PoliticalContext,OrganizationalMissionandtheQualityofSocialServices:InsightsfromtheHealthSectorinLebanon

MelaniCammett(HarvardUniversity)

and

AytuğŞaşmaz(HarvardUniversity)

Abstract

Inmanydevelopingcountries,non-stateactorsareimportantprovidersofsocialwelfare.In

partsoftheMiddleEast,SouthAsiaandotherregions,religiouscharitiesandpartiesand

NGOshavetakenonthisrole,withsomeprecedingindependentstatehoodandothers

buildingparalleloralternativewelfareinfrastructurealongsidethemodernstate.Howwell

dothesegroupsprovidewelfaregoods?Dosomeexhibita“welfareadvantage,”ora

demonstratedsuperiorityinthequalityandefficiencyofprovidingsocialservices?Inthis

paper,weexplorewhetherdistinctorganizationaltypesareassociatedwithdifferentlevelsof

thequalityofcare.BasedonastudyinGreaterBeirut,Lebanon,wherediversetypesof

providersoperatehealthcenters,weproposeandtestsomehypothesesaboutwhycertain

organizationsmightdeliverbetterservices.Wefindlittleempiricalsupportforafaith-based

welfareadvantage,assomeresearchcontends.Instead,thedataindicatethatsecularNGOs

exhibitsuperiormeasuresofhealthcarequality,aseeminglycounterintuitivefindingin

Lebanonwherereligiousandsectarianactorsdominatepoliticsandthewelfareregimeand

commandthemostextensiveresources.Ourpreliminaryexplanationforthisfinding

emphasizesthewaysinwhichthesociopoliticalcontextshapesthechoicesofqualified

providerstoselectintosecularorganizationsandwhycitizensmightperceivetheseproviders

tobebetter,irrespectiveoftheactualqualityofservicesdelivered.

1

Introduction

Inmanydevelopingcountries,non-stateactorsareimportantprovidersofsocialwelfare,with

someprecedingindependentstatehoodandothersbuildingparalleloralternativewelfare

infrastructurealongsidethemodernstate.Awidearrayofactors,includingNGOs,religious

charitiesandevenpoliticalparties,areinthebusinessofprovidinghealthservices,schooling,

vocationaltrainingandotherimportantservices,andthusgreatlyaffectthestandardsof

livingandwell-beingoflowandmiddleincomepeople(CammettandMacLean,2014).Yet

littleresearchexploresthequalityofwelfaregoodssuppliedbyNSPs.Docertaintypesexhibit

a“welfareadvantage,”orademonstratedsuperiorityinthequalityofsocialservice

provision?

Inthisarticle,weproposeandassessavarietyofhypothesesrelatedtoorganizational

typeandthequalityofservicesanddevelopsomepropositionsabouttheeffectsof

organizationalmissiononservicedelivery.Basedonevidencefromanoriginalsetofsurveys

inprimaryhealthcentersaffiliatedwithdiversepublicandnon-stateactorsinGreaterBeirut,

Lebanon,weshowthatsecularNGOsdemonstrateanapparentwelfareadvantageoverother

providertypesinbothobjectiveandsubjectivemeasuresofhealthquality.1Further,patient

evaluationsofhealthcentersrunbydistinctorganizationsaredrivenlargelybyperceptions

ofdoctors,anddoctorswhoworkinsecularorganizationsreporthigherlevelsofsatisfaction

withtheorganizationswheretheywork.Thisapparentsecularwelfareadvantagecontradicts

manytheoreticalandempiricalexpectations,aswedetailbelow.Ourproposedexplanation

forthisresultcentersonthewaysinwhichthepoliticalcontextaffectsboththeobjectiveand

subjectivequalityofcarebysecular,religiousandpoliticalgroupsthroughbothsupplyand

1Whilewerecognizethattheterm“secular”iscontestedandhasmultiplemeanings(Asad,2003),hereweuse

thetermtorefertoorganizationsthatarenotconnectedtoanyreligiousgrouporcommunityandarenotlinked

topoliticalparties,religiousorotherwise.IntheLebanesecontext,secularorganizationsoftenexplicitly

distinguishthemselvesfromreligiousandsectariangroupsandideologies.

