EstimatingHospitaThisworkwasfundedbytheDeHealthServicesResearchandconclusionsoftheauthBostonVAHealthcareSystem•Phone:(857)364-6058lCostShiftRates:APractitioners’GuideAustinFrakt,,VeteransHealthAdministration,&&Economics,150SouthHuntingtonAvenue,MailStop152H,Boston,MA02130•Fax:(857)364-4511•Web:
EstimatingHospitalCostShiftRates:APractitioners’GuideAcompanionto“Howmuchdohospitalscostshift?Areviewoftheevidence,”byAustinFrakt,(TheMilbankQuarterly89(1),2011).Aworkingpaperversionisavailableat
(2011),whichisanon-technicalreviewoftheacademicliteratureonhospitalcostshiftingoverthelast15years(since1996).Thepurposeofthispaperistoserveasaguidetoresearchersconsideringestimatingcostshiftratesorwishingamoredetailed,,thoughthispapercoverssomeofthesamematerialasFrakt(2011),itdoesnotprovideanypolicyorhistoricalcontextorbackground,,IdonotdrawanypolicyconclusionsfromtheliteraturesurveyedbelowandencouragereaderstoconsultFrakt(2011).Infact,IassumethereaderhasreadFrakt(2011)andisthereforefamiliarwiththeterm“costshifting”andhowitdiffersfrompricediscrimination,,Iuseeconomicandeconometrictermsfreelyinthisdocument,(reviewedinSection3),,“public”and“private.”,(formcontractingnetworks)(GlazerandMcGuire2002).3
Thisdistinctionhighlightstheroleof“excludability”—whichcanbehospitals,physiciangroups,nursinghomes,oranyfirmprovidinghealthservices—inmarketsforwhichgoodsubstitutesexistaremoreeasilyexcludedfrominsurers’,highprestigeor“musthave”,ahospitalwithalocalmonopoly(due,say,toalargedistancetotheclosestcompetitor)isnotexcludablefrominsurers’networks,(Ho2009).,aphenomenonthatpushesinsurers’inputprices(providerprices)downward,thoughisambiguouswithrespecttooutputprices(premiums)(Dafny,Duggan,Ramanarayanan2009).Putanotherway,ifinsurers’demandforproviderservicesisperfectlyelastic(demandcurveishorizontal),,thereisnoscopeforcostshifting(Morrisey1996).Thus,thesizeofstaticpricemarkups(ordegreeofpricediscrimination)(Ginsburg2003).Therefore,ontheoreticalgrounds,,,,,theycanonlygodownifpricesareincreased(Morrisey1996).4
Mosteconomistsrejectthepossibilityofcostshiftingbyappealingtoaprofitmaximizationhypothesis(MorriseyandCawley2008).Showalter(1997),,asimpleresultofsupplyanddemand:,,theprovidermustloweritsperpatientprivateprice,,inresponsetolowerpublicpayments,profitmaximizationpredictsavolumeshift(lowerpublicvolume,higherprivatevolume)andapricespillover(lowerprivatepaymentsaswell).Thisistheantithesisofthecostshiftingtheory(;McGuireandPauly1991).Morrisey(1993,1994,1996)’,it’splausiblethatcausalityrunstheotherway—(2002),,,publicprograms“freeride”,Medicareismotivatedtosetpriceslow—solowthat,wereMedicaretheonlypayer,theresultingqualitywouldbesociallyinefficient—andthentorelyontheprivatesectorto“repair”(2002)modelisthatthedegreetowhichaprofit-maximizingproviderrespondstoMedicarepaymentchangesisafunctionofitspublic/,(2009)characterizesthe“reversecausality”storyofGlazerandMcGuire(thatpublicpricesrespondtoprivateones)asa“strategy”,shedubsthemorestandardstory—thathospitalswithunexploitedmarketpower—asthe“marketpower”
hypothesis,,thestrategyhypothesissuggeststhathospitalswithalargershareofprivatepatientswouldcostshiftlessbecausetheyarelesssensitiveto(lessrelianton),Gaumer,andMiller(2010),allowingthemtorise.(Alternatively,itcouldbethehighcoststructureisitselfafactorinhighmarketpower,perhapsduetohighquality.),,iftheydonotfullyexploittheirmarketpower,theorysuggeststhatthescopeforcostshiftingisstillrelatedtotheirdegreeofmarketpower,aswellascostsandquality,public/privatepayermix,,:,inamodelwithprivatepaymentasthedependentvariable,’vealreadytouchedontheimplicationsforcostshiftingifprovidersdonotmaximizeprofit,-fivepercentofbedsincommunityhospitalsareinnon-profitinstitutions(Ginsburg2003).,forexample,,thereisnoscopeforcostshifting(Morrisey1993,1994,1996).IntheremainderofthissectionIconsidertheoppositecaseinwhichsuchhospitalsdonotmaximizeprofit,(1997/1998),citingpriorworkinagencytheory,(1988),providedthehospital6
