1

NUDGINGMEDICALPROVIDERSTOADOPTANDSUSTAINBETTERQUALITYCAREPRACTICES

ByPABLOCELHAY,PAULGERTLER,PAULAGIOVAGNOLIANDCHRISTELVERMEERSCH*

February21,2017Abstract:Weshowthatcostsofadjustmentasopposedtolowreturnslikelyexplainwhybetterqualitycarepracticesdiffuseslowlyinthemedicalindustry.UsingarandomizedfieldexperimentconductedinArgentina,wefindthattemporaryfinancialincentivespaidtohealthclinicsfortheearlyinitiationofprenatalcare‘nudged’providerstotestanddevelopnewdatadrivenstrategiesto locateandencourage likelypregnantwomentoseekcare inthe first trimesterofpregnancy.Theseinnovationsraisedtherateofearlyinitiationofprenatalcareby34%whiletheincentiveswere being paid in the treatment period. We followhealth clinics over time and find that thisincreasepersistedforatleast24monthsaftertheincentivesended.Intheabsenceofincentives,eventhoughitisintheclinics’interesttostimulateearlyinitiationofcare,thepresenceofhardtochangehabitsandcostofexperimentationmadeittooexpensivetodevelopandimplementnewmethods to increase early initiation of care. Despite the large increases in early initiation ofprenatalcare,wefindnoeffectsonhealthoutcomes.(JELI12,I13,I15,I18)*Celhay:AssistantProfessoratPontificiaUniversidadCatólicadeChile (email:pacelhay@uc.cl);Gertler:LiKaShingProfessorofEconomicsattheUniversityofCalifornia,Berkeley(emal:gertler@haas.berkeley.edu);Giovagnoli: Economist at The World Bank (email: pgiovagnoli@worldbank.org); Vermeersch: SeniorEconomist atTheWorldBank (cvermeersch@worldbank.org). The field experimentwasdevelopedundertheleadershipofMartinSabignoso,NationalCoordinatorofPlanNacerandHumbertoSilva,NationalHeadofStrategicPlanningofPlanNacer,MinistryofHealth,Argentina.LuisLopezTorresandBettinaPetrellafromtheMisionesOfficeofPlanNaceroversawtheimplementationofthepilotandfacilitatedaccesstoprovincialdata.AlvaroS.Ocarizof the InformationTechnologyunitatCentral ImplementationUnitat theMinistryofHealth provided support in identifying and gaining access to other sources of data. Vanina Camporeale,Fernando Bazán Torres, Ramiro Flores Cruz, Santiago Garriga, Alfredo Palacios, Daniela Romero, RafaelRamirez, SilvestreRios Centeno,GabrielaMoreno, andAdamRoss provided excellent projectmanagementsupport and research assistance. The authors acknowledge the contribution of Sebastian Martinez to thedesignof thepilot. The authors also thankNavaAshraf,NedAugenblick,OrianaBandiera,DanBlack,NickBloom, Megan Busse, Stefano DellaVigna, Damien de Walque, Nicolás Figueroa, Emanuela Galasso, JeffGrogger, Yona Rubenstein, John Van Reenan, and Petra Vergeer, as well as participants in seminars atInstitute for Fiscal Studies, LSE, Northwestern University, Stanford University, UC Berkeley, University ofChicago, Pontificia Universidad Católica de Chile, and Universidad Adolfo Ibañez for helpful comments.Finally,theauthorsgratefullyacknowledgefinancialsupportfromtheHealthResultsInnovationTrustFund(HRITF)andtheStrategic ImpactEvaluationFund(SIEF)of theWorldBank.Theauthorsdeclare that theyhavenofinancialormaterialinterestsintheresultsofthispaper.

2

I. Introduction

A well-documented feature of technological change is its remarkably slow diffusion.1 One

reason could be that innovation may have large costs of adjustment above and beyond

acquisition expenditures. Firms have to design, test, and learn how to best incorporate an

innovation into existing practices. Firms may also need to purchase complementary

technology (Rosenberg 1982, David 1990, Bresnahan and Trajtenberg 1995). Productivity

mightalsobe lowerduringaperiodofadjustmentwhile the firm learnshowbest touse the

innovation and themarginal productivity of the new technology. Managementmay have to

overcome costly informational deficits (Bloom et al. 2012) and behavioral barriers such as

presentbias(Dufloetal.2011). Innovationmayconfrontworker’sresistanceifpartoftheir

wage varieswith performance (Lazonick 1979, Atkin et al. 2015) or if they have developed

strongworkhabits. Thoughitmayseemthatsuchcostsofadjustmentareeasytoovercome,

theycanbe largeenough topreventproductiveandprofitable innovations.2 Change ishard,

andevensmallcostsofadjustmentmayinhibitchangesinfavorofmaintainingthestatusquo

(DellaVigna2009;ThalerandSunstein2009).

Slowdiffusionofbetterqualitymedicalcarepracticesisaglobalissueasevidencedby

theremarkablylowlevelofcompliancewithstateoftheartClinicalPracticeGuidelines(CPGs)

(Figure 1).3 While low CPG compliance may in part reflect a lack of knowledge or slow

diffusion of information (Phelps 2000), evidence shows that practitioners often provide a

standardof carewellbelowtheir levelofknowledgeofCPGs.4 Inasystematic reviewof the

literature,Cabanaetal. (1999)report thatpsychologicalcostssuchasresistance tochanging

1Slowadoptionofnewtechnologiesbyfirmshasbeenextensivelydocumentedinagriculture,manufacturingand medicine. Surveys and studies of slow diffusion include Ryan and Gross (1943), Griliches (1957),Mansfield(1961),ColemanandMenzel(1966),Rosenberg(1972),ParenteandPrescott(1994),FosterandRosenzweig(1995),Geroski(2000),HallandKhan(2003),Hall(2005),CominandHobijn(2010),ConleyandUdry(2010).2 The organizational literature refers to the phenomenon of high fixed costs of preventing the adoptingprofitableinnovationsasorganizationalinertia(HannanandFreeman1984;CarrollandHannan2000).3CPGsdefinemedicalcareproductionpossibilityfrontiersinthattheyprescribetheclinicalcontentofcarethat maximizes the likelihood of successful health outcomes based on medical science, clinical trials, andpractitionerconsensus.LocalCPGsareregularlyupdatedandserveasthebasisoftraininginmedicalschoolsandpractitionerrefreshercourses.4SeeDasandHammer(2005);DasandGertler(2007);Das,HammerandLeonard(2008);BarberandGertler(2009);LeonardandMasatu(2010);GertlerandVermeersch(2012);andMonahanetal.(2015).

3

existing practice patterns are one of the most important barriers to CPG adherence.5 For

example, Grol and Grimshaw (2003) report that 49% of UK nurses and doctors said that

resistance to changingoldhabitswas amajorobstacle to complyingwithnewhandhygiene

guidelines. InarecentstudyofU.S.hospitals,SkinnerandStaiger(2015)statethatthe large

differencesobservedintheadoptionofaspirinsandbeta-blockersareinpartduetoresistance

tochangingoldhabitsofphysicians.6

In this paper we examine the short-term and long-term effects of paying temporary

financialincentivestomedicalcareprovidersinArgentinatoincreasetheshareofwomenwho

initiatedprenatalcareintheirfirsttrimesterofpregnancy.Beforetheintervention,theshare

ofwomenwhoinitiatedcareearlywaslowandmedicalcareclinicshadnospecificactivitiesor

practices devoted to identifying and encouraging pregnant women to start care early.

Providersbeganprenatalcarewheneverwomenchosetofirstshowupattheclinic.

We show that the temporary incentives motivated clinics to invest time and effort to

developandtestnewdatadrivenmethodsofhowtobestidentifynewlypregnantwomenand

encouragethemtoseekcareearly.Inotherwords,clinicsinnovatedinthesensethattheyhad

to experimentwith new outreach strategies in order to learn the productivity of alternative

strategiesandwhatworkedbest.Thetemporaryincentivesnudgedproviderstochangetheir

oldmedicalcarepracticepatternhabits. Thetemporary fee increasecompensatedproviders

forthecostsoftestingnewoutreachstrategiesaswellasthecostsofovercomingpsychological

barriersofchangingpracticepatternsanduncertaintyaboutthenewservice.

Given the well-known market failures in health care, a major policy issue facing both

publicandprivatehealthcaresystemsishowtoimproveaccesstohigherqualitymedicalcare

(Chaudhury et al. 2006; Das et al. 2008). Both public (government) and private (insurance

plans) payers are turning to pay for performance incentivemechanisms to encouragebetter

qualitycare(Eldridgeetal.2009;MillerandBabiarz2014).Howbesttousetheseincentives,

however, depends on whether slow adoption of better quality care practices is due to low

perceivedvalueortohighcostsofadjustment.Ifprovidersvalueanewclinicalservicebutare

5FormoreevidenceofresistancetochangeasabarriertoCPGcomplianceseeGrol(1990);Hudak,O’DonnellandMazyrka(1995);Main,CohenandDiClemente(1995);andPathmanetal.(1996).6 For other references on technology adoption in themedical care industry see Baker (2001), Baker andPhibbs(2002),Berwick(2003),CutlerandHuckman(2003),Cutler(2007),andBechetal.(2009).

4

reluctant to adopt because of high costs of adjustment, then a temporary incentive large

enoughtocoverthesecostsshouldnudgeproviderstoadopttheserviceandcontinueitafter

theincentiveisremoved.7Ifontheotherhand,costsofadjustmentarelowbutprovidershave

alowperceivedvalueoftheservice,thenthepriceincreaseshouldalsoleadtoadoptionwhile

the increase is active, but the servicewill bedroppedonce the incentivedisappears. Hence,

studying provider behavior both while incentives are active and after the incentives are

removed provides a test of whether costs of adjustment versus low returns are inhibiting

innovation.

That early initiation of prenatal care is an important priority is common wisdom and

widelydisseminatedinthemedicalprofessional.Earlyinitiationofprenatalcarehaslongbeen

part of the Argentine CPGs for prenatal care, and is part of standard training in Argentine

medical and nursing schools as well as throughout the world (WHO 2006). Providers are

taught that prenatal care by skilled health professionals beginning in the first trimester of

pregnancy isessential forgoodmaternalandnewbornhealthoutcomesas it isargued in the

medical literature (Schwarcz et al. 2001; Carroli et al. 2001a; Campbell and Graham 2006).

Through early initiation of care, providers are able to detect and correct importantmedical

conditionssuchasmaternalinfectionsoranemiaintheperiodinwhichthefetusismostatrisk

and before these conditions can jeopardize maternal or newborn outcomes (Carroli et al.

2001b;Hawkeset al. 2013). Earlyprenatal care alsoallowsproviders to advisemotherson

proper prenatal nutrition and prevention activities in the period in which the fetus is

developingmostrapidly.

There is also strong empirical support for the relationship between early initiation of

prenatal care and birth outcomes. Rosenzweig and Schultz (1983) andGrossman and Joyce

(1988)showthatdelayingthestartofmedicalcarewhilepregnantsignificantlyreducesbirth

weight. Evans and Stech-Lien (2005) further show that missing medical visits in the early

stagesofthepregnancysignificantlyreducesbirthweight,whilemissingvisitsatlaterperiods

hasnoeffect.

7Inpractice,thisamountstopayingprovidersatime-limitedperunitincentiveforthenewservice.Payinganupfront lump sum amount is another option. However, it may be harder to ensure and verify the actualchangeinpracticepatterns.Bypayingbasedonactualperformancetheincentivesalsoincludeacommitmentdeviceforcompliance.

5

We use data from a field experiment conducted with Plan Nacer, an Argentine

government program similar to Medicaid in the U.S. and Seguro Popular in Mexico that

provides health insurance to otherwise uninsuredpregnantwomen and children.8 The field

experiment randomized temporary financial incentives to health care clinics in which

treatment clinics were paid a 200% premium for the first prenatal care visit if the visit

occurredbeforethe13thweekofpregnancy.Thefeeincreasewaspaidfor8monthsandthen

removed.Clinicswereexplicitlymadeawarethatthefeewastemporary.

Wefindthattherateofearlyinitiationofprenatalcarewas34%higherinthetreatment

groupthaninthecontrolgroup(0.42versus0.31)andthattheaverageweekspregnantatthe

timeofthefirstprenatalcarevisitfellbyabout1.5weekswhiletheincentiveswerebeingpaid.

We then show that that thehigher levelsof early initiationofprenatal care in the treatment

grouppersistedforatleast24monthsaftertheincentivesended.

We also document that clinics developed specific data driven strategies to find newly

pregnant women and encourage them to start care early. Clinics designed and tested new

beneficiaryoutreachstrategiessuchas(i)coordinatingwithlocalpharmaciestokeeptrackof

womenwhostoppedusingbirthcontrolpills,(ii)meetingwithteenagerswhileparentswere

lesslikelytobeathomesothattheywouldbemorepronetorevealpregnancies,(iii)talking

withmotherswhen they come topick freemilk for childrenand (iv)modifyinggynecologist

schedules to be able tomore easilymake appointments. These new strategies took time to

develop and test, and involved opportunity costs to clinical staff beyond marginal costs of

actual implementation. We show that all outreachactivitiesdoubled in the treatment group

relativetothecontrolgroupduringtheinterventionperiod,andthatthisincreasepersistedat

least15monthsaftertheincentivesended.

We also provide evidence that it was in the clinics’ interest to have provided these

outreach services absent the costs of adjustment. First, in a survey discussed later, clinic

medical directors reported that early initiation of care was ranked as one of the highest of

healthpriorities amongallprenatal care services. Second,PlanNacer reimbursed clinics for

beneficiary outreach activities at a rate higher than the cost of delivering those activities.

8In2013,PlanNacerwasexpandedtootherpopulationsandrenamedProgramaSumar.

6

Finally,prenatalcarevisitsbyPlanNacerbeneficiarieswereprofitabletoclinicstaffsince50%

ofthefeesobtainedfromprenatalvisitswereusedtopaywagebonuses.

Despitethefactthattheincentivessucceededininducingclinicstoinnovateanddevelop

newoutreach activities thatwere effective in increasing early initiation of prenatal care, the

incentives had no ultimate effect on birth weight. This is likely due to the fact that early

initiationofcaremayhavecomefromprimarilylow-riskmotherswhoarelesslikelytobenefit

fromearlyinitiationofcare.Indeed,onewouldthinkthatitwouldbeeasiertopersuadelow-

riskmothers to comea littleearlier than to convincehigh-riskmotherswhoare reluctant to

come for any care at all. In fact, this is consistentwith the small reduction in the average

weekspregnantatthetimeofthefirstprenatalvisit.Iftheinterventionhadinducedhigh-risk

womenwhootherwisewould havehad their 1st visitmuch later in the pregnancy, then the

incentivesmighthavehadameasurableimpactonbirthoutcomes.Onesolutionthenwouldbe

toconditionthe incentivesontheearly initiationofhigh-riskwomen,butrisk isdifficultand

expensivetoidentifyandverify,andthereforemaynotbecontractible.

Taken together these resultsareconsistentwith thepresenceof costsofadjustmentas

opposedtolowperceivedvalueinhibitingthediffusionofbetterqualityofcarepractices.First,

earlyinitiationofcarewasbothprofitableatthelowerfeesandperceivedtobeimportantfor

health outcomes; yet, absent the incentives it was being provided at a low rate. Second,

temporary incentives leadtothedevelopmentofnewoutreachactivitiesdesignedto identify

andencouragepregnantwomentoseekcareearlyresultinginalargeandsignificantincrease

in the early initiation of care that persisted long after the incentives ended. Third, the new

outreachactivitieswereinandofthemselvesprofitable.

Our results suggest that costs of adjustment maybe the mechanism behind recent

evidence that permanent performance incentives improve access to higher quality medical

care.9Thestandardexplanationisthatprovidersarereallocatingtheireffortacrossservicesin

responsetotheincreasedprofitopportunities.10However,previousstudieshavebeenunable

to distinguish between this mechanism and a cost of adjustment story. We are able to

9SeeforexampleBasingaetal. (2011);Floresetal. (2013);Bonfreretal.(2013);DeWalqueetal. (2015);Gertler and Vermeersch (2013); Gertler et al. (2014); and Huillery and Seban (2014). Miller and Babiarz(2013)provideareview.10SeeBakeretal.(1988);HolmstromandMilgrom(1991);Gibbons(1997);andLazear(2000).

7

distinguish between the two mechanisms by observing what happens when incentives are

removed.Whiletheincentivesareinplaybothmodelspredictapositiveresponse.However,

once the incentives are removed, practice patterns should revert to prior levels in standard

modelsbutcontinueatthehigherlevelsunderthecostsofadjustmentmodel.Understanding

the mechanism by which financial incentives work is also policy relevant. If temporary

financial incentivesareable to induceproviders toadoptpermanentchangesto theirclinical

practicepatterns,thentemporaryincentivescanachievealong-termboostinperformanceata

lowercostthanpermanentincentives.

Temporary incentives for technology adoption have been rarely studied.11 Notable

exceptions include Atkin et al. (2015)who examine how temporary financial incentives can

overcomeworkersresistancetoadoptanewandmoreefficienttechnologytoproducesoccer

balls.Theauthorsfindthatinitialslow-downsinproductivityfromlearninganewproduction

processinhibitedtheadoptionofthetechnologyinfirmswhereworkerswherecompensated

for performance. The results show that a short run financial incentive large enough to

compensate workers for their short-term loss generated long run gains in productivity. In

anotherrelatedpaper,Dufloetal.(2011)studytheeffectofprovidingshort-termsubsidiesto

purchasenewmoreeffectivefertilizerbysmall farmers. Theauthorsarguethateventhough

newfertilizerishighlyprofitable,theremightbeimportantbehavioralbarriersanddirectcosts

that inhibit their adoption. They show that small and temporary subsidies generated large

increases inadoption,especiallyamong impatientfarmers. Finally,Bloometal. (2012)show

thatmanagementpracticesexplainagreatpartofthedifferencesinproductivityamongIndian

firms in the textile industry. They show that providing managers with free short-term

consultingonbettermanagementskillscancreatelargegainsinproductivityinthelongrun.