2

demandprocesses.Inapolitystructuredexplicitlyalongreligiouslines,beinganavowed

secularistgoesagainstdominantsocialandpoliticaltrendsandoffersfewifanymaterial

rewards.Asaresult,secularNGOsthatprovidehealthservicesmayattractdoctorswhoare

notincorporatedinpatronagenetworksassociatedwithmorepoliticallyconnectedreligious

andsectarianorganizationsand,therefore,maybemoremotivatedbycharitable

considerationsoracommitmenttoprofessionalism.Second,widespreadcitizen

dissatisfactionwithreligiousandsectarianorganizations,2whichareoftenviewedascorrupt

andself-serving,mayresultininferiorevaluationsofwelfareprogramsrunbysuchgroups

and,conversely,expressmorefavorableassessmentsofservicesprovidedbyorganizations

thatexplicitlydissociatethemselvesfrompoliticalsectarianism.

Inthenextsection,wejustifyourfocusonthehealthsector,presenta

multidimensionaldefinitionof“quality”inprimaryhealthcare,andreviewargumentsabout

whysometypesofprovidersmaybeespeciallyadeptatprovidinghealthcareandothertypes

ofsocialservices.Thethirdsectionofthepaperprovidesessentialbackgroundinformation

onLebanonandonthetypesoforganizationsinquestionanddescribesthedataandkey

variablesusedintheanalyses.Sectionfourpresentsdescriptiveandstatisticalanalyses

followedbyadiscussionoftheimplicationsofthefindingsfortherelationshipsbetween

organizationalmission,politicalcontextandthequalityofservicedelivery.Intheconclusion,

wesummarizethefindingsandsuggestabroaderresearchagendaonpoliticalcontext,

organizationalmissionandthequalityofservicedelivery.

2Werefertoorganizationsas“religious”whentheyhavenoformallinkagetopoliticalpartiesormovementsand

“sectarian”whentheyareexplicitlylinkedtoapoliticalpartywithlinkstoaparticularreligiouscommunity.

3

Politics,healthanddimensionsofhealthcarequality

Thehealthsectorisanappropriatearenaforexaminingwhetherdifferenttypesof

organizationsexhibitawelfareadvantagebecausemanyNSPsareinvolvedinthedeliveryof

medicalservicesandaccesstohealthcareisimportanttowell-being(Cammett, 2014; Thachil,

2014).Furthermore,foravarietyofreasons,socialscientists–andnotjustpublichealthand

medicalspecialists–shouldbeconcernedwiththepoliticsofhealth.First,accesstohealth

careisimportanttowell-being.Asaresult,peoplemayfeelindebtedtoinstitutionsthat

provideormediateaccesstomedicalservicesand,cognizantofthesepotentialpayoffs,

politicalorganizationsfaceincentivestodeliverorclaimcreditfortheprovisionofhealth

care.Thehealthsystemisalsoacriticallocusofcitizeninteractionswithgovernments,which

playanimportantroleinthefinancingandprovisionofhealthcareinmiddle-income

countries(Rockers,KrukandLaugesen,2012)andwithnon-stateproviders,whichareeither

wellestablishedorincreasinglyimportantinwelfareregimesindevelopingcountries

(CammettandMacLean,2014;GoughandWood,2006).Insocietieswithpoliticized

ethnoreligiousidentities,asinLebanon,theprovisionofbasicservicesalsohelpstoconstitute

asenseofgroupmembershipbyestablishingboundariesofinclusionandexclusionin

politicalcommunities(Cammett 2014, 2, 13-14).Thus,theprovisionofhealthcarecan

intersectwithpoliticsinbothdirectandindirectways.

