The Diversity-Innovation Paradox in Science

Abstract

Prior work finds a diversity paradox: Diversity breeds innovation, yet underrepresented groups that diversify organizations have less successful careers within them. Does the diversity paradox hold for scientists as well? We study this by utilizing a near-complete population of ∼1.2 million US doctoral recipients from 1977 to 2015 and following their careers into publishing and faculty positions. We use text analysis and machine learning to answer a series of questions: How do we detect scientific innovations? Are underrepresented groups more likely to generate scientific innovations? And are the innovations of underrepresented groups adopted and rewarded? Our analyses show that underrepresented groups produce higher rates of scientific novelty. However, their novel contributions are devalued and discounted: For example, novel contributions by gender and racial minorities are taken up by other scholars at lower rates than novel contributions by gender and racial majorities, and equally impactful contributions of gender and racial minorities are less likely to result in successful scientific careers than for majority groups. These results suggest there may be unwarranted reproduction of stratification in academic careers that discounts diversity's role in innovation and partly explains the underrepresentation of some groups in academia.

Keywords: diversity; inequality; innovation; science; sociology of science.

Fig. 1.

The introduction of innovations and their subsequent uptake. (A–F) Examples drawn from the data illustrate our measures of novelty and impactful novelty. Nodes represent concepts, and link thickness indicates the frequency of their co-usage. Students can introduce new links (dotted lines) as their work enters the corpus. These examples concern novel links taken up at significantly higher rates than usual (e.g., 95 uses of Schiebinger’s link after 1984). The mean (median) uptake of new links is 0.790 (0.333), and ∼50% of new links never gets taken up. (A) Lilian Bruch was among the pioneering HIV researchers (25), and her thesis introduced the link between “HIV” and “monkeys,” indicating innovation in scientific writing as HIV’s origins are often attributed to nonhuman primates. (C) Londa Schiebinger was the first to link “masculinity” with “justify,” reflecting her pioneering work on gender bias in academia (26). (E) Donna Strickland won the 2018 Nobel Prize in Physics for her PhD work on chirped pulse amplification, utilizing grating-based stretchers and compressors (27).

Fig. 2.

Gender and race representation relate to novelty and impactful novelty. (A) Introduction of novelty (# new links) by the percentage of peers with a similar gender in a discipline (n = 808,375). Specifically, the results suggest that the more students’ own gender is underrepresented, the more novelty they introduce. (B) Similarly, the more students’ own race is underrepresented, the more novelty they introduce. (C) Binary gender and race indicators suggest that historically underrepresented groups in science (women, nonwhite scholars) introduce more novelty (i.e., their incidence rate is higher). (D) In contrast, impactful novelty decreases as students have fewer peers of a similar gender and suggests underrepresented genders have their novel contributions discounted (n = 345,257). (E) There is no clear relation between racial representation in a discipline and impactful novelty. (F) Yet the novel contributions of women and nonwhite scholars are taken up less by others than those of men and white students (their incidence rate is lower).

Fig. 3.

Underrepresented genders introduce distal novelty, and distal novelty has less impact. (A and B) Apparent network communities (colors) of concepts and their linkages. (A) The link between “fracture_behavior” and “ceramic_composition” arises within a semantic cluster. Both concepts are proximal in the embedding space of scientific concepts, and as such, their distal novelty score is low. (B) In contrast, the conceptual link between “genetic_algorithm” and “hiv-1” spans distinct clusters in the semantic network. As such, the concepts are distal in the embedding space of scientific concepts, and their distal novelty score is high. (C) Students of an overrepresented gender introduce more proximal novelty, and students from an underrepresented gender introduce more distal novelty in their theses. (D) In turn, the average distance of new links introduced in a thesis is negatively related to their future uptake.

Fig. 4.

The novelty and impactful novelty minorities introduce have discounted returns for their careers. (A–H) Each of the observed patterns holds with and without controlling for distal novelty. (A–D) Correlation of gender- and race-specific novelty with becoming research faculty or continued researcher (n = 805,236). As novelty increases, the probabilities of becoming faculty (for gender and race) and continuing research (for race) have diminished returns for minorities. For instance, a 2SD increase from the median level of novelty (# new links) increases the relative difference in probability to become research faculty between gender minorities and majorities from 3.5 to 9.5%. (E–H) Correlation of gender- and race-specific impactful novelty with becoming research faculty and a continued researcher (when novelty is nonzero, n = 628,738). With increasing impactful novelty, the probabilities of becoming faculty (for gender and race) and continuing research (for gender) start to diverge at the expense of the career chances of minorities. For instance, a 2SD increase from the median of impactful novelty (uptake per new link) increases the relative difference in probability of becoming research faculty between gender minorities and majorities from 4.3 to 15%.