,whichisacomponentofthehospital’,Melnick,andBamezai(2000)developatheoreticalmodelalsosimilartothatofDranove(1988).Thelatterincludesfixedaveragecostsacrosspayers,,providersthatmaximizeutilitythatdependsonprofitandvolumemayormaynotcostshift,,,Li,andFriesner(2000)(highprivatepricesandlowerprivatevolume)ortheopposite(lowerprivatepricesandhigherprivatevolume)’(2002)provideasimilaramodelofhospitalprestigemaximization(maximizingrevenuesubjecttotheconstraintthatitatleastmeetscosts).(1998)providesanintuitive,,,,costsvarybypayerbecausepublicprogramsandprivateinsurerscoverdifferentpopulationswithdifferentneeds(Morrisey1993,1996).Costs7
,costshiftinganalysisbasedonmargin(revenuedividedbycost),theliteratureoncostshiftingassumingutility(notmerelyprofit) αmit+ßtxit+γt + δi+ εit(1)8
wheremitisthepublicpayerprice(MedicareorMedicaidoravectorofboth),γtand δi,areyearandproviderfixedeffects,respectively,andε(1)wouldincludeinteractionsbetweenmitandothertermsand/ßtispotentiallytimevaryingisimportant,asdiscussedbelowItisessentialtobepreciseaboutwhatImeanby“price,”’slistpriceorchargesasthosearenotrepresentativeofwhattheproviderisactuallypaid(Rosenman,Li,andFriesner2000).Instead,(1)isamodelofprice,notofprice-to-costratio(ormargin)orpricelesscost(profit).Sinceapossibleresponsetochangesinpublicpriceisachangeincost,(1).-sectionalstudiesonlyrevealevidenceofcostshiftingunderanassumptionthatgeographicalvariation(orvariationacrossproviders),cross-sectionalstudiesidentifyaneffectbyexploitingvariationinpricediscrimination,whichisastaticphenomenon,notcostshifting,(δi)-invariantfactors,soissimilartoafixedeffectsspecification(Wooldridge2002).SuchamodelisfoundbytakingthedifferenceofEquation(1)attimetwithitselfattimet-1sothat∆pi=α∆mi+ßt∆xi + ∆ßxit-1+∆γ + ∆εi(2)where ∆ is the first temporal difference operator. All time-invariant effects difference out, but because the coefficientsßtarepotentiallytimevarying,atermlinearinlevels(asopposedtochanges)ßorxistimeinvariant(Wu(2009)isanotableexception).9
,,toobtainanestimateofthecausaleffectofpublicpricesonprivateonesusingEquations(1)or(2)(IV)approachinstrongerstudies,,,Melnick,andBamezai(2000),(;Showalter1997;Rosenman,Li,andFriesner2000).Rosenman,Li,andFriesner(2000)analyze1995CaliforniaprimarycareclinicdatawithrelativelysophisticatedmethodologicallythatconsidersthepotentialendogeneityofMedicarepricesandtheroleofgovernmentgrantsinaseeminglyunrelatedregression(SUR),becausethedataarefromasingleyear,—andthatitismitigatedbygovernmentgrants—,,Showalter(1997)investigatespricediscriminationinastudybasedon1983-1985cross-sectionalPhysicians’(OLS)modelsofphysicianfeesandMedicaidvolumewithMedicaidreimbursementsasthekeyindependentvariable,hefindsevidenceconsistentwithprofitmaximizingbehaviorbyphysicians,.(1996)studiedtheeffectofreductionsinMedicarephysicianpaymentratesmandatedbytheOmnibusBudgetReconciliationActsof1989and1990usingafixedeffectsspecificationwithmarket-yearasthe10
—privatebilledchargesandprivatebilledchargeslesspaymentrates(called“excesscharges”)—wereestimatedwithMedicarepaymentrateasthekeyindependentvariable,controllingfornursewagelevels,providerdensity,HMOmembershiprate,,%foreach10%,-1996FloridaMedicaidnursinghomefinancialdataandOnlineSurveyCertificationandReporting(OSCAR)data,Troyer(2002)-daysbypayer,wageandpriceindices,numberofbeds,casemix,percentelderly,profitandownershipstatus,measuresofquality,,,nursinghomearecompensatingfortheriskofresidents’::accordingtocross-sectional,fixed-effects,,two,andfourstudiesofthesetypes,respectively,,,qualitatively,withoneexception,(1998),forthe11