The paper is organized as follows. Section II describes a simple model of technology

adoptionunderfixedcosts.SectionIIIdescribestheinterventionandtheexperimentaldesign.

Section IV describes the data. Section V explains the identification strategy and estimation

methods. Section VI shows ourmain results on different outcomes and discusses themain

11 There is alsowork on temporary incentives in the form of sales and coupons tomarket products—e.g.BlattbergandNeslin(1990);KirmaniandRao(2000);andDupas(2014).Similarly,thereisaliteratureontheeffect of temporary incentives for individuals todevelopbetterhealthhabits suchas exercise andquittingsmoking--e.g.Volppetal.(2008);Volppetal.(2009);CharnessandGneezy(2009);Johnetal.(2011);Royeretal.(2012);CawleyandPrice(2013);andAclandandLevy(2015).

8

mechanism to explain the effects we find as well as alternative explanations to our results.

Section XI and section XII discusses spill over effects and effects on birth outcomes,

respectively.SectionXIIIconcludes.

II. ConceptualFramework

We develop a stylizedmodel where clinics incur in different costs of adjustment to change

clinicalpracticepatterns.Weconsider3typesofcostsofadjustment:(i)monetarycostssuch

as the purchase of complementary expenditures on design and testing, purchases of

complementary technology and reduced productivity during a period of learning; (ii)

psychologicalcostssuchaspresentbiasandworkerresistancetochange;and(iii)uncertainty

aboutthetruemarginalproductivityandhenceprofitabilityoftheinnovationtothefirm.We

usethismodeltosimplyillustratetherearecostsofadjustmentwhenfirmsdecidetoadopta

new service into their set of practice patterns. We assume that patients are identical, that

clinicsprovidethesameservicestoallpatients,andthatdemandisexogenouslydetermined.

ObjectiveFunction:Clinicshaveapay-offfunction𝑅 = 𝜋 + 𝛼𝐻𝑁,where𝜋isprofits,His

healthof therepresentativepatient,N is thenumberofpatients,and𝛼 ≥ 0 is theprovider’s

intrinsic value of a unit of patient health.12 When 𝛼 takes on value 0, the clinic is purely

extrinsicallymotivatedandas𝛼risestheclinic iswillingtosacrificemoreincomeforpatient

health. While we allow for both extrinsic and intrinsic motivation in the model, all of the

resultsfollowevenwithpureextrinsicmotivation. Allowingforintrinsicmotivationdoesnot

changethedirectionofthepredictionsjustthemagnitude.13

Health Production Function-- Treatment technology, as defined by CPGs, involves two

services, 𝑆,and𝑆0 where 𝑆1 = 1 if the clinic provides the service and 0 if not. If the clinic

providesbothservices,thenitisoperatingattheproductionpossibilitiesfrontier.Thehealth

productionfunctionfortherepresentativepatientis

𝐻 = 𝜆,𝑆, + 𝜆0𝑆0 + 𝜀 (1)

where𝜀isameanzerorandomshock.12There isevidence to support intrinsicmotivationasat leastpartiallymotivatingmedical careproviders.SeeforexampleLeonardandMasatu(2010);Kolstad(2013);andClemenesandGotlieb(2014).13Withoutsomefixedcostsofadjustment,bothintrinsicallyandextrinsicallymotivatedproviderswouldstilloperateattheefficientfrontier.

9

ClinicalPracticePatterns--Consideraclinicwhosecurrentclinicalpracticepatternisto

provide𝑆,toallpatients.Inthiscase,𝑆,istheclinic’sexistingclinicalpracticepattern,and𝑆0

isanadditionalservicethatthecliniccouldchoosetoaddtoitspracticeroutine. Iftheclinic

wantstointegratetheprovisionof𝑆0intoitspracticepatternthenitmustovercomedifferent

barriers or costs. Clinics may have to incur an upfront one-time fixed cost F. Fixed costs

includedesigning,testing,andlearninghowtobestincorporatethedeliveryoftheserviceinto

existing practice patterns, retraining, purchase of complementary medical equipment, and

reduced productivity during a period of adjustment. Clinics have to learn how to best

implementthenewpracticeandhencetheymayberisk-aversetowardsitinthatthemarginal

costsexantearenotknownorknownimperfectly.Moreover,clinicsmaybepresentbiasedin

thattheymaydiscountfutureprofitshighly.Weincorporatethesedifferentcoststoadoption

intotheprofitfunctionofeachclinic.

Profits--Clinicsarepaid𝑝1 for𝑆1 andthemarginalcostofproviding𝑆1 toapatientis𝑐1 .

Clinic’sprofitscanthenbeexpressedas:

𝜋 = 𝛽8 𝑝, − 𝑐, + 𝐸𝑈<= 𝑝0 − 𝑐0 𝑆0 𝑁>8?, − 𝐹𝑆0,(2)

where𝛽 is the clinic’s discount rate. Thediscount ratemay inpart reflect present bias and

psychologicalresistancetochange;discountingfuturereturnstoaninnovationatahigherrate

thereby lowering the present value of an innovation. In this equation, 𝑐0 is the stochastic

componentoftheprofitfunction,wheretheexpectedutilityofproviding𝑆0is𝐸𝑈<= 𝑝0 − 𝑐0 =

𝑢(𝑝0 − 𝑐0)𝑑𝐺(𝑐0), 𝑢(∙)isanon-decreasingconcavefunction,and𝑐0isdistributedaccording

to𝐺(∙). To illustrate a measure of relative risk aversion we let𝐺(∙) be a mean preserving

spreadofanotherdistribution𝐹(∙).14

Adoption--Theclinicadopts𝑆0if

𝑅 𝑆0 = 1 − 𝑅 𝑆0 = 0 ≥ 0 . (3)

Substitutionof(1)and(2) intothepay-off functionandrearrangingtermsallowsustowrite

theconditionin(3)as:

14 It can be shown that since 𝐺(∙) is a mean preserving spread of another distribution 𝐹(∙), 𝑈 𝐹 =𝑢(𝑝0 − 𝑐0)𝑑𝐹(𝑐0) ≥ 𝑢 𝑝0 − 𝑐0 𝐺 𝑐0 = 𝑈(𝐺). Since𝐺(∙) is riskier than𝐹(∙), the latter is preferred for

everyriskaverter(Mas-Colell,Winston,andGreen,1995).

10

𝛽8 𝐸𝑈<= 𝑝0 − 𝑐0 + 𝛼𝜆0 𝑁>8?G − 𝐹 ≥ 0(4)

Clinics are more likely to adopt 𝑆0 if the profit margin from 𝑆0 is higher, they have higher

patient volumes, and they have lower discount rates. Clinics who are more intrinsically

motivated(i.e.higher𝛼)arealsomorelikelytoadoptandmaybeevenwillingtolosemoneyin

ordertoadopt𝑆0,especiallyif𝑆0 isveryproductive(i.e.higher𝜆0). Finally,theprobabilityof

adopting𝑆0decreaseswithlotteriesof𝑐0thatareofhigherriskormoreuncertain.

Temporary Incentives-- Without loss of generality we can simplify the model to 2

periodswith𝛽asthediscountrate.Inthiscase,basedon(3),anincentive𝜃thatcompensates

fortheutilitydifferentialofadoptingthenewserviceinperiod1is:

𝜃 ≥ IJ− (1 + 𝛽) 𝐸𝑈<= 𝑝0 − 𝑐0 + 𝛼𝜆0 (4)

Thetemporaryincentive,𝜃,atminimumcoverstheremainderofthecostofadjustmentthatis

notpaidby thediscountedpresentvalueof the future streamofexpectedsurplusgenerated

from theprovision of𝑆0. The incentive goesdownwith scale𝑁, the expectedprofitmargin

𝐸𝑈<= 𝑝0 − 𝑐0 , increases with uncertainty about 𝑐0, and decreases with the extent to which

clinics are extrinsicallymotivated times themarginal product of𝑆0 in the health production

function 𝛼𝜆0 ,andthediscountrate.

Cross-PriceEffects--Oneconcernvoicedintheliteratureisthatpriceincreasesforsome

services might lead to a reallocation of effort from other services that remain unchanged

leading to negative cross-price effects. The implicit underlyingmodel in these papers is an

individualphysicianallocating timebetweenactivitieswitha timebudget constraint. Inour

modelofamedicalcareorganizationthatcanhiremorestaff,cross-priceeffectsaregenerated

basedonthenatureofeconomiesofscopeineitherthehealthcareproductionfunctionorcost

function.Ifboththeproductionandcostfunctionsareadditivelyseparable,thenthereareno

cross-priceeffects.Ifthefunctionsarenotseparable,thenitispossibletohaveeithernegative

or positive cross-price effects depending thenature of substitutability in theproduction and

costfunctions.

III. ContextandExperimentalDesign

ThefieldexperimentwasconductedbyPlanNacer,apublic insuranceprogramthatbeganin

2005 to improveaccess toqualityhealth care forotherwiseuninsuredpregnantwomenand

11

children less than 6 years old (Gertler et al. 2014). Like Medicaid in the U.S. and Seguro

Popular inMexico, thenationalPlanNacer program transfers funds to local governments, in

this case Provinces, who are then responsible for enrolling beneficiaries, organizing the

provision of services, and paying medical care providers. An innovative feature of the

Argentineprogramisthatitusesfinancialincentivestoensurethatbeneficiariesreceivehigh-

quality care. Financing from the National level to Provinces is based for 60% on program

enrollmentandfor40%onperformance.

Provinces thenuse those funds topaypublichealthcare facilitiesona fee-for-service

basisforhealthcareprovidedtoprogrambeneficiaries.Thenationalgovernmentdetermines

the content of the benefits package, which is uniform across provinces, while provincial

governmentssetthepricetheywillpaytoprovidersforeachserviceinthatpackage.Revenues

fromPlanNacer are on topof clinic budgets that cover salaries aswell asmedical andnon-

medicalsuppliesandmaterials.Inpractice,PlanNacerpaymentstopupthesebudgetsby5to

7%. Health facilitiesare free tochoosehowtouserealizedrevenueswithinrelativelybroad

guidelines, and inMissiones clinics can and do use 50% of thePlan Nacer payments to pay

bonusestoclinicstaff.Inthissense,allservices,includingprenatalcarevisits,providedtoPlan

Nacerbeneficiariesareintheinterestofclinicstaffas50%ofthepaymentisusedtoincrease

staffbonuses.

Plan Nacer scaled up by first recruiting and training clinics in the operations of its

program, including feestructure,billing,andotherrules. Theprogramregularlyretrains the

clinics to keep themup todate on any changes and reinforce areas that areperceived to be

weak.Afterclinicsareenrolled,cliniccommunityoutreachstaffidentifieseligiblewomenand

children in order to enroll them into the program. Enrollment activities usually consist on

door-to-doorvisitsacrossadeterminedgeographicareaassignedtoeachclinicanddefinedby

theProvince.

Clinics can only provide services to the population within their area and enrolled

beneficiariescanonlyobtaincare fromtheirassignedclinic. Outreachstaff regularlycontact

beneficiariestoencouragethemtotakeadvantageofprogrambenefits.PlanNacerreimburses

clinicsforalloutreachactivitiestothebeneficiarypopulationataratehigherthantheclinic’s

costofoutreach.

12

ThefieldexperimentwasconductedwithprimaryhealthcareclinicsintheProvinceof

Misiones,oneofthepoorestinthecountryandwithhighratesofmaternalandchildmortality.

InMisiones, each clinic is allowed to use up to 50%of revenue fromPlanNacer fees to pay

bonusestofacilitypersonnelatthediscretionofthefacilitydirector.TherolloutofPlanNacer

inMisioneswascompleted in2008 longbefore thepilot study. As such,bothprovidersand

beneficiarieswereknowledgeableoftheoperationofPlanNacerbeforetheexperimentbegan.

The experimental interventionwas designed to encourage early initiation of prenatal

care for Plan Nacer beneficiaries, thereby aligning the incentives in Plan Nacer with official

Argentine clinical practice guidelines, medical school training, and international scientific

evidence.Beforetheexperiment,onlyone-thirdofPlanNacerbeneficiarieswereinitiatingcare

in the first trimester (National Ministry of Health, 2009 and 2010). The experiment

randomized temporary financial incentives toprimaryhealth care clinics inwhich treatment

clinicswerepaida200%premiumforearly initiationofprenatalcare, i.e.beforeweek13of

pregnancy.

Table 1 presents the payment schedule for the periods before, during and after the

intervention. Prior to the intervention period, the province paid facilities $40 ARS for each

prenatalvisitregardlessofwhenitoccurredorwhetheritwasthefirstorasubsequentvisit.15

Atthisinitialpriceprenatalcarevisitswereprofitableas50%ofthisfeewasusedtoincrease

staffbonuses. Duringtheinterventionperiodthefeewasincreasedto$120ARSfor1stvisits

that occurred before week 13 but remained at $40 ARS for subsequent visits. Every other

componentofthePlanNacerprogramremainedthesame.Afterthat,theinterventionperiod

feesrevertedtotheoriginalpaymentof$40ARSforallvisits.Themodificationamountedtoa

3-fold increase in the fee for 1st visits before week 13. The modified fee structure was

implementedfor8months-fromMay2010toDecember2010.

Facilitiesselectedtoreceivethemodifiedfeestructurewereinvitedtoparticipateand

notifiedofthetime-limitedimplementationonApril14,2010.Facilitydirectorswererequired

tosignaformalmodificationoftheirexistingcontractwithPlanNacerinordertoreceivethe

modifiedfeestructure.

15Theexchangeratefor$1ARSwasaround$0.25USDbetween2009through2011.

13

Thestudydesignincluded37clinicsoutof262primarycarefacilitiesoftheprovince,of

which18wererandomlyassigned to the treatmentgroupandwereoffered themodified fee

schedule.Theother19formedthecontrolgroup.Compliancewithtreatmentassignmentwas

not perfect: out of 18 facilities assigned to the treatment group, 14were actually treated as

three refused to sign the agreement and a fourth closedbefore the intervention started. In

addition,oneof the facilitiesoriginally assigned to the control groupwasmistakenlyoffered

thetreatmentandagreedtothemodifiedfeestructure.

IV. Data

TheProvinceofMisionesmaintainsawell-developedandlong-establishedautomatedmedical

recordinformationsystemmanagedbytheprovincialauthorities.Personnelatpublicprimary

healthclinicsandhospitalsdigitizearecordofeachserviceprovidedtoeachpatient.Thedata

areofunusuallyhighquality in thatkeyoutcomes suchasdatesof visits, servicesdelivered,

andbirthweightarerecordedatthetimeeachserviceisprovided;thereforewedonotneedto

relyonmaternalrecallofthesevariablesusuallycollectedbysurveyslongafterthevisit.The

datausedintheanalysisareextractedfromindividualclinicalrecordsandcontaininformation

on the universe of patients for the 36 clinics in the study. The records also include the

individual’snationalidentitynumber,whichisusedtolinktheindividualclinicmedicalrecords

from primary health facilitieswith the registry of health insurance coverage, the registry of

PlanNacerbeneficiaries,andhospitalmedicalrecords.Inall,97%oftheprimaryclinicmedical

recordsweremergedwith the data on insurance status and program beneficiary status. In

addition, 75% of these were successfully merged withmedical records data from hospitals.

Therefore,eachobservationinoursamplecorrespondstoauniquepregnancybywomenwho

initiatedtheirprenatalcareinoneoftheprimarycareclinicsofthesample.

A. AnalysisSample

The timeline of the study and the availability of data is divided into 4 different sub-

periods: (i) a16-monthspre-interventionperiod from January2009 toApril 2010, (ii) an8-

month intervention period from May 2010 to December 2010, (iii) a 15-month “post-

interventionperiodI”fromJanuary2011toMarch2012and(iv)a9-month“post-intervention

periodII”fromApril2012toDecember2012.

14

Prenatalcaredatawasconsistentlycollectedforthefirst3periodsfromJanuary2009

throughMarch 2012. Starting in April 2012, however,Misiones adopted a new information

systemandasaresultdatafrompost-interventionperiodIIcannoteasilybecomparedtodata

fromtheearlierperiods.Inparticular,thenewsystemchangedthecodesusedtoclassifythe

reason for visits in order to facilitate billing. If in the first visit the attending physician

requested an ultrasound to confirm a pregnancy, this first visitwas labeled as a “care visit”

while the subsequent (second) visit, was labeled as the first prenatal visit, if indeed the

ultrasoundconfirmedthepregnancy.Onaverage,thiswouldleadtoareductionintheshareof

womenwhohadavisit labeledas“firstprenatalvisit”beforeweek13andanincreaseinthe

weekspregnantatthetimeofthisvisit. Ifthenewcodingsystemaffectedthetreatmentand

control groups in the same way, the differences between the treatment and control groups

wouldstillcapturetheimpactoftheincentives,albeitpossiblywithsomemeasurementerror.

Therefore,weanalyze thedata frompost-interventionperiod II separately,and interpret the

resultswithcaution.

