of the interview pool is reviewed by a dean or other committee external to the search committee. We proceeded as we did when analyzing the percentage of female applicants. We first fitted a logistic regression model to the probability of no women in the interview pool. We then considered all positions and fitted a Poisson multiple regression model to the number of women in the interview pool to investigate whether institutional or position-level attributes are associated with the representation of women in the interview pool. We used the size of the interview pool as an exposure in the model, since the range in interview pool size was quite large, from 1 to 22. (The mean number of candidates interviewed for a position was 5.) In both cases, we accounted for the possible correlation among positions advertised by the same institution by computing standard errors of parameter estimates using the GEE method. The total number of cases considered for these analyses was 667. Of the 667 cases, there were no women in interview pools in 188 cases.
We have argued earlier that the probability of no women in interview pools is below what might be expected across many of the disciplines we reviewed. Results from the logistic regression modeling suggest further that the probability of female interviewees increases when the percentage of female applicants increases, as would be expected (p < 0.0001), with the percentage of women in the search committee (borderline significant, p = 0.06) and with the number of family-friendly policies advertised by the university (borderline significant, p = 0.07). When we account for all covariates, the adjusted mean probability that a woman who has applied to a position receives an invitation to interview is lowest in biology and not significantly different in any of the other disciplines. This would be expected given that biology has significantly more female applicants than other disciplines. The probability of women in the interview pool is significantly lower when the position is advertised as tenured than when it is advertised as tenure-track (p-value = 0.013). No other factor was significantly associated with the probability of having at least one woman in the interview pool.
Adjusted means of the probability of at least one woman in the interview pool, with the corresponding 95 percent confidence interval for the true mean probability, are presented in the table in Appendix 3-4. The values in the table corresponding to differences between levels of an effect represent the ratio of the odds ratios in each of the two levels. For example, if the probability that a woman will be interviewed in biology is 0.51, the odds ratio 0.51/0.49 is 1.04, meaning a female applicant is 4 percent more likely to be interviewed than not. If for chemistry the corresponding odds ratio is 4 (0.8/0.2, according to Appendix 3-4) then the ratio of odds ratios between biology and chemistry is 1.04/4 = 0.26. In other words, the “advantage” of a female applicant in biology is only 26 percent of that of a female applicant in chemistry. Calculation of all standard errors (and consequently, confidence intervals) in the table in Appendix 3-4 required using the Delta method. (The Delta method is described in Appendix 3-7.)
When we focused on the number of women in each interview pool, we found