Measuringhealthcarequality

Intheliteratureonhealthpolicyandmanagement,itiswidelyacceptedthatquality

encompassesmultipledimensions,includingobjectiveandsubjectivemeasuresaswellas

technicalandnon-technicalfactors.Inbroadterms,healthcarequalityincludesthree

componentsrelatedtothestructure,processandoutcomeofthedeliveryofhealthservices,

respectively(Donabedian,1988;Klassenetal.,2010).Thestructuraldimensionofquality

referstotheenvironmentinwhichhealthcareisprovided,orthematerialandhuman

4

resourcesandcharacteristicsofthefacilitywhereservicesaredeliveredaswellasthe

organizationofthedeliveryofmedicalservices.Thisincludestheavailabilityandconditionof

medicalequipmentandtrainedmedicalstaff,medicationsandrelevantinfrastructureaswell

asthewaysinwhichphysicalandhumanresourcesaremanagedupanddownthesupply

chaininthedeliveryofcare.Theprocess-orientedcomponentofqualityaddressesthe

methodbywhich healthcareisprovided,focusinginparticularonthewaysinwhich

providersinteractwithpatientsaswellasprovidercapabilitiesandeffort.Processmeasures

assessdoctorknowledgeandtrainingaswellasthedegreetowhichtheyapplythis

knowledgetodeliverappropriatecaretopatientsinatimelyandrespectfulmanner.Finally,

outcomesdenotetheresultsofhealthcare,notablythehealthstatusofpatientsandpatient

satisfaction,amongotherfactors(StelfoxandStraus,2013;Tuanetal.,2005).

Twopointsrelatedtotheconceptualizationandmeasurementofhealthcarequality

shouldbeemphasizedandguideourchoiceofindicators.First,healthoutcomesresultfroma

varietyoffactorsaboveandbeyondthedeliveryofservices(MarmotandWilkinson,2004),

complicatingeffortstolinkthemdefinitivelytotheprovisionofmedicalcare.Asaresult,our

analysesdonotaimtoexplainhealthoutcomes.Second,publichealthresearchshowsthatthe

processdimensionsoutweighthestructuralaspectsofqualityinaffectinghealthoutcomes

(DasandHammer,2014).Adoctorwhoiswell-trained,regularlyshowsuptowork,and

practicesmedicineattheir“knowledgefrontier”hasagreaterimpactonpatienthealththan

themereavailabilityofmedicalsuppliesandnewmachines.Withoutcapableandcommitted

professionalstaff,state-of-the-artmedicalequipmenthaslittleeffectonpatienthealth.

Likewise,patientsaremorelikelytoreportmorefavorableviewsoftheirserviceproviders

whentheyseemcompetent,engagedandattentive,evenwhenthefacilityinwhichthecareis

providedislessattractiveandlesswellappointed.Thus,whileweaccountforthestructural

dimensionsofqualityinouranalyses,wefocusmostcentrallyonprocessquality.

Furthermore,mostofstatisticalanalysesaimtoexplainsubjectivemeasuresofquality,

5

notablypatientsatisfaction,becauseperceptionsofperformanceratherthanobjective

measuresofqualityaremoregermanetocitizenevaluationsofprovidersand,therefore,are

likelytohaveamoredirectimpactonpoliticalattitudesandpreferences(Cammett,Lynch

andBilev,2015;ChristensenandLægreid,2005).Indeed,ourhypotheses,whichhighlightthe

reasonswhycompetentdoctorsselectintosomeproviderorganizationandwhysome

patientsreportmorefavorableviewsofsomeprovidertypes,aremoredirectlyrelevantto

theprocess-orienteddimensionsofmedicalcare

OrganizationalMissionandtheQualityofServiceDelivery

Distinctsocialscienceapproaches,whichwereviewbrieflybelow,eitherdirectlyorindirectly

suggestthatdifferenttypesoforganizationsarelikelytoexhibitawelfareadvantage(or

disadvantage).

