此為英文文檔 nber -An Experimental Evaluation of Deferred Acceptance - Evidence from Over 100 Army Officer Labor Markets

NBERWORKINGPAPERSERIES

ANEXPERIMENTALEVALUATIONOFDEFERREDACCEPTANCE:EVIDENCEFROMOVER100ARMYOFFICERLABORMARKETS

JonathanM.V.Davis

KyleGreenberg

DamonJones

WorkingPaper31612

/papers/w31612

NATIONALBUREAUOFECONOMICRESEARCH

1050MassachusettsAvenue

Cambridge,MA02138

August2023

AllviewsexpressedinthismanuscriptarethoseoftheauthorsanddonotrepresenttheviewsorofficialpositionsoftheUnitedStatesMilitaryAcademy,theUnitedStatesArmy,ortheDepartmentofDefense.AndrewHooverandNinaKerkebaneprovidedexceptionalresearchassistance.WethankAlexRees-JonesandparticipantsatLERAatASSA,SOLE,NBERSILaborStudiesandPersonnelEconomicsprograms,andStanfordSITEforhelpfulcomments.AlthoughKyleGreenbergisalieutenantcolonelintheU.S.Army,noneoftheauthorsreceivedfinancialcompensationfromtheDepartmentofDefenseforthepurposeofcompletingthisproject.ThisresearchissupportedbyNSFGrantSES-2018246andaJ-PALWorkforceoftheFutureGrant.ThisRCTwasregisteredintheAmericanEconomicAssociationRegistryundertrialnumberAEARCTR-0004718.TheviewsexpressedhereinarethoseoftheauthorsanddonotnecessarilyreflecttheviewsoftheNationalBureauofEconomicResearch.

NBERworkingpapersarecirculatedfordiscussionandcommentpurposes.Theyhavenotbeenpeer-reviewedorbeensubjecttothereviewbytheNBERBoardofDirectorsthataccompaniesofficialNBERpublications.

?2023byJonathanM.V.Davis,KyleGreenberg,andDamonJones.Allrightsreserved.Shortsectionsoftext,nottoexceedtwoparagraphs,maybequotedwithoutexplicitpermissionprovidedthatfullcredit,including?notice,isgiventothesource.

AnExperimentalEvaluationofDeferredAcceptance:EvidencefromOver100ArmyOfficer

LaborMarkets

JonathanM.V.Davis,KyleGreenberg,andDamonJones

NBERWorkingPaperNo.31612

August2023

JELNo.D47,J01,M5

ABSTRACT

Wepresentevidencefromarandomizedtrialoftheimpactofmatchingworkerstojobsusingthedeferredacceptance(DA)algorithm.OursettingistheU.S.Army’sannualmany-to-onemarketplacethatmatches10,000officerstounits.Officersandjobsarepartitionedintoover100distinctmarkets,ourunitofrandomization.MatchingwithDAreducedofficers’attritionintheirfirstyearintheirnewmatchby16.7percent,butwecanruleoutmorethana10percentreductioninattritionbytheendoftheirsecondyear.MatchingwithDAhadprecisezeroeffectsonperformanceevaluationsandpromotions.AlthoughmatchingwithDAincreasedtruthfulpreferencereportingbyastatisticallysignificant10percent,manyofficersmatchedbyDAmisreporttheirtruepreferences.WepresentnewevidencesuggestingthatcommunicationandcoordinationofpreferencesmaylimitthebenefitsofDAinmatchingmarketswhereeachsideactivelyrankstheother.

JonathanM.V.Davis

UniversityofOregon

DepartmentofEconomics

516PLC

1415KincaidStreet

Eugene,OR97403

jdavis5@

DamonJones

HarrisSchoolofPublicPolicy

UniversityofChicago

1307E60thStreet

Chicago,IL60637

andNBER

damonjones@

KyleGreenberg

U.S.MilitaryAcademyatWestPoint

DepartmentofSocialSciences

606ThayerRoad

WestPoint,NY10996

kyle.greenberg@

Arandomizedcontrolledtrialsregistryentryisavailableat

/trials/4718

1

1Introduction

Overthelastthreedecades,insightsfrommarketdesignhavehadasubstantialinflu-enceonmatchingmarketswithoutprices.Resultsfromthisfieldhavebeenusedtohelpredesignmechanismstomatchnewdoctorstohospitals(

