lowing institutional, departmental, and position-level variables measured in our survey were used as explanatory variables: discipline, type of position (tenured, tenure-track), whether the institution is private or public, the prestige level of the department advertising the position, the proportion of females in the search committee, the number of family-friendly policies advertised by the institution, whether the search committee chair is a man or a woman, the percentage of female faculty in the department, and the size of the metropolitan area in which the institution is located.
We first investigated whether any of these factors are associated with the probability that no women apply to a position.9 To do so, we first created a binary variable with the value 0 if there were no female applicants and the value 1 if at least one woman applied to the position. We excluded for this analysis those positions identified as target of opportunity and open rank positions. We fitted a logistic regression model to the binary outcome variable and included as predictors in the model the institutional, departmental, and position-level variables listed above, as well as two-way interactions between discipline and the other predictors to investigate whether any of the potential effects of predictors is discipline-dependent. To account for possible correlations within positions advertised by the same institution, we implemented the method of generalized estimating equations (GEE) to compute standard errors for all parameter estimates that account for possible correlations across positions in the same institution.
We found the probability that at least one woman would apply to a position is associated with the set of discipline indicators (p = 0.03), type of position (p < 0.0001), type of institution (p = 0.08), prestige of the institution (p = 0.04), and the number of family-friendly policies in effect at the institution (p = 0.001). No other factor was statistically associated with the probability of at least one female applicant. Results can be more easily understood by looking at the adjusted means of the differences in the probability of no female applicant across levels of some of the statistically significant factors. These adjusted means are the means computed after “adjusting for” or “accounting for” all other effects in the model. Technical details and the tables are given in Appendix 3-2. We then focused on all positions and modeled the number of female applicants as a function of the same independent variables listed above. To do so, we fitted a Poisson regression model to the number of female applicants and used total number of applicants as an exposure variable. Possible correlation across positions advertised by the same institution was accounted for when computing standard errors of parameter estimates via the method of generalized estimating equations method. Again, we only included positions that were advertised as tenured or tenure-track.
As expected, we found statistically significant differences across disciplines in the proportion of females in the applicant pool. Biology, chemistry, and math-