Faith-BasedOrganizationsandCharitableMotivations

Asubstantialliteratureonfaith-basedorganizations(FBOs)holdsthatthecharitable

dimensionsofreligionmotivatethepioustovolunteerorworkforminimalcompensationto

dosocialgood(ClarkeandJennings,2008;Cnaan,2002;DeHavenetal.,2004;Unruhand

Sider,2005;Wuthnow,2004).Theseapproachesholdthatreligiousorganizationstendto

attractpersonnelwhoarecommittedtotheirmissionsonspiritualgrounds,makingthem

willingtoputinlonghours,oftenforrelativelyminimalcompensation.Inaddition,staff

membersandvolunteersinreligiouscharitiesmaychoosetoserveothersasawaytoensure

thesurvivalofthecongregationthroughincome-generatingactivitiesorinordertofoster

acceptanceofthereligiousgroupinthecommunitywhereitisbased.Socialserviceprovision

mayalsoaidinproselytism,apotentiallypowerfulincentivefortheleadershipandstaffof

religiousorganizationstoofferhighqualityservicesandonethatisrelativelyuniqueto

6

religiousgroups.ArecentspecialissueoftheLancetonreligionandhealthcareechoessome

oftheseclaims(Karametal.,2015;SummerkillandHorton2015).3

TheEconomicsofReligionand“Strict”Churches

Theliteratureontheeconomicsofreligionpointstoarelatedyetdistinctreasonwhyatleast

someFBOsmaydeliversuperiorwelfareservices.“Strictchurches”(Iannaccone1994)or

religiousgroupsthatrequiremajorsacrificesfromtheirmembersandcallonadherentsto

visiblydistinguishanddistancethemselvesfromtherestofsociety,exhibithigherratesof

volunteerismandattractmoredevotedpersonnelthanothers.4Thehighlevelsof

commitmentoftheirmembersenablessuchgroupstoweedoutlesscommittedindividuals,

therebyovercomingthefreeriderproblemsthatplaguemostorganizations,includingless

stringentFBOs.Theselectioneffectsatthecoreofthisapproachinturnmayaffectthequality

ofservicesbyincentivizingstafftodevotemoreefforttotheirworkforlittleorno

compensation.

Organizationalstrictnessmaybeassociatedwithhigherlevelsofsubjectiveand

objectivequality.Ontheonehand,organizationsthatexpectbigsacrificesonthepartoftheir

membersmayattractespeciallycommittedprofessionals,whoarelikelytoworktotheir

“knowledgefrontier”(Das,HammerandLeonard,2008),leadingtohigherlevelsof

objectivelymeasuredqualityofservicedelivery.Ontheotherhand,beneficiariesand

communitymembersmayperceivethatstaffmembersatfacilitiesrunbystrictgroupsare

morelikelytobeself-sacrificing,toworkespeciallyhard,andtoremaincommittedtotheir

cause,leadingtohighersubjectivemeasuresofservicequality.

3ReinnikaandSvensson(2010)provideevidenceofafaith-basedwelfareadvantageintheirstudyofreligious

non-profitorganizationsinUganda.Astheynote,“Thesefindingsareconsistentwiththeviewthatreligious

nonprofitprovidersareintrinsicallymotivatedtoserve(poor)people—workingforGodseemstomatter!”4AprimeexampleinIannaccone’swork(1994)istheChurchoftheLatterDaySaints.

7

EthnoreligiousPartiesandPoliticalIncentives

Whenadaptedtothepoliticalarena,similarlogicsmayapplytoethnicorsectarianparties,

whichcombinecommunalandpoliticalmessages.Attheindividuallevel,identity-based

partieswithaffiliatedsocialservicewings,suchasHezbollahinLebanon,HamasinPalestine,

theBharatiyaJanataPartyinIndiaandotherethnicandreligiousparties,mayattract

volunteersandstaffmemberswhoarewillingtoputinlonghoursatparty-linkedinstitutions,

whetherbecauseofgenuinecommitmenttothecause,integrationinpartypatronage

networksorboth.Attheorganizationallevel,thedrivetowinvotesortogalvanizenon-

electoralmobilizationconstitutesastrongincentiveforpoliticalgroupstoofferhigh-quality

services(Cammett,2014;Thachil,2014).Ethnicandsectarianpartiesmaythereforefacehigh

incentivestoofferattractiveandwell-runsocialprograms.