RothandPeranson

,

1999

),as-signstudentstoschoolsinseverallargeschooldistricts(

Abdulkadiro

luandSnmez,

2003

;

Abdulkadiro

luetal.,

2009

;

PathakandS

nmez,

2008

),matchnewlycommissionedArmyofficerstomilitaryoccupations(

S

nmez,

2013

;

S

nmezandSwitzer,

2013

;

Green-

bergetal

.,

2023

),andmatchkidneydonorstorecipients(

Rothetal.

,

2003

).Inaddi-tion,manyorganizationshaveadoptedtoolsfrommarketdesigntofacilitateinternalworker-to-divisionmatchingmarkets(

BarronandVardy

,

2005

;

CowgillandKoning

,

2018

;

Cowgilletal

.,

2022

;

Davis

,

2022

).

Stability,thematchpropertyinwhichnounmatchedpairofagentspreferbeingmatchedtogetherovertheirassignedmatch,isconsideredacrucialcharacteristicofsuccessfulmar-ketdesigninterventions(

Roth

,

1984

,

1990

,

1991

;

RothandXing

,

1994

;

RothandPeranson

,

1999

;

KagelandRoth

,

2000

).Whenamatchisunstable,someagentshave“justifiedenvy”whichcreatesopportunitiestodeviatefromtheassignedmatch.Thedeferredacceptance(DA)algorithmisthesimplestandmostwidelyusedexampleofastablematchingal-gorithm(

GaleandShapley

,

1962

;

Roth

,

2008

).MarketdesignershavealsoadvocatedforDAbecauseitisbothstrategyproof—thatis,truthfulreportingofrankingsisadominantstrategy—andityieldstheoptimalstablematchfortheproposingsideofthemarket(

Ab-

dulkadiro

luandSnmez,

2003

;

Abdulkadiro

luetal.,

2006

;

ChenandS

nmez,

2006

;

PathakandS

nmez,

2008

).However,thesetheoreticalbenefitsmightnotberealizedinpractice.Forexample,ifparticipantsdonotunderstandortrustthatthealgorithmisstrat-egyproof,theymaystillmisreporttheirpreferences(

Rees-Jones

,

2017

).Ortheymaytrytoreducetheiruncertaintyabouttheirmatchbyenteringintoinformalagreementswithagentsontheothersideofthemarketto“rankeachotherfirst”(

RothandXing

,

1994

).

Thispaperprovidesevidencefromarandomizedcontrolledtrialevaluatingtheimpacts

2

ofmatchingworkerstojobsusingtheDAalgorithm.OursettingistheinternalmarketwithintheUnitedStatesArmywhereofficersarematchedtounits.OfficersgenerallyrotateunitswithintheArmyeverythreeyears.TheArmy’sHumanResourcesCommand(HRC)coordinatesthisreorganizationinanannualmatchingmarketthatincludesover14,000officerstobematchedacrossroughly500units.Officersandpotentialpositionsatunitsaresegmentedintodisjointmarketsbasedontheirrankandmilitaryoccupation.Werandomlyassignedasubsetofmarketstoatreatmentgroupwhereofficersandunitswerematchedusinganofficer-proposingDAalgorithm.TheremainderareassignedtoacontrolgroupwhereofficersandunitswerematchedaccordingtotheArmy’straditionalprocess,whichinvolvescareermanagersmanuallymakingmatchdecisionsusingofficerandunitpreferenceswithouttheaidofanyspecificalgorithm.

Oursettingisidealfortworeasons.First,ourexperimentalsampleincludes115disjointmarkets.Thisprovidesalargeenoughsampletohavecomparabletreatmentandcontrolgroupsandthestatisticalpowertopreciselyestimatetreatmenteffectsofinterestatthemarketlevel.Second,theArmycollectshigh-qualitydataonparticipants’preferencesandrelevantmarketoutcomes.Weareabletolinkthesedatatosurveysthataskpartici-pantsaboutstrategicpreferencemanipulationsandtheirsatisfactionwiththeirassignedmatches.Wealsolinktoadministrativeperformanceandretentiondata.Together,thesefeaturesprovideuswiththeuniqueopportunitytocrediblymeasuretheimpactofDAonimmediateandlonger-runmarketoutcomes.

