Are Manufacturing Jobs Still Good Jobs? An Exploration of the Manufacturing Wage Premium

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Are Manufacturing Jobs Still Good Jobs? An Exploration of the Manufacturing Wage Premium Kimberly Bayard, Tomaz Cajner, Vivi Gregorich, and Maria D. Tito 2022-011 Please cite this paper as: Bayard, Kimberly, Tomaz Cajner, Vivi Gregorich, and Maria D. Tito (2024). “Are Manufacturing Jobs Still Good Jobs? An Exploration of the Manufacturing Wage Premium,” Finance and Economics Discussion Series 2022-011r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2022.011r1. 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.

Are Manufacturing Jobs Still “Good” Jobs? An Exploration of the Manufacturing Wage Premium* Kimberly Bayard† Tomaz Cajner‡ Genevieve Gregorich§ Maria D. Tito¶ September 3, 2024 Abstract Thispaperexploresthefactorsbehindthedisappearanceofthemanufacturingwagepremium—theadditionalpayamanufacturingworkerearnsrelativetoacomparablenonmanufacturingworker. Withsubstantiallylargerdeclinesacrossunionmembers,wequantifytheroleofunionizationbyexploiting the heterogeneity in membership status across manufacturing industries. Wefindthatthedeclineinunionmembershipexplainsmorethan70percentof thedeclineinthewagepremiumsincethe1990sforunionmembersbutdoes not affect nonunion premia. Our findings suggest that the erosion of “good” manufacturingjobshascontributedtotheincreaseinoverallwageinequality. Keywords: WageInequality,WagePremia,Manufacturing,ProductionWorkers,UnionMembership. JELclassification: E24,J31,J51. *WewouldliketothankJoelElvery,LeiFang,andparticipantsoftheWISER2021conference. Theviewspresentedinthispaperrepresentthoseoftheauthorsanddonotnecessarilycoincide withthoseoftheFederalReserveBoardortheFederalReserveSystem. †FederalReserveBoard ‡FederalReserveBoard §ColumbiaBusinessSchool ¶FederalReserveBoard. Correspondingauthor: 20thStreetandConstitutionAve. NW,Washington,DC,20551. E-mail: maria.d.tito@frb.gov.

Thereisalong-standingconventionalwisdomthatmanufacturingjobsare“good” jobs.1 These perceptions reflect the historical role of the manufacturing sector in offering higher wages and benefits than elsewhere in the private sector. In contrasttotheseperceptions,themanufacturingwagepremium—theadditionalpaya manufacturingworkerearnsrelativetoacomparablenonmanufacturingworker— disappeared several years ago, and manufacturing wages currently rank in the bottom half of the wage distribution across all industries in the United States. Though the hourly earnings of manufacturing workers rose 2.3 percent per year, on average, between 2006 and 2019, overall private sector earnings increased 2.6 percent per year over the same period. After the levels of the two series converged in April 2018, manufacturing wages remained below those in the private sector.2 The erosion of the manufacturing wage premium has primarily affected workersemployedinproductionoccupations,asalsorecentlynotedbyElveryand Dunn [2021]: we find that the wage premium in production occupations—which account for about 70 percent of all manufacturing employment—disappeared in 2006, while nonproduction workers in manufacturing have typically suffered a wagediscountwhichfurtherincreasedonlyinveryrecentyears. This paper explores the factors that contributed to the disappearance of the manufacturing wage premium. First, we rely on worker-level data from the Current Population Survey (CPS) to address the role of worker observable characteristics and the decline in unionization rates on the wage premium. We find that changes in the demographic characteristics of the manufacturing workforce over time have been broadly similar to changes in the demographic characteristics in 1Accordingto2017pollconductedbyDeloittefortheManufacturingInstitute,Americans perceivedmanufacturingjobsas“goodjobs.” SeeRuckelshausandLeberstein[2014],Giffietal. [2017],andLangdonandLehrman[2012]. 2Recentanecdotessuggestthat,withcontinueddifferentialgainsinwagegrowth,evenrelativelylowerpayingjobsintheleisureandhospitalitysectorhavebeenofferingcompetitive wagesandhavebeenabletoattractworkerspreviouslyemployedinfactories. See,forexample, HuffordandNaughton[2021]. 1

other sectors and do not materially affect the trend in the manufacturing wage premium. After controlling for demographic characteristics, we estimate that the manufacturing wage premium declined 2.5 percentage points (pp) between the 1990s and the 2010s, which is similar to our raw estimate without those controls. Next, we explore the role of unionization rates, which have declined dramatically over the past few decades across most sectors, but the decline has been much more pronounced for manufacturing. In our analysis, we decompose the residual premia—thatis,thewagepremiumthatexistsaftercontrollingforworkerobservable characteristics—across unionized and non-unionized workers. We find that, althoughpremiadeclinedsignificantlyinbothgroups,thedeclinewasmuchlarger across unionized workers: the wage premium for unionized workers in manufacturing moved down 5.5 pp relative to unionized workers in other sectors, while wages for non-unionized manufacturing workers declined only 2.5 pp relative to similar groups outside of manufacturing. Because these estimates could still be affectedbychangesinsectoralcompositionandunobservableworkercharacteristics, we also look at the relationship between changes in the wage premium and changesinunionmembershipswithinmanufacturing. Overall,changesinpremia associated with workers joining or leaving a union account for about 70 percent of the decline in the premium. These effects, however, are identified from a very smallnumberofobservationsandtendtobemarginallysignificant. In the second main empirical exercise, we look within the manufacturing sector and explore the variation in premia across different manufacturing industries toquantifytheroleofunionizationconditionalonotherindustrycharacteristics— such as productivity, trade exposure, firm size and age distributions, and capital intensity. This is our core empirical strategy, and it relies on lagged union membership rates to address concerns of endogeneity and wage-stickiness. We find that the decline in unionization rates is the most important factor for changes in 2

wage premia; in particular, a one standard deviation (sd) decline in unionization rates—which corresponds to a decrease of 13 pp—is associated with a decline in thewagepremiumof1.4pp. Ouranalysisisalsorobusttotheimpactofotherfactors, such as the increasing adoption of temporary help services, the presence of exporters, and the imputations in the CPS sample. To put our quantification exercise in historical context, the decline in unionization rates since the 1990s explains more than 70 percent of the decline in premia across industries within manufacturing. In addition, looking at the effect on wage premia separately for unionized and non-unionized workers, we find that changes in unionization rates tend to be positively associated with changes in premia for worker groups, as in the literature pioneered by Freeman and Medoff [1981], but the effect is significant only for unionizedworkers. Fornon-unionmembers,capitalintensityappearstodrivethe variationin(residual)wagedifferences. Finally,ourpaperconnectsthetrendsinmanufacturingwagestothedispersion in wages across sectors and occupations. Borrowing the methodology on decomposing wage dispersion from Davis and Haltiwanger [1991], we find that the declinesinmanufacturingwageshavecontributedtotheincreaseinwageinequality between sectors and occupations and, ultimately, have affected overall wage inequality. Furthermore, while the decline in relative wages could lower labor costs all else equal, manufacturing firms have been also facing increasing competition from other—particularly low-wage—sectors in attracting and retaining workers. Thus, with a limited pool of available talent, the decline in the wage premium could presage an additional channel for the secular decline of the manufacturing sector,complementarytowhathasbeenproposedbyGould[2019]. This paper builds on the evidence on wage differentials across manufacturing and other industries documented by Harris and McCall [2019], Levinson [2019], and Elvery and Dunn [2021]. Our work contributes to the literature on wage 3

inequality—specificallytotheworkthatreliesontheidentificationofworker-and firm-levelcharacteristicsinthespiritofAbowdetal.[1999]—andontheroleofdeunionization.3 AsintheworkbyLangdonandLehrman[2012]andMishel[2018], we first investigate the role of worker observables on changes in the manufacturing premium. While we also lack firm-level data, we improve upon those papers as we leverage sector-level controls to capture the impact of the average firm; as a result, our estimates are robust to the variation in worker wages due to sectoral characteristics. 1 Data Our analysis relies on the Bureau of Labor Statistics’ (BLS) Current Population Survey (CPS) basic monthly data, a household survey with worker-level characteristics that allows us to disentangle the impact of demographic trends and other observablesfromotherfactorsindrivingthetrajectoryofwagesacrosssectorsover time. Indeed,werelyon(log)hourlywagesasthemaindependentvariableofour analysis,whichweconstructfollowingtheCEPRmethodology.4 Werestrictoursampletoworkersemployedinthebusinesssector(NAICS11– 81); while we do not apply further restrictions, our analysis controls for full-time status—that is, for workers who are employed for at least 35 hours per week— and age, two important determinants of hourly wages. Furthermore, as industry andoccupationclassificationshavebeenrevisedovertime,weconstructconsistent industry and occupation codes. We primarily follow Pollard [2019] to build our concordance.5 3Foramorecomprehensiveliteraturereviewontrendsinwageinequality,seeKatzandAutor [1999]andGoldinandKatz[2001]. Fortheroleofde-unionization,seeFreeman[1992],DiNardo etal.[1996],Card[1996],andFortinandLemieux[1997] 4Detailsonthemethodologyareavailableathttps://ceprdata.org/cps-uniform-dataextracts/cps-outgoing-rotation-group/. 5ForthemappingofmanufacturingindustriesbetweenCPSandNAICScodes,weintroduce 4

