<<

Downloaded by guest on September 26, 2021 a Huang disciplines Junming and countries scientific across in careers inequality gender of comparison Historical eeaietefidn oalo cec.Afrhrcomplication further www.pnas.org/cgi/doi/10.1073/pnas.1914221117 A and . compare of countries, to all specific difficult to in finding it scientists the making generalize active institutions, stud- of or case subsets on the disciplines, to based of is limited differences much ies gender Consequently, on evidence population. available across academic genders both diverse of scientists the for careers full struct puzzle, productivity effective it. enact address the to to of policies ability community’s consequences scientific differentiate the the to complicating ability from our causes limited interre- has the deep factors the these systemic Indeed, of that academic. latedness evidence female increasing the impede dismisses simplistic barriers data female such the However, of equivalent. of programs not reading are perpetuate (28). scientists the could male climate that and differences work interpretation gender naive and these the aca- (27), of (25), specialization persistence stereotypes The (26), role peer of 24), rank role (23, demic the collaboration (21), to (22), allocation (17–19), their review explanations resource responsibilities of (20), possible family absences course of in career plethora the differences a over from inspired women (14–16), has than which more career, publish men evidence persistent that (9–13)—the differ- puzzle” gender “productivity of the epitome is The ence (1–8). countries across and documented disciplines well all are salary, and recognition, , ity, G inequality gender and institutions for careers consequences women’s makers. important policy of rephrase with sustainability help academia, can the in academia around in conversation comprehen- inequality the gender This of remain. picture still impact, sive and differences productivity productivity in although of pub- differences portion career-wise in large a reported differences size explain the rates that same dropout demonstrate and the annual lengths we career for comparable Finally, lishing impact a work. career-wise of at equivalent body publish invariants, have women and gender and rate two men impact. uncover that and we finding productivity though, both an surprisingly, by in accompanied Most differences was years gender 60 of past increase the over of science participation of in increase women the disci- paradoxically, career 13 that, and find countries publishing We 83 plines. covering whose 2010, and 1955 authors between ended gender-identified over million of 1.5 publication publishing complete academic the reconstructing of by analysis careers per- bibliometric in a differences through a gender offer formance we longitudinal Here, of citations. picture fewer comprehensive acquires work their throughout and articles career, differ- fewer a publish underrepresented and gender disciplines are scientific of women most that in 2019) evidence 15, suggesting August fragmented, academia review in for yet (received ences 2020 extensive, 22, January is approved and There NJ, Princeton, University, Princeton Fiske, T. Susan by Edited and 02115; MA Boston, Hungary School, Budapest, Medical 1051 Harvard University, Hospital, Women’s and Brigham Medicine, China; Italy; 611731, Turin, Chengdu 10126 08540; China, NJ of Princeton, University, and Princeton Science Electronic of University Engineering, ewr cec nttt n eateto hsc,Nrhatr nvriy otn A02115; MA Boston, University, Northeastern Physics, of Department and Institute Science Network e ehdlgclosal a entedfclyt recon- to difficulty the been has obstacle methodological key A h ubro eaeadml uhr,terproductiv- their authors, male in and disparities female by of captured number academia, the in differences ender f | hnigDvso fNtokMdcn,BihmadWmnsHsia,HradMdclSho,Bso,M 02115; MA Boston, School, Medical Harvard Hospital, Women’s and Brigham Medicine, Network of Division Channing cec fscience of science a,b,c,1 lxne .Gates J. Alexander , | STEM d eateto optrSine TUiest fCpnae,20 oehgn Denmark; Copenhagen, 2300 Copenhagen, of University IT Science, Computer of Department | cetfi careers scientific a,1 oet Sinatra Roberta , d,e doi:10.1073/pnas.1914221117/-/DCSupplemental. at online information supporting contains article This 2 1 under distributed BY).y is (CC article authors access open other This Submission.y All Direct Scipher health. PNAS a in and is article networks Foodome, This of Nomix, role interest. y of the competing no founder explore declare that is companies A.-L.B. wrote Medicine, statement: A.-L.B. and interest R.S., Competing A.J.G., J.H., and data; per- analyzed paper. A.J.G. y A.J.G. the and J.H. and research; J.H. designed research; A.-L.B. formed and R.S., A.J.G., J.H., contributions: Author and 1955 lacks systematically between limitations, published methodological papers to due all but of 2010 extensive 33% This careers. covers female sample and S2.H male 2010 complete and and compare S1 1955 cally 1,110,194 sections between Appendix , and ended (SI careers female publishing countries (412,808 whose Appendix scientists 83 male) 1,523,002 (SI spanning on disciplines male) focused major 2,146,926 authors 13 million and and 3 Sci- over female of a gender of Appendix , (856,889 the deploying (SI identified Web By we identification S2.E), 2016. the gender section and for in 1900 method record between state-of-the-art database publication (WoS) their ence from scientists boundaries. national careers and publishing disciplinary complete across capture scientists we all if for only possible landscape, is survey geographical to which truly and need disciplinary, To we longitudinal, inequality, (30). whole gender sizes (29), the the sample citations of small roots the in of the exacerbated of understand fraction is majority large that the effect a receive dispropor- an produce and a authors publications academia: of the of number nature small heavy-tailed tionately the from arises ..adAJG otiue qal oti work. this to equally contributed A.J.G. and J.H. owo orsodnemyb drse.Eal [email protected] Email: addressed. be may correspondence whom To n Albert-L and , esto rudtessanblt fwmnscresin and careers institutions for women’s makers. policy consequences of important sustainability gender with con- academia, the the of rephrase around help picture can versation comprehensive lengths publishing academic career This in publishing inequality rates. differ- different gender dropout by of and explained impact and are productivity ences publications. the of equivalent suggests number have total This same and a the year publish for gen- per impact women career-wise these papers and increased of men that number has find comparable women years we Yet, of 60 differences. number past der the the in over (STEM) increase academics mathematics the across Paradoxically, and careers fields. engineering, academic of technology, impact science, and in productivity differences total gender the significant suggests evidence Empirical Significance ee ercntutdtefl ulsigcre f7,863,861 of career publishing full the reconstructed we Here, h eateto ewr n aaSine eta European Central Science, Data and Network of Department c aladMri yhsCne nCneprr China, Contemporary on Center Wythes Marcia and Paul aszl b ´ ope a,Sho fCmue cec and Science Computer of School Lab, CompleX Barab o ´ asi ´ a,f,g,h,2 raieCmosAtiuinLcne4.0 License Attribution Commons Creative y y ,alwn st systemati- to us allowing ), https://www.pnas.org/lookup/suppl/ .W then We S2). section , NSLts Articles Latest PNAS e g S Foundation, ISI eatetof Department | f8 of 1

