Measuring Inclusion: Gender and Coauthorship at the Federal Reserve Board

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Measuring Inclusion: Gender and Coauthorship at the Federal Reserve Board Deepa D. Datta, Robert J. Vigfusson 2024-091 Please cite this paper as: Datta, Deepa D., and Robert J. Vigfusson (2024). “Measuring Inclusion: Gender and Coauthorship at the Federal Reserve Board,” Finance and Economics Discussion Series 2024-091. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2024.091. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Measuring Inclusion: Gender and Coauthorship at the Federal Reserve Board∗ Deepa D. Datta† Robert J. Vigfusson‡ November 7, 2024 Abstract Relativetodiversity,inclusionismuchhardertomeasure. Wemeasureinclusionofwomen in economics using novel data on coauthoring relationships among Federal Reserve Board economists. Individual coauthoring relationships are voluntary, yet inclusion in coauthoring networks can be central to research productivity and career success. We document gender affinity in coauthoring, with individuals up to 34 percent more likely to have a same-gender coauthor in the data relative to what would be predicted by random assignment. Because womenaccountforunder30percentofFederalReserveBoardeconomists, genderaffinityin coauthoring relationships may reduce research opportunities for women relative to their men peers. Whereascommonalityofresearchinterestsisnotsufficienttoexplainobservedgender affinity in coauthoring, we find that paper outcomes may encourage gender affinity, in that papers authored by only men are more downloaded and more likely to be published than papers by mixed-gender teams. Gender affinity may contribute to the gender gap in authoring as well: women make up only 23 percent of authors in the later part of our sample, about 4 percentage points below their share of the economist population. We estimate that reducing gender affinity by men could eliminate between 1.5 to 3 percentage points of the gender gap inobservedresearchoutputbywomen. Ourfindingsongenderaffinityincoauthoringprovide anempiricalassessmentofthestateofinclusivityineconomics. JELClassification: A14,J16,E58. Keywords: central banks, coauthoring networks, diversity, gender affinity, inclusion, leakypipeline. ∗We thank Carolyn Davin, Joshua Hu, Jessica Liu, Grace Lofstrom, and Aditya Pande for excellent research assistance. We also thank Anusha Chari, Paul Goldsmith-Pinkham, Erin Hengel, Sune Karlsson, Shelly Lundberg, andJennaStearnsforgenerouslysharingtheirdatawithus,andthankGiuliaSestieriandAbigailWozniakforhelpful commentsanddiscussions. ThisworkisnotrelatedtoRobert’spositionatAmazon. Theviewsinthispaperaresolely theresponsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsofAmazonanditssubsidiaries ortheBoardofGovernorsoftheFederalReserveSystemorofanyotherpersonassociatedwiththeFederalReserve System. †BoardofGovernorsoftheFederalReserveSystem. Email: deepa.d.datta@frb.gov. ‡Amazon. Email: robvig@amazon.com. 1

1 Introduction Relativetodiversity,inclusionismuchhardertomeasure. Yet,inclusioniscentraltosuccessinthe economics profession. Professional networks provide crucial support including much needed critical feedback on research, conference and seminar invitations, and access to a potentially “hidden curriculum”fornavigatingissuesrelatedtothepublicationprocessorcareeradvancement. In this paper, we measure inclusion of women in economics using coauthoring relationships among economists at the Board of Governors of the Federal Reserve System (Federal Reserve Board, or FRB). Coauthoring relationships are an informative measure for inclusion. Coauthoring relationships are voluntary, so they measure individual preferences. Coauthoring relationships are time intensive, as individual projects take months to years from inception to realization. As such, coauthoring provides a deep and costly measure of true inclusion. Finally, coauthoring relationships are important. Finding coauthors is a big help to research productivity, and, conversely, barriers to coauthorship may impede productivity and career progression. Given that only 28 percentofeconomistsattheFederalReserveBoardarewomen,genderaffinityincoauthorshipselections may lead to fewer collaboration opportunities for women, resulting in women having lower observedproductivityasmeasuredbythenumberofworkingpapersperperson. Although we document evidence of gender affinity among economists at the Federal Reserve Board, our work should not be interpreted as showing that the Federal Reserve Board faces greater inclusion challenges than at universities or other institutions. Indeed, we show that, unlike universities—wheretheso-called“leakypipeline”resultsinlowerwomen’srepresentationamong senior professors than among junior professors—the Federal Reserve Board has roughly the same level of women’s representation at all levels of seniority. In addition, the large population of over 100womeneconomistsattheFederalReserveBoardalsoprovidesmoresame-gendercoauthoring opportunities for women than at smaller institutions. Further, we identify that one potential motivation for gender affinity by men could be that gender affinity seems to improve men’s chances of journal publication. Given that others have found broad evidence of gender bias in the journal publication process (Hengel, 2022; Hengel and Moon, 2023; Alexander et al., 2023; Card et al., 2020),thisdriverofgenderaffinitylikelyappliesatallresearchinstitutionsanduniversities. Using web-scraped data, we construct a new data set of economists at the Federal Reserve Boardoverthepasttwodecades. WecombinethesedatawithinformationonFRBworkingpapers tostudygenderinclusivityincoauthorshippatterns. Wefindthat,overthepast20years,FRBmen have consistently displayed a high degree of gender affinity, coauthoring more frequently with other men than would be expected based on random matching. Among two- and three-authored papers, about 25 percent more of them are authored by men-only teams relative to what would be expectedbyrandomchance. Wefindevidenceofgenderaffinityamongwomenaswell,especially 2

inthepastdecade,whenthenumberofFRBwomeneconomistsincreasedrelativetoearlieryears. Overall, both men and women are up to 34 percent more likely to have a same-gender coauthor in the data relative to what would be predicted by random assignment. The net effect of gender affinity is to increase the gender gap in authoring with women making up only 23 percent of authorsinthelaterpartofoursample,about4percentagepointsbelowtheirshareoftheeconomist population. Motivatedbythesefindings,wedevelopamodelofcoauthorteamformationandshow that eliminating gender affinity could reduce the gender gap in research paper production by up to 3percentagepoints. One might question whether the observed gender affinity is driven by men and women being different either in their research interests or the frequency of their social or professional interactions. We find that these gender affinity results hold up even after controlling for other variables thatpredictcoauthorship,mostnotablyusingdataonorganizationalstructurethatreflectcommonalityofresearchinterestsaswellashighlikelihoodofcollaboratingonpolicyquestions. We also examine paper outcomes, including both downloads in the first year after release and probability of publication in a journal. We find that outcomes may encourage gender affinity. Papersauthoredbyonlymengetmoredownloadsandaremorelikelytobepublishedthanmixedgender teams. Although women-only teams used to get fewer downloads than those by mixedgenderteams,thatpenaltyhasdisappearedoverthepastdecade. However,papersbywomen-only teamsarestilllesslikelytobepublishedthanthosebymixed-genderormen-onlyteams. Lastly, in addition to measuring coauthoring relationships and paper outcomes, we examine how FRB economists’ authoring profiles change over time, conditioning on early-career paper writing. We find little difference between men and women in the rate of coauthoring and writing in the first three years at the Federal Reserve Board. However, we find that women who either do not write papers in their first three years or write papers but do not coauthor with other FRB economistsaremuchlesslikelytowritepapersinsubsequentyearsthanmenwithsimilarprofiles. Although we are sensitive to the risks of selection, these findings highlight a potential upside of encouragingcoauthorshipforearly-careereconomists. Tosupportourempiricalwork,weconstructadatasetofover3,600workingpaperspublished bytheFederalReserveBoardandover2,000individuals,includinginternalandexternalcoauthors on these papers and additional information on all FRB economists since 2003. Consequently, an additional contribution of our work is this constructed data set of economists, gender tags, and someprofessionalhistory,includingPh.D.yearandworkingpapers. Thenumberofeconomistsat the Federal Reserve Board makes it at least one of the largest concentrations of Ph.D. economists atoneinstitution. FRBeconomistsareactiveresearchers,andtheFederalReserveBoardtypically 3

is among the top 20 economic institutions, when ranked by research output.1 A benefit of using our data for this and future research is that FRB economists constitute a large population within a relatively homogeneous environment. In particular, the variation in responsibilities and resources across the hundreds of FRB economists is much smaller than differences across a similar-sized groupofuniversityprofessorsthatwouldbespreadacrossmanyinstitutions. Ourfindingsofgenderaffinityandalackofinclusionhighlightchallengesforboththecareers ofindividualwomeneconomistsandtheFederalReserveBoardasaninstitution. Although there are many ways to progress at the Federal Reserve Board, research output is an important factor in promotion decisions for economists.2 Given the importance of research, we have identified two barriers in producing and publishing research that can harm women’s careers at the Federal Reserve Board. First, gender affinity reduces women’s opportunities to build coauthorship networks, which as shown in Ductor (2015) and Ductor et al. (2023) can harm research productivity. In particular, Ductor et al. (2023) estimates that, for their sample, differences in coauthorship networks explain 18 percent of the gender gap in research output. These effects fromsmallernetworksmaybeparticularlyharmfulgiventhetrendtowardsmorecoauthorship.3 Second, our finding that women-only teams are less likely to be published highlights a challenge with basing internal promotion decisions on external validation such as peer-reviewed publications. Even when an internal goal is to be gender neutral, relying too heavily on external validation for promotion decisions may introduce gender bias if the external validation process is biased. Beyondbeingaproblemforindividualwomeneconomists,limitedgenderinclusivityincoauthoring could be a broader concern for the Federal Reserve Board as an institution. As Hospido et al. (2022) note, “given the influence central banks wield over the economic well-being of the public at large, a better understanding of the factors that hold back women at these institutions is of great importance.” More broadly, Chair Jerome Powell has said that diversity and inclusion make the Federal Reserve System stronger by providing a richer pool of experience that reflects more points of view and can help Federal Reserve leadership make better decisions. He notes: “If we are inclusive in our work, listening hard to a wide range of views and learning from different 1See RePEc (n.d.), “Top 10% Economic Institutions, as of June 2024,” IDEAS database, webpage, https://ideas.repec.org/top/top.inst.all.html. 2As highlighted in a recent FRB Office of the Inspector General report, it is a common view within the Federal ReserveBoardthattheskillsacquiredthroughindependentresearchstrengthenothereconomicanalysis(Lyonsetal., 2021). 3In our data, we observe a steady trend toward more coauthorship. Solo-authored papers have declined from 57 percent of papers before 1990 to just 23 percent of papers since 2010, and three-authored papers have increased from6to28percent. 4

experiences,thenwewillreapthebenefitsofdiversity.”4 Consequently,thelackofinclusivitysuggestedbycoauthoringrelationshipsattheFederalReserveBoardisworthinvestigating,asitcould potentially diminish women’s contributions or career progression and thus the overall strength of theFederalReserveBoard. Although our paper is unique in its use of FRB coauthoring relationships to measure inclusion relative to a random benchmark, other papers have also studied gender representation amongst economists at either the Federal Reserve System or in academia. For example, Datta and Tzur- Ilan (2024) use our data on FRB economists and research papers combined with analogous data fromthe12regionalFederalReserveBankstostudyresearchandpolicyoutput. Theyfindsimilar levels of women’s representation as well as a similar research output gap as we find here, but find no gender gap in policy output, highlighting the gender gap in independent research rather than all economic analysis. Azzimonti-Renzo et al. (2023) also study the 12 regional Federal Reserve Banks and find similar levels of women’s representation, as do Auriol et al. (2022) in their study comparing U.S. and European economics departments. Davies (2022) illustrates similar patterns ofgendersortingamongauthorsofworkingpaperspublishedbytheNationalBureauofEconomic Research (NBER). Sherman and Tookes (2022) report that women in finance tend to have fewer coauthors overall, but more women coauthors, than their men peers. Ductor et al. (2023) find that women in economics have fewer collaborators, collaborate more often with the same coauthors, and have more clustering within coauthors. Further, they find that all these characteristics are associatedwithloweroverallproductivity. Lastly,McDowelletal.(2006)studycoauthorshipsinpublishedpapersamongAmericanEconomic Association members for six years between 1964 and 1998. They model the joint decision to coauthor and publish and then attribute women’s lower coauthoring rates as arising from their lower rate of publication. By contrast, we view the lower rate of publication among women as potentiallyanoutcomeoflowercoauthoring. Our work complements these papers by studying economists all working within a single institution and by reviewing working papers, which are not subject to the potential gate-keeping elements of peer-reviewed publications. Previous findings of gender affinity within broad academic networks may plausibly be driven by differences in institutional resources or social networking constraints. Bycontrast,ourfindingsapplywithinasingleinstitutionandevenwithinsmallteams of economists sorted by research interests and so point to inclusion affecting collaboration and hence research productivity even beyond factors such as institutional rank or social network opportunities. 4SeeJeromeH.Powell(2018)“ChairmanPowell’sMessagetoFederalReserveSystemStaffonDiversityandInclusion,”October9,quotedtextonp. 2,https://www.federalreserve.gov/mediacenter/files/chairman-powell-diversitytranscript-20181009.pdf. 5

