Women's Labor Force Exits during COVID-19: Differences by Motherhood, Race, and Ethnicity

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Women’s Labor Force Exits during COVID-19: Differences by Motherhood, Race, and Ethnicity Katherine Lim, Mike Zabek 2021-067 Please cite this paper as: Lim, Katherine, and Mike Zabek (2023). “Women’s Labor Force Exits during COVID- 19: Differences by Motherhood, Race, and Ethnicity,” Finance and Economics Discussion Series 2021-067r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2021.067r1. 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.

Women’s Labor Force Exits during COVID-19: Differences by Motherhood, Race, and Ethnicity Katherine Lim* Mike Zabek† June 14, 2023 First posted: September 23, 2021 Abstract While the descriptive impacts of the pandemic on women have been well documented in the aggregate, we know much less about the impacts of the pandemic on different groups of women. Aftercontrollingfordetailedjobanddemographiccharacteristics,includingoccupation and industry, we find that the pandemic led to significant excess labor force exits among women living with children under age six relative to women without children. We also find evidence of larger increases in exits among lower-earning women. The presence of children predictedlargerincreasesinexitsduringthepandemicamongLatinaandBlackwomenrelative to White women. Overall, we find evidence that pandemic induced disruptions to childcare, includinginformalcarefromfamilyandfriends. Ourresultssuggestthattheuniqueeffectof childcaredisruptionsduringthepandemicexacerbatedpre-existingracialandincomeinequalitiesamongwomen. Keywords: Women,LaborForceParticipation,Race,Ethnicity,LaborSupply,COVID-19 JELNumbers: J16,J70,H31,I14,I18 *UnitedStatesDepartmentofAgriculture,EconomicResearchService;1400IndependenceAve,SW;Washington, DC;20250;katherine.lim@usda.gov. †Correspondingauthor;BoardofGovernorsoftheFederalReserveSystem;MailStopI-303;20thStreetandConstitutionAvenueN.W.; Washington,DC20551; mike.zabek@frb.gov. Additionalresultsandcopiesofthecomputer programsusedtogeneratetheresultspresentedinthearticleareavailablefromthecorrespondingauthor. Theopinions,analysis,andconclusionsarethoseoftheauthorsanddonotindicateconcurrencebytheFederalReserveBoard, theFederalReserveSystem,anyoneassociatedwiththeseorganizations,oranyoneelse.Additionally,thefindingsand conclusionsinthispresentationarethoseoftheauthorsandshouldnotbeconstruedtorepresentanyofficialUSDAor U.S.Governmentdeterminationorpolicy. ThisresearchwassupportedinpartbytheU.S.DepartmentofAgriculture, EconomicResearchService. DataAvailabilityStatement: DatausedforthestudyarepubliclyavailableviaFlood etal.(2020). Programsanddatasetsgeneratedduringand/oranalysedduringthecurrentstudyarealsoavailablefrom the corresponding author on reasonable request. Other Statements and Declarations: The authors declare that no funds,grants,orothersupportwerereceivedduringthepreparationofthismanuscript. Theauthorshavenorelevant financial or non-financial interests to disclose. All authors contributed to the data analysis, writing, and revising of thepaper. Acknowledgments: ThisprojecthasbenefitedfromcommentsfromJohnBound,DavidBuchholtz,Curie Chang, Jeff Larrimore, Alicia Lloro, Joshua Montes, Ryan Nunn, Jessica Ott, Christopher Smith, and Erin Troland, amongothers.

Disruptions during the first year of the COVID-19 pandemic affected women’s working lives in many ways. Restrictions had direct effects on the ability to safely work in person while families’ formal and informal childcare arrangements were interrupted. Daycare centers were closed, schoolsswitchedtoremotelearning,andhealthconcernsrestrictedtheabilityoffamilyandfriends to care for children. These disruptions were not felt uniformly across women, and this paper demonstratestheirunequaleffects. We build on previous work by isolating the role that children had on women’s labor force exits relative to other women with comparable jobs in the United States. We find that the pandemic led more women with children under age six to leave the labor force. The pandemic also had unequal effectsbyincome,race,andethnicity. Ourresultssuggestthatchildcaredisruptionsandincreased caregivingresponsibilitiesledtolargerincreasesinexitsamonglower-earningwomenlivingwith children compared to higher-earning mothers with similarly aged children. While factors relating to employment explain the largest share of the differences in exit rates between women of color and White women both before and during the pandemic, the effect of having children and the interactions of having children with women’s pre-pandemic earnings and marital status became important explanatory factors during the pandemic. We find that industry and occupation played a smaller role, in contrast to their major role in employment losses during the early months of the pandemic(Cortes&Forsythe,2023). We use panel data to identify labor force exits of previously employed women during the first year of the pandemic to study the role that children and increased caregiving responsibilities played. We compare the likelihood of labor force exit between employed women with children and observably similar women who do not have children. We allow for the effect of children to differ based on interactions of children’s ages, previous earnings, and marital status. The strategy allowsustoconditiononinitialemploymentandtocontrolforarichvarietyofpre-pandemiccharacteristics, including earnings, occupation and industry level effects, and the differing geographic effectsofthepandemic. Our main specification compares women’s exits during the pandemic with exits before the pandemic,becausechildrenalsoaffectedwomen’slaborforceexitsbeforethepandemic. Specifically, we identify “excess exits” from the labor force during the pandemic above and beyond those expected based on the existing empirical relationships directly before the pandemic and during the GreatRecession. Ourfindingssuggestthatwomen’slaborforceexitsduringthepandemicamplifiedexistingeconomicinequalitybasedontwopiecesofevidence. First,weestimatealargereffectofthepandemic on lower-earning mothers’ labor force exits. Second, a decomposition shows that living with childrenhelpstoexplainlargerincreasesinlaborforceexitsamongLatinasandBlackwomenrelative to White women. Notably, differential effects of children by marital status and previous earnings 1

meaningfullycontributetolaborforceexitgapsduringthepandemic,butarelessimportantbefore it. Education,occupation,andindustryalsoexplainalargeportionoftheracialdifferencesinexits, including the increased gaps during the pandemic. However, a substantial portion of the overall gapsaswellastheincreaseingapsremainsunexplained,whichcouldbeattributabletodifferences intheeffectsofcovariatesbyraceandethnicity,covariatesthatarenotincludedintheanalysis,or additionalunmeasureddifferences. Our results speak to the effects of childcare, including informal childcare provided by grandparents, on women’s labor force participation and economic inequality. A substantial literature in laboreconomicsstudiestheeffectsofformalchildcare,focusingonvariationincostsandavailability of pre-K schooling.1 A broader literature has emphasized the importance of often unmeasured non-market work, including informal childcare and some studies have used variation in workers’ proximity to their children’s grandparents to show that availability of informal care also affects parents’labormarketoutcomes.2 Weusethepandemicasanaturalexperimenttoshowwhathappened to mothers when all forms of childcare were severely constrained. The labor force exits we identify in the paper may have lasting negative effects on women’s future earnings as workforce interruptions and lower levels of experience, which are more common for women, still contribute tothegenderearningsgap(Blau&Kahn,2017). The results also add to the COVID-19 literature by focusing on heterogeneity between women thatisotherwiseobscuredbyafocusonoveralldifferencesbetweenmenandwomen. Thepatterns that we investigate by parenthood, race, and ethnicity are not as apparent when looking at men.3 The differences between groups of women are also larger than those between men and women overall(C.Goldin,2022).4 Our analysis of labor force participation a year into the pandemic also contextualizes the studies examining gender and demographic differences early on in the pandemic. In particular our results demonstrate the gendered effect of informal and formal child care disruptions as a result of of women’s frequently larger role in informal caregiving.5 Our general finding is that the labor supply related effects of childcare disruptions were larger contributors to labor force exits than 1SeeMorrissey(2017)forasurvey. 2PonthieuxandMeurs(2015)provideasummaryofliteraturesongenderandnon-marketworkandComptonand Pollak(2014)andKrolikowskietal.(2020)havelinkedparentslaboroutcomestotheirproximitytotheirchildren’s grandparents. 3Althoughmenofcolorweremorelikelytoleavethelaborforceintheearlymonthsofthepandemic,thatdifferencemoderatedinthefallof2020suchthatlaborforceparticipationdeclinesweresimilarformenofcolorandWhite meninMarch2021. Menlivingwithchildrenlookedsimilartomenwithnochildrenathomeintermsoflaborforce participationratesthroughoutthepandemic. Seeappendixfigures1and2fordetails. 4Other papers investigating overall gender differences include: Leigh et al. (2021); Luengo-Prado (2021); Pitts (2021);Couchetal.(2022),GarciaandCowan(2022);andHansenetal.(2022). 5Augustine and Prickett (2022) document patterns in childcare time by gender in the U.S., showing increases amongmenandwomenwhileCostoyaetal.(2022)showincreasesinunpaidactivitieswerelargerforwomenthan meninArgentina. 2

were differences by women’s occupation and industry. This finding is distinct with studies of the early pandemic months, which found very large differences in employment related to exposure by occupationandindustry–includingamongwomencaregivers.6 Data We study the labor force participation of women during the first year of the COVID-19 pandemic, withafocusonoutcomesfromSeptember2020toFebruary2021. Ouranalysesusemonthlydata from the Current Population Survey (CPS) from the U.S. Census Bureau and the U.S. Bureau of LaborStatisticsaccessedfromIPUMS(Floodetal.,2020).7 Nearly all of our analysis uses a linked longitudinal sample of individuals where we use an exact 12 month lag. Individuals in this main sample are observed twice: first at year and month t and second at year and month t−12. While the linking reduces the sample size substantially, it allows us to create a sample of individuals who were employed in their first sample observation andprovidesinformationontheirjobcharacteristicsinyear-montht. Focusing on previously employed women allows us to measure the job characteristics of previously employed women and focuses the analysis on women who are attached to the labor force. The linking is particularly meaningful during the pandemic, when we are able to observe women in jobs before the pandemic’s onset. So as to only include pre-pandemic jobs we we only include observationsthroughFebruaryof2021. Thereforethepre-pandemicobservationisfromFebruary 2020orbefore.8 Our sample includes prime-working-age women aged 25 to 54 who were employed twelve months before. We use information on employment and labor force status using the standard CPS definitions to categorize respondents as employed, unemployed, or not in the labor force.9 AllresultsusinglinkedobservationsareweightedusinglongitudinalweightsprovidedbyIPUMS. We characterize respondents’ race and ethnicity by calling them Latino if they say they are of Hispanic, Latino, or Spanish origin. Among those who answer that they are not Latino, we characterize them according to their (single) reported race as Black, White, or other. We focus on Black, White and Latino respondents in this paper because the other racial groups have sample sizestoosmalltoseparatelyanalyze. 6Heggeness(2020);RussellandSun(2020);andAlbanesiandKim(2021) 7The Current Population Survey currently contains only information about sex, not gender. So we use sex as an imperfectproxyforgender. 8Forourmainsample,wearelinkingrespondents’surveyswhentheyareintheoutgoingrotationgroupsincewe relyonpre-pandemicearnings,whichareonlyobservedduringcertainmonthsoftheCPS. 9SinceoursampleofemployedworkersisfrombeforeMarchof2020andourmainresultsuselaborforceparticipation,notunemployment,ourmeasuresarenotsubjecttoissuesarisingfromthemisclassificationofworkerswho areontemporarylayoffduringtheCOVIDpandemic. 3

