Equal-opportunity hiring practices have been implemented repeatedly to promote fairness and represent human diversity; but could they also lead to the production of higher quality work? This argument has been suggested repeatedly by policy makers and advocacy groups, but lacks empirical support [1]–[4]. While gender diversity is known to improve internal group processes, there is ambiguous evidence (at best) for the effect of gender diversity on group performance. Gender diverse groups tend to collaborate more effectively and exhibit higher collective intelligence [5]; and this effect is primarily explained by benefits to group processes, like better morale [6], different interpersonal styles promoting greater social sensitivity, conversational turn-taking, etc. [5], [7]–[9]. The effect of gender diversity on team performance appears more complex and context-dependent [2]–[4], [10]–[16]. Specifically, studies have often revealed no effect or a negative effect of gender diversity on team performance [2]–[4]. Otherwise, the effect of gender diversity depends upon team demography, task difficulty, etc. [2], [16]. Finally, in groups in which women have more perceived expertise than other group members, the productivity of the team might be negatively affected [17].

Recently, however, theoretical work by Hong and Page [18], [19] revealed the potential truth behind the workplace folklore that gender diversity can lead to improved productivity. Specifically, groups of problem-solvers randomly selected from a large group of intelligent operators outperformed similar groups composed of the smartest individuals selected from the same group. The failure of women to flourish in academic settings has been routinely blamed on their “different availability of aptitude at the high end”[20], [21]. Therefore, one might surmise that groups containing women might not perform as well as groups without women. However, simulations that grouped individuals with diverse problem-solving skills led to the generation of more diverse solutions, from which the best solutions were more likely to be identified than from the pool of solutions created by smarter, but more uniform groups of problem-solvers [18], [19]. Therefore, these models predict the opposite outcome predicted by many opponents of equal opportunity hiring practices [20]–[22], assuming gender diversity can approximate diversity in the model. Based on the results provided by the studies mentioned above [2]–[16], problem-solving diversity appears to be related to within-group gender-diversity, and therefore we would expect gender- or race-diverse groups to outperform homogeneous groups in their attempts to solve problems of import (e.g., [23]–[25]).

Ecology and environmental sciences increasingly involve collaborative research efforts [26], [27]. Collaboration can benefit academics by increasing early career prospects [28] and the citation rate of papers [29], [30]. It is still unclear whether collaboration affects the recognition of a paper’s academic importance. For instance, some studies have suggested that women’s articles might be cited more than their male colleagues in certain fields [31], [32]. However, a more recent study of geography journal articles found that single, male authored publications tend to be cited more frequently than papers produced by collaborative male groups, gender diverse collaborative groups, or all female (single- or multi-authored) author groups [26]. In other studies, results suggest that gender doesn’t affect citation rate [33], [34]. And finally, some studies suggest that women are less frequently cited than their male colleagues[35].

Here, we evaluated the quantity of gender diversity at the level of leadership (PIs) and working group (WG) participants within one of the most influential ecological institutions worldwide, the National Center for Ecological Analysis and Synthesis (NCEAS). Because women are relatively common but not ubiquitous in Ecological Science [36], it is a particularly suitable scientific sub-discipline in which to explore the consequences of gender diversity for research productivity. Therefore, we also explored the relationship between gender diversity in these working groups and the perceived quality of science (as measured by citations) these groups produce.

First, we compared the recent h-indexes (2005–2007) of male and female NCEAS PIs with those of the preceding decade (1996–1998). For the first three years of NCEAS, the proportion of female PIs was significantly lower than the proportion of female academics employed by US universities (χ² = 25.64, P<0.001, Figure 1A). Further, the h-indexes of these early female PIs were approximately half that of their male counterparts (Figure 1B). Over time, the proportion of female PIs at NCEAS increased such that, recently, the proportion of female PIs was virtually identical to the proportion of female academics employed by US universities (χ² =  0.0008, P>0.05). On average, h-indexes evaluated more recently should be lower than h-indexes evaluated a decade ago because younger scientists have had less time to publish articles and accumulate citations, as observed with male PI h-indexes (MANOVA: Gender x Time: F_1,123 = 4.22, P = 0.042). In contrast, the h-index of female PIs increased to equal that of their male counterparts, coincident with their increased prevalence as PIs. Therefore, as gender parity was reached in leadership roles, we observed an increase in the participation rate of highly (and equally) qualified women.