(Wu2009)findsanaverage21%(GowrisankaranandTown1997),hospitalcostreportdatafromtheHealthCareFinancingAdministration(nowtheCentersforMedicareandMedicaidServices),andAmericanHospitalAssociationdata(allfrom1991),theauthorsestimatebygeneralmethodofmoments(GMM)-andnon-profitscompeteandmaximizedifferentobjectivefunctions,havedifferentpreferencesforinvestment,-profitsmaximizeprofitswhilenon-profitsmaximizeamixofprofitsandquality,wherequalityisimplementedasareducedformabstractionbutthoughtofascharacterizedbylevelsofphysicalandhumancapital,,exit,investment,andmulti-payerpricingdecisions,,incomethresholdforfreecare,co-payment,Medicaredeductible,Medicarereimbursementrate,corporatetaxrates,’%reductioninqualityanda1%-profits,-SectionalStudiesStensland,Gaumer,andMiller(2010),,promotedbythehospitalandinsuranceindustriesorconsultingfirmsontheirbehalf(PWC2009,FoxandPickering2008;),
:hospitalswithstrongmarketpowerandaprofitablepayermixhavestrongfinancialresources,highcosts,,strictlyspeakingStansland,Gaumer,andMilleronlytestsstaticversionsofthem,thatis,(weakly),,hospitalswithhighermarketpowerhadhighercosts,lowerMedicaremargins,,,theydonotfindevidenceofdynamiccostshifting,indeed,theynevertestforit(though,tobefair,nordotheindustry-fundedstudiestheauthorsaimtorefute).Dobson,DaVanzo,andSen(2006),,theyexploityear2000statevariationinpayment-to-costmarginsforprivatepayers,relatingthemtovariationsinMedicare,Medicaid,anduncompensatedcaremargins,,theiranalysisdoesn’,theresultsconfoundpricewithcosteffects,,manycostshiftingstudiesfocusontheCaliforniamarket,,beginningwithZwanziger,Melnick,andBamezai(2000),(1982)
ofthestudyperiod(1990),over80%,,duringthe1980sMedicareandMedicaidreimbursementstoCaliforniahospitalsfellrelativetocosts(DranoveandWhite1998).Zwanziger,Melnick,andBamezai(2000),measuresofhospitalcompetition,ownershipstatus,averagecost,,profitstatus,andtimeperiod(1983-1985,1986-1988,1989-1991).—bothfor-andnon-profit—,—asisthefor-profitcostshiftingfinding—andnotconsistentwiththefindingsofotherstudies,(eachhospital’scostrelativetoaveragehospitalcostscomputedoverthestateandoverthehospital’smarket)(2006)isafollow-upstudyinwhichtheauthorsimplementasimilarfixed-effectsspecification,,Melnick,andBamezai(2000):,aslightlydifferentsetofcontrolsareapplied:averagecosts(instrumented,asdescribedabove),levelofhospitalcompetition(HHI),:(1)Californiahospitalpricecompetitionincreasedoverthe1990s;and(2)theBalanced14
BudgetAct(BBA),(2006)are,inmanyways,-andnon-profithospitals,nodifferencebeforeandaftertheBBA,%decreaseinMedicare(Medicaid)%(%),overthe1997-2001period,%,,California-wide,log-logmodelofprivaterevenue-costratioasafunctionofMedicareandMedicaidrevenue-costratios,,(1997/1998)examinedtherelationshipbetweenprivaterevenue-costmarginsandMedicareandMedicaidmarginsduringthreeCaliforniafiscalyears(1985-1986,1988-1989,1991-1992)relativetoabaselineyear(1982-1983).UsingCaliforniaOfficeofStatewideHealthPlanningandDevelopmenthospitaldischargedata,(notlogarithmically),’stotalmargin,ameasureofotherrevenue,anhistoricalaverageofassetvalue,hospitalcompetition,HMOmarketstrength,privateoccupancyrate,servicemix,profitandownershipstatus,,,becausethemodelisofmarginsandnotpayment,onecannotseparatelyidentifyeffectsonpaymentand15
,(1998)alsoexaminechangesinprivateprice-costmargins,aswellasinservicelevelsandhospitalclosings,(1997/1998),however,,theytakeadifferentapproachbasedonthenotionthatifhospitalscanshiftcosts,,(numberofservicesperday,controllingforDRG).Using1983and1992CaliforniaOfficeofStatewideHealthPlanningandDevelopmenthospitaldischargedata,DranoveandWhiteestimatehospital-levelOLS,SUR,andlogit(forclosings)modelsoftheeffectofMedicareandMedicaidcaseloads(proportionsofbilledcharges)onchangesinprivatemargins,servicelevelstoMedicare,Medicaid,orprivatepatients(threedifferentequations),andhospitalclosings,controllingforhospitalcompetition,hospitalsize,ahigh-techhosp1italindicator,profitstatus,,changes,,becausemargin,notprice,isthedependentvariable,,-caseloadcross-payercorrelationassupportforthehypothesisthatquality(asproxiedbyservicelevel),,-techindicatortheauthorswritethatit“equals1ifthehospitalisinroughlythetopone-quartertoone-thirdofallhospitalsinthestateinthebreadthofhigh-techserviceofferings,includingneonatology,openheartsurgery,cardiaccatheterization,traumacenter,magneticresonanceimaging,
