professors compared 7

8 to men, they are 70% less likely to be full professors compared to men (with associate professor as the reference category). Since this large and significant difference holds even after controlling for modified service years and years pre-hire/post-degree, this suggests that something other than these observable characteristics drives the difference between male and female rank decisions. To the degree that some of this difference might be explained by discrimination or bias, including rank in a salary equation will underestimate the discrimination or bias that is present in salary decisions. The remaining variables included in the rank regression are of less interest but are of plausible magnitudes. The parameter estimates for time pre-hire/post-degree indicates that an additional year of pre-hire/post-degree time is associated with a diminished likelihood of being an assistant professor and an increased likelihood of being a full professor, compared to being an associate professor (by factors of 0.81 and 1.24 respectively). Similarly, an additional year of service prior to current rank is associated with a diminished likelihood of being an assistant professor and increased likelihood of being a full professor, compared to being an associate professor (by factors of 0.56 and 1.4 respectively). Though not reported here, when both squared terms are included they are very close to 1 indicating that these effects are relatively constant for each additional year and that estimating a linear model is appropriate. Now that we have measured the impact of gender on rank, we turn to the crux of this analysis: the linkage between gender and salary. For this multivariate model, we rely upon OLS regression. We aim to assess the impact of gender on faculty salary while controlling for the effects of: years pre-hire/post-degree, modified service years, years in rank, rank, prior administrative appointment, highest degree earned, and college. We also include an interaction term for gender and rank, to test whether men and women earn different salaries over time in their rank. Table 6 reports the results of this analysis. The main variable of interest is female. Column 1 of Table 6 shows that being female at UNH is associated with an average decrease in salary of $ compared to men, even after controlling for all other factors potentially associated with salary. However, this difference is not statistically significant. Since this regression is based on a universe of cases (i.e., all faculty at UNH) and not a sample, the parameter for female is not an estimate but reflects the true difference in average salaries between men and women at UNH conditional on all other variables included in the model. Thus, statistical significance is less important than it would be if we were using a sample of UNH faculty to estimate the difference between men and women for all UNH faculty. Given the large standard error associated with the coefficient of the female variable, the lack of statistical significance is most likely due to the wide variation in salaries at UNH. That is, there is a large amount of variation around the average faculty salary at UNH, even after controlling for all variables in the model. This large amount of individual variation around the average salary, and accompanying lack of significance of the female variable, indicates that the distributions of men s and women s salaries overlap considerably. On average, women earn less than men after controlling for all other variables in the model, but there is so much variation around this average estimate that it cannot be used to make reliable predictions about the salaries of male and female faculty. If new male and female faculty were to join UNH, we could not predict with a reasonable degree of statistical certainty who would earn more. 8

9 Table 6: OLS Regression to Predict Faculty Salary (FTE Adjusted 9 Month Salary) Model 1 Model 2 Female (1476.5) (1726.2) Years pre-hire/post-degree ** (100.5) ** (103.5) Modified Service Years 263.6** (108.5) ** (87.2) Years in Rank 873.2** (68.6) ** (73.9) Female*Years in Rank 63.7 (123.3) (142.3) Administrative Appointment ** (3866.3) ** (4506.1) Non-Doctorate ** (1723.6) ** (1998.2) Associate Professor ** Omitted (1595.4) Professor ** Omitted (2171.8) COLSA ** (1486.3) ** (1729.7) CEPS ** (1343.5) ** (1567.3) PAUL ** (1770.3) ** (2070.4) CHHS ** (1697.3) * (1986.0) LIBR ** (3316.2) ** (3876.1) UNHM (2324.7) (2693.3) Constant ** (1657.6) ** (1752.9) N Adjusted R Squared Standard errors in parentheses; *p<0.05, **p<0.01 We conducted a series of additional statistical tests to account for the large amount of variation around the average estimate for salaries. We ran the regression model in Table 6 within individual 9

10 colleges, and controlling for departments. Despite all our attempts, we were not able to improve the reliability of the coefficient for the female variable. Within individual colleges and departments, there was still a large amount of variation around the average estimates of men s and women s salaries. This large amount of variation rendered the female variable insignificant, and thus not a reliable predictor of faculty salaries at UNH. Given the significant differences between men and women on the basis of rank, we also estimated the regression from column 1 without including current rank. This regression is summarized in column 2. When rank is omitted from the regression the average difference between men and women s salaries increases from $ to $ Given that we know rank is correlated with gender and that salaries increase with rank, this is what we expect; however, the magnitude of the change between column 1 and column 2 is not particularly large. Once again, the coefficient for the female variable is not statistically significant. While we can say that on average female faculty earn less than male faculty after controlling for all other variables in the model, there is too much variation around this estimate to use it as a reliable predictor of men s and women s salaries. Furthermore, our model allows us only to report the difference between the average salaries of men and women after controlling for all other factors in the model. We cannot identify the source of this difference. Without better data describing why hiring, firing, promotion, tenure, and salary decisions are made we are unable to do more than describe the average difference between men and women conditional on all available observable characteristics. This difference ranges from $ and $ , but is not statistically significant. The effect of the number of years prior to being hired at UNH after finishing the highest degree is positive. This makes sense if faculty are using this time to gain experience elsewhere. Similarly, years of service prior to current rank (modified years of service) is also positive suggesting experience at UNH leads to higher salaries. The same holds for the number of years in current rank. The interaction between female and years in rank is less than $100 and not significant, indicating that men and women receive roughly the same returns to additional years in their current rank. Prior administrative appointment is significantly associated with increased salaries that persist even after the appointment is complete. As we would expect, earning a doctoral degree is associated with significantly higher salary compared to faculty without a doctorate. Associate professors are paid on average $8, more than assistant professors and full professors are paid on average $28, more than assistant professors, controlling for all other variables in the model. These differences among assistant, associate, and full professors are statistically significant. In terms of colleges, the College of Liberal Arts (COLA) is used as the comparison group and is the lowest paid besides the library and UNH Manchester. Faculty in the Paul College of Business and Economics (PAUL) are paid the most on average, earning $38, more than faculty in COLA. The College of Engineering and Physical Sciences (CEPS) followed, as faculty in CEPS on average earned $20, more than faculty in COLA, holding all other independent variables constant. The faculty in the College of Health and Human Services (CHHS) and the College of Life Science and Agriculture (COLSA) earned $ and $ more than colleagues in COLA, respectively. Faculty in the library earned on average $15, less than faculty in 10

11 COLA. All of these differences are statistically significant. There was no significant difference between faculty salaries at UNHM and COLA. Recommendations We find that there is no statistically significant difference between male and female faculty salaries at UNH. According to this analysis, gender does not have a statistically significant direct impact on faculty compensation. However, our analysis does indicate that gender has an indirect impact, as women are 70% less likely to be full professors than men. This is a large and statistically significant difference. Given the relationship between rank and salary, we recommend that UNH continue with its efforts to improve the promotion process. We also recommend that UNH further study the reasons why women are underrepresented at the ranks of full professor, so that additional measures can rectify this imbalance. 11