Weuseinformationontheagesofotherindividualsinthehouseholdindependentfromfamilial relationships to create indicators for the presence of children of different ages. This measure has the benefit of including caregiving responsibilities for children in the household even if they are notone’sownchildren,althoughitmaydifferfromotheranalysesthatfocusonlyonrespondents’ children. We also consider industry- and occupation-level impacts of COVID-19 as measured by special questions added to the CPS about COVID-19 in the summer of 2020. Specifically, we construct industry- and occupation-level indices of the percentage change in employment from one year earlier,theshareofworkerswhoareworkingfromhome,andtheshareofworkerswhoresponded thattheyhadlostworkinthepastfourweeksbecauseofthepandemic(regardlessofwhetherthey were paid). To remove a mechanical correlation in our measures as they are applied to women’s labor force exits, we construct the measures using labor market experiences of men. We also increase our sample size in sometimes small (four digit) industries and occupations by pooling observationsfromMaytoAugust2020.10 Finally,wenormalizeindividuals’usualweeklyearningstocontrolforearningsoneyearpriorto year-montht. Alldemographicandemploymentvariablesaremeasuredfromthefirstobservation, att−12,whileouroutcomesarefromthesecondobservation. Pandemic Patterns in Labor Force Exits Policymakers and researchers have focused on the role of caring for children in explaining higher declines in labor force participation among women during the pandemic (Albanesi & Kim, 2021; Furman et al., 2021). As we show in figure 1, previously employed women in households with young children saw the sharpest increases in labor force exits, followed by women living with school-aged children. Women in households with no children under age 13 have exit rates that are only around 1 percentage point higher than before the pandemic.11 The plot also shows that labor force exits were declining more so for women with young children before the pandemic meaning their larger pandemic increases represented a sharp break from previous trends.12 In contrast to the patterns among women, there were relatively small differences in labor force exits betweenmenwithandwithoutchildrenasshowninappendixfigure1. Together,theresultssuggest that the presence of children influenced women’s labor force participation more than they did for men, highlighting the effects of childcare disruptions combined with many women’s larger roles 10Ourexerciseismeanttobedescriptive.However,theseimpactsarequiteplausiblyexogenousinthatitisunlikely thatthedifferencesareduetotheselectionofwomenintooccupationsandindustriesforotherreasons. 11Weuseseasonally-adjustedthreemonthaveragevaluescomputedfromJanuary2003toFebruary2020toadjust formonthlyseasonalityinouroutcomevariables. Alloutputsareweightedusingsamplingweights. 12SeeC.Goldin(2022)forfurtherdiscussionofrecentgainsinparticipationamongwomenwithyoungchildren. 4

inchildcarecomparedwithmen. Pandemic labor force exits also differed by race and ethnicity. Women of color saw larger and more persistent increases in exits relative to White women. Men of color also experienced large increasesinexits,butthedifferencesbyraceandethnicitynarrowedafterthesummerof2020. As weshowinfigure2,BlackwomenandLatinas,whowereworkingoneyearprior,sawbetweena4 and5percentagepointincreaseintheirlaborforceexitscomparedtobetweena1and2percentage pointincreaseforWhitewomen.13 Againweseedifferencesbetweenmenandwomeninpandemic exit changes by race and ethnicity. As shown in appendix figure 2, exit rates for Black and Latino menincreasedbymorethantheydidforWhitemenbutthosedifferencesclosedsubstantiallyafter summerof2020whilethedifferencesforwomenremained. Thelaborforceexitsthatwedocumentmirrorwhatpreviousstudieshaveshownforemployment patterns. In the early months of the pandemic, employment losses were larger for women relative to men (Albanesi & Kim, 2021) and for workers of color (Cortes & Forsythe, 2023; Couch et al., 2020). The patterns of labor force exits by race and ethnicity suggest that employment losses translatedintolaborforceexitsforwomeningeneralandBlackwomenandLatinasinparticular. Figure1: LaborForceExits,byPresenceofChildren .06 .05 .04 .03 .02 .01 0 −.01 0202 yraunaJ morf egnahC Aged 0 to 5 Aged 6 to 12 None under age 13 Jul 18 Jan 19 Jul 19 Jan 20 Jul 20 Jan 21 Note: Plotted are three-month moving average changes in labor force exits for prime-working-age workers, by the presence of children aged 0 to 5 and 6 to 12 before the pandemic among workers who were employed one year prior. Each is adjusted for monthly seasonality based on average monthly values from January 2003 to February 2020. Statisticsareweightedusingsamplingweights. DataarefromtheCurrentPopulationSurveydownloadedfrom IPUMSFloodetal.(2020). 13TherewerealsolargerdeclinesinexitsinthemonthsbeforethepandemicamongLatinasandBlackwomenthan therewereamongWhitewomen. Sothepandemicrepresentedabreakfromthesepre-existingtrends. 5

Figure2: LaborForceExits,byRaceandEthnicity .05 .04 .03 .02 .01 0 −.01 −.02 −.03 0202 yraunaJ morf egnahC Black Latina White Jan 19 Jul 19 Jan 20 Jul 20 Jan 21 Note: Plotted are three-month moving average changes in labor force exits for prime-working-age workers, by race andethnicityamongworkerswhowereemployedoneyearprior. Eachisadjustedformonthlyseasonalitybasedon averagemonthlyvaluesfromJanuary2003toFebruary2020. Statisticsareweightedusingsamplingweights. Data arefromtheCurrentPopulationSurveydownloadedfromIPUMSFloodetal.(2020). Explaining Women’s Labor Force Exits In order to understand what drove differences in labor force exits, we conduct two interrelated analyses of women’s labor force exits by the presence of children and across racial groups. First, we further examine patterns in women’s labor force exits during the pandemic to test if they are most plausibly related to childcare disruptions and increased caregiving responsibilities or other differencesbetweenwomenwithandwithoutchildren. Inourpreferredspecification,weestimate theadditionalrolethatthesefactorsplayedinlaborforceexitsduringthepandemicrelativetothe period immediately before the pandemic. We find evidence that childcare disruptions and general increases in caregiving did lead to additional labor force exits by women with children during the pandemic. Next, we use a modified, non-linear Oaxaca decomposition technique (Fairlie, 2005) to show the extent to which covariates – like motherhood, wages, and occupational sorting – can explain Latinas and Black women’s higher rates of labor force exits prior to and during the pandemic. One result from the decomposition analysis is that the additional impacts of children on labor force exits during the pandemic, particularly among low earning women, were the single biggest explanatory factor for the larger increases in labor force exits among Latinas and Black women. However,theincreasesarestilllargelyunexplainedbythecovariatesweobserve. 6

EmpiricalMethodology Our analysis of labor force exits uses linear probability models to predict the likelihood that a woman who was previously employed will have exited the labor force in the previous 12 months. Wecharacterizeobservationsintothreecategoriesbasedonthemonthofthelastinterview.14 First, pandemic observations includes women whose second observation occurred between September 2020 and February 2021. The period coincides with the beginning of the typical 2020-21 school year and the conclusion of the first six months of the pandemic. We intentionally exclude the first six months of the pandemic and begin in September 2020 so that the results will not be overly influenced by relatively short duration spells out of the labor force following job losses in March 2020. WeendinFebruary2021becauseitisthelatestwecanmeasurelabormarketexitsfromprepandemic jobs using the CPS. The second category of observations we call pre-pandemic, which include those where the woman is last observed between September 2015 and February 2020, again including only women who can be linked to their 12 month prior observation and who were employed in that earlier observation. As a robustness exercise, we also include observations from theGreatRecession.15 Eachsampleonlyincludesprime-working-age(aged25to54)women. We first estimate equation 1, which predicts whether or not a woman has exited the labor force measured in year-month t based on characteristics measured at her previous interview in t−12. The β coefficients show the impact of our characteristics of interest, Z , on the probability 0 t−12 that a woman has exited the labor force conditional on having been working one year before.16 Our covariates of interest are living with a child aged 0 to 5, living with a child aged 6 to 12, and interactions of living with the two age ranges of children with marital status and weekly earnings again all measured in the t−12 month interview.17 The interactions allow for different effects of children on women’s labor force exits based on marital status and earnings. We control for other characteristics in X. These include a cubic for the age of the woman, industry and occupation COVID-19effectsasdescribedinthedatasection,stateandmonthfixedeffects,raceandethnicity indicators, and educational attainment controls. The specification also includes month and state levelfixedeffects(γ andγ )tofurthercontrolforeffectsthatvaryacrosstimeandgeography. We t j estimatethisspecificationseparatelyforbothourpandemicsampleandourpre-pandemicsample. 14Individualsareonlyincludedintheanalysisiftheirobservationinmonth-yeart canbelinkedtothesameperson 12monthspriorint−12. 15ThisincludeswomenwhosesecondobservationisobservedduringtheNationalBureauofEconomicResearch datedrecessionfromDecember2007toJune2009. Weusetheentireperiodtoprovidemoreprecision. Resultsare similarusingonlythefirstyearwhenweobserveoursampleasbeingemployedbeforetherecession’sonset. 16Becauseweonlyobserveindividualstwice,oneyearapart,exitsarethosethathaveoccurredatanypointduring thepreviousyearandhavepersisteduntilthesecondobservation. 17Wealsoincludethemaineffectsofmaritalstatusandweeklyearningsbothascontrolsandforeaseofinterpretation. 7

Exit =Z β +X γ +γ +γ +ε (1) t t−12 0 t−12 0 t j 1 Next, we estimate effects on labor force exits in excess of historical trends using both the prepandemicand the duringpandemic sampleswith equation2. Thespecification isan interactionof the single time period specification (equation 1) with an indicator for whether the observation is during the pandemic or not (1 ). Our coefficient of interest, β , is on the interaction term pandemic 1 between the pandemic indicator and our coefficient of interest. Its interpretation is the additional effect of each variable on exits during the pandemic in excess of the variable’s effects before the pandemic. Note that the interaction terms apply to the controls as well, so we are controlling for additionalimpactsofothervariablesduringthepandemic–includingvaryinggeographicimpacts ofthepandemiconlaborforceparticipation. Wealsoestimateequation2usingourpandemicand Great Recession samples to examine whether exits during the pandemic differed from a recession moregenerally. Exit =Z 1 β +X 1 γ +γ 1 (2) t t−12 pandemic 1 t−12 pandemic 1 j pandemic +Z γ +X γ +γ +γ +ε t−12 2 t−12 3 j t 2 Finally we use an Oaxaca-Blinder-Fairlie non-linear decomposition to quantify how much observed characteristics can explain Latinas’ and Black women’s higher rates of labor force exit during the COVID-19 pandemic.18 The decomposition consists of two steps. First, we estimate a regression model relating labor force exits to relevant covariates including measures of household composition and information on previous employment. Second, we use the model’s estimated parameters to evaluate the effects that differences in the covariates across racial groups have on the probabilityofexitingthelaborforce.19 Wedothisdecompositionseparatelyforthepre-pandemic sampleandthepandemicsampleforcomparison. The idea behind the decomposition is most easily shown in the classical, linear case. Here the model is simply a linear probability model predicting labor force exits for individuals (i): Exit =Xβ +ε. i i i Exit −Exit =(X¯ −X¯ )β ˆ +X¯ (β ˆ −β ˆ )+X¯ (β ˆ −β ˆ ) (3) j j(cid:48) j j(cid:48) j j j(cid:48) j(cid:48) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) Overall Explained Unexplained 18ThemethodwasintroducedbyKitagawa(1955),predatingitsuseineconomics. 19Fortinetal.(2011)provideanexcellentoverviewofdecompositionmethodsgenerally,includingOaxaca-Blinder decomposition, and Fairlie (2005) provides more details on our specific methodology. A recent example using this techniqueforasimilarquestionisCouchetal.(2020). 8