TheanalysissampleincludespregnantwomenwhowerebeneficiariesofPlanNacerat

thetimeofthefirstprenatalvisit.16Whileinformationonprenatalcareutilizationisavailable

for the full sampleperiod, informationrelated tobirthoutcomes isonlyavailable forwomen

whogavebirthinapublichospitalthrough2011,i.e.womenthatbecamepregnantbeforeMay

2011.

B. MeasurementofWeeksPregnantat1stPrenatalVisit

Each observation in our sample corresponds to a different pregnancy that initiated

prenatalcareinoneoftheclinicsincludedintheexperiment.Foreachpregnancyweobserve

thedateofthefirstprenatalvisitandthedateofthelastmenstruationperiodasrecordedby

thephysician.Weconstructthenumberofweeksofpregnancyatthetimeofthefirstprenatal

visitasthedifferencebetweenthedateofthefirstvisitandthelastmenstrualdate(LMD).The

LMD is routinely collected at the timeof the visit to calculate the estimateddate of delivery

(EDD)andbothareroutinelyrecordedinthepatient’smedicalrecordattheclinic.17

16Weexcludednon-beneficiariesbecausemostofthemhaveprivatehealthinsuranceandassucharelikelytoreceivesomeofcareanddeliveratprivatefacilities.Sincewedonothavedatafromprivatefacilities,theoutcomesofmostoftheseobservationsarecensored.17For10%ofthesampleLDMwasnotrecorded.Forthosecases,weusetheEDDtorecovertheLMD.

15

Onepotentialproblemisthatmedicalpersonnelintreatmentfacilitiesmightmisreport

thedateofthefirstvisitasoccurringbeforeweek13sothattheycouldbillittotheprogramat

a higher amount. We think this is unlikely for the following reasons. First, the week of

pregnancyatthefirstvisitisconstructedfromthedateofthefirstprenatalvisitandtheLMD,

bothofwhichalongwiththeEDDarerecordedinrealtimeinthemedicalrecord.Inorderto

falselyreportthatafirstvisitoccurredinthefirst12weeks,theproviderwouldhavetoalter

the date of the first visit relative to either the LMD or the EDD in themedical record. This

wouldrequiresomeeffortifdoneinrealtimeandwouldbenoticeablebyauditorsifalteredex

post. Second, Plan Nacer uses external auditors to verify the accuracy of clinic billing. The

auditors compare the detailed clinical records to the billing requests to find inconsistencies

that could turn into substantial financialpenalties for theprovinces. Third,while theremay

have been an incentive to misreport during the intervention period, there was no financial

return tomisreporting in thepost-interventionperiodonce the incentiveswere removed. It

alsowasunlikelythatitwasworththeclinics’timetosetupelaborateproceduresforfalsifying

records when they knew the incentives were only in place for 8 months. Finally, clinical

recordsare legaldocuments inArgentinaandpractitionerscould losetheirmedical license if

caughtsystematicallymisreportingforfinancialgains.

Tocorroborateourbeliefthatfalsereportingofrecordsisunlikely,weempiricallytest

whether there is any evidence of systematic misreporting using data from an alternative

source. Specifically, we use gestational age at birth measured by physical examination

obtainedfromhospitalrecordstoconstructasecondestimateoftheLMDandweekspregnant

atthetimeofthefirstprenatalvisit.Thehospitalpersonnelthatattendthebirthdonothave

anyincentivetomisreporthospitalrecords.Wethencomparetheestimatedweekoffirstvisit

basedongestationalageatbirthtotheweekoffirstvisitreportedbythehealthfacilities.The

results donot showany evidence of systematicmisreportingdue to incentives. AppendixA

providesadetaileddiscussionoftheanalysisandresults.

Wealsoexplorewhetherthere isanymanipulationof thedataat thethresholdof the

13th week of pregnancy. Appendix Figure A2 shows that there is no discontinuity at this

threshold using the test proposed by McCrary (2008) for manipulation at the threshold in

studiesthatuseRegressionDiscontinuityastheirresearchdesign.

16

C. DescriptiveStatisticsandBaselineBalance

Table 2 reports the descriptive statistics for the key outcomes of interest and

demographiccharacteristicsatbaseline,i.e.inthe16-monthpre-interventionperiod(Jan2009

–April2010). Outcomesarebalancedatbaselineinthattherearenostatisticallysignificant

differences in themeansofvariablesbetweenthetreatmentandcontrolgroups. Onaverage

women had their first prenatal visit about 17.5weeks into their pregnancywith about one-

thirdofwomenhavingthatvisitbeforeweek13.Womencompletedabout4.7prenatalvisits

over the course of their pregnancy andmore than 80% of them received a tetanus vaccine.

Newbornsweighedapproximately3,300gramsonaverage,whileabout6%ofthemwereborn

with lowbirthweight (i.e. less than2,500grams), and slightlymore than9%ofbirthswere

bornprematurely.

V. IdentificationandEstimation

Weestimateboththeintent-to-treat(ITT)andlocalaveragetreatment(LATE)effectsof

the incentives on outcomes (Imbens andAngrist 1994). The ITT is the effect of assigning a

clinic to treatmentonoutcomes, regardlessof compliance. TheLATE is the effect of a clinic

actually receiving the incentives. In both cases, the treatment effect is identified off the

variationinducedbytherandomizedassignmentstatus.Inthediscussionofresultsinthenext

section,wereporttheLATEestimates.18

TheITTestimatecomparesthemeanoutcomeofthegroupassignedtotreatmenttothe

mean outcome of the group assigned to control and is estimated by regressing the outcome

against an indicator of whether the clinic was assigned to treatment using the following

specification

𝑦1L8 = 𝛼8 + 𝛽8𝑇L8 + 𝜖1L8, (5)

where𝑦1L8 is the outcome of individual i receiving care in clinic j in period t,𝑇L is a dummy

variabletakingonthevalue1iftheclinicwasassignedtothetreatmentgroupand0otherwise,

and𝜖1L8 isazeromeanrandomerror. Notice thatparametersareallowedtovarybyperiod.

Weworkwith fourdifferentperiodsof analysis: an18-monthpre-interventionperiod, an8-

18 The ITT results are almost identical to the LATE estimates, which is expected given the relatively highcomplianceratestotheoriginalassignment.TheITTresultsarepresentedinAppendixC.

17

month intervention period, a 15-month post intervention period I, and an 8-month post

interventionperiod II. Weestimate separatemodels for eachof theseperiods. In theLATE

modelwe replace𝑇L the “assigned to treatment” variablewith an indicator of being actually

treated and use the clinic’s randomized assignment status as an instrumental variable for

actualtreatment.

Our sample is clustered within 36 health clinics since the random assignment of

treatmentoccurredatthecliniclevel.Assuch,theremaybeintra-clustercorrelationthatmust

beconsideredforstatisticalinference.Standardmethodsofcorrectingstandarderrorsrelyon

largesampletheorybothinthenumberofobservationsandinthenumberofclusters. Given

thesmallnumberofclustersinoursample,weinsteaduserandomizationinferencemethods

that are robust to randomized assignment of treatment among a small number of clusters.

Specifically, we use the Wild bootstrap to generate p-values for hypothesis testing in ITT

models (Cameron et al. 2008) and an analogousmethod for hypothesis testing in the LATE

models (Gelbachetal.2009). OurWildbootstrapprocedureassignssymmetricweightsand

equal probability after re-sampling residuals, and uses 999 replications (Davidson and

Flachaire2008).

VI. TimingofFirstPrenatalVisit

Inthissectionwereporttheresultsofanalysesoftheeffectsofthetemporaryincentivesonthe

timingofthefirstprenatalvisitandmechanismsbywhichclinicsachievedthoseresults.

A. Densities

Figure2comparesthedensitiesofweekspregnantatthetimeofthefirstprenatalvisit

for the clinics assigned to the treatment and control groups and reports p-values for

Kolmogorov-Smirnov tests of equality of the distributions. Panel A shows that there is no

differencebetween thedensitiesof the treatmentandcontrolgroups in thepre-intervention

period. Panel B shows that the treatment group density is to the left of the control group

densityduringtheinterventionperiod. Finally,PanelCandDshowthatthetreatmentgroup

densityisplacedtotheleftofthecontrolgroupdensityduringpost-interventionperiodsIand

II. Kolmogorov-Smirnovtests forequalityof thedistributionscannotberejectedforthepre-

interventionanalysis,butarerejectedfortheinterventionandbothpost-interventionperiods

18

withp-valuesof0.031,0.004,and0.009respectively. Theseresultsimplythatthetemporary

incentivesledtoearlierinitiationofcareinthetreatmentgroupcomparedtothecontrolgroup

in the intervention period and that these higher levels of care persisted for at least for 24

monthsandmoreafterthehigherfeeswereremoved.

B. ShortRunEffects

Table3reportstheestimatesoftheeffectsofthetemporaryfeesontheearlyinitiation

ofcare.PanelAreportstheresultsforweekspregnantatthetimeofthefirstprenatalvisitand

Panel B reports the results for whether the first visit occurred before week 13. The first

columnreportstheresultsfortheinterventionperiodandthesecondandthirdcolumnsreport

the results for the post-intervention periods. During the intervention period, on average,

women in the treatmentgrouphad their1stvisit about1.5weeksearlier in theirpregnancy

thanwomeninthecontrolgroup.Theshareofwomeninthetreatmentgroupwhohadtheir

1stvisitbeforeweek13is11percentagepointshigherthanthecontrolgroup;approximately

35% higher than the control group. Both estimates are significantly different from zero at

conventionalp-values.

C. LongRunEffects

Our model in Section II provided clear predictions about provider behavior once

temporaryincentivesdisappear:i.e.ifthefeeincreaseisenoughtoovercomethefixedcostsof

adaptinganewpractice,clinicsshouldmaintainhigherlevelsofprenatalcareafterincentives

areremoved.Column2ofTable3reportstheestimatedimpactofthetemporaryfeeincrease

on early initiationof care in the15-monthperiod after the feeswere removed. On average,

pregnantwomeninthetreatmentgroupstartedtheircare1.6weeksearlierthanthoseinthe

controlgroup.Thedifferencebetweenthetreatmentandcontrolgroupsintheshareofwomen

whohadtheir1stvisitbeforeweek13was8percentagepoints.Bothestimatesarestatistically

different fromzeroatconventional levels. Further,wecannotreject thenullhypothesis that

theimpactisdifferentintheinterventionandpost-interventionperiods.

Whilethereisnosignificantdifferencebetweentheeffectduringtheinterventionand

thepost-interventionperiods,oneconcernmaybethattheeffectoftreatmentslowlytrended

towardszeroaftertheincentivesended.Toexplorethishypothesis,weplotthemeannumber

ofweekspregnantat thetimeof firstprenatalvisit for treatmentandcontrolgroups,before,

19

duringandafter the intervention (Figure3).19 Wesplit thepre-interventionperiod into two

sub-periodsof6-monthseachandthepost-interventionperiodinto3sub-periods:thefirsttwo

are6monthsandthethirdis3months.Thetreatmenteffectisthedifferencebetweenthetwo

lines.Whilethetreatmentandcontrolgroupshavesimilartrendsbeforetheintervention,the

treatment group appears to receive earlier care during the intervention, and the change

persistsaftertheendoftheintervention.Noticethatthereislittleifanyfalloffoverthepost-

interventionperiod. Rather, the treatmenteffects remain fairlyconstantover the15 -month

post-interventionperiodI.Figure4depictsthesamerelationshipfortheshareofwomenwho

receive carebeforeweek13of pregnancy.20 Again, the effects of the intervention appear to

continueatasteadyrateafteritisdiscontinued.

D. Longerruneffects

Theperiodof analysis in ourmain results is restricted to Januaryof 2009 toMarchof

2012. Recall that starting in April 2012, the visit coding system changed. Hence starting in

April2012whatisreportedasfirstvisitinthedataisactuallyamixoffirstandsecondvisits.

Asaresulttheaverageofweekspregnantatthefirstvisitincreasesandtheshareofpregnant

womenwhose first visitwasbeforeweek13 falls relative topreviousperiods. Column3 in

Table3showstheresultsforthislastperiod.Themeanaverageofweekspregnantatthetime

of the firstvisit for thecontrolgroup issubstantiallyhigher for thisperiodthan forprevious

periods and themean share that had their first visit before week 13 is substantially lower,

suggestingthatthereismeasurementerrorinourmainoutcomeinthisperiod.However,this

difference in coding should have a similar effect in treatment and control clinics given the

randomized assignment of the treatment. Therefore the difference between treatment and

control clinics should cancel out the measurement error and provide us with unbiased

estimatesoftheimpact.

Theresults inTable3showastatisticallysignificantreduction in thenumberofweeks

pregnant at the time of the first visit and a statistically significant increase in the share of

19 As discussed above, the information from post-intervention period II (April-December 2012) uses adifferentmetricandisthereforenotincludedinthisfigure.20Ibidem.

20

pregnant women who had their first visit before week 13. These results suggest that the

improvedproductivitypersistedatleast24monthsafterthefeeswereremoved.

E. Robustness

We implement a number of different robustness checks to our results. First, themain

samplemayincludepregnanciesthatstartinoneperiodandendinanother,whichcouldcloud

theeffectoftheincentivesontimingofthefirstvisit.Forexample,awomanwhois6months

pregnant and had not had a prenatal visit when the intervention starts and subsequently

receives her first prenatal checkup during the intervention, would be counted as a third

trimesterfirstvisitduringtheinterventionperiod,eventhoughtheinterventioncannotaffect

whether she receives prenatal care beforeweek13. Hence,we re-estimate themodels on a

restrictedsamplewherewomenarenomorethanonemonthpregnantinthefirstmonthofthe

period and no less than 3-months pregnant in the last month of the period. The results,

reportedinPanelsBofAppendixTablesB1andB2,areverycloseinmagnitudeandstatistical

significancetothemainresultsinTable3.

Second, in studies involving a small sample of clusters there is a concern that a few

outliersmaydrivetheaverageeffectfoundintheprevioussections.Weexplorethispossibility

intwoways.First,were-estimatethemodelsbydroppingalltheobservationsfromoneclinic

one at a time. This produces 36 different estimated treatment effects, which we picture in

appendix Figures B1 and B2 for the outcomes of weeks pregnant at the time of the first

prenatal visit (B1) and for the probability that the first visit occurred beforeweek 13 (B2),

respectively. Theresultsaresortedalongthex-axisfromthelowesttothehighestestimated

effect, while the dashed blue line is the intent-to-treat effect calculated by pooling the

intervention and the first post intervention period. The solid black line represents a zero

treatmenteffect. Thevertical linesare95%confidence intervals constructedusingstandard

errorsobtainedfromtheWildbootstrapprocedure.Noticethatthereisalmostnodifferencein

anyoftheestimatesimplyingnooneclinicdrivestheestimatedeffectsinTable3.

Finally,weestimateclinic-specifictreatmenteffectswherebywecompareeachtreated

clinic individually to the control clinics as a group. Appendix Figures B3 and B4 plot these

individualclinic treatmenteffects for theoutcomesofweekspregnantat thetimeof the first

prenatalvisit(B1)andfortheprobabilityofthatthefirstvisitoccurredbeforeweek13(B2),

respectively. The results are again sorted along the x-axis from the lowest to the highest

21

estimatedeffect,while thedashedblue line is the intent-to-treat effect calculatedbypooling

theinterventionandthefirstpostinterventionperiodandthesolidblacklinerepresentsazero

treatmenteffect.Thefiguresshowthatthehypothesisofnotreatmenteffectisrejectedfor11

outof17clinicsinFigureB1and12outof17clinicsinFigureB2.Inaddition,thetreatment

effectshavetheexpectedsignin15out17clinicsinFigureB1and14outofclinicsinFigure

B2.Thisprovidesevidencethatourresultsarenotdrivenbyafewlarge-effectclinics.

VII. Mechanisms

Inordertobetterunderstandhowclinicswereabletoachievesuchlargeincreasesintheshare

of women who initiated prenatal care before week 13, we conducted a series of in-depth

interviewswithmedicalprofessionalsanddirectorsof the treatedclinics. In this sectionwe

first reportwhatwe learned from these interviews. In summary, all of the clinics reported

developing a new set of community outreach activities designed to identify Plan Nacer

beneficiary women early in their pregnancies and reach out to them to encourage early

initiationofprenatal care. Thedesignand installationof theseoutreachactivities into clinic

routinesinvolvednontrivialfixedcostsandthedeliveryofthoseservicescreatednewvariable

costs. Moreover, the interviews reflect thatduring an experimentationperiod, clinicswhere

abletodesignprofitablestrategiestoimplementinthefuturesothattheysignificantlyreduced

theuncertaintyembeddedinincorporatingandsustainingthenewpractice.Knowledgeabout

which practicesworked best also allows reducing present bias towards of old routines. We

thenuse thewhole sample to analyze the impactof the temporary incentiveson community

outreachactivities.

Developing New Outreach Strategies— All of the clinics reported organizing a team

meetingwiththestaffatthebeginningoftheinterventioninordertodiscussandbrainstorm

strategies to respond to the new incentive scheme. They developed innovative data driven

strategies to identifywomenwhowere likely tobepregnant. The clinics then typically sent

stafftoinquireaboutlastmenstruationdateandofferaninstant-readpregnancytesttothose

womenwhosemenstruationwas overdue. If pregnant, they then encouraged the expectant

motherstostartprenatalcarequickly.