Forourmainoutcomes,weselectedandpre-registeredproxiesforwell-knowngoalsofmarketdesignmechanisms:retentionwithintheArmy,whichproxiesforofficersatis-faction;officers’evaluationreportsandpromotions,whichmeasureofficerperformance;andsurveyevidenceontruthfulpreferencereporting,whichreflectswhetherDAisstrate-gyproofforofficers.Overall,wefindlittleevidencethatDAleadstomeaningfulimprove-mentsinretention,performanceevaluations,orpromotionstwoyearsintotheirnewpo-sition.AndwhilewefindevidencethatDAincreasestruthfulpreferencereporting,effectsizesaremodestandmostofficersinDAmarketsadmittomisreportingpreferences.

3

Theofficer-proposingDAyieldstheofficeroptimalstablematch,whichmeansthatallofficerspreferthematchselectedtoallotherstablematches.Consistentwiththis,match-ingwithDAinoursettingreducesattritioninthefirstpost-matchyearbyastatisticallysignificant1.1percentagepoints(p=0.03),a16.7percentreductionrelativetothecontrolgroup’sattritionrate.However,thisissensitivetohowweadjustformultiplehypothesistestingandbytwoyearsoutmatchingwithDAonlyreducesattritionbyastatisticallyinsignificant0.3percentagepoint(pp).Ourresultsattwoyearsarepreciseenoughtoruleouta1.5ppreductioninattrition,lessthanone-tenthoftwo-yearattritioninthecontrolgroup(15.1percent).

TotheextentthatmatchqualitycanbeimprovedbyDA,wemightexpectimprove-mentsinproductivityandpromotion.However,matchingwithDAhasprecisezeroeffectsontheprobabilitythatanofficerreceivesthehighestpossibleperformanceeval-uation.

1

Roughlyhalfofofficersinthecontrolgroupreceivethehighestpossibleper-formanceevaluationinthefirstandsecondyearafterstartingtheirnewpositions.OurestimatessuggestthatmatchingwithDAhaslittleimpactonthisperformanceoutcome,withconfidenceintervalsrangingfroma2.2ppreductionuptoa1.0ppincreaseinthefirstyearandfroma3.2ppreductionuptoa1.2ppincreaseinthesecondyear.Wefindsimilarlyprecisezeroeffectsonthelikelihoodthatanofficerispromotedtothenextrankandonofficers’promotionboardpercentilerankingrelativetootherofficersconsideredforpromotion.

AnotheradvantageofDAisthatitisstrategyprooffortheproposingsideofthemarket,meaningthattruthfulpreferencereportingisadominantstrategyforofficers.Wefindmoderateevidenceinsupportofthisprediction.Inasurveyadministeredthreeweeksbeforethemarketplaceclosed,officersinDAmarketsare2.4ppmorelikelytoreportal-wayssubmittingtheirtruepreferences(p=0.001),a10percentincreaseonthe24percentofofficersincontrol.OfficersinDAmarketsarealso1.2ppmorelikelytostatethatthey

1Officers’evaluationreportshaveasignificantimpactonwhetheranofficerispromotedtothenextrank.Weonlyobservepromotionoutcomesforthesubsetofofficersinoursamplewhowereconsideredforpromotionwithintwoyearspost-match.

haveaccuratelyreportedtheirtopchoice(p=0.061),relativetothe84percentofofficerswhoindicatedtruthfulreportingoftheirtopchoiceincontrolgroupmarkets.However,inapost-marketsurveyadministeredwhenofficerslearnedoftheirmatches,only69per-centofofficersstatedthattheytruthfullyreportedtheirtopchoice,withnostatisticallysignificantdifferencebetweenthetreatmentandcontrolmarkets.