The divergence between predicted and observed gender sorting in coauthorship may also provideinsightintothestateofinclusivityintheprofession. Ourfindingthatalackofgenderinclusivity in coauthoring harms women suggests that a lack of inclusivity in a broader set of professional interactions, such as conference and seminar invitations or informal social networking, could also beharmful. Indeed,ChariandGoldsmith-Pinkham(2017)findthatamongprestigiousconferences organized by the National Bureau of Economic Research, when a woman organizes the program, theshareofwomenontheprogramishigher. Additionally,Wu(2018)findsthatindiscussionson a popular online economics forum, postings about women tend to highlight physical appearance, personal information, and sexism, whereas those about men are more academically or professionallyoriented. Ourresultsarealsolimitedtofindingsongenderduetodatalimitationsbutmayindicatelackof inclusivityacrossawiderrangeofpersonalandprofessionalcharacteristics,suchasrace,ethnicity, nationality, university affiliation, and even socioeconomic diversity. Such lack of inclusivity may be related to the finding by Stansbury and Schultz (2023) that U.S. economics Ph.D. recipients, especiallythosefromhigher-rankedprograms,aresubstantiallylesslikelytohaveparentswithout acollegedegreethanPh.D.recipientsinotherdisciplines. Thislackofdiversitymaybedrivenby lack of inclusion in our profession and also may be indicative of the importance of inclusion and accessforprofessionalsuccess. The remainder of the paper is organized as follows: Section 2 describes our data collection and construction, and Section 3 summarizes our findings on gender gaps in representation and research output. We outline our models of coauthorship team formation in Section 4 and then furtherexaminehowteamformationmaybeaffectedbyinstitutionalorganizationandoverlapping research interests in Section 5. Section 6 studies paper outcomes, and Section 7 presents how coauthoringcanaffectcareerprogression. Section8concludes. 2 Data construction Our data on coauthorship comes from observing working papers in the International Finance Discussion Papers (IFDP) and Finance and Economics Discussion Series (FEDS) working papers series. Papers released in the FEDS and IFDP series are subject to internal peer-review, and the viewsexpressedrepresentviewsofthestaffanddonotindicateconcurrenceeitherbyothermembers of FRB staff or by the Board of Governors. We obtain the paper titles and authors from the RePEc (Research Papers in Economics) website, which lists more than 1,300 IFDP and 2,300 FEDS working papers, or about 3,650 papers total (LogEc, 2022a,b). Using the FRB website, we alsoassociateeachofthesepaperswithapublicationyear,rangingfrom1971to2022fortheIFDP series and 1987 to 2022 for the FEDS series (Board of Governors of the Federal Reserve System, 6

2022a,b). Each paper is written by between 1 and 14 authors, and, in sum, our data set has 2,379 uniqueauthorsacrossthese3,650papers. Although FRB economists do not have a specific requirement to produce working papers, research output does factor into economist promotion decisions. Research output includes both FEDS and IFDP releases, as well as papers released in other working papers series (such as those affiliated with the Federal Reserve Banks or the National Bureau of Economic Research), conferencepresentations,andpeer-reviewedpublications. Althoughourdataonlyincludesthoseworking papers and publications which are at some point released as a FEDS or IFDP working paper, our exclusionofthosepeer-reviewedpublicationsandworkingpaperswhicharenotpartoftheFEDS andIFDPserieslikelydoesnotskewouranalysis. AnecdotallywefindthatmostFRBeconomists releasemostoftheirworkingpapersthroughtheFEDSorIFDPseries. Next, we generate a list of past and present FRB economists, based on archived versions of the FRB public website (Board of Governors of the Federal Reserve System, 2022c). We use the Wayback Machine, which crawls the internet to capture and archive webpages as they are at certain points in time, often multiple times a month (Internet Archive, 2019). Using Python, we scrape each of these captures to collect the names of all FRB economists listed at each point in time. Given the appearance and disappearance of individuals from the website over time, the web scraping allows us to track economists’ starting years and departure years (if applicable), giving us individuals’ length and timing of service. We scrape over 200 captures of the FRB websitefromMarch2003,whenitisfirstlistedintheWaybackMachine,toApril2022,givingus 730 economists who were employed at the Federal Reserve Board at some point over this period (BoardofGovernorsoftheFederalReserveSystem,2002,2022c). Thewebpagesalsoprovideimportantinformationonadditionalcovariates,includingdetailing theFRBorganizationalstructureandeconomists’Ph.D.yearsanduniversities. Inparticular,websitesreporteconomists’divisionandsectionaffiliations. Duringoursampleperiod,themajorityof FRB economists are assigned to one of four divisions: Research and Statistics, Monetary Affairs, International Finance, and (since 2011) Financial Stability. Additionally, our data set includes economists working in three other divisions: Consumer and Community Affairs, Supervision and Regulation,andReserveBankOperationsandPaymentSystems.5 Within divisions, economists are divided into sections that are organized by research interests and policy responsibilities, with section membership typically assigned based on economists’ primary research interests. For example, both of the current authors were previously members of a sectionthatcoversdevelopmentsinU.S.tradeandcommoditypricesandhavecoauthoredworkon 5The website also lists the Division of Board Members, which includes advisers to members of the Board of Governors.AsalltheindividualspubliclylistedintheDivisionofBoardMembersalsohaveotherdivisionaffiliations, weomitthisdivisionfromouranalysis. 7

these topics (e.g., Datta et al. (2021)). In contrast, other sections focus on other areas of interest, such as developments in labor markets or capital markets. Economists can switch sections and divisions over time, though many economists spend large parts of their careers in a single section. Asisevidentfromthispaper,sectionmembershipdoesnotrequireeconomiststoworkexclusively ontopicsrelevanttothesection’sfocus. As we will show below, having common section membership is an important determinant of whethertwoeconomistscoauthor. Itsimportanceislikelydeterminedbytwofactors: greatercommon research interests and greater familiarity. Section membership increases familiarity through assigned policy collaborations, as economists typically work most closely with other members of theirownsection. Additionally,commonsectionmembershiptypicallyincreasesphysicalproximity. Aneconomist’sofficeisgenerallylocatednearthoseofothersectionmembers,andespecially beforetheCOVID-19pandemic,economistsspentmuchoftheworkweekintheiroffices. Overall, the web-scraped data have several advantages and some disadvantages relative to administrativedata,e.g. fromhumanresources. Themainadvantageisthattheweb-scrapeddataare not confidential and so can be shared much more readily. The main disadvantage is that website updates beyond dates of FRB employment are self-managed. As such, an economist’s promotion and section affiliation may not be reflected on the website until some time has passed. Consequently, we use all of the information from the FRB website that we can trust as accurate and timely, such as service dates and section and division affiliation. However, to avoid introducing biases that could arise from potential gender differences in self-reporting, we do not use any information that could be subject to self-reporting bias or delay, such as promotions or the precise timing of section and division affiliation. Importantly, this feature of the data requires us to define section and division membership or affiliation as non-time-varying and inclusive of everyone who was ever listed as being in that section or division. Future work that uses these data should be mindfulofthesechallenges. Combining our set of working paper authors with our set of FRB economists, we have about 2,400 unique individuals in our sample. Of the 730 FRB economists in our data, 617 are also authorsofworkingpapers. We gender tag these individuals using a variety of methods. In addition to using personal knowledge and picture identification from the current FRB website, we supplement the data with the gender tags generated by Chari and Goldsmith-Pinkham (2017) and Hengel (2022). Next, as in Chari and Goldsmith-Pinkham (2017), we gender tag individuals using the Tang et al. (2011) gender dictionary. The dictionary lists first names along with a count of individuals self-reporting as men and women with each name, using Facebook as an underlying source. From these counts, we generate the probability that a name is associated with being a man or a woman. If a name is associated with one gender at least 95 percent of the time, we use that gender assignment. 8

Finally, we do web searches to identify individuals who remain untagged by these sources. Of the 2,400 individuals in our data set, around one-third are identified by existing economist data sets,one-thirdareidentifiedusingtheTangetal. dictionary,andone-thirdaremanuallyidentified, meaningtheyaretaggedusingourpersonalknowledgeoftheindividuals,pictureidentification,or byfindingpronounreferencesininternetsearches. Overall,wehave536womenand1,843men. To investigate the accuracy of our gender tagging, we consider the subset of individuals listed asFRBeconomistsontheWaybackMachinecapturefromJanuary20,2017(BoardofGovernors of the Federal Reserve System, 2017). For these individuals, we generate manual gender tags using personal knowledge and picture identification, and compare these tags with those produced by the Tang et al., Chari and Goldsmith-Pinkham, and Hengel dictionaries. We find that of these 406 economists, 171 were tagged by the Chari and Goldsmith-Pinkham and Hengel dictionaries, and 184 were tagged by the Tang et al. dictionary, leaving 51 untagged. Our manual gender tagging (based on personal knowledge as well as picture identification) agreed with all 44 of the overlapping tags generated from the Hengel dictionary, with 147 of 149 of the overlapping tags generated from the Chari and Goldsmith-Pinkham dictionary and with 314 of 316 of the gender tags generated from the Tang et al. dictionary.6 We conclude from this investigation that the use of the three dictionaries provides fairly accurate gender tags, making our data set of economists with gender tags a substantial contribution to further research on gender issues in the economics profession. 3 Data summary 3.1 Gender of potential coauthors Before considering the patterns of coauthorship across genders within FRB working papers, we firstdocumenttheshareofwomeneconomistsattheFederalReserveBoard. As42percentofFRB working paper authors are external coauthors, we also document the share of women economists inafewotherrelevantexternalpopulations. As shown in Table 1, of the 730 FRB economists in our sample, 191 are women, equivalent to about 26 percent. Figure 1 shows that the number of economists at the Federal Reserve Board roughlydoublesoveroursampleperiod,fromabout225in2004tonearly450in2021. Asseenin Figure2,overthisperiodtheshareofwomeneconomistsrisesfromjustbelow25percentin2004 to nearly 30 percent in 2013 before falling back slightly to 27 percent in 2021. Given this slightly 6A common error was for individuals with the first name “Michele,” prompting us to manually gender tag all individualswiththisfirstnameinourdataset. 9