Much of this involved experimenting with different strategies until they found what

workedbest. Forinstance,healthworkersstartedtomonitorwomenwhousedbirthcontrol

22

pills21andprioritizehomevisitstowomenwhowerelateinpickinguptheirpillrefills.Second,

clinics targeted mothers who already have children, as they are less likely to initiate their

prenatalvisitsearly inanewpregnancy,bymeeting themat freemilkdistributioncenters.22

Third, health workers noted that adolescents are less willing to reveal a pregnancy in the

presenceoftheirparents.Clinicsthereforechangedthetimingofhomevisitssoastoincrease

thechanceoffindingadolescentsbythemselves.Clinics’workscheduleswerealsomodifiedso

as to ensure predictable availability of a gynecologist on certain days of theweek so health

workers could better schedule patient appointments. Other clinics started keeping track of

visits to “at risk” patient and map clinic catchment areas with corresponding (potential)

pregnanciessoastomoreefficientlyorganizehomevisitroutes.

Clinics invested in testing a number of different strategies designed to reach out

pregnantwomen and encourage them to initiate care early in order to learnwhich of these

strategiesworkedbest. Absent the incentives, suchexperimentationwasperceivedasbeing

toocostlyandislikelyanimportantcostsofadjustment.Thesetypesofcostsincludemonetary

investments toput inextrahoursofwork thinkingandplanningnewstrategies,overcoming

uncertaintyabouttheeffectivenessofnewpractices,andadjustingtoworkingunderanewset

ofoutreachactivities.

Thesetypesofcostsofadjustmentareanchoredbyoldpractices(habits)similartothe

statusquobias(ThalerandSunstein,2009,pp.35).Overcominguncertainty,adjustingtonew

practices, and theexpenseof investing time toplan them, areall simultaneouslydetermined

and we are not able to separately identify one or the other. In fact, we interpret that the

temporary incentivesmotivatingclinics toovercomesuchcostsbyproviding theclinicswith

theopportunitytotestandexperimenttherebyovercomingallofthesepotentialbarriers.

Implementation—ClinicsusedCommunityHealthWorkers(CHWs)toimplementthese

newoutreachactivities.23 Normally,CHWscarryoutcommunityoutreachactivitiesincluding

21Birthcontrolpillsaredispensedfreeofchargebyeachhealthfacility.Womencannotcollectmorethanamonthssupplyatanyonetimeandmustreturneachmonthforrefill. Thepharmacyunitkeepsrecordsofbirthcontrolpillcollections.22PlanNacerbeneficiarieswithyoungchildrenareeligibleforfreemilkweeklyandmotherscollectthemilkatdistributioncenters.23TheMinistryofHealthcreatedCHWasajobcategoryin2005aspartofa3-yearassociatesdegreeprograminPrimaryHealthSectorManagement fromtheMinistryofHealth.CHWshaveclassesat least4hoursper

23

promotionofpreventivehealth,follow-upofpatientsintreatmentincludingpregnantwomen,

follow-up of immunization status of children, health data management, early detection of

malnutritioninchildren,amongothersaswellasperiodicallyupdatingtherosterofresidents

intheclinic’scatchmentareas.Sinceitsrolloutin2005,PlanNacerhasreimbursedclinicsfor

outreach activities at a profitable rate.24 CHWswork under temporary contracts of variable

lengthwiththefacilitiesandarenotpartoftheformalcivilservicesubjecttomorerigidlabor

laws.Assuch,clinicscaneasilyandquicklyexpandandcontracttheamountofCHWlaborthey

employ. Duringtheintervention,clinicsreportedexpandingCHWactivitiesbyincreasingthe

hoursofexistingCHWsasopposedtohiringnewCHWsandpaidincentivebonusestoCHWs

forgettingpregnantwomenintoprenatalcare.25

ImpactofTemporaryIncentivesonOutreachActivities—Basedonqualitativeevidence,

inthelastsectionwelearnedthatclinicsreportedtohavedevelopedandimplementedvaries

strategiesthatincreasedtheamountofoutreachactivitiesoncetheincentiveswereactive.To

investigatethisfurther,weaccessedadministrativedatathatrecordsallcommunityoutreach

activitiesperformedbyclinicsinMisiones,andtestwhetherthereweresignificantincreasesin

the amount of outreach activities to pregnantwomen in treatment clinics relative to control

clinics.26 Figure 5 displays the average andmedian number of CHWoutreach activities that

resulted inmaternalcarevisits forthepre-intervention, intervention,andpost-interventionI

periods. In this case themediansarebettermeasuresof central tendencyas thedensitiesof

bothactivitiesareasymmetricheavily skewed to the right.The results show that there isno

differenceinoutreachactivitiesbetweentreatmentandcontrolclinicsinthepre-intervention

period. Intheinterventionperiodthetreatmentgrouphadsubstantiallymoreactivitiesthan

thecontrolgroup,andthisdifferencecontinuedthroughthepost-interventionperiod.

weekandarerequiredtoworkatleast21hoursaweekasinternsinalocalclinicorhospital.Theinternsarepaidanhourlystipendthatislessthantheminimumwage.24Fromadministrativerecordswetheaveragecostofoutreachactivitiesto$1USDasCHWsarepaid$2USDper hours and complete on average2 outreach activities per hour.PlanNacer pays $2.5USD for outreachactivitiestopregnantwomen,sothateachoutreachactivitygeneratesaprofitof$1.5USD.25Until2013healthfacilitiesparticipatinginPlanNacerinMisioneswasabletouseupto50%oftheirofPlanNacerfundstopaybonusestohealthprofessionals.Thebonusescouldbeassignedtoanypersonworkingatthehealthfacility,includingCHWs26 Plan Nacer finances clinic outreach activities on a fee-for-service basis and employs an externalindependentauditortoauditclinicactivityreports.Treatmentandcomparisonclinicswerepaidthesamefeefortheseactivitiesbefore,duringandaftertheexperiment.

24

Weuse thesedata toestimate thedifferences in the logarithmofnumberofactivities

between the treatment and control groups using the samemethods in Table 3. The results

shownodifferences inactivities in thepre-interventionperiod, andpositiveand statistically

significanthigherlevelsofactivitiesinthetreatmentclinicsinboththeinterventionandpost-

interventionperiods(Table4).Outreachactivitiesdoubledinthetreatmentclinicsrelativeto

the controls in both the intervention and post intervention periods suggesting that the

temporary incentives significantly raisedCHWoutreach activities to a level that persisted at

least15monthsafterthetemporaryincentiveswereremoved.

VIII. ProfitabilityofPrenatalCareandOutreachActivities

We have shown so far that the temporary incentives led to a long-term increase in early

initiationofprenatalcarethroughincreasedoutreachactivitiesbyCHWs.Wehavealsoshown

thatclinics investedindevelopingandtestingnewdatadrivenoutreachactivities inorderto

locateandencouragepregnantwomentoseekcareearly.Tobecompletelyconsistentwithour

modelwealsoneedtoshowthatprenatalcareandtheoutreachactivitiesusedtoencourage

early initiation of care were known to be profitable before the temporary incentives were

rolledout.

PrenatalcarehasalwaysbeenprofitableforclinicssincePlanNacerstarted.PlanNacer

paysclinicsanadditionalAR$40foreachprenatalcarevisitandhalfofthatwasusedtopay

bonusestothemedicalcareproviders.RevenuesfromPlanNacerareontopofclinicbudgets

thatcoversalariesaswellasmedicalandnon-medicalsuppliesandmaterials.Clinicshadbeen

enrolled inPlanNacer forover5yearsbefore the intervention.PlanNacer trainedclinicson

billing procedures and clinics had been billing Plan Nacer for prenatal care visits over the

wholeperiod.

Clinics reliedonCommunityHealthWorkers (CHWs) foroutreachactivities including

locatingandencouragingpregnantwomentoinitiateprenatalcareinitiation. TheMinistryof

HealthcreatedCHWasajobcategoryin2005aspartofa3-yearassociatesdegreeprogramin

PrimaryHealthSectorManagementfromtheMinistryofHealth.CHWshaveclassesatleast4

hoursperweekandarerequiredtoworkatleast21hoursaweekasinternsinalocalclinicor

hospital. The interns are paid an hourly stipend that is less than theminimumwage. Their

normal activities include promotion of preventive health, follow-up of patients in treatment

25

including pregnant women, follow-up of immunization status of children, health data

management,earlydetectionofmalnutritioninchildren,amongothersaswellasperiodically

updatingtherosterofresidentsintheclinic’scatchmentareas.

Since itsrollout in2005,PlanNacerhasreimbursedclinics foroutreachactivitiesata

profitable rate. From administrative recordswe calculated that the average cost of outreach

activities is $1USDasCHWsarepaid$2USDperhourand completeonaverage2outreach

activitiesperhour.PlanNacerpays$2.5USDforeachoutreachactivitytopregnantwomen,so

thateachactivitythengeneratesaprofitof$1.5USD.

Theprofitabilityofprenatalcareandoutreachactivitieswerewellknowntotheclinics

longbeforetheimplementationofthetemporaryincentives.ClinicshadbeenusingCHWsfor

other outreach activities such as promotion of preventive health, follow-up of patients in

treatmentincludingpregnantwomen,follow-upofimmunizationstatusofchildren,healthdata

management,earlydetectionofmalnutritioninchildren,amongothersaswellasperiodically

updating the roster of residents in the clinic’s catchment areas. Clinics hadbeenbillingPlan

Nacer for outreach activities from the beginning of the program, so that clinics should have

beenawareoftheprofitabilitylongbeforetheintervention.Indeed,figure5reportsthatclinics

wereheavilyengagedinoutreachintheyearbeforethetemporaryincentiveswererolledout.

Finally,while clinicsmayhaveknown thepricespaid for these services ex ante, they

may have not known the costs prior to experimentingwith alternatives outreach strategies.

Specifically,themaynothaveknownhowmanywomenthatCHWscouldvisitandimplement

the strategy per day. Indeed, clinics could have been inhibited from trying these strategies

becauseoftheuncertaintyincosts. Atthehigherreimbursementrateforearlyprenatalcare

during the incentives, clinicsmighthave found it cost-effective to tryoutnewstrategies that

were otherwise too risky. After learning the costs of the best strategies and confirming

profitability, clinics continued to perform it in the longer run. As such, the fact that clinics

expandedoutreachactivitiesisalsoconsistentwithaprocessoflearningabouttheprofitability

ofsuchstrategies.

IX. KnowledgeandSalienceoftheImportanceofEarlyInitiationofPrenatalcare

A key aspect of the argument that fixed costs of adoption inhibited clinics from adopting

services to increase the early initiation of prenatal care is that clinics valued early initiation

26

enoughtohavehadadoptedtheseserviceswithoutthefixedcosts.Itispossible,however,that

fixed costs of adoption were not inhibiting adoption, but rather clinics did not know the

importancethatthehealthprofessionplacesontheearlyinitiationofcare.

The temporary incentives couldhave informedproviders about thehealthbenefitsof

early initiation of prenatal care. Hence, the incentives might have worked through a

“knowledge channel” to change practice patterns. This seems unlikely as early initiation of

prenatalcarehaslongbeenpartoftheArgentineCPGsforprenatalcare,andispartofstandard

training inArgentinemedicalandnursingschools.Providersaretaughtthatprenatalcareby

skilledhealthprofessionalsbeginning inthe first trimesterofpregnancy isessential forgood

maternalandnewbornhealthoutcomesasitisarguedinthemedicalliterature(Schwarczetal.

2001;Carrolietal.2001a;CampbellandGraham2006;WHO2006).

Relatedtotheknowledgechannelissaliency.Providersmayhaveintellectuallyknown

theimportanceofearlycare,butmaynothavesufficientlyvalueditenoughtoinvestinthese

services without the increased fees. The temporary incentives might have just made early

initiationofcaremoresalient27andtherebyincreasedtheimportanceofearlyinitiationofcare

inthestaff’smindssothatitbecameahigherpriorityforaction.Kahneman(2012,pp8)states

that“…frequentlymentionedtopicspopulatethemind…”morethanothersand“…peopletend

to assess the relative importance of issues by the ease with which they are retrieved from

memory”.Assuch,salience“…isenhancedbymerementionofanevent”(Kahneman2012,pp

331).Ifincompleteornon-adoptionofataskisamatterofsalienceasopposedtofixedcostsof

adoption then the observed treatment effects may be explained by the fact that temporary

incentiveshelptoovercomethistypeofpsychologicalresistancetochange.

Whilewedonothaveinformationontheknowledgeandsalienceofearlyinitiationof

carebeforetheexperiment,weareabletoexplorewhetherthetemporaryfeeincreasemade

early initiation of care more important in the minds of the clinic staff after the end of the

experiment.Totestforthesehypothesesweadministeredasurveytothechiefmedicalofficer

27 Taylor and Thompson (1982) define salience as, “…the phenomenon that when one's attention isdifferentiallydirectedtooneportionoftheenvironmentratherthantoothers,theinformationcontainedinthat portion will receive disproportionate weighting in subsequent judgments”. See Bordalo et al. (2012,2013)foramorerecentdiscussionofsalienceandchoicetheory.SeeDeMeletal.(2013),andKarlanetal.(2015)forempiricalanalysisofsalienceeffectsthroughinformationalreminders.

27

(CMO)ofeachclinicabouttheabsoluteandrelativeimportanceofsevendifferentprenatalcare

proceduresincludinginitiatingprenatalcarepriortoweek13ofpregnancy(seeAppendixD).28

Figure6comparestheabsolutescoreandrelativerankingoftheproceduresintermsof

importanceforprenatalcare.Theabsolutescoresrangesfrom0to5,with5beingthehighest

while the relative ranking sorts the seven practices from 1 to 7, with 1 being the highest

ranking.Ouroutcomesofinterestaretheabsolutescoreandrelativerankingassignedtoearly

initiation of prenatal care. Panel A in Figure 6 shows that the absolute score assigned by

medicaldirectorstoearlyprenatalcareisonaverage4.8inthetreatmentgroupand4.7inthe

controlgroup.PanelBinFigure6showsthatonaveragetherelativerankingforthispractice

is also similar between the two groups, 2.0 for the treatment group and 1.9 for the control

group. Moreover, thesedifferencesarenotstatisticallysignificantatconventional levels(see

Appendix D). These results suggest that the early initiation of prenatal care is of very high

absoluteandrelativeimportance,andthatthetemporaryfeesdidnothaveaneffectoneither

theabsoluteorrelativeimportanceofthispractice.

X. OtherAlternativeExplanations

PerformanceIndicator.-Financialincentivescouldhavemadetreatedclinicsbelievethat

earlyprenatalcarewouldbeaqualityindicatoronwhichtheywillbeheldaccountableinthe

future. This is unlikely for a number of reasons. First, when the temporary fees were

introducedtothetreatmentclinicsnomentionwasmadeoftheimportanceofearlyinitiation

of prenatal care or that thiswas an indicator of quality. Theywere only informed about the

changeinthefeestructureanditstiming.Second,thereisapublishedwell-establishedlistof

criteriaonwhichclinicsandpersonnelareevaluated(NationalMinistryofHealth2009,2010).

Earlyinitiationofprenatalcareisnotonthislist.PlanNacerregularlyretrainsclinicsonthese

criteria.Neitherthelistofcriterianorthetrainingchangedaroundthetimeoftheintervention.

Substitution.-One alternative explanation for the short-term treatment effects is that

the incentivesarecausing treatmentclinics to try toattractpregnantwomenwhootherwise

28Wewanttostudythebehavioroftheclinicasaunitinsteadofatypicalprenatalcareprovider,suchasanOBGYN,sincethe financial incentivesweredesignedtoaffect thebehaviorof thewhole teamratherthanaparticular individual. As such, CMOs are more representative within a clinic, and were involved in bothmanagingtheclinicalteam(e.g.communityhealthworkersandmedicalstaff)andprovidinghealthcare.

28

wouldhaveusedotherclinics.Thisisunlikelytobetrueasbeneficiarywomenareassignedto

specificclinicswhenenrolled inPlanNacerandcannotsimplygotoanotherclinic toreceive

care. Moreover, clinics and their CHWs have specific geographic areas assigned and do not

conductoutreachactivitiesoutsideofthoseareas. Finally, thenumberofpatientspermonth

andthesharethatinitiatecarebeforeweek13arethesameinthepre-andpost-intervention

periodsforcontrolclinics,andtheaveragemonthlynumberofpatientsisalsothesameinthe

pre-andpost-interventionperiodsforthetreatmentclinics.

InformationSpillovers.-Anotheralternativeexplanationforlong-runresultsisthatafter

the temporary incentivesended,womenwhowerepregnantduring the interventionperiods

passedthemessageoftheimportanceofearlyinitiationofcareontootherwomenwhobecame

pregnant during the post-intervention period. Hence, the persistence of the effect of the

incentivesmightbecausedbyaninformationalspillover.Thisisapossibleargumentandone

thatwouldbeconsistentwithfindinghigherratesofeffectsofearlyprenatalcareinthelong

run in treated clinics. However, the higher amount of community outreach activities in

treatment clinics, themechanism used to generate higher early initiation of care, continued

intothepost-experimentalperiodatthesamelevelasintheinterventionperiod. Ifthelong-

run effects where to be explained by informational spillovers the production of community

outreachactivitieswouldhavetopresentrapiddiminishingmarginalreturnswhichisunlikely

giventheirhighproductivityasevidencedbythe field interviews. Hence, if therewere large

information spillovers in the post-intervention period, then one would expect to see higher

treatment effects in the post-intervention period than in the intervention period, since they

wouldaddtotheeffectsofoutreachactivities. Moreover, ifbeneficiarywomenwerepassing

information to others, then we could expect to see some changes in control clinics as well.

However,theempiricalanalysisoflong-runtrendssuggeststhatthereisno“catch-up”effectof

controlclinics.