FriesnerandRosenman(2002)isthefinalstudybasedonCaliforniaOfficeofStatewideHealthPlanningandDevelopmenthospitaldischargedata(from1995and1998).(-payments/charges).Withhospital-levelOLSmodels,theauthorsestimatetheeffectsofchangesinMedicareorMedicaidchargesandtheproportionunpaidonchangesinprivatepricesandpublicandprivateserviceintensity(lengthofstay),controllingforchangesinnumberofbeds,race,ethnicity,outpatientprices,,onprivatepricechanges,,,theyalsofindthatthechangeinpublicchargesispositivelycorrelatedwithchangesinprivatecharges,,(notallofwhichisreceivedinpayment),again,,(1998)asks,towhatextentdolowerMedicarepaymentsleadtolowercosts(reducedservicesandlowerquality)andtowhatextentisthecostlevelmaintainedandtheburdenofcoveringthemshiftedtotheprivatesector?Hisanswerdependsinpartonthenatureoftheprivatemarket,whichvariedconsiderablyoverthetwotimeperiodsheexamined—,includingthoseestablishedbytheConsolidatedOmnibusBudgetReconciliationActof1985,OmnibusBudgetReconciliationActsof1987,1989,1990and1993,
Forthekeyindependentvariable,CutlerconstructsanarguablyexogenousmeasureofMedicarepayment2reduction,whichhecallsthe“Medicarebite.”HenotesthatMedicare’,,CutlerestimatesbyOLStheeffectoftheMedicarebiteonhospital’schangesinperpatientnon-Medicareprivaterevenue,hospitalclosures,numberofhospitalbeds,changesinnursestaffinglevels,anddiffusionoftechnology,controllingforchangesincost,managedcareenrollment,profitandownershipstatus,numberofbeds,andMSAsize,but,notably,-1985periodhospitalsshiftedcostsdollar-for-dollar,amuchgreatercostshiftratethanfoundbyClement(1997/1998)andZwanziger,Melnick,andBamezai(2000)whostudiedthesametimeperiod(thoughthosetwostudiesfocusonCaliforniaonly,asdescribedabove.),,’,,,(2009)hasprovidedwhatis,perhaps,,,sheconsiderstheheterogeneityofthateffectacrossprivate-publicpayermix(atestofthe“market”vs.“strategy”hypotheses,),,,theendogeneityofthetyperaisedbyGlazerandMcGuire(2002)
,Wu’sprovidesthestrongestmitigationagainstandtestofthepotentialendogeneityofMedicarepayment,:a“BBAbite”(similartoCulter’s(1998)“Medicarebite”),butnotthesecond,-Medicareprice,againsimilartothatofCulter(1998).Twotypesofmodelsareestimated,,abargainingpowermeasure(theshareofdischargesthatareprivatepaylessthatforMedicarepatien4ts),hospitalownershiptype,levelandchangeinHMOmarketpenetration(alsoinstrumented),changeincasemix,hospitaloccupancyrate,levelandchangeinMedicaid-to-Medicarephysicianfeeratio,shareofforprofithospitals,,thekeyindependentMedicarepriceorrevenuechangevariablesareinteractedwiththebargainingpowervariable(totestthemarketpowerversusstrategyhypotheses).Inothermodels,theMedicarerevenuechangeisfurtherinteractedwithhospitalcharacteristics(profitstatus,teachinghospitalindicator,publichospitalindicator,HMOmarketpenetrationlevelandchange,levelandchangeinproportionofdischargesinthemarketrepresentedbyforprofithospitals).Shefoundthat,:-profit,teaching,,Wuobtainedverydifferentresultsusingun-instrumentedchangesinMedicarerevenueandprice,therebyjustifyingtheneedforinstruments(Wu2010).,Wuconductsafalsificationtest,findingthattheinstrumentedvariablesarenotstatisticallysignificantinthesamesecondstagemodelsusingdatafromapriorperiod(1992to1996).19
-shiftingliteraturesince1996providedintheforegoingsections,,the1980sthroughearly1990s(fivestudies),themid-1990sthroughearly2000s(fourstudies),,fourfoundsomeevidenceofcostshifting,,,,,asitwouldbebyaprofit-maximizingfirm,,,,butnotwithoutcarefulattentiontotheformofdependentvariables,theinclusionofrelevantobservablefactors,theattentiontopotentialendogeneityofothers,(2011).20
PWC(PriceWaterhouseCoopers).,StearnsS,DesHarnaisS,PathmanD,Tai-SealeM,