SOCIAL AB

C

D

Fig. 1. Gender imbalance since 1955. (A) The number of active female (orange) and male (blue) authors over time and the total proportions of authors (Inset). (B and C) The proportion of female authors in several disciplines (B) and countries (C); for the full list, see SI Appendix, Tables S3 and S4.(D) The career of a scientist is characterized by his or her temporal publication record. For each publication, we identify the date (gold dot) and number of citations after 10 years c10 (gold line, lower). The aggregation by year provides the yearly productivity (light gold bars), while the aggregation over the entire career yields the total productivity (solid yellow bar, right) and total impact (solid yellow bar, right). Career length is calculated as the time between the first and last publication, and the annual productivity (dashed gold line) represents the average yearly productivity. Authors drop out from our data when they published their last article.

authors from China, Japan, Korea, Brazil, Malaysia, and Singa- using different criteria for publication inclusion and methodolo- pore (SI Appendix, section S2). To demonstrate the robustness gies for career reconstruction (SI Appendix, sections S1 and S6). of our findings to database bias and author disambiguation Our focus on bibliometric data limits our analysis to publishing errors, we independently replicated our results in two addi- careers and is unable to capture the career dynamics of teaching, tional datasets: the Graph (MAG) (31) administrative, industrial, or government related research activ- and the Digital Bibliography & Library Project (DBLP), each ities. Nevertheless, our efforts constitute an extensive attempt