largershareandgrowingeconomistpopulation,thenumber ofFRBwomeneconomistsmorethan doublesoverthis20-yearperiod. The share of women within each hiring cohort has varied widely, reaching below 15 percent andabove40percentinmultipleyears. Figure3showsthenumberofmenandwomenhiredbythe Federal Reserve Board in each year, while Figure 4 shows the number of net hires by the Federal Reserve Board in each year, which accounts for the departures of men and women economists. Between 2003 and 2013, the share of hired economists that are women is well above 25 percent, whiletheshareofdepartingeconomiststhatarewomenisaround25percent. Inthelastpartofthe sample, however, the women’s share of departing economists rises and the women’s share of new hires falls back, contributing to the slight decline in the share of women economists at the Federal ReserveBoard. How do these numbers compare with the share of women economists in the profession? As noted in the 2023 Annual Report by the Committee on the Status of Women in the Economics Profession(CSWEP),theshareofwomentenuredandtenure-trackeconomistsatU.S.universities was about 23.9 percent in 2023, ranging from 17.5 percent of full professors to 33.5 percent of assistant professors (Chari, 2023). Women were 37.2 percent of all non-tenure-track faculty. Additionally, in 2023, 31.9 percent of U.S. Ph.D.’s in economics were granted to women. Based on these numbers, the share of women economists at the Federal Reserve Board seems to be similar totheshareofwomeneconomistsincomparableacademicinstitutions.7 One key difference between the Federal Reserve Board and academia, however, is the share of women at various levels of seniority. Others have found that the share of women declines when moving from more junior to more senior faculty in academic departments (Chari, 2023; Lundberg and Stearns, 2019). In contrast, we find that women are similarly represented at the lower and higher levels of seniority at the Federal Reserve Board (see Figure 5). Having less of a “leaky pipeline” at the Federal Reserve Board than in academia may reflect the structure of the Federal Reserve Board. Because career success at the Federal Reserve Board can be gained through high performance measured using both external and internal metrics (such as briefings to policy-makers, forecast memos, or analytical contributions to current policy questions), biases in external validation such as publication and tenure review processes—as found by Hengel (2022), Sarsons (2017), Sarsons et al. (2021), and Card et al. (2020)—may be less of a headwind than elsewhereintheprofession. Another important characteristic of economists at the Federal Reserve Board is the high share of economists in the fields of macroeconomics and finance. These fields have a lower share of women than economics as a whole. For example, Chari and Goldsmith-Pinkham (2017) show thatthemacroeconomicsandfinancesubfieldsofeconomicstendtohavealowershareofwomen 7BayerandRouse(2016)haveadditionaldiscussiononthestateofdiversityintheeconomicsprofession. 10

than microeconomics at the NBER Summer Institute. Based on data from Lundberg and Stearns (2019),we observethatabout25 percentofdissertationsin macroeconomics andfinancebetween 1990 and 2015 were written by women. The Federal Reserve Board having just over 25 percent womeninrecentyearsseemsaboutinlinewiththesestatisticsandisalsoconsistentwithfindings byAzzimonti-Renzoetal.(2023)ongenderrepresentationacrossFederalReserveBanks. Finally,whenconsideringtheshareofwomenamongexternalcoauthors,wemightexpectthat becauseFRBeconomistswritemorepapersinmacroeconomicsandfinancethaninothersubfields, the share of women among external coauthors is likely to be a bit below the share of women economists in all fields. Additionally, given that FRB economists are sorted into groups based on similar fields of interest and expertise, we might expect to find a lower propensity to overcome gender affinity among external coauthorships than internal coauthorships. Indeed, among external coauthorsofFRBworkingpapers,wefindthatabout18percentarewomen. 3.2 Section demographics Havingcommonsectionmembershipisanimportantdeterminantofwhethertwoeconomistscoauthor,asindividualsinthesamesectionarelikelytohavesharedinterests,highlikelihoodofpolicy work collaboration, and physical proximity of offices. We find that, on average, FRB economists have about nine unique coauthors, of whom about 56 percent are other FRB economists. Among an individual’s set of FRB economist coauthors, we find that, on average, about 80 percent share a division affiliation and about 50 percent share a section affiliation. Similarly, in the observed coauthorshippairsamongFRBeconomists,wefindthatover85percentshareadivisionaffiliation andnearly60percentshareasectionaffiliation. Given the high rates of coauthorship within sections, understanding the size and gender distribution across sections is informative. Though the number of sections has varied over time, the number of economists per section has remained relatively constant, with most sections having around 5 to 10 economists each. Sections do get reorganized from time to time, infrequently merging with other sections or being renamed, and, more frequently, being divided into multiple sections. Additionally,theDivisionofFinancialStabilitywasestablishedonlyin2011. Atthestart of our sample in 2003, the three main research divisions ranged from 6 to 12 sections each.8 At theendofoursamplein2022,thefourmainresearchdivisionsrangeinsizefrom5to17sections each, and the three smaller divisions range from 3 to 6 sections each. Including all 7 divisions in our sample, we have 56 sections at the end of the sample and about 60 sections over our whole sampleperiod. 8Though all these divisions have administrative and technology sections, we count here sections that have economistsasmembers. 11

Economistssometimesmoveacrosssectionsanddivisionsovertheircareer. Onaverage,about 85 percent of economists have just one division affiliation, though some have two or three. About 60 percent of economists have just one section affiliation, nearly 25 percent have two, and the remainder have three or more. Additionally, because our definition of section affiliation includes anyone who is ever affiliated with a section, the median section size is 14, and the interquartile rangeis9to26. Women economists’ share of section membership varies across sections. The distribution of the share of women across sections ranges fairly evenly between 0 and 50 percent. The share of women in the section is less than 10 percent for about 15 percent of sections. Another 17 percent ofsectionshavemorethan40percentwomen. AswediscussfurtherinSection5.1,thisdispersion across sections has implications for the likelihood that men and women coauthor. However, this dispersiondoesnotexplainourfindingsofgenderaffinity. 3.3 Number of papers To study the rate of paper production, we first define an “authorship” as a unique paper-author observation. That is, a solo-authored paper produces one authorship, while a two-authored paper producestwoauthorships,oneforeachauthor. Underthisdefinition,weabstractfromthenumber ofcoauthorsoneachpaperinthesensethatanindividualgetsasingleauthorshipfromeachpaper, whetherthatpaperhas1or10coauthors. As shown in panel A of Table 1, of the 730 FRB economists in our sample, 191 are women, equivalent to about 26 percent. Including data through 2022, the median number of authorships is 4, while the means are about 6.3 for men and 4.5 for women. The table also reports statistics in 2003 and 2021 in panels B and C. Of the 221 economists employed in 2003, 53 are women, equivalentto24percent. Forthissampleofindividuals,themediannumberofauthorshipsis8for menand5forwomen,andthemeansare10.8formenand6.5forwomen. Thesestatisticsindicate thattherearesomeextremelyproductivemeneconomists,raisingthemeannumberofauthorships. In addition, these numbers are higher than in panel A because they reflect all the working papers that these economists have produced before 2003 and through the past 20 years. Panel C provides summarystatisticsforeconomistsemployedin2021. Here,112ofthe425economistsarewomen, or 26 percent of the population. Given that many of the economists in this sample are just starting theirpublishingcareers,themediannumberofpapersis4formenandwomen,andthemeansare 5.9formenand5.0forwomen. Figure 6 shows the distribution of the number of authorships by gender. For this figure, we restrict the sample to the 307 FRB economists employed at the start of 2022 who have at least three years of FRB service. We find that the share of women with no authorships is substantially 12

higher than the share of men with no authorships, while the reverse is true for individuals with 11 ormoreauthorships.9 Figure7comparestheshareofwomeneconomistsattheFederalReserveBoardinagivenyear towomen’sshareofauthorshipsforpaperspublishedthatyear. Women’sshareofallauthorshipsin ourdatasetandwomen’sshareofauthorshipsbyFRBeconomistsbothrisefromabout15percent in 2004 to about 25 percent most recently. This increase takes the women’s share of authorships fromwellbelowtonearlyequaltotheshareofwomeneconomistsattheFederalReserveBoard. Overall,thesestatisticspointtoahigherrateofpaperproductionbymenthanbywomenatthe FederalReserveBoard. Inourinvestigationofcoauthorship,thesedifferencesmaybeanimportant related variable. Namely, the lower rate of observed paper production by women may be driving the coauthorship result or could be the result of it. Further, these outcomes are shaped by many covariatesthatmayalsodifferbygenderwithinthesetofFRBeconomists,includingage,yearsat theFederalReserveBoard,yearssinceearningaPh.D.,likelihoodofremainingatordepartingthe Federal Reserve Board, and the split between policy and research efforts and assignments. In our models of coauthoring that follow, we can choose the parameter representing the share of women among potential coauthors. Though other choices are possible, our benchmark metrics are based on the assumption that the share of available women coauthors is equal to the share of women in theFRBeconomistpopulation. 4 Modeling coauthorship team formation Inthissection,wefirstdocumentthesizeandgendercompositionofcoauthoringteams. Next,we modelcoauthoringteamformation. Table 2 presents the distribution of papers for the 2004-12 period and the 2013-21 period.10 Notably, the average number of authors per paper has increased over time. The share of soloauthored papers has fallen from 33 to 22 percent, while the share of papers with three or more authorshasrisenfrom30to43percent. Table 2 shows a modest increase in the share of two- and three-authored papers with majority or all women authors and the associated decline in the share of papers written by majority or all men. Thesestatisticsareconsistentwiththeincreasingshareofwomeneconomists,aswellasthe riseintheshareofwomen’sauthorshipsovertime. 9Inthissample,21menand11womenhave0authorships,and53menand13womenhave11ormoreauthorships. 10We separated the sample after 2012 for two reasons. First, the step-up in hiring beginning in 2011 (seen in Figure 3) did increase the share of women economists by 2013 (as seen in Figure 2). Second, tests for a structural break in estimating our model described in Section 4.2 indicated a break in the sample in the middle of the sample range. Giventhatthebreakdatewasestimatedimprecisely,wefoundbreakingourfullsampleatthehalfwaymarkto beareasonableandconvenientchoice. 13