LaborContractFrictions.-Anadditionalcandidateexplanationfornotreducingoutreach

activities after the temporary fees disappearedwas not that clinics valued early initiation of

prenatalcarewithoutthefeeincrease,butratherthatCHWemploymentcontractsweresticky

soitwashardandcostlytoreduceCHWs.Thisisunlikelytobetrueforanumberofreasons.

First,continuingtoprovideunproductiveoutreachserviceswascostlyandclinicscouldhave

reassignedCHWstoothertasksorreducedtheiruse.Second,mostoftheCHWexpansionwas

29

throughincreasingCHWhoursandnothiringnewCHWssoitshouldhavebeeneasytoreduce

hourstopre-interventionlevels.Third,anynewCHWhiredcouldhavebeeneasilydismissed.

CHWsworkundertemporarycontractsofvariablelengthwiththefacilitiesandarenotpartof

theformalcivilservicesubjecttomorerigidlaborlaws.Assuch,clinicscaneasilyandquickly

expandand contract the amountof CHW labor that they employ. Fourth, since clinics knew

that the temporary feesonly lasted eightmonthswhen theprogramstarted, theywouldnot

havehirednewCHWsoncontractsforlongerthanthatperiod.Contractswouldthenhavehad

toberenewedornewCHWshiredinordertocontinuethehigherlevelofoutreachactivities.

Finally, even if contracts were sticky clinics should have been able to reduce some of the

outreachactivitiesinthefollow-upperiodbutweseenoreduction.

Career incentives.- It is also possible that even if clinic directors did not value early

initiation of prenatal care, they did not reduce outreach activities in the post intervention

periodbecausetheywereworriedthatthesereductionswouldharmtheircareers(Ashrafetal.

2105). This is highly unlikely for a number of reasons. First, the Government regularly

monitored clinic performance on a large number of indicators, none of which was early

initiationofprenatalcare. Withcareer incentivesatplaymoneyspentonoutreachactivities

couldhavebeenbetterusedonservicesforwhichtheclinicwasexplicitlyaccountable.Second,

theinterventiononlychangedthefeesforashort8-monthperiodoftimeandnotapermanent

changethatmightalsosignalalong-termchangeinprioritiesthatmighthaveshiftedbeliefsof

clinical directors and staff about the importance of early prenatal care. Third, therewas no

accompanying informationexplaining thereason for the temporary feechangeor thatclinics

wouldbeassessedonthisindicator.

XI. Cross-PriceEffects

Whilethemodifiedfeeschedulewasdesignedtoaffectthetimingofthefirstprenatal

visit, providers may have reduced effort supplied to other services, resulting in a lower

provisionofsuchservicestopatients.Wetestforthisbyestimatingtheeffectoftheincentives

on the probability of pregnant women having a valid tetanus vaccine, and the number of

prenatal visits. The results presented in Table 5 report no evidence of cross-price effects,

positiveornegative,ineithertheinterventionperiodorinpost-interventionperiodI. Infact,

the levels of these services appear to be constant over time. While the concern about

30

crowding-out is typically for a context of individual providers facing time and effort

constraints,ourresultsareconsistentwithafirmsettingwheretherearenooveralleffortor

timeconstraints.

XII. BirthOutcomes

Nextweaddressthequestionofwhethertheeffectoftheincentivesforearlyinitiationof

prenatalcaretranslatedintoimprovedbirthoutcomesasmeasuredbybirthweight,lowbirth

weight,andprematurebirth.AsshowninFigure7andreportedinTable6wefindnoeffectof

the incentives on birth outcomes in either the intervention period or the post-intervention

period.

Thereareanumberofpossiblereasonsforthis.First,thesamplecouldbetoosmalltobe

able todetectastatisticallysignificanteffectonoutcomes. However, thepointestimatesare

verysmall,halfofthemarenegativeandtheyareofsimilarmagnitudetodifferencesbetween

treatmentandcontrolgroupsinthepre-interventionperiod.Second,giventhattheresultson

birth outcomes are obtained from an analysis of a subsample of beneficiaries forwhomwe

wereabletomergeprenatalcarerecordswithhospitalmedicalrecords,itispossiblethatthe

results in Table 3 do not hold for this subsample. We therefore replicate the prenatal care

analysisusingonlythesubsampleofwomenforwhomhospitalmedicalrecordsareavailable.

Overall,weobtainsimilarresults tothoseobtainedwiththe fullsample.29 Third,despitethe

medicalliteratureandCPGrecommendation,itispossiblethatearlyinitiationofcarematters

only for a small amount of the general population of pregnant women, such as high-risk

patients. High risk patients include, among others, smokers, substance abusers, those with

poormedicalandpregnancyhistories,andthosewhostartprenatalcareverylateintheirthird

trimesteroronlywhenaproblemoccurs.Itmaybethattheincreaseinearlyinitiationofcare

comes fromprimarily low-riskmotherswhoare less likely tobenefit fromearly initiationof

care. Onewould think that itwouldbe easier topersuade low-riskmothers to comea little

earlierthantoconvincehigh-riskmotherswhoarereluctanttocomeforanycareatall.

Infact,thisisconsistentwiththesmallreductionintheaverageweekspregnantatthe

timeof the firstprenatalvisit. Onaverage,women in the treatmentgroup initiatedprenatal

29Resultsofthisanalysisareavailableuponrequest.

31

careabout1.5weeksearlierthanwomeninthecontrolgroup.Prenatalcaremayaffectbirth

outcomesbydiagnosingandtreating illnesssuchashypertensionandgestationaldiabetesas

wellastryingtochangematernalbehaviorthroughpromotingactivitiessuchasgoodnutrition,

notsmokingandnotconsumingalcohol.Iftheinterventionhadinducedhigh-riskwomenwho

otherwisewouldhavehadtheir1stvisitmuchlaterinthepregnancy,thentheincentivesmay

havehadameasurableimpactonbirthoutcomes.Hence,whiletheincentiveswereeffectivein

increasing early initiation of care, they did notmanage to sufficiently affect the groupmost

likelytobenefit fromit. Thesolutionmightbetoconditionincentivesonattendinghigh-risk

women, but risk is difficult and expensive to identify and verify and therefore may not be

contractible.

XIII. Discussion

In this paper we examined the effects of temporary financial incentives for medical care

providerstoadoptbetterqualitypractices.Weusedthisanalysistoinvestigatewhetherslow

diffusion of better quality practices is driven by perceived low-returns or high costs of

adjustment for adoptinghigh returnpractices. The results suggest that the slowdiffusion is

drivenbyhighcostsofadjustmentasopposedtolowreturns.

WeaddressedthisquestioninthecontextofarandomizedfieldexperimentinMisiones,

Argentina.Theinterventionrandomlyallocatedathree-foldincreaseinthefeepaidtohealth

facilitiesforeachinitialprenatalvisitthatoccursbeforeweek13ofpregnancy.Thispremium

wasimplementedforaperiodof8monthsandthenended.Usingdataonhealthservicesand

birthoutcomesfrommedicalrecords,weestimatedboththeshort-termeffectsoftheincentive

and whether the effects persisted once the direct monetary compensation disappears. We

foundthatpregnantwomenwhoattendedclinicsinthetreatmentgroupwere34%morelikely

to initiate prenatal care beforeweek 13 and that the higher levels of early initiation of care

persistedforatleast24monthsaftertheincentivesended.

We also showed that the temporary incentives motivated clinics to design and

experiment with new outreach strategies to locate and encourage pregnant women to start

careearly. For instance, somecoordinatedwith localpharmacies to findoutwhenawoman

was late in picking up contraceptive pills, then send community health workers to inquire

about last menstruation date, offer instant-read pregnancy tests, and finally encourage the

32

expectant mothers to start prenatal care quickly. We show that outreach activities for

pregnantwomendoubledinthetreatmentgroup.

Finally,weprovidedevidencethatintheabsenceofadjustmentcostsofadoption,itwas

intheclinics’interesttohaveprovidedtheseoutreachservices.First,clinicmedicaldirectors

rankearly initiationof careasoneof thehighestofhealthprioritiesamongallprenatal care

services.Second,outreachactivitiesarereimbursedatahigherratethantheircostforalong

periodbeforetheexperiment. Likewise,beforethetemporary fee increase,clinicswerepaid

for eachprenatal care service, and50%of these additional resourceswereused topay staff

bonuses. As such, the temporary incentives helped to overcome adjustment costs of

developing and experimenting with new outreach strategies for early prenatal care. Once

clinics learnedwhatworkedbest they continued toprovideoutreach activities to encourage

earlyprenatalcareservicesbecausetheyareprofitableandvaluable.

Our results have a number of important policy implications. First, they suggest that

temporary incentives may be effective in motivating long-term provider performance at a

substantiallylowercostthanpermanentincentives.Second,whilewefindthatincentivesare

abletomotivatechangesinclinicalpracticepatterns,wedidnotfindimprovementsinhealth

outcomes.Themonetaryincentivesthatwereimplementedwerenotabletosufficientlyreach

thosewomenforwhomearlyinitiationofprenatalcarewouldhavethelargesthealthimpact.

Therefore, incentives may be made more effective by defining ex-ante the population most

likely to benefit, and tailoring incentives towards this population. However, tailoring

incentivestohighriskpopulationsorthosemostlikelytobenefitfromtheservicesmaynotbe

contractible as these characteristics are typically not observable. This is maybe a major

limitationofusingincentivecontractstoimprovehealthoutcomes.

References

Acland, D., & Levy, M. R. (2015). “Naiveté, projection bias, and habit formation in gymattendance,”ManagementScience,61(1),146-160.

Ashraf, N., Bandiera, O., & Lee S. (2014). “Do-gooders and go-getters: career incentives,selection, and performance in public service delivery,” STICERD - EconomicOrganization and Public Policy Discussion Papers Series 54, Suntory and ToyotaInternationalCentresforEconomicsandRelatedDisciplines,LSE.

Atkin, D., Chaudhry, A., Chaudry, S., Khandelwal, A., & Verhoogen, E. (2015). “OrganizationalBarriers to Technology Adoption: Evidence from Soccer-Ball Producers in Pakistan,”WorkingPaper21417.NationalBureauofEconomicResearch.

33

Baker,G.P., Jensen,M.C.,&Murphy,K. J. (1988). “Compensationand incentives:practicevs.theory,”TheJournalofFinance,43(3),593-616.

Baker, L. C. (2001). “Managed care and technology adoption in health care: evidence frommagneticresonanceimaging,”JournalofHealthEconomics,20(3),395-421.

Baker,L.C.,&Phibbs,C. S. (2002). “Managedcare, technologyadoption, andhealth care: theadoptionofneonatalintensivecare,”TheRANDJournalofEconomics,33(3),524-548.

Barber,S.L.,&Gertler,P.J.(2009).“Empoweringwomentoobtainhighqualitycare:evidencefromanevaluationofMexico'sconditionalcashtransferprogramme,”HealthPolicyandPlanning,24(1),18-25.

Basinga, P., Gertler, P. J., Binagwaho, A., Soucat, A. L., Sturdy, J., & Vermeersch, C.M. (2011).“EffectonmaternalandchildhealthservicesinRwandaofpaymenttoprimaryhealth-care providers for performance: an impact evaluation,”The Lancet,377(9775), 1421-1428.

Bech,M.,Christiansen,T.,Dunham,K.,Lauridsen,J.,Lyttkens,C.H.,McDonald,K.,&McGuire,A.(2009).“Theinfluenceofeconomicincentivesandregulatoryfactorsontheadoptionoftreatmenttechnologies:acasestudyoftechnologiesusedtotreatheartattacks,”HealthEconomics,18(10),1114-1132.

Berwick,D.M.(2003).“Disseminatinginnovationsinhealthcare,”JAMA,289(15),1969-1975.

Blattberg, R. C. & Neslin, S. A. (1990).“Sales promotion: concepts, methods, and strategies,”EnglewoodCliffs,PrenticeHall,NewJersey.

Bloom,Nicholas,BennEifert,AprajitMahajan,DavidMcKenzie,andJohnRoberts.2012.“DoesManagementMatter?EvidencefromIndia,”TheQuarterlyJournalofEconomics,128(1),1-51.

Bonfrer,I.,Soeters,R.,vandePoel,E.,Basenya,O.,Longin,G.,vandeLooij,F.,&vanDoorslaer,E.(2013).“Theeffectsofperformance-basedfinancingontheuseandqualityofhealthcareinBurundi:animpactevaluation,”TheLancet,381,S19.

Bordalo, P., Gennaioli, N. & Shleifer, A. (2012). “Salience theory of choice under risk,” TheQuarterlyJournalofEconomics,127(3):1243-1285.

Bordalo,P.,Gennaioli,N.&Shleifer,A.(2013).“Salienceandconsumerchoice,”TheJournalofPoliticalEconomy,121(5),803-843.

Bresnahan, T., and Trajtenberg, M. (1995). “General Purpose Technologies ‘Engines ofGrowth’?,”JournalofEconometrics65(1):83–108.

Cabana,M.D.,Rand,C.S.,Powe,N.R.,Wu,A.W.,Wilson,M.H.,Abboud,P.A.C.,&Rubin,H.R.(1999). “Why don't physicians follow clinical practice guidelines?: A framework forimprovement,”JAMA,282(15),1458-1465.

Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008). “Bootstrap-based improvements forinferencewithclusterederrors,”TheReviewofEconomicsandStatistics,90(3),414-427.

Campbell,O.M.&Graham,W.J.(2006).“Strategiesforreducingmaternalmortality:Gettingonwithwhatworks,”TheLancet,368(9543),1284-1299.

34

Campbell, S., Reeves, D., Kontopantelis, E., Middleton, E., Sibbald, B., & Roland, M. (2007).“QualityofprimarycareinEnglandwiththeintroductionofpayforperformance,”TheNewEnglandJournalofMedicine,357(2),181-190.

Carroll, G. R., & Hannan, M. T. (2000).“The demography of corporations and industries,”PrincetonUniversityPress.

Carroli, G., Villar, J., Piaggio, G., Khan-Neelofur,D., Gülmezoglu,M.,Mugford,M.,&Bersgjø, P.(2001). “WHO systematic review of randomized controlled trials of routine antenatalcare,”TheLancet,357(9268),1565-1570.

Carroli, G., Rooney, C., & Villar, J. (2001). “How effective is antenatal care in preventingmaternal mortality and serious morbidity? An Overview of the Evidence,”PaediatricandPerinatalEpidemiology,15(s1),1-42.

Cawley, J., & Price, J. A. (2013). “A case study of a workplace wellness program that offersfinancialincentivesforweightloss,”JournalofHealthEconomics,32(5),794-803.

Charness,G.&Gneezy,U.(2009).“Incentivestoexercise,”Econometrica,77(3),909-931.

Chaudhury, N., Hammer, J., Kremer, M., Muralidharan, K., and Rogers, H. 2006. “Missing inAction: Teacher andHealthWorker Absence inDeveloping Countries.”The Journal ofEconomicPerspectives20(1):91–116

Clemens,J.&Gottlieb,J.D.(2014).“Dophysicians'financialincentivesaffectmedicaltreatmentandpatienthealth?”TheAmericanEconomicReview,104(4),1320-1349.

Coleman, J. S., Katz, E., & Menzel, H. (1966). Medical innovation: A diffusion study. SecondEdition,Indianapolis:Bobbs-Merrill.

Comin,D.&Hobijn,B.(2010).“Anexplorationoftechnologydiffusion,”TheAmericanEconomicReview,100(5):2031-59.

Conley,T.G.,&Udry,C.R.(2010).“Learningaboutanewtechnology:PineappleinGhana,”TheAmericanEconomicReview,35-69.

Cutler,D.M.,&Huckman,R.S.(2003).“Technologicaldevelopmentandmedicalproductivity:thediffusionofangioplastyinNewYorkstate,”JournalofHealthEconomics,22(2),187-217.

Cutler,D.M.(2007).“Thelifetimecostsandbenefitsofmedicaltechnology,”JournalofHealthEconomics,26(6),1081-1100.

Das, J., & Gertler, P. J. (2007). “Variations in practice quality in five low-income countries: aconceptualoverview,”HealthAffairs,26(3),w296-w309.

Das, J. & Hammer, J. (2005). “Which Doctor? Combining vignettes and item response tomeasureclinicalcompetence,”JournalofDevelopmentEconomics,78(2),348-383.

Das, J., Hammer, J., & Leonard, K. (2008). “The quality of medical advice in low-incomecountries,”TheJournalofEconomicPerspectives,22(2),93-114.

Davidson, R. & Flachaire, E. (2008). "The Wild bootstrap, tamed at last," Journal ofEconometrics,146(1),162-169.

David, P. (1990). “The dynamo and the computer: an historical perspective on the modernproductivityparadox,”TheAmericanEconomicReview355-361.

35

de Mel, S., McIntosh, C., & Woodruff, C. (2013). “Deposit collecting: Unbundling the role offrequency,salience,andhabitformationingeneratingsavings,”TheAmericanEconomicReview,103(3),387-92.

DeWalque,D.,Gertler,P.J.,Bautista-Arredondo,S.,Kwan,A.,Vermeersch,C.,deDieuBizimana,J.,&Condo, J. (2015). “Usingproviderperformance incentives to increaseHIV testingandcounselingservicesinRwanda,”JournalofHealthEconomics,40(2),1-9.

DellaVigna,S.(2009).“Psychologyandeconomics:evidencefromthefield,”JournalofEconomicLiterature,47(2),315-372.

Duflo,E.,Kremer,M.,andRobinson,J.(2011).“NudgingFarmerstoUseFertilizer:TheoryandExperimentalEvidencefromKenya,”TheAmericanEconomicReview101(6):2350–90.

Dupas,P.(2014).“Short-runsubsidiesandlong-runadoptionofnewhealthproducts:Evidencefromafieldexperiment,”Econometrica,82(1),197-28.

EldridgeC and Palmer,N. 2009. “Performance-based payment: some reflections on thediscourse,evidenceandunansweredquestions.”HealthPolicyandPlanning,24:160-6.

EvansW.N.andStechLienD.2005."Thebenefitsofprenatalcare:EvidencefromthePATBusStrike,"JournalofEconometrics,125(1),207-239.

Flores,G., Ir,P.,Men,C.R.,O’Donnell,O.,&vanDoorslaer,E. (2013). “Financialprotectionofpatients through compensation of providers: The impact of health equity funds inCambodia,”JournalofHealthEconomics,32(6),1180-1193.

Foster,A.D.,&Rosenzweig,M.R.(1995).“Learningbydoingandlearningfromothers:Humancapitalandtechnicalchangeinagriculture,”JournalofpoliticalEconomy,1176-1209.

Gelbach,J.B.,Klick,J.,&Stratmann,T.(2009).“Cheapdonutsandexpensivebroccoli:theeffectofrelativepricesonobesity,”WorkingPaper.

Geroski,P.A.(2000).“Modelsoftechnologydiffusion,”Researchpolicy,29(4),603-625.

Gertler, P., Giovagnoli, P. I., &Martinez, S.W. (2014). “Rewarding provider performance toenableahealthystarttolife:evidencefromArgentina'sPlanNacer,”WorldBankPolicyResearchWorkingPaper,6884,WorldBank,Washington,DC.

Gertler, P., & Vermeersch, C. (2012). “Using performance incentives to improve healthoutcomes,”WorldBankPolicyResearchWorkingPaper.

Gertler, P. & Vermeersch, C. (2013). “Using performance incentives to improvemedical careproductivity and health outcomes,”NBERWorking Papers 19046, National Bureau ofEconomicResearch,Cambridge,MA.

Gibbons, R. (1997). “An introduction to applicable game theory,” Journal of EconomicPerspectives,11(1),127-149.

Griliches, Z. (1957). “Hybrid corn: An exploration in the economics of technological change,”Econometrica,501-522.

Grol, R. P. T. M. (1990). “National standard setting for quality of care in general practice:attitudesofgeneralpractitionersandresponsetoasetofstandards,”BritishJournalofGeneralPractice,40(338),361-364.

36

Grol,R.(2001).“Successesandfailuresintheimplementationofevidence-basedguidelinesforclinicalpractice,”MedicalCare,39(8),11-46.

Grol,R.,&Grimshaw,J.(2003).“Frombestevidencetobestpractice:effectiveimplementationofchangeinpatients'care,”TheLancet,362(9391),1225-1230.

Grossman,M., and Joyce, T. (1988). "Unobservables, pregnancy resolutions, and birthweightproductionfunctionsinNewYorkCity,"JournalofPoliticalEconomy,98:983-1007.

Hall,B.,&Khan,B.(2003).“AdoptionofNewTechnology,"inNewEconomyHandbook,ed.byD.C.Jones.AcademicPress,SanDiego.

Hall, B. (2005). “Innovation and Diffusion,” in Oxford Handbook of Innovation, ed. by J.Fagerberg,D.C.Mowery,andR.R.Nelson,pp.459-484.OxfordUniversityPress.

Hannan,M.T.,&Freeman, J. (1984). “Structural inertiaandorganizational change,”AmericanSociologicalReview,149-164.

Hawkes,S.J.,Gomez,G.B.,&Broutet,N.(2013).“Earlyantenatalcare:doesitmakeadifferenceto outcomes of pregnancy associated with syphilis? A systematic review and meta-analysis,”PloSone,8(2),e56713.

Holmstrom,B.&Milgrom,P. (1991). “Multitaskprincipal-agentanalyses: Incentivecontracts,assetownership,andjobDesign,”JournalofLaw,Economics,&Organization,7(SpecialIssue),24-52.

Hudak,B.B.,O'Donnell,J.,&Mazyrka,N.(1995).“Infantsleepposition:pediatricians'advicetoparents,”Pediatrics,95(1),55-58.

Huillery,E.&Seban,J.(2014).“Pay-for-Performance,motivationandfinaloutputinthehealthsector:ExperimentalevidencefromtheDemocraticRepublicofCongo,”WorkingPaper,DepartmentofEconomics,SciencesPo,Paris.

Imbens,G.W.&Angrist,J.D.(1994).“IdentificationandestimationofLocalAverageTreatmentEffects,”Econometrica,62(2),467-475.

John, L. K., Loewenstein, G., Troxel, A. B., Norton, L., Fassbender, J. E., & Volpp, K. G. (2011).“Financial incentives forextendedweight loss: a randomized, controlled trial,”JournalofGeneralInternalMedicine,26(6),621-626.

Kahneman,D.(2012).“Thinking,fastandslow,”Farrar,StrausandGiroux,NewYork.

Karlan, D., M. McConnell, S. Mullainathan & Jonathan Zinman (2015). “Getting to the top ofmind:Howremindersincreasesavings,”ManagementScience,forthcoming.

Kirmani,A.&Rao,A.R.(2000).“Nopain,nogain:Acriticalreviewoftheliteratureonsignalingunobservedproductquality,”JournalofMarketing,64(2),66–79.

Kolstad, J. T. (2013). “Information and quality when motivation is intrinsic: Evidence fromsurgeonreportcards,”TheAmericanEconomicReview,103(7),2875-2910.

Lazear, E. P. (2000). “Performance pay and productivity,” The American Economic Review,90(5),1346-1361.

Lazonick,W. (1979). “Industrial Relations andTechnical Change: The Case of the Self-ActingMule,”CambridgeJournalofEconomics3(3):231–62.

37

Leonard, K. L. & Masatu, M. C. (2010), “Professionalism and the know-do gap: Exploringintrinsic motivation among health workers in Tanzania,” Health Economics, 19(12),1461-1477.

Main,D.S.,Cohen,S. J.,&DiClemente,C.C. (1995).“Measuringphysicianreadinesstochangecancerscreening:preliminaryresults,”AmericanJournalofPreventiveMedicine.

McCrary,Justin.“Manipulationoftherunningvariableintheregressiondiscontinuitydesign:Adensitytest,”JournalofEconometrics142,no.2(2008):698-714.

Mansfield,E.(1961).“Technicalchangeandtherateofimitation,”Econometrica,741-766.

Mas-Colell, A., Whinston, M., & Green, J. (1995). Microeconomic Theory, Oxford, England:OxfordUniversityPress.

Miller,G.&Babiarz,K.S. (2013). “Pay-for-performance incentives in low-andmiddle-incomecountryhealthprograms,”NBERWorkingPapers18932,NationalBureauofEconomicResearch,Inc.

Mohanan,M., Vera-Hernández,M., Das, V., Giardili, S., Goldhaber-Fiebert, J. D., Rabin, T. L., &Seth,A.(2015).“Theknow-dogapinqualityofhealthcareforchildhooddiarrheaandpneumoniainruralIndia,”JAMAPediatrics.

National Ministry of Health (2009). "Informe de gestión Plan Nacer," Área Técnica, UnidadEjecutoraCentral.BuenosAires,Argentina.

National Ministry of Health (2010). "Informe de gestión Plan Nacer," Área Técnica, UnidadEjecutoraCentral.RevisedversionMarch.BuenosAires,Argentina.

Pathman,D.E.,Konrad,T.R.,Freed,G.L.,Freeman,V.A.,&Koch,G.G.(1996).“Theawareness-to-adherencemodelof thesteps toclinicalguidelinecompliance: thecaseofpediatricvaccinerecommendations,”MedicalCare,34(9),873-889.

Parente, S. L., & Prescott, E. C. (1994). Barriers to technology adoption and development.JournalofpoliticalEconomy,298-321.

Phelps, C. E. (2000). “Information diffusion and best practice adoption,”Handbook of HealthEconomics,1,223-264.

Rosenberg,N.(1972).“Factorsaffectingthediffusionoftechnology,”Explorationsineconomichistory,10(1),3-33.

___________(1982).Insidetheblackbox:technologyandeconomics.CambridgeUniversityPress.

Rosenzweig, M.R., and Schultz P. T. (1983). "Estimating a household production function:Heterogeneity, the demand for health inputs, and their effects on birth weight."TheJournalofPoliticalEconomy,723-746.

Rous,J.J.,Jewell,R.T.andBrown,R.W.(2004).“Theeffectofprenatalcareonbirthweight:afull-informationmaximumlikelihoodapproach,”Healtheconomics,13(3),pp.251-264.

Royer, H. M. Stehr, and J. Sydnor (2012). “Incentives, commitments and habit formation inexercise: evidence from a field experimentwithworkers at a Fortune-500 company,”NBERWorking Paper 18580, forthcoming in American Journal of Economics: AppliedEconomics.

38

Ryan,B.,&Gross,N.C. (1943). “Thediffusionofhybridseedcorn in twoIowacommunities,”Ruralsociology,8(1),15.

Schuster,M.A.,McGlynn,E.A.,&Brook,R.H.(1998).“HowgoodisthequalityofhealthcareintheUnitedStates?,”MilbankQuarterly,76(4),517-563.

Schwarcz,R.,Uranga,A.,Lomuto,C.,Martinez,I.,Galimberti,D.,García,O.M.,Etcheverry,M.E.,& Queiruga, M. (2001). "El cuidado prenatal: Guía para la práctica del cuidadopreconcepcionalydelcontrolprenatal."NationalMinistryofHealth,Argentina.

Skinner,J.,&Staiger,D.(2009).“Technologydiffusionandproductivitygrowthinhealthcare,”ReviewofEconomicsandStatistics,97(5),951-964.

Taylor, S. E., & Thompson, S. C. (1982). “Stalking the elusive ‘vividness’ effect,”PsychologicalReview,89(2),155.

Thaler, R. H. & Sunstein C.R. (2009). “Nudge: Improving decisions about health, wealth, andhappiness,”PenguinBooks,NewYork.

Volpp, K. G., John, L. K., Troxel, A. B., Norton, L., Fassbender, J., & Loewenstein, G. (2008).“Financial incentive–based approaches for weight loss: a randomizedtrial,”JAMA,300(22),2631-2637.

Volpp,K.G.,Troxel,A.B.,Pauly,M.V.,Glick,H.A.,Puig,A.,Asch,D.A.,...&Audrain-McGovern,J.(2009). “A randomized, controlled trial of financial incentives for smokingcessation,”TheNewEnglandJournalofMedicine,360(7),699-709.

Wooldridge, J. M. (2007). “Inverse probability weighted estimation for general missing dataproblems,”JournalofEconometrics,141(2),1281-1301.

World Health Organization (2006). “Standards for maternal and neonatal care: Provision ofeffectiveantenatalcare,”WorldHealthOrganization,Geneva.

39

FiguresandTables

FIGURE1:PROVIDERCOMPLIANCEWITHCLINICALPRACTICEGUIDELINES

Notes: The Figure shows the percentage of adherence to clinical practice guidelines for different countriesandconditionsobtained fromseveralstudiesonthe topic.Authors’elaborationbasedon(-)Schusteretal.(1998); (+) Grol (2001); (++) Campbell et al. (2007); (*) Das and Gertler (2007); and (#) Gertler andVermeersch(2012).CHD:CoronaryHeartDisease.

75%$

26%$

18%$

46%$

58%$

24%$

38%$

45%$

67%$

50%$

70%$

60%$

84%$

81%$

85%$

0%$ 10%$ 20%$ 30%$ 40%$ 50%$ 60%$ 70%$ 80%$ 90%$

Mexico$3$Prenatal$Care$*$

India$3$Tuberculosis$$*$

India$3$Diahrrea$*$

Indonesia$3$Tuberculosis$$*$

Indonesia$3$Diahrrea$*$

Tanzania$3$Malaria$*$

Tanzania$3$Diahrrea$*$

Rwanda$3$Prenatal$Care$#$

Netherlands$3$Family+$

USA3PrevenQve$Care$3$

USA3Acute$Care$$3$

USA3Chronic$CondiQons$$3$

UK3Asthma$++$

UK3Diabetes$++$

UK3CHD$++$

Adherence$To$Protocol$

40

FIGURE2:DENSITIESOFWEEKSPREGNANTAT1STPRENATALVISIT

Notes: Densities estimated using anEpanechnikov kernelwith optimal bandwidth. P-vales of Kolmogorov-Smirnov tests of equality of distributions between groups reported below figure. The two vertical linesindicateweeks13and20ofpregnancy.Source:Authors’ownelaborationbasedondatafromtheprovincialmedicalrecordinformationsystem.

0.0

2.0

4.0

6De

nsity

0 10 13 20 30 40Weeks Pregnant at First Prenatal Visit

Treatment ControlK-S Test: p-value = .823

Panel A: Pre-Intervention Period

0.0

2.0

4.0

6De

nsity

0 10 13 20 30 40Weeks Pregnant at First Prenatal Visit

Treatment ControlK-S Test: p-value = .031

Panel B: Intervention Period0

.02

.04

.06

Dens

ity

0 10 13 20 30 40Weeks Pregnant at First Prenatal Visit

Treatment ControlK-S Test: p-value = .004

Panel C: Post-Intervention Period I

0.0

2.0

4.0

6De

nsity

0 10 13 20 30 40Weeks Pregnant at First Prenatal Visit

Treatment ControlK-S Test: p-value = .009

Panel D: Post-Intervention Period II

41

FIGURE3:MEANNUMBEROFWEEKSPREGNANTAT1STPRENATALVISIT

Notes:Thefirsttwopoints(circles)aremeansfor6-monthperiodspriortotheinterventionperiod.Thethirdpoint (Diamond) corresponds to the 8-month intervention period. The fourth and fifth points (triangles)correspondto6-monthsperiodsaftertheinterventionperiod,whilethelastpoint(triangle)isfora3-monthperiod.

FIGURE4:PROPORTIONOFMOTHERSWITH1STPRENATALVISITBEFOREWEEK13OFPREGNANCY

Notes:Thefirsttwopoints(circles)aremeansfor6-monthperiodspriortotheinterventionperiod.Thethirdpoint (Diamond) corresponds to the 8-month intervention period. The fourth and fifth points (triangles)correspondto6-monthsperiodsaftertheinterventionperiod,whilethelastpoint(triangle)isfora3-monthperiod.

1516

1718

19

Wee

ks P

regn

ant a

t Firs

t Pre

natal

Visi

t

Jan-Jun 2009

Jul-Dec 2

009

Jan-Apr 2010

May-Dec 2

010

Jan-Jun 2011

Jul-Dec 2

011

Jan-Mar 2

012

Period

Treatment ControlPre-Int. period Int. periodPost-Int. period

.25

.3.3

5.4

.45

Firs

t Visi

t Bef

ore

Wee

k 13

Jan-Ju

n 200

9

Jul-Dec

2009

Jan-Apr

2010

May-Dec

2010

Jan-Ju

n 2011

Jul-Dec

2011

Jan-M

ar 20

12

Period

Treatment ControlPre-Int. period Int. periodPost-Int. period

42

Figure5:NumberofClinicOutreachActivities

Notes:Thebarsreportthemeanandmediannumberofoutreachactivitiesthatresultedinactualmaternal-child service at the clinic, per trimester for the pre-intervention period (January 2009-April 2010), theinterventionperiod(May-December2010),andpost-interventionperiodI(January2011-March2012)

27.6

18.7

49.8

27.1

42.0

26.0

010

2030

4050

60

Num

ber o

f out

reac

h ac

tivtie

s

Pre-Int. Intervention Post-Int.

Mean

Treatment Control

9.5

10.8

21.2

8.8

22.4

9.5

05

1015

2025

Num

ber o

f out

reac

h ac

tiviti

es

Pre-Int. Intervention Post-Int.

Median

Treatment Control

43

FIGURE6:IMPORTANCEOFPRENATALCARESERVICES

Notes:PanelAandPanelBreport theaverageof theabsolutescoreandrelativeranking, respectively, thatmeasures the importance given by clinics to seven different prenatal care procedures including initiatingprenatal care prior toweek13 of pregnancy (AppendixD). The absolute scores range from1 to 5,with 5beingthehighestscoreintermsofimportance.

FIGURE7:BIRTHWEIGHTDENSITIES

Notes: Densities estimated using an Epanechnikov kernelwith optimal bandwidth.P-vales of Kolmogorov-Smirnov tests of equality of distributions between groups reported below figure. Source: Authors’ ownelaborationbasedonmedicalrecordinformationsystem.