Equation 3 shows that the overall difference in labor force exits for women of race j compared to women of race or ethnicity j(cid:48), (Exit −Exit ), can be divided into terms that are explained by j j(cid:48) the model and terms that remain unexplained. Hats denote estimated coefficients and bars denote average values in the data. We follow Oaxaca and Ransom (1994) in using the coefficients from a ˆ ˆ pooledmodelwithallwomen(β),asopposedtoonlyWhitewomen(β ),fortheexplainedresult. j(cid:48) The explained portion estimates the difference that would result due to the observed average differences in covariates and the relationships in the data as estimated by the pooled regression. Theunexplainedportionincludeseffectsthataredueeithertodifferencesinrelationshipsbetween covariates and outcomes for women of the specified race or ethnicity (e.g. motherhood and lower earnings leading to a larger number of labor force exits for women in group j) or differences in laborforceexitsthatareunrelatedtocovariates(e.g. unobservedinstitutionalfactors). We follow Fairlie (2005) in using a logit specification (denoted F(Xβ)) to constrain predicted i probabilities to between zero and one. The specification leads to a modification of equation 3 to present the difference in average probabilities of a labor force exit due to differences in covariates inthemodel,asshownbelow. ˆ ˆ Exit −Exit =F(X β)−F(X β) j j(cid:48) ij ij(cid:48) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) Overall Explained (4) ˆ ˆ ˆ ˆ +F(X β )−F(X β)+F(X β)−F(X β ) ij j ij ij(cid:48) ij(cid:48) j(cid:48) (cid:124) (cid:123)(cid:122) (cid:125) Unexplained We present the differences in the average predicted probability of exit from changing the distributionofthecovariatesofinterestfromthevaluesforthereferencegroup j(cid:48)withthoseofourgroup ofinterest j,whilekeepingthedistributionsofallothercovariatesfixed. Sincethedetaileddecompositionintocategoriesofexplanatoryfactors(thoughnottheresultintermsofoverallexplanatory power) is sensitive to the order that variables are introduced into the model, we introduced variables in a random order and averaged the effects over 1,000 iterations for each specification.20 This allows us to decompose the observed differences in exits seen in the data into parts that are explainedoverallandbydifferentobservedcovariatesandapartthatremainsunexplained. CovariatesandSummaryStatistics Table 1 provides information on how exits and covariates differ in our sample across previously employed women with and without children. Around 8 percent of the sample left the labor force 20Notethatthisistheprimaryreasonwepreferthistypeofdecompositionrelativetoanalternativethatsequentially addscovariatestothelinearprobabilitymodelasdescribedintheAppendix. 9

during the pandemic. Women living with children had higher pandemic era labor force exits than those without, mirroring our earlier figures looking at differences relative to the period before the pandemic. Women living with children are also younger on average, more likely to be married, andhaveslightlylowerearningsthanthosewithoutchildren. There also is substantial overlap between our two categories of women with children. Table 1 shows that around 40 percent of women living with a child aged 0 to 5 also live with a child 6 to 12. Around 30 percent of women living with a child aged 6 to 12 also live with a child under 6. Half of the women in our sample have a Bachelor’s degree, around 20 percent live with a child under6,andnearly30percentlivewithachildbetween6and12yearsofage. The Role of Children in Women’s Labor Force Exits MainResults Before the pandemic, living with young children was associated with higher likelihoods of exit among married women and women who earned less as shown in column 1 of table 2. A married womanlivingwithatleastonechildaged0to5,withaverageearnings,was1.9percentagepoints more likely to exit the labor force relative to a married woman without children under 13 in the home who also has average earnings. In contrast, the analogous effect for an otherwise identical unmarriedwomanwastobeastatisticallyinsignificant0.6percentagepointslesslikelytoexitthe laborforce. Exits during the pandemic were much more common among women with young children even after controlling for observed characteristics. As shown in column 2 of table 2, the direct effect of having a young child was 4.1 percentage points, meaning that unmarried women with young children were more likely to exit the labor force than those without. Married women with young childrenalsoexperiencedasimilarincreaseinexitsduringthepandemicrelativetomarriedwomen withoutchildren. Column three of table 2 shows our main finding – women who lived with children under six experienced greater increases in their exit rates during the pandemic relative to observably similar women. Living with a child under age 6 was associated with a 3.5 percentage point increase in excess exits among single women with average earnings relative to similar single women with no children in the household. The effect for married women was very similar at 3.6 percentage points.21 Theestimatedeffectsizesarequitelargerelativetotheoverall2percentagepointdecline in labor force participation during the pandemic. The direct effect of having a child aged 6 to 12, and the interaction between married status and living with a child aged 6 to 12 are statistically 21Theimpliedeffectformarriedwomenisstatisticallysignificantattheonepercentlevel. Howeverthedifference betweenmarriedandunmarriedwomenisnotstatisticallydetectable. 10

NONCONFIDENTIAL // EXTERNAL #Table1: SummaryStatistics None Overall Age 0 to 5 Age 6 to 12 Under 13 Labor force exits 0.07 0.11 0.09 0.06 Less than high school 0.05 0.05 0.07 0.04 High school or GED 0.19 0.20 0.21 0.19 Some college 0.26 0.26 0.26 0.26 Bachelor's degree (only) 0.30 0.28 0.27 0.32 More than a bachelor's degree 0.20 0.22 0.18 0.20 Lived with a child aged 0 to 5 0.21 1.00 0.30 0.00 Lived with a child aged 6 to 12 0.29 0.41 1.00 0.00 Was married 0.57 0.71 0.68 0.49 Black 0.13 0.14 0.14 0.13 Latina 0.17 0.19 0.21 0.16 White 0.59 0.57 0.55 0.61 Age 39.75 35.71 39.57 40.82 (8.54) (6.25) (6.52) (9.42) Previous weekly wage 988 953 938 1010 (659) (668) (655) (654) Occupation employment change -0.04 -0.03 -0.04 -0.04 (0.13) (0.13) (0.13) (0.13) Occupation share working from home 0.36 0.35 0.35 0.36 (0.23) (0.23) (0.23) (0.23) Occupation share unable to work due to COVID-19 0.17 0.17 0.17 0.17 (0.10) (0.10) (0.10) (0.10) Industry employment change -0.05 -0.05 -0.05 -0.05 (0.08) (0.07) (0.07) (0.08) Industry share working from home 0.37 0.38 0.37 0.37 (0.17) (0.16) (0.17) (0.17) Industry share unable to work due to COVID-19 0.17 0.16 0.17 0.17 (0.09) (0.08) (0.08) (0.09) Note:Thistablepresentsthemeanvaluesandstandarddeviations(onlyforcontinuousvariables)ofcovariatesineach ofourcategoriesofraceandethnicity. Theestimationsampleisprime-working-agewomenfromSeptember2020to February2021intheCurrentPopulationSurveywhowereemployedoneyearearlier,whichiswhenthevariablesare measured(besidesage,education,andexits). 11

Table2: EffectsofChildrenonLaborForceExits (1) (2) (3) (4) Excess: Excess: pandemic pandemic Pre- and pre- and Great Variables pandemic Pandemic pandemic Recession Lived with a child aged 0 to 5 0.006 0.041 0.035 0.034 (0.005) (0.017) (0.016) (0.018) Lived with a child aged 6 to 12 0.004 0.018 0.015 0.019 (0.004) (0.018) (0.017) (0.017) Was married 0.011 0.018 0.006 0.012 (0.002) (0.006) (0.006) (0.007) Previous weekly earnings (normalized) -0.010 -0.010 -0.000 0.006 (0.001) (0.005) (0.004) (0.005) Earnings (normalized) by lived with child aged 0 to 5 -0.012 -0.029 -0.017 -0.014 (0.002) (0.006) (0.006) (0.007) Earnings (normalized) by lived with child aged 6 to 12 -0.002 -0.018 -0.016 -0.017 (0.002) (0.005) (0.005) (0.006) Married by lived with child aged 0 to 5 0.013 0.014 0.002 0.001 (0.007) (0.016) (0.017) (0.018) Married by lived with child aged 6 to 12 0.005 -0.009 -0.014 -0.009 (0.004) (0.023) (0.022) (0.022) Observations 86,377 8,787 95,164 45,919 Age cubic X X X X Race and ethnicity indicators X X X X Month fixed effects X X X X State fixed effects X X X X Industry and occupation controls X X X X Education controls X X X X Note:ThepandemicledtomoreexitsamongwomenwithchildrenundersixrelativetobothbeforethepandemicandtheGreatRecession. Lowearningwomenwithchildren6to12werealsomorelikelytoexitduringthepandemic. Shownarecoefficientsfromlinearprobabilitymodels predictinglaborforceexits(columnsoneandtwo)andexcessexitsduringthepandemic(β2inequation2)relativetotheperiodbeforeit(column 3)andtheGreatRecession(column4). Theestimationsamplesforthefirsttwocolumnsareprime-working-agewomenfromSeptember2020to February2021whowereemployedoneyearearlier.ThelasttwoalsoincludewomenobservedfromSeptember2015toFebruary2020(column3) andDecember2007toJune2009.Standarderrorsareclusteredbymonth. 12

insignificant,thoughpotentiallyeconomicallymeaningful. Our results also suggest that the effect on excess exits of living with a child under the age of 13 were larger for lower-earning women. Specifically, we estimate that a woman with earnings one standard deviation below the pre-pandemic average earnings living with a child had a statistically significant 1.6 or 1.7 percentage point larger increase in labor forceexits relative to a woman with the same aged children with average earnings.22 The effect of the interaction between earnings and the presence of children stands in contrast to the small and statistically insignificant coefficient on the direct effect of weekly earnings. The small direct effect of earnings suggests that the mechanisms go beyond factors that affected all low earning women equally, like more generous governmental benefits. One explanation for the higher rates of exit among lower-earning women with school-aged children is a loss of school as an inexpensive mode of childcare. Additionally rates of homeschooling increased during the pandemic, and homeschooling may have been more difficult to combine with work for lower-income women or women who were unable to work remotely(Musaddiqetal.,2021). Additionally, we find no evidence that the pandemic increased labor force exits among married women with young childrenmore than it did among unmarried women. If anything, our estimates suggest that the pandemic led to more excess exits among unmarried women than among married women at least for school aged children. One hypothesis voiced early in the pandemic was that childcare disruptions could lead more women with small children and working husbands to drop outofthelaborforce,inresponsetothegapinmen’sandwomen’swagesandthedemandsoftwo parents working full time.23 However, the negative coefficient on the interaction between being married and living with school-aged children and the zero coefficient on being married and living with children under 5 years old suggests that the pandemic has not resulted in larger increases in laborforceexitsformarriedwomenwithkidsrelativetounmarriedwomen. Our result that women who live with children had excess exits during the pandemic is also true when we use the Great Recession as our comparison. This provides additional evidence that the effect is due to the pandemic induced increases in childcare responsibilities or loss of childcare access rather than an economic downturn more generally. Column 4 of table 2 shows that the pandemic led to a 3.4 percentage point increase in the likelihood that an unmarried woman, with average earnings, living with children under six would exit the labor force. The interaction terms of living with a child and earnings in column four are also of similar magnitude to column three. The takeaway is that the results are not driven by the comparison with the relatively strong labor market before the pandemic’s onset, since they also apply when we compute excess exits relative toarecession. 22Thisincludesthedirecteffectoflowerearningsaswell. 23Alonetal.(2020)andC.D.Goldin(2020)mentionthishypothesis. 13

