The first two years of recruitment to the academic internship track show a higher success rate for male candidates compared to female candidates: while both male and female students initiated the application process in similar numbers, successful male candidates outnumbered female by 3:1. Following the introduction of anonymised applications for some reviewers, gender balance improved. There were no other changes to the recruitment process, and changes in the distribution of male and female reviewers and interviewers does not explain this effect. However, anonymising applications did not make a statistically significant difference to scores.
Removing identifiers from applications seems like a reasonable approach to mitigating unconscious bias: reviewers cannot be biased if they are unaware of the applicant’s gender. However, anonymising job applications can have unpredictable effects, with some studies finding that the practice resulted in increasing women’s chances of interview, and others finding a reduced chance [21, 22]. An investigation of the effects of anonymisation on recruitment to an ophthalmology residency programme found no significant effect on applicant scores overall or specifically for female candidates [16]. When applications are anonymised, reviewers can seek implicit signals to categorise applicants according to gender, and in doing so, use stereotypes of employment patterns and communications styles, thus activating biases which the anonymisation procedure sought to suppress [21]. Our study is in keeping with these findings that anonymising applications alone may not be sufficient to tackle implicit bias.
One possible explanation for our finding is that our email communications with reviewers and interviewers which included information on mitigating gender bias might have constituted “cues for control” – a prompt to override prejudiced responses not in keeping with the individual’s beliefs and values (e.g., that people should be treated equally regardless of gender) [23]. Further investigation would be required to explore this hypothesis.
Female candidates were more likely than male candidates to identify a female supervisor for their project. Most candidates identify a supervisor that they have already built up a relationship with and have worked with in the past, so the supervisor may also be considered a mentor. The proportion of female supervisors selected by male candidates (25%) is the same as the known distribution of professorial posts among women in higher level institutions in Ireland (25%) [3] and internationally in academic medicine (25%) [4]. This might suggest that male candidates’ choice of supervisor is more in keeping with the proportion of senior faculty who are female, and potentially less influenced by the gender of the supervisor. This finding is supported in the literature: in a study of the impact of gender on mentor–mentee success in dermatology, < 40% of male participants (mentees) indicated that they would prefer a mentor of the same gender, while 80% of female participants reported that they would prefer a female mentor [24].
Female mentors can act as role models and share their experiences on issues specific to women e.g., balancing a career with the family responsibilities that usually fall to women. Protégés may also feel they have a greater connection and find it easier to communicate with same-gender mentors [24]. There are also a small number of female-dominated specialties (e.g. Child and adolescent psychiatry, public health) [25], and it is possible that female candidates are drawn more to these specialties than male candidates, hence will meet a higher proportion of potential supervisors who are female. However, with senior female faculty currently outnumbered approximately 3:1 in Ireland, and junior female faculty equalling junior male faculty in terms of numbers [3], same-gender mentoring risks over-burdening female faculty with the work of mentoring potentially at the expense of other work which would further their own careers, e.g., publication, creating a paradoxical barrier. Moving away from more traditional dyadic or 1:1 mentoring towards other models such as peer-mentoring, group mentoring or networking models may provide part of the solution because these models are less reliant on individual senior faculty member [26], although even these models typically require input from senior faculty e.g., facilitators in peer mentoring groups. Protégés benefit from mentorship regardless of gender concordance [24], so providing opportunities for mentoring relationships to develop without emphasising a need for same-gender mentoring is likely to benefit early career researchers while avoiding inequitable distribution of mentoring responsibilities.
Limitations
Our study has the advantage of including data from real-life job applications and reviewer scores, however there are potential drawbacks to this approach. One limitation is that all the score differences in the analysis of the anonymisation process were treated as if they came from independent reviewers, whereas in reality, the same reviewer will have scored multiple applications. This creates a potential source of bias. It is difficult to fully anonymise academic applications where publications are included, furthermore gender-specific information can be inadvertently revealed in the CV section (e.g., captaincy of a camogie team, a women’s sport). Supervisor nominations were handled the same whether a candidate identified one or more supervisors, even when the genders of supervisors differed, creating another potential source of bias. Finally, implicit bias is an issue that doesn’t affect only women, there are recognised minority groups who are underrepresented in medicine (URiM). Due to a reliance on data gathered by the Irish Health Service Executive’s recruitment body, the Health Business Services (HBS), which currently only collects binary data relating to sex (M/F), it was not possible to analyse data related to other applicant characteristics including representation of groups who are URiM. Overlooking intersectionality is a cognitive pitfall which limits our ability to understand women’s experiences of discrimination [27].
Outlook
Priorities for future work could include qualitative work exploring candidate’s choice of supervisor and how it may be influenced by gender; understanding barriers and incentives to applying for the AIT and how they might differ according to gender; and investigating the effect of anonymisation from the reviewer’s perspective, e.g., reviewer’s ability to correctly identify the gender of a candidate based on an anonymised CV, and their views on the effectiveness of anonymisation.
Even though we did not find that anonymisation made a significant difference to scoring, we decided to continue the practice under close monitoring because of the apparent effect on gender balance. At a minimum, receiving an anonymous application reminds reviewers of the risk of unconscious bias and may provide a cue for control. Given the lack of robust evidence for this approach, we recommend continuous monitoring and careful evaluation to avoid unintended consequences.
Measures recommended to enhance gender equality are often focussed on women, e.g., leadership or professional development training and gender-concordant mentoring [27]. Women are not the source of gender inequality, and even well-intentioned initiatives can paradoxically create barriers by demanding more of women’s time. We recommend investigation into solutions which tackle gender inequality at an organisational and societal level, giving appropriate recognition to intersectionality and the needs of groups who are URiM.
