Abstract
Gender disparity continues to be an issue in STEM, with progress requiring consistent and focused efforts. Here, we present a data-driven approach to promote high quality, gender balanced invited speaker selection for neuroscience conferences. We have targeted invited speaker opportunities because underrepresentation of female speakers at international neuroscience conferences remains a major problem, and such opportunities are critical for career development. First, we audited the top ten neuroscience journals (indexed by SCImago Journal and Country Rank; SJR), identifying (1) highly cited papers, (2) gender of first and last authors, and (3) field-weighted citation impact and total publications of first and last authors. Second, we used these data to establish a database of high quality scientists that could be used to select speakers for conferences. We found that research quality (as indexed by field-weighted citation impact and total publications) of authors of highly cited publications in the top 10 neuroscience journals did not differ significantly for females and males. The comparison between the gender base rate in neuroscience and authors publishing highly cited papers in high-quality neuroscience journals shows that female representation, particularly at last author level, is less than the estimated base rate for neuroscience. In summary, we present a data-driven approach to invited speaker selection that would facilitate gender balanced conference programs while maintaining the highest of scientific standards. This approach minimizes the influence of implicit gender bias in speaker selection decisions by using scientific quality metrics that STEM researchers are familiar with, and indeed use to evaluate their own performance. Having an immediate effect on reducing gender disparity in conference programs, our approach would generate a positive spiral for more long-term reduction of gender disparity in STEM.
Significance Statement Gender disparity is a persistent issue in STEM. We present a data-driven approach to invited speaker selection, based on scientific quality metrics that researchers use to evaluate their own and their peers’ performance. We targeted invited speaker opportunities because underrepresentation of female speakers at international conferences remains a major problem, and such opportunities are critical for career development. Research quality of authors of highly cited publications in top neuroscience journals did not differ between females and males. This approach minimizes implicit gender bias in speaker selection, which will immediately reduce gender disparity in conference programs, as well as generate a positive spiral for more long-term reduction of gender disparity in STEM.
Introduction
Gender disparity in academia has been acknowledged for some time. In neuroscience, females represent approximately half of PhD graduates but only 25 - 30% of tenure-track faculty in the US1,2. Although many have called for potential solutions to the problem, the disparity persists and progress towards gender balance is slow 3,4.
The persistence of gender disparity in neuroscience is likely due, at least in part, to implicit bias5: the covert attitudes that influence our understanding, actions, and decisions in an unconscious manner. Evidence suggests that implicit gender bias in science negatively affects outcomes for females in terms of hiring, promotion, funding, and invitations for conference presentations and editorial roles6–12. For example, in a randomized double-blind study in which laboratory manager applications were randomly allocated male or female names, faculty at research intensive universities rated male applicants as more competent and offered a higher starting salary than the identical applicant with a female name6. It is important to note, however, that within this growing literature investigating gender bias in STEM, some studies show a bias against female scientists6–12 whilst other studies suggest little bias or, more recently, affirmative treatment of female scientists13–20.
The negative effects of implicit gender bias can be overcome by either reducing the bias itself, or implementing protocols that minimize the influence of the bias. Here we have developed an approach to minimize the influence of bias in the process of selecting invited speakers. To this end, we present a data-driven approach to promote high quality, gender-balanced invited speaker selection for neuroscience conferences. We have targeted invited speaker opportunities because underrepresentation of female speakers at international neuroscience conferences remains a major problem2, and such opportunities are critical for career development. Whilst there have been some recent suggestions for ensuring gender balance in invited speaker programs (including guidelines for selection21 and diversity policies22), the selection of invited speakers remains largely subjective, leaving it open to negative effects of implicit gender bias.
We developed a two-step approach to minimize the influence of implicit gender bias in invited speaker selection. First, we audited the top ten neuroscience journals (indexed by SCImago Journal and Country Rank; SJR), identifying (1) highly cited papers, (2) gender of first and last authors, and (3) field-weighted citation impact and total publications of first and last authors. Second, we used these data to establish a database of high quality scientists (irrespective of their gender) that could be used to select speakers for conferences. If the quality of scientists on this database is comparable across gender, this approach enables gender balance in invitations that is based on established metrics of quality frequently used by researchers, hiring committees, and funding bodies, thereby minimizing the influence of implicit gender bias on selection decisions. Notably, this approach can have an immediate effect to improve the underrepresentation of female invited speakers at neuroscience conferences, and will likely have a medium- to long-term effect to improve the progression of female scientists to senior levels within STEM.
Method
The study was approved by the Murdoch University Human Research Ethics Committee (2017/206). Figure 1 shows the study procedure. The journal ranking data and citation reports were extracted on the November 26, 2017.
Journal Selection
Neuroscience journals were ranked using the SJR indicator system and Web of Science. The top ten journals comprising ≥50% original research articles were selected for auditing (see Figure 1). (Note: Molecular Psychiatry was excluded because more than 60% of publications reported authors’ initials only).