2 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1914221117 Huang et al. Downloaded by guest on September 26, 2021 Downloaded by guest on September 26, 2021 un tal. et Huang (S rank affiliation (K percentile ( two-sided country the (G), by discipline (E established decade as and significant, (D), statistically rank are differences gender (see relative stated the, otherwise cases all In authors. (orange) 2. Fig. Russia in 50% with parity reaches and 1C). Germany (Fig. as in be can 28% scientists as female 33% low of by proportion variations reaches as the significant that and low observe finding also science country, as We computer 1B). is (Fig. and psychology women physics, in of math, fraction in by the considerable 15% 35% hide as numbers reaching frac- differences, aggregate century, that these disciplinary last Yet, authors, 1A). the 1955 active (Fig. over in 2005 of all increased while of 27% steadily trends: 12% for tion important only account hides represented women that women disciplines, number a and authors, years all Across Gap Gender Persistent and Increasing The and national across perspective boundaries. career-wise disciplinary offer- longitudinal, citations, engineer- and a publications technology, ing (STEM) science, mathematics in and inequality ing, gender quantify to P K GHIJ H FG DE C AB edrgpi cetfi ulsigcres h edrgpi unie yterltv ifrnebtentema o ae(le n female and (blue) male for mean the between difference relative the by quantified is gap gender The careers. publishing scientific in gap Gender ,cuty(M country (L), discipline ), ,addcd (T decade and ), eto S4.A section Appendix, SI H .Tegne a npoutvt a enicesn rmte15st h 00.(F 2000s. the to 1950s the from increasing been has productivity in gap gender The ). ,afiito ak(I rank affiliation ), Q L ). ,afiito ak( rank affiliation ), ,addcd (J decade and ), o etsaitc) ( statistics). test for N nulpoutvt snal dnia o aeadfml uhr hnsbiie by subdivided when authors female and male for identical nearly is productivity Annual (K–O) ). ,addcd ( decade and ), R M A–E ,cuty( country (B), discipline (A), percentile by down broken productivity Total ) .(P O). –T 09TmsHge dcto ol nvriyRankings; University World S2.D). the section Education from Appendix, for Higher (determined productivity 2 total 2 D) Times in (Fig. (Fig. 2019 gap countries affiliations per- gender large highest-ranked all productivity a the almost total observe in also and We gap disciplines C). gender all authors The for the 20%. for sists median bottom reverses for the and disappears 20%) in gap (middle gender authors authors the productive female Interestingly, pro- than pro- 2A). 20% papers top more (Fig. total pronounced the 37% in in publish particularly authors bracket gap male is ductivity gender as difference authors, 27% productive The among a 2A). authors in (Fig. female resulting career, ductivity sci- 9.6, their male only average, during on papers publish while, 13.2 that publish find between We entists differences 1D). impact (Fig. and genders the productivity persistent important Equally the inequality. gender are of aspect one only captures aerlnt rkndw ypretl (P percentile by down broken length Career ) h o rprino oe cieypbihn nSTEM in publishing actively women of proportion low The S N –J oa matsbiie ypretl (F percentile by subdivided impact Total ) t et with test, T O NSLts Articles Latest PNAS ,cuty(R), country (Q), discipline ), P values < C 10 ,affiliation ), −4 | B unless , f8 of 3 and SI ),

SOCIAL SCIENCES A B

C

DE

Fig. 3. Controlling for career length. (A and B) The gender gap in career length strongly correlates with the gender gap in productivity across disciplines (Pearson correlation, 0.80) (A) and countries (Pearson correlation, 0.56) (B). A gender gap of 0.0% indicates gender equality, while negative gaps indicate the career length or productivity is greater for male careers, and positive gaps indicate the feature is greater for female careers. (C) In a matching experiment, equal samples are constructed by matching every female author with a male author having an identical discipline, country, and career length. (D) The average productivity provided by the matching experiment for career length compared to the population; the gender gap is reduced from 27.4% in the population to 7.8% in the matched samples. (E) The average impact provided by the matching experiment for career length compared to the original unmatched sample. Where visible, error bars denote 1 SD.