0.0

005

.001

Dens

ity

1000 2000 3000 4000 5000Birth weight

Treatment ControlK-S Test: p-value = .376

Panel A: Pre-Intervention Period

0.0

005

.001

Dens

ity

1000 2000 3000 4000 5000Birth weight

Treatment ControlK-S Test: p-value = .54

Panel B: Intervention Period

6.65.9

3.34.1

4.74.1

2.43.5

1.83.4

2.63.0

1.92.0

0 1 2 3 4 5 6 7

Thorax X-Ray

Combined Diphtheria/Tetanus vaccine

Blood test without serology

Prenatal ultrasound

Bio-psycho-social counseling visit

Blood test with serology

First prenatal visit before week 13

Panel B. Relative ranking of services

Treatment Control

4.74.8

4.54.7

4.44.4

3.54.4

3.53.1

1.62.7

0.20.2

0 1 2 3 4 5 6 7

First prenatal visit before week 13

Blood test with serology

Bio-psycho-social counseling visit

Prenatal ultrasound

Combined Diphtheria/Tetanus vaccine

Blood test without serology

Thorax X-Ray

Panel A. Absolute value of services

Treatment Control

0.0

005

.001

Den

sity

1000 2000 3000 4000 5000Birth weight

Treatment ControlK-S Test: p-value = .825

Panel C: Post-Intervention Period

44

TABLE1:PAYMENTSFOR1STPRENATALVISITTimePeriod Dates Paymentfor1stPrenatalVisit

Begin End BeforeWeek

13ofpregnancy

Atweek13ofpregnancyor

afterPre-Intervention January2009 April2010 $40ARS $40ARS

Intervention May2010 December2010 $120ARS $40ARSPostIntervention January2011 December2012 $40ARS $40ARS

Notes:NationalMinistryofHealth,Argentina(2010b)

TABLE2:BASELINEDESCRIPTIVESTATISTICS Assigned

TreatmentGroup AssignedControlGroup p-Valuefortestof

equalityofmeans

Mean(s.d.) N Mean

(s.d.) N

Largesample

WildBoot-

StrappedWeeksPregnantat1stPrenatalVisit 17.5 743 17.6 497 0.89 0.84

(7.48) (7.74)

1stVisitbeforeWeek13ofPregnancy 0.35 743 0.33 497 0.57 0.56 (0.48) (0.47)

TetanusVaccineDuringPrenatalVisit 0.80 743 0.84 497 0.34 0.41 (0.40) (0.37)

NumberofPrenatalVisits 4.68 743 4.28 497 0.39 0.45 (2.94) (2.77)

BirthWeight(grams) 3,328 552 3,291 379 0.36 0.37 (519) (558)

LowBirthWeight(<2500grams) 0.06 552 0.06 379 0.96 0.98 (0.23) (0.23)

Premature(gestationalage<37weeks) 0.09 319 0.10 249 0.83 0.82 (0.29) 0.30

Notes:Thistablepresentsmeansandstandarddeviationsinparenthesesforthetreatmentandcontrolgroupsduringthe16-monthpre-interventionperiodfromJanuary2009throughApril2010.P-valuesfortestsequalityoftreatmentand control groupsmeans are presented in the last 2 columns.We present both the p-value computed for largesamplesandaWildbootstrappedp-value that is robust insampleswithsmallnumbersofclusters (Cameronetal.2008).OurWildbootstrapprocedureassigns symmetricweightsandequalprobabilityafter re-sampling residuals(DavidsonandFlachaire2008)anduses999replications.

45

TABLE3:EFFECTSONTEMPORARYINCENTIVESONTIMINGOF1STPRENATALVISIT (1) (2) (3)

InterventionPeriod

Post-InterventionPeriodI

Post-InterventionPeriodII

A.WeeksPregnantat1stPrenatalVisitTreatment -1.47** -1.63** -2.47**

(0.71) (0.75) (1.02)LargeSamplep-value 0.04 0.03 0.02

WildBootstrappedp-value 0.08 0.03 0.03ControlGroupMean 17.80 17.90 20.10

SampleSize 769 1,296 710B.FirstPrenatalVisitBeforeWeek13ofPregnancy

Treatment 0.11** 0.08** 0.08** (0.04) (0.04) (0.04)

LargeSamplep-value 0.01 0.02 0.04WildBootstrappedp-value 0.03 0.05 0.06

ControlGroupMean 0.31 0.34 0.27SampleSize 769 1,296 710

Notes:ThistablereportsLATEestimatesofthetreatmenteffectestimatedfrom2SLSregressionsofthedependent variable on actual treatment status instrumentedwith clinic treatment assignment type.Thep-valuesarefortestsofthenullthatthedifferenceisequaltozero.Wepresentboththep-valuecomputed for large samples and a Wild bootstrapped p-value that is robust in samples with smallnumbersofclusters(Cameronetal.2008).OurWildbootstrapprocedureassignssymmetricweightsand equal probability after re-sampling residuals (Davidson and Flachaire 2008) and uses 999replications.*p<0.10,**p<0.05,***p<0

TABLE4:IMPACTONLOGNUMBEROFOUTREACHACTIVITIES (1) (2) InterventionPeriod Post-InterventionPeriodI

Treatment 0.47** 0.56** (0.23) (0.22)

LargeSamplep-value 0.04 0.01WildBootstrappedp-value 0.04 0.02Log(ControlGroupMean) 1.93 1.93

SampleSize 324 324

Notes:ThistablereportsLATEestimatesofthetreatmenteffectestimatedfrom2SLSregressionsofthedependentvariableonactual treatment status instrumented with clinic treatment assignment type. The p-values are for tests of the null that thedifference is equal to zero.We present both thep-value computed for large samples and aWild bootstrappedp-value that isrobust in samples with small numbers of clusters (Cameron et al. 2008). Our Wild bootstrap procedure assigns symmetricweightsandequalprobabilityafterre-samplingresiduals(DavidsonandFlachaire2008)anduses999replications.Column(1)reportstheresultsforthesampleobservedinan8-monthinterventionperiod(May2010–December2010).Column(2)reportstheresultsforthesampleobservedinthe15-monthperiodfollowingtheendoftheintervention(January2011–March2012).).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

46

TABLE5:CROSS-PRICEEFFECTS(SPILLOVER) (1) (2) InterventionPeriod Post-InterventionPeriodI

A.TetanusVaccine Treatment 0.02 -0.02

(0.08) (0.05)LargeSamplep-value 0.76 0.62

WildBootstrappedp-value 0.75 0.67ControlGroupMean 0.79 0.84

SampleSize 769 1,053B.Numberofvisits

Treatment 0.39 0.51 (0.33) (0.58)

LargeSamplep-value 0.24 0.38WildBootstrappedp-value 0.27 0.41

ControlGroupMean 4.05 4.40SampleSize 769 1,053

Notes Notes: This table reports LATE estimates of the treatment effect estimated from 2SLS regressions of the dependentvariableonactualtreatmentstatusinstrumentedwithclinictreatmentassignmenttype.Thep-valuesarefortestsofthenullthatthedifferenceisequaltozero.Wepresentboththep-valuecomputedforlargesamplesandaWildbootstrappedp-valuethat is robust in samples with small numbers of clusters (Cameron et al. 2008). Our Wild bootstrap procedure assignssymmetricweightsandequalprobabilityafterre-samplingresiduals(DavidsonandFlachaire2008)anduses999replications.Column (1) reports the results for the sample observed in an 8-month intervention period (May 2010 – December 2010).Column(2)reportstheresultsforthesampleobservedinthe12-monthperiodfollowingtheendoftheintervention(January2011–December2011).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

47

TABLE6:BIRTHOUTCOMES (1) (2) InterventionPeriod Post-InterventionPeriodI

A.BirthWeight Treatment -37.34 25.11

(48.61) (40.67)LargeSamplep-value 0.44 0.54

WildBootstrappedp-value 0.49 0.51ControlGroupMean 3,304 3,279

SampleSize 555 802B.LowBirthWeight

Treatment 0.01 -0.01 (0.02) (0.02)

LargeSamplep-value 0.63 0.60WildBootstrappedp-value 0.61 0.56

ControlGroupMean 0.05 0.06SampleSize 555 802

C.PrematureBirth Treatment 0.03 -0.04

(0.03) (0.02)LargeSamplep-value 0.31 0.08

WildBootstrappedp-value 0.28 0.12ControlGroupMean 0.09 0.12

SampleSize 414 708

Notes:ThistablereportsLATEestimatesofthetreatmenteffectestimatedfrom2SLSregressionsofthedependentvariableonactual treatment status instrumentedwith clinic treatment assignment type. The p-values are for tests of the null that thedifference isequal tozero.Wepresentboththep-valuecomputed for largesamplesandaWildbootstrappedp-valuethat isrobust in sampleswith small numbers of clusters (Cameron et al. 2008). OurWild bootstrap procedure assigns symmetricweightsandequalprobabilityafterre-samplingresiduals(DavidsonandFlachaire2008)anduses999replications.Column(1)reports the results for the sample observed in an 8-month intervention period (May 2010 – December 2010). Column (2)reports the results for the sample observed in the 12-month period following the end of the intervention (January 2011 –December2011).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

48

ForOnlinePublication

AppendixA:TestofMisreportingWeeksPregnantat1stPrenatalVisit

One concern is that the financial incentives may cause clinics to misreport the week ofpregnancy at the first visit. In this appendix we report the results of test for this behavior.Recallthatinourmainanalysisweconstructtheweekofpregnancyatthefirstvisitusingthedateofthefirstvisitandthelastmenstrualdate(LMD)asreportedbythewomen.Ifthelatterisnotavailableweuse theestimateddateofbirth(EDD)asrecordedbythephysician in thefirstvisit.TheEDDiscalculatedoff theLMDasreportedby thewomenduringher firstvisit.Whileclinicmedicalrecordsshouldcontainbothdates,about10%ofrecordsaremissingtheLMD.

Onepossiblewayofmisreporting theweekofpregnancyat the firstvisit is tochangetheLMDandtheEDDinthepatient’sclinicalmedicalrecord.Forinstance,ifawomanisinher21stweekofpregnancyatthefirstvisit,thephysiciancouldadd7daystotheLMDandEDDsothatthevisitfallsintothe20thweekofpregnancy.Bothwouldhavetobechangedinordertodeceivetheauditors.

Totestforthispossibilityweusegestationalageatbirth(GAB)inweeksmeasuredbyphysical examinationat the timeof birth, registered in thehospitalmedical record.We thencomparetheweekselapsedfromthe firstprenatalvisit to thedeliverydatebasedonGABtoweekselapsedfromfirstvisittothedeliverydatebasedonEDD.WhileEDDiscollectedbytheclinicwhohasanincentivetomisreport,theGABiscollectedbythehospitalattimeofdeliverywherethereisnoincentivetomisreport.

FigureA1plotsthenumberofweekstodeliveryfromthetimeofthe1stvisitbasedonGAB (y-axis) to the one based on EDD (x-axis). If there is no difference between the twomeasures, thenall of thedates should fall on the45-degreeblue line.There shouldbe somedifferencesasEDDisanestimatethatassumesnoprematurityatbirth,andtherecouldbedataentry inGAB andEDDand recall errors inEDD. FigureA1 shows that almost all of thedataembracetheblue45-degreelineandmostoftheobservationsoffthelinearesituatedaboveit,consistentwithprematurityexplainingthedifferences.

Wealsoexplorewhetherthere isanymanipulationof thedataat thethresholdof the13thweekofpregnancy.FigureA2showsthatthereisnodiscontinuityatthisthresholdusingthe test proposed by McCrary (2008) for manipulation at the threshold in studies that useRegressionDiscontinuityastheirresearchdesign.Thep-valueatthediscontinuityis0.838.

If the clinic changes the EDD in order to capture higher payments, wewould expectgreater differences, for the treatment group, between GAB and EDD below the 12-weekthresholdsthanaboveitduringtheinterventionperiodwhentheincentivesareinforce,butnodifferences in the pre-intervention period. In order to test this, we estimate the followingdifferenceindifferenceregression:

𝑊1LPQR = 𝛼L + 𝛽𝑊1L

STT + 𝛾𝐼 𝑊1LSTT < 13 + 𝛿𝐼 𝑊1L

STT < 13 𝑇L + 𝜀1L (A1)

49

where𝑊1LSTTisweeksofpregnantatthefirstvisitbasedonEDDforindividualigettingcarein

clinicj,𝑊ZLPQRisthenumberofweeksatthefirstvisitbasedonGABforindividualigettingcare

inclinicj,𝛼L isaclinicfixedeffect,𝐼 𝑊1LSTT < 13 isanindicatorofwhethertheclinicreported

thefirstvisittobeinthefirst12weeksbasedonEDD,𝑇L isanindicatorofwhethertheclinicwasactuallytreated,and𝜀1Lisanerrorterm.

In the absence of misreporting and no prematurity there should be no differencebetweenthetwomeasuresand𝛽wouldhaveacoefficientof1.However,becauseprematurebirthsoccurbeforeEDD,weexpect𝛽tobeclosetobutlessthanone.Thenwecaninterprettheothercoefficientsastheeffecton𝑊1L

PQR−𝛽𝑊1LSTTaccountingforaverageweeksofprematurity.

SothedependentvariableistheerrorinEDDinforecastingactualdeliverydate.Equation(A1)takes on a difference in difference interpretation in the sense the we are differencing thechange in the forecast error between the pre-intervention and intervention periods for thegroupofpregnantwomenforwhichaclinicreportsashavingtheirfirstvisitbefore13weeksandthegroupofpregnantwomenforwhichaclinicreportshavingthefirstvisitinweek13orlater.Ifthereisnodifferenceintheerrorforthetreatmentgroupinthepostperiodthen𝛿,theinteraction between treatment and reported having the first period beforeweek 13,will bezero.Wefindnoevidenceofmisclassificationbytreatedclinics(SeeTableA1).

FIGUREA1:COMPARISONOFWEEKSPREGNANTAT1STPRENATALVISITBASEDONGESTATIONALAGEATBIRTHAND

BASEDONDATEOFLASTMENSTRUATION

Notes:Authors’ownelaborationbasedondatafromtheprovincialmedicalrecordinformationsystem.

010

2030

40

Wee

ks P

regn

ant a

t Firs

t Pre

nata

l Vis

it co

nstru

cted

usi

ng G

AB

0 10 20 30 40

Weeks Pregnant at First Prenatal Visit constructed using EDD

50

FIGUREA2:TESTFORMISREPORTINGWEEKSOFPREGNANCYATTHETHRESHOLDOFTHE13THWEEK

BASEDONTHE“MANIPULATION”TESTINMCCRARY(2008)

Notes:Authors’ownelaborationbasedonMcCrary(2008).Thep-valueatthediscontinuityis0.838.

TABLEA1:TESTFORMISREPORTINGWEEKSPREGNANTAT1STPRENATALVISITDependentVariable:WeeksPregnantat1stPrenatalVisit,byGestationalAgeatBirth

WeeksPregnantbyEDD 0.90*** (0.02)

1(WeeksPregnantbyEDD<13) -0.13 (0.31)

1(WeeksPregnantbyEDD<13)x1(Treated=1) -0.03 (0.44)

Constant 1.33*** (0.39)

Observations 1730

AdjustedR2 0.82

0.0

5.1

.15

−20 0 20 40 60Weeks of pregnancy at the first visit

51

Notes: The dependent variable is weeks pregnant at the first prenatal visit constructed usinggestationalageatbirth.Theindependentvariableisweekspregnantatthefirstvisitconstructedbyusingthelastdayofmenstruationorestimateddeliverydate(EDD).Theinteractionterminteractsadichotomous indicator forwhether the visitwas beforeweek13 and a dichotomous indicator forwhether theclinicwasactually treated.Theregressioncontrols forclinic fixedeffectsbyaddingabinaryindicatorforeachclinicinthesample.Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

AppendixB:Robustnesstestresults

FIGUREB1:ESTIMATESOFIMPACTONWEEKSPREGNANTAT1STPRENATALVISITDROPPINGTHEOBSERVATIONSFOREACHCLINICONEATATIME

Notes:Thisfigureplotsdifferenttreatmenteffectscomputedbydroppingoneclinicatatimeforweekspregnantatthefirstvisitprenatalvisit.WerunOLSregressionoftheoutcomecomparingeachclinicassignedtothetreatmentgrouptoallclinicsassignedtothecontrolgrouppoolingtheinterventionperiodandpostinterventionperiodI(henceMay2010-March2012).Thex-axisissortedfromthelowesttothehighesttreatmenteffect.Thedashedbluelineistheintent-to-treateffectcalculatedbypoolingtheinterventionandthefirstpostinterventionperiod.Theverticallinesare95%confidenceintervalsconstructedusingstandarderrorsobtainedfromtheWildbootstrapprocedure.

52

FIGUREB2:ESTIMATESOFIMPACTONWEEKS1STPRENATALVISITBEFOREWEEK13DROPPINGTHEOBSERVATIONSFOREACHCLINICONEATATIME

Notes:Thisfigureplotsdifferenttreatmenteffectscomputedbydroppingoneclinicatatimeforfirstprenatalvisitbeforeweek13.WerunOLS regressionof theoutcomecomparingeachclinicassigned to the treatmentgroup toall clinicsassigned to the controlgrouppoolingtheinterventionperiodandpostinterventionperiodI(henceMay2010-March2012).Thex-axisissortedfromthelowesttothehighest treatment effect. The dashed blue line is the intent-to-treat effect calculated by pooling the intervention and the first postinterventionperiod.Theverticallinesare95%confidenceintervalsconstructedusingstandarderrorsobtainedfromtheWildbootstrapprocedure.

FIGUREB3:INDIVIDUALCLINICTREATMENTEFFECTSFORWEEKSPREGNANTAT1STPRENATALVISIT

Notes: This figure plots individual clinic treatment effects for the outcome of weeks pregnant at first prenatal visit. We run OLSregressionoftheoutcomecomparingeachclinicassignedtothetreatmentgrouptoallclinicsassignedtothecontrolgrouppoolingtheintervention period and the post-intervention period I (May 2010-March 2012). One treatment clinic is not included because of itsinsufficientsamplesize.Thiscliniccorrespondstooneofthetwothatdidnottakeuptreatment.Thetrianglesymbolreferstotheclinicthatwas assigned to treatment but did not take up the treatment. The x-axis is sorted from the lowest to the highest clinic-specificimpact.Thedashedbluelineistheintent-to-treateffectcalculatedbypoolingtheinterventionandthefirstpostinterventionperiod.Theverticallinesare95%confidenceintervalsconstructedusingstandarderrorsobtainedfromtheWildbootstrapprocedure.

-6-4

-20

2W

eeks

Pre

gnan

t at F

irst P

rena

tal V

isit

Treated Not treated95% C.I.

-.10

.1.2

.3Fi

rst V

isit B

efor

e W

eek

13

Treated Not treated95% C.I.

53

FIGUREB4:INDIVIDUALCLINICTREATMENTEFFECTSFOR1STPRENATALVISITBEFOREWEEK13OFPREGNANCY

Notes:Thisfigureplotsindividualclinictreatmenteffectsfortheoutcomeoffirstprenatalvisitbeforeweek13.WerunOLSregressionoftheoutcomecomparingeachclinicassignedtothetreatmentgrouptoallclinicsassignedtothecontrolgrouppoolingtheinterventionperiod and post intervention period I (henceMay 2010-March 2012). One treatment clinic is not included because of its insufficientsamplesize.Thiscliniccorrespondstooneofthetwothatdidnottakeuptreatment.Thetrianglesymbolreferstotheclinicthatwasassignedtotreatmentbutdidnottakeupthetreatment.Thex-axis issortedfromthelowesttothehighestclinic-specific impact.Thedashedblue line is the intent-to-treateffect calculatedbypooling the interventionand the firstpost interventionperiod.Theverticallinesare95%confidenceintervalsconstructedusingstandarderrorsobtainedfromtheWildbootstrapprocedure.

54

TABLEB1:ROBUSTNESSTESTSFORWEEKSPREGNANTAT1STPRENATALVISIT

(1) (2) (3)

InterventionPeriodPost-

InterventionPeriodI

Post-InterventionPeriodII

A.ResultsfromTable4

Treatment -1.47** -1.63** -2.47**

(0.71) (0.75) (1.02)

LargeSamplep-value 0.04 0.03 0.02WildBootstrappedp-value 0.08 0.03 0.03

ControlGroupMean 17.80 17.90 20.10SampleSize 769 1,296 710

B.EstimatesUsingRestrictedSample

Treatment -1.47* -2.01*** -2.01* (0.77) (0.70) (1.11)

LargeSamplep-value 0.06 0.00 0.07WildBootstrappedp-value 0.09 0.02 0.12

ControlGroupMean 17.96 18.32 17.01SampleSize 760 1,326 425

Notes:ThistablereportsLATEestimatesofthetreatmenteffectofthemodifiedfeescheduleonweekspregnantat1stprenatalvisit.Thep-valuesarefor2-sidedhypothesistestsofthenullthatthedifferenceisequaltozero.Wepresentboththep-valuecomputedforlargesamples and aWild bootstrappedp-value that is robust in sampleswith small numbers of clusters (Cameron et al. 2008). OurWildbootstrapprocedureassignssymmetricweightsandequalprobabilityafter re-samplingresiduals (DavidsonandFlachaire2008)anduses999replications.Column(1)reportstheresultsforthesampleobservedinan8-monthinterventionperiod(May2010–December2010).Column(2)reports theresults for thesampleobserved in the15-monthperiod following theendof the intervention(January2011–March2012).Column(3)reportstheresultsforthe9-monthperiodafterthechangeinthecodingofthefirstprenatalvisit(April2012–December2012).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

55

TABLEB2:ROBUSTNESSTESTSFOR1STPRENATALVISITBEFOREWEEK13 (1) (2) (3)

Intervention

PeriodPost-Intervention

PeriodI

Post-InterventionPeriodII

A.ResultsfromTable4

Treatment 0.11** 0.08** 0.08**

(0.04) (0.04) (0.04)

LargeSamplep-value 0.01 0.02 0.04WildBootstrappedp-value 0.03 0.05 0.06

ControlGroupMean 0.31 0.34 0.27SampleSize 769 1,296 710

B.EstimatesUsingRestrictedSample

Treatment 0.09** 0.10** 0.10* (0.04) (0.04) (0.06)

LargeSamplep-value 0.03 0.01 0.08WildBootstrappedp-value 0.08 0.02 0.11

ControlGroupMean 0.31 0.33 0.36SampleSize 760 1,326 425

Notes:ThistablereportsLATEestimatesofthetreatmenteffectofthemodifiedfeescheduleanindicatorofwhetherthe1stprenatalvisitoccurredbeforeweek13ofpregnancy.Thep-valuesarefor2-sidedhypothesistestsofthenullthatthedifferenceisequaltozero.Wepresentboththep-valuecomputedforlargesamplesandaWildbootstrappedp-valuethatisrobustinsampleswithsmallnumbersofclusters (Cameron et al. 2008). Our Wild bootstrap procedure assigns symmetric weights and equal probability after re-samplingresiduals (Davidson and Flachaire 2008) and uses 999 replications. Column (1) reports the results for the sample observed in an 8-monthinterventionperiod(May2010–December2010).Column(2)reportstheresultsforthesampleobservedinthe15-monthperiodfollowing the end of the intervention (January 2011 –March2012). Column (3) reports the results for the 9-monthperiod after thechangeincodingofthefirstprenatalvisit(April2012–December2012).Standarderrorsinparentheses.*p<0.10,**p<0.05,***p<0.01.

56

AppendixC:ITTResults

TABLEC1:ITTESTIMATESOFTHEEFFECTOFTEMPORARYINCENTIVESONTIMINGOF1STPRENATALVISIT (1) (2) (3)

InterventionPeriod

Post-InterventionPeriodI

Post-InterventionPeriodII

A.WeeksPregnantat1stPrenatalVisit

Treatment -1.39** -1.59** -2.47** (0.67) (0.73) (1.02)

LargeSamplep-value 0.04 0.03 0.02WildBootstrappedp-value 0.09 0.03 0.03

ControlGroupMean 17.80 17.90 20.10SampleSize 769 1,296 710

B.FirstPrenatalVisitBeforeWeek13ofPregnancy

Treatment 0.10*** 0.08** 0.08** (0.04) (0.04) (0.04)

LargeSamplep-value 0.01 0.02 0.04WildBootstrappedp-value 0.03 0.05 0.08

ControlGroupMean 0.31 0.34 0.27SampleSize 769 1,269 710

Notes:This table reports ITTestimatesof the treatmenteffectof themodified feescheduleon indicatorsof the timingof the1stprenatalvisit.TheLATEestimatesarereportedinTable4.ThedifferencesareestimatedfromOLSregressionsofthedependentvariableonanindicatorforclinictreatmentrandomassignment.Thep-valuesarefor2-sidedhypothesistestsofthenullthatthedifferenceisequaltozero.Wepresentboththep-valuecomputedforlargesamplesandaWildbootstrappedp-valuethatisrobustin sampleswithsmallnumbersof clusters (Cameronetal.2008).OurWildbootstrapprocedureassignssymmetricweightsandequal probability after re-sampling residuals (Davidson and Flachaire 2008) and uses 999 replications. Column (1) reports theresultsforthesampleobservedinan8-monthinterventionperiod(May2010–December2010).Column(2)reportstheresultsforthe sample observed in the 15-month period following the end of the intervention (January 2011 – March 2012). Column (3)reportstheresultsforthe9-monthperiodafterthechangeinthecodingofthefirstprenatalvisit(April2012–December2012).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

57

TableC2:ITTofCross-PriceEffects(Spillover) (1) (2) InterventionPeriod Post-InterventionPeriod

A.TetanusVaccine Treatment 0.02 -0.02

(0.07) (0.05)

LargeSamplep-value 0.76 0.62WildBootstrappedp-value 0.80 0.59

ControlGroupMean 0.79 0.84SampleSize 769 1,053

A.Numberofvisits

Treatment 0.37 0.50 (0.32) (0.57)

LargeSamplep-value 0.24 0.38WildBootstrappedp-value 0.27 0.40

ControlGroupMean 4.05 4.40SampleSize 769 1,053

Notes: This table reports ITT estimates of the treatment effect of themodified fee schedule onindicators of other services. The LATE estimates are reported in Table 5. The differences areestimated from OLS regressions of the dependent variable on an indicator for clinic treatmentrandomassignment.Thep-valuesarefor2-sidedhypothesistestsofthenullthatthedifferenceisequaltozero.Wepresentboththep-valuecomputedforlargesamplesandaWildbootstrappedp-value that is robust in sampleswith smallnumbersof clusters (Cameronet al. 2008).OurWildbootstrapprocedureassignssymmetricweightsandequalprobabilityafterre-samplingresiduals(DavidsonandFlachaire2008)anduses999replications.Column(1)reportstheresults forthesample observed in an 8-month intervention period (May2010 –December 2010). Column (3)reports the results for the sample observed in the 12-month period following the end of theintervention(January2011–December2011).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

58

TABLEC3:ITTEFFECTSOFINCENTIVESONBIRTHOUTCOMES (1) (2)

InterventionPeriod Post-InterventionPeriod

A.BirthWeight Treatment -34.88 24.48

(45.38) (39.63)

LargeSamplep-value 0.44 0.54WildBootstrappedp-value 0.46 0.57

ControlGroupMean 3304.82 3279.13SampleSize 555 802

B.LowBirthWeight Treatment 0.01 -0.01

(0.02) (0.01)LargeSamplep-value 0.63 0.60

WildBootstrappedp-value 0.61 0.63ControlGroupMean 0.05 0.06

SampleSize 555 802B.Premature

Treatment 0.03 -0.04* (0.03) (0.02)

LargeSamplep-value 0.31 0.08WildBootstrappedp-value 0.32 0.09

ControlGroupMean 0.09 0.12SampleSize 414 708

Notes: This table reports ITT estimates of the treatment effect of themodified fee schedule for on indicators of birthoutcomes.TheLATEestimatesarereportedinTable6.Theobservationsincludewomanforwhomweareabletoobtaininformation on birth outcomes provided in public hospital birth records. The differences are estimated from OLSregressionsof thedependentvariableonan indicator forclinic treatmentrandomassignment.Thep-valuesare for2-sidedhypothesis testsof thenull that thedifference isequal tozero.Wepresentboth thep-valuecomputed for largesamplesandaWildbootstrappedp-valuethatisrobustinsampleswithsmallnumbersofclusters(Cameronetal.2008).OurWildbootstrapprocedureassignssymmetricweightsandequalprobabilityafter re-sampling residuals (DavidsonandFlachaire2008)anduses999replications.Column(1)reports theresults for thesampleobserved inan8-monthinterventionperiod (May2010–December2010).Column (2) reports the results for the sampleobserved in the12-monthperiodfollowingtheendoftheintervention(January2011–December2011).Standarderrorsareinparentheses.*p<0.10,**p<0.05,***p<0.01.

59

AppendixD:SurveyofClinicMedicalDirectors

In collaboration with the Provincial Management Unit of the program (UGPS), weconducted a short of clinics that participated in the pilot. The survey aimed tomeasure theabsolute and relative importance of seven different prenatal care procedures includinginitiatingprenatalcarepriortoweek13ofpregnancy.Theabsolutescoresrangefrom1to5,with 5 being the highest score in terms of importance, and an additional option of zeroindicating that the procedure is not appropriate for a pregnantwoman. Hence, the absolutescorerangesfrom0to5points.Therelativerankingaimedtosortthesevenpracticesfrom1to7,with1beingthehighestranking.Inpracticehowever,thesurveyinstrumentallowedtherespondenttorepeatnumbers.

The surveywas sentout toby email to clinicsdirectors (or thenextperson in rank).Fifty-fivepercentoftheclinicsrespondedtothesurvey,whichreducesthesampleto20clinicsfromthe36clinicsconsideredinitiallyintheanalysis.AppendixTableD1showsthatthereareno significant differences in baseline characteristics between clinics that responded to thesurveyandclinicsthatdidnotrespond.Inaddition,weaccountforsurveynon-responseusingInverse Probability Weighting based on the logistic regression reported in Table D2(Wooldridge2007).WereportresultsforbothIPWandnon-IPWregressions.

Figures7donotsuggestanydifferenceintheabsolutescoreandrelativerankingoftheproceduresbetweentreatmentandcontrolclinics.Totestforthesignificanceofthedifferencesbetween the two groups, we run an OLS regression of the absolute score and the relativerankingagainstabinaryindicatorfortreatment.Toaccountforthesmallsamplesizewealsocompute the p-value for the differences in means permuting our data and using a randomsampleof10,000permutations.TheresultsareshowninTablesD3.

SurveyQuestionnaireWeaskforyourcollaborationincompletingabriefsurveyaboutprenatalcareservicesprovidedatyourhealthfacility.Important:Whenanswering the survey,please thinkofahypothetical caseofawomanwith thefollowingcharacteristics:

• 25yearsold• Livinginthesameneighborhoodwhereyourhealthfacilityislocated• Withoutanyapparentsignofdisease• 6weekspregnant• Hadapreviouslow-riskpregnancy

1. Pleaseassignascorebetween1to5toeachofthefollowingservicesthatcouldbedelivered

tothepregnantwomanpresentedinthehypotheticalcase.

60

1correspondstoaservicetowhichyouassignthelowestimportance5correspondstoaservicetowhichyouassignthehighestimportance

1 2 3 4 5

Notappropriatefor apregnantwoman

Prenatalultrasound

ThoraxX-Ray

Firstprenatalvisitbeforeweek13ofpregnancy

Bio-psycho-socialpregnancycounselingvisit

CombinedDiphtheria/Tetanusvaccine

Bloodtestwithserology

Bloodtestwithoutserology

2. Pleaserankinorderofpriority(from1to7)thefollowing7healthservicesthatcouldbedeliveredtothepregnantwomanofthehypotheticalcase.

1correspondstotheserviceyouwouldprioritizethemost7correspondstotheserviceyouwouldprioritizetheleast

Prenatalultrasound

ThoraxX-Ray

Firstprenatalvisitbeforeweek13ofpregnancy

Bio-psycho-socialpregnancycounselingvisit

CombinedDiphtheria/Tetanusvaccine

Bloodtestwithserology

Bloodtestwithoutserology

TABLED1:BASELINECHARACTERISTICSOFCLINICS,BYONLINESURVEYRESPONSESTATUS

Non-respondent Respondent P-value Obs.

PercentageinTreatmentGroup 0.38 .62 0.15 36

61

NumberofPregnantWomenAttendedperYear 48.60 54.90 0.33 36

WeeksPregnantat1stPrenatalVisit 17.04 16.77 0.15 36

1stVisitbeforeWeek13ofPregnancy 0.34 0.36 0.27 36

%ofPregnantWomenwhoarePlanNacerBeneficiaries 0.61 0.64 0.59 36

TetanusVaccineDuringPrenatalVisit 0.76 0.81 0.22 36

NumberofPrenatalVisits 4.26 4.42 0.72 36

BirthWeight(Grams) 3,283 3,320 0.33 36

GestationalAge(Weeks) 38.65 38.47 0.57 31

LowBirthWeight(<2500Grams) 0.06 0.07 0.73 31

Premature(GestationalAge<37Weeks) 0.10 0.12 0.60 31

Notes:ThistablereportsthemeansofbaselinecharacteristicsforclinicsthatrespondedtotheMay2015onlinesurveyandforclinicsthatdidnotrespond.Thecharacteristicsare taken fromthemedical records informationsystem(2009).Thep-values for the testsofdifferencesinmeansarecomputedusingpermutationteststhatarerobustforsmallsamplesizes.

TABLED2:PROBABILITYOFRESPONDINGTOTHEONLINESURVEY,

LOGITCOEFFICIENTSANDMARGINALEFFECTS

Coefficient Marg.Eff.

TreatmentGroup 1.498 0.274 (1.111) (0.180)

BirthWeight(grams) 0.100 0.018 (1.076) (0.196)

WeeksPregnantat1stPrenatalVisit -0.594 -0.109 (0.648) (0.121)

1stVisitbeforeWeek13ofPregnancy -3.590 -0.657 (9.026) (1.670)%ofPregnantWomenwhoarePlanNacerBeneficiaries 1.620 0.296 (4.359) (0.774)

TetanusVaccineDuringPrenatalVisit 3.350 0.613 (3.817) (0.646)

NumberofPrenatalVisits -0.099 -0.018 (0.559) (0.101)

Constant 7.644 (18.248) Observations 36 36

Notes: This table reports the coefficients andmarginal effects from a Logit regression that estimates theprobabilitythataclinicrespondedtotheMay2015onlinesurvey.

TABLED3:DIFFERENCESINABSOLUTESCOREANDRELATIVERANKINGOFEARLYPRENATALCARE

62

AbsoluteScore RelativeRanking

(1)OLS

(2)OLS-IPW

(3)OLS

(4)OLS-IPW

Difference(Treatment–Control) 0.20 0.13 0.10 0.14 (0.22) (0.92) (0.21) (0.89)

LargeSamplep-value 0.38 0.89 0.65 0.88Permutationp-value 0.35 1.00 0.46 0.99

Observations 20 20 20 20Controlgroupmean 4.57 1.88 4.66 1.88

Notes:Column(1)showsthedifferencesbetweentreatmentandcontrolclinicsintheabsolutescoreassignedtothepracticeofearlyprenatalcarewithoutanyadjustmentofsampleloss.Column(2)adjustsforsamplelossbyInverse ProbabilityWeighting. Column (3) shows the differences between treatment and control clinics in therelativerankingassignedtoearlyprenatalcareamongsevendifferentpractices.Column(4)isthesameasColumn(3)butadjustsforsamplelossbyInverseProbabilityWeighting.(Wooldridge2007)ThecoefficientsareobtainedfromanOLSregressionofeachoutcomeagainsta treatmentbinary indicator.The thirdrowshows theP-valueobtained from permuting the data using a random sample of 10,000 permutations. Standard errors are inparentheses.Weloseoneobservationineachcasebecauseofmissingdataineachspecificquestion.