Nowthatwehavedocumentedthedistributionofpapercoauthorships,anobviousnextquestion iswhethertheobservedoutcomesdisplaygenderaffinity. Toaddressthisquestion,wefindituseful topresentsomemodels,which,undercertainassumptions,createmappingsbetweentheobserved distributionofcoauthorshipteamsandtheshareofwomeneconomistsinthepopulation. We present two such models of increasing complexity. We present first a simple model of random matching and follow with a more complex model with coauthor team formation. The estimatedmodelsimplythattheobservedoutcomesshowgreatergenderaffinityinteamformation thanwhatmightbeexpectedwithrandommatching. 4.1 Random assignment conditional on number of coauthors The first model is simple. It takes as given the number of authors on an individual paper. Conditional on this number of authors, we assume that the authors for a paper are chosen at random and independently from the pool of possible authors, with a fixed probability of an author being a woman. Wethendeterminehowfrequentlydifferentgendercombinationsareobservedunderthis modelofrandomassignmentandcomparewiththeobservedoutcomesinourdata. The calculations underlying the model are a straightforward application of the binomial distribution for a random variable X, representing the number of women authors on a paper, with parametersnforthetotalnumberofauthorsonapaperandf fortheprobabilitythatasingledraw fromthepoolofauthorsisawoman. Consider the case for a two-authored paper (n = 2). The probability that both authors are women(X = 2)is Pr(X = 2|n = 2) = f2. Further, the probability of both authors being men is just the probability of no authors being women, Pr(X = 0|n = 2) = (1−f)(1−f). Finally,theprobabilityofoneauthorbeingawomanis Pr(X = 1|n = 2) = 2f(1−f). Moregenerally,forann-authoredpaper,theprobabilityofhavingk womenauthorsis (cid:18) (cid:19) n Pr(X = k|n) = fk(1−f)n−k. k For very large n, a hypergeometric distribution would better reflect that we draw from the pool of authors without replacement (i.e., an author can be listed only once per paper). However, as n is 14

typicallysmallrelativetothepopulation,thebinomialdistributionseemsareasonableapproximation. Given this model of random assignment, Table 2 reports the model share for each coauthoring teamsize. Thesesharesarecalculatedasthepredictedprobabilityofobservingthedifferentgender groupings, under the assumption that f equals the observed population value. For these predicted values, we use the observed 25.4 percent women share of all FRB economists from 2004 to 2012 and 27.6 percent women share of FRB economists from 2013 to 2021. The table also reports the observed counts and shares of coauthor groupings. Relative to the benchmark of random assignment, the observed frequencies of two- or three-authored papers by all men are about 10 to 12 percentage points higher than would be expected. In sum, about 25 percent more of these papers areauthoredbymen-onlyteamsrelativetowhatwouldbeexpectedbyrandomchance. The next section presents an expanded model, which can fit all of the observed groupings jointly. 4.2 Modeling the joint distribution To help us better understand the observed outcomes, our next model enriches the assumption of random assignment by moving toward a model of team formation for coauthoring relationships. Additionally, this model allows us to estimate the joint distribution across different sizes of coauthorteamsratherthanseparatelyestimatingthedistributionforeachfixednumberofauthors. In this model, we first assume that every coauthoring relationship begins with a single author initiating a project. With a defined gender-specific probability, this author either solo-authors the paper or coauthors the paper with others. The coauthors are again chosen at random. However, conditional on the gender of the first author, we include a preference parameter that can increase ordecreasetheprobabilityofchoosingawomancoauthor. 4.2.1 Modelsetup Given a population of authors, let a fraction f be women and a fraction 1 − f be men. Assume that a first author initiates a project. With probability g, the first author is a woman, and with probability1−g,thefirstauthorisaman. Aspecial(butreasonable)caseofthismodelisthatthe probability of the first author being a woman is equal to the fraction of women in the population, g = f. In our more general model setup, we allow for heterogeneity in the rate at which men and women initiate projects. In particular, this setup allows us to model the case in which the share of women-initiatedprojectsislowerthanwomen’spopulationshare. A woman who initiates a project will, with probability c (0), write the paper by herself. With w probability1−c (0),shewillseekoutcoauthors. Letc (n)betheprobabilityofawomanhaving w w 15

n coauthors. A man with a project will, with probability c (0), write the paper just by himself. m With probability 1 − c (0), he will seek out coauthors. Let c (n) be the probability of a man w m havingncoauthors. A first-author woman has a propensity p to favor women coauthors, and a first-author man w hasapropensityp tofavorwomencoauthors. Thesepropensitiesimplythatawomanwillmatch m with a woman coauthor with probability p f and a man will match with a woman coauthor with w probability p f. Values p and p below 1 imply that the probability of matching with a woman m w m coauthor is below the frequency of women in the population. In addition, the values of p and w p can vary from 0 (will not have a woman as a coauthor) to 1/f (will only have a woman as a m coauthor). Given this notation, we can calculate the probability of having k women authors on an nauthored paper. As a first step, it is useful to calculate separately the probabilities of having k women authors on an n-authored paper, conditional on the gender of the author who initiates the project(i.e.,the“first”author): (cid:32) (cid:33) n−1 Pr(X = k,n|First = W) = c (n−1) (1−p f)n−k(p f)k−1 w w w k −1 (cid:32) (cid:33) n−1 Pr(X = k,n|First = M) = c (n−1) (1−p f)n−k−1(p f)k m m m k Combiningthesetwoconditionaldistributions,wedenotetheunconditionaldistributionsas: Pr(X = k,n) = gPr(X = k,n|First = W)+(1−g)Pr(X = k,n|First = M). To better illustrate this model, consider a special case in which the choice is between having oneortwoauthorsonapaper. Insuchacase,thevalueofc (1)equals1−c (0)andc (1)equals m m w 1 − c (0). Additionally, in our data, we cannot tell if a woman asks a man to coauthor or a man w asksawomantocoauthor;wejustseeamanandawomanworkingtogether. Assuch,wewilluse {w,m}toindicatethestateswmandmw. Insuchascenario,wehavethefollowingoutcomes. State Symbol Probability Manwritesalone m (1−g)c (0) m Manwriteswithman mm (1−g)c (1)(1−p f) m m Womanandmanwritetogether {w,m} gc (1)(1−p f)+(1−g)c (1)p f w w m m Womanwriteswithwoman ww gc (1)p f w w Womanwritesalone w gc (0) w 16

4.2.2 Empiricalapplication UsingthedatareportedinTable2,wecancalculatetherelativefrequencyofeachoutcome. Define the value of z(n,k) equal to the observed frequency of n-authored papers that have k women authors. For example, based on Table 2, for the early sample, the value of z(1,0) is 0.27, which is thenumberofsolo-authoredpapersbymen(227)dividedbythetotalnumberofpapers(854). Define a vector θ to represent the model parameters {f,g,p ,p ,c (.),c (.)}. We estimate w m w m the value of θ as the solution to a two-step generalized method of moments (GMM) estimation exercise. Wefirstestimatetheθ thatminimizesthefollowinglossfunction,L(θ): 3 n (cid:88)(cid:88) L(θ) = (z(n,k)−Pr(X = k,n|θ))2. n=1 k=0 Thislossfunctionisthesumofsquareddifferencesbetweentheobservedfrequencyofn-authored papers that have k women authors and the model-implied frequency with parameter vector θ. Our estimated model parameters are the elements of the value of θ that minimizes this sum of squared differences. We then use this estimated θ to construct a weighting matrix W based on a stacked vector version of (z(n,k)−Pr(X = k,n|θ)) for n from 1 to 3 and k from 1 to n.11 We then estimate the value of θ that minimizes the standard second-stage GMM loss function using the weightingmatrixW andthestackedvectorofmoments. Wedothiscalculationforthefullsample andthenseparatelyforourearlyandlatesamples. TheresultsarereportedinTable3. First, consider the results reported for the early sample, in row 2 of Table 3. For this sample, the estimate for g is substantially lower than f, suggesting that the share of projects initiated by womenislowerthantheirrepresentationinthepopulation. Thevaluesofp andp arebothbelow w m 1. Astheestimatedvaluesofp andp arebotharound0.8,wefindthatbothmenandwomenare m w abitlesslikelytomatchwithawomancoauthorrelativetothewomen’sshareofthepopulation. In particular,thesevaluesofp andp areconsistentwithwomencoauthoringatafrequencyequivm w alent to them making up only p f ≈ p f ≈ 0.8∗0.254 of the population, which would be only m w about 20 percent. These results suggest that both men and women are making choices that result in a lack of inclusion of women in coauthoring relationships. The last column of the table reports the estimated share of authorships by women, given the estimated model parameters. Consistent withTable2andFigure7,inthisearliersample,womenareunderrepresentedascoauthors. Asweturntotheresultsforthe2013-21sampleinrow3,afewchangesarenotable. Although f andg arebothsomewhathigher,theestimatedshareofprojectsinitiatedbywomenisg = 0.241, which remains lower than the share of women in the population, f = 0.276. The value of p is w 11We use data on papers with up to three authors. With four authors, the large number of moments relative to parametersresultedinthelossfunctionbecomingmulti-modal. Assuch,wefocusonthethree-authorcasehere. 17

now above 1, suggesting that women are more likely to coauthor with each other than would be suggested by chance and perhaps pointing to more inclusive behavior by women towards other women. These results are again consistent with Table 2. Additionally, we see that p is still well m below1,whichimpliesthatmeninitiatingprojectsarelesslikelytomatchwithawomancoauthor thanwouldbesuggestedbyrandomassignment. One open question is whether a low value of p represents men displaying less inclusive bem havior, by not asking women to coauthor, or else some other form of an unsuccessful match, including women declining to coauthor with men, or projects failing. If women are declining proposed matches, then, in the 2004-2012 sample, where the estimated value of p is also less than w 1, it may be the case that either both men and women were not asking women to coauthor or else womenwererejectingoverturesfrombothmenandwomen. Incontrast,inthe2013-2021sample, the estimated value of p is greater than 1, while the estimated p is still below 1. In this period, w m both men and women seem to be making choices that exhibit gender affinity and therefore limit the number of mixed-gender teams. Given our estimated p of 1.34, the probability of a woman w havingawomancoauthor is34percenthigher,reaching37 percent ascomparedtothepopulation shareof28percent. Likewiseformen,p being0.64booststheprobabilityofamanhavingaman m asacoauthorfrom72to82percent. Thesemodelestimationscanbeusedtoprovideinsightsonpolicystrategiestoboostwomen’s authorships. AsreportedinrowAofTable3,giventhelargeshareofprojectsinitiatedbymen,an interventionthatincreasesp to1wouldboosttheshareofwomen’sauthorshipsby3.8percentage m pointsto26.8percent,closertowomen’s27.6percentshareofthepopulation. AsshowninrowB, asimilarinterventionincreasingp to1whilealsoreducingp to1wouldsimilarlyboosttheshare m w of women’s authorships by 3.2 percentage points to 26.2 percent. Although such outcomes may beappealing,designinginterventionsaimedatboostingp wouldrequirefurtherunderstandingof m why it is currently low. Recent research has found a downtick in mixed-gender collaboration after the focus on sexual harassment triggered by the “Me Too” movement (Gertsberg, 2022), which the author attributes to senior men potentially avoiding working with more junior women to avoid the risk of being accused of improper behavior. Our own results above do find an increase in women’s gender affinity that could also reflect this evolution. If a lowp results from men failing m to invite women to coauthor, then efforts to increase inclusivity in coauthoring relationships could behelpful. Alternatively,alowp resultingfromwomendecliningopportunitiestocoauthorwith m men may be optimal for women given evidence in the literature that women get less credit for projectscoauthoredwithmoreseniormencolleagues(Sarsons,2017). Other policy intervention options would be to boost either g or p . Boosting g to equal f w seemsareasonablegoalandworthinvestigating. However,thisstrategyresultsinasmallerchange than boosting p . Setting g equal to f (row C) boosts women’s authorships to 25.1 percent, 2.1 m 18

percentagepointsabovethelatesampleestimate. Here,too,understandingwhyg iscurrentlyless thanf shouldbedonebeforelaunchinganintervention. Womenmaybechoosingtoinitiatefewer researchprojectsthanmen(resultinginavalueofg lessthanf)becausetheyfaceadifferentriskreturn trade off between investment in research and policy than men. Supporting this hypothesis, Datta and Tzur-Ilan (2024) report that, in contrast to the observed gender gap in research output, there is little to no gap in policy output between men and women in the Federal Reserve System. One possible reason for this difference is that women may find that their policy work is evaluated more fairly than external academic research. Indeed, it has been found that women authors are oftenheldtohigherstandardsinacademicresearch(Hengel,2022;HengelandMoon,2023;Card et al., 2020). Alternatively, g may be less than f because women have fewer hours to devote to research projects, either because they may be assigned more policy work or because they have fewer additional hours outside of the workday for research, perhaps due to to greater parenting or otherdomesticresponsibilities.12 Finally, boosting p would also increase women’s authorships. Boosting p could be interw w pretedasonepurposeofnetworkingeventssuchastheCSWEP’sCeMENT:MentoringforJunior Facultyprogram(Gintheretal.,2020). However,asubstantialincreaseinp wouldberequiredto w materially change the share of women’s authorships. For example, as reported in row D, boosting p by50percenttoabout2.0wouldresultinwomen’sshareofauthorshipsincreasing3.4percentw agepointsto26.4percent. Suchaninterventionwouldreducethegendergapinpaperproduction, butthewomen’sshareofauthorshipswouldremainbelowwomen’sshareoftheeconomistpopulation. Ifp remainswellbelow1,onlyinterventionsthatboostbothg andp (rowE)wouldresult m w inwomen’sshareofauthorshipsbeingclosetothe27.6percentwomen’sshareofthepopulation. 5 The determinants of coauthorships Giventheseestimatesofobservedgenderaffinity,wenowdivedeeperintotheassociationbetween gender affinity and other determinants of coauthorship. In particular, we use the structure of the FederalReserveBoardtounderstandwhetherobservedgenderaffinitymayresultfromconfoundingfactorsincludingcommonresearchinterestsandincreasedinteraction. This section provides empirical evidence regarding the likelihood that two FRB economists coauthorapaper,whenconditioningonseveralcharacteristicsincludinggender,seniority,andsectionmembership. Wefindthatcoauthorshipincreaseswhenpotentialcollaboratorsfinishgraduate school at a similar time. We also find that coauthorship rises with common section membership, 12ThoughoursampleofFRBeconomistsaregenerallysalariedemployeeswhoareexpectedtowork40hoursper week,itiscommonforemployeestoreportworkingmorethan40hoursperweekforthepurposeofdevotingmore hours to research work. If men work more additional hours than women, they may have more hours to devote to research. 19

whichreflectsahigherlikelihoodthattwoindividualswillhavecommonresearchinterestsandbe assigned to work together on policy assignments. Lastly, we find that even when controlling for thesefactors,westillobserveagreaterlikelihoodofcoauthoringbetweensame-genderpairs. 5.1 Data We define the set of observed coauthorships as all the pairwise collaborations in our set of FRB working papers. For example, a two-authored paper generates one observed pair of coauthors, or one coauthorship observation. A three-authored paper generates three pairwise coauthorships and so forth. In our data on over 3,600 FRB working papers, we have nearly 2,250 unique authors and about 4,700 unique coauthorships. Of these, 617 are FRB economist authors and around 1,500 coauthorships are between two FRB economists; the remainder are associated with external collaboratorsorFRBcoauthorswhoarenotFRBeconomists,suchasresearchassistants. Toinvestigatethedeterminantsofcoauthoringrelationships,weneedthesetofpotentialcoauthorshippairs,includingthosethatarenotobservedintheworkingpapersdataset. Totakeadvantage of the rich data set we have constructed, we focus on coauthorships among FRB economists for whom we have covariates such as section affiliation and year of Ph.D. completion. Additionally,becausewefindthat83percentofobservedFRBcoauthorshipsareforindividualswhoshare a division affiliation and that 99 percent are for individuals whose tenure at the Federal Reserve Board overlaps, we further limit the set of potential coauthorships to pairs of FRB economists whohaveoverlappingdivisionaffiliationsandFRBtenure. Giventhesedefinitions,wehaveabout 40,000 unique potential coauthorships, of which about 1,200, or about 3 percent, are observed in the data. These 1,200 coauthorships result from matching among 487 FRB economists. These 487 individuals exclude those economists who never author any papers, have only solo-authored papers,orwhosecoauthorsincludeonlyexternalcollaborators. Before estimating a model to study the likelihood of observing potential coauthorships in the data, we make a few observations on the joint variation between gender and two other characteristics that strongly influence coauthorships: joint section membership and similarity of Ph.D. cohort. First, just over half of FRB coauthorships are between individuals who share section membership. Further, among same-division coauthorships, 63 percent are between same-section individuals. The share of women differs widely across sections, and men are more likely to be in sections with a lower share of women. This dispersion has implications for coauthorship matching. Although the Federal Reserve Board is 26.5 percent women, restricting the set of coauthors toonlyone’sownsectionwouldresultinsomeobservedgenderaffinityforbothmenandwomen. Suppose a man randomly chooses to coauthor with another economist in his own section. The 20

fraction of those random coauthor selections that would be with a woman is 0.263. In contrast, if a woman randomly matches with another economist in her own section, the fraction of those randommatchesthatwouldbewithawomanis0.280. (Theseresultsarecalculatedassumingselfmatching is not allowed.) As such, although section affinity does explain some gender affinity, its role does not seem to be dominant. For example, as reported in Table 3, the estimated values of p f andp f inthelatesampleareequivalenttomendrawingfromapoolthatisonly17.7percent m w womenandwomendrawingfromapoolthatis37.0percentwomen. Thesesharesaremuchmore disparate than what would be implied by gender dispersion across sections. Furthermore, as we showbelow,observedgenderaffinitypersistsevenwithinsections. Second, we observe that 30 percent of coauthorships are between individuals whose Ph.D. cohortsarethreeorfeweryearsapart. Assuch,weincludedataoncohortsimilarityasanadditional explanatory variable, so that we do not attribute to gender affinity that which may be driven by cohort affinity. We show below that cohort similarity predicts coauthorship. However, it does not explain gender affinity. Although cohort affinity is an interesting topic, it is beyond the scope of thecurrentpapertoexplorefully.13 5.2 Linear probability model We now study the determinants of coauthorship using a linear probability model. This model will allow us to consider multiple determinants of coauthorship at once, so that we can investigate gender affinity in the context of other features of the data, including section affinity and cohort affinity. Formally, we let Y be the binary outcome variable for coauthorship, where Y = 1 if ij ij individualsiandj coauthoratleastonepapertogetherandY = 0otherwise. ij Inourmodel,weletY dependonvariouspredictorvariables. First,inourbaselinemodel,we ij include an indicator variable for whether individuals i and j have overlapping section affiliation (IS). Second, because we observe that individuals are more likely to coauthor with others in a ij similar Ph.D. cohort, we include in our model the variable CohortDifference = |PhDyear − ij i PhDyear |,whichmeasurestheabsolutevalueofthedifferenceinPh.D.graduationyearbetween j thetwoindividuals. Turningtothegender-relatedterms,wefirstincludeindicatorvariablesforwhetherindividuals i and j are both men (IM) or both women (IW). These variables allow us to estimate whether ij ij same-gender affinity differs between men and women. Next, we also include interaction terms betweenthesame-genderindicatorvariablesandthesame-sectionindicatorvariabletoinvestigate 13In particular, our evidence of cohort affinity would need to be reconciled with previous studies that have found thathavingmoreseniorcoauthorsisassociatedwithhigherresearchoutput(Ductoretal.,2023). 21

whethergenderaffinityisstrongerwithinsections: Y = β +β IS +β IM +β (IM ∗IS) ij 0 1 ij M ij MS ij ij +β IW +β (IW ∗IS)+β CohortDifference +(cid:15) . (1) W ij WS ij ij 2 ij ij 5.3 Linear probability model results Overall, we find that the strongest determinant of coauthoring is whether individuals are in the same section. Beyond this variable, we also find that the probability of coauthoring increases when individuals are the same gender. In addition, the likelihood of coauthoring increases when two individuals are from similar Ph.D. cohorts. We look at each of these results in turn using the baselineresultfromourlinearprobabilitymodel,asreportedincolumn2ofTable4. Additionally, we note that the marginal effects from our variables of interest cannot be read directly from the estimated coefficients of the linear probability model given the interaction terms included in the regression. Consequently, in Table 5, we report the likelihood of coauthorship given same or differentsectionaffiliations,conditionalonthepairofindividualsbeingthesamegenderordifferent genders. Using these likelihoods, we then report the increase in the likelihood of coauthorship whenmovingfromdifferenttosamesections. First, we find that the likelihood of coauthorship is most affected by overlapping section affiliation. As shown in column 2 of Table 4, on average, two different-gender individuals are 7.2 percentage points more likely to coauthor when they are in the same section. Overall, being in the same section is by far the strongest predictor of observed coauthorship: It makes two individuals aroundfivetimesaslikelytocoauthor,asseeninthefinalcolumnofTable5. The importance of section affiliation is not surprising, as the structure of the Federal Reserve Board is such that individuals with similar research interests are likely to be in the same section. Furthermore, being in the same section increases the likelihood of collaboration on policy work as well as physical office proximity, both of which would tend to increase research collaboration. Indeed, 52 percent of unique FRB coauthorships (for which we have section information) are betweenindividualswhosharesectionmembership. Second, we also find that individuals are more likely to coauthor with their peers. The likelihood of coauthorship declines as individuals’ Ph.D. years grow further apart: We find that a pair withanadditionalyearbetweenPh.D.cohortsisabout0.1percentagepointlesslikelytocoauthor together. Although the cohort effect is statistically significant, gender and section affinity seem to playalargerrole. Third, even when focused on individuals within the same section, and after controlling for Ph.D.cohort,westillfindthatthelikelihoodofcoauthorshipincreasesfurtherwhentheindividuals 22

are the same gender. Relative to section members that are a mixed gender pair, the probability that two men in the same section coauthor is 0.9 percentage point higher and the probability that two women in the same section coauthor is 1.5 percentage points higher. These are statistically significantandsizablechangesintheprobabilityofcoauthoring. The gender affinity within a section can be illustrated by the marginal effects between counterfactuals. To calculate these marginal effects, we use the average value of three years of cohort difference. As reported in Table 5, when using these values, we determine the predicted likelihood of coauthorship for a different gender pair in the same section is 9.0 percent. This estimate increases to 10.1 percent for two men and 10.3 percent for two women. These differences are a sizable change in the probability of coauthoring. Based on Table 5, the probability of matching with anyone outside of one’s section is low and does not actually differ much whether the other person is the same or different gender. However, within a section, the probability of matching increasessubstantiallywhenthepersonisthesamegender. As such, although section membership is an important determinant of coauthorship, it does not fully explain observed gender affinity. Instead, even when economists are in the same section, genderaffinityseemstoaffecttheprobabilityofmatching. In response to our initial finding of gender affinity in Table 3, it has been suggested that men and women might have different and incompatible research interests. However, as Table 5 establishes, even if two economists have been assigned to the same section in large part due to their research interests and, as such, are much more likely to coauthor, gender affinity still plays a role in determining the probability that they will coauthor. Though observed gender affinity may still reflect differences in research interests, controlling for section does narrow the extent to which researchinterestscanexplaintheobservedpatterns. Figure 8 provides an illustration of this gender affinity. In this figure, each section is represented by two observations, or data points. For each point, the y-value represents the mean share of women coauthors, aggregated across women section members (red triangles) or men section members (black circles). The x-value of each observation is the fraction of women affiliated with thatsection. Wecanseeapositivecorrelationintheobservations,indicatingthatforbothmenand women, as the fraction of women in the section increases, the fraction of women among section members’ coauthors also tends to increase. The red and black regression lines affirm this association,giventheirpositiveslopecoefficients. Thatsaid,wecanalsoseethattheredtrianglestendto havehighery-valuesthantheblackcircles,indicatingthatforagivenlevelofwomen’srepresentation within a section (x-values), we tend to see women in the section (in red) having more women coauthors than men in the section (in black). Overall, the higher y-values for the red triangles relativetotheblackcirclesillustrategenderaffinityincoauthorships,evenwithinsections. 23

5.4 Counterfactuals Given these results, it is natural to ask the questions: If we were able to reduce the gender divide andencouragemoreinclusionincoauthorshipselections,howmanymorecollaborationswouldbe observed,andhowmanymorepapersmightbeproduced? One way to estimate the role of the gender divide would be to take the set of potential collaborations between mixed-gender pairs and apply higher probabilities of them resulting in a coauthorship, based on the estimates for same-gender pairs. For example, we have 12,390 pairs in our sample with different sections and different genders. If these pairs were to coauthor at the same rate as the different section and same gender pairs, we would see 19 additional coauthorships, which is an 11.6 percent increase in coauthorships among these pairs. If we apply a similar counterfactualtothe3,010pairsinoursamplewiththesamesectionanddifferentgenders,andassume that they were to coauthor at the same rate as the same section and same gender pairs, we would see another 33 additional coauthorships. Combining these with the 19 additional coauthorships from among different-section pairs, we obtain a total boost of 52 additional coauthorships among different-genderpairs,equivalenttoanincreaseof12.4percent. Notably,these52additionalpairsofcoauthorscouldeachcollaborateonmorethanonepaper. Among the coauthoring pairs in our regression sample, we observe, on average, about 1.5 papers per unique pair. Consequently, we might expect these 52 new unique coauthoring relationships to be associated with 78 new coauthorships and as many as 78 new papers. Furthermore, adding these 78 new authorships for men and women would eliminate about 1.5 percentage points of the 4.8 percentage point gap between women’s share of authorships and their representation in the economistpopulation,asubstantialchangegiventhatwomenare26.5percentofthepopulation. We can also evaluate this counterfactual in the context of the gender affinity observed in the distribution of teams in Table 2. We noted earlier that among two- and three-authored papers, the observed count of all-men groups was about 10 to 12 percentage points higher than the model prediction. Under thecounterfactual describedhere, byadding 78new papersauthored bymixedgender teams, we would reduce the gap between observed and model predictions by about 3.5 percentage points. Overall, this indicates that overcoming gender affinity would result in about a 30 percent reduction in the gap between observed outcomes and model predictions under random assignment. Of course, these numbers are only an estimate and do not reflect any resource constraints in producing papers. However, these counterfactuals are a good way to quantify the potential effect of reducing gender affinity and increasing inclusion in coauthorship selections on the rate of paper production by men and women. Additionally, we note that these counterfactual estimates are somewhat smaller than those one we obtain in Section 4.2.2, reflecting the differences in counter- 24

factualsandmodels. However,themainfindingthatreducinggenderaffinitywouldboostwomen’s shareofauthorshipsby1.5to3percentagepointsiscommonacrosscounterfactuals. 6 Coauthorship and paper outcomes We have shown that observed gender affinity in the coauthorships data cannot be explained by similarityoffieldofinterestorlikelihoodofpolicyinteractions,asproxiedbysectionmembership. GenderaffinityalsocannotbeexplainedbysimilarityinPh.D.cohort. Wenextturntoconsidering whetherobservedgenderaffinityisconsistentwithchoicesincoauthorselectionthatmaybetaken to maximize research outcomes. We consider two outcomes: first, how many times a paper is downloadedinitsfirst12monthsofbeingmadepubliclyavailableand,second,whetheraworking paperevergetspublishedinajournal.14 Ifpaperswithsame-gendercoauthorshipgroupingstendto havemoredownloads,orahigherlikelihoodofpublication,thengenderaffinitycouldbedrivenby adesiretomaximizeresearchoutcomesratherthantheauthorshavinggenderbiaspersonally. We find that both downloads and probability of publication could lead men to favor men-only teams, potentially resulting in less inclusion of women in coauthoring networks. Though women do not pay much of a price in terms of downloads for a women-only team, there may be some penalty to women-onlyteamsassociatedwiththeprobabilityofpublication. For our first investigation of outcomes, we use data on how frequently each working paper is downloaded in the first 12 months after the paper is posted. The data on downloads are particularly useful because they can provide a metric of research success or impact in the absence of thepotentialgatekeepingeffectsassociatedwiththepeer-reviewprocess,whichhasbeenshownin somecasestorequirehigherstandardsforwomenauthors(Hengel,2022;HengelandMoon,2023; Alexander et al., 2023; Card et al., 2020).15 Our data on downloads are from the LogEc website. LogEc reports downloads for all papers that are available through the RePEc web services, which includes all FRB working papers.16 A disadvantage of using LogEc data is that they only count downloads that originate via a paper’s RePEc listing. As such, we do not know about downloads thatoccurbysomeonegoingdirectlytotheFRBworkingpaperwebsite. However,thisrestriction is offset by several advantages. First, the download counts are available to the public, including the number of downloads over the first 12 months after posting to the RePEc website, which, for 14Tomatchsampleperiodsinourearlieranalysis,wefocusonworkingpaperspostedinorafter2004. Wealsoend thesamplein2019sothatwemayobservepublicationoutcomesinthefewyearsafterworkingpapersareposted. 15Althoughdownloadsarelikelylessaffectedbygatekeepingbehaviorintheeconomicsprofession,differencesin downloads can also reflect gender differences in the size, reach, or use of professional networks. Given that Ductor etal.(2023)havefoundthatwomentendtohavesmallernetworks,downloadcountsmaystillbelowerforwomen, evenconditionalonresearchquality. 16FormoreonLogEc,seeLogEc(2022),”AboutLogEc,”webpage,https://logec.repec.org/about. htm. 25

FRBworkingpapers,closelymatchesreleasedates.17 Second,andmoreimportantly,LogEctakes steps to ensure data quality, including making sure that reported downloads reflect downloads by humans rather than by automated web mapping services (“robots”).18 As reported in Table 6, we findthatpapersaredownloadedabout38times,onaverage,intheirfirstyearofpublication. We regress downloads per paper on an intercept term along with indicator variables for menonly and women-only teams. We control for the number of coauthors as well, with a variable that counts “additional authors” (after the first one). Table 7 reports the results. In the first column, we see that a paper by a two-authored mixed-gender team would have about 32 downloads, while a men-onlyteamwouldhavealmost9moredownloadsandawomen-onlyteamwouldhaveabout2 fewerdownloads. Havingadditionalcoauthorsdoesboostthenumberofdownloads. Inthesecond column, we can see that the penalty in terms of downloads for being a women-only team was greaterintheearlysample. Inthethirdcolumn,wecanseethatforthelatesample,thepenaltyfor beingawomen-onlyteamdisappears. As such, in the early sample, a coauthoring team of two men gets about 25 percent more downloads in the first year than a mixed-gender or women-only team of two. Additionally, by the late sample, downloads for a women-only team do not differ greatly from those for a mixedgender team. As such, these results are consistent with the explanation that publication outcomes may be driving choices that result in gender affinity. In particular, these results are consistent with our estimated model findings of women not having gender affinity in the early sample and havinggenderaffinityinthelatesampleandwithmenhavinggenderaffinityinbothearlyandlate samples. We next look at how gender groupings shape publication success. Our publications data are from RePEc. These data provide a mapping between FRB working papers and peer-reviewed publications, allowing us to measure whether each paper is eventually published and the time elapsed between the posting of a working paper and its eventual publication. We repeat the same regressions as for downloads but change the outcome variables to, first, an indicator variable for whetherapaperiseverpublishedand,second,totwoadditionaloutcomesthatmeasurewhethera paper is published within or after two years from the working paper publication year. Results are reportedinTable8. First, to understand the baseline, Table 6 shows summary statistics for paper outcomes. We find that about one-third of working papers are eventually published, and the majority of these are 17AspecialthankstoSuneKarlsson,whogaveustheabilitytoquerythedataondownloadsinthefirst12months fromLogEc. Subsequenttoourrequest,LogEcnowprovidesthisinformationforallworkingpaperseries. 18We compared internal FRB data from Google Analytics with RePEc data and found that the RePEc data are more plausible. For example, Google Analytics frequently reported that a number of older vintage papers each got downloadedthesamenumberoftimesamonth,whichseemedunlikely. FormoreonRePEc,seeRePEcteam(2009), “HowAbstractViewsandDownloadsAreCounted,”RePEcBlog,September,19,https://blog.repec.org/ 2009/09/19/how-abstract-views-and-downloads-are-counted/. 26

published within two years of the working paper being posted. The number of authors on each paper has risen from about 2.1 authors per paper, on average, in the early sample to 2.4 in the late sample. Turningtotheregressionresults,weseeinthelastlineofTable8thateachadditionalcoauthor increasesthelikelihoodofpublicationbyaround10percentagepoints,whichisalargeeffectsince it boosts the likelihood of publication from 28 percent for a two-authored, mixed-gender team, to 38 percent for a three-authored team. We also find that the men-only teams have an 11 percentage pointhigherlikelihoodofpublication. Theseresultsholdupevenwhenwelookattheearlysample orthelatesample. The regression results also suggest that women-only teams have a 7 percentage point lower likelihoodofpublication,thoughthisestimatedpenaltyisnotstatisticallysignificant. Additionally, it does seem notable that the reduction in downloads for women-only teams relative to mixedgender teams diminishes greatly between the early and late sample. In contrast, the publication outcomes for women-only and mixed teams are little changed between the early and late sample. This discrepancy between downloads and publication outcomes is suggestive that the peer-review process may be changing more slowly than other aspects of the profession and is consistent with otherpapersfindinggreaterobstaclesforwomeninthepublicationprocess(Hengel,2022;Hengel andMoon,2023;Cardetal.,2020;Alexanderetal.,2023). To better understand these final results, we also report in Table 8 the results for the same regressions but with outcome variables of either the likelihood of publication within two years or for publication after two years. (In the latter set of regressions, we limit the sample to those papers that are not published within two years.) Focusing on the late sample, we find that papers with men-only authoring teams do not have a higher likelihood of publication within two years relative to mixed teams. However, turning to the likelihood of publication after two years, we do see a positive coefficient for the men-only indicator and a near-zero coefficient for the womenonly indicator. That is, we find that men-only teams may have more persistence in publication effortsrelativetomixedandwomen-onlyteams,andthispersistencedrivesthehigherlikelihoodof publicationforpapersoverall.19 Thisfindingofgreaterpersistenceisconsistentwiththefindingby Shastry and Shurchkov (2024) that after receiving a paper rejection from a peer-reviewed journal, women assistant professors perceive a significantly lower likelihood of subsequently publishing thepaperinanyleadingjournalthancomparablemaleassistantprofessors. 19Although beyond the scope of this paper, it would be good to understand whether this increased persistence is rewardedwithhigher-rankingpublicationsorjustmorepublications. Forexample,GintherandKahn(2004)findthat women in the 1989 cohort of assistant professors have 0.3 fewer top-10 publications and 3.8 fewer articles in other journals 10 years into their careers. Sherman and Tookes (2024) find that women finance professors have a similar numberoftoppublicationsasmenfinanceprofessorsbutfewerpublicationsoutsideofthetopthreefinancejournals andtopfiveeconomicsjournals. 27

Overall, our results suggest that gender affinity in coauthoring by men may be consistent with a desire to maximize research outcomes as measured by downloads, as well as likelihood of publication. Notably, these results point to a perhaps lesser-studied consequence of potential gender biasinthepublicationprocess. Ifbiasesintherefereeandcitationprocessleadtoworseoutcomes for women, then as collaborators seek to maximize outcomes, they may be swayed toward less inclusivepracticestoavoidexposingtheirprojecttothesebiases. 7 Career progression Given our results, what advice can we give central banks and others that want to increase both diversity and inclusion in their institutions? Although the number of women economists has increased at the Federal Reserve Board, the share of economists that are women has not increased. TheBoardhastakenanumberofstepstosupportdiversity. However,ifonemeasuresinclusionby whether women and men are likely to coauthor papers together, then inclusion has not increased muchattheFederalReserveBoard. Ourworkcannotdeterminewhatcausesthelackofinclusion; however,itclearlyhasconsequences. Our model shows how gender affinity in coauthorship selections may result in lower observed productivityamongwomen,asmeasuredbythenumberofworkingpapersperperson. Thisfinding is particularly important for understanding career progression, as research output is often cited as an important input into promotion decisions. Additionally, to the extent that there is learning by doing in research, barriers to finding coauthors in early years may result in lower productivity throughout a person’s career. Our data in Table 9 provide some supportive evidence. As seen in row 2c of the table, of the men and women who write papers but do not coauthor with other FRB economists in their first three years of FRB employment, women are much more likely to not write additional papers in the next three years. Furthermore, as seen in row 3c of the table, of the men and women who do not produce papers in the first three years of FRB employment, womenaremuchmorelikelytonotwriteinthenextthreeyearsaswell. Theseresultssuggestthat policies encouraging coauthorship, especially for early-career economists, could boost research productivity. Finally, given that our work in Section 6 shows that the external market may reward men-only teams with more downloads and publications, the Federal Reserve Board and other similar institutions should evaluate their internal evaluation processes to identify and reward quality research to ensure that internal processes do not merely magnify external biases. Internal rewards could encourage inclusion by countering the way in which external markets reward gender affinity in coauthorshipgroupings. 28