Beyond variables related to children we find that other factors were not very predictive of additionalexitsduringthepandemic. Appendixtable1showsthatexitsweremorecommonforwomen with less education both before and during the pandemic. Excess exits during the pandemic were alsomonotonicallydecreasingbyeducationalattainment,thoughpointestimatesareoftennotstatistically different from zero. The direct effect of earnings on excess exits is small in magnitude andstatisticallyindistinguishablefromzero. After controlling for education and earnings, we find that occupation and industry measures of the impact of COVID-19 play only a minor role in predicting excess labor force exits. We find small and statistically insignificant associations with the industry and occupational impacts of COVID-19, which suggests that occupation- or industry-specific human capital and adjustment frictions explain little of the increase in women’s pandemic labor force exits, at least for women with similar educational attainment and previous earnings. Interestingly, the occupations where ourconstructedpandemic-eraemploymentdisruptionswerelargestalsoweretheoccupationswith higherpre-pandemicratesofexit(column1ofappendixtable1). Womenwhoworkedinindustries and occupations that had higher shares of workers working form home during the pandemic were lesslikelytoleavethelaborforceevenbeforethepandemicinourspecification.24 Our findings are based on the presence of children in the home, but we interpret the effects as arising from disruptions to families’ childcare arrangements that increased some women’s caregiving responsibilities. The disruptions also likely go beyond formal school and daycare closures since there also were disruptions to informal care networks due to health concerns like grandparents’ concerns about exposure from their grandchildren. Interruptions to informal care could have been particularly difficult because of the unreliability of formal childcare and the possibility that changing work arrangements could lead to temporary gaps in childcare that would be easier to fill informally.25 Another piece of evidence that caring for children led to labor force exits during the pandemic is a rise in the share of women who left the labor force and said that they are out of the labor force for caregiving reasons. As shown in appendix figure 3, women living with children had largerincreasesinthe share ofexitsassociatedwithcaregivingduringthe pandemic. Theshareof women living with kids under age 13 who said they exited the labor force because of caregiving responsibilities increased by between 3 and 4 percentage points during the pandemic compared to lessthanonepercentagepointforwomenwithoutchildreninthehousehold. 24Ofcourse,itisalsopossiblethatoccupationandindustryaremeasuredwitherrorintheCPS. 25Anotherfactorcouldbeconcernsaboutchildren’sexposuretoCOVID-19inchildcaresettings. 14

RobustnessandAlternativeSpecifications In this section, we first examine how our modeling assumptions affect our estimates, and second testsomealternativewaysofapproachingourresearchquestion. Across a number of specifications, we find that the effects of children on labor force exits are qualitatively robust. Column one of table 3 shows our baseline estimates. In column two, we estimate a sparse model that includes only indicators for the presence of children. Without any controls, having a preschool aged child increased labor force exits by 3 percentage points while having a school aged child increased them by 1.7 percentage points. In column three, we add the demographic individual level controls and find that the effect of school aged children on excess exits declines. In column four, we add the interactions between having children and earnings and marital status in addition to the earnings control. The estimate for living with a school aged childchangesappreciablyfromcolumnthreetocolumnfour,highlightingtheheterogeneityinthe effect of children on women with different marital status and earnings. Columns four and one look nearly identical suggesting the limited role state, time, and occupation or industry controls play. In column five, we add month-year fixed effects, and state by pandemic fixed effects with little change to our estimates. In column six, we run our baseline regression without weights and the magnitude of our estimates decline although they remain consistent in sign. Finally in column seven, we control for the number of children under age 13 in the household. Excess exits increase as the number of children increases. The direct effect of having children in each age group is slightly smaller, as some of the effect for each age group is taken into account in the number of childrencoefficient. However,theoverallresultsarequalitativelysimilartothebaselineestimates. Although we focus on excess exits in our main specification, it would also be of interest to test if women remained employed through the first year of the pandemic. We find similar if not slightly stronger results if we focus on employment as the outcome rather than participation. This impliesthatthesamecharacteristicsthatpredictleavingthelaborforcealsoappeartopredictbeing unemployed,sincethisspecificationlumpsbeingoutofthelaborforcewithbeingunemployed.26 Another interesting extension is to see if similar patterns apply to women who were not employed one year previously. So in column nine we include all women and we consequently drop the controls for characteristics of women’s previous jobs. The specification includes both entry and exit effects because women who were not employed could decide to participate.27 The esti- 26Onereasonthisisnotourpreferredspecificationisbecausethemisclassificationofemployedworkerswhowere unabletoworkduringthepandemicasbeingunemployedontemporarylayoffwouldaffecttheseresults,unlikeour mainspecificationforlaborforceexits. Theeffectsofthisphenomenon,however,arelikelytobesomewhatmodest becauseoursampleperiodbeginssufficientlylatethatitexcludestheearlymonthsofthepandemicwhenthisissue wasthemostacute. 27Note that some of the women who were not employed were already in the labor force, since they were unemployed. 15

mates are qualitatively similar, but larger in magnitude suggesting that declines in entry operated similarly to increases in exits although some of the increase may be due to the lack of interaction termsinourmainspecification.28 Next, we test different ways of measuring our covariates of interest in our main specification thatestimatesexcesspandemicexitsrelativetopre-pandemicyears. While being married can signify a greater degree of resource sharing, we also tried focusing on partnered individuals rather than married individuals. In column two of table 4, we show that our results look very similar if instead of marital status we use the presence of a partner living in the household. Additionally, we tried using education level as a proxy for earning potential rather than using earnings directly in column 3. We see these results as qualitatively similar, although the evidence for less-educated workers with children having greater increases in labor force exits is not as strong as for lower-earning women in our baseline specification. While both education and earnings are a function of previous decisions the women have made, educational attainment is a coarser measure. In our main specification, we control for education and focus on previous earnings, which allows forwomen of the same educationlevel to have different earningsbased on unobservedcharacteristicsanddifferentemploymentchoices. Incolumnfour,weusethesamespecification,witheducationinteractiontermsratherthanearnings, to predict non-participation where we do not impose the sample restriction that the women had to be employed 12 months prior. As in table 3, we see that the effects are larger in magnitude suggesting that declines in entry may have affected the same groups that saw increases in labor forceexits. Finally, we run our baseline specification using a logit model rather than a linear probability model. As shown in appendix table 2, the results remain qualitatively similar with the exception thatthedifferenceinexcessexitsbasedonpreviousearningsforwomenlivingwithchildrenunder agesixbecomessmallerandstatisticallyinsignificant. Overall our estimated effects of children on excess exits of previously employed women during the pandemic are qualitatively robust to alternative measures and specifications. When we expand the sample to include women who were not working one-year prior, the results are similar suggesting that the effect of children on labor force exits may have been similar to their effect on decreasinglaborforceentryaswell. 28Includingmovementsfromunemploymenttononparticipationmayalsostrengthenthiseffect. 16

Table3: RobustnessofEffectsofChildrenonLaborForceExits (1) (2) (3) (4) (5) (6) (7) (8) (9) Excess exits: Excess non- Excess non- Variables Baseline Excess exits Excess exits Excess exits Excess exits Excess exits Excess exits employment participation Lived with a child aged 0 to 5 0.035 0.030 0.033 0.035 0.034 0.021 0.028 0.047 0.059 (0.016) (0.009) (0.006) (0.017) (0.017) (0.010) (0.017) (0.020) (0.012) Lived with a child aged 6 to 12 0.015 0.017 0.005 0.016 0.013 0.011 0.007 0.028 0.018 (0.017) (0.007) (0.008) (0.017) (0.017) (0.012) (0.019) (0.020) (0.017) Was married 0.006 0.003 0.007 0.005 0.006 0.006 0.013 0.010 (0.006) (0.006) (0.007) (0.007) (0.006) (0.006) (0.006) (0.009) Previous weekly earnings (normalized) -0.000 -0.000 -0.001 -0.003 -0.000 -0.006 (0.004) (0.003) (0.003) (0.004) (0.004) (0.003) Earnings (normalized) by lived with child aged 0 to 5 -0.017 -0.016 -0.016 -0.013 -0.016 -0.009 (0.006) (0.006) (0.006) (0.004) (0.006) (0.006) Earnings (normalized) by lived with child aged 6 to 12 -0.016 -0.015 -0.016 -0.013 -0.016 -0.022 (0.005) (0.005) (0.005) (0.005) (0.006) (0.007) Married by lived with child aged 0 to 5 0.002 0.002 0.003 0.011 0.001 -0.025 -0.045 (0.017) (0.017) (0.017) (0.012) (0.016) (0.023) (0.013) Married by lived with child aged 6 to 12 -0.014 -0.016 -0.013 -0.009 -0.014 -0.014 -0.006 (0.022) (0.023) (0.022) (0.018) (0.022) (0.025) (0.020) Number of children aged 0 to 12 0.006 (0.004) Observations 95,164 103,283 95,210 95,210 95,210 95,164 95,164 95,164 140,636 Direct effects - not shown X X X X X X X X X Weights X X X X X X X X Race and ethnicity X X X X X X X X Age and education X X X X X X X X Month by year indictors X X X X X X State indicators separately pre and post X X X X X X Occupation and industry effects X X X X Employment 12 months ago X Note:Theeffectofhavingchildrenundersixandofhavingchildrenbetween6and12amongwomenwithlowerearningsisapparentacrossspecifications.Columns1through7predictlaborforceexits amongoursampleofpreviouslyemployedwomen. Samplesizeschangeacrossspecificationsduetomissingdata. Column1isourbaselinespecification. Column2includesonlythemaineffectsof havingchildinthehomewithnoindividuallevelcontrols. Column3includesraceandethnicity,maritalstatus,earnings,age,andeducationcontrols,butnointeractionswiththepresenceofchildren. Column4addstheinteractionterms.Column5addsmonthbyyearindicatorsandstateindicators.Column6isanunweightedversionofthebasespecification,Column7includesacontrolforthenumber ofchildrenundertheageof13livinginthehousehold.Column8isaregressionpredictingnon-employmentasopposedtolaborforceexit.Column9isaspecificationincludingwomenwhowereworking ayearearlier. EachisavariationofthemainspecificationincolumnthreeofTable2presentingβ2termsinequation2). Standarderrorsareclusteredforeachmonth. SeethenotesinTable2formore details“Directeffects”refertothenon-interactedtermswhile“Employment12monthsago”referstotheinclusionofacontrolforemploymentstatus. 17