However,ifsectarianpartiesoperateaccordingtoaclientelistlogicratherthanan

ideologicalvision,thenstaffmembersatparty-linkedinstitutionsmaybelessinclinedto

makepersonalsacrificesonbehalfofparty.Incomparisonwithmoreintrinsicmotivations,

suchextrinsicincentivespotentiallyreducethedrivetoprovidehighqualityservices.

Similarly,communitymembersmayviewthewelfareagencieslinkedtocorrupt,patronage-

basedpartieswithcynicism,reducingsubjectiveevaluationsofthequalityofservicesoffered

bysuchinstitutions.

ThesedistinctapproachessuggestthatFBOsmaydeliversuperiorsocialservicesthan

othertypesofproviders,whetherbecausetheirreligiousmissionsincentivizestaffmembers

toprovidehighqualitycharitableservicesorbecausetheyattractespeciallycommitted

personneland,therefore,moreeffectivelyovercomefreeriderproblemsplaguingother

organizations.Someevidencealsosuggeststhatsectarianparties–particularlythosethat

emphasizeastrongideologicalmission–mayproviderelativelyhighqualityservicesunder

someconditions.Furthermore,thesacrificesthatstaffmembersmakebyworkingat

8

charitableorganizationsratherthanfor-profitinstitutionsmayalsogarnerhighersubjective

measuresofquality.

Apublicsectorwelfaredisadvantage?

Muchdevelopmentresearchfocusesontheroleofthepublicsectorinservicedelivery,

particularlyinthecontextofthegovernmentfiscalcrisesindevelopingcountriesandthe

emphasisontheprivatesectorandpublic-privatepartnershipsindevelopmentpolicysince

the1980s(CITES).Indeed,somestudiesoftheprovisionofservicesbyFBOsandidentity-

basedpartiesbenchmarkservicedeliverybytheseorganizationsagainstthatofstateagencies

(CITES;LANCET2015,ETC.).Otherworkcomparestheextentandqualityofservices

providedbygovernmentinstitutionswiththoseofthefor-profitprivatesector,whichisthe

fastestgrowingproviderofbasicservicesinmanydevelopingcountries(CITES).Whilealarge

bodyofworkexaminestheconditionsunderwhichstateagenciesprovidebetterservices

(WorldBank,2004;CITES),anoverarchingthemeisthatthepublicsectorfacesconstraintsin

effectiveservicedelivery.Theextenttowhichthisistrueisanempiricalquestionthatis

contingentonspecificsociopoliticalandeconomicconditionsandmayvarydependingonthe

typeofserviceinquestionanddimensionofquality,asourresultssuggest.

Inthenextsection,wedescribethesampleanddatausedtoassesswhethercertain

providertypesinLebanonexhibitawelfareadvantage,whethermeasuredinobjectiveor

subjectiveterms.

SampleDesignandDataCollection

Lebanonisanappropriatesiteforthisresearchbecauseabroadrangeofprimaryhealthcare

providersandnon-stateactorsoperateinthewelfareregimeandmostarewellestablished.

TheLebanesegovernmentisbasedonapower-sharingarrangement,whichenshrines

religioninthepoliticalsystemandstipulatesthatgovernmentpostsareallocatedbysect

9

accordingtoapre-establishedformula,effectivelyleadingtothedistributionofpublic

resourcesalongsectarianlines(SaltiandChaaban,2010).

Thesectarianpower-sharingsysteminLebanonhasshapedthepost-independence

welfareregime,whichinvolvesminimalstateinterventionandreliesheavilyonprivate,non-

stateactors,includingreligiouscharities,sectarianpartiesandNGOs.Asaresult,the

Lebanesecaseismostdirectlycomparabletocontextswithpoliticizedethnicorreligious

cleavages,aphenomenonthatisincreasinglycommonintheMiddleEastandSouthAsia,

amongotherplaces.However,theLebaneseexperienceofferspertinentlessonsforMiddle

Easternandotherdevelopingcountriesinthecontemporaryperiod,whenpublicwelfare

infrastructureisdeclining,non-stateprovisionisontheriseandsystemsbasedonhybrid

governancