NBER WORKING PAPER SERIES

CONTROLLING HEALTH CARE COSTS THROUGH LIMITED NETWORK INSURANCE PLANS:

EVIDENCE FROM MASSACHUSETTS STATE EMPLOYEES

Jonathan Gruber
Robin McKnight

Working Paper 20462

http://www.nber.org/papers/w20462

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
September 2014

We are extremely grateful to Dolores Mitchell, Ennio Manto, Catherine Moore, and Diane McKenzie
at the GIC for their enormous assistance in providing the data for this project and to seminar participants
at Brigham Young University, UC Irvine and NBER Summer Institute for helpful comments. This
research was supported by the National Institute on Aging through Grant #P30AG012810 to the National
Bureau of Economic Research. Jonathan Gruber is a member of the Commonwealth Health Connector
Board. The views expressed herein are those of the authors and do not necessarily reflect the views
of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-
reviewed or been subject to the review by the NBER Board of Directors that accompanies official
NBER publications.

2014 by Jonathan Gruber and Robin McKnight. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit, including
notice, is given to the source.
Controlling Health Care Costs Through Limited Network Insurance Plans: Evidence from
Massachusetts State Employees
Jonathan Gruber and Robin McKnight
NBER Working Paper No. 20462
September 2014
JEL No. I13

ABSTRACT

Recent years have seen enormous growth in limited network plans that restrict patient choice of provider,
particularly through state exchanges under the ACA. Opposition to such plans is based on concerns
that restrictions on provider choice will harm patient care. We explore this issue in the context of
the Massachusetts GIC, the insurance plan for state employees, which recently introduced a major
financial incentive to choose limited network plans for one group of enrollees and not another. We
use a quasi-experimental analysis based on the universe of claims data over a three-year period for
GIC enrollees. We find that enrollees are very price sensitive in their decision to enroll in limited
network plans, with the states three month premium holiday for limited network plans leading 10%
of eligible employees to switch to such plans. We find that those who switched spent considerably
less on medical care; spending fell by almost 40% for the marginal complier. This reflects both reductions
in quantity of services used and prices paid per service. But spending on primary care actually rose
for switchers; the reduction in spending came entirely from spending on specialists and on hospital
care, including emergency rooms. We find that distance traveled falls for primary care and rises for
tertiary care, although there is no evidence of a decrease in the quality of hospitals used by patients.
The basic results hold even for the sickest patients, suggesting that limited network plans are saving
money by directing care towards primary care and away from downstream spending. We find such
savings only for those whose primary care physicians are included in limited network plans, however,
suggesting that networks that are particularly restrictive on primary care access may fare less well
than those that impose only stronger downstream restrictions.

Jonathan Gruber
Department of Economics, E17-220
MIT
77 Massachusetts Avenue
Cambridge, MA 02139
and NBER
gruberj@mit.edu

Robin McKnight
Department of Economics
Wellesley College
106 Central Street
Wellesley, MA 02481
and NBER
rmcknigh@wellesley.edu
 2

Asemployersandgovernmentslooktocontrolrunawayhealthcarecosts,oneplacetheyare

turningistolimitednetworkplans.Recognizingthatthecostofcomparableservicescanvarywidely

acrossproviders,insurersareofferingplansthatexcludethehighestcostprovidersandthereby

significantlyreduceinsurancepremiums.Theseplansoftendonotvaryintheirenrolleecostsharingor

otherplancharacteristics,relyingonlyontherestrictiontolowercostproviderstoensuresavings.Asa

result,theyhaveproventobeincreasinglypopular,andtheyappeartobeamainstayoftheplan

offeringsonstateandfederalexchangesundertheAffordableCareAct(ACA).Inparticular,theexplicit

tyingofACAinsurancesubsidiesforlowincomefamiliestothe(second)lowestcostplanintheareais

likelytoinduceenormousmovementintolimitednetworkplans,whichareoftentheleastexpensive.

Buttheselimitednetworkplansarenotwithouttheirdetractors.Manyareconcernedthat

individualswillsufferadisruptionincareiftheyswitchtoalimitednetworkplan.Thiscouldleadto

deteriorationinthequalityofcarewherethecontinuityofcareismostvaluable,suchasforthosewith

chronicillness.Recentexpansioninlimitedandtierednetworkplans(thelatterincludeabroader

networkbutchargedifferentialsforuseofmoreexpensiveproviders)inMassachusetts,forexample,

wasstronglyopposednotonlybyhighercostprovidersbutalsobypatientadvocacygroups.1Indeed,

thisconcernpromptedtyingtheACAsubsidiestothesecondlowestcostplaninanarea,toensurethat

patientswouldnotbeforcedintonetworkswhichdidnotincludetheirprovider.Despitethis,ACA

criticshaverecentlyfocusedonthedominanceoflimitednetworkplansonthenewexchanges.2

Assessingtheimplicationsforenrolleesoflimitednetworkplanenrollmentisthereforean

importantissueforevaluatingboththefutureofemployersponsoredinsurance(ESI)andtheefficacyof

theACAexchanges.Yetthereisvirtuallynoworkontheimplicationsofenrollmentinalimitednetwork

1
WeismanandConaboy(2011)
2
Forexample,theCEOofCedarsSinaiHospital,ThomasPriselac,recentlytoldTimemagazine,Werevery
concernedwiththeimpact[thatasmallernetwork]hasonpatients(Pickert2014).AnarticleonCNN.com
describedpatientswhoaredismayedthattheircurrentdoctorsarentintheplansorthattheycantgotothe
onestheythinkarebestforthem(Luhby(2014).Similarly,TheBostonGlobequotedpatientNancyPetro,who
saidNowIhavetodrive50milesforbloodworkwhentheresahospitalthreemilesfrommyhouse(Jan2014).
 3

planforenrolleewellbeing.Thereisanolderandmuchlargerliteratureontheimpactsofmanaged

careplans,whichincludeasoneoftheirkeyaspectsnetworklimitations(e.g.Glied,2000).Butthis

literaturewasnotfocusedondistinguishingtheimpactsofnetworklimitationsfrommanyoftheother

differencesinvolvedinmanagingcare.Thereisalsoasmallrecentliteratureonhowlimitednetworks

impactchoiceofproviders(seeFranketal.forareview),butthisliteraturedoesnotaddressthe

impactsonoverallspendingandutilizationpatterns.

Toaddressthisissue,weturntotheexampleoftheMassachusettsGroupInsurance

Commission(GIC),thehealthinsuranceproviderforstateemployees,whichintroducedsizeablenew

incentivesforlimitednetworkplansaspartoftheiropenenrollmentforfiscalyear2012.Inparticular,

thestateofferedathreemonthpremiumholidayforenrollmentinlimitednetworkplansbystate

employees.Atthesametime,theGICprovidesinsuranceforanumberofmunicipalities,towhomthis

premiumholidaywasnotextended,providinganaturalcontrolgroup.WehaveobtainedfromtheGIC

acompletesetofclaimsdataforthe2009through2012periodwhichallowustoassesstheimplications

ofthissizeablenewincentiveforenrollmentinlimitednetworkplans.

Weusethesedatatoanswertwosetsofquestionsaboutlimitednetworkplans.First,how

responsiveareindividualstofinancialincentivestousesuchplans?Wehavesizablevariationin

financialincentivesinourdata,withthesavingsfromchoosingalimitednetworkplanrisingbyover

$500peryearonaverage.Thisallowsustoobtainprojectionsforpricesensitivitythatarehighly

relevanttoemployerplansandexchanges.Wecanalsoassesswhichenrolleesaremostpricesensitive.

Dofinancialincentivesinduceonlyhealthyenrolleestojoinlimitednetworkplans,leadingtoincreasing

sortingbyhealthacrossinsurers?

Wethenestimatetheimplicationsoflimitednetworkenrollmentforhealthcareutilization,

spendingandoutcomes,forthoseenrolleeswhododecidetoswitch.Weareparticularlyinterestedin
 4

assessingtheextenttowhichsuchswitcherschangetheirpatternofphysicianutilization,andwhether

thisimpactsbroaderhealthcareutilization.

Ourfindingssuggestthatswitchingtoalimitednetworkplanisverysensitivetofinancial

incentives;thethreemonthpremiumholidayofferedbytheGICcaused10%ofenrolleestoswitchto

limitednetworkplans,withanimpliedelasticityofswitchingwithrespecttothepremiumsavingsof1.3.

Thehealthiestindividualsarethelikeliesttoswitch,althoughthedifferencesbyhealtharenotlarge.

Wefindthatincentivestoswitchtoalimitednetworkplaninducedasizeablereductionin

spendingfortheGICof4.2%,implyingthatthemarginalpersoninducedtoswitchplansbythisincentive

spent36%less.Spendingfallssignificantlyformostcategoriesofspendingandthespendingdecline

appearstobecausedbyreductioninboththequantitiesofcarereceivedandthepricespaidforcare.

Mostimportantly,however,thereisanincreaseinprimarycarephysicianvisitsandspendingthatis

morethanoffsetbyadecreaseinspecialistvisitsandspending.Fallsinemergencyroomandhospital

spendingsuggestthatanyreductioninphysicianaccessthroughnetworklimitationsdidnotcausean

increaseinuseoftertiarycare,andthereisnoevidenceofanydeteriorationalongmeasuresofhospital

quality.Distancetraveledtoprovidersfallsforprimarycarephysicians,butrisesforspecialistsandin

particularhospitals;thereis,however,noevidencethatpatientsareusinglowerqualityhospitals.There

isalsonoevidenceofparticularlyharmfuleffectsforchronicallyillpatients.Butwedofindthatthe

savingsareconcentratedinthoseindividualswhocanretaintheirprimarycarephysicianwhenmoving

toalimitednetworkplan,suggestingthatlimitsonprimarycareaccessmaynotbeascostreducingas

aredownstreamlimitsonotherproviders.Overall,thefindingssuggestthattheswitchtolimited

networkplansreducedspendingwithoutharmingaccesstoprimarycareorinducingshiftstomore

expensivetertiarycare.

Ourpaperproceedsasfollows.SectionIdescribestheGICpolicychange.SectionIIlaysoutthe

detaileddatamadeavailabletousbytheGIC,andSectionIIIdescribesourempiricalstrategyforusing
 5

theserichdatatoidentifytheimpactoflimitednetworkplans.SectionIVshowstheresultsforplan

choice,andSectionVpresentstheresultsforutilizationresponses.SectionVIconsidersimpactson

patientaccessasmeasuredbydistancetraveledandhospitalquality,whileSectionVIIconsiders

heterogeneityinourfindings.SectionVIIIconcludes.

PartI:TheGICPolicyChange

TheGICprovideshealthinsuranceoptionsforallstateemployeesaswellasemployeesofa

numberoflocalmunicipalitieswhohavechosentobuyintothestateplan.Atthestartofoursample

period,theGICinsured81,420stateemployeesand109,343dependents.Inaddition,therewere23

municipalitiespurchasingtheirinsurancethroughtheGIC,with14,232employeesand19,160

dependents.MunicipalitiesmayfindthebroadernegotiatingpoweroftheGICamoreattractive

alternativetotheirlocalpurchasingoptions,althoughunionsoftenopposethehigheremployee

contributionsandmorelimitedchoicesassociatedwithGICenrollment.Asaresultoftheseconflicting

interests,about10%ofthemunicipalitiesinthestatewereenrolledintheGICby2012.

Infiscalyear2011,theyearbeforethelimitednetworkincentiveplanbegan,theGICoffered11

planoptions.TheseplanoptionsaresummarizedinTable1.Oftheseplans,fivewereclassifiedas

broadnetworkplans(mostlyPreferredProviderOrganizations(PPOs))thatallowedemployeesafree

choiceofproviderinthecountiesinwhichtheplansoperated.Theothersixwerecategorizedas

narrownetworkplanswithmorelimitedchoiceofproviders..Thecostsharingfeaturesofthebroad

networkandlimitednetworkplansarenearlyidentical,sotheonlysubstantivedifferenceacrossthe

plansistheirnetwork.3AccordingtoinformationfromGICinsurers,thesenarrownetworkplansare

3
Thereareminimaldifferencesincostsharingacrossplans;mostofthevariationisacrossinsurers,notacross
plantypes.Forinstance,Tuftsplanshaveslightlyhigherhospitalcopaymentsthanotherplans,butthesehigher
copaymentsapplytoboththeirbroadnetworkplan(TuftsNavigator)andtheirlimitednetworkplan(TuftsSpirit).
Asaresult,averagehospitalcopaymentsinlimitednetworkplansareidenticaltoaveragehospitalcopaymentsin
broadnetworkplans.Foraspecialist(inTier1),copaysrangefrom$20to$25,withanaverageof$24.17in
 6

establishedtoexcludethemostexpensiveproviderswhilestillmaintainingsufficientcoverageofthe

plansservicearea.

Ofcourse,narrownetworkisavaguetermthatcanhavemultiplemeanings.Toprovidea

richerinterpretationofthemeaningofanarrownetworkintheGICcontext,weconsiderempirical

measuresofnetworkbreadth.Inparticular,wetakeourfullsetofdataoverthreeyears(described

furtherbelow)andfocusoncountiesinwhichplansoperate.Inthosecounties,weconsiderall

providersforwhichweobserveatleast5(or10)innetworkclaimsoverthethreeyearperiodacrossall

insurers.Wethencalculate,foreachinsuranceplan,theproportionoftheseprovidersforwhichwe

observeatleast5(or10)innetworkclaimsforthatparticularplan.Whilethismeasureundoubtedly

incorporatesmeasurementerror,itnonethelessprovidessomerelativeinformationaboutthebreadth

ofeachplansnetworkwithinthecountiesinwhichitoperates.

TheresultsofthisexercisearepresentedinTable2.Theysuggestthatlimitednetworkplans,

onaverage,haveclaimsfromasmallerfractionofthephysiciansandhospitalsinacountythanthe

broadnetworkplansdo.Overall,ourmeasuressuggestthatbroadnetworkplanscovernearlytwiceas

manyphysiciansandabout50%morehospitalsthanarecoveredbynarrownetworkplans.Whileone

ofthelimitednetworkplans,HealthNewEngland,appearstohavearelativelybroadhospitalnetwork,

itisworthnotingthatthisplandoesnotoperateineasternMassachusetts,soitsnetworkdoesexclude

manyhospitalsinthestate.4

limitednetworkplansand$23.00inbroadnetworkplans.Forprimarycarephysicians,copaymentsrangefrom
$15to$20,withanaverageof$17.50inlimitednetworkplansand$18.00inbroadnetworkplans.
4
WehaveconfirmedtheresultsinTable2byusingdatafrominsurerswhereavailable.Wehavecomparedthe
listsofinnetworkhospitalsforlimitednetworkplansandfoundthattheresultsaresimilartothosethatwereport
inTable2.Forexample,HealthNewEnglandincludes92.3%ofhospitalsintheclaimsbasedmeasureand100%of
thehospitalsinthelistbasedmeasure.Likewise,TuftsSpiritincludes32.9%ofhospitalsintheclaimsbased
measureand25.7%inthelistbasedmeasure.Forphysicians,weentereddataonallinnetworkphysiciansfor
twolimitednetworkplans,HarvardPrimaryChoiceandHealthNewEngland.Forbothplans,wefoundthatfor
25%ofphysicianswedidnothaveenoughclaimstoclassifytheminourclaimsdata;fortheremaining75%of
physicians,60%(Harvard)or68%(HealthNewEngland)wereclassifiedaslimitedinbothourdataandtheinsurers
lists,aneffectivematchrateof8090%.
 7

Priortothepremiumholiday,therewasanexistingfinancialbenefittochoosinglimited

networkplans,reflectingdirectlythelowercostofthoseplanstothestate.Inparticular,thestate

contributed20%ofthecostofinsuranceplansforactiveemployeeshiredbeforeJuly1st,2003(and25%

forthosehiredafterthatdate),sothataportionofthelowercostsoflimitednetworkplanswere

passedontoemployees.In2011,theemployeeshareofthemonthlypremiumforindividualcoverage

rangedfrom$81.32forUnicareCommunitytoChoiceto$153.36forUnicareBasic(forworkerspaying

20%ofthepremiumcost).Sixteenpercentofenrolleesatthestatelevelchosetoenrollinlimited

networkplans.

Thefinancialincentivetoswitchtoalimitednetworkplanvariedsignificantlyacrossemployees,

forseveralreasons.First,thereweredifferentcontributionrulesforthelocalmunicipalities.For

example,whileSaugusrequiredthatemployeescontributeonly10%towardsmostplans,Swampscott

requiredthatemployeescontribute35%towardsmostplans.Second,withinmunicipalities,different

contributionrulesapplytodifferenttypesofemployees.Forinstance,teachersfacedifferentratesthan

otheremployeesinsomemunicipalities.Third,thestatechargesdifferentratestoemployeeswith

differenthiringdates,asdescribedabove.Finally,somemunicipalitiesprovidedmoregenerous

coverageoflowercostinsuranceoptions.Forexample,Springfieldrequiredthatemployeescontribute

25%towardsUnicareBasic,butonly15%towardsotherplans.Thesedifferencesincontributionrules

generatemeaningfuldifferencesintheincentivetoswitchtoalimitednetworkplan.Forexample,the

savingsfromswitchingfromfamilycoveragethroughTuftsNavigator(abroadnetworkplan)tofamily

coveragethroughTuftsSpirit(alimitednetworkplan)rangedfrom$29permonthinthethree

municipalitiesthatrequireda10%employeecontributionforbothplans,to$160permonthina

municipalitythatrequireda7%higheremployeecontributionforTuftsNavigatorthanforTuftsSpirit.
 8

Forthefiscalyear2012openenrollment,theGICdecidedtoaddanextraincentiveforstate

employeestoenrollinlimitednetworkplans.5Inparticular,stateemployeeswhodecidedtoenrollin

limitednetworkplanswereofferedathreemonthpremiumholiday,withnoemployeecontribution

requiredforthefirstthreemonthsoftheyear.Fortheaffectedplans,thispremiumholidayamounted

toa25%reductionincontributions,rangingfromsavingsof$268permonthforindividualcoverage

fromUnicareCommunityChoiceto$764forfamilycoveragefromHarvardPilgrimPrimaryChoice.This

incentivewasnotofferedbylocalitieswhichusetheGICsystem.Thispolicychangeinducedamajor

differentialincentiveforlimitednetworkplansforstateversusmunicipalemployees.

PartII:Data

OurdataforthisanalysisincludeacompletesetofclaimsandenrollmentrecordsforallGIC

enrolleesforthethreefiscalyears,spanningtheperiodfromJuly2009throughJune2012.GICs

insurancecontractsrunfromJulythroughJune,sothesedataallowustoobservethreefiscalyears.The

premiumholidayaffectsfiscalyear2012,whichrunsfromJuly2011toJune2012.

Forthepurposesofouranalysis,welimitthesampletoactiveemployeesandtheirdependents

whowerecontinuouslyenrolledoverthethreeyearsofoursampleperiod.Therestrictionto

continuouslyenrolledindividualsensuresthatthecompositionofoursampledoesnotchangeover

time.However,thissamplerestrictionprecludestheinclusionofeightmunicipalitiesthatjoinedGIC

duringoursampleperiod.Becausewecannotidentifywhichemployeesareteachers,weexcludedata

fromamunicipalitythathasdifferentcontributionratesforteachersthanforotheremployees.The

resultingsampleincludesdataon159,732enrollees,ofwhom86%obtainedcoveragethroughthestate

and14%obtainedcoveragethroughoneof21municipalitiesinoursample.

5
AtthesametimethestateimposedmandatoryreenrollmentinGICplans,withanautomaticdefaulttolimited
networkplansifindividualsdidntreenroll.Butreenrollmentrateswereover99%,sothisdidntenduphavinga
verylargeimpact.(CommonwealthofMassachusettsGroupInsuranceCommission,2012)
 9

Our(deidentified)dataincludeverylimiteddemographicinformationsuchasageandgender,

informationonenrollmentchoices,andinformationonhealthcareutilizationandspendingoverthis

timeperiod.Intheclaimsdata,weidentifydifferenttypesofservicesandconstructannualmeasuresof

utilization.Ourmeasuresincludecountsofmedicalencountersinayear(e.g.officevisits,prescription

drugpurchases,etc.)andtotalexpendituresbyallpayersforthosemedicalservices.Whileitispossible

thatoutofnetworkproviderschoosenottofileclaimswiththelimitednetworkinsurers,wedo

observeclaimsthatindicatethattheproviderisoutofnetwork,includingsomeclaimsforwhichthe

insurerpaysnothing.Whilewecantruleoutthepossibilitythatwearemissingsomeclaimsfromout

ofnetworkproviders,anymissingclaimsarelikelytobeforlowcosteventsandwouldthereforehave

minimaleffectonouroverallfindings.Withthreeyearsofclaimsdataforeachenrollee,ourfinaldata

setincludes479,196annualobservationsonthe159,732continuouslyenrolledindividualsinour

sample.

Table3providessummarystatisticsonoursample,includinginformationonaverageannual

medicalexpendituresandutilization.Theaveragememberincurs$4,811intotalmedicalexpenses

duringayear.About23%oftheseexpensesareincurredinofficevisits,18%ininpatient

hospitalizations,and30%inoutpatientvisits.Prescriptiondrugsaccountfor19%ofthecosts.The

remainderofthecostsincludesemergencydepartmentvisits,labs,andothercosts,whichinclude

homehealthcare,supplies,ambulances,andavarietyofotherservices.

PartIII:EmpiricalStrategy

FollowingthediscussionoftheGICpolicychangeabove,thereareseveralsourcesofvariationin

thecostoflimitedandbroadnetworkplans.Atanypointintime,thereissignificantvariationacross

statevs.municipalworkers,aswellasacrossstateworkersbydateofhire.Overtime,aspremiums

change,thesedifferencesinpolicygiverisetodifferentialchangesintheoutofpocketpremiumcostof
 10

limitednetworkplans.Andthepremiumholidayinfiscalyear2012createdasharpdiscontinuityinthe

costoflimitednetworkplansforstateemployeesrelativetolocalemployees.

Asaresult,wepursuetwoidentificationstrategiesinouranalysis.Thefirstisadifferencein

differencesanalysisaroundthe2012policychange,comparingstatetomunicipalemployeesovertime.

Thisisalegitimateidentificationstrategyiftherearenopreexistingdifferentialtrendsbetweenthese

twogroups,andifthereisnocontemporaneousshocktooneofthesetwogroups.

Tocarryoutthisstrategy,weestimateregressionsoftheform:

(1) Yimt=+STATEm*AFTERt+MUNIm + YEARt+Ximt+imt

whereiindexesindividuals,mindexesmunicipalities(andstate),andtindexesyears.STATEisadummy

forobtainingcoveragethroughastateemployee,andAFTERisadummyforfiscalyear2012.MUNI

representsafullsetoffixedeffectsformunicipality,andYEARrepresentsafullsetofyearfixedeffects.

Xisasetofindividualcontrolswhichincludesage,gender,familycoveragetier(individualorfamily),

andanindicatorforstateemployeeswhowerehiredbeforeJuly1st,2003.Thecoefficientcaptures

theimpactofbenefittingfromthepremiumholiday,relativetoearlierstateworkers,andcomparedto

thechangeoverthesametimeperiodformunicipalworkers.

Thesecondidentificationstrategyincorporatesthebroaderpricevariationthatarisesfromthe

differentialoutofpocketpremiumcostoflimitednetworkplansacrossemployeetypesandovertime.

Theadvantageofusingthisapproachisthatitprovidesmorepowertoidentifytheeffectofincentives

tomovetoalimitednetworkplan.Thedisadvantageisthattherecouldbepotentialendogeneityfrom

severalsourcesinthisbroadervariation.Forexample,thesharethatthestateandmunicipalities

requiretheiremployeestopayforhealthinsurancecouldberelatedtounderlyinginsurancedemand,or

dateofemployeehirecouldbecorrelatedwithindividualinsurancedemand.

Weaddresstheseconcernsthroughoursecondempiricalspecification:

(2) Yimt=+LIMSAVmt+MUNIm + YEARt+Ximt+imt

 11

whereLIMSAVisthesavingsfromswitchingtoalimitednetworkplanforworkeri.Thismeasureis

computedasthedifferenceintheweightedaverageofemployeecontributionstobroadnetworkplans

andtheweightedaverageofemployeecontributionstolimitednetworkplans,measuredasa

percentageoftheemployeecontributionstobroadnetworkplans.Theweightsrepresentthefraction

ofenrolleesineachtypeofplanwhochoseeachspecificplaninabaseperiod,sothatmoreweightis

placedontheemployeecontributionstothemorepopularplans.Becauseweweighttheemployee

contributionsbybaselineenrollmentshares,thecalculationexcludesinformationaboutHarvard

PrimaryChoiceandTuftsSpirit,whichwereaddedasnewinsuranceplanchoicesinfiscalyear2011.

Theuseofaweightedaverageacrossalloftheinsuranceplanoptionsmeansthatthismeasuredoesnot

varyacrossindividualswithinamunicipalityyear.Instead,thismeasurereflectsthesourcesofvariation

thatwereoutlinedabove,includingvariationacrossthestateandmunicipalities,acrosshiredategroups

(i.e.,groupsfacingthesamepremiumsharingrules),andovertime.Theuseofapercentagedifference

insavingsmeansthatwearenotusingvariationinthelevelofsavingsthatarisesfromdifferencesin

premiumsacrossindividualandfamilyplans;thedollarvalueofsavingsfromswitchingtoalimited

networkplanisalwayshigherforfamilyplansthanthedollarvalueofsavingsforindividualplans,but

thepercentagesavingsisalwaysthesame.

ToaddressthepotentialendogenietyofLIMSAV,weincludefixedeffectsforeachmunicipality

andcontrolsforthehiringperiodforstateemployeestocapturethosecorrelatesofinsurancedemand.

Inpractice,sincethemostsignificantvariationinoursamplecomesfromthepremiumholiday,our

resultsaresimilarusingeithermethod.

Itisveryimportanttobeclearontheinterpretationofthekeycoefficient. Ourestimatesof

theimplicationsoflimitednetworkplansforutilizationandoutcomesareidentifiedsolelybythe

compliersthatswitchplansinresponsetofinancialincentives.Thatis,ourestimatesarenota

populationaverageestimateoftheimpactofforcingallenrolleestoenrollinalimitednetworkplan.
 12

Butcurrentpolicyconversationscenteraroundemployeeandexchangechoice,whichconsiderlimited

networkplansasachoiceoption,notthemandateddefault.Thatis,ourestimatesprovidetherelevant

estimatesoftheimpactsofofferingfinancialincentivesoftherangedescribedaboveonutilizationand

outcomes.

PartIV:EnrolleePlanChoiceResults

Webeginbyexaminingtheeffectsoffinancialincentivesonthedecisionsofenrolleestoenroll

inlimitednetworkplans.Weestimatetheequationsabove,usingasadependentvariableadummyfor

enrollinginalimitednetworkplan.Weestimateallmodelsaslinearprobabilitymodels,althoughour

resultsareverysimilarifweuseProbits.Standarderrorsareclusteredatthelevelofthemunicipality.

Figure1previewsourfirststagefindings.PanelAgraphsthesavingsfromchoosingalimited

networkplanbyyear,separatelyformunicipalitiesandthestate.Asthefigureshows,municipal

employeesin2010facedlargerpotentialsavingsfromswitchingtoalimitednetworkplan,becausethe

employeeshareofthepremiumswastypicallyhigherforthemthanforstateemployees.Because

limitednetworkplanshavelowerpremiumsthanbroadnetworkplans,ahigheremployeeshareof

premiumsgeneratesmorepotentialsavingsfromchoosingalimitednetworkplan.From2010to2011,

therewasanincreaseinsavingsinbothgroupsfromchoosingalimitednetworkplan.Thisincreasein

savingswaspartiallyattributabletothefactthatbroadnetworkplanshadrelativelylargeincreasesin

premiums,whereasseveralofthelimitednetworkplanshadpremiumincreasesthatwereclosetozero.

Inaddition,increasesintheemployeeshareofpremiumswereimplementedbythestateandbyseveral

municipalities.6From2011to2012,therewasalargeriseinthesavingsfromlimitednetworkplansfor

stateemployees,duetothepremiumholiday,thatwasnotpresentformunicipalemployees.

6
UndertheFY10AppropriationAct,premiumcontributionratesforstateemployeesincreasedby5percentage
points.ThesechangeswerefirstproposedinJune2009andimplementedinAugust2009(Commonwealthof
MassachusettsGroupInsuranceCommission,2011).Thus,whilethepricechangewasimplementedpartway
 13

PanelBofthefigureshowsenrollmentratesinlimitednetworkplansovertimeforthestateand

municipalities.Theinitialrateofenrollmentishigherinmunicipalities,whichisconsistentwiththe

higherinitialmunicipalitydiscountforenrollinginsuchplans.From2010to2011,enrollmentinlimited

networkplansrisesinbothgroupsbysimilarmagnitudes,onceagainconsistentwiththeparallelrisein

financialincentivesovertheseyears.From2011to2012,enrollmentinlimitednetworkplansinthe

municipalitiesisfairlyconstant,whereasthereisanenormousjumpinthestateplans,mirroringthe

increaseinpotentialsavingsforstateemployees.Forstateemployees,enrollmentinlimitednetwork

plansrisesbyabout50%between2011and2012.

Table4presentsthesefirststageresultsinregressionform,confirmingwhatisshowninthe

figures.Weestimatethatthepremiumholidayraisedenrollmentinlimitednetworkplansinthestate

byover11percentagepoints,relativetothemunicipalities.Ouralternativeregressionapproach,using

thefullvariationoverthisperiod,yieldsanestimatethateach1percentagepointincreaseinthe

discounttolimitednetworkplansgivesrisetoa0.7percentagepointincreaseintheshareofenrollees

inlimitednetworkplans.Thediscountroseby16percentagepointsfrom2011to2012asaresultof

thepremiumholiday,whichwouldpredictthesame11percentagepointriseinlimitednetwork

enrollment.Theestimatedelasticityoflimitednetworkenrollmentwithrespecttoitspremium

discountis1.28(or0.007*(36.55/0.201)).

Table4alsoshowsthecoefficientsonmanyofthecovariatesincludedinthesefirststage

regressions.Thecoefficientssuggestthatmalesareslightlymorelikelytochooselimitednetworkplans

thanfemalesare.Enrollmentinlimitednetworkplanspeaksbetweentheagesof30and39,and

decreasesconsiderablyasadultsage.

Table5exploresheterogeneityinpricesensitivity,ineachcaseshowingthekeycoefficients

fromthesamespecificationshowninTable4,butestimatedononlythesubsampleofinterest.Wefirst

throughfiscalyear2010,theprimaryimpactonenrollmentchoicesshouldhaveoccurredduringopenenrollment
forfiscalyear2011.
 14

considerheterogeneitybyunderlyinghealth,dividingthepopulationintothosewhoareandarenot

chronicallyill.Weidentifythechronicallyillasindividualswithadiagnosis(inanofficesetting)of

hypertension,highcholesterol,diabetes,asthma,arthritis,affectivedisorders,andgastritis,following

Goldmanetal.(2004).Wefindthathealthierenrolleesaremorepricesensitive:thosewhoarenot

chronicallyillareabout20%moreresponsivethanthosewhoarechronicallyill.Thissuggeststhat,

whentheGICofferedfinancialincentivesforenrollmentinlimitednetworkplans,theyimprovedthe

healthmixofthoseplansandworsenedthehealthmixofthoseremaininginbroadnetworkplans,

althoughthedifferentialeffectisnotlarge.

Twokeydeterminantsofswitchingcouldbewhetherindividualscanremainwiththeirinsurer,

andwhethertheycankeeptheirdoctor.WeexploretheseissuesintheremainderofTable5.Wefirst

separatethesamplebyinitialchoiceofinsurer,limitingoursampletothosewhowereenrolledinbroad

networkplansatthebeginningofoursampleperiod.Weseethatthereissomeheterogeneityin

switchingbyinsurer.ThefirststageislargestforFallonenrollees,andsmallestforUnicareenrollees.

Theredoesnotappeartobeanysystematiccorrelationacrosscompanieswiththerestrictivenessof

thenetworksasillustratedearlier.

For79%oftheenrolleesinoursample,wecanidentifytheirprimarycarephysicianbasedon

claimsduringthefirsttwoyearsofoursample.Weidentifytheprimarycarephysicianasthephysician

withaprimarycarespecialtywithwhomtheenrolleehadthemostofficevisitsduringthefirsttwoyears

ofoursampleperiod.Wethenconsiderthreemutuallyexclusivegroups,amongthoseenrolleeswho

werenotyetenrolledinalimitednetworkplaninfiscalyear2011:(a)thosewhoseprimarycare

physicianisinthelimitednetworkversionofthesameinsuranceplaninwhichtheyarenowenrolled,

(b)thosewhoseprimarycarephysicianisinalimitednetworkplan,butnotthatofferedbytheircurrent

insurer,(c)thosewhoseprimarycarephysicianisnotavailablethroughanylimitednetworkplan.We
 15

expectenrollmentinalimitednetworkplantobedecliningacrossthesegroups,andindeedthatisin

thecase.

Table5showsthatthatthecoefficientonswitching(thatisidentifiedprimarilybythepremium

holiday)isdecliningacrossthesegroups.7Theeffectforthosewhocanswitchwithoutchanginginsurer

orphysicianisabout60%largerthanforthosewhomustswitchbothinsurerandphysician.Itis

interestingtonotethatover90%oftheseenrolleeshaveaprimarycarephysicianwhoparticipatedinat

leastonelimitednetworkplan.Thissuggeststhatswitchingtoalimitednetworkplandoesnot

necessarilydisruptprimarycarerelationshipsformanyenrollees.

PartV:ResultsforUtilizationandSpending

Havingestablishedourfirststagefactthatfinancialincentivesforlimitednetworkplans

stronglyinfluencechoiceofsuchplans,wenowturntoestimatingtheimpactonpatientoutcomes.The

regressionframeworkisthesameasthatusedintheprevioussection,butourdependentvariablesnow

relatetohealthcareutilizationandoutcomes.

Theseregressioncoefficientscanbeinterpretedasreducedformestimatesoftheimpactof

financialincentivesonpatientoutcomes.Ifwenormalizebythefirststageestimatesoftheimpactof

financialincentivesonplanchoice,wecanobtainanimplicitinstrumentalvariablesestimateofthe

effectofenrollinginalimitednetworkplanonoutcomes.Asnotedearlier,thisIVestimateisvalidonly

forthemarginalindividualsinducedtoswitchplansbyfinancialincentives,andnotforindividuals

randomlyenrolledinalimitednetworkplan.

Sinceourdependentvariablesfeaturebothmanyzerosandalargedegreeofskewness,we

estimateourmodelsbyagenerallinearmodel(GLM).Intheabsenceofzeroes,wewouldwanttouse

7
Ofcourse,theseresultssufferfromsomecensoringbiasthosewhosephysicianisinalimitednetworkplanmay
havebeenmostlikelytoswitchbeforethepremiumholiday.Thiswouldmostlikelyleadustounderstatethe
impactofthefinancialincentivedifferentialacrossgroups.
 16

ln(spending)asourdependentvariableandestimateourmodelsusingOLS.However,thepresenceof

observationswithzerospendingmakesthisaproblematicstrategy,sowefollowtheliteratureand

estimateGLMwithaloglinkfunction(BuntinandZaslavsky,2004).Inthisapproach,theconditional

meanismodeledas:

(3) LnE(Spending|X)imt=+LIMSAVmt+MUNIm + YEARt+Ximt+imt

Thisapproachallowsustoavoidselectiononthedependentvariableandalsogeneratescoefficients

thatareinterpretableaspercentagechanges.8

TotalImpactonSpending

WepreviewtheresultsfortotalspendingwithFigure2,whichshowsthetrendsintotal

spendingforstateandmunicipalemployees.Forthisfigure,weusequarterlydataontotalspending,in

ordertoshowmorepreciselythetimingofanychangesintotalspending.Thisfigureshowsthat

spendingtrendsverycloselyforstateandmunicipalemployeesovertimebeforethepremiumholiday,

withstateemployeeshavingconsistentlyhighertotalspending(onceagainconsistentwithhigher

enrollmentinlimitednetworkplansbymunicipalemployees).Thereisthenaveryclearrelativedecline

intotalspendingforstateemployees,whichbeginsinthefirstquarteroffiscalyear2012andbecomes

evenstrongerinsubsequentquarters.Thefactthatthespendingpatternsforstateemployees,as

comparedtomunicipalemployees,mirrorsthepatternsofenrollmentinlimitednetworkplansseenin

Figure1isquitesuggestiveofacausallink.

InTable6,weformalizethisanalysisusingourregressionframework.Therearetwocolumnsin

thetable,representingourtwoidentificationstrategies:theDDstrategyisusedinthefirstcolumn,

8
IncontrasttoGLM,OLSmodelsE(lnSpending|X)imt=+LIMSAVimt+MUNIimt+YEARimt+Ximt+
imt.Asaresult,OLSgeneratescoefficientsthataredifficulttotranslateintostatementsaboutE(Spending).The
traditionalsolutionhasbeentouseasmearingestimatortoconvertpredictionstotheunloggedscale(Manning
etal.,1987).However,theseretransformationsarebiasedinthepresenceofheteroskedasticity(Buntinand
Zaslavsky(2004),ManningandMullahy(2001)).

 17

whilethepricevariationstrategyisusedinthesecondcolumn.Ineachcell,wepresenttheGLM

estimateoftheeffectonaspendingmeasure.Ifonewishestointerprettheseasstructuralestimates

fortheimpactofbeingenrolledinalimitednetworkplanthen,asnotedabovetheDDestimatesshould

bemultipliedby1/0.116=8.6,andthepricevariationestimatesshouldbemultipliedby1/0.0070=143.

Webegininthefirstrowbymodelingtotalspending.Weestimateamarginallysignificant4%

declineinthelevelofmedicalspending.Normalizingbythefirststageeffect,thisimpliesthat,forthe

marginalswitcher,thereisa36%declineintotalspendingwhenmovingtoalimitednetworkplan.This

isaverysizeableimpact.Ontheotherhand,themagnitudeofthedeclineisquiteconsistentwiththe

differenceintotalpremiumsforthetwotypesofplans.Indeed,themeanreductioninoutofpocket

premiumforanenrolleetoswitchfromanaveragebroadnetworkplantoanaveragelimitednetwork

planinoursamplewas36.55%,asshowninTable3.Ourresultssuggestthatthepremiumdifferential

forlimitednetworkplansisdrivennotmerelybypositiveselection,butbyasubstantivedifferencein

costspercapita.9Theresultsusingfullvariation,whennormalizedbytherelevantfirststageresults,

generateverysimilarresults:forthemarginalswitcher,thereisa41%declineintotalspending.The

factthatourtwoidentificationstrategiesgeneratesuchsimilarresultshere,andthroughoutthe

subsequentresults,isreassuring.

Wethenturntomodelingspendingbysubcategoryofmedicalutilization.Wefindalarge

(albeitonlymarginallysignificant)declineinofficespending,withtheDDestimateimplyingaroughly

16%declineinofficespendingforthemarginalswitcher.Weestimateareductioninhospitalspending

whichisalsoverylarge,butstatisticallyinsignificant.Unfortunately,theconfidenceintervalsforour

9
Asimpleregressionofspendingonlimitednetworkplanenrollment,ignoringtheendogeneityoflimitednetwork
enrollment,yieldsacostsavingsestimateof30%.Itissurprisingthatthecausalimpactonspendingislargerthan
theOLSeffect,giventhatthelatterincludesselectioneffects.Thispartlyreflectsrelativelylimitedselection:the
averageageofthoseenrolledisonly3yearsyoungerthanthosenotenrolled,andtheoddsofchronicillnessis
onlyabout10%lower.Thismayalsopartlyreflectdifferencesbetweenthemarginalcomplierwiththepolicy
changeandtheaverageenrolleeinlimitednetworkplans,ordifferencesinthelongruneffectofenrollment
versusfirstyeareffects.Nevertheless,theresultssuggestthatmostofthesavingsfromlimitednetworkplansis
fromreducedspendingbyenrollees,notjustpositiveselection.
 18

inpatienthospitalspendingresultsarefairlywide,leavingusunabletoruleoutalargefallorrisein

hospitalspending.

Wedo,however,findalargeandmarginallysignificant5%declineinoutpatienthospital

spendingintheDDspecification,implyinga43%reductioninoutpatientspendingforthemarginal

switcher.Resultsforoutpatientspendingaresimilarinmagnitude,andstatisticallysignificant,inthe

specificationthatusesthefullpricevariation.Weseeanevenlargerreductioninlabspendingof8%in

theDDspecification,implyinga71%reductioninlabspendingforthemarginalswitcher.Wehave

furtherexploredthelabresultsandfoundthatlabspendingassociatedwithofficevisitsdoesntfall

significantly;rather,themajorchangeisinlabspendingassociatedwithoutpatientandemergency

roomsettings.

Wefindacorrespondinglylargereductioninemergencyroomspending.Thisisstrikingbecause

itdefiesthenaturalhypothesisthatwhenindividualshavetheirphysicianchoiceslimitedtheywould

tendtousemoreemergencyroomcare.Wehavefurtherbrokenemergencyroomvisitsbytypeof

visit,andfindthatabouttwothirdsofthereductionisfromreduceduseoftheemergencydepartment

fortreatmentofinjuries.

Wefindnomeaningfulimpactonprescriptiondrugspending,butalargeimpactonother

spending.Whenwebreakdownotherspendingintoitsconstituentcomponents,theresultsimply

thatthereareparticularlylargereductionsinhomehealthcarespendingandambulancespending,

althoughneitherissignificantonitsown.Itisworthnotingthatotherspendingconstituteslessthan

5%oftotalspending,sodeclinesinthiscategoryofspending,whilelargerinpercentageterms,are

relativelyunimportantinexplainingtheaggregatedeclineinspending.Rather,ifonetakesthepoint

estimatesseriously,thechangesthatappeartobemostimportantindrivingthe4.2%declineintotal

spendingincludethe5.0%declineinoutpatientspendingandthe(statisticallyinsignificant)5.6%decline

ininpatientspending,whichtogetheraccountformorethanhalfoftheoveralldeclineintotalspending.
 19

GiventherobustnessofourfindingstothetwoempiricalstrategiesusedinTable6,forthe

remainderofthepaperwepresentonlythemoreeasilyinterpretabledifferenceindifferenceresults.

Allresultspresentedbeloware,notsurprisingly,consistentusingthefullermeasureofincentives.

DecomposingSpendingImpacts

ThelargeimpactsontotalspendingthatareapparentinFigure2canbefurtherdecomposedto

assesswhetherthesavingsarisefromareducedquantityofcareorlowerpricespaidforafixedamount

ofcare.WeshowtheresultsofthisdecompositioninFigure3.InPanelA,weshowthechangesin

costsperservicetype,weightedbyfixedquantitiesofcareforeachservices.Thesefixedquantitiesare

basedonaverageutilizationforeachservicecategory(primarycareofficevisits,specialistofficevisits,

otherofficevisits,inpatientcare,outpatientcare,ERvisits,labs,prescriptiondrugs,andothercare)for

broadnetworkenrolleesinFY2010.Costsperserviceareallowedtovary,onaverage,overtimeand

acrossstatevs.municipalenrollees.Interestingly,averagecostsperservicearesimilarforstateand

municipalenrolleesatthebeginningofthesampleperiod,withcostsforstateenrollees,ifanything,

lowerthancostsformunicipalenrolleesinFY2011.Butcostsperserviceseemtodeclinerelatively

moreforstateenrolleesafterthepremiumholiday,especiallytowardstheendofFY2012.Thisfigure

suggeststhatthedeclineinspendingamongstateenrolleesispartiallyattributabletoarelativedecline

inperservicecosts.

PanelBofFigure3showsthechangesinquantitiesofeachtypeofservice,weightedbyfixed

costsofcare.Thefixedcostsarebasedonaverageperserviceallowablecostsforeachtypeofservice

forbroadnetworkenrolleesinFY2010,whilequantitiesareallowedtovaryovertimeandacrossstate

vs.municipalenrollees.Stateenrolleeswereclearlyusingagreateramountofcarethanmunicipal

enrolleesinFY2010andFY2011,butthemagnitudeofthegapdeclinesvisiblybeginningatthetimeof
 20

thepremiumholiday.Basedonthesefigures,itappearsthatdeclinesinpricesandquantitiesbothplay

aroleinreducingspendingforstateenrolleesafterthepremiumholiday.

Table7examinesthisissue,inaregressionframework,fordifferentcategoriesofservice.It

extendsTable6bypresenting,inadditiontoourGLMspendingresults,resultsformeasuresofquantity

ofcareandforcostspervisit.Specifically,itshowsOLSresultsforadummyvariableforanyutilization

ofthattypeofservice,fornumberofinstancesofutilization(visits,stays,tests,drugprescriptions,etc),

andforaveragecostperinstanceofutilization(conditionalonsomeutilization).Theresultsshownhere

areallsimilarifweuseProbitmodelsfortheanyvisitsoutcome,orifweusePoissoncountmodelsfor

thenumberofvisitsoutcome.

Wefindthatthereisnochangeintheoddsofhavinganofficevisit,butthatthenumberofvisits

fallssignificantly,by0.15offabaseof7.36visitsonaverage.Thereisnoeffectonpervisitcost.For

inpatientutilization,weonceagaindonotfindanyresultsofstatisticalsignificance.Thepoint

estimates,however,aremoreconsistentwithadeclineincostspervisitthanwithadeclineininpatient

hospitalvisits.Foroutpatientutilization,wedofindareductionintheoddsofanyoutpatientvisit,in

thenumberofvisits,andinthecostpervisit.Foremergencyroomutilization,therearereductionsinall

measures,buttheonlysignificantchangeisareductioninthenumberofvisits.Forlabs,wefinda

reductioninboththeleveloflabresultsandthecostperresult,whichproducesthesignificanttotal

dropinspending.Theresultsforprescriptiondrugutilizationareinteresting.Asnotedearlier,thereis

nonetimpactonprescriptiondrugspending.Butthisappearstoreflectasignificantreductioninthe

numberofprescriptions(areductionof0.4prescriptionsonabaseof11.7prescriptions),offsetbya

large(althoughnotsignificant)riseinthecostperprescription(ariseof$2.08offabaseof$77).We

alsoseealargeandstatisticallysignificantdeclineinthequantityofothercare.

Overall,theresultsareconsistentwiththevisualimpressionfromthetwopanelsofFigure3,

whichsuggestedthatthereducedspendingisattributabletoreductionsinbothquantityandprice.The
 21

resultsinTable7indicatethatthepremiumholidayisassociatedwithdeclinesinthequantityofcare

acrossallcategoriesofservice,withstatisticallysignificanteffectsonofficevisits,emergencyroom

visits,prescriptiondrugs,andothervisits.Theresultsindicatethatpremiumholidayisalsoassociated

withdeclinesinthepervisitcostforeverycategoryofcare,exceptprescriptiondrugs,withstatistically

significantresultsforoutpatientvisits.

TypeofPhysician

Theresultsforphysicianofficeutilizationsuggestthepotentialvalueofafurtherdecomposition

bytypeofphysician.Table8followsthesameformatasTable7,butfurtherdecomposestheresults

forofficevisitsbytypeofprovider.

Thefirstpaneldecomposesofficevisitsintovisitstoprimarycarephysicians,specialists,and

others,usingdataonproviderspecialtyforallprovidersinMassachusetts.Theotherprovider

category,whichaccountsfor7%ofofficevisits,includesoutofstateprovidersaswellasother

providersthatwewereunabletomatchtoaspecialty.Thedifferencesacrossthesegroupsisstriking.

Wefindthatthatprimarycareofficevisitsincrease,withastatisticallysignificant3%riseinspendingin

thereducedform,implyingaroughly28%riseinspendingonprimarycareforthosewhomoveinto

narrownetworks.10Atthesametime,visitstospecialistsfallsignificantly,withalarge5%declinein

spendinginthereducedformimplyingaroughly45%reductioninspecialistspending.Forother

physicians,thereisasignificantdeclineinvisits,butanoffsettingsignificantriseincostpervisit,

resultinginaninsignificanteffectonspending.Theseresultssuggestthatenrollmentinalimited

networkplanisassociatedwithashifttowardsprimarycareandawayfromspecialistcare.

10
In2013,GICintroducedaCenteredCareIntegratedRiskBearingOrganizationInitiativethatmayhaveaffected
themixofprimaryandspecialistcareuse.However,thatinitiativewasintroducedafteroursampleperiodended,
socannotexplainourfindingofincreasedprimarycareanddecreasedspecialistcareuse.
 22

Thesecondpaneldecomposestheofficevisitsinsteadintovisitstophysiciansthatthepatient

hadseenpreviously,andvisitstothosethatthepatienthadnotseen.Ofcourse,wecanonlyidentify

whetherapatienthaspreviouslyseenaphysicianiftheencounterappearsinourclaimsdata;sincewe

donotobserveanentirelifetimeofclaimsdataforeachindividual,wewilloverstatethenumberof

newproviders.Weknowthattheproportionofprovidersthatweidentifyasnewwillbeartificially

highatthebeginningofourclaimsdata,andwilldecreasemechanicallyovertime.Wethereforerelyon

afulladditionalyearofclaimsdata,extendingbacktoJuly2008,toidentifynewandoldproviders,

whilecontinuingtofocusouranalysisontheperiodbeginninginJuly2009.Thedisadvantageofthis

approachisthatourregressionsincludeonlyindividualswhowerecontinuouslyenrolledforfouryears

(or84%ofourbaselinesample).Buttheadvantageisthatwedecreasethemeasurementerrorinour

classificationofnewvs.oldproviders.Importantly,thereisnoreasontoexpectdifferential

measurementerrorinourclassificationofnewandoldprovidersforstateandmunicipalemployees,so

wedonotbelievethatourcoefficientsarebiasedbythisissue.

Thereisasizeablereductioninvisitstoproviderswhomthepatienthadpreviouslyseen,aswell

asareductionincostpervisit,sothattotalspendingonsuchphysiciansfallsby3.4%.Fornew

providers,thereisasizeableriseintheoddsofavisit,thenumberofvisits,andcostpervisit,sothat

totalspendingonnewprovidersrisesby5.6%.Giventhatthemeanspendingonoldprovidersis$771,

whileonnewprovidersitis$304,thisisconsistentwithanoverallfallinphysicianspending.Overall,

shiftingtolimitednetworkplansappearstocauseashiftawayfromtraditionalproviderstowardsnewer

(lowercost)options.Thenewproviderschosenbyindividualsaremarginallymoreexpensivethannew

providerschosenpreviously,butnotenoughtooffsetthecostdifferencebetweennewandold

providers.

 23

PartVI:ImpactonPatientAccess

Afullanalysisoftheimpactoflimitednetworkplanswouldincludeeffectsonpatientoutcomes.

Unfortunately,ourdatadonotcontainhealthoutcomemeasures.Typicallyusedprocessmeasures,

suchasavoidablehospitalizations,relyoninpatientdatawhereourprecisionislimited.Oureffortsto

investigatesuchvariableshavefoundnoeffectsbutverywideconfidenceintervals.Wehavealsotried

toassessimpactsonpatientmortality,andonceagainourestimatesweresimplytooimprecise.11

Thereisoneimportantoutcomethatisthecurrentfocusofmuchdebateoverlimitednetwork

plans,however:patientaccesstoprovidersasproxiedbydistancetraveled.Amajorconcernraised

aboutlimitednetworkplansisthatitwillleadpatientstohavetotravelmuchfurthertoseetheir

providers.Wecanaddressthisconcernwithourdatabyexaminingthedistancebetweenpatientsand

theproviderstheydoseewhentheyjoinlimitednetworkplans.Todoso,weusethedistancebetween

thecentroidofpatientandproviderzipcodesinourdata,foreveryproviderpatientpairthatwe

observe.

TheresultsofthisanalysisareshowninthetoppanelofTable9.Wefindthatoverallthereis

nosignificantimpactondistancetraveledforanofficevisit.Butwefindthatthismasksimportant

heterogeneitybytypeofofficevisit:distancetraveledforprimarycarevisitsfallsby0.65milesinour

reducedformestimates,orbyabout5.6milesasourimpliedIVcoefficient;thisismorethanhalfofthe

baselinedistancetraveledforprimarycare.12Ontheotherhand,distancetraveledrisesforspecialists,

11
Whilewedontobservemortalitydirectlyinourdata,wecanexaminetheprobabilityofexitfromthefull
sampleand,inparticular,exitfromafamilyplanwhentherestofthefamilyremainsenrolledatagesthatarenot
associatedwithexitsduetocollegegraduationorMedicareenrollment.Inthefullsample,thepremiumholidayis
associatedwithastatisticallyinsignificant0.014percentagepointdeclineintheprobabilityofsuchanexit.Witha
standarderrorof0.04percentagepointsandameanexitrateof0.4percent,welacktheprecisiontorejecta
meaningfulincreaseordecreaseinsuchexits.Wealsoexaminedthisexitrateforavarietyofsubsamplesand
foundnostatisticallysignificanteffects.
12
Itispossibletoobserveanegativeimpactonprimarycaredistanceasaresultofcompositionalchangein
primarycarevisits.Weobservethatlimitednetworkenrollmentiscorrelatedwithanincreaseinprimarycare
visitsand,totheextentthatthoseadditionalvisitsaretoproviderswhoarerelativelyshortdistancesawayfrom
thepatients,itispossibletofindthatlimitednetworkenrollmentisassociatedwithadeclineinaverageprimary
caredistance.
 24

althoughnotsignificantly.Wefindthatthosepatientswhocontinuetoseetheiroldprovidersare

travelingshorterdistancestodoso,butthatthosepatientswhoseenewprovidersaretravelingfarther;

thelattereffectisfairlysizeable,withanimpliedIVcoefficientofabout7miles,orabouttwothirdsof

themeandistancetraveledtonewproviders.

WefindthatpatientstravelshorterdistancesforoutpatientandERvisits,butthattheytravel

muchfarthertothehospital.TheimpliedIVcoefficientontraveltohospitalssuggeststhatthemarginal

patientswitchingtoalimitednetworkplantravelsalmost40milesfurthertothehospital.However,

thiscoefficienthasalargeconfidenceintervalaroundit,sowecannotexcludeconsiderablysmaller

increasesinthedistancetravelled.

Arelatedconcernisthatlimitednetworkplansrestrictpatientaccesstohighqualityproviders.

Inparticular,onemightbeconcernedthatlowercostsarecorrelatedwithlowerquality.Toassessthis

concern,weobtainedeightmeasuresofhospitalqualityforMassachusettshospitalsandmatchedthem

tothehospitalizationsinoursample.Ourqualitymeasuresincludethe30daymortalityrates

associatedwithhospitalizationsforheartattacks,heartfailure,andpneumoniaforeachhospital.In

addition,ourmeasuresincludedthe30dayreadmissionratesassociatedwithhospitalizationforheart

attack,heartfailure,pneumonia,hiporkneesurgery,andallcauses.

InthebottompanelofTable9,wereportresultsfromregressionsthatareexactlylikethe

regressionsinthetoppanelofTable9,exceptthattheyusequalitymeasuresforeachenrollee

(conditionalonhospitalization)asthedependentvariable.Theresultsarenotsuggestiveofanyimpact

onquality.Theresultsareuniformlystatisticallyinsignificant,andtheyareequallylikelytobenegative

(suggestinghigherquality)astheyaretobepositive(suggestinglowerquality).Inaddition,thepoint

estimatesareverysmallrelativetothemeans.Asaresult,weconcludethatenrollmentinlimited

networkplansisnotassociatedwithanychangeinthequalityofaccessibleinpatienthospitalcare.

 25

PartVII:HeterogeneityInResponses

Itispossiblethattheimpactsoflimitednetworkplansvaryconsiderablybytypeofpatient.We

considerinparticularthreetypesofheterogeneityintheanalysis.

Aparticularconcernisthatmovingtoamorelimitednetworkplanmayhavesignificant

negativeconnotationsforthemostillpatients.Table10ashowstheresultsseparatelybychronicillness

status.Wefindthattheoveralleffectonspendingissimilarforthechronicallyillandnonchronicallyill,

althoughgiventhelargerfirststageforthelattergroup,thisimpliessomewhatlargereffectsforthe

marginalchronicallyillindividualwhowasinducedtoswitchbythefinancialincentives.Most

importantly,wedonotfindanyevidencethatlimitednetworkplanscauseddifficultiesinphysician

accessforthechronicallyill.Indeed,wefindastrongshiftinspendingfromspecialiststoprimarycare

physicians,withspendingfallingconsiderablyfortheformerandrisingforthelatter.Wealsofindthat

forthechronicallyillthereisasignificantreductionininpatientspending,andnostatisticallysignificant

impactonemergencyroomuse.Takentogether,theseresultsdonotindicateanyparticularcausefor

concernforthechronicallyillfromswitchingtoalimitednetworkplan.

Thenextdimensionofheterogeneitythatweconsiderisbywhetherpatientsprimarycare

physicianswereincludedinlimitedplannetworks.Asdescribedearlier,wedividethesampleintothose

whocouldkeepboththeirinsurerandtheirprimarycareprovider,thosewhocouldkeeptheirprimary

careproviderbuttodosowouldhavetoswitchtoadifferentinsurer,andthosewhocouldnotkeep

theirprimarycareprovideriftheyswitchedtoalimitednetworkplan.

TheresultsofthisanalysisareshowninTable10b.Wefindthatthelargestdeclinesinspending

areforthosewhoareabletokeeptheirprimarycarephysician,eitherbymovingtothelimitednetwork

versionofthecurrentinsurersplanorbymovingtoanotherlimitednetworkplan.Forthesegroups

primarycarevisitsrise,andspecialistvisitsfall.Forthosestayingwiththesameinsurer,thereisamore

modestchangeinbothmeasures;forthosekeepingtheirdoctorbutswitchinginsurers,thereisamuch
 26

moresizeablereductioninspecialistuse.Thisisaninterestingfindingwhichsuggeststhatdoctors

changetheiruseofspecialistsdependingonthenetworkavailabilityofthosespecialists.

Forpatientswhocankeeptheirphysician,wealsoseenochangeinhospitalizationrates,buta

declineinhospitalspending,consistentwiththeideathatthelimitednetworkcausesthemtochoosea

lowercosthospital.Inaddition,therearestatisticallysignificantdeclinesinoutpatientspendingand

utilizationforthesegroups.

Forthosewhoseprimarycarephysiciandoesnotparticipateinalimitednetwork,incontrast,

theimpactonspendingispositiveandstatisticallyinsignificant,withlittlechangeofsignificanceinmost

categoriesofspending.Thisisdespitethefactthatthereisasignificantresponsetothefinancial

incentivesinthisgroupintermsofswitchingtoalimitednetworkplan.Thesefindingssuggestthatthe

poweroflimitednetworkplanstolowercostsdependscriticallyonthosewhoretaintheirprimarycare

physicianthroughtheswitchingprocess.Thisfurtherimpliesthattheabilitytoextrapolateourfindings

dependscriticallyonhowlimitednetworkplanslimitaccesstoprimarycarephysicians;thoseplansthat

haveverynarrownetworksofprimarycarephysiciansmaybelesssuccessfulincontrollingcosts.

Finally,weconsiderheterogeneitybytypeofillness:whichtypesofillnessaredrivingthese

results?Weclassifiedallofthespendinginoursamplein19majordiagnosticcategoriesbasedonthe

primaryICD9codes.13AsshowninTable11,therearenegativeimpactsonspendingforabouttwo

thirdsofourdiagnosiscategories.Noneofthecategorieswithpositivespendingimpactsaresignificant.

Incontrast,wefindfourcategoriesofspendingforwhichtheeffectsarenegativeandsignificant:

Neoplasms,RespiratoryDiagnoses,MusculoskeletalDiagnoses,andInjuriesandPoisonings.Thus,our

findingsarenotdrivenbyjustonecategoryofspending,butappearbroadlyspreadacrossthediagnosis

spectrum.

13
Spendingforanencountercouldbeclassifiedintomorethanonecategoryifthereweremultiplediagnoses.
Thiscouldhappenif,forexample,anindividualwashospitalizedwithmorethanoneprimarydiagnosisacrossthe
claimsgeneratedbythehospitalization.Asaresult,thesumofspendingacrossall19diagnosticcategoriesis
greaterthantotalspending.
 27

PartVIII:Conclusions

Thedebateovertheimpactoftheshifttonarrownetworkplanshaslargelyproceededinan

evidencevacuum.Thispaperattemptstomoveforwardourunderstandingofhowindividualschoose

suchplansandtheirimplicationforutilizationofhealthcare.

Wefirstfindthatpatientsareverypricesensitiveintheirdecisionstoswitchtolimitednetwork

plans,withapriceelasticityaboveone.Thereismodestadverseselectionassociatedwithsuchprice

incentives,asthosewhoaremosthealthyarethemostpricesensitive.

Wethenshowthatthelargepremiumdifferentialbetweenbroadandlimitednetworkplansis

drivennotbyselectionbutbyrealreductionsinspendingamongthoseinducedtoswitchplans.This

reductioninspendingcomesfrombothreductionsinpricespaidandquantityofcareused.The

reductioninspendingdoesnotappeartocomefromreducedaccesstoprimarycare;indeed,useof

primarycareandspendingonsuchservicesrisesforthoseswitchingtolimitednetworkplans.Rather,

thereductionarisesfromlessuseofspecialistsandhospitalcare.Thefactthatprimarycareuseis

rising,whileemergencyroomandhospitalspendingisfalling,suggeststhatthemovetolimitednetwork

plansisnotadverselyimpactinghealth,althoughweareunabletodemonstratehealtheffectswithany

certainty.Wefindthatdistancetraveledfallsforprimarycareandrisesfortertiarycare,althoughthere

isnoevidenceofadecreaseinthequalityofhospitalsusedbypatients.

Wealsofoundthatthepositiveeffectsonprimarycareandreductionsinspendingon

specialist/hospitalcareoccurforbothmoreandlesshealthypatients,andthatthespendingreduction

holdsforabroadspectrumofillnesses.Wedofind,however,thatthespendingreductionisdriven

primarilybythosewhoareabletokeeptheirprimarycarephysicianwhenmovingtoalimitednetwork

plan.Takentogetherwithouroverallfindingsonprimarycare,weconcludethattherealsavingsfrom

limitednetworkplansarisesfromrestrictionsdownstreamfromtheprimarycareprovider.
 28

Onenaturalquestionthatarisesfromourfindingsiswhetherthepremiumincentivesprovided

bytheGICwerefiscallybeneficialtothestateofMassachusetts.Theanswertothisquestionappearsto

beYes.Wecalculatethattheemployerpremiumcontributionsthatwerepaidforallindividualand

familyplanenrolleesinfiscalyear2012was1.2%percentlowerthanitwouldhavebeenintheabsence

ofthepremiumholiday.This1.2%reductioninemployerpremiumcostscombinesa2.8%reductionin

theemployershareofthelowerpremiuminlimitednetworkplansarisingfromthe11.6percentage

pointincreaseinlimitednetworkenrollmentthatwecalculatedinourfirststageanda1.6%increasein

theemployersshareofallpremiumsduetothepremiumholiday.Infutureyears,ifthesamegroupof

enrolleeswhorespondedtothepremiumholidayweretoremainenrolledinlimitednetworkplans,the

savingswouldbeexpectedtobeapproximately2.8%.Whilesomeofthemarginalenrolleesmight

switchbacktobroadnetworkplans,thereisagreatdealofinertiaininsuranceplanenrollment.14We

haveanalyzedaggregateenrollmentdatafortheyearsfollowingthepremiumholiday,andthepatterns

areconsistentwiththeideathatthevastmajorityofthenewlimitednetworkenrolleeswhowere

inducedtoswitchbythepremiumholidayremainedinlimitednetworkplansinsubsequentyears.

While31.2%ofallenrolleescoveredbyactivestateemployeeswereenrolledinlimitednetworkplansin

FY2012(theyearthatthepremiumholidaytookeffect),30.4%wereenrolledinlimitednetworkplans

inFY13and31.1%wereenrolledinFY14.Bycomparison,25.1%ofallenrolleescoveredbyactive

municipalemployeeswereinlimitednetworkplansinFY12,24.5%ofthemwereenrolledinlimited

networkplansinFY13,and26.9%wereenrolledinFY14.Thisinertiainplanenrollmentssuggeststhat

thefiscalbenefitsofthe2012premiumholidaywerelikelymuchlargerinsubsequentyearswhenthe

premiumholidaywasnolongerinplace,sincethestatebenefitedfromareducedpremiumbillwithany

offsettingincreaseintheemployershareofthepremium.

14
Inoursample,only3%ofthosewhowereenrolledinbroadnetworkplansinFY2010switchedtolimited
networkplansforFY2011;similarly,only1%ofthosewhowereenrolledinlimitednetworkplansinFY2010
switchedtoabroadnetworkplaninFY2011.
 29

Themostimportantcaveattoourresultsisthattheyapplytooneparticularexample,andthat

wemaynotbeabletoextrapolatethemtootherlimitednetworkplans,forexamplethosefeaturedon

stateexchanges.Animportantgoalforfutureworkshouldbetoextendthisanalysistothoseother

examples.ThisshouldbefeasiblegiventhatthetaxcreditsavailableundertheACAprovidedistinctly

nonlinearpricedifferentialsacrosshealthinsuranceoptions,allowingfutureresearcherstoassesshow

thoseinducedintolimitednetworkplansonexchangesarefaringintermsofhealthcarespendingand

outcomes.

 30

References

Buntin,MelindaBeeuwkesandZaslavsky,AlanM.,2004."Toomuchadoabouttwopartmodelsand
transformation?:ComparingmethodsofmodelingMedicareexpenditures,"JournalofHealth
Economics,23(3):525542.

CommonwealthofMassachusettsGroupInsuranceCommission(2012).FiscalYear2012Annual
Report:LeadingtheWayinHealthCare,December2012.Downloadedfrom
http://www.mass.gov/anf/docs/gic/annualreport/arfy2012.pdfonMarch5,2014.

CommonwealthofMassachusettsGroupInsuranceCommission(2011).FiscalYear2010Annual
Report:SuccessfullyNavigatingRoughFiscalandHealthCareSeas,January2011.Downloaded
fromhttp://www.mass.gov/anf/docs/gic/annualreport/arfy2010.pdfonMarch7,2014.

Frank,Matt,JohnHsu,MaryBethLandrumandMichaelChernew(2014).TheEffectsofaTiered
NetworkonHospitalchoiceandUtilizationofDiscretionaryandNondiscretionaryCare,mimeo,
HarvardUniversity.

Glied,Sherry(2000).ManagedCare,inCulyer,AnthonyJ,Newhouse,JosephP.(Eds.),Handbookof
HealthEconomics.Amsterdam:ElsevierNorthHolland:707753.

Goldman,DanaP.,GeoffreyF.Joyce,JoseJ.Escarce,JenniferE.Pace,MatthewD.Colomom,Marianne
Laouri,PamelaB.Landsman,andStevenM.Teutsch(2004).PharmacyBenefitsandtheUseof
DrugsbytheChronicallyIll,JournaloftheAmericanMedicalAssociation291:23442351

Jan,Tracy(2014).WithHealthlaw,LessEasyAccessinN.H,TheBostonGlobe,January20,2014,
availableathttp://www.bostonglobe.com/news/nation/2014/01/20/narrowhospitalnetworks
newhampshiresparkoutragepoliticalattacks/j2ufuNSf9J2sdEQBpgIVqL/story.htmlAccessed
May7,2014

Luhby,Tami(2014).GotObamacare,CantFindDoctors,cnn.com,March19,2014,Availableat
http://money.cnn.com/2014/03/19/news/economy/obamacaredoctors/AccessedMay7,
2014.

Manning,WillardG.andJohnMullahy(2001).EstimatingLogModels:ToTransformorNotto
Transform?,JournalofHealthEconomics20(4):461494.

Manning,WillardG.,JosephP.Newhouse,NaihuaDuan,EmmettB.Keeler,andArleenLeibowitz(1987).
HealthInsuranceandtheDemandforMedicalCare:EvidencefromaRandomizedExperiment,
TheAmericanEconomicReview77(3):251277.

Pickert,Kate(2014).KeepingYourDoctorUnderObamacareisNoEasyFeat,Time,January1,
2014,availableathttp://swampland.time.com/2014/01/01/keepingyourdoctorunder
obamacareisnoeasyfeat/AccessedMay7,2014.

Weisman,RobertandChelseaCanaboy(2011).TieredHealthPlansCuttingCosts,RestrictingOptions,
TheBostonGlobe,November28,2011,p.A1.

 31

Figure1

PanelA:TheMonthlySavingsfromSwitchingtoaLimitedNetwork,
asaPercentageofAverageBroadNetworkPremiumContribution

MonthlySavingsfromSwitching
55

50

45
Percent

40

35

30

25

20
2010 2011 2012
FiscalYear

Municipalities State


PanelB:EnrollmentinLimitedNetworkPlans,asaPercentageofTotalEnrollment

EnrollmentinLimitedNetworkPlans
35
30
25
Percent

20
15
10
5
0
2010 2011 2012
FiscalYear

Municipalities State
 32

Figure2:TotalQuarterlySpendingperCapita

TotalSpendingPerCapita
1500

1400

1300

1200

1100

1000

900
2010q1

2010q2

2010q3

2010q4

2011q1

2011q2

2011q3

2011q4

2012q1

2012q2

2012q3

2012q4
FiscalYear

Municipality State


 33

Figure3:DecompositionofChangesinTotalQuarterlySpendingperCapita
PanelA

QuantitiesFixedat2010BroadNetwork
Level,PricesVarying
1450

1400

1350

1300

1250

1200

1150

1100
2010q1

2010q2

2010q3

2010q4

2011q1

2011q2

2011q3

2011q4

2012q1

2012q2

2012q3

2012q4
Municipality State


PanelB

QuantitiesVarying,
PricesFixedat2010BroadNetworkLevel
1250

1200

1150

1100

1050

1000
2010q1

2010q2

2010q3

2010q4

2011q1

2011q2

2011q3

2011q4

2012q1

2012q2

2012q3

2012q4

Municipality State
 34

Table1:Detailsof2010GICPlanOptions

Limited
Enrollmentin TypeofPlan Network
PlanName June2010 Plan
FallonCommunityHealthPlanDirectCare 1% HMO Yes
FallonCommunityHealthPlanSelectCare 3% HMO No
HarvardPilgrimIndependencePlan 26% PPO No
HarvardPilgrimPrimaryChoicePlan 0% HMO Yes
HealthNewEngland 6% HMO Yes
NeighborhoodHealthPlan 1% HMO Yes
TuftsHealthPlanNavigator 31% PPO No
TuftsHealthPlanSpirit 0% HMOtype Yes
UniCareStateIndemnityPlanBasic 17% Indemnity No
UnicareStateIndemnityPlanCommunityChoice 6% PPOtype Yes
UnicareStateIndeminityPlanPLUS 9% PPOtype No

 35

Table2:MeasuresofNetworkBreadth

Physician Hospital
>5Claims >10Claims >5Claims >10Claims

Averageacrossallplans
Broad 0.250 0.212 0.776 0.710
Narrow 0.135 0.107 0.541 0.419
HarvardPilgrim
Broad:Independence 0.367 0.315 0.963 0.901
Narrow:PrimaryChoice 0.110 0.077 0.570 0.418
Tufts
Broad:Navigator 0.351 0.312 0.827 0.815
Narrow:Spirit 0.054 0.034 0.329 0.158
Unicare
Broad:Basic 0.263 0.220 0.926 0.864
Broad:Plus 0.199 0.160 0.802 0.728
Narrow:CommunityChoice 0.166 0.128 0.650 0.563
Fallon
Broad:Select 0.069 0.052 0.360 0.240
Narrow:Direct 0.066 0.051 0.400 0.200
OtherNarrow
HealthNewEngland 0.353 0.313 0.923 0.923
NeighborhoodHealthPlan 0.059 0.041 0.373 0.253
Notes:Thistableshowstheproportionofproviderslocatedinthecountieswheretheinsuranceplanoperatesfor
whomweobserveatleast5(or10)innetworkclaims.

 36

Table3:SummaryStatistics

Mean
Variable (StandardDeviation)
EnrolledinLimitedNetworkPlan 0.201
(0.400)

Savingsfromswitchingtolimitednetworkplan 36.55%
(asa%ofemployeecontributiontobroadnetwork (9.64)
plan)

Spending Visits
Totalexpenses $4,811
(15,132)
Officevisits $1,084 7.36
(2,155) (9.69)
PrimaryCare $323 2.17
(653) (2.92)
Specialist $676 4.60
(1799) (8.31)
Other $85 0.55
(762) (2.56)
OldProvider $771 5.64
(1,937) (8.37)
NewProvider $304 1.43
(546) (1.68)
InpatientHospitalization $864 0.053
(8,117) (0.297)
OutpatientHospital $1,443 3.76
(7,200) (8.12)
EmergencyRoom $235 0.220
(995) (0.635)
Lab&Xrays $69 0.550
(336) (1.463)
Drugs $900 11.69
(4,417) (17.03)
Other $210 0.70
(3,324) (4.38)
Numberofobservations 479,196

 37

Table4FirstStageRegressions

Differenceindifference Fullvariation
Stateemployees*Post 0.1165**
(0.0036)
RelativePriceofLimitedPlans 0.0070**
(0.0002)
Male 0.0011** 0.0011**
(0.0004) (0.0004)
Age1929 0.0067** 0.0068**
(0.0010) (0.0010)
Age3039 0.0236** 0.0236**
(0.0036) (0.0036)
Age4049 0.0019 0.0019
(0.0020) (0.0020)
Age5059 0.0212** 0.0212**
(0.0037) (0.0037)
Age6069 0.0546** 0.0545**
(0.0035) (0.0035)
Age69+ 0.0812** 0.0810**
(0.0069) (0.0069)
Familyplan 0.0006 0.0092**
(0.0022) (0.0022)
NumberObs 479,196 479,196
Notes:Eachcolumnshowscoefficients(andstandarderrors)fromasingleOLSregression.Othercontrolvariables
includeafullsetofmunicipalityandyearfixedeffectsandcontrolsfordateofhire.Theomittedagecategoryis
<19yearsold.Standarderrorsareclusteredonmunicipality.Thesampleincludesallcontinuouslyenrolledactive
employeesoverthethreeyearperiodfromfiscalyear2010tofiscalyear2012;theunitofobservationisaperson
year.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level

 38

Table5HeterogeneityinFirstStage

Differenceindifference Fullvariation
BaseEstimates 0.116** 0.0070**
(0.004) (0.0002)

ByChronicIllness
NoChronicIllness 0.104** 0.0063**
(N=132,727) (0.003) (0.0002)

ChronicIllness 0.121** 0.0073**
(N=346,469) (0.004) (0.0002)

BybroadnetworkinsurancecompanyinFY2010
Fallon 0.236** 0.0139**
(N=13,695) (0.012) (0.0008)

Harvard 0.199** 0.0117**
(N=121,992) (0.007) (0.0005)

Tufts 0.109** 0.0068**
(N=169,065) (0.009) (0.0005)

Unicare 0.081** 0.0056**
(N=102,381) (0.006) (0.0007)

ByPCPsavailabilityinalimitednetwork
PCPisinthelimitednetworkplan 0.168** 0.0100**
offeredbycurrentinsurer(N=187,656) (0.006) (0.0003)

PCPisinalimitednetworkplan 0.127** 0.0077**
offeredbyadifferentinsurer (0.010) (0.0006)
(N=76,125)

PCPisnotinalimitednetworkplan 0.101** 0.0061**
(N=43,197) (0.002) (0.0002)
Notes:Eachcellshowsthecoefficient(andstandarderror)fromasingleregression.Inthefirstcolumn,the
coefficientisontheinteractionbetweenstateemployeeandpost;inthesecondcolumn,thecoefficientison
therelativepriceoflimitednetworkplans.Eachrowshowsresultsforadifferentsubsample.Controlvariables
includegender,agegroup,enrollmentinafamilyplan,dateofhire,andafullsetofmunicipalityandyearfixed
effects.CoefficientsareestimatedusingOLS.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level

 39

Table6BasicSpendingResults

Differenceindifference FullVariation
TotalSpending 0.042* 0.0029**
(0.022) (0.0013)
OfficeVisits 0.018* 0.0012*
(0.010) (0.0006)
InpatientHospitalization 0.056 0.0048
(0.071) (0.0043)
OutpatientHospital 0.050* 0.0033**
(0.025) (0.0015)
EmergencyRoom 0.095* 0.0054*
(0.055) (0.0032)
Lab&XRay 0.083* 0.0047
(0.049) (0.0029)
Drugs 0.003 0.0003
(0.017) (0.0011)
Other 0.111** 0.0074**
(0.054) (0.0036)
Numberofobservations 479,196 479,196
Notes:Eachcellshowscoefficients(andstandarderrors)fromasingleregression.Inthefirstcolumn,the
coefficientisontheinteractionbetweenstateemployeeandpost;inthesecondcolumn,thecoefficientison
therelativepriceoflimitednetworkplans.Eachrowshowsresultsforspendingonadifferenttypeofservice.
Controlvariablesincludegender,agegroup,enrollmentinafamilyplan,dateofhire,andafullsetofmunicipality
andyearfixedeffects.Thesampleincludesallcontinuouslyenrolledactiveemployeesoverthethreeyearperiod
fromfiscalyear2010tofiscalyear2012;theunitofobservationisapersonyear.Coefficientsareestimatedusing
GLM.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level

 40

Table7BroaderMeasuresofUtilizationforDDModel

TotalSpending AnyVisits NumberofVisits CostPerVisit

(GLM) (OLS) (OLS) (OLS)
OfficeVisits 0.018* 0.0001 0.154* 0.127
(0.010) (0.0026) (0.083) (2.087)
InpatientHospitalization 0.056 0.0005 0.0006 861.59
(0.071) (0.0020) (0.0027) (845.44)
OutpatientHospital 0.050* 0.0086 0.103 20.00*
(0.025) (0.0053) (0.071) (11.51)
EmergencyRoom 0.095* 0.0026 0.0090* 67.24
(0.055) (0.0029) (0.0046) (42.15)
Lab&XRay 0.083* 0.0019 0.036 4.60
(0.049) (0.0073) (0.022) (4.05)
Drugs 0.003 0.0039 0.386** 2.08
(0.017) (0.0042) (0.113) (1.82)
Other 0.111** 0.034** 0.075** 4.19
(0.054) (0.010) (0.027) (21.45)
Numberofobservations 479,196 479,196 479,196 Varies
Notes:Eachcellshowsthecoefficient(andstandarderror)ontheinteractionbetweenstateemployeeand
postfromasingledifferenceindifferenceregression.Eachrowshowsresultsforadifferentservice;each
columnshowsadifferentmeasureofutilizationforthatservice.Controlvariablesincludegender,agegroup,
enrollmentinafamilyplan,dateofhire,andafullsetofmunicipalityandyearfixedeffects.Thesampleincludes
allcontinuouslyenrolledactiveemployeesoverthethreeyearperiodfromfiscalyear2010tofiscalyear2012;the
unitofobservationisapersonyear.CoefficientsinthefirstcolumnareestimatedusingGLM;resultsintheother
columnsareestimatedusingOLS.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level

 41

Table8ResultsforOfficeVisitUtilizationbyTypeofPhysician

TotalSpending AnyVisits NumberofVisits CostPerVisit

(GLM) (OLS) (OLS) (OLS)

PrimaryCarevs.Specialistvs.Other
PrimaryCare 0.030** 0.002 0.040* 1.95
(0.015) (0.005) (0.023) (2.09)
Specialist 0.051** 0.007 0.153** 3.27
(0.013) (0.007) (0.069) (3.54)
Other 0.014 0.0001 0.027* 18.87**
(0.077) (0.0046) (0.015) (6.38)

Oldvs.NewProviders
OldProviders 0.034** 0.004 0.142** 2.27
(0.011) (0.003) (0.042) (1.83)
NewProviders 0.056** 0.016** 0.051* 7.13**
(0.013) (0.007) (0.028) (1.40)
Numberofobservations 479,196 479,196 479,196 Varies
Notes:Eachcellshowsthecoefficient(andstandarderror)ontheinteractionbetweenstateemployeeand
postfromasingledifferenceindifferenceregression.Eachrowshowsresultsforadifferenttypeofofficevisit;
eachcolumnshowsadifferentmeasureofutilizationforthatservice.Controlvariablesincludegender,agegroup,
enrollmentinafamilyplan,dateofhire,andafullsetofmunicipalityandyearfixedeffects.Thesampleincludes
allcontinuouslyenrolledactiveemployeesoverthethreeyearperiodfromfiscalyear2010tofiscalyear2012;the
unitofobservationisapersonyear.CoefficientsinthefirstcolumnareestimatedusingGLM;resultsintheother
columnsareestimatedusingOLS.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level

 42

Table9:ImpactonDistanceTraveledandHospitalQuality

DependentVariable MeanofDependent DDCoefficient

Variable
MeasuresofDistanceTravelled
DistancetoOfficeVisits 9.82 0.114
(9.45) (0.131)
DistancetoPrimaryCareVisits 8.19 0.659**
(10.69) (0.278)
DistancetoSpecialists 10.53 0.038
(10.11) (0.183)
DistancetoOtherOfficeVisits 9.88 0.151
(15.59) (0.447)
DistancetoOldProviders 9.49 0.363**
(10.27) (0.147)
DistancetoNewProviders 12.59 0.857**
(12.82) (0.377)
DistancetoInpatientHospitalization 28.10 4.538**
(26.81) (2.149)
DistancetoOutpatientHospital 14.58 1.193**
(13.00) (0.333)
DistancetoEmergencyRoom 22.23 0.774
(22.43) (0.485)
MeasuresofHospitalQuality
30DayMortalityRate,AMI 13.81 0.002
(1.24) (0.040)
30DayMortalityRate,Heartfailure 10.34 0.031
(1.28) (0.078)
30DayMortalityRate,Pneumonia 11.04 0.062
(1.50) (0.112)
30DayReadmissionRate,AMI 19.07 0.054
(1.25) (0.067)
30DayReadmissionRate,HeartFailure 23.68 0.016
(1.46) (0.041)
30DayReadmissionRate,Pneumonia 18.24 0.044
(1.27) (0.050)
30DayReadmissionRate,HiporKneeSurgery 5.51 0.026
(0.68) (0.018)
30DayReadmissionRate,AllCause 16.46 0.035
(1.05) (0.039)
Notes:Eachcellshowsthecoefficient(andstandarderror)ontheinteractionbetweenstateemployeeand
postfromasingledifferenceindifferenceregression.Eachrowshowsresultsfordistancetoadifferenttypeof
provider.Controlvariablesincludegender,agegroup,enrollmentinafamilyplan,dateofhire,andafullsetof
municipalityandyearfixedeffects.Thesampleincludesallcontinuouslyenrolledactiveemployeesoverthethree
yearperiodfromfiscalyear2010tofiscalyear2012;theunitofobservationisapersonyear.Coefficientsare
estimatedusingOLS.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level
 43

Table10a:HeterogeneityinResultsbyChronicIllness

NotChronicallyIll ChronicallyIll
Spending Visits Distance Spending Visits Distance
(GLM) (OLS) (OLS) (GLM) (OLS) (OLS)
TotalSpend 0.039* 0.043
(0.023) (0.033)
OfficeTotal 0.023** 0.066 0.137 0.013 0.349** 0.083
(0.011) (0.092) (0.183) (0.016) (0.168) (0.103)
Primary 0.022 0.060** 0.0851** 0.035* 0.006 0.474
(0.020) (0.026) (0.139) (0.018) (0.039) (0.463)
Specialist 0.053** 0.085 0.210 0.045** 0.303** 0.095
(0.015) (0.071) (0.244) (0.021) (0.143) (0.133)
Other 0.032 0.031 3.314** 0.051 0.015 1.733**
(0.103) (0.021) (0.370) (0.057) (0.035) (0.417)
Old 0.054** 0.124 0.467** 0.009 0.069 0.252
(0.021) (0.130) (0.178) (0.024) (0.182) (0.163)
New 0.037 0.046 0.566* 0.087** 0.079** 1.282**
(0.024) (0.033) (0.329) (0.009) (0.033) (0.453)
Inpatient 0.053 0.001 5.603* 0.137* 0.005 3.99*
(0.088) (0.002) (2.840) (0.073) (0.006) (2.117)
Outpatient 0.045 0.109* 1.072** 0.051 0.088 1.291**
(0.038) (0.057) (0.253) (0.034) (0.143) (0.457)
ER 0.139** 0.007 2.432** 0.016 0.012 1.151
(0.059) (0.005) (0.768) (0.061) (0.008) (0.725)
Lab 0.040 0.024* 0.159** 0.064
(0.055) (0.012) (0.065) (0.057)
Drugs 0.00003 0.250** 0.007 0.710**
(0.0282) (0.110) (0.029) (0.242)
Other 0.151* 0.049* 0.106* 0.141**
(0.077) (0.029) (0.055) (0.044)
N 346,469 346,469 Varies 132,727 132,727 Varies
Notes:Eachcellshowsthecoefficient(andstandarderror)ontheinteractionbetweenstateemployeeand
postfromasingledifferenceindifferenceregression.Eachrowshowsresultsforadifferentservice;each
columnshowsadifferentmeasureofutilizationforthatserviceforoneofthetwosubsamples.Controlvariables
includegender,agegroup,enrollmentinafamilyplan,dateofhire,andafullsetofmunicipalityandyearfixed
effects.CoefficientsinthefirstandfourthcolumnsareestimatedusingGLM;resultsintheothercolumnsare
estimatedusingOLS.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level

 44

Table10b:HeterogeneityinResultsbyPCPsavailabilityinlimitednetworkplans
Limitednetwork,sameinsurer Limitednetwork,differentinsurer Nolimitednetwork
Spending Visits Distance Spending Visits Distance Spending Visits Distance
(GLM) (OLS) (OLS) (GLM) (OLS) (OLS) (GLM) (OLS) (OLS)
TotalSpend 0.072** 0.130** 0.047
(0.024) (0.055) (0.045)
OfficeTotal 0.012 0.199 0.119 0.047** 0.414** 0.347 0.006 0.0001 0.025
(0.015) (0.116) (0.156) (0.019) (0.192) (0.301) (0.053) (0.207) (0.569)
Primary 0.032** 0.072* 0.637* 0.046 0.068 0.225 0.053 0.018 0.789
(0.010) (0.038) (0.338) (0.036) (0.061) (0.311) (0.065) (0.075) (0.658)
Specialist 0.039* 0.196* 0.131 0.122** 0.422** 0.573 0.033 0.004 0.354
(0.021) (0.114) (0.112) (0.027) (0.185) (0.474) (0.072) (0.183) (0.615)
Other 0.204 0.066** 6.548** 0.168 0.025 0.820 0.041 0.064 0.409
(0.159) (0.029) (2.370) (0.149) (0.169) (0.717) (0.225) (0.066) (2.101)
Old 0.007 0.042 0.317 0.071** 0.118* 0.269 0.189* 0.674* 0.370
(0.017) (0.114) (0.261) (0.022) (0.067) (0.337) (0.097) (0.396) (0.569)
New 0.086** 0.081* 1.011** 0.055 0.129** 0.023 0.059 0.067 0.105
(0.025) (0.046) (0.342) (0.086) (0.053) (0.456) (0.069) (0.107) (0.480)
Inpatient 0.270** 0.001 2.547 0.097 0.006 6.720 Insufficient 0.002 7.221
(0.133) (0.005) (2.703) (0.179) (0.005) (4.044) data (0.009) (6.340)
Outpatient 0.095** 0.164* 3.152** 0.202** 0.271* 0.920* 0.171** 0.451* 0.301
(0.036) (0.086) (1.229) (0.086) (0.146) (0.481) (0.085) (0.219) (0.712)
ER 0.121 0.012 0.207 0.289** 0.030 0.875 Insufficient 0.020 0.770
(0.074) (0.008) (0.802) (0.086) (0.019) (1.983) data (0.017) (1.739)
Lab 0.110 0.051 0.134 0.028 0.019 0.011
(0.082) (0.029) (0.140) (0.028) (0.120) (0.052)
Drugs 0.021 0.223 0.002 0.602** 0.054 0.371
(0.024) (0.206) (0.056) (0.276) (0.064) (0.336)
Other 0.041 0.060* 0.174 0.050 0.190* 0.141**
(0.038) (0.032) (0.175) (0.086) (0.104) (0.055)
N 187,656 187,656 Varies 76,125 76,125 Varies 43,197 43,197 Varies
Notes:Eachcellshowsthecoefficient(andstandarderror)ontheinteractionbetweenstateemployeeandpostfromasingledifferenceindifferenceregression.Eachrow
showsresultsforadifferentservice;eachcolumnshowsadifferentmeasureofutilizationforthatserviceforoneofthreesubsamples.Controlvariablesincludegender,age
group,enrollmentinafamilyplan,dateofhire,andafullsetofmunicipalityandyearfixedeffects.Coefficientsinthefirst,fourth,andseventhcolumnsareestimatedusing
GLM;resultsintheothercolumnsareestimatedusingOLS.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level
 45

Table11:HeterogeneityinResultsbyDiagnosis

Meanof Effecton
Dependent Total
DiagnosticCategory Variable Spending
Infectiousandparasiticdiseases $102 0.215
(2,496) (0.181)
Neoplasms $507 0.348**
(6,294) (0.139)
Endocrine,nutritionalandmetabolicdiseases&immunitydisorders $298 0.010
(5,087) (0.100)
Diseasesofthebloodandbloodformingorgans $96 0.044
(3,407) (0.128)
Mentaldisorders $255 0.055
(2,828) (0.046)
Diseasesofthenervoussystem $237 0.208
(3,501) (0.164)
Diseasesofthesenseorgans $139 0.038
(1,141) (0.048)
Diseasesofthecirculatorysystem $484 0.036
(6,048) (0.061)
Diseasesoftherespiratorysystem $371 0.140*
(5,225) 0.077)
Diseasesofthedigestivesystem $361 0.053
(4,246) (0.096)
Diseasesofthegenitourinarysystem $379 0.013
(4,726) (0.103)
Complicationsofpregnancy,childbirth,andthepuerperium $108 Insufficient
(1,527) data
Diseasesoftheskinandsubcutaneoustissue $137 0.022
(1,840) (0.080)
Diseasesofthemusculoskeletalsystemandconnectivetissue $643 0.155**
(4,275) (0.054)
Congenitalabnormalities $66 0.065
(2,432) (0.202)
Certainconditionsoriginatingintheperinatalperiod $13 Insufficient
(544) data
Symptoms,signs,andilldefinedconditions $893 0.072
(6,775) (0.086)
 46

Injuryandpoisoning $393 0.090*

(4,632) (0.048)
Externalcausesofinjuryandsupplementalclassification $891 0.052
(6,557) (0.042)
N 479,196 479,196
Notes:Eachcellinthesecondcolumnshowsthecoefficient(andstandarderror)fromasingleregression;meansofthe
dependentvariablearereportedinthefirstcolumn.Eachrowshowsresultsforspendingonadifferentdiagnostic
category.Controlvariablesincludegender,agegroup,enrollmentinafamilyplan,dateofhire,andafullsetof
municipalityandyearfixedeffects.Thesampleincludesallcontinuouslyenrolledactiveemployeesoverthethreeyear
periodfromfiscalyear2010tofiscalyear2012;theunitofobservationisapersonyear.Coefficientsareestimatedusing
GLM.Standarderrorsareclusteredonmunicipality.
*denotessignificanceatthe10%level
**denotessignificanceatthe5%level