Aggregateeffects In addition to the direct impacts on women and their careers, labor force exits due to childcare interruptions could have contributed to lower levels of aggregate labor force participation during our sample period. While the impacts of caregiving on employment levels is beyond the scope of our examination, we can use our estimates to calculate the share of exits attributable to having childreninthehouseholdduringtheFallof2020. Thecalculationassumesthatwomenlivingwith childrenwouldotherwisehavehadthesameincreasesinlaborforceexitsassimilarwomenwithout children under 13 in the household. This estimate requires that there are no general equilibrium or “crowding” effects of women with children on women without, and we need to assume that thedifferencesweestimateareduetochildcaredisruptionsandvirusconcernsrelatingtochildren and not unobserved differences between women with kids and those without. Despite the strong assumptions, the exercise still provides a useful means of understanding the size of our estimated effects. Based on our regression estimates, the increase in labor force exits among prime-working-age women would be 0.8 percentage points smaller if women living with children experienced the same increases in exits as those without children under 13 in the home. A 0.8 percentage points smaller increase would roughly halve the 1.5 percentage point increase in excess exits comparing our pandemic sample period of September 2020 to February 2021 to our comparison period of February 2015 to 2020. While our estimates do not suggest that all of the increase in exits among womenduringthepandemicwasrelatedtochildcare,theydosuggestthatchildcareplayedamajor role. Decomposing Differences by Race and Ethnicity In this section, we use an Oaxaca-Blinder-Fairlie (Fairlie, 2005) decomposition to quantify the effects of pandemic induced changes on disparities in labor force exit rates for Latinas and Black women relative to White women. An Oaxaca-Blinder-Fairlie decomposition is valuable in this context for two reasons. First, it allows us to perform a detailed decomposition, as described in Fortin et al. (2011), that shows the effects of specific variables on the level of exits women of different races experienced during the pandemic.29 Second, Oaxaca-Blinder-Fairlie decompositions have been commonly used to study racial and gender gaps in employment outcomes, including 29An alternative approach, which we present in Appendix Table 5, is to sequentially add covariates to our linear specification and examine how the coefficients on race and ethnicity change. Sequentially adding coefficients does notallowustoperformthisdetaileddecompositionwherewecanattributedifferencestospecificvariables,sincethe resultsaresensitivetotheordervariablesareintroduced. 18

Table4: EffectsofChildrenonLaborForceExits (1) (2) (3) (4) Baseline: Excess non- Variables Excess exits Excess exits Excess exits participation Lived with a child aged 0 to 5 0.035 0.037 0.037 0.057 (0.016) (0.022) (0.019) (0.013) Lived with a child aged 6 to 12 0.015 0.018 0.023 0.021 (0.017) (0.016) (0.017) (0.018) Was married 0.006 0.007 0.010 (0.006) (0.005) (0.008) Married by lived with child aged 0 to 5 0.002 -0.008 -0.046 (0.017) (0.011) (0.015) Married by lived with child aged 6 to 12 -0.014 -0.010 -0.005 (0.022) (0.017) (0.021) Previous weekly earnings (normalized) -0.000 -0.000 (0.004) (0.004) Earnings (normalized) by lived with child aged 0 to 5 -0.017 -0.016 (0.006) (0.006) Earnings (normalized) by lived with child aged 6 to 12 -0.016 -0.016 (0.005) (0.005) Had partner 0.008 (0.007) Had partner by lived with child aged 0 to 5 -0.001 (0.024) Had partner by lived with child aged 6 to 12 -0.017 (0.022) Bachelor's degree or more -0.009 -0.012 (0.005) (0.007) Bachelor's or more by lived with child aged 0 to 5 -0.002 0.002 (0.013) (0.012) Bachelor's or more by lived with child aged 6 to 12 -0.011 -0.006 (0.008) (0.008) Observations 95,164 95,164 103,234 140,636 Weights X X X X Race and ethnicity X X X X Direct effects - not shown X X X X Month by year indictors X X X X State indicators separately pre and post X X X X Occupation and industry effects X X X Age cubic term X X X X Education controls X X Employed 12 months prior X Includes previously non-employed X Note:Theeffectsofhavingchildrenundersixandofhavingchildren6to12amongwomenwithlowerearningsaresimilarwhenseparatingout byhavingapartner(includingunmarriedpartnersandthoseofanygender)andofhavingaBachelor’sdegreeormoreasopposedtohavinglow earnings.Resultsarealsosimilarinlookingatnon-participation,includingwomenwhowerenotemployedayearearlier.Presentedareestimated effectsofeachvariableonexcesslabormarketexitsduringthepandemic(β2 inequation2)alongsidestandarderrorsclusteredbyeachmonth. OtheraspectsfollowTable2. 19

duringthepandemic,sotheyincreasethecomparabilityofourresults.30 As shown in figures 3 and 4, a sizeable share of the racial gaps in labor force exits remain unexplained both before and especially during the pandemic. While it’s difficult to attribute the unexplained portion to factors outside of our analysis, other unobserved differences across racial groups, differences in how observed characteristics affect labor force participation, and discrimination all may contribute. Before the pandemic, covariates explained three-quarters of the gap between Latinas and White women’s exit patterns. This share fell to a little over half of Latinas’ six percentage point gap in labor force exits during the pandemic. For Black women, covariates explainroughlyonethirdoftheirhigherlikelihoodofexitcomparedtoWhitewomenbothduring thepandemicandbefore.31 Education, industries, occupations, and earnings describe the largest proportion of the cross sectional differences in exits among women of color relative to White women both during and beforethepandemic. Togethertheyaccountfor80percentoftheexplaineddifferencesforLatinas and around 100 percent of explained differences for Black women. As we show in appendix table 3, Latinas and Black women were more likely to be employed in occupations and industries that wereadverselyeffectedbyCOVID-19. AdditionallyLatinasandBlackwomenhadlesseducation on average and lower earnings relative to White women. Differences in marital status between Black women and White women actually suggest that Black women should have lower rates of laborforceexitbeforeandduringthepandemic. Looking at the differences in contributions during the pandemic relative to the years before, the biggestchangesrelatetotheinteractiontermsbetweenmaritalstatus,earnings,andthepresenceof children shown under “Interactions with children”. Importantly, the effect of these characteristics during the pandemic was to increase exits for women of color relative to White women. Prior to the pandemic they were associated with lower levels of exit for Black women and very slightly higher rates of exit for Latinas. Black women in our sample were less likely to be married than White women, and prior to the pandemic, married women with children were more likely to exit the labor force. Additionally, women who earned less and lived with children had increases in their excess exits, and Latinas and Black women earned less on average relative to White women. Finally children were associated with larger increases in exits during the pandemic and Latinas and Black women are over-represented among women with children relative to their shares in the overall population. Latinas make up 19 percent of women with children under 6 and 21 percent 30Forexample,Zafar(2013)usedthetechniquetostudygenderdifferencesincollegemajorchoice,(Couchetal., 2022) study gender gaps during COVID-19, and Couch et al. (2020) study gaps by race in the first months of the pandemic,beforeouranalysis. 31Inanefforttobetterunderstandtheunexplainedportion,weaugmentourbaselinespecificationtoallowtheeffect ofchildrentodifferbyraceandethnicity,buttheresultsareimprecise. Howeverthepointestimatesarelarge,andwe cannotruleoutlargedifferencesintheeffectsofchildrenbyrace. SeeAppendixTable6. 20

Figure3: DecomposingtheLatina-WhiteExitGap NONCONFIDENTIAL // EXTERNAL # 100% 80% 60% 40% 20% 0% Not Children Interactions Earnings Education Industry Married Age State explained with and children occupation -20% Pre-pandemic Pandemic Difference Note: Observed covariates explained a smaller share of the of the higher rate of labor force exits among Latinas relativetoWhitewomenduringthepandemicrelativetobefore.Additionally,thepresenceofchildrenandinteractions of earnings and marital status with the presence of children were more explanatory during the pandemic, leading to essentiallyalloftheincreasesduringthepandemicthatarepredictedbyvariables.Initialearnings,education,industry, andoccupationhoweverarethemostexplanatoryinbothperiods. Shownistheproportionofdifferencesinexitsby Latinas relative to White women that are not explained by variables and explained by the specified categories of variables according to the decomposition. Shades represent the pre-pandemic decomposition, the pandemic period decomposition,andthedifferences(inlevels)betweenpre-pandemicandpandemicperioddecompositions. 21

Figure4: DecomposingtheBlack-WhiteExitGap NONCONFIDENTIAL // EXTERNAL # 100% 80% 60% 40% 20% 0% Not Children Interactions Earnings Education Industry Married Age State explained with and children occupation -20% Pre-pandemic Pandemic Difference Note: ObservedcovariatesexplainedasmallershareoftheofthehigherrateoflaborforceexitsamongBlackwomen relativetoWhitewomenduringthepandemicrelativetobefore.Additionally,thepresenceofchildrenandinteractions of earnings and marital status with the presence of children were more explanatory during the pandemic, leading to essentiallyalloftheincreasesduringthepandemicthatarepredictedbyvariables.Initialearnings,education,industry, andoccupationhoweverareveryexplanatoryinbothperiods. ShownistheproportionofdifferencesinexitsbyBlack women relative to White women that are not explained by variables and explained by the specified categories of variables according to the decomposition. Shades represent the pre-pandemic decomposition, the pandemic period decomposition,andthedifferences(inlevels)betweenpre-pandemicandpandemicperioddecompositions. 22

of women with children 6 to 12 despite making up only 17 percent of the overall population of women. Black women make up 14 percent of both categories of women with children compared with 13 percent of the population. Notably, the explanatory value of the state of residence falls during the pandemic and can no longer appreciably explain racial differences in exit rates. This maycontributetotheoverallincreaseinthesharethatisunexplainedduringthepandemic. The presence of children and their interaction with earnings and marital status stand out as the largestcontributorstotheexplainedportionoftheincreaseinexitsduringthepandemicrelativeto pre-pandemic levels for women of color relative to White women. These results suggest that the childcare disruptions during the pandemic were either larger for women of color or they were less abletonavigatethemwhileremainingemployed. The decompositions suggest that the higher rates of exit of women living with young children and with relatively low earnings heightened differences in exit rates for Latina and Black women relativetoWhitewomen. Ouranalysisalsosuggeststhatdifferencesineducationandoccupational sorting as well as unobserved factors, like discrimination or unobserved labor supply factors, play meaningful roles in Latinas and Black women’s higher rates of labor force exits both before and duringthepandemic. Conclusion This paper shows that the COVID pandemic led to larger increases in labor force exits among womenlivingwithchildrenandwomenofcolorthroughitsfirstyear. Womenlivingwithchildren under age 6, particularly single women, were more likely to exit the labor force during the pandemicthaninpreviousyears. Thepandemicalsohaddetectablylargereffectsonwomenwhoboth worked at low-earning jobs and were living with children before the pandemic. Finally, increases in exits among women with children contributed to the larger increases in labor force exits among LatinaandBlackwomenduringthepandemic. Examiningvariationinlaborforceexitsbetweendifferentgroupsofwomenprovidesavaluable lensforunderstandingthepandemicandtheimportanceofformalandinformalcaremorebroadly. Ourresultssuggestthatearlierpatternsoflabormarketoutcomesbyrace,ethnicity,education,and pre-pandemicincomeextendedthroughlaterperiodsofthepandemic.32 However,ourfindingthat increasesinlaborforceexitswerelargerforwomenwithchildrenstandsincontrasttotheemphasis that earlier studies placed on occupation and industry level differences. Our finding of larger increases in exits among women with children compared to similar women without children also suggests that the exits were not solely attributable to the effect of government support programs 32EarlierstudiesincludeCouchetal.(2020),Holderetal.(2021),andCortesandForsythe(2023). 23