Second, we evaluated changes in the proportion of women WG participants through time and the number of citations received by publications relative to the proportion of authors who were female, with or without women in authorship positions of status (i.e., first or last author). Between 1996 and 1998, the proportion of female WG participants was significantly less than the national proportion of female faculty at US universities (χ² = 123.48, P<0.001, Figure 2A). This low proportion of female participants rose slightly but continued to be significantly less than the national proportion of female faculty between 2005 and 2007 (χ² = 35.02, P<0.001). We then compared the average number of citations received by publications with and without female coauthors, controlling for journal impact factor (F_1,200 = 32.377, P<0.0001); Figure 2B). Despite the discrepancy between the proportion of WG participants and the availability of women to participate, publications with at least one female coauthor tended to be cited 87% more often than publications without female coauthors (F_1,200 = 14.86, P<0.001). Interestingly, as the proportion of authors that were female increases, the number of citations received decreases (β = –3.46, F_1,200 = 18.86, P<0.001). This particular result may reinforce the concept that the presence of diversity, rather than the presence of one gender or the other has a significant effect on group function and effectiveness. Publications with women in both first and last author positions tended to receive marginally significantly more citations than publications without women in authorship positions of status or in only one position (either first or last position; F_2,200 = 2.31, P = 0.10). Finally, time period (and its interaction with either the presence/absence of diversity or the presence of women in authorship positions of status) did not significantly affect citation rate (Time period: F_1,200  =  0.12, P = 0.7; Time period x Diversity: F_1,200  =  0.6, P = 0.5; Time period x Status: F_2,200 = 0.2, P = 0.8). Therefore, though women in non-leadership positions continue to be under-represented as participants, the perspectives provided by both genders within a working group appear to play a fundamental role as authors in increasing the quality of publications produced.

Finally we explored the proportion and visibility of women as authors relative to their male colleagues. Over our observation period, the proportion of publications with at least one female author increased significantly over time from 42.8% (SE = 5.8) to 76.2% (SE = 7.2, F_{1, 48} = 12.489, P = 0.001). Furthermore, the proportion of female authors has increased significantly from 40.6% (SE = 5.9) to 44.3% (SE = 7.6) over that same time period (F_{1, 48} = 5.364, P =  0.025). However, there were no significant changes through time in the proportion of women acting as first authors (Early: mean  =  22.4%, SE  =  0.05; Late: mean  =  20.0%, SE  =  0.07; Kolmogorov-Smirnov test: Z = 1.155, df = 60, P = 0.139) or last authors (Early: mean  =  18.5%, SE  =  0.04; Late: mean  =  22.5%, SE  =  0.05; Kolmogorov-Smirnov test: Z = 0.695, df = 60, P = 0.720). Finally, there were no significant changes to the average relative rank of female authors over time (Kolmogorov-Smirnov test: Z = 0.981, df = 60, P = 0.291). Even though the proportion of female authors increased coincidently with the increase in female PIs, the proportion of women occupying more prestigious authorship roles did not change.

Our study also revealed that as the proportion of women in leadership positions increased, the quality of women as experienced scientists filling those positions also increased to equal that of their male counterparts, resulting in an overall increase in average leadership quality (measured by a change in h-index). This result is consistent with a recent congressional report measuring the productivity of women at US universities [42]. As leadership gender-diversity increases, as it did in our study, we expect that this may create a more welcoming social environment that, in turn, might have strong influences on the retention of women in science [43], [44]. Despite this hopeful trend, women continue to be minorities in faculty positions and leadership roles in academic science [45]. Perhaps more puzzling is the continued low proportion of who women participate in working groups, given that participants often include graduate student and post-doctoral populations - both of which include high proportions of women (although there may be other reasons for this pattern [46]–[49]). Differences in the increased proportion of women PIs versus WG participants through time may be partially a consequence of differences in their relative experience in scientific collaboration and knowledge of their ecological field. However, it is also possible that institutional efforts to increase gender diversity are more focused on highly-visible leadership positions, such as PIs, than working group participant populations, and that bias against including women as WG participants still exists [44], [50]. Promoting women, not only as PIs, but also as participants and coauthors in prestigious collaborative groups like those hosted by NCEAS, is likely to substantially alter the trajectory of their careers [44], [51].

Our study is not without limitations. We confined our analysis to a ten-year period of NCEAS’ history. This may be insufficient to fully explore the nature of publication productivity in the ecological sciences, but it does provide a benchmark for consequent studies. Second, we determined gender by name or picture recognition; some people, including academics, project gender ambiguity making our method susceptible to mistakes. Further our dataset included at least one male-to-female transgendered scientist whose publication rate was higher than the average publication rates of the other female scientists in our dataset (but was not an outlier to the dataset). This may point to interesting research questions into the effect of early-gendered socialization on academic success in the scientific community. In addition, we compared the frequency of female PIs and working group members to the proportion of female academics who identify themselves as working in the Biological Sciences across all departments at U.S. institutions because this was the finest resolution provided by NSF. Given that there may be differences in the proportion of women in ecology versus cellular biology, female ecology academics may be under- or over-represented in the NCEAS working groups. Finally, we used two indices to measure scientific output. The h-index is an imperfect measure of a scientist’s productivity for a variety of reasons (including gender and age bias, [52]–[56]) and yet it is the most frequently used (with wide general acceptance) numerical indicator that provides a reasonable method for ranking scientific productivity[53].

Few female authors participate in prestigious authorship roles, even though their frequency as authors has increased over time. This dataset is the first to document the positive consequences of gender diversity on the quality of science produced by collaborative working groups. Gender-diverse groups (specifically authorship groups with at least one woman) tend to receive more citations from their peers, suggesting that peers perceive the publications produced by gender-diverse groups to be higher quality. Bringing together the collective abilities of diverse thinkers need not be thought of as an exercise in tokenism but rather as the best opportunity to address the biggest scientific puzzles of the day.