8 Conclusion This paper shows that the observed distribution of coauthorship groupings across genders differs from predictions based on random assignment. By modeling coauthor team formation, we show that, in recent years, the observed groupings are consistent with parameters that imply sorting withingender. Thatis,relativetorandomassignment,mentendtocoauthormorewithothermen, while women coauthor more with other women. That said, the modal coauthor for both a woman andamanisstillaman,consistentwithmen’sgreatershareoftheFRBeconomistpopulation. Althoughwefocusongenderinclusivityinthispaperduetodatalimitations,theresultsmaybe informativeforothertypesofinclusivity(orinsularity)incoauthorships,oreveninclusivityingeneral within the economics profession. We note that assortative matching in coauthor relationships may also occur along other characteristics, including common language, nationality, university, physical location, or interests. As it has been previously documented that diverse groups have better outcomes in a variety of settings, barriers to coauthorship with dissimilar individuals may resultinworseoutcomesovertime. 29

References Alexander, D., Gorelkina, O., Hengel, E., and Tol, R. (2023). Gender and the Time Cost of Peer Review. TinbergenInstituteDiscussionPapers2023-044/V,TinbergenInstitute. Auriol, E., Friebel, G., Weinberger, A., and Wilhelm, S. (2022). Underrepresentation of Women in the EconomicsProfessionmorePronouncedintheUnitedStatesComparedtoHeterogeneousEurope. ProceedingsoftheNationalAcademyofSciences,119(16). Azzimonti-Renzo, M., Jarque, A., and Wyckoff, A. (2023). How Are Women Represented in Economic ResearchattheFed? EconomicBrief,23(40). Bayer, A. and Rouse, C. E. (2016). Diversity in the Economics Profession: A New Attack on an Old Problem. JournalofEconomicPerspectives,30(4):221–42. Board of Governors of the Federal Reserve System (2002). Economists, by Name. webpage, http://www.federalreserve.gov/research/name.htm. Accessed December 13, 2019, via https://web.archive.org/web/20030306075802/http://www. federalreserve.gov:80/research/name.htm. Board of Governors of the Federal Reserve System (2017). The Economists. webpage, https: //www.federalreserve.gov/econresdata/theeconomists.htm. Accessed December 13, 2019, via https://web.archive.org/web/20170120163259/https://www. federalreserve.gov/econresdata/theeconomists.htm. Board of Governors of the Federal Reserve System (2022a). Finance and Economics Discussion Series(FEDS). webpage,https://www.federalreserve.gov/econres/feds/all-years. htm. accessedApril21,2022. Board of Governors of the Federal Reserve System (2022b). International Finance Discussion Papers (IFDP).webpage,https://www.federalreserve.gov/econres/ifdp/all-years.htm. accessedApril21,2022. Board of Governors of the Federal Reserve System (2022c). Meet the Economists. webpage, https://www.federalreserve.gov/econres/theeconomists.htm. Accessed May 12, 2022, via https://web.archive.org/web/20220401063559/https://www. federalreserve.gov/econres/theeconomists.htm. Card,D.,DellaVigna,S.,Funk,P.,andIriberri,N.(2020). AreRefereesandEditorsinEconomicsGender Neutral? QuarterlyJournalofEconomics,135(1):269–327. Chari,A.(2023).ReportoftheCommitteeontheStatusofWomenintheEconomicsProfession.InJohnson, W.R.andLippert,T.,editors,AEAPapersandProceedings,volume113,pages815–839. Chari, A. and Goldsmith-Pinkham, P. (2017). Gender Representation in Economics across Topics and Time: EvidencefromtheNBERSummerInstitute. WorkingPaper23953,NationalBureauofEconomic Research. 30

Datta, D. D., Johannsen, B. K., Kwon, H., and Vigfusson, R. J. (2021). Oil, Equities, and the Zero Lower Bound. AmericanEconomicJournal: Macroeconomics,13(2):214–253. Datta,D.D.andTzur-Ilan,N.(2024). GenderGapsintheFederalReserveSystem. Workingpaper. Davies, B. (2022). Gender Sorting among Economists: Evidence from the NBER. Economics Letters, 217:110640. Ductor, L. (2015). Does Co-authorship Lead to Higher Academic Productivity? Oxford Bulletin of EconomicsandStatistics,77(3):385–407. Ductor, L., Goyal, S., and Prummer, A. (2023). Gender and Collaboration. Review of Economics and Statistics,105(6):1366–1378. Gertsberg,M.(2022). TheUnintendedConsequencesof#MeToo: EvidencefromResearchCollaborations. CambridgeWorkingPapersinEconomics. Ginther, D. K., Currie, J. M., Blau, F. D., and Croson, R. T. A. (2020). Can Mentoring Help Female AssistantProfessorsinEconomics? AnEvaluationbyRandomizedTrial. InJohnson,W.R.andHerbert, G.,editors,AEAPapersandProceedings,volume110,pages205–09. Ginther,D.K.andKahn,S.(2004). WomeninEconomics: MovingUporFallingOfftheAcademicCareer Ladder? JournalofEconomicPerspectives,18(3):193–214. Hengel, E. (2022). Publishing while Female: Are Women Held to Higher Standards? Evidence from Peer Review. EconomicJournal,132(648):2951–2991. Hengel,E.andMoon,E.(2023). GenderandEqualityatTopEconomicsJournals. Workingpaper. Hospido,L.,Laeven,L.,andLamo,A.(2022).TheGenderPromotionGap: EvidencefromCentralBanking. ReviewofEconomicsandStatistics,104(5):981–996. Internet Archive (2019). Wayback Machine. webpage, https://web.archive.org/. accessed December13,2019andApril21,2022. LogEc (2022a). International Finance Discussion Papers - Top 5000 Working Papers for the 12 Months from First Access/Download through RePEc. webpage, https://logec.repec.org/ scripts/seritemstat.pf?topnum=5000&mrange=No&fm=&lm=&fromstart=Yes& ms=12&sortby=ld&h=repec%3Afip%3Afedgif&.submit=New+List&.cgifields= mrange&.cgifields=fromstart. accessedApril21,2022. LogEc (2022b). International Finance Discussion Papers - Top 5000 Working Papers for the 12 Months from First Access/Download through RePEc. webpage, https://logec.repec.org/ scripts/seritemstat.pf?topnum=5000&mrange=No&fm=&lm=&fromstart=Yes& ms=12&sortby=ld&h=repec%3Afip%3Afedgif&.submit=New+List&.cgifields= mrange&.cgifields=fromstart. Accessed: 2022-04-21. Lundberg,S.andStearns,J.(2019). WomeninEconomics: StalledProgress. JournalofEconomicPerspectives,33(1):3–22. 31

Lyons,C.,Wilderman,A.,Fortson,M.,Lewis,S.,Luckman,A.,Shapiro,E.,Rogers,T.,andVanHuysen,M. (2021). TheBoardEconomicsDivisionsCanEnhanceSomeofTheirPlanningProcessesforEconomic Analysis. TechnicalReport2021-MO-B-001,FederalReserveBoardOfficeoftheInspectorGeneral. McDowell, J. M., Singell Jr, L. D., and Stater, M. (2006). Two to Tango? Gender Differences in the DecisionstoPublishandCoauthor. EconomicInquiry,44(1):153–168. Sarsons, H. (2017). Recognition for Group Work: Gender Differences in Academia. American Economic Review,107(5):141–145. Sarsons, H., Ge¨rxhani, K., Reuben, E., and Schram, A. (2021). Gender Differences in Recognition for GroupWork. JournalofPoliticalEconomy,129(1):101–147. Shastry,G.K.andShurchkov,O.(2024). RejectorRevise: GenderDifferencesinPersistenceandPublishinginEconomics. EconomicInquiry,62(3):933–956. Sherman,M.G.andTookes,H.(2022).FemaleRepresentationintheAcademicFinanceProfession.Journal ofFinance,77(1):317–365. Sherman, M. G. and Tookes, H. (2024). Gender Balance in the Academic Finance Profession. CSWEP News,2:15–18. Stansbury, A. and Schultz, R. (2023). The Economics Profession’s Socioeconomic Diversity Problem. JournalofEconomicPerspectives,37(4):207–30. Tang,C.,Ross,K.,Saxena,N.,andChen,R.(2011). What’sinaname: Astudyofnames,genderinference, andgenderbehaviorinfacebook. InXu,J.,Yu,G.,Zhou,S.,andUnland,R.,editors,DatabaseSystems forAdvancedApplications,pages344–356.SpringerBerlinHeidelberg. Wu, A. H. (2018). Gendered Language on the Economics Job Market Rumors Forum. In Johnson, W. R. andMarkel,K.,editors,AEAPapersandProceedings,volume108,page175–79. 32

A Figures Figure1: NumberofeconomistsattheFederalReserveBoard Count 500 Women 400 Men 300 200 100 0 2005 2009 2013 2017 2021 Figure2: WomeneconomistsattheFederalReserveBoard Percent 50 40 30 20 New hires 10 All economists 0 2005 2009 2013 2017 2021 33

Figure3: HiringattheFederalReserveBoard Count 60 Women 50 Men 40 30 20 10 0 2005 2009 2013 2017 2021 Figure4: NethiringattheFederalReserveBoard Count 50 Women 40 Men 30 20 10 0 −10 2005 2009 2013 2017 2021 34

Figure5: Economistexperiencedistribution,2004-21 Percent 70 Total 60 Women 50 Men 40 30 20 10 0 Rookies Midcareer Seasoned Note: Barsshowthedistributionsofyearsofexperienceforeconomist-yearobservationsbetween 2004and2021. ExperienceisdefinedasyearssincePh.D.graduation. Rookiesaredefinedas thosewith3orfeweryearssincePh.D.graduation,midcareersarethosewith4to7yearsof experience,andseasonedeconomistsarethosewithatleast8yearsofexperience. Figure6: Distributionofpapersperperson,bygender Percent 40 Women 35 Men 30 25 20 15 10 5 0 0 1 2 3−4 5−10 11+ Note: Percentageofwomenandmeneconomists,categorizedbythenumberofpapers. Sampleis restrictedtoFederalReserveBoardeconomistsemployedatthestartof2022,withatleast3years ofservice. 35