SowhatexplainstherelativelylimitedeffectsofDAonretention,performance,andtruthfulpreferencereporting?Onepossibilityisthatstrategiccross-marketcommuni-cationbetweenofficersandunitsmaylimitthebenefitsofhavingastrategyproofmecha-nism.InbothDAandcontrolmarkets,roughly45%ofmatchesare“first-to-first”pairings—thatis,amatchwhereanofficerhasrankedajoblistingastheirfirstchoiceandwheretheunithasrankedthesameofficerastheirfirstchoiceforthatparticularlisting.Thisistrue,eventhoughofficers’andunits’exactrank-orderedpreferencesarehidden.Wederiveatestforthepresenceofstrategiccommunicationbyleveragingcaseswhereunitshavemultiple,nearlyidenticaljobpostings.Inthesecases,wecansimulatetherateoffirst-to-firstmatchesunderthenullhypothesisofnocoordination.Thehighobservedrateofsuchpairingsfarexceedswhatwewouldexpectunderthenull,andbecausewefocusonsetsofidenticaljobsatthesameunit,thepatternsarenotsimplytheresultofhighlycorrelatedpreferencesbetweenofficersandunits.WefurthershowthatthispatternisnotsimplyanartifactoftheinitialrolloutofDA:thesamepatternsprevailtwoyearsafterourstudy,bywhichtimeDAhadbeenusedinallmarketsbytheArmyandmessagingofDA’spropertieshadimproved.

2

Ifcareermanagerstypicallyhonorfirst-to-firstpairingsbecauseoftheeasewithwhichtheycanbeidentifiedwithoutthehelpofacomputeralgorithm,thentheywillachievemanyofthesamematchesasaDAalgorithm.However,ifsomefirst-to-firstmatchesinDAmarketsarestrategicanddifferfromthetruepreferencesofofficersandunits,thebenefitsofswitchingtoDAmaybeattenuated,akintotheprocessdescribedby

Rothand

2InAppendixA,weshow,usingasimpletheoreticalexample,thatofficersmaybenefitfromstrategic

communicationandmisreportingtheirpreferencesifunits’preferencesareresponsivetothiscommunica-

tion(Antler,2015).

4

5

Xing

(

1994

)as“stage4unraveling.”Surveyevidencecorroboratesthatofficersfrequentlydeviatefromtheirtruepreferencestoachieveafirst-to-firstpairing.

Theselimitationshaveimplicationsforothersettingswhereproximityandrepeatedin-teractionbetweenthoseoneithersideofthemarketarenon-trivial,suchasinotherinter-nallabormarketsoreventheNationalResidentMatchingProgram(NRMP)formedicaldoctors.Infact,studiesofphysiciansandresidencyprogramdirectorsreportfrequentcommunicationaboutrankingsafterinterviews,eventhoughthistypeofcommunica-tionisprohibitedbytheNRMPcodeofconduct(

Andersonetal.

,

1999

;

Careketal.

,

2000

;

Teichmanetal.

,

2000

;

Sbiccaetal.

,

2010

;

Berriochoaetal.

,

2018

).

AnalternativeexplanationforthelimitedeffectsofDAfoundinoursettingisthattheofficersmaynothaveunderstoodthattruthfulpreferencereportingisaweaklydominantstrategy(

ChenandS

nmez,

2002

;

Hassidimetal.

,

2017

;

Rees-Jones

,

2018

;

Rees-Jonesand

Skowronek

,

2018

).Althoughcareermanagersknewwhethertheofficerstheymanagedwerepartoftreatmentorcontrol,wewerenotabletoobservetheirexactmessagingtoofficersinDAmarkets,whichcouldresultinconsiderableheterogeneityinawarenessofDA’sproperties.However,wedonotbelievethatthiscompletelyexplainsourresults.Insubsequentyears,asDAwasrolledouttotheentiremarketplaceandwhenwehavebet-terinformationhowDA’spropertieswerecommunicatedtoofficers,weobservesimilarpatternsin1-to-1matchingandself-reportedtruthfulness.