Article Selection
Total citations and average citations per year were calculated for each original research article in the selected journals (Citations from 2012-2016 for all journals except Lancet Psychiatry, for which citation data were only available from 2014-2016) Articles were selected for the author gender audit if their total citation count was greater than the average total citations for the journal in which the article was published.
Gender identification
The gender of first and last authors of the selected articles was determined to be male, female, or unknown (last author was selected because it typically represents the senior author in neuroscience). Gender determination (using western naming convention) was completed independently by two investigators, and then cross-referenced. If gender could not be determined using this method, or the name was indeterminate or androgynous, an electronic search was conducted using institutional and academic networking websites: gender was determined if the online resources included the author’s name, photo (with clear gender identification) and either a reference to the article or the author’s affiliation (listed in the article). If gender of first or senior authors could not be determined using either of these methods (6.9%), the corresponding author was emailed to request gender identity information (email response rate: 20%). (In total, the gender of 163 author could not be determined.)
Database for speaker selection
The weighted total citations (2012-2016) were obtained by dividing the total citation counts for each paper by the number of years since its publication. The weighted total citations were then used to rank all articles; the first and last authors of the top 100 ranked articles were included on our ‘potential speakers’ lists. The field-weighted citation impact (FWCI) and their total number of career publications were obtained for these authors, and the rank order of the lists was then adjusted based on FWCI. (Note: if an author did not have an identifiable FWCI using SciVal they were not included in the database.)
Results
The lists of top 100 first and senior authors based on weighted total citations and subsequently re-ranked based on FWCI showed that 32% of first authors and 21% of last authors were female (supplementary material: Table 1 and Table 2). Figure 2 shows the gender breakdown of authors in the top 100 list for FWCI and total publications. FWCI did not differ between males and females for either first or last authors (p>0.49, Cohen’s d<0.15, Bayes Factor, BF10<0.29), indicating no difference in the impact of research between males and females irrespective of career stage. All of the data are available online (see supplementary material ‘speaker database’).
The percentage of female first authors from our database (32%) was significantly less than the base rate of female trainees within neuroscience (49%), as determined by biaswatchneuro based on ~20,000 attendees at the Society for Neuroscience conference in 2017 (https://biaswatchneuro.com/base-rates/neuroscience-base-rates/; p <.001, Cohen’s w= 0.32); however, the associated Baye’s Factor (BF10=2.39) suggests that the empirical data do not provide strong evidence to distinguish the observed percentage from the base rate. In contrast, the percentage of female last authors from our database (21%) was significantly less than the base rate of female faculty within neuroscience (39%), as determined by biaswatchneuro (https://biaswatchneuro.com/base-rates/neuroscience-base-rates/; p <.001, Cohen’s w= 0.35, BF10=6.27), suggesting that the observed female representation is less than the estimated base rate in the field of neuroscience.
Figure 3 shows the percentage of female invited speakers across the 387 neuroscience conferences from 2014-2019 that are listed on www.biaswatchneuro.com. The mean percentage of female invited speakers across these conferences was 27%: this percentage of female invited speakers differs substantially from the overall base rate of 45% females in neuroscience (averaged across trainees and faculty from www.biaswatchneuro.com; p<0.001, Cohen’s w=4.26, Bayes Factor, BF10=4.05) but does not differ significantly from the observed percentage of female first or last authors as determined from our audit (both p>0.16, both Cohen’s w <0.14, both Bayes Factor, BF10<0.34). It is important to note (and indeed it is clear from Fig. 3) that whilst some neuroscience conferences are attaining, or exceeding, the gender base rate of 45% female invited speakers, more than 75% of conferences have less than 40% females in their invited speaker programs. That is to say, more than 75% of conferences are not attaining the gender base rate of females in neuroscience in their invited speaker programs.
Discussion
The data driven approach presented here enables speaker selection based on scientific impact, thereby minimizing the influence of implicit bias. Notably, this approach can ensure gender balance, given that the current results show that scientific impact does not differ between males and females in the potential speaker database.
Implicit gender bias is widespread and is proving challenging to overcome, and gender bias in STEM is no exception. Indeed, in a series of randomized, double-blind experiments, males and females evaluated the quality of scientific journal abstracts reporting gender bias in a STEM context: males evaluated abstracts less favorably than females, with male STEM faculty evaluating abstracts less favorably than female STEM faculty and male and female members of the general community23. If evidence demonstrating gender bias in STEM is not convincing to a subgroup in STEM who serve as panel members that make decisions regarding hiring, promotion, speaking invitations, and editorial invitations, we have to develop new approaches to negate, or at least minimize, the effects of implicit gender bias. Our approach purposefully includes established metrics of quality that are frequently used by researchers, hiring committees, and funding bodies. The benefits of this approach are twofold. First, it provides a data driven method for selecting invited speakers (both senior researchers as well as early career researchers), which can have an immediate effect on reducing gender disparity at scientific conferences. Second, establishing a database of high quality researchers based on these metrics provides convincing evidence of parity in scientific quality between males and females at the highest level. These benefits should, in turn, lead to a positive spiral in which invited speaking opportunities for females facilitate career development through recognition of high-quality research, providing greater opportunity for collaborative outreach, which will increase likelihood of academic promotion and female leadership within STEM, as well as providing an environment in which implicit bias should become less pervasive. The existence of equity and diversity policies in a growing number of scientific societies provides evidence of a willingness to engage in protocols that ensure more equitable conference programs. Therefore, a data-driven approach to facilitate equitable conference programs is likely to be useful and used by the rapidly growing number of societies that are considering equity and diversity in their decision making for the selection of speakers.