provided to all women, like expanded unemployment insurance and stimulus payments.33 It is also possible, however, that the effects of children on labor force exits would be smaller in an environment with fewer financial supports for women outside of the labor force, particularly for unmarriedwomen. Our results highlight the importance of formal and informal childcare for women’s labor force participation, and particularly for lower-earning women and women of color. The wide ranging disruptions to childcare caused by the COVID-19 pandemic went beyond the cost shocks to formal childcare explored in some previous work (Morrissey, 2017). So the coincident increase in laborforceexitsamongwomenlivingwithchildrenprovidesadditional,plausiblycausalevidence of the wider role that care and unpaid work plays in the economy (Compton & Pollak, 2014; Himmelweit, 2002). The characteristics of the women who experienced excess exits during the pandemic also mirror previous examinations of the disproportionate effects of childcare costs on labor force participation, with larger effects for single women, women with children under age 6, andlower-earningwomen(Morrissey,2017). Our results are also important for policymakers trying to develop measures to increase labor force participation and address economic disparities. A back of the envelope calculation suggests that around half of the increase in prime-working-age women’s labor force exits during the pandemicwasduetolargerincreasesinexitsattributabletolivingwithchildren. Women’spandemiceraexperienceshighlighttheimportanceofchildcareinstitutionsinsupportingemploymentforall women,butparticularlyforlowerearningwomenandwomenofcolor. References Albanesi, S., & Kim, J. (2021). Effects of the covid-19 recession on the us labor market: Occupation,family,andgender.JournalofEconomicPerspectives,35(3),3–24.https://doi.org/10. 1257/jep.35.3.3 Alon, T., Doepke, M., Olmstead-Rumsey, J., & Tertilt, M. (2020). The impact of covid-19 on gender equality (Working Paper No. 26947). National Bureau of Economic Research. https: //doi.org/10.3386/w26947 Augustine, J. M., & Prickett, K. (2022). Gender Disparities in Increased Parenting Time During the COVID-19 Pandemic: A Research Note. Demography, 59(4), 1233–1247. https://doi. org/10.1215/00703370-10113125 Blau,F.D.,&Kahn,L.M.(2017).Thegenderwagegap:Extent,trends,andexplanations.Journal ofEconomicLiterature,55(3),789–865.https://doi.org/10.1257/jel.20160995 Compton, J., & Pollak, R. (2014). Family proximity, childcare, and women’s labor force attachment.JournalofUrbanEconomics,79(100),72–90.https://doi.org/10.1016/j.jue.2013.03. 007 33Itisalsopossiblethatwomenwithchildrenweremoreresponsivetochangesinthoseprograms. 24

Cortes, G. M., & Forsythe, E. (2023). Heterogeneous labor market impacts of the covid-19 pandemic.ILRReview,76(1),30–55.https://doi.org/10.1177/00197939221076856 Costoya, V., Echeverr´ıa, L., Edo, M., Rocha, A., & Thailinger, A. (2022). Gender gaps within couples: Evidence of time re-allocations during covid-19 in argentina. Journal of Family andEconomicIssues,43,213–226.https://doi.org/10.1007/s10834-021-09770-8 Couch, K. A., Fairlie, R. W., & Xu, H. (2020). Early evidence of the impacts of covid-19 on minority unemployment. Journal of Public Economics, 192. https://doi.org/10.1016/j. jpubeco.2020.104287 Couch,K.A.,Fairlie,R.W.,&Xu,H.(2022).Theevolvingimpactsofthecovid-19pandemicon genderinequalityintheuslabormarket:Thecovidmotherhoodpenalty.EconomicInquiry, 60(2),485–507.https://doi.org/10.1111/ecin.13054 Fairlie, R. W. (2005). An extension of the blinder-oaxaca decomposition technique to logit and probitmodels.Journalofeconomicandsocialmeasurement,30(4),305–316. Flood, S., King, M., Rodgers, R., Ruggles, S., & Warren, J. R. (2020). Integrated public use microdataseries,currentpopulationsurvey:Version7.0.https://doi.org/10.18128/D030.V7.0 Fortin, N., Lemieux, T., & Firpo, S. (2011). Chapter 1 - decomposition methods in economics. In O. Ashenfelter & D. Card (Eds.), Handbook of labor economics (pp. 1–102). Elsevier. https://doi.org/10.1016/S0169-7218(11)00407-2 Furman, J., Kearney, M., & Powell III, W. (2021). How much have childcare challenges slowed the us jobs market recovery? https://www.piie.com/blogs/realtime-economic-issueswatch/how-much-have-childcare-challenges-slowed-us-jobs-market Garcia, K. S., & Cowan, B. (2022). The impact of school and childcare closures on labor market outcomes during the covid-19 pandemic (Working Paper No. 29641). National Bureau of EconomicResearch.https://doi.org/10.3386/w29641 Goldin, C. (2022). Understanding the Economic Impact of COVID-19 on Women. Brookings PapersonEconomicActivity,2022(1),65–139.https://doi.org/10.1353/eca.2022.0019 Goldin, C. D. (2020). Journey across a century of women. NBER Reporter, 2020(3), 1–7. https: //www.nber.org/reporter/2020number3/journey-across-century-women Hansen, B., Sabia, J., & Schaller, J. (2022). Schools, job flexibility, and married women’s labor supply (Working Paper No. 29660). National Bureau of Economic Research. https://doi. org/10.3386/w29660 Heggeness, M. L. (2020). Estimating the immediate impact of the covid-19 shock on parental attachmenttothelabormarketandthedoublebindofmothers.ReviewofEconomicsofthe Household,18(4),1053–1078.https://doi.org/10.1007/s11150-020-09514-x Himmelweit, S. (2002). Making visible the hidden economy: The case for gender-impact analysis of economic policy. Feminist Economics, 8(1), 49–70. https://doi.org/10.1080/ 13545700110104864 Holder, M., Jones, J., & Masterson, T. (2021). The early impact of covid-19 on job losses among black women in the united states. Feminist Economics, 27(1-2), 103–116. https://doi.org/ 10.1080/13545701.2020.1849766 Kitagawa, E. (1955). Components of a difference between two rates. Journal of the American StatisticalAssociation,50(272),1168–1194.https://doi.org/10.2307/2281213 Krolikowski,P.,Zabek,M.,&Coate,P.(2020).Parentalproximityandearningsafterjobdisplacements.LabourEconomics,65(100).https://doi.org/10.1016/j.labeco.2020.101 25

Leigh,I.,Montes,J.,&Smith,C.(2021).Caregivingandcovid-19(tech.rep.).BoardofGovernors oftheFederalReserveSystem.https://doi.org/10.17016/2380-7172.2984 Luengo-Prado, M. J. (2021). Covid-19 and the labor market outcomes for prime-aged women (Current Policy Perspectives Working Paper). Federal Reserve Bank of Boston. https:// EconPapers.repec.org/RePEc:fip:fedbcq:90899 Morrissey, T. W. (2017). Child care and parent labor force participation: A review of the research literature. Review of Economics of the Household, 15(1), 1–24. https://doi.org/10.1007/ s11150-016-9331-3 Musaddiq, T.,Stange, K., Bacher-Hicks,A., & Goodman,J. (2021). ThePandemic’s effect ondemandforpublicschools,homeschooling,andprivateschools.NBERWorkingPaper29262. Oaxaca, R. L., & Ransom, M. R. (1994). On discrimination and the decomposition of wage differentials. Journal of Econometrics, 61(1), 5–21. https://doi.org/10.1016/0304-4076(94) 90074-4 Pitts, M. M. (2021). Where are they now? workers with young children during covid-19 (Policy Hub Working Paper No. 2021-10). Federal Reserve Bank of Atlanta. https://doi.org/10. 29338/ph2021-10 Ponthieux,S.,&Meurs,D.(2015).Genderinequality.Handbookofincomedistribution(pp.981– 1146).Elsevier.https://doi.org/10.1016/B978-0-444-59428-0.00013-8 Russell, L., & Sun, C. (2020). The effect of mandatory child care center closures on women’s labor market outcomes during the covid-19 pandemic. Covid Economics, Vetted and Real- Time Papers, 62(18), 124–54. https://cepr.org/system/files/publication-files/101414covid economics issue 62.pdf#page=129 Zafar, B. (2013). College major choice and the gender gap. Journal of Human Resources, 48(3), 545–595.https://doi.org/10.3368/jhr.48.3.545 26

Online Appendix to Women’s Labor Force Exits during COVID-19: Differences by Motherhood, Race, and Ethnicity Katherine Lim Mike Zabek June 14, 2023 Trends in Exits Among Men In order to compare the trends we see across women to pandemic patterns among men, we show labor force exits among previously employed men broken out by the presence of children in the householdinappendixfigure1andbyraceandethnicityinappendixgigure2. Figure1: MaleLaborForceExits,byPresenceofChildren .06 .05 .04 .03 .02 .01 0 −.01 0202 yraunaJ morf egnahC Aged 0 to 5 Aged 6 to 12 None under age 13 Jul 18 Jan 19 Jul 19 Jan 20 Jul 20 Jan 21 Note: Plotted are three-month moving average changes in labor force exits for prime-working-age workers, by the presence of children aged 0 to 5 and 6 to 12 before the pandemic among workers who were employed one year prior. Each is adjusted for monthly seasonality based on average monthly values from January 2003 to February 2020. Statisticsareweightedusingsamplingweights. DataarefromtheCurrentPopulationSurveydownloadedfrom IPUMSFloodetal.(2020). These figures show that differences in exit rates between men with children and those without aremuchsmallerthanthedifferencesforwomen. Menofcolorhadlargerexitratesduringthepan- 1

Figure2: MaleLaborForceExits,byRaceandEthnicity .05 .04 .03 .02 .01 0 −.01 −.02 −.03 0202 yraunaJ morf egnahC Black Latino White Jan 19 Jul 19 Jan 20 Jul 20 Jan 21 Note: Plotted are three-month moving average changes in labor force exits for prime-working-age workers, by race andethnicityamongworkerswhowereemployedoneyearprior. Eachisadjustedformonthlyseasonalitybasedon averagemonthlyvaluesfromJanuary2003toFebruary2020. Statisticsareweightedusingsamplingweights. Data arefromtheCurrentPopulationSurveydownloadedfromIPUMSFloodetal.(2020). demicrelativetoWhitemen,butthedifferencesdecreasedlaterinthepandemic. Thisconvergence didnotoccurforwomen. 2

Women Outside the Labor Force: Caregiving Trends Thelaborforcepatternsforwomendescribedinthepaperaremirroredbyincreasesintheshareof women not in the labor force stating that caregiving is their primary reason for non-participation. In particular, we find that there were large increases in the share of women who were living with childrenwhoexitedthelaborforceandcontinuedtobeoutofthelaborforcebecauseofcaregiving, asshowninpanelAofappendixfigure3. Womenwhowerelivingwithchildrenunderage6sawa roughlyfourpercentagepointincreasethatpersistedthroughearly2021. Womenwhowereliving with children aged 6 to 12 saw a highly persistent jump of around three percentage points. While it is reassuring that the largest jumps are among women with children, the graph also shows a slight increase of around half a percentage point for other women. This could reflect additional caregivingresponsibilities,forexampleforelderlyrelatives.1 Latinas and Black women also had larger increases in the share of exits that women report as being due to caregiving. Panel B of figure 3 shows a roughly two percentage point increase in the share of Latinas and Black women who exited the labor force and said that they were not in the laborforcebecauseofcaregivingreasons. Thetwopercentagepointincreasesamongthesewomen of color was roughly double the one percentage point increase among White women. This pattern provides additional support for our main finding that the burden of caregiving affected women of color more than it did White women, at least in terms of women’s ability to remain in the labor force. Figure3: PreviouslyEmployedWomenNotintheLaborForce: CaregivingReasons PanelA:PresenceofChildren .05 .04 .03 .02 .01 0 0202 yraunaJ morf egnahC PanelB:RaceandEthnicity .05 .04 .03 .02 .01 0 Aged 0 to 5 Aged 6 to 12 None under age 13 −.01 Jan 19 Jul 19 Jan 20 Jul 20 Jan 21 0202 yraunaJ morf egnahC Black Latina White Jan 19 Jul 19 Jan 20 Jul 20 Jan 21 Note: Plotted are three month moving average changes in the rates of respondents not in the labor force stating caregiving as a reason among prime-working-age women who were employed 12 months ago by the presence of children aged 0 to 5 and aged 6 to 12 before the pandemic and by race and ethnicity. Each is adjusted for monthly seasonality based on average monthly values from January 2003 to February 2020. Statistics are weighted using samplingweights. DataarefromtheCurrentPopulationSurveydownloadedfromIPUMSFloodetal.(2020). 1The plot uses a question asked in the Current Population Survey (CPS) of women who are outside of the labor forceandsaythattheyaretakingcareofhouseorfamilywhenaskediftheywere“disabled,ill,inschool,takingcare ofhouseorfamily,orsomethingelse.” 3

Additional Results Inthissection,wereportadditionalresultsfromourmainspecificationsintable2andthemarginal effectsfromalogitspecificationratherthanalinearprobabilitymodel. Fullresultsofbaselinespecification As briefly discussed in the main text, labor force exits decline as educational attainment increases both before and during the pandemic. Excess pandemic exits exhibit the same pattern both when comparedwithpre-pandemicexitsandGreatRecessionexitsalthoughmanyofthepointestimates are not statistically different from zero. Generally the occupation and industry measures are not predictive of excess exits. They would likely be predictive of employment losses in the initial months of the pandemic, but these results suggest that occupation and industry COVID-19 effects didn’t drive excess exits as measured in September 2020 to February 2021. Interestingly, occupations that were hard hit by employment losses during the pandemic had greater base levels of exitsduringtheyearspriortothepandemicwhileoccupationsandindustriesthatusedmoreworking from home had fewer exits. These correlations likely reflect existing differences above and beyond women’s education and earnings in the occupations and industries that were affected by COVID-19. Agedoesnotappeartobepredictiveofexcessexits,butitwaspredictiveofpre-pandemicexits. Latinas and Black women were more likely than White women to exit the labor force both before and during the pandemic, but when we compare their excess exits we see positive but statistically insignificanteffectsaftercontrollingformanyobservablecovariates. Resultsfromlogitspecification An alternative specification would be to use a non-linear logit model to predict labor force exits. Weshowthemarginaleffectsatthemeanofeachvariableintable2. Theestimatesarequalitatively similar to those from the linear probability model. Women with small children were more likely toleavethelaborforceduringthepandemicthanobservablysimilarwomenwithoutchildren. The wage gradient for women with older children remains while the wage gradient for women with younger children is economically small and statistically insignificant. Women with older children who earned lower wages before the pandemic had larger increases in their exits than women with olderchildrenwhoearnedhigherwages. Additional Decomposition Results Summarystatistics Appendixtable3showssummarystatisticsforoursamplebyraceandethnicity,providingcontext forthedecompositionresults. The first big difference is that Latinas and Black women had higher rates of exits relative to Whitewomenduringthepandemic. Andtheselargerdifferencesarelikelytobeexpectedbecause Latinas and Black women generally were working in occupations and industries that were harder hit by the pandemic than White women. Latinas and Black women additionally had lower levels ofeducationthanWhitewomen. 4

Table1: EffectsofChildrenonLaborForceExits: AdditionalVariables (1) (2) (3) (4) Excess: Excess: pandemic pandemic Pre- and pre- and great Variables pandemic Pandemic pandemic recession Less than High School 0.054 0.077 0.023 0.035 (0.008) (0.017) (0.017) (0.017) High School or GED 0.010 0.016 0.006 0.004 (0.003) (0.003) (0.004) (0.004) Bachelor's degree (only) -0.005 -0.007 -0.001 -0.009 (0.003) (0.004) (0.004) (0.005) More than a bachelor's degree -0.007 -0.015 -0.008 -0.016 (0.003) (0.003) (0.004) (0.005) Occupation employment change -0.043 -0.002 0.041 0.013 (0.010) (0.044) (0.042) (0.045) Occupation share working from home -0.014 -0.026 -0.012 -0.026 (0.004) (0.021) (0.019) (0.020) Occupation share unable to work due to COVID-19 -0.004 0.059 0.062 0.025 (0.016) (0.049) (0.048) (0.051) Industry employment change 0.026 -0.063 -0.089 0.030 (0.035) (0.142) (0.135) (0.142) Industry share working from home -0.017 0.001 0.017 0.006 (0.007) (0.028) (0.027) (0.028) Industry share unable to work due to COVID-19 0.044 0.026 -0.018 0.060 (0.031) (0.139) (0.131) (0.135) Age (normalized) -0.187 -0.348 -0.162 -0.432 (0.068) (0.396) (0.369) (0.391) Age squared (normalized) 0.284 0.652 0.368 0.899 (0.135) (0.796) (0.743) (0.787) Age cubed (normalized) -0.108 -0.307 -0.199 -0.462 (0.069) (0.404) (0.377) (0.400) Latina 0.013 0.024 0.011 0.010 (0.003) (0.011) (0.011) (0.011) Black 0.012 0.027 0.015 0.018 (0.003) (0.008) (0.008) (0.008) Asian 0.025 0.007 -0.018 -0.001 (0.004) (0.013) (0.013) (0.014) Other race/ethnicity 0.010 -0.010 -0.019 -0.014 (0.006) (0.022) (0.021) (0.022) Observations 86,377 8,787 95,164 45,919 Household comosition X X X X Weekly earnings X X X X Interacted household composition and earnings X X X X Month fixed effects X X X X State fixed effects X X X X Note: Differencesbyindustryandoccupationplayedasurprisinglysmallroleinwomen’sexcesslaborforceexitsduringthepandemic. Table showsadditionalcoefficientsthatarenotshowninTable2duetospaceconstraints.SeefootnotesofTable2formoredetails. 5

Table2: EffectsofChildrenonLaborForceExits;LogitSpecification (1) (2) (3) (4) Excess: Excess: pandemic pandemic Pre- and pre- and Great Variables pandemic Pandemic pandemic Recession Lived with a child aged 0 to 5 0.002 0.026 0.020 0.013 (0.005) (0.013) (0.011) (0.014) Lived with a child aged 6 to 12 0.003 0.005 0.001 0.008 (0.004) (0.017) (0.014) (0.014) Was married 0.012 0.019 0.004 0.010 (0.002) (0.005) (0.005) (0.007) Previous weekly earnings (normalized) -0.019 -0.023 -0.000 0.015 (0.002) (0.006) (0.005) (0.007) Earnings (normalized) by lived with child aged 0 to 5 -0.008 -0.015 -0.005 -0.010 (0.004) (0.011) (0.009) (0.012) Earnings (normalized) by lived with child aged 6 to 12 0.001 -0.030 -0.026 -0.023 (0.003) (0.014) (0.011) (0.012) Married by lived with child aged 0 to 5 0.009 0.009 -0.001 -0.001 (0.006) (0.011) (0.011) (0.012) Married by lived with child aged 6 to 12 0.002 -0.013 -0.013 -0.010 (0.004) (0.019) (0.015) (0.016) Observations 86,377 8,787 95,164 45,919 Age cubic X X X X Race and ethnicity indicators X X X X Month fixed effects X X X X State fixed effects X X X X Industry and occupation controls X X X X Education controls X X X X Note:MarginaleffectsfromalogitmodelrunonourmainbaselinespecificationdescribedinTable2.SeefootnotesofTable2formoredetails. 6

Relevant to our results, we also find differences in fertility, marital status, and pre-pandemic earnings. Latinasweremuchmorelikelytohavechildren–particularlysmallchildren–thanwere White women. Black women were also slightly more likely to have children than were White women. Black women were also much less likely to be married than White women or Latinas. LatinasandBlackwomenalsoearnedlessperweekthanWhitewomen. Eachofthesedifferences points to the channels of how childcare interruptions could differentially affect these groups of women. NONCONFIDENTIAL // EXTERNAL Table3: Sum# maryStatisticsbyRaceandEthnicity Overall Latina Black White Labor force exits 0.07 0.12 0.10 0.06 Less than high school 0.05 0.15 0.05 0.01 High school or GED 0.19 0.28 0.25 0.16 Some college 0.26 0.25 0.28 0.26 Bachelor's degree (only) 0.30 0.21 0.24 0.33 More than a bachelor's degree 0.20 0.10 0.18 0.23 Lived with a child aged 0 to 5 0.21 0.23 0.22 0.20 Lived with a child aged 6 to 12 0.29 0.35 0.31 0.27 Was married 0.57 0.57 0.57 0.57 Age 39.75 39.06 39.67 40.06 (8.54) (8.61) (8.54) (8.54) Previous weekly wage 988 778 879 1042 (659) (536) (592) (669) Occupation employment change -0.04 -0.08 -0.05 -0.03 (0.13) (0.14) (0.14) (0.12) Occupation share working from home 0.36 0.29 0.31 0.39 (0.23) (0.23) (0.22) (0.22) Occupation share unable to work due to COVID-19 0.17 0.20 0.18 0.16 (0.10) (0.12) (0.11) (0.09) Industry employment change -0.05 -0.07 -0.05 -0.05 (0.08) (0.08) (0.08) (0.07) Industry share working from home 0.37 0.33 0.37 0.39 (0.17) (0.18) (0.17) (0.16) Industry share unable to work due to COVID-19 0.17 0.18 0.17 0.16 (0.09) (0.10) (0.09) (0.08) Note:Thistablepresentsthemeanvaluesandstandarddeviations(onlyforcontinuousvariables)ofcovariatesineach ofourcategoriesofraceandethnicity. Theestimationsampleisprime-working-agewomenfromSeptember2020to February2021intheCurrentPopulationSurveywhowereemployedoneyearearlier,whichiswhenthevariablesare measured(besidesage,education,andexits). 7

Fulldecompositionresultsasatable Appendix table 4 shows the results from our Oaxaca style decomposition. Reported in the table are the proportions of the difference in exit rates explained by each group of variables. Columns one though three show the results for the Latina-White gap while columns four through six show the results for the Black-White gap. The Before column explains differences in exit rates prior to the pandemic, the During column is for pandemic era exits, and the Difference column shows the differenceinexplanatorypowerforthevariablesbetweenthetwotimeperiods. Factors relating to employment explain a large share of the differences in exit rates before and during the pandemic for both Latinas and Black women. When we look at the differences between the two time periods we see that the household interactions stand out as being much more important in explaining exits during the pandemic than prior to the pandemic. For Black women their lower rates of marriage actually predict lower exit rates so this covariate actually increases the unexplained portion. A woman’s state of residence is predictive of exits before the pandemic but is not predictive during the pandemic. While we include this control for completeness, one could argue that states with higher levels of workers of color may have higher exit rates due to discriminationandthereforethestateitselfisnotagoodcontrol. FinallyalargershareoftheLatina-WhitegapisexplainedthantheBlack-Whitegap. Mechanically, much of the difference in the explained effects is due to the explanatory power of the larger differences in educational attainment and earnings between Latinas and White women. Higher rates of fertility as well as lower earnings (interacted with higher fertility) also make the variables relatingtochildrenmoreexplanatoryofthegapsbetweenLatinasandWhitewomen. Alternativedecompositionusinglinearregression Another conceptually similar approach to understanding the role of covariates in explaining racial and ethnic differences in exits is to break down our baseline regression into steps where we first include the race/ethnicity coefficients alone, and then we add categories of covariates to examine howtheychangethemeandifferencesbetweenrace/ethnicitygroups. Equation1formalizesthisapproachintermsofcoefficientsonanindicatorofwhethersomeone is a Black woman (α ) or a Latina (α ). Changes in α and α show to what extent correlations 1 2 1 2 with controls in the matrix X are able to account for different rates of exits by race and ethnicity. Conceptually, if differences in the covariates themselves explained all of the racial and ethnic differencesinexits,thenthecoefficientsonraceandethnicitywouldgotozero. Exit=α Black+α Latina+βX+ε (1) 1 2 Table 5 presents this approach both in the pre-pandemic period and during the pandemic. When looking across the table at how the coefficients change, we can again see the importance of education, industry and occupation, and earnings in explaining differences in exits both before the pandemic and during between women of color and White women. Looking further across the columns,weseesomeroleformaritalstatus,children,andchildreninteractionvariables. Weincludethisapproachforcompleteness,butweprefertheOaxaca-Blinder-Fairliedecompositionbecausetheexactcontributionofeachcovariateisdifficulttoassessasitheavilydependson the order that the coefficients are added in. In contrast, the main results in Table 4 are averaged so 8

Table4: DecompositionofGapsinExitsDuringthePandemic Latina Black Variable groupings Before During Difference Before During Difference Children 0.001 0.002 0.002 0.000 0.001 0.001 (0.0007) (0.0019) (0.0002) (0.0015) Interactions with children 0.001 0.005 0.004 0.000 0.004 0.004 (0.0009) (0.0025) (0.0005) (0.0021) Earnings 0.007 0.007 0.000 0.003 0.004 0.001 (0.0011) (0.0024) (0.0005) (0.0015) Education 0.012 0.012 0.000 0.003 0.004 0.002 (0.0018) (0.0040) (0.0005) (0.0015) Industry and occupation 0.005 0.006 0.001 0.001 0.003 0.002 (0.0010) (0.0025) (0.0003) (0.0015) Married -0.001 -0.002 -0.001 -0.001 -0.002 -0.001 (0.0003) (0.0011) (0.0007) (0.0025) Age 0.001 0.000 0.000 0.001 0.000 0.000 (0.0003) (0.0004) (0.0002) (0.0004) State 0.003 0.001 -0.002 0.002 0.000 -0.002 (0.0013) (0.0036) (0.0006) (0.0023) Level 0.087 0.117 0.031 0.067 0.096 0.029 Difference 0.039 0.060 0.021 0.019 0.039 0.020 Explained 0.030 0.033 0.003 0.009 0.011 0.002 Observations 86,377 8,787 86,377 8,787 Note:AroundhalfofthedifferencebetweenLatinasandWhitewomen,andaroundaquarterofthedifferencebetweenBlackandWhitewomen,is explainedbycovariatesbothbeforeandafterthepandemic.Educationandjobcharacteristicsdothemosttoexplaindifferences.Theexplanatory poweroftheinteractionsbetweenthepresenceofchildrenandearningsaswellasmaritalstatusincreasedmarkedlyduringthepandemic.Thistable showsadetaileddecompositionofthecontributionsofthevariousgroupingsofvariablesshownineachrowinexplainingthehigherrateofexits amongLatinaandBlackwomenrelativetoWhitewomenbeforeandintheperiodendingfromSeptember2020toFebruary2021,labeledPandemic. Anadditionalcolumnshowsthedifferenceinthecontributionsinthetwoperiods.Thenextgroupofrowsshowthenumberofobservationsusedto estimatethemodel,thelevelofexitsforLatinasorBlackwomen,thedifferencewithWhitewomen,andthedifferencewithWhitewomenthatis explainedbythemodel.Thelastrowgivesthenumberofpotentialexitsusedtoestimatethemodel,includingwomenofallracesandethnicities. 9

thattheordervariablesareintroduceddoesnotmatter.2 2More precisely the results are based on an averaging of 1,000 different random orderings due to the nonlinear formofthespecification. 10

Table5: ExcessExitsSequentiallyAddingCovariatesasDecomposition NONCONFIDENTIAL // EXTERNAL # (1) (2) (3) (4) (5) (6) (7) (8) (9) Pre-pandemic Black 0.0193 0.0133 0.0129 0.0111 0.0138 0.0128 0.0130 0.0107 0.0119 (0.0024) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0026) (0.0026) Latina 0.0393 0.0216 0.0193 0.0178 0.0184 0.0171 0.0170 0.0140 0.0131 (0.0026) (0.0029) (0.0029) (0.0028) (0.0028) (0.0028) (0.0028) (0.0032) (0.0033) Pandemic Black 0.0391 0.0292 0.0270 0.0247 0.0297 0.0274 0.0276 0.0270 0.0271 (0.0089) (0.0075) (0.0079) (0.0080) (0.0086) (0.0081) (0.0080) (0.0083) (0.0081) Latina 0.0600 0.0348 0.0319 0.0293 0.0306 0.0286 0.0278 0.0243 0.0238 (0.0101) (0.0108) (0.0115) (0.0110) (0.0110) (0.0112) (0.0109) (0.0111) (0.0111) Pre-pandmic observations 86,377 86,377 86,377 86,377 86,377 86,377 86,377 86,377 86,377 Pandemic observations 8,787 8,787 8,787 8,787 8,787 8,787 8,787 8,787 8,787 Education X X X X X X X X Industry and occupation effects X X X X X X X Weekly earnings (standardized) X X X X X X Married X X X X X Children X X X X Children interactions X X X State X X Age X Note: Thistablepredictsexcessexitsstartingbycontrollingonlyforraceandethnicity. Eachcolumnaddsadditionalcovariatessequentiallyuntilthefinalcolumnmatchesourbaselinespecificationin Table2.SeefootnotesofTable2formoredetails. 11

Differentialeffectsofchildrenbyrace-ethnicity OurOaxaca-Blinder-Fairliedecompositionshowsthatalargeproportionofthegapinexcessexits between women of color and White women remains unexplained. An underlying assumption of the decomposition is that the effect of the covariates on exits is the same across racial and ethnic groups. In order to test whether the effects of children were larger for Black women and Latinas, we augment our baseline model to include interactions between the presence of children and the race and ethnicity of the woman. Unfortunately we lack precision to estimate these interaction terms definitively. The point estimates suggest that the effect of living with a child under 6 was greateronwomenofcolor’sexcessexits,whilelivingwithachildaged6to12waseitherthesame or perhaps less important. Interpreting these effects; however, requires us to hold constant marital status and earnings, which we know differ by race and have their own effects on excess exits by race. 12

Table6: EffectsofChildrenonLaborForceExitsInteractedwithRace-Ethnicity (1) (2) (3) (4) Excess: Excess: pandemic pandemic Pre- and pre- and Great Variables pandemic Pandemic pandemic Recession Latina 0.009 0.021 0.012 0.002 (0.004) (0.012) (0.012) (0.013) Latina by lived with child aged 0 to 5 0.000 0.033 0.032 0.048 (0.007) (0.028) (0.027) (0.031) Latina by lived with child aged 6 to 12 0.013 -0.014 -0.027 -0.011 (0.007) (0.012) (0.013) (0.018) Lived with a child aged 0 to 5 0.008 0.026 0.018 0.020 (0.006) (0.015) (0.015) (0.017) Lived with a child aged 6 to 12 -0.001 0.026 0.027 0.027 (0.005) (0.019) (0.018) (0.019) Black 0.011 0.022 0.011 0.015 (0.004) (0.009) (0.009) (0.010) Black by lived with child aged 0 to 5 -0.007 0.029 0.037 0.019 (0.010) (0.019) (0.020) (0.023) Black by lived with child aged 6 to 12 0.009 -0.006 -0.014 -0.005 (0.008) (0.023) (0.023) (0.025) Earnings (normalized) by lived with child aged 0 to 5 -0.013 -0.029 -0.016 -0.013 (0.002) (0.006) (0.006) (0.007) Earnings (normalized) by lived with child aged 6 to 12 -0.001 -0.018 -0.017 -0.017 (0.002) (0.006) (0.005) (0.006) Married by lived with child aged 0 to 5 0.011 0.017 0.007 0.003 (0.007) (0.016) (0.016) (0.017) Married by lived with child aged 6 to 12 0.007 -0.011 -0.018 -0.011 (0.004) (0.022) (0.021) (0.021) Previous weekly earnings (normalized) -0.010 -0.011 -0.000 0.006 (0.001) (0.005) (0.004) (0.005) Was married 0.011 0.017 0.006 0.012 (0.002) (0.005) (0.005) (0.007) Observations 86,377 8,787 95,164 45,919 Standard controls X X X X Note: TableshowsaugmentedspecificationfromTable2thataddsinteractiontermsbetweenrace-ethnicindicatorsandthepresenceofchildren indicators.SeefootnotesofTable2formoredetails. 13