4 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1914221117 Huang et al. Downloaded by guest on September 26, 2021 Downloaded by guest on September 26, 2021 un tal. et Huang gen- first the male representing and indistinguishable, female largely year, per when are publications words, authors of other fundamental number the In no to 2O). comes consistently (Fig. it genders is the there low between yet, difference and time; median over of decreased authors for reversed for (4% is authors and productivity. productive 20%) most the top produc- among the annual even in small difference is gender tivity The S6). section Appendix, 2 (0.9%, (Fig. considerably disparities is year, 2K gender Fig. per significant, other papers statistically than 1.32 year, while smaller per average, that papers on difference, 1.33 publish, a average, authors negligi- on male are publish, while women authors and female publication men ble: last between and differences first productivity scientist’s a defined 1D between length, publishing (Fig. span career full and time productivity the the annual as her to or productiv- his Access total into author’s dur- (39–46). each ity decompose observable to frame us patterns allows to time career data career productivity to fixed annual or a authors of ing of study pub- subset the small full limited a reconstructing has two of these careers separate difficulty lishing must the we gap, Traditionally, to gender factors. the length, con- of career a roots publishing represent the and identify career productivity a annual over of impact volution and productivity total As Length Career and Productivity Annual impact? and characteristics productivity differences What gender-based total ask: observed in the to impact drive us careers and academic prompts of universal- productivity phenomenon The the in years. 60 of gaps last ity the gender over More- the significantly regions. increased that have geographic find most we in over, and in persisting discipline phenomenon gender STEM universal global every a that is perception academia the in sup- manner differences and quantitative 35–38) a 16, 12, in 11, port 9, (2, scientific in countries observations and the unified previous disciplines specific science, a and in of using most careers confirm, 1 spanning results productive methodology to These more impact. have 2 more average, accumulate women on outnumber and, workforce to continue men female participating of dis- number authors. the gender observations increasing achieve These by can simply academia frame. favoring equality that gap time wisdom 34% conventional same the a the rupt to in in 1950s the authors from gap) switches in male (35% actually impact impact productivity female total more male in the slightly gap toward in gender 10% bias The 2000s. near strong the from a rose in 2 to productivity increase (Fig. total 1950s steady gaps in gap gender a gender impact by The and accompanied productivity is the 1A) both (Fig. (5) science in rank affiliation of countries regardless all 1I affiliations (Fig. all almost across 1 in extent, (Fig. persists lesser counter- disciplines impact female all in their and career disparity than in citations The median- 20% more parts. top for 36% the reverses in receive authors and impact male authors authors: low-impact high-impact and female for than largest publications the their 2F for (Fig. scientists citations scientists male more inflation that find 30% We for receive S2.F). section account remov- , Appendix to after (SI 1D), rescaling (32–34) (Fig. and career self-citations a during ing published paper each by h ubro iain cre 0yasatrpbiain(c publication after years 10 accrued citations of number the h vrg nulpoutvt fsinit a slightly has scientists of productivity annual average The field, playing the level to attempts recent despite summary, In women of fraction the in increase gradual the Paradoxically, emauetettlipc uiga cdmccre by career academic an during impact total the measure We ). .Ti euti bevdi l onre n disciplines and countries all in observed is result This ). L and and .W n htteannual the that find We S3). section Appendix, SI M ,adw elctdi naltredtst (SI datasets three all in it replicated we and ), .Oc gi,tettlipc ifrneis difference impact total the again, Once ). G and H ,adcnb on,t a to found, be can and ), P value E < and 10 −9 10 J ). ) ; eerhr u esssa iia ae hogotscientific throughout rates junior similar to limited at not careers. persists is gap but a dropout authors researchers observa- the risk male this that Moreover, other higher demonstrates giving time. tion scientists, In 19.5% over advantage male a 4A). cumulative than have (Fig. major of academia scientists 10.8% leave 9.0% women nearly to average, year, is on each women yearly that, the words, for while find year, rate published each We publishing dropout just yearly stop 47). have scientists the (42, male who active as paper population defined the last in rate, their authors dropout of the fraction calculated we career, in differences gender of correlate significant publication academia. a that gender is suggests impact length length career and career for here, productivity control we unobserved the when in are gaps decrease that significant variables are the differences unmatched gender that by much out influenced rule to cannot Fig. us and matching limit S4.B While section S1). they Appendix , (SI although populations results, do matched smaller these rank affiliation affect and greatly strenuous country not More for observable gaps. controlling the gender on criteria the mitigate matching on based we variables discipline, authors these their of of influence as pairs such the 12.0% variables, matching Furthermore, to confounding 38.4 By 3D). from reduced 3E). (Fig. also is (Fig. 7.8% impact total to the in 31.0 gap produc- gender the total from in from reduces gap these scientist gender tivity In the length. male samples, career which a same length-matched in the career to exactly first, with matched the and discipline of is same discipline subset scientist a same female as the each population, matched from female second each scientist a for male population, 3C a first (Fig. select the gender In we the length. eliminate per- scientist, we career to this, designed in For experiment gaps length. the matching career and in a variation productivity form the 2 total by (Fig. the explained above of observed be gaps much gender how impact ask total that next gap we gender terns, productivity con- total In (7.8%). a careers longer productivity in 35.1%. 19.2% exceeds total resulting have in (2.6%), men average, chemistry, physics gap on and applied gender biology in in the smallest trast, gen- is is the so length example, small career as For a in have productivity. gap also with total der length difference countries in career gender difference or the large gender disciplines in a differences those gender have while small also differ- productivity, gender length total correlation large career in a strong the with this in countries words, ence or disciplines other Pear- that In B; 3 implies (Fig. 0.58). country correlation, and discipline 0.80) son by correlation, scientists Pearson subdivide 3A; we (Fig. when career- gap the a productivity and by rooted gap wise supported career-length is is the conclusion between productivity correlation This total strong lengths. career in in variation variations the in of fraction 2 nificant (Fig. affiliation 2T or (Fig. country, 2P discipline, pub- either (Fig. Q, to by ceasing years grouped authors before female 9.3 are of years authors age only academic 11.0 male terminal is average average, of sci- the on age while each lish, that, of academic is 1D) finding an this (Fig. database, if reach length the test career To in the length. entist measured career we in case, differences the productivity by career total determined in gap is gender large the that suggests tists of roots possible the gaps. probe gender us observed next, helps the show result, we key As our invariant, metrics. this performance in quantity invariant der oadestefcosgvrigteedo publishing a of end the governing factors the address To pat- productivity annual indistinguishable largely the Given scien- female and male of productivity annual comparable The and R, .Tkntgte,Fg 2 Fig. together, Taken ). and n a enicesn vrteps 0years 60 past the over increasing been has and S) .W hnconstructed then We ). S4.B section Appendix , SI .Ti a essswe authors when persists gap This ). K and NSLts Articles Latest PNAS T ugssta sig- a that suggests A and F could ) | f8 of 5

SOCIAL SCIENCES AB

CD

EF

Fig. 4. Author’s age-dependent dropout rate. (A) Dropout rate for male (blue) and female (orange) authors over their academic ages. (B) The cumulative survival rate for male and female authors over their academic ages. (C and D) The effect of controlling for the age-dependent dropout rate on the gender gaps in total productivity (C) and impact (D). (E) The total impact gap is eliminated in the matched sample based on total productivity. (F) The gender gap in the total number of collaborators is eliminated in the matched sample based on total productivity.

The average causal effect of this differential attrition is accounting for about 67% of the productivity and impact gaps. demonstrated through a counterfactual experiment in which we Yet, the differential dropout rates do not account for the whole shorten the careers of male authors to simulate dropout rates effect, suggesting that auxiliary disruptive effects, from percep- matching their female counterparts at the same career stage tion of talent to resource allocation (15, 21), may also play a (Fig. 4 C and D and SI Appendix, section S4.F). We find that potential role. under similar dropout rates, the differences in total productiv- The reduction of the gender gaps in both total productiv- ity and total impact reduce by roughly two-thirds, namely from ity and total impact by similar amounts suggests that total 27.4 to 9.0% and from 30.5 to 12.1%, respectively. This result, impact, being the summation over individual articles, may be combined with our previous matching experiment (Fig. 3 D and primarily dependent on productivity (15). To test this hypoth- E), suggests that the difference in dropout rates is a key fac- esis, we conducted another matching experiment in which we tor in the observed total productivity and impact differences, selected a male author from the same discipline and with exactly