Our analysis of long-term trends in hourly wages covers the period between 1983 and 2019, while our main empirical strategy, presented in section 3, covers theyearsfrom1990to2019.6 Finally, we match the CPS industry aggregates to different (3-digit NAICS) industry-leveldatasources—BLSdataonlaborproductivity,FederalReserveBoard dataoncapitalexpenditures,andtheCensusBureau’sinternationaltradedataand QuarterlyWorkforceIndicators—toaccountfortheeffectofindustry-levelcharacteristicsonwages. 2 The Manufacturing Wage Premium: Descriptive Analysis Drawing on the CPS data, we construct the percent difference in average hourly wages between workers in the manufacturing sector vs. workers in the rest of the economy—the so-called manufacturing wage premium—shown in figure 1. Specifically, weconstructthe manufacturingwage premiumusing thecoefficients onthe interaction between the manufacturing dummy and the year dummy. As shown in figure 1, manufacturing average hourly wages for all employees were about 14 percent above wages in the overall private sector in the mid-1980s, but the differencegraduallydeclinedovertime,reaching7percentin2019,thuspointingtothe threemainmodifications. First,wesplittheElectricalProductManufacturingaggregate(NAICS 334-335)intotwoseparateindustries,ComputersandElectronicProductsManufacturing(NAICS 334)andElectricalEquipment,Appliance,andComponentManufacturing(NAICS335),usingthe post-2002averageemploymentsharesacrossthose3-digitNAICSindustries. Second,wegroup PrintingandRelatedSupportActivities(NAICS323),whichisclassifiedunderNAICS511inthe Pollard[2019]classification,withPaperManufacturing(NAICS322). Third,weexcludefromthe MiscellaneousandnotSpecifiedManufacturinggrouping(NAICS339,31-33)thoseindustrieswith unspecifiedmanufacturingNAICScodes(NAICS31-33)toseparatelyidentifyMiscellaneousManufacturing,(NAICS339). 6WhileBLSdataareavailablefrom1979throughmid-2023,weexcludeearlieryears,when individualandfirm-levelinformationaremorelimited,andthemostrecentyears,whenfluctuationsinwagesandworkerflowsmightbeheavilyinfluencedbytheCOVID-19pandemic. 5

erosionofthemanufacturingpremium,consistentwiththeaggregatetrends.7 Figure1: ManufacturingWagePremium Figures 2 and 3 show, respectively, the manufacturing wage premia for both production and nonproduction occupations in the CPS data. The premia estimates shown in the charts are based on the coefficients of interactions between time indicators and a dummy for the manufacturing sector. To easily identify long-time trends, our baseline estimates look at changes in average premia over four decades, from the 1980s to the 2010s.8 The manufacturing premium for nonproduction workers has been little changed in recent decades, fluctuating around 25 percent. Production workers employed in the manufacturing sector, instead, have experienced a significant decline in wages relative to similar occupations in other sectors; since the 1990s, the manufacturing wage premium for production workers has declined 2.5 pp to 11/4 percent in the 2010s, a point estimate that has remained significantly different from zero. Thus, the CPS data show that the manufacturing premium across production occupations has declined, but it has not yet disappeared.9 7TheaggregatetrendsdescribedintheintroductionarecharacterizedinmoredetailsinsectionB.1. 8SeefiguresB4andB5foryearlyestimatesofthepremium. 9Lookingatsectorsmoreconnectedwithmanufacturing—specifically,Construction(NAICS 6

Differencesbetweenmanufacturingworkersandworkersinothersectorsalong otherdimensions—suchasweeklywagesandbenefits—havealsobeennarrowing inrecentyears.10 Figure2: ManufacturingWagePre- Figure3: ManufacturingWagePremium,ProductionOccupations mium,NonproductionOccupations 2.1 The Role of Worker Observables This section explores the role of worker observables on wage differences between manufacturing and other sectors. First, we control for demographic trends and other observable characteristics; then, we explore in more details how trends in unionizationrateshaveshapedthemanufacturingpremium. DemographicTrendsandOtherObservables The presence of a manufacturing wage premium appears even more remarkable after considering that manufacturing has historically employed worker groups 23),WholesaleTrade(NAICS42),RetailTrade(NAICS44-45),Professional,Scientific,andTechnical Services(NAICS54),AdministrativeandSupportandWasteManagementandRemediationServices (NAICS56),andAccommodationandFoodServices(NAICS72),whichaltogetherrepresentabout60 percentoftheworkerinflowsandabout50percentoftheworkeroutflowsforthemanufacturing sector—wecontinuetofinddeclinesinthewagepremium,butouranalysisrevealsdifferences inlevelsandtimingpatterns. Infact,asshowninfigureB6,thewagepremiumofmanufacturing productionworkersishigherrelativetomoreconnectedsectors,butstarteddecliningsincethe 1980sandpartiallyrecoveredinthe2010s. Thedeclineinthepremiumissignificantaftercontrollingfordemographiccharacteristicsandotherworkerobservables,asshownbyredbars. 10Theonlineappendixexamines,indetails,thepatternsforweeklywagesandbenefits;these data,however,areavailableonlyattheaggregatelevel,andweareunabletopursueadetailed analysissimilartowhatwedowithhourlywages. 7

that typically receive lower wages, such as less educated workers. In particular, in the 1980s, the average manufacturing production worker was white, male, between the ages of 25 and 34, and had a high school diploma.11 All told, the demographiccharacteristicsofproductionworkeremploymentexhibitonlysmalldifferences in comparison with other sectors, with manufacturing production workers slightlymorelikelytobeolder,lesseducated,andlesslikelytobemale. Sincetheearly1990s,demographictrendswithinmanufacturinghaveledtoan increaseinthesharesofworkerswhoareaged35orolder,andwhoaremale;over thesametime,therehasbeenacorrespondingdeclineintheshareofworkerswho arehighschoolgraduatesandwhoarewhite. Changesindemographicsformanufacturinghavebeenverysimilartochangeselsewhereintheeconomy;themodest differences in demographic trends—with a relatively larger increase in the share of male employment and of older workers and a relatively larger decrease in the shareofhighschoolgraduates—actuallysuggestthatthemanufacturingpremium should have increased since the 1990s, a prediction that is not consistent with our estimates. Figure 4 compares the raw production worker premium with the residual premium, an estimate that controls for demographic characteristics—such as, race and etnicity, sex, age, and education—as well as several other observables— thatis,unionmembership,maritalstatus,tenure,metropolitanarea,andstate-year dummies,whichabsorballthevariationatthestate-yearlevel,suchasincomeand migrationflows. Twomainfindingsemergefromouranalysis. First,controllingfordemographics,wefindthatresidualpremiainthe1980sweresignificantlyhigherthanrawestimates,consistentwiththedisproportionatelyhigheremploymentoflower-wage workers in manufacturing at that time. Indeed, we find a premium of 3.5 percent after controlling for demographics, 1.5 pp higher than the raw estimate. Second, 11SeefiguresB10-B13intheonlineappendix. 8

controlling for demographics does not affect the magnitude of the overall decline in the manufacturing wage premium; adjusting for level differences, the residual manufacturingpremiumstilldeclined2.5pp,fromapeakof4percentinthe1990s to 11/2 percent in the 2010s. Thus, trends in demographics and other worker observablescannotaccountforthedeclineofthemanufacturingwagepremium. Figure4: ManufacturingWagePremium,ProductionOccupations: Comparing PremiawithandwithoutWorkerObservables UnionizationandthePremium Sofar,ouranalysishasincludedinformationonunionmembershipamongworker observablesandhasidentifiedaveragedifferencesinhourlywagesbetweenmanufacturing and other sectors controlling for union memberships status as well as other observable characteristics. However, union membership deserves a more detailed analysis, as it has undergone significant changes over time and has been typically associated with large differences in wages across workers. In fact, in manufacturing, union membership rates dropped almost 20 pp since the mid- 1980s, a significantly larger decline compared to what happened in the rest of the 9

economy—asshowninfigure5. Inaddition,whileunionizedproductionworkers continuetoenjoyhigherwagesthannon-unionizedworkers,therawwagepremia ofthe unionized productiongroup—thatis,without controllingforobservables— summarized in figure 6, also experienced significant declines, moving down from 17percentinthe1980sto12percentinthe2010s. Figure5: UnionizationRates,Produc- Figure6: UnionPremium,ManufactionWorkers turing,ProductionOccupations The timing of those declines and the relative movements in unionization rates seems consistent with the evolution of the manufacturing wage premium. To quantifytheimplicationsofthechangesintheunionpremiumonthemanufacturing wage premium, we rely on a model à la Mincer [1974] that directly compares wages by union status between manufacturing and other sectors across the four decadesofouranalysis, j ∈ {1980s,1990s,2000s,2010s}, lnw = α +α Mfg +α Union +α Mfg ·Union +γX +(cid:101) (1) it 0 1,j it 2,j it 3,j it it it it Inourmodel, α representsthepercentdifferenceinwagesbetweenmanufactur- 1,j ing non-union workers relative to non-union workers in other sectors in decade j (Mfg Nonunion Premium), α denotes the union premium outside of manufac- 2,j turing (Non-Mfg Union Premium), while α + α + α identifies the percent 1,j 2,j 3,j difference in wages between manufacturing union workers relative to the same 10