Thispapercontributestoseveralliteratures.First,wecontributetotheliteratureonmar-ketdesignbyprovidingthefirstevidencefromarandomizedtrialontheimpactofmatch-ingworkerstojobsusingDA.Despitetherapidlygrowingadoptionofmarketdesigntoolsoverthepastthreedecades,thereislittleempiricalevidenceoftheimpactoflabormarketdesignonmarketandmatchoutcomes.Theleadingstudiestodateusetimeseriescomparisonswithinasinglemarket(

NiederleandRoth

,

2003b

),cross-sectionalcompar-isonsbetweenmarkets(

NiederleandRoth

,

2003a

),ordifference-in-differencemethods(

Davis

,

2022

).Onechallengeisthatexperimentalorquasi-experimentalvariationinmech-anismsacrossmultiple,comparablemarketsisuncommon.Whatismuchmorecommon

6

isaone-timeconversiontoanewmechanismwithinasinglemarket(

RothandPeranson

,

1999

;

Abdulkadirog?luetal.

,

2005a

,

b

).Counterfactualoutcomesatthemarketlevelarethereforedifficulttoestimatewhenthereisonlyonemarketunderobservation.

Ourpreferenceandsurveydataallowustoanalyzehowmatchingmechanismsdeviatefromtheoreticalidealswhenimplementedinpractice.Recentresearchsuggeststhatdoc-torsmisreporttheirtruepreferencesintheincentive-compatibleNRMP(

Hassidimetal.

,

2017

;

Rees-Jones

,

2018

;

Rees-JonesandSkowronek

,

2018

).Inoursetting,wecantakead-vantageofthepresenceofidenticalpostingswithinaunittotestforstrategicbehaviorbasedonsubmittedrankings,whichcomplementsevidencefromself-reportedsurveys.Relatedly,

Echeniqueetal

.(

2022

)positthatdoctors’preferencesforhospitalsdeviatefromthetruthinpartbecausedoctorscanonlypreferencehospitalstheyinterviewwith.

3

Ourresultssuggestthatparticipantsintwo-sidedmarketsmaystrategicallycommunicate(of-tenuntruthful)preferencesinordertoobtainhigherrankingsfromparticipantsontheothersideofthemarket,apossibilitythatisconsistentwithevidencefromseveralsurveysthattypicallyconsistofafewhundreddoctorsandresidencyprogramdirectors(

Ander-

sonetal

.,

1999

;

Careketal.

,

2000

;

Teichmanetal.

,

2000

;

Sbiccaetal.

,

2010

,

2012

;

Berriochoa

etal.

,

2018

).Wecontributetopriorevidenceusingasurveythatisadministeredtoallof-ficersparticipatinginthemarketplaceandthathasahigherresponseratethanistypicalinsurveysofdoctorsintheNRMP.

Second,ourstudycontributestoorganizationalandpersonneleconomicsbyprovidingnewevidenceonthetrade-offsofdifferentassignmentmechanismswithinorganizations.Marketdesigntoolsaredesignedtoproducematcheswithcertainproperties,suchasincentive-compatibility,stability,transparency,andstrategicsimplicity,butorganizationsthatorganizethesemarketsandparticipantsinthemarketmayhaveothercompetingob-jectives(

Cowgilletal

.,

2022

).Forexample,

Haegele

(

2021

)showsmangers’incentivestohoardtalentontheirteamscreatesamisallocationofworkerstojobswithinfirms.Our

3Inthecontextofcentralizedschoolchoicemarketswithstrategy-proofassignmentmechanisms,

Arteagaetal

.(

2022

)findthatbeliefsaboutadmissionsprobabilitiesinfluencechoicesbyshapinghowap-plicantssearchforschools.

7

studyoffersauniqueopportunitytoestimateimprovementsinofficersatisfactionduetoDA,asmeasuredbyretentionintheArmy,againstanychangesinofficerperformanceevaluations,whichwebroadlyinterpretasaproxyformoregeneralorganizationalob-jectives.Wefocusprimarilyonhorizontaljobchangeswithinanorganization.

Huitfeldt

etal.

(

2023

)studytheverticalstructureoflabormarketswithinafirm.Relatedly,

Benson

etal.