6 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1914221117 Huang et al. Downloaded by guest on September 26, 2021 Downloaded by guest on September 26, 2021 un tal. et Huang ymn ntttosadfnigaece ospottepar- efforts the recent support to of agencies impact funding the and detect institutions to many Therefore, needed by careers. is active work currently further eliminating 2010, by ended faculty female of advancement perpetuating and 53–55). retention academia, (10, lower in of mentors the cycle senior increases the dramatically for effect disparity career-wide gender to this sufficient of cumula- be impact The imbalance. not tive gender may career-wise alone observed suggesting the scientists careers, reduce their junior of on stage focusing every at that rate losing higher is a system at academic women the female that junior however, nurture find, We to researchers. efforts make must community scientific in gen- careers 48–53). implications 37, about woman’s policy (16, and of conversation administrative important sustainability the with the academia, rephrase around finding must This inequality productivity. der we total that and length suggests gaps career gender publishing subsequent rooted the are in gaps and to gender rates observed dropout us the gender-specific allowed of in portion quantities size large same a gender-invariant that the These show for work. citations of of also number body and we comparable female Yet, a receive active performance them. yearly and that indistinguishable for largely revealing have models scientists invariants, male role gender as two serve found and generations new scientists the the of train among pronounced who most authors—those the productive is highly the gap gender in particularly in the is increase that It increase an impact. troubling and gradual by productivity accompanied in the disparities was gender STEM the that in showed women impact of We coun- and fraction and 1955. productivity disciplines since across total scientists tries in male and differences female the between confirm allowed to scientists of us careers publishing full of reconstruction The Discussion 4F and Summary (Fig. authors authors female male of of S4.E). favor favor collabora- section in of in 13.3% number 16% from total to career, the a in productivity gen- throughout gap for tors gender impact controlling the total the Interestingly, flips main the academia. similarly drives drive in our consequently body gap that reinforces der which size differences invariant career-length gap, same is gender productivity the it second for gender- that This finding scientists second favor work. female a in impact of 0.8% and reveals in to difference male This discernible authors between 4E). no male is (Fig. quantity—there of drop- invariant authors the favor eliminated, female samples, completely in is of matched 38.4% impact author from these total female the ping In each in S4.D). gap as gender section publications total Appendix, of (SI number same the .Ntoa cdm fSine,Ntoa cdm fEgneig n Institute and Engineering, of Academy National Sciences, of https://www. 2017; Academy , National Rep., (Tech. 8. landscape” research global the in until “Gender long How 7. science: in science: gap Zippel, in gender effect The K. Hauser, Matilda 6. E. C. The Stuart-Fox, Leboy, D. Holman, S. P. L. Koster, 5. B. J. Pincus, S. gap. Lincoln, gender E. the A. Mind 4. in quantified: disparities Inequality gender Global Shen, Sugimoto, R. H. C. Cronin, 3. B. Gingras, Y. Ni, C. Lariviere, applications. V. grant NIH in 2. gap gender The Hamilton, H. B. Ley, J. T. 1. u ou ncoe aer iie u td ocresthat careers to study our limited careers closed on focus Our the gap, gender the reduce to order in that argued often is It ei cec n Engineering 2007). and Science demic Medicine, of elsevier.com/research-intelligence/campaigns/gender-17). Collaboration International 2000s. and represented? equally are 1990s women US, the in prizes (2012). and Awards (2013). science. (2008). 1474 Nature oe nGoa cec:AvnigAaei aer Through Careers Academic Advancing Science: Global in Women eodBa n ares ufiln h oeta fWmni Aca- in Women of Potential the Fulfilling Barriers: and Bias Beyond 1–1 (2013). 211–213 504, Safr nvriyPes tnod A 2017). CA, Stanford, Press, University (Stanford LSBiol. PLoS TeNtoa cdme rs,Wsigo,DC, Washington, Press, Academies National (The 2096(2018). e2004956 16, o.Su.Sci. Stud. Soc. aueNews Nature and Science IAppendix, SI 307–320 42, 22–24 495, 1472– 322, 1 .S og esrso e ifrne nsinicproductivity. scientific in differences sex of science. Measures in differences Long, sex S. of J. origins The 11. Long, S. J. 10. 6 .Xe .A Shauman, A. K. Xie, differences Y. Sex 16. Gingras, Y. Gelinas, P. Villeneuve, C. Vignola-Gagne, E. Lariviere, V. 15. productiv- research in differences Gender Caprasecca, A. D’Angelo, A. C. Abramo, work- G. scientific the 14. and equity, “Gender,(in) Linkova, M. Whittington, about K. evidence Fox, New F. productivity: M. research in 13. differences Sex Shauman, A. K. Xie, Y. 12. ..akolde upr rmArFreOfieo cetfi Research Scientific of by Office part Force in Air FA9550-15-1-0364. from and supported DARPA-BAA-15-39. FA9550-15-1-0077 Grants Contract support were Agency acknowledges A.-L.B. Projects Research R.S. Scientific Research and of Advanced Office J.H. Defense Force Temple- Air by FA9550-19-1-0354. for part and University in Award 61066 supported Contract Northeastern were Foundation at A.-L.B. ton Zippel and A.J.G. Khorramzadeh, Kathrin Yasamin suggestions. and valuable particular discussions, Center in the helpful and at for Research, community Network research wonderful Complex the for thank also We alizations. ACKNOWLEDGMENTS. corresponding the from request. available upon are author code and related data Other relevant Appendix). and (SI academic websites source their entire from Availability. available an Code of and Data course the gender- over in effort sustain- gender-specific rooted career. that gaps the research of be by gender conduct but ability the not academics publications of contribute may which and component through impact processes key and specific a productivity that in find departments. consequences academic we important within Namely, structures suggest (13, organizational do researchers the organizational results individual for the our to in However, available variation 58). resources potential and test context can- and we observe yet and our not differences, large-scale gender Furthermore, the reflecting DBLP) into patterns analysis. insight and statistical deep our draw MAG, can biasing approach (WoS, bibliometric possibly publications sources language thereby English data and (57), our STEM overrepresent some of important, to tend in equally three are outputs; all academic and and possible the disciplines, publications of academic Scientific one only or publications. represent industry research their in through communicate career; to research findings pressure academic less productive with an conduct positions, admin- governmental of or or end teaching duties retain an often istrative publishing imply stopped always who authors not in does inequality career gender particular, study- in for controls and, academia. temporal careers of academic gen- importance accurate ing the our more limitations, suggests develop these a work to Despite is methodologies. work there identification future discourse, der for scientific need global the pressing to contributions increased sci- significantly their in countries these differences from scientists gender Since of ence. more perspective a Brazil, provide global also Korea, would future. comprehensive were inclusion Japan, whose China, the we Singapore, for and disambiguation, in gender Malaysia, gender author studies assess of to experimental reliability could unable of such gap the gender analysis to for the Due Our baseline dominate 56). that a factors (41, offer the minorities and careers and all women of ticipation .J .Cl,H ukra,Tepoutvt uze essec n hne npat- in changes and Persistence puzzle: productivity The Zuckerman, H. Cole, R. J. 9. ti motn oepaieta h n fapublishing a of end the that emphasize to important is It en fpbiaino e n oe scientists. women (1984). and men of publication of terns nrsac udn,poutvt n mat naayi fQu of 2003). Press, analysis University An impact: and productivity professors. funding, research in system. academic Italian (2009). the of analysis bibliometric A ity: 2017). in force” puzzle. old an (1992). 178 adoko cec n ehooyStudies Technology and Science of Handbook m oi.Rev. Socio. Am. oe nSine aerPoessadOutcomes and Processes Career Science: in Women etakAieGihhnofrhl ihtevisu- the with help for Grishchenko Alice thank We 8–9 (2011). 483–498 87, 4–7 (1998). 847–870 63, h BPadMGaepublicly are MAG and DBLP The d.Mtvto Achiev. Motivation Adv. o.Forces Soc. NSLts Articles Latest PNAS MTPes abig,MA, Cambridge, Press, (MIT Scientometrics 2711 (1990). 1297–1316 68, o.Forces Soc. bcuniversity ebec ´ 517–539 79, 217–256 2, | (Harvard 159– 71, f8 of 7

SOCIAL SCIENCES 17. P. L. Carr, Relation of family responsibilities and gender to the productivity and career 39. K. Rorstad, D. W. Aksnes, Publication rate expressed by age, gender and academic satisfaction of medical faculty. Ann. Intern. Med. 129, 532–538 (1998). position – a large-scale analysis of Norwegian academic staff. J. Informetrics 9, 317– 18. S. Stack, Gender, children and research productivity. Res. High. Educ. 45, 891–920 333 (2015). (2004). 40. M. R. E. Symonds, N. J. Gemmell, T. L. Braisher, K. L. Gorringe, M. A. Elgar, Gender dif- 19. M. F. Fox, Gender, family characteristics, and publication productivity among ferences in publication output: Towards an unbiased metric of research performance. scientists. Soc. Stud. Sci. 35, 131–150 (2005). PLoS One 1, e127 (2006). 20. E. Z. Cameron, A. M. White, M. E. Gray, Solving the productivity and impact puzzle: 41. P. van Arensbergen, I. van der Weijden, P. van den Besselaar, Gender differences in Do men outperform women, or are metrics biased? BioScience 66, 245–252 (2016). scientific productivity: A persisting phenomenon? Scientometrics 93, 857–868 (2012). 21. J. Duch et al., The possible role of resource requirements and academic career-choice 42. D. Kaminski, C. Geisler, Survival analysis of faculty retention in science and risk on gender differences in publication rate and impact. PLoS One 7, e51332 (2012). engineering by gender. Science 335, 864–866 (2012). 22. R. M. Borsuk et al., To name or not to name: The effect of changing author gender 43. J. M. Box-Steffensmeier et al., Survival analysis of faculty retention and promotion in on . BioScience 59, 985–989 (2009). the social sciences by gender. PLoS One 10, e0143093 (2015). 23. M. Jadidi, F. Karimi, H. Lietz, C. Wagner, Gender disparities in science? Dropout, pro- 44. S. F. Way, D. B. Larremore, A. Clauset, “Gender, productivity, and prestige in computer ductivity, collaborations and success of male and female computer scientists. Adv. science faculty hiring networks” in Proceedings of the 25th International Confer- Complex Syst. 21, 1750011 (2017). ence on World Wide Web (International World Wide Web Conferences Steering 24. K. M. Uhly, L. M. Visser, K. S. Zippel, Gendered patterns in international research Committee, 2016), pp. 1169–1179. collaborations in academia. Stud. High Educ. 42, 760–782 (2015). 45. S. F. Way, A. C. Morgan, A. Clauset, D. B. Larremore, The misleading narrative of the 25. A. H. Eagly, C. Nater, D. I. Miller, M. Kaufmann, S. Sczesny, Gender stereotypes have canonical faculty productivity trajectory. Proc. Natl. Acad. Sci. U.S.A. 114, E9216–E9223 changed: A cross-temporal meta-analysis of U.S. public opinion polls from 1946 to (2017). 2018. Am. Psychol., 10.1037/amp0000494 (2019). 46. L. A. Hechtman et al., NIH funding longevity by gender. Proc. Natl. Acad. Sci. U.S.A. 26. P. van den Besselaar, U. Sandstrom,¨ Vicious circles of gender bias, lower positions, and 115, 7943–7948 (2018). lower performance: Gender differences in scholarly productivity and impact. PLoS 47. S. Milojevic,´ F. Radicchi, J. P. Walsh, Changing demographics of scientific careers: One 12, e0183301 (2017). The rise of the temporary workforce. Proc. Natl. Acad. Sci. U.S.A. 115, 12616–12623 27. E. Leahey, Gender differences in productivity: Research specialization as a missing (2018). link. Gend. Soc. 20, 754–780 (2006). 48. H. Etzkowitz, C. Kemelgor, B. Uzzi, Athena Unbound: The Advancement of Women 28. P. Bronstein, L. Farnsworth, Gender differences in faculty experiences of inter- in Science and Technology (Cambridge University Press, 2000). personal climate and processes for advancement. Res. High. Educ. 39, 557–585 49. S. J. Ceci, W. M. Williams, Understanding current causes of women’s underrepresen- (1998). tation in science. Proc. Natl. Acad. Sci. U.S.A. 108, 3157–3162 (2011). 29. S. Fortunato et al., Science of science. Science 359, eaao0185 (2018). 50. J. M. Sheltzer, J. C. Smith, Elite male faculty in the life sciences employ fewer women. 30. A. Clauset, C. R. Shalizi, M. E. J. Newman, Power-law distributions in empirical data. Proc. Natl. Acad. Sci. U.S.A. 111, 10107–10112 (2014). SIAM Rev. 51, 661–703 (2009). 51. W. M. Williams, S. J. Ceci, National hiring experiments reveal 2:1 faculty prefer- 31. A. Sinha et al., “An overview of Microsoft Academic Service (MAS) and Applica- ence for women on stem tenure track. Proc. Natl. Acad. Sci. U.S.A. 112, 5360–5365 tions” in Proceedings of the 24th International Conference on World Wide Web (2015). (International World Wide Web Conference Committee, 2015), pp. 243–246. 52. M. W. Nielsen et al., Opinion: Gender diversity leads to better science. Proc. Natl. 32. R. Sinatra, D. Wang, P. Deville, C. Song, A.-L. Barabasi,` Quantifying the evolution of Acad. Sci. U.S.A. 114, 1740–1742 (2017). individual scientific impact. Science 354, aaf5239 (2016). 53. E. A. Cech, M. Blair-Loy, The changing career trajectories of new parents in stem. Proc. 33. D. Wang, C. Song, A.-L. Barabasi,´ Quantifying long-term scientific impact. Science 342, Natl. Acad. Sci. U.S.A. 116, 4182–4187 (2019). 127–132 (2013). 54. National Research Council, Gender Differences at Critical Transitions in the Careers 34. F. Radicchi, S. Fortunato, C. Castellano, Universality of citation distributions: Toward of Science, Engineering, and Mathematics Faculty (The National Academies Press, an objective measure of scientific impact. Proc. Natl. Acad. Sci. U.S.A. 105, 17268– 2010). 17272 (2008). 55. L. R. Martinez, K. R. O’Brien, M. R. Hebl, Fleeing the ivory tower: Gender differ- 35. I. E. Broder, Professional achievements and gender differences among academic ences in the turnover experiences of women faculty. J. Wom. Health 26, 580–586 economists. Econ. Inq. 31, 116–127 (1993). (2017). 36. G. Abramo, C. A. D’Angelo, A. Caprasecca, The contribution of star scientists to overall 56. A. J. Stewart, V. Valian, An Inclusive Academy: Achieving Diversity and Excellence (MIT sex differences in research productivity. Scientometrics 81, 81–137 (2009). Press, 2018). 37. D. Maliniak, R. Powers, B. F. Walter, The gender citation gap in international relations. 57. P. Mongeon, A. Paul-Hus, The journal coverage of and : A Int. Organ. 67, 889–922 (2013). comparative analysis. Scientometrics 106, 213–228 (2016). 38. J. D. West, J. Jacquet, M. M. King, S. J. Correll, C. T. Bergstrom, The role of gender in 58. C. Wennerc, A. Wold, Nepotism and sexism in peer-review. Nature 387, 341–343 scholarly authorship. PLoS One 8, e66212 (2013). (1997).

8 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1914221117 Huang et al. Downloaded by guest on September 26, 2021