Figure7: Authorshipsbyyearandgender Percent women 30 25 20 15 10 FRB economists FRB authorships 5 Total authorships 0 2005 2009 2013 2017 2021 Note: FederalReserveBoard(FRB)authorshipsandtotalauthorshipsseriesrepresent3-year movingaverages. Figure8: Shareofwomenamongcoauthorsandsectionmembers 60 y = 14+0.46 x y = 13+0.29 x 50 40 30 20 10 0 0 10 20 30 40 50 60 Percent women in section srohtuaoc gnoma nemow tnecreP Women Men Note: Eachsectionisdenotedbyaredtrianglethatrepresentswomenaffiliatesandablackcircle thatrepresentsmenaffiliates. Thex-valueofeachobservationisthefractionofwomenaffiliated withthatsection. They-valuerepresentsthemeanshareofwomencoauthors,whenaveraging acrosssectionmembersofaparticulargender. Wecanseeapositivecorrelationinthe observations,indicatingthatforbothwomenandmen,asthefractionofwomeninthesection increases,thefractionofwomencoauthorsalsotendstoincrease. Theredandblacklines representgender-specificregressions,whichindicatethatthefractionofwomencoauthorsrises withthewomen’sshareofsectionmembership. Thedashedbluelinerepresentsthelinealong whichthepercentwomenamongcoauthorsisequaltothepercentwomeninthesection. 36

B Tables Table1: FRBeconomistsandworkingpaperproduction A.Economistseveremployed,2003–22 Cumulativeauthorships Gender Economists Mean Median Men 539 6.3 4 Women 191 4.5 3 All 730 5.8 4 B.Economistsemployedin2003 Cumulativeauthorships Gender Economists Mean Median Men 168 10.8 8 Women 53 6.5 5 All 221 9.8 7 C.Economistsemployedin2021 Cumulativeauthorships Gender Economists Mean Median Men 313 5.9 4 Women 112 5.0 4 All 425 5.7 4 Note: Tableshowsmeanandmediannumberofworkingpapersauthoredbymenandwomen. In panelA,wereportaveragevaluesforworkingpaperspublishedthrough2022forallFederal ReserveBoard(FRB)economistsinourdataset,whichincludesanyoneemployedatanytime over2003-22. InpanelsBandC,wereportaveragevaluesforFRBeconomistsemployedin2003 and2021,respectively. WhereaspanelBcapturesforonecohorttheauthorshipscumulatedovera longperiod,panelCdemonstratesthedistributionofobservedauthorshipsamongthepopulation atapointintime. 37

Table2: Coauthorshipdistribution Earlysample(2004–12) Latesample(2013–21) ObservedPapers Model Diff. ObservedPapers Model Diff. Authors Count Share Share Count Share Share 1author 278 33 254 22 Man 227 82 75 7 193 76 72 4 Woman 51 18 25 -7 61 24 28 -4 2authors 319 37 411 35 Men 216 68 56 12 264 64 52 12 Mixed 88 28 38 -10 115 28 40 -12 Women 15 5 6 -2 32 8 8 0 3authors 193 23 344 29 Men 99 51 42 10 171 50 38 12 Majoritymen 74 38 42 -4 108 31 43 -12 Majoritywomen 19 10 14 -5 52 15 17 -1 Women 1 1 2 -1 13 4 2 2 4+authors 64 7 163 14 Men 23 36 31 5 52 32 28 4 Majoritymen 29 45 42 3 78 48 42 6 Mixed 7 11 22 -11 24 15 24 -9 Majoritywomen 5 8 5 3 9 6 6 -1 Women 0 0 0 0 0 0 1 -1 All 854 100 25 75 1172 100 28 72 Note: Theboldedrowsreporttheobservedcountandshareofpapersforeachpossibleteamsize. Theremainingrowsreport,foreachteamsize,theobservedcountandshareforeachpossible gendercomposition,conditionalonteamsize. Theserowsalsoreportthemodelpredictedshare foreachpossiblegendercomposition,conditionalonteamsize. Forthesepredictedvalues,we usetheobservedparametersof25.4percentwomenin2004-12and27.6percentwomenin 2013-21. Lastly,the“Diff.”columnreportsthedifferenceinobservedminusmodelsharesin percentagepoints. Positivevaluesforthesedifferencesindicatetheobservedshareishigherthan themodelprediction. 38

Table3: Estimatedmodel Estimation f g p p W-Authorships w m Unrestrictedmodelestimates 1. FullSample(2004-21) 0.265 0.209 1.282 0.716 0.219 2. Earlysample(2004–12) 0.254 0.184 0.853 0.739 0.193 3. Latesample(2013–21) 0.276 0.241 1.341 0.641 0.230 Counterfactualsusinglatesampledata A.p = 1 0.276 0.241 1.341 1.000 0.268 m B.p = p = 1 0.276 0.241 1.000 1.000 0.262 m w C.g = f 0.276 0.276 1.341 0.641 0.251 D.p = 2.012 0.276 0.241 2.012 0.641 0.264 w E.g = f,p = 2.012 0.276 0.276 2.012 0.641 0.280 w F.g = f,p = p = 1 0.276 0.276 1.000 1.000 0.280 m w Note: Modelparametersareestimatedusingdataonwomen’srepresentation,asreportedinthe firstcolumn,andthedatareportedinTable2forpaperswithupto3authors. Thefinalcolumnof thetablereportstheestimatedshareofauthorshipsbywomenbasedonthemodelparameters. 39

Table4: Linearprobabilitymodelofcoauthorship 1 2 SameSection 7.78 7.18 (37.39) (21.05) BothMen 0.39 0.21 (2.19) (1.04) BothMen*SameSection 0.90 (2.04) BothWomen 0.12 −0.18 (0.33) (−0.45) BothWomen*SameSection 1.47 (1.68) CohortDifference −0.07 −0.07 (−7.19) (−7.18) Constant 1.95 2.06 (11.01) (11.19) Observations 39,976 39,976 Note: ThepoolofpotentialcoauthorsincludesallFederalReserveBoard(FRB)economistsfor whomwehavePh.D.yearandsectionaffiliationdataandwhohavecoauthoredwithanotherFRB economist. Weconstructpairwisecombinationsoftheseindividualsandrestrictthepoolof potentialpairstothosepairsofindividualswithoverlappingdivisionaffiliationandoverlapping FRBtenure. T-statisticsarereportedinparentheses. 40

Table5: Sectionandgenderaffinity Marginaleffectofmovingfrom Likelihoodofcoauthorship,given: Sectionaffiliation differenttosamesection Different Same Percentagepoints Percent Differentgender 1.84 9.02 7.18 389.23 Bothmen 2.05 10.13 8.08 394.04 Bothwomen 1.67 10.31 8.65 518.90 Memo: Weightedaverage 1.95 9.72 7.77 399.45 Note: Usescolumn2ofTable4asbaselineandconditionsonaveragevalueof3yearsofcohort difference. Table6: Summarystatisticsforpublicationoutcomes Fullsample Earlysample Latesample Mean St.dev. Mean St.dev. Mean St.dev. Downloads 38.23 32.73 40.08 34.48 36.52 30.94 2-authormixedteams 31.56 28.01 33.47 31.96 29.69 23.54 Publication 0.36 0.48 0.39 0.49 0.33 0.47 Publicationwithin2years 0.22 0.42 0.24 0.43 0.21 0.41 Numberofauthors 2.23 1.18 2.08 1.04 2.37 1.28 Count 1776 NA 854 NA 922 NA Note: Fullsampleincludespaperspostedfrom2004to2019;earlysampleincludespapersposted from2004to2012;andlatesampleincludespaperspostedfrom2013to2019. NA:Notapplicable. 41

Table7: 12-monthdownloads Fullsample Earlysample Latesample Intercept 28.92 31.15 26.31 (12.73) (8.93) (8.84) Men-only 8.88 7.99 9.62 (4.58) (2.63) (3.88) Women-only −1.58 −4.45 1.37 (−0.48) (−0.84) (0.33) Additionalauthor 3.31 3.80 3.36 (3.53) (2.59) (2.78) Observations 1,776 854 922 Note: Fullsampleincludespaperspostedfrom2004to2019,earlysampleincludespapersposted from2004to2012,andlatesampleincludespaperspostedfrom2013to2019. 42

Table8: Likelihoodofpublication Fullsample Earlysample Latesample Ever Within After Ever Within After Ever Within After publish 2years 2years publish 2years 2years publish 2years 2years Intercept 0.18 0.12 0.05 0.18 0.11 0.06 0.16 0.12 0.03 (5.47) (4.22) (1.68) (3.70) (2.52) (1.22) (3.58) (3.14) (0.74) Men-only 0.11 0.04 0.10 0.15 0.08 0.12 0.08 0.01 0.09 (3.90) (1.70) (3.86) (3.53) (2.17) (2.87) (2.05) (0.28) (2.74) Women-only −0.07 −0.07 0.004 −0.04 −0.05 0.02 −0.08 −0.08 0.004 (−1.57) (−1.77) (0.09) (−0.61) (−0.81) (0.35) (−1.29) (−1.52) (0.09) Additionalauthor 0.10 0.07 0.06 0.11 0.08 0.07 0.10 0.07 0.06 (7.52) (5.98) (4.65) (5.53) (4.20) (3.64) (5.71) (4.55) (3.49) Observations 1,776 1,776 1,378 854 854 652 922 922 726 Note: Fullsampleincludespaperspostedfrom2004to2019,earlysampleincludespapersposted from2004to2012,andlatesampleincludespaperspostedfrom2013to2019. 43

Table9: Coauthoringandcareerprogression Years1-3 Years4-6 Men Women Men Women 1. Coauthors 27 25 1a. Coauthors 71 74 1b. Writes 5 0 1c. Doesnotwrite 11 11 1d. Exits 12 15 2. Writes 40 37 2a. Coauthors 58 50 2b. Writes 16 15 2c. Doesnotwrite 15 25 2d. Exits 11 10 3. Doesnotwrite 19 23 3a. Coauthors 39 20 3b. Writes 9 4 3c. Doesnotwrite 39 64 3d. Exits 12 12 4. Exits 13 14 Note: IncludesFederalReserveBoardeconomistsstartingafterMarch2003andbeforeJanuary 2017. 44

C Additional figures and tables Table A.1 reports summary statistics for downloads within the first year of a paper being posted, tabulated by coauthor composition.1 The distribution of downloads is heavily right skewed. For thefullsample,themediannumberofdownloadsis30,themeannumberofdownloadsis38,and the top 10 percent of papers have at least 75 downloads. Because papers with more coauthors automaticallyhaveabroadernetworkofcolleagues,itisperhapsunsurprisingthatweseedownloads rising slightly with the number of coauthors. Although we observe that solo- and two-authored papers by men tend to have more downloads than those authored by women, this pattern does not holdonceweexpandtopaperswithmorecoauthors. TableA.1: Distributionofdownloadsperpaper Paper 12-monthdownloads Authors Count Mean Median P90 1Author 497 36 28 68 Man 392 39 30 76 Woman 105 26 23 44 2Authors 655 37 29 75 Men 436 40 33 76 Mixed 178 32 23 70 Women 41 33 27 53 3Authors 454 40 33 81 Men 233 36 32 75 Majoritymen 151 28 21 59 Majoritywomen 59 45 35 93 Women 11 44 15 54 4+Authors 170 43 35 81 Men 58 42 36 74 Mixed 112 43 34 83 All 1776 38 30 75 Men 1119 36 32 75 Majoritymen 151 28 21 59 Mixed 290 41 32 79 Majoritywomen 59 36 26 72 Women 157 29 23 47 Note: Sampleincludespaperspostedfrom2004to2019. Summarystatisticsareroundedto nearestinteger. 1AlthoughwehavedataondownloadsforallFRBworkingpapers,thetrendsinobserveddownloadsindicatethat this measure is most likely useful for papers posted after 2002. We further limit the sample to begin in 2004, to be consistentwiththeperiodforwhichwehavethehighest-qualitydataonindividualcovariates,andtoendin2019,to beconsistentwiththeperiodforwhichwestudypublicationoutcomes. A-0