(

2019

)studyhowfirmsmakepromotiondecisionsandshowthattheymaypriori-tizecurrentperformanceattheexpenseofothercharacteristicsthataremorepredictiveofmanagerialtalent.

Third,thisresearchaddstotheliteratureonpersonnelconsiderationswithinmilitaryor-ganizations.Severalpapershavestudiedtheassignmentofcadetstobranchesofthemil-itary(

So¨nmez

,

2013

;

So¨nmezandSwitzer

,

2013

;

Schlegel

,

2015

;

Jagadeesan

,

2019

;

Green-

bergetal

.,

2023

).

Lewisetal.

(

2022

)studytheassignmentofCoastGuardservicementoshipsandrecommendfocusingontheassignmentsofofficers.

Greenbergetal

.(

2022

)studythelong-termimpactsofvoluntaryenlistmentintheU.S.Army,

Bruhnetal.

(

2023

)studytheeffectsofcombatdeployments,andseveralothersstudythelong-termeffectsofcompulsorymilitaryserviceintheU.S.andelsewhere(e.g.,

Angrist

,

1990

;

Boundand

Turner

,

2002

;

BedardandDesche?nes

,

2006

;

Angristetal

.,

2010

;

AngristandChen

,

2011

;

Angristetal

.,

2011

;

CardandCardoso

,

2012

;

Bingleyetal

.,

2020

).Moregenerally,ourpa-percontributestothebroadliteratureonexperimentsinlaboreconomics(

Hortonetal.

,

2011

;

CharnessandKuhn

,

2011

;

ListandRasul

,

2011

).Ourstudyisuniqueinthatwerandomizeentiremarketstounderstandhowdifferentmarketclearingrulesaffectout-comesandwefocusontheimpactofdifferentmatchingmechanismsusingexperimentalvariation.

8

2InstitutionalDetailsoftheInternalMatchingMarket

Since2017,theU.S.Armyhasusedanonline,interactivemoduletomatchmostofficers topositionsatArmyunitswithinaninternallabormarket.

4

Theonlinemarketplace,knownastheArmyTalentAlignmentProcess(ATAP),

5

allowsofficerstobuildprofiles thatunitscansee(andvice-versa),

6

permitsofficerswhoarescheduledtochangeassign-mentswithin6to9monthstosubmitpreferencesoveravailablejobs,andallowsunits tosubmitpreferencesoverofficersexpectedtomove.TheArmy’shumanresourcesdi-vision,knownastheHumanResourcesCommand(HRC),managesthemarketplaceandpartitionsallofficersandjobswithinthemarketplaceintodistinctmarkets.Eachmar-ketisdefinedbyacombinationofofficerrankandofficeroccupation(e.g.amarketfor“infantrycaptains”andaseparatemarketfor“militaryintelligencemajors”).

Inadditiontobeingassociatedwithaspecificrankandaspecificoccupation,eachjobalsobelongstooneofroughly500differentArmyunits.Althoughthedistinctmarketswithinthemarketplacearemany-to-oneinthesensethatmultipleofficerscanmatchtoasingleunit,inpracticeofficerssubmitpreferencesoverspecificjobsataunitaspartofaone-to-onemarket.Forexample,ifamarkethas10unitsthateachhave5distinctjobswithinthemarket,theneachofficerinthatmarketcanrankupto50jobs.Unitsprovidedescriptionsforeachjoblistedintheonlinemarketplace,andjobdescriptionsmayvarywhenamarketcontainsmultiplejoblistingsthatbelongtothesameunit.Unitswithmultiplejoblistingsinthesamemarketmustsubmitseparaterank-orderlistsofofficers

4ThisinternallabormarketdoesnotincludenewSecondLieutenants,whoreceivetheirinitialassign-mentthroughtheirrespectivesourcesofcommission(e.g.OfficerCandidatesSchool,theReserveOfficerTrainingCorps,ortheUnitedStatesMilitaryAcademy).

5Themarketplaceisalsofrequentlyreferredtobythenameoftheonlineplatform,theAssignmentInteractiveModule—version2.0(AIM2).