Here we use the broad discipline of neuroscience as an exemplar. The comparison between the gender base rate in neuroscience and authors publishing highly cited papers in high quality neuroscience journals shows that female representation, particularly at senior author level, is less than the estimated base rate for neuroscience. This, together with the data showing that more than 75% of neuroscience conferences are not attaining the gender base rate for female invited speakers, suggests underrepresentation of female scientists is a real problem in the field of neuroscience. It is important to note that we recommend refining the data-driven approach using keywords and/or the selection of specialist journals to ensure that the resultant database of potential speakers is suitable for the target conference or focussed symposia within conferences. Indeed, we have previously shown that the proposed approach would be effective in the sub-discipline of brain stimulation24. Establishing the database of potential speakers is largely automated (exportation of publications, citations, and FWCI can be automated, and ranking of authors can be performed with simple code), and the identification of gender could be automated if journals request gender information. Indeed, we call on publishers to collect these data at the proofing stage of publication and make them available post-publication.
Our data-driven approach to speaker selection takes an important step in addressing the complex issue of gender disparity in STEM, and extends beyond tools that are already available by (i) identifying individuals as potential speakers and (ii) overcoming the criticism that selection based on policies and quotas is not merit-based. For example, online calculators can provide estimates of equitable gender representation, and equity and diversity policies can prescribe equitable gender representation, but neither provide any information regarding who to invite to deliver conference or departmental presentations. Furthermore, equity and diversity policies are often subject to the criticism that the selection process is not merit-based. Our approach purposefully includes established metrics of quality that are frequently used by researchers, hiring committees, and funding bodies to overcome this criticism. In addition, the combination of metrics used in our approach provides a list of potential speakers with a recent and relevant high-quality publication, whilst ensuring some stability in terms of career research performance. Some proposed approaches to reduce gender disparity in STEM are data-driven, such as www.biaswatchneuro.com, however these approaches use data to increase accountability for gender disparity in conference programs, not to select speakers as per our approach.
Nonetheless, it is important to acknowledge some limitations with the approach. First, achieving gender balance is not equal to achieving diversity and inclusion: our approach should be extended to ensure representation of minority groups. Indeed, our approach could easily be expanded to include information regarding geographical location, ethnicity, and career stage, which would provide an opportunity to reduce the underrepresentation of minority groups in STEM. Second, our approach relies on the citations of publications in high impact journals. Evidence suggests that female scientists submit fewer manuscripts than male scientists to high quality journals, have fewer manuscripts accepted for publication in high quality journals, and that publications with a female senior author are cited less than publications with a male senior author8,25,26. Therefore, although our approach is data-driven, the data themselves are likely to be affected by implicit gender bias that negatively affects female scientists 3,4: our approach should be continually refined to include the most reliable and well-accepted quality metrics for STEM researchers. We have made the data from this study available online, and we recommend that program committees use these data, as well as continue to collect data, to reduce the underrepresentation of women in speaking programs at conferences. In light of the strengths and limitations of our approach, we argue strongly that a combination of approaches will be most effective at reducing the persistent gender disparity and preventing the emergence of gender bias in STEM, as well as increasing diversity in STEM more generally.
Funding
AMV is supported by a National Health and Medical Research Council Early Career Fellowship (GNT1088295). MRH is supported by an Australian Research Council Future Fellowship (FT150100406).
Conflict of interest statement
The authors have no potential conflicts of interest to disclose.
Acknowledgments
We would like to thank Brigid Bolton, Ellika Carson, Courtney McAuliffe, Chelsea Moran, and Tayla Stucke for their assistance with this project. We would like to thank www.biaswatchneuro.com for providing data on invited speaker programs at neuroscience conferences.
Footnotes
This version of the manuscript includes comparisons between the percentage of female authors publishing highly cited papers in the high-quality neuroscience journals that we audited and (i) the gender base rate in neuroscience conferences, and (ii) the percentage of females in invited speaker programs (from www.biaswatchneuro.com); the revised manuscript includes the results and discussion of these new comparisons. In addition, this version of the manuscript has undergone considerable revisions in relation to the feasibility and broad application of our approach, as well as how our approach extends on currently available approaches aimed at increasing representation of women in STEM.