Table1: DecomposingtheResidualAverageManufacturingWagePremium MfgNonunion Non-MfgUnion MfgUnion MfgUnion Non-MfgUnion Res. Avg Premium Premium Premium Share Share MfgPremium 1980s 6.2% 30.3% 23.6% 34% 21% 4.3% 1990s 5.7% 24.7% 21.0% 25% 16% 6.5% 2000s 3.2% 18.4% 15.1% 18% 12% 3.1% 2010s 2.4% 18.2% 14.4% 14% 11% 2.1% ∆ -3.3pp -6.5pp -6.6pp -11pp -5pp -4.4pp 2010s,1990s Source: CPS. Notes: Wagedifferencesareestimatedfromworker-levelregressionsthatcontrolforagevariables,sex,marital status,education,race,tenure,metro-area,andstate-yeardummies,1983-2019. comparisongroupindecadej(MfgUnionPremium)—controllingfordemographicsandotherworkerobservables, X ,asinthepreviouspartofouranalysis. it Usingtheseestimates,theresidualmanufacturingwagepremium—thatis,the premium that conditions on worker observables—can be proxied by a weighted averageoftheresidualpremiaacrossunionizedandnon-unionizedworkers, Res. Avg. MfgPremium =α +α (MfgUnionShare −Non-MfgUnionShare ) j 1,j 2,j it it +α MfgUnionShare 3,j it Table 1 summarizes the estimates from our decomposition. Two main findings emerge from this analysis. First, the decline in relative wages across union workershasbeentwofoldthatacrossnon-unionizedmembers.12 Second,thedeclinein theshareofunionizedworkersinmanufacturing—whichdropped11ppsincethe 1990s—haspartlyoffsetthechangeinthemanufacturingunionpremium: Ifmanufacturing union shares had remained at the same average level as in the 1990s, averagepremiainthe2010swouldhavebeen1.4percent,translatingintoanadditionaldeclineintheresidualpremiumof0.7pp. Our premia estimates, however, include the effect of sectoral composition and 12Thefasterdeclineofrelativeunionwagescomparedtorelativenonunionwagesinmanufacturingisconsistentwiththeevidencedocumentedby,amongothers,MacphersonandHirsch [2021],whichfindadeclineintheratioofunionwagestononunionwagesforthewholeeconomyandwithinmanufacturing. 11

other time-invariant worker unobservable characteristics. To isolate the impact of changes in union memberships from those factors, we separately looked at flows ofmanufacturingworkersintooroutofunionsbyusinglongitudinallylinkedobservations in the CPS microdata; the results from this analysis are shown in figure 7. Overall,theflowsassociatedwithchangesinunionmembershiparelinkedtoa significantdeclineinthemanufacturingpremiumofabout4ppbetweenthe1990s and the 2010s. This decline combines two effects. First, workers joining unions in the 2010s tend to have lower wage premia than in the previous decades. Second, workers leaving unions in the 2010s seem not to suffer much of a wage discount; in contrast, workers who left unions in the 1990s were offered (significantly, although marginally so) lower wages.13 This analysis, however, relies on very few observations and, thus, the associated estimates tend to be estimated with very largestandarderrors. Figure7: ChangesinWagePremiaofUnionizedWorkersinManufacturing,ProductionOccupations The drop in unionization rates and the wage outcomes observed after workers 13FigureB18showsadecompositionofthe“inflow”and“outflow”effects. 12

join or leave a union point to a decline in the bargaining power of manufacturing workersastheunderlyingcauseoftrendsinwages. However,withoutdirectmeasures on bargaining power, our analysis focuses on a quantifiable dimension, the patternofunionizationrates. 3 Empirical Strategy Our analyses so far have relied on the variation across individuals and over time to identify differences in wages between manufacturing production workers and production workers in other sectors of the economy. We view that part of the investigation,however,asmostlydescriptivesinceitlackscontrolsonfirmorsector characteristics, which the literature has shown to be important determinants of wage premia. Furthermore, an economy-wide analysis of the patterns of wages would not be an ideal setting for further investigations since sector premia are calculated relative to aggregate wages; in fact, while manufacturing wages have declinedrelativetotherestoftheeconomy,wagegainsinsomeothersectorshave been rising faster than in the aggregate.14 To make progress on the factors behind the secular decline of manufacturing wages and to be able to control for various proxies of firm-level characteristics, we will look at industries within manufacturing, leverage differences in wage premia across those industries, and relate those trendstochangesinunionmembership. Sincetherelativedeclineinmanufacturingwagesstartedinthe1990s,thissectionfocusesontheperiodfromthe1990stothe2010s,aperiodalsocoveredbyall themaincontrolsincludedinouranalysis. While manufacturing production workers have enjoyed, on average, a 4 percent wage premium since the 1990s, there have been large differences in premia 14SeesectionB.4intheonlineappendixfortheaggregateresults. 13

across manufacturing industries. In the 1990s, average wage differences ranged from a 25 percent premium for the Petroleum and Coal Products Manufacturing industry (NAICS 324) to a 15 percent penalty for Textile, Apparel, and Leather Manufacturing (NAICS 313-316). By the 2010s, the range of the average premia had contracted, with wages declining in industries that enjoyed higher premia and increasing in industries that suffered wage discounts. The negative relationship between initial wage differences and subsequent changes, shown in figure 8, points toconvergenceinpremiaacrossindustries.15 Figure8: ConvergenceinWagePremiaacrossManufacturingIndustries The differential trends across industries have been positively correlated with the magnitude of the decline in union memberships. Although unionization rates declined across all manufacturing industries, figure 9 shows that those industries thatexperiencedthelargestdeclinesinunionmembershipsbetweenthe1990sand the 2010s—such as, Beverage and Tobacco Product Manufacturing (NAICS 312) and Transportation Equipment Manufacturing (NAICS 336)—also displayed the largest 15Convergencetrendsarequalitativelysimilarifcomparingthe2010stothe1980s. 14

Figure9: WagePremiaandUnionizationRatesacrossManufacturingIndustries declinesinwagesrelativetotherestoftheeconomy.16 Thelargeheterogeneityinpremiaandunionizationrateswithinmanufacturing industries and reduced-form evidence on long-term trends suggest that those industriesprovideanaturalsettingtoextendourinvestigation. Thus,ourmainempiricalstrategywillexploitthevariationinthedeclinesinwagesandunionization rates across manufacturing industries. In particular, our baseline equation relates the industry-level residual wage premium—that is, the estimate of wage differences relative to sectors outside of manufacturing that controls for demographic characteristics and other available worker observables—to the unionization rates, conditionalonavarietyofotherfactors, Res. Premium = β +β UnionRate +γX +d +d +ε (2) st 0 1 s,t−1 st s t st where s indicates a 3-digit NAICS industry within manufacturing, Res. Premium denotes the residual wage premium of an industry, and UnionRate represent the 16Aswiththepremiaconvergence,therelationshipbetweenchangesinpremiaandchangesin unionizationratesisqualitativelysimilariflookingrelativetothe1980s. 15

unionization rate of production workers within the same industry.17 β is our co- 1 efficient of interest: We exploit within-industry variation in union membership to identify average changes in wages across manufacturing industries relative to other sectors. To address concerns of endogeneity as well as possible time lags between changes in union membership status and effects on wages, we rely on a 1-yearlaggedunionmembershipratesasourmainregressor.18 In addition, our specification includes industry fixed effects, time dummies, the industry employment share relative to total manufacturing employment—to control for the dynamics introduced by shrinking industries that could depress wage and unionization membership—and various other industry-level controls that capture average firm-level differences across manufacturing industries. In particular, following the trends in wage premia at large firms documented by Bloometal.[2018],wecontrolfortheshareoflarge(500+employees)firmsacross manufacturing industries. Similarly, Haltiwanger et al. [2012] point to an increase intheemployer-agepremium;therefore,weincludetheshareofyoung(5yearsor less)firmswithinanindustryinourmodel. Furthermore, recent strands of the trade literature have advocated for a simultaneousroleofexportexposureandtechnologyindetermininglabordemandpatternsandhavehighlightedtheimportanceofoutsourcingandimportcompetition on wage conditions.19 Thus, (2) includes the share of exports out of total production and the import share in domestic absorption–that is, the share of imports out ofdomesticconsumption. 17SeesectionB.3formoredetailsonhowRes. Premiumisestimated. 18Withcontractnegotiationsoccurringlessfrequentlythaneveryyearinsomeindustries, wealsoconsideredaspecificationthatreliesonlongerlagsfortheunionmembershiprate— specifically,a3-yearlag. Theresultswiththe3-yearlaggedmainregressor,reportedintableB1 intheonlineappendix,arenotstatisticallydifferentfromourbaselineresultsshownintable2. 19See,forexample,Verhoogen[2008]andBustos[2011]ontheimpactofexportingonlabor demandandFeenstraandHanson[1996]andAutoretal.[2013]onthechannelsfortheimpactof imports. 16

Finally, we control for labor productivity and capital intensity, “traditional” labor market features that tend to drive wage differences across manufacturing industries. 4 Results Table 2 presents our main results. We find that the lagged unionized share of employmentisstronglypositivelyassociatedwiththemanufacturingwagepremium. Addingindustry-levelcontrolstoourbasicspecificationlowersonlymodestlyour estimatedeffectoftheunionizedshareandkeepsitstatisticallysignificant. Theeffect of changes in unionization on wages is also economically significant: Using the coefficient from column (7)—our preferred specification—reducing the share of union members in an industry from 100 percent to 0 would lead to a 10 percent decline in the wage premium. While this hypothetical shock may not be considered realistic, expressing it in terms of standard deviations of the explanatory variables—a common approach to quantify results—still points to economically meaningfulmagnitudes. Aone-standard-deviationdecreaseinthe(lagged)unionization rate—which correspond to a decline of almost 13 pp—implies that the dependentvariablewouldmovedown1.4pp. Toconvertthisresultintermsofstandard deviations, we divide this effect by the standard deviation of the dependent variable, Res. Premium, implying a decline of 0.17 (or 17 percent) standard deviation. As a final quantification exercise, we looked at how our results map into historical trends. Between 1990 and 2019, unionization rates declined, on average, almost18pp. Ourpreferredestimateimpliesadeclineinthewagepremiumof1.9 pp, thus explaining more than 70 percent of the observed decline in the premium overthesameperiod. Among explanatory variables that we control for, we find that capital intensity 17

Table2: Sector-LevelRegressions: WagePremiaandUnionizationinManufacturing (1) (2) (3) (4) (5) (6) (7) Variables ResidualWagePremium t UnionShare t−1 0.211*** 0.158** 0.157** 0.128** 0.148** 0.118** 0.110** (0.065) (0.055) (0.054) (0.056) (0.054) (0.044) (0.043) EmplShare -0.399 -0.332 -0.329 -0.246 -0.284 -0.167 -0.190 t (0.442) (0.328) (0.330) (0.253) (0.296) (0.236) (0.232) TopShare 0.109 0.108 0.119 0.109 0.121 0.128 t (0.162) (0.162) (0.146) (0.159) (0.131) (0.125) YoungShare -0.220 -0.220 -0.161 -0.202 -0.155 -0.134 t (0.240) (0.242) (0.209) (0.229) (0.194) (0.193) GLabProd -0.012 -0.013 t (0.050) (0.044) Exp/Prod 0.090 0.101* t (0.054) (0.051) Imp/Abs 0.035 -0.070 t (0.059) (0.043) CapInt 0.066*** 0.055** (0.021) (0.020) Year y y y y y y y SectorFE y y y y y y y Obs. 464 464 464 464 464 464 464 R2 0.237 0.272 0.272 0.289 0.274 0.304 0.313 NumberofSectors 16 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualWagePremium: wagepremiumaftercontrollingforagevariables,sex,education,race,tenure,maritalstatus,stateandmetropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019,across3-digitNAICSsectorswithinmanufacturing. Standarderrorsareclusteredatthesectorlevel. 18

and the export share of production have a positive effect on the manufacturing wage premium. While the export share is only marginally statistically significant, capital intensity appears to be an important factor driving the variation in residual premia: with a calculation similar to what we have outlined in the case of our main regressor, a one-standard-deviation decline in capital intensity is associated with a decline in wages of 23 percent of a standard deviation. The effect of capital intensity appears slightly larger than that of our baseline regressor; however, capital intensity has remained roughly unchanged between 1990 and 2019 and, thus, cannotexplainthedynamicsinthewagepremiaacrossmanufacturingindustries. Unionvs. Non-unionPremia: DecomposingtheImpactofUnionization Theestimatesintable2combinetheeffectsonwagesforunionandnon-union members. Intheanalysisthatfollows,weestimateequation(2)separatelyforeach group. In particular, in table 3, we show results for the manufacturing wage premiumofunionizedworkers. Thelaggedunionizedshareofemploymenthaseven strongereffectsonthemanufacturingwagepremiumforthisgroup,withthepoint estimates in the specification with all controls included being about twice as large as the point estimate in table 2. In terms of magnitudes, a one-standard-deviation decline in union membership is associated with a 26 percent of a standard deviation(sd)declineinthewagepremiumofunionizedworkers,alargerdeclinecompared with our baseline regressions, but not twice as large since the variability of residual wages across union members tends to be higher. The explanatory power of the change in unionization rates over time appears to be particularly important for union members: the almost 18 pp drop in membership rates is able to explain about 40 percent of the reduction in the residual wages of union members.20 Interestingly, the effect of capital intensity is negative for unionized workers—in contrastwithourexpectationandearlierresults—buttheeffectisonlymarginally 20Between1990and2019,residualwagepremiaofunionizedworkersdeclinedabout10pp. 19

Table3: TheImpactofUnionizationonUnionWagePremia (1) (2) (3) (4) (5) (6) (7) Variables ResidualUnionPremium t UnionShare t−1 0.230*** 0.216*** 0.226*** 0.225*** 0.242*** 0.249*** 0.263*** (0.057) (0.057) (0.058) (0.062) (0.050) (0.070) (0.059) EmplShare -0.456 -0.486 -0.519 -0.512 -0.604* -0.618* -0.726** t (0.280) (0.318) (0.297) (0.301) (0.291) (0.332) (0.309) TopShare 0.056 0.068 0.053 0.055 0.047 0.064 t (0.152) (0.151) (0.154) (0.160) (0.148) (0.143) YoungShare -0.009 -0.012 -0.027 -0.053 -0.062 -0.066 t (0.216) (0.213) (0.210) (0.245) (0.223) (0.221) GLabProd 0.128 0.102 t (0.100) (0.106) Exp/Prod -0.027 0.093 t (0.049) (0.068) Imp/Abs -0.087 -0.120 t (0.059) (0.094) CapInt -0.053* -0.055* (0.028) (0.028) Year y y y y y y y SectorFE y y y y y y y Obs. 464 464 464 464 464 464 464 R2 0.227 0.228 0.231 0.228 0.231 0.233 0.239 NumberofSectors 16 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualUnionPremium: wagepremiumforunionizedworkersaftercontrollingforage variables,sex,education,race,tenure,maritalstatus,stateandmetropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. ImportRatio: ratioofimportstorevenues. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019,across3-digitNAICSsectorswithinmanufacturing. Standarderrorsareclusteredatthesectorlevel. 20

significant. Beyond changes in employment shares, which display a somewhat non-intuitive negative correlation with the dependent variable, no other control appearstohaveameaningfulimpactontheresidualwagepremiumacrossunionizedmembers. The falling rate of unionization may have lowered wages not only because workers may lose higher wages after leaving the union, but also because there islesspressureonnonunionemployerstoraisewages. Table4containsresultsforthemanufacturingwagepremiumofnon-unionized workers. While the basic specification shown in column 1 finds similar results as in the case of overall manufacturing and unionized workers, the effect of the lagged unionized share of employment remains positive but becomes statistically insignificantafterincludingmostothercontrols. Among other variables, capital intensity emerges as the most relevant factor influencing wages for non-union members. A one-standard-deviation decline in capital intensity is associated with a 35 percent standard deviation decline in the wage premium of non-union members. However, the historical trend in capital intensity is not consistent with the dynamics in the wage premium; indeed, over the period of analysis, the nonunion wage premium moved down, while capital intensityedgedup. RobustnessChecks Ourspecificationincludesseveralsector-specificfactorsthattendtoaffectwage patterns. In this section, we show that our results are robust to three additional characteristics. First, as documented in figure 8, wage premia converged across industries within manufacturing. To account for possibility that the declines in wage premia are due to this convergence dynamics, we have added the lagged premium to our baseline model; the results are shown in table A1. Columns (2), (4), and (6) report 21

Table4: TheImpactofUnionizationonNonunionWagePremia (1) (2) (3) (4) (5) (6) (7) Variables ResidualNonunionPremium t UnionShare t−1 0.182** 0.110 0.109 0.077 0.089 0.046 0.039 (0.068) (0.069) (0.068) (0.071) (0.075) (0.053) (0.055) EmplShare -0.287 -0.240 -0.238 -0.145 -0.145 0.020 0.030 t (0.510) (0.370) (0.374) (0.289) (0.315) (0.245) (0.248) TopShare 0.173 0.173 0.184 0.174 0.192 0.196 t (0.204) (0.206) (0.186) (0.199) (0.147) (0.143) YoungShare -0.255 -0.255 -0.189 -0.220 -0.152 -0.134 t (0.253) (0.254) (0.219) (0.233) (0.181) (0.181) GLabProd -0.007 0.004 t (0.052) (0.046) Exp/Prod 0.100 0.047 t (0.064) (0.073) Imp/Abs 0.070 -0.011 t (0.072) (0.065) CapInt 0.103*** 0.096*** (0.028) (0.031) Year y y y y y y y SectorFE y y y y y y y Obs. 464 464 464 464 464 464 464 R2 0.234 0.280 0.281 0.296 0.287 0.337 0.339 NumberofSectors 16 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualNon-UnionPremium: wagepremiumfornon-unionizedworkersaftercontrollingforagevariables,sex,education,race,tenure,maritalstatus,stateandmetropolitan area. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019,across3-digitNAICSsectorswithinmanufacturing. Standarderrorsareclusteredatthesectorlevel. 22

the specification that controls for the lagged premium; we also report comparable estimates in columns (1), (3), and (5), respectively. The impact of unionization on wage premia is little changed in the augmented specification compared with our baseline estimates. The impact of other covariates is also little changed in the new specification. Interestingly,afterincludingourlargesetofcontrols,pastpremiaare positively related to the current premium, in contrast with a story of convergence, butonlyforthenonuniongroup. Second, the trade literature since the seminal contribution by Bernard et al. [1995] has highlighted that exporters are not only larger, more productive, and more capital intensive compared to non-exporting establishments, but also pay higher wages and benefits. Thus, while we control for the export share of production in our baseline specification, cross-sector differences in the prevalence of exporters could account for further differences in wages. Table A2 addresses this concernbycontrollingforthe(log-)numberofexportersaswellastheaverageexport value across exporters.21 Since those variables are available only since 1996, thefirstcolumnacrossalltablesreplicatesthebaselineresultsfortheshortersampleandcanbedirectlycomparedtothemagnitudesofestimatesincolumns(2)-(4), which includes the export-related controls. After including those controls, the impact of unionization on the wage premium is not significantly different from the resultsincolumn(1). Among other controls, table A2 points to a role for trade variables and for the share of young firms in the sector; these results, however, are not robust to estimatingthepremiaseparatelyforunionizedandnon-unionizedworkers. A third factor we consider in our robustness exercises is the role of temporary helpworkers. Infact,ashiftinthenumberofjobsthatarefilledthroughtemporary help workers could affect the wage premia of other workers in the sector; the im- 21Theonlineappendixalsoincludesadditionalresults,decomposingbetweenunionand nonunionpremiaintablesB2andB3. 23

pactonwageslargelydependsonwhichtypesofoccupationsarefilledwiththose workers. To understand the effect of temporary help workers on wages, we rely onyearlydatafromtheQuarterlySurveyofPlantCapacity(QSPC),whichcollects theshareoftemporaryhelpworkersacrossmanufacturingindustriessince2013.22 Figure A1 plots the share of temporary help workers in each 3-digit NAICS industry against the industry-specific residual wage premia. As highlighted by the fitted line, the correlation between temporary help employment and wage premia is positive: intuitively, firms are likely to fill the least skill-intensive–and likely least well-paying–occupation with outside help, implying that the wage premia for other workers in the sectors would raise. As a result, the increasing reliance ofthemanufacturingsectorontemporaryhelpworkerssincethe1990sisunlikely to account for the decline in the manufacturing wage premium over the period of ouranalysis. We perform a final robustness check following the work by Hirsch and Schumacher [2004] and Bollinger and Hirsch [2006], which point to biased estimates forunionwagegapswhenimputedCPSwagedataareincludedintheestimation. Following Bollinger and Hirsch [2006], we restrict our analysis to the respondent samplewithobservationsweightedby(theinverseof)theprobabilityofresponse. Tables A3 and present our results.23 The decline in unionization continues to be the most important factor for the decline in the wage premium for union workers acrossmanufacturingindustries;lookingatcolumn(7),thecoefficientestimatesis only marginally significantly different from our baseline estimates shown in table 3. Reweighted regressions highlight a more prominent role of international trade, butthehistoricalmovementsinimportandexportshares–withimportandexport sharesrisingovertime–arecounterfactualtothepatternsinmanufacturingwages. 22Weexclude2013q2–2013q4datafromouranalysissincetheydonotcoverafullyear. 23SeetableB4intheonlineappendixfortheestimateonthenonunionpremium. 24

5 Implications for Wage Inequality The trends in unionization and the resulting patterns in wages have also important implications in terms of wage inequality, the focus of this section. In fact, the patterns that we document imply that, within the manufacturing sector, there has been a widening of the gap between the higher-paid nonproduction workers and the relatively lower-paid production workers and point to rising inequality across occupations. Furthermore, the declines in the wages of manufacturing productionworkersrelativetoworkersinothersectorspointtorisinginequalitybetween manufacturing and the rest of the economy. All told, those results point to rising wageinequalitybetweensectorsandoccupations. The data confirm these conjectures. Looking first at general trends, shown in figure 10, measures of aggregate inequality have increased since the 1980s, while inequalitywithinsectorsandforagivenoccupation—hereafter,withininequality— has moved down.24 As a result, the inequality between sectors and occupations— or between inequality, defined as the difference between aggregate inequality and the within sector-occupation measure—rose 0.12 log point (or about 12 percent). Measures of inequality that control for demographic characteristics reveal similar patterns,asshowninfigure11,withanincreaseinbetweeninequalityofabout4.5 percent. 24OurwageinequalitydecompositionlargelyfollowsDavisandHaltiwanger[1991]. 25

Figure10: WageInequality Figure11: ResidualWageInequality Beyond consistent trends, changes in wage premia have quantitatively contributed to the recent trends in wage inequality. Results from table 5 quantify the impact of changes in the premium by occupation group on the increase in aggregateandbetweenwageinequality, Ineq = δ +δ MfgProd. Premium +δ MfgNonprod. Premium +d +u (3) t 0 1 t 2 t t t whereIneq denotesameasureofwageinequality,MfgProd. Premium represents t t thewagepremiumformanufacturingproductionworkers,andMfgNonprod. Premium t isthewagepremiumformanufacturingnonproductionworkers. Whiletheimpact of premia for nonproduction workers on inequality is not robust across specifications,wefindthataone-standard-deviationdeclineinthepremiumforproduction workers is associated with an increase in overall inequality of about 10 percent of asdandanincreaseinbetweeninequalityofabout15percentofasd. Thefactthat onlychangesinproductionworkerpremiamatterforinequalityisconsistentwith thefindingsfromtheprevioussections. Tying these effects to historical trends, the decline in manufacturing production worker wages since the 1990s explains 10 percent—or about 0.5 pp—of the increasein(betweenandoverall)wageinequalityoverthesameperiod,aftercon- 26

trollingforchangesindemographicscharacteristics. Inturn,theeffectofthewage premium on inequality is primarily driven by the dynamics in unionization rates; in fact, changes in unionization rates ultimately account for about 0.35 pp of the changesinwageinequality. While changes in unionization rates and manufacturing wages explain only a smallpartofthetotalincreaseinwageinequality,thefactthatasimilarmechanism might be at play also in some other sectors suggest that our estimate should be interpretedasalowerbound. Table5: InequalityandWagePremia (1) (2) (3) (4) Inequality Variables Total BetweenSectors-Occupations Premium,Production -0.102*** -0.099** (0.038) (0.047) Premium,Nonproduction 0.124* 0.075 (0.065) (0.097) ResidualPremium,Production -0.125*** -0.180*** (0.044) (0.059) ResidualPremium,Nonproduction 0.025 -0.191* (0.077) (0.110) MonthDummies y y y y Obs. 432 432 432 432 R2 0.218 0.224 0.226 0.243 Source: BureauofLaborStatistics. Premium,Production: wagepremiumforproductionworkers. Premium,Nonproduction: wagepremiumfornonproductionworkers. ResidualPremium,Production: wagepremiumforproductionworkersaftercontrollingfor agevariables,sex,education,race,tenure,maritalstatus,stateandmetropolitanarea. ResidualPremium,Nonproduction: wagepremiumfornonproductionworkersaftercontrollingforagevariables,sex,education,race,tenure,unionstatus,maritalstatus,stateand metropolitanarea. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Time-seriesregressions,1980m1-2019m12. Robuststandarderrorsarereportedin parenthesis. 27

6 Conclusions The conventional wisdom that manufacturing jobs are “good jobs” is less accurate than it used to be. While manufacturing workers used to receive a premium relative to workers in other sectors, that premium has significantly declined in recent years for most manufacturing jobs. Our results indicate that the decline in unionizationratesisresponsibleformorethan70percentofthedropinthemanufacturingwagepremium. Notably,theunionizationeffectremainssignificanteven afteraccountingforalargesetofworkerandsectoralcharacteristics. Our findings also point to a widening of wage inequality across occupations withinmanufacturingandwithrespecttotheprivatesector. Inparticular,wefind that, after controlling for demographics characteristics, the decline in manufacturing production worker wages since the 1990s explains 10 percent—or about 0.5 pp—of the increase in wage inequality over the same period, with the largest portionofthiseffectattributabletothedeclineinunionizationrates. Ourestimatesof the impact on wage inequality, however, are likely only a lower bound, as while the wages of manufacturing production workers were above the median of the distribution of all production worker wages, they remained typically below that ofnonproductionworkers. Beyondsuggestiveevidencethatasimilarrelationship between unionization and wages could be at play in other sectors, the decline in unionization rates and in the wage premium might exacerbate the structural declineofthemanufacturingsector—oneofthesectorsthatmadethemiddleclass— andfurtherraisewageinequality. References JohnMAbowd,FrancisKramarz,andDavidNMargolis. HighWageWorkersand HighWageFirms. Econometrica,67(2):251–333,1999. 28

H Autor, David, David Dorn, and Gordon H Hanson. The China Syndrome: Local Labor Market Effects of Import Competition in the United States. American EconomicReview,103(6):2121–68,2013. Andrew B Bernard, J Bradford Jensen, and Robert Z Lawrence. Exporters, Jobs, andWagesinUSManufacturing: 1976-1987. Brookingspapersoneconomicactivity. Microeconomics,1995:67–119,1995. NicholasBloom,FatihGuvenen,BenjaminSSmith,JaeSong,andTillvonWachter. The Disappearing Large-Firm Wage Premium. In AEA Papers and Proceedings, volume108,pages317–22,2018. ChristopherRBollingerandBarryTHirsch. MatchBiasfromEarningsImputation in the Current Population Survey: The Case of Imperfect Matching. Journal of LaborEconomics,24(3):483–519,2006. BureauofLaborStatistics. CurrentEmploymentStatistics,2019a. BureauofLaborStatistics. CurrentPopulationSurvey,2019b. BureauofLaborStatistics. EmployeeCostofEmployerCompensation,2019c. BureauofLaborStatistics. TotalFactorProductivity,2021. PaulaBustos. TradeLiberalization,Exports,andTechnologyUpgrading: Evidence on the Impact of MERCOSUR on Argentinian Firms. The American economic review,101(1):304–340,2011. DavidCard. TheEffectofUnionsontheStructureofWages: ALongitudinalAnalysis. Econometrica,64(4):957–979,July1996. CensusBureau. U.S.InternationalTradeData,2019. CensusBureau. QuarterlyWorkforceStatistics,2021. Steven J Davis and John Haltiwanger. Wage Dispersion Between and Within US ManufacturingPlants,1963-86. BrookingsPapersonEconomicActivity,pages115– 180,1991. John DiNardo, Nicole M. Fortin, and Thomas Lemieux. Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach. Econometrica,64(5):1001–1044,1996. 29

Joel Elvery and Julianne Dunn. Manufacturing Wage Premiums Have Diverged between Production and Nonproduction Workers. Regional policy report, FederalReserveBankofCleveland,2021. FederalReserveBoard. ManufacturingInvestmentandCapital,2021. Robert C Feenstra and Gordon H Hanson. Globalization, Outsourcing, and Wage Inequality. TheAmericanEconomicReview,86(2):240–245,1996. Nicole M. Fortin and Thomas Lemieux. Institutional Changes and Rising Wage Inequality: IsThereaLinkage? JournalofEconomicPerspectives,11(2):75–96,June 1997. Richard B. Freeman. How Much Has De-Unionisation Contributed to the Rise in Male Earnings Inequality? In Sheldon Danziger and Peter Gottschalk, editors, UnevenTides,pages133–163.SagePress,1992. RichardB.FreemanandJamesL.Medoff. TheImpactofthePercentageOrganized on Union and Nonunion Wages. The Review of Economics and Statistics, pages 561–572,1981. Craig Giffi, Michelle Drew Rodriguez, and Sandeepan Mondal. A Look Ahead: How Modern Manufacturers Can Create Positive Perceptions with the US Public. Technicalreport,Deloitte,2017. ClaudiaGoldinandLawrenceF.Katz. Decreasing(andthenIncreasing)Inequality in America: A Tale of Two Half-Centuries. In F. Welch, editor, The Causes and Consequences of Increasing Income Inequality, pages 37–82. University of Chicago Press,2001. Eric D. Gould. Explaining the Unexplained: Residual Wage Inequality, Manufacturing Decline and Low-Skilled Immigration. The Economic Journal, 129(619): 1281–1326,2019. John Haltiwanger, Henry R Hyatt, Erika McEntarfer, and Liliana Sousa. Business Dynamics Statistics Briefing: Job Creation, Worker Churning, and Wages at Young Businesses. Worker Churning, and Wages at Young Businesses (November 1,2012),2012. Katelynn Harris and Michael D. McCall. The Relative Weakness in Earnings of Production Workers in Manufacturing, 1990–2018. Monthly Labor Review, pages 1–15,2019. 30

BarryTHirschandEdwardJSchumacher. MatchBiasinWageGapEstimatesdue toEarningsImputation. JournalofLaborEconomics,22(3):689–722,2004. AustenHuffordandNoraNaughton. WageGainsatFactoriesFallBehindGrowth inFastFood. WallStreetJournal,2021. Lawrence F. Katz and David H. Autor. Changes in the Wage Structure and Earnings Inequality. In Orley C. Ashenfelter and David Card, editors, Handbook of LaborEconomics,pages1463–1555.Elsevier,1999. David Langdon and Rebecca Lehrman. The Benefits of Manufacturing Jobs. Issue Brief1-12,U.S.DepartmentofCommerce,EconomicsandStatisticsAdministration,2012. Marc Levinson. Job Creation in the Manufacturing Revival. CRS Report R41898, 2019. DavidAMacphersonandBarryTHirsch. Fivedecadesofunionwages,nonunion wages,andunionwagegapsatunionstats.com. 2021. Jacob A Mincer. Schooling, experience, and earnings. Technical report, National BureauofEconomicResearch,1974. LawrenceMishel. Yes,ManufacturingStillProvidesaPayAdvantage,butStaffing FirmOutsourcingisErodingit. EPIWorkingPaper141193,2018. EmilyPollard. ANewApproachtoIndustryandOccupationRecodingintheCPS. TechnicalBriefingsTB19-02,FederalReserveBankofKansasCity,2019. Catherine Ruckelshaus and Sarah Leberstein. Manufacturing Low Pay: Declining Wages in the Jobs that Built America’s Middle Class. Technical report, NELP, 2014. EricA.Verhoogen. Trade,QualityUpgrading,andWageInequalityintheMexican ManufacturingSector. TheQuarterlyjournalofeconomics,123(2):489–530,2008. 31

A Appendix FigureA1: TemporaryHelpServicesandWagePremiaacrossManufacturingIndustries 32

TableA1: WagePremiaandUnionizationinManufacturing,ControllingforConvergenceinPremia (1) (2) (3) (4) (5) (6) Variables ResidualPremium Res. UnionPremium Res. NonunionPremium t t t UnionShare t−1 0.110** 0.089** 0.263*** 0.271*** 0.039 0.002 (0.043) (0.033) (0.059) (0.060) (0.055) (0.042) EmplShare -0.190 -0.178 -0.726** -0.773** 0.030 -0.030 t (0.232) (0.196) (0.309) (0.330) (0.248) (0.201) TopShare 0.128 0.117 0.064 0.071 0.196 0.192 t (0.125) (0.114) (0.143) (0.147) (0.143) (0.138) YoungShare -0.134 -0.103 -0.066 -0.053 -0.134 -0.080 t (0.193) (0.163) (0.221) (0.225) (0.181) (0.135) GLabProd -0.013 -0.018 0.102 0.111 0.004 -0.004 t (0.044) (0.042) (0.106) (0.106) (0.046) (0.041) Exp/Prod 0.101* 0.094** 0.093 0.096 0.047 0.055 t (0.051) (0.043) (0.068) (0.073) (0.073) (0.057) Imp/Abs -0.070 -0.064* -0.120 -0.124 -0.011 -0.018 t (0.043) (0.036) (0.094) (0.104) (0.065) (0.048) CapInt 0.055** 0.037* -0.055* -0.055* 0.096*** 0.071** (0.020) (0.019) (0.028) (0.030) (0.031) (0.027) ResidualPremium t−1 0.195** (0.076) ResidualUnionPremium t−1 -0.074 (0.063) ResidualNonunionPremium t−1 0.194*** (0.055) Year y y y y y y SectorFE y y y y y y Obs. 464 464 464 464 464 464 R2 0.313 0.336 0.239 0.243 0.339 0.361 Numberofsectors 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualPremium: wagepremiumaftercontrollingforagevariables,sex,education,race,tenure,marital status,stateandmetropolitanarea. Res. UnionPremium: wagepremiumforunionizedworkersaftercontrollingforagevariables,sex,education,race,tenure,maritalstatus,stateandmetropolitanarea. Res. Non-UnionPremium: wagepremiumfornon-unionizedworkersaftercontrollingforagevariables,sex, education,race,tenure,maritalstatus,stateandmetropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1996-2019,across3-digitNAICSsectorswithinmanufacturing. Standard errorsareclusteredatthesectorlevel. 33

TableA2: WagePremiaandUnionizationinManufacturing,AdditionalExporters’Controls (1) (2) (3) (4) Variables ResidualWagePremium t UnionShare t−1 0.075** 0.072** 0.075** 0.069** (0.032) (0.029) (0.032) (0.026) EmplShare -0.033 -0.039 -0.021 -0.004 t (0.204) (0.202) (0.207) (0.212) TopShare 0.096 0.103 0.098 0.116 t (0.130) (0.131) (0.130) (0.135) YoungShare -0.318** -0.319** -0.319** -0.321** t (0.141) (0.140) (0.142) (0.141) GLabProd -0.014 -0.015 -0.016 -0.024 t (0.045) (0.048) (0.049) (0.057) Exp/Prod 0.137** 0.139** 0.138** 0.144** t (0.050) (0.052) (0.051) (0.054) Imp/Abs -0.156*** -0.146** -0.159*** -0.149** t (0.051) (0.056) (0.054) (0.058) CapInt 0.043** 0.038 0.042** 0.036 (0.019) (0.023) (0.019) (0.025) lnNumExporters -0.257 -0.401 (0.406) (0.592) lnAvg. ExportValue -0.067 -0.219 (0.179) (0.359) Year y y y y SectorFE y y y y Obs. 384 384 384 384 R2 0.210 0.211 0.210 0.212 NumberofSectors 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualWagePremium: wagepremiumaftercontrollingforage variables,sex,education,race,tenure,maritalstatus,stateand metropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthe sector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. lnNumExporters: lognumberofexportersinthesector;because ofdataavailability,before2008,thenumberofexportersforeach sectoristhenumberforoverallmanufacturing. lnAvg. ExportValue: logaverageexportvalueperexporter withinthesector;becauseofdataavailability,before2008,the exportvalueandthenumberofexportersforeachsectoraresetto thevaluesforoverallmanufacturing. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1996-2019,across3-digit NAICSsectorswithinmanufacturing. Standarderrorsareclusteredatthesectorlevel. 34

TableA3: UnionizationandUnionWagePremiainManufacturing,Re-weighted ObservableSample (1) (2) (3) (4) (5) (6) (7) Variables ResidualUnionPremium t UnionShare t−1 0.160* 0.140** 0.147** 0.135** 0.165*** 0.160*** 0.170*** (0.078) (0.054) (0.056) (0.058) (0.054) (0.051) (0.048) EmplShare -0.024 -0.174 -0.202 -0.159 -0.299 -0.264 -0.412 t (0.341) (0.476) (0.454) (0.460) (0.453) (0.476) (0.454) TopShare 0.147 0.157 0.149 0.146 0.141 0.163 t (0.142) (0.139) (0.141) (0.150) (0.148) (0.138) YoungShare 0.092 0.090 0.103 0.046 0.057 0.064 t (0.457) (0.455) (0.470) (0.456) (0.470) (0.456) GLabProd 0.104 0.067 t (0.084) (0.085) Exp/Prod 0.016 0.183** t (0.061) (0.068) Imp/Abs -0.091 -0.203** t (0.076) (0.093) CapInt -0.036 -0.045 (0.034) (0.037) Year y y y y y y y SectorFE y y y y y y y Obs. 464 464 464 464 464 464 464 R2 0.148 0.150 0.153 0.151 0.154 0.153 0.165 NumberofSectors 16 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualUnionPremium: wagepremiumforunionizedworkersaftercontrollingforage variables,sex,education,race,tenure,maritalstatus,stateandmetropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. lnNumExporters: lognumberofexportersinthesector;becauseofdataavailability,before2008,thenumberofexportersforeachsectoristhenumberforoverallmanufacturing. lnAvg. ExportValue: logaverageexportvalueperexporterwithinthesector;becauseof dataavailability,before2008,theexportvalueandthenumberofexportersforeachsector aresettothevaluesforoverallmanufacturing. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019,across3-digitNAICSsectorswithinmanufacturing,ontherespondentsampleonly,withobservationweightedbytheprobabilityof response. Standarderrorsareclusteredatthesectorlevel. 35

B Online Appendix B.1 The Manufactuing Wage Premium in the Current Employment Statistics (CES) Data WhileourmainempiricalanalysisdrawsontheCurrentPopulationSurvey(CPS), thissectionpresentsthemanufacturingwagepremiumusingtheCurrentEmployment Statistics (CES) survey, which is based on establishment-level data and is designedtomeasurebroadpatternsinsector-levelemploymentandearnings. AsmeasuredintheCESdata,manufacturingaveragehourlywagesforallemployees were 3 percent above wages in the overall private sector in 2006, a difference commonly known as the manufacturing wage premium. Since then, manufacturing wages have averaged gains of 2.3 percent per year, while wages in the private sector have risen 2.6 percent per year. Because of this differential growth, thelevelofwagesintheprivateeconomycaughtupwithmanufacturingwagesin April2018andhasbeenhighereversince(figureB1). FigureB1: ManufacturingWagePremium - Notably, the erosion of the manufacturing wage premium has been a phenomenon that has affected production workers, which, in 2019, accounted for about 70 percent of manufacturing employment. Figures B2 and B3 report data 36

on the manufacturing premium as a percent difference relative to private-sector wagesbymajoroccupationgroups: figureB2showsthemanufacturingwagepremium for production workers, and figure B3 shows the premium for nonproduction/supervisoryworkers. Whilenonsupervisoryworkersinmanufacturinghave traditionallysufferedawagediscount,thatdifferencehasincreasedonlyinrecent years. In contrast, the manufacturing wage premium across production occupations declined steadily beginning in the late 1990s, and it had disappeared by the mid-2000s. In 2019, manufacturing wages for production workers were 5 percent belowthoseofproductionworkersintherestoftheprivatesector. FigureB2: ManufacturingWagePre- FigureB3: ManufacturingWagePremium,ProductionOccupations mium,SupervisoryOccupations The patterns for production worker premia in the CES data are roughly consistent with our findings for CPS, while the empirical evidence for supervisory workers is different across the two datasets. There are two main reasons for such differences. First, the CPS analysis focuses on production occupations, a smaller set than the group of nonsupervisory occupations in the CES sample. Second, the CPS comparison are relative to other sectors outside of manufacturing, while the CES total wage measures include manufacturing. All told, the restriction to productionoccupationsexplainsmostofthedifferentpatterns. B.2 The Manufacturing Premium: Weekly Wages and Nonwage Compensation Differencesbetweenmanufacturingandothersectorshavealsobeennarrowingin weekly wages. Figure B7 shows the percent difference in average weekly wages between manufacturing production workers and production occupations in other 37

sectors as measured in CES data. After rising in the 1980s, differences in production worker weekly wages gradually declined in subsequent decades but still averaged 18.6 percent in the 2010s. While the weekly manufacturing premium remains sizeable, the declining trend appears to be the result of the behavior of averagehourlywagesratherthanthenumberofhoursworkedinaweek. Indeed, differences in average weekly hours between manufacturing and other sectors, as shown in figure B8, have increased over time. In the 1980s, a production worker inmanufacturingworkedanaverageof40.4hoursperweekversus36.2hoursfor production workers in other sectors. The workweek difference rose to 8.1 hours in the 2010s, with average weekly hours increasing for manufacturing production workers but decreasing for production workers in other sectors. The increase in hours has partly offset the decline in average hourly wages, resulting in weekly wagesinmanufacturingremainingstillabovethoseofothersectors. Wages, however, represent only one dimension of manufacturing compensation. Other characteristics—such as job security and benefits (the nonwage component of compensation)—have historically been higher for manufacturing workers and, thus, have contributed to the view of a manufacturing premium. However, while separation rates for manufacturers remained significantly below those of other sectors through the 2010s, differences in nonwage compensation between manufacturingandothersectorshavebeennarrowinginrecentyears.25 Figure B9 reports the percent difference in the monetary value of benefits betweenmanufacturingandtheprivatesector;thesedata,drawnfromtheEmployer CostofEmployeeCompensationsurvey,areavailableonlysince2004andprovide some separate details for full-time and production workers.26 The relative decline in benefits is particularly striking for full-time workers: While in the mid-2000s full-time manufacturing workers—a group that includes both production and supervisory occupations—received 20 percent higher benefits (by value) relative to full-time employees in other sectors of the economy, that premium declined to 10 percentby2019andcontinuedtomoveddowninthefollowingyears. Production workers in manufacturing, instead, have continued to receive benefits that are 10 percent higher than those in other sectors since the mid-2000s. The trend in the 25UsingCPSdata,wefindthatseparationrateswithinmanufacturingare5ppbelowthoseof othersectors. 26Dataonbenefitsareavailableonlyeitherforfull-timestatusorbyoccupation(production vs. non-productionworkers),andthus,arenotfullycomparableforthemorerefinedcellsthat weuseinourwageanalysis. Furthermore,dataforproductionworkersareonlyavailablesince 2006Q4. 38

benefitsformanufacturingproduction(full-andpart-time)workershaspartlyoffset the declines in wages and salaries still implying a 5 percent premium in total compensation relative to production workers in other sectors. All told, as we focusonfull-timeproductionworkersinmanufacturing,thesefindingsstillsuggest mildlydecliningpatternsinthebenefitsandtotalcompensationformanufacturing workersrelativetoothersectors. All told, several characteristics of manufacturing jobs have deteriorated over timerelativetoothersectors. B.3 Estimating Residual Industry-Level Wage Premia Residual industry-level wage premia are estimated from a wage equation that includes a manufacturing industry dummies, worker demographics, and other observables. In particular, we estimate the following model for each manufacturing industry, lnw = α +α MfgInd +δX +d +ε , ist 0 1,st ist ist t it wherew denotesthehourlywageforworkeriinindustrysattimetandMfgInd ist ist isadummyindicatorforanindustrywithinmanufacturing,andRes. Premium ≡ st α .27 In each regression, we include a dummy for a single manufacturing in- 1,st dustry and drop the observations for all other manufacturing industries in order to keep the same comparison group (private sector outside of manufacturing). As in the descriptive analysis, X includes demographics and other worker ist observables; specifically, we control for full-time status, union membership, gender, education—with the return to education allowed to vary across economic regions, age group, race and ethnicity, marital status, tenure—using a second-order polynomial—metropolitan statistical area dummies, and state-year fixed effects.28 Finally,ourresidualpremiaestimatesvarynotonlybysectorbutalsobyyear. B.4 Aggregate Effects Whilethepatternsinwagesandunionizationwithinthemanufacturingsectorand therelateddataavailabilitysuggestthatmanufacturingindustriesareanidealsetting to study the importance of unionization trends on the wage premium, this 27Specifically,s ∈ {311,312,313−316,321,322−323,324,325,326,327,331−332,333,334,335,336,337,339}. 28Inrobustnessanalysis,wehavealsoadoptedaspecificationthatallowsthereturnstodemographiccharacteristicstovaryovertime–thatis,weestimateδ foreachdemographiccharacterist tics. Theresults,availableuponrequest,areinsignificantlydifferentfromourbaseline. 39

section extends our empirical analysis to the entire private sector and evaluate whether similar dynamics occurred in sectors outside of manufacturing. Table B5 presentstheestimatesfromourbaselinemodelwhentheunitofobservationisa2digitNAICSsector. Inthissetting,ourmodelincludesamorerestrictedsetofcontrol;infact,weexcludecontrolsonexportsandimportpenetrationbecausedataon those characteristics for most service sectors are extremely limited. The first three columnsincludeallsectors.29 Ourresultscontinuetopointtoapositivecorrelation between the lagged unionization rate and the wage premium within sectors, but thecoefficientsaremuchlesspreciselyestimated. Indeed,whileunionizationrates have been largely declining across most sectors—with few cases where rates have remained little changed—many service sectors have, instead, experienced a relativeriseintheirwagepremium. Furthermore,thesampleofunionworkersinmost sectors outside of manufacturing appears fairly small, leading to large variability in the estimates of unionization rates from one year to the next. If limiting the analysis to the sectors that have experienced above-the-median declines in unionization over time and smoothing through the volatility in unionization rates—as shownincolumns(3)-(6)—wefindthatthemagnitudeoftheeffectofunionization on wage premia is notably higher—suggesting that declines in unionization rates could explain around 70 percent of the decline in the wage premia even across thosesectors—althoughsignificantonlyatthe10percentconfidencelevel.30 Amongotherfactors,onlylaborproductivityemergesasasignificantdeterminantofwagedifferencesacross2-digitNAICSsectors. 29Specifically,ourregressionincludes18sectors: Agriculture,Forestry,Fishing,andHunting (NAICS11),Mining(NAICS21),Utilities(NAICS22),Construction(NAICS23),Manufacturing (NAICS31-33),WholesaleTrade(NAICS42),RetailTrade(NAICS44-45),TransportationandWarehousing(NAICS48-49),Information(NAICS51),Finance(NAICS52),RealEstate(NAICS53),Professional,Scientific,andTechnicalServices(NAICS54),AdministrativeandSupportandWasteManagementandRemediationServices(NAICS56),EducationServices(NAICS61),HealthCareandSocial Assistance(NAICS62),Arts,Entertainment,andRecreation(NAICS71),AccommodationandFood Services(NAICS72),andOtherServices(NAICS81). 30Above-the-mediandeclinesinunionizationratesoccurredinthefollowingsectors: Agriculture,Forestry,Fishing,andHunting(NAICS11),Mining(NAICS21),Utilities(NAICS22),Construction(NAICS23),Manufacturing(NAICS31-33),WholesaleTrade(NAICS42),RetailTrade(NAICS 44-45),TransportationandWarehousing(NAICS48-49),andInformation(NAICS51). 40

B.5 Additional Results FigureB4: ManufacturingWagePre- FigureB5: ManufacturingWagePremium,Yearly,ProductionOccupa- mium,Yearly,NonproductionOccutions pations FigureB6: ManufacturingWagePremiumrelativetoConnectedSectors 41

FigureB7: ManufacturingWagePremium,WeeklyWages FigureB8: AverageWeeklyProductionWorkerHours 42

FigureB9: ManufacturingBenefitPremiumbyWorkerGroups 43

FigureB11: ProductionWorkerEm- FigureB10: EmploymentbyGender ploymentbyAge FigureB12: ProductionWorkerEm- FigureB13: ProductionWorkerEmploymentbyEducation ploymentbyRaceandEthnicity 44

FigureB14: EvolutionofProduction FigureB15: EvolutionofProduction WorkerEmploymentbyGender WorkerEmploymentbyAge FigureB17: EvolutionofProduction FigureB16: EvolutionofProduction WorkerEmploymentbyRaceand WorkerEmploymentbyEducation Ethnicity 45

FigureB18: ManufacturingPremia: DecompositionofUnionInflowsandOutflows 46

TableB1: Sector-LevelRegressions: WagePremiaand3-PeriodLaggedUnionizationinManufacturing (1) (2) (3) (4) (5) (6) (7) Variables ResidualWagePremium t UnionShare t−3 0.202*** 0.153** 0.152** 0.124** 0.143** 0.111** 0.104** (0.052) (0.056) (0.056) (0.056) (0.054) (0.047) (0.045) EmplShare -0.419 -0.378 -0.374 -0.282 -0.324 -0.201 -0.219 t (0.436) (0.323) (0.324) (0.244) (0.287) (0.232) (0.226) TopShare 0.134 0.132 0.139 0.132 0.139 0.145 t (0.164) (0.163) (0.147) (0.160) (0.135) (0.128) YoungShare -0.191 -0.191 -0.136 -0.174 -0.134 -0.113 t (0.232) (0.234) (0.201) (0.221) (0.189) (0.187) GLabProd -0.016 -0.016 t (0.051) (0.044) Exp/Prod 0.092 0.101* t (0.054) (0.051) Imp/Abs 0.038 -0.069 t (0.058) (0.041) CapInt 0.066*** 0.055** (0.021) (0.020) Year y y y y y y y SectorFE y y y y y y y Obs. 464 464 464 464 464 464 464 R2 0.232 0.270 0.270 0.288 0.273 0.302 0.312 NumberofSectors 16 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualWagePremium: wagepremiumaftercontrollingforagevariables,sex,education,race,tenure,maritalstatus,stateandmetropolitanarea. UnionShare t−3 : unionizedshareofemployment,3-periodlag. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019,across3-digitNAICSsectorswithinmanufacturing. Standarderrorsareclusteredatthesectorlevel. 47

TableB2: UnionizationandUnionWagePremiainManufacturing,Additional Exporters’Controls (1) (2) (3) (4) Variables ResidualUnionPremium t UnionShare t−1 0.296*** 0.288*** 0.297*** 0.290*** (0.070) (0.073) (0.072) (0.074) EmplShare -0.517 -0.533 -0.588 -0.567 t (0.406) (0.396) (0.429) (0.452) TopShare -0.127 -0.108 -0.143 -0.121 t (0.278) (0.271) (0.278) (0.262) YoungShare -0.423 -0.426 -0.421 -0.425 t (0.355) (0.354) (0.355) (0.357) GLabProd 0.073 0.070 0.088 0.079 t (0.123) (0.126) (0.130) (0.139) Exp/Prod 0.121 0.126 0.114 0.121 t (0.075) (0.074) (0.072) (0.073) Imp/Abs -0.201 -0.175 -0.185 -0.172 t (0.162) (0.146) (0.157) (0.147) CapInt -0.096** -0.106*** -0.095** -0.103*** (0.041) (0.035) (0.041) (0.031) lnNumExporters -0.636 -0.495 (0.763) (1.001) lnAvg. ExportValue 0.402 0.214 (0.519) (0.685) Year y y y y SectorFE y y y y Obs. 384 384 384 384 R2 0.161 0.162 0.162 0.162 NumberofSectors 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualUnionPremium: wagepremiumforunionizedworkers aftercontrollingforagevariables,sex,education,race,tenure, maritalstatus,stateandmetropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthe sector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. lnNumExporters: lognumberofexportersinthesector;because ofdataavailability,before2008,thenumberofexportersforeach sectoristhenumberforoverallmanufacturing. lnAvg. ExportValue: logaverageexportvalueperexporter withinthesector;becauseofdataavailability,before2008,theexportvalueandthenumberofexportersforeachsectoraresetto thevaluesforoverallmanufacturing. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1996-2019,across3-digitNAICS sectorswithinmanufacturing. Standarderrorsareclusteredatthe sectorlevel. 48

TableB3: UnionizationandNon-UnionWagePremiainManufacturing,AdditionalExporters’Controls (1) (2) (3) (4) Variables ResidualNon-UnionPremium t UnionShare t−1 -0.036 -0.035 -0.037 -0.039 (0.034) (0.031) (0.031) (0.028) EmplShare 0.028 0.030 0.076 0.083 t (0.222) (0.229) (0.220) (0.210) TopShare 0.278 0.276 0.289 0.296 t (0.170) (0.172) (0.174) (0.183) YoungShare -0.157 -0.157 -0.158 -0.159 t (0.179) (0.179) (0.181) (0.179) GLabProd 0.012 0.012 0.002 -0.001 t (0.043) (0.044) (0.042) (0.049) Exp/Prod 0.093 0.093 0.098 0.100 t (0.075) (0.076) (0.072) (0.078) Imp/Abs -0.086 -0.089 -0.097 -0.093 t (0.066) (0.059) (0.057) (0.056) CapInt 0.095*** 0.096** 0.094** 0.092** (0.032) (0.039) (0.033) (0.042) lnNumExporters 0.062 -0.155 (0.604) (0.784) lnAvg. ExportValue -0.271 -0.330 (0.163) (0.398) Year y y y y SectorFE y y y y Obs. 384 384 384 384 R2 0.261 0.261 0.263 0.263 NumberofSectors 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualNon-UnionPremium: wagepremiumfornonunionizedworkersaftercontrollingforagevariables,sex, education,race,tenure,maritalstatus,stateandmetropolitan area. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsin thesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. lnNumExporters: lognumberofexportersinthesector;becauseofdataavailability,before2008,thenumberofexporters foreachsectoristhenumberforoverallmanufacturing. lnAvg. ExportValue: logaverageexportvalueperexporter withinthesector;becauseofdataavailability,before2008,the exportvalueandthenumberofexportersforeachsectorare settothevaluesforoverallmanufacturing. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1996-2019,across3-digit NAICSsectorswithinmanufac4t9uring. Standarderrorsare clusteredatthesectorlevel.

TableB4: UnionizationandNon-UnionWagePremiainManufacturing,ReweightedObservableSample (1) (2) (3) (4) (5) (6) (7) Variables ResidualUnionPremium t UnionShare t−1 0.119 0.056 0.049 0.013 0.045 0.022 0.004 (0.097) (0.070) (0.071) (0.065) (0.055) (0.065) (0.063) EmplShare 0.086 -0.152 -0.127 -0.028 -0.098 -0.005 -0.041 t (0.328) (0.256) (0.256) (0.182) (0.248) (0.180) (0.247) TopShare 0.323 0.314 0.338* 0.323 0.333* 0.340* t (0.197) (0.203) (0.171) (0.192) (0.164) (0.162) YoungShare 0.052 0.055 0.141 0.072 0.111 0.151 t (0.171) (0.179) (0.134) (0.148) (0.121) (0.118) GLabProd -0.096** -0.106** t (0.036) (0.038) Exp/Prod 0.132** 0.199** t (0.047) (0.077) Imp/Abs 0.040 -0.146 t (0.065) (0.089) CapInt 0.058** 0.038* (0.020) (0.021) Year y y y y y y y SectorFE y y y y y y y Obs. 464 464 464 464 464 464 464 R2 0.250 0.295 0.299 0.320 0.297 0.312 0.342 NumberofSectors 16 16 16 16 16 16 16 Source: BureauofLaborStatisticsandCensusBureau. ResidualNon-UnionPremium: wagepremiumfornon-unionizedworkersaftercontrollingforagevariables,sex,education,race,tenure,maritalstatus,stateandmetropolitan area. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofmanufacturingemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Exp/Prod: exportshareofproduction. Imp/Abs: importshareofdomesticabsorption. CapInt: logratioofcapitalstocktoshipments. lnNumExporters: lognumberofexportersinthesector;becauseofdataavailability, before2008,thenumberofexportersforeachsectoristhenumberforoverallmanufacturing. lnAvg. ExportValue: logaverageexportvalueperexporterwithinthesector;because ofdataavailability,before2008,theexportvalueandthenumberofexportersforeach sectoraresettothevaluesforoverallmanufacturing. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019,across3-digitNAICSsectorswithinmanufacturing,ontherespondentsampleonly,withobservationweightedbytheprobability ofresponse. Standarderrorsareclusteredatthesectorlevel. 50

TableB5: Sector-LevelRegressions: WagePremiaandLaggedUnionization, Economy-wideEffects (1) (2) (3) (4) (5) (6) ResidualWagePremium t Variables AllSectors SectorswithLargestDeclines UnionShare t−1 0.075 0.066 0.048 0.717 0.759* 0.749* (0.110) (0.110) (0.102) (0.478) (0.394) (0.398) EmplShare 0.236 0.307 0.315 -0.198 -0.033 -0.047 t (0.474) (0.712) (0.711) (0.430) (0.777) (0.755) TopShare -0.082 -0.097 -0.153 -0.148 t (0.372) (0.367) (0.462) (0.454) YoungShare -0.273 -0.315 0.020 -0.031 t (0.375) (0.350) (0.577) (0.556) GLabProd 0.168** 0.148* t (0.075) (0.068) Year y y y y y y SectorFE y y y y y y Obs. 440 440 440 239 239 239 R2 0.049 0.053 0.060 0.135 0.140 0.150 NumberofSectors 18 18 18 9 9 9 Source: BureauofLaborStatisticsandCensusBureau. ResidualWagePremium: wagepremiumaftercontrollingforagevariables,sex, education,race,tenure,maritalstatus,stateandmetropolitanarea. UnionShare t−1 : unionizedshareofemployment,lagged. EmplShare: sectorshareofprivateemployment. TopShare: shareoflarge(500+employees)firmsinthesector. YoungShare: shareofyoung(lessthan5yearsold)firmsinthesector. GLabProd: laborproductivity,growthrate. Legend: ∗∗∗ significantat1%,∗∗ at5%,∗ at10%. Notes: Sector-levelFEregressions,1990-2019. Thefirstthreecolumnsinclude all2-digitNAICSsectors;columns(3)-(6)includeonly2-digitNAICSsectors thathaveexperiencedabove-the-mediandeclinesintheirunionizationrates. Standarderrorsareclusteredatthesectorlevel. 51