6OfficerprofilesincludeallinformationonthestandardOfficerRecordBrief(includingassignmenthis-tory,civilianeducationdegreeinformation,militaryeducation,andmilitaryawards)andadditionalself-reporteddetails,includingpreviouscivilianandmilitaryemploymentandeducation,professionalskillsandcertifications,culturalexperiences,andtravel.Unitscanprovidespecificdescriptionsforeachjobinthemarketplace,contactinformationforthejob’scurrentincumbentorpointofcontact,andgeneralinfor-mationabouttheirunit.

9

foreachlisting,andthesepreferencesneednotbeidenticalacrosslistings,evenwhensuchjoblistingshaveidenticaldescriptions.

7

Eachcycleoftheonlinemarketplaceisopenfor6to8weeks,duringwhichtimeofficersmaysubmitpreferencesforalljobswithintheir(rank-by-occupation)market.

8

Officerscanadjusttheirpreferencesforjobsatanytimewhilethemarketplaceisopen.Likewise,unitsmaysubmitpreferencesoverofficersandcanchangetheirpreferencesatanytime.Officersarenotrequiredtorank-orderalljobsintheirmarketandjobsarenotrequiredtorank-orderallofficers.Thefinalversionoftheirpreferencelistsatthemarket’sscheduledclosingdateareusedtodetermine(forDA)orinform(forcontrolmarkets)matches.

Officers’exactpreferencesoverjobsandunits’exactpreferencesoverofficersarehiddenfromeachother.However,unitscanobserveasignalifanofficerranksoneoftheunit’sjobsamongtheofficer’stop10percentofallpossiblechoices.Forexample,ifamarkethas200jobs,thentheofficerinterestsignalwillappearnextto20jobsregardlessofhowmanyjobstheofficerleavesunranked.Thissignalisessentiallycostlessbecauseofficerscanchangewhichjobstheylistintheirtop10percentofchoicesatanytime,andonlypreferencessubmittedatthetimethemarketplaceclosesarerelevanttoeventualmatches.Ontheothersideofthemarket,officersobserveasignalifaunitranksthemanywhereontheirrank-orderedlistforajobintheofficer’smarket.Officersandunitsarepermittedtoconductinformalinterviewsandtocommunicateoutsideoftheonlinemarketplace,butthereisnostrictrequirementtodoso.

9

Relatedly,officerscansubmitpreferencesoveralljobsintheirmarketregardlessofwhethertheyhaveinterviewedwithanyoftheunits(andvice-versa).

CareermanagersatHRCareresponsibleforclearingmarketsbymatchingofficersto

7Inwhatfollows,weoftenusetheterms“ajob’spreferencesoverofficers”and“aunit’spreferencesoverofficers”interchangeably.

8Asapracticalmatter,thereislittlevariationinwagesforjobswithinthesamemarketasmilitarybasepayisafunctionofanofficer’smilitaryrankandyearsofservice.Officersreceiveahousingallowancethatvariesaccordingtolocalhousingpricesnearthebaseanofficerisassigned.Officersassignedtobaseswithhighcostsoflivingmayalsoreceiveanadditionalcostoflivingallowance.

9TheArmy’sonlineplatformdoesnothaveafunctionalitythatallowsofficersandunitstorequestandscheduleinterviews.Assuch,weareunabletoobservewhichofficersinterviewedwithwhichunits.

10

jobswithindistinctmarkets.ThesemanagersareofficerswhoservetwotothreeyearsatHRC,beforetypicallyreturningtoanon-HRCpositionwithintheirnormalmilitaryoccupation.Aftercareermanagersclearadistinctmarket,theyplaceofficersonorderstomovetotheirassignedunitsinthecomingmonths.Between6and9monthsafterthemarketplacecloses,officersreporttoanewunit.Dependingonthetimingoftheircontracts,officerscandecidenottorenewandexittheArmyiftheyareunhappywiththematch.Inparticular,somemayexitbeforetheybegintheirnextassignment.

10

3ExperimentalDesign

Drawingonpriorresearchontheimpactsofthedeferredacceptancealgorithmonmatchoutcomes(

Davis

,

2022

)andpersonneleconomicswithinthecontextofthemilitary(

Green-

bergetal

.,

2022

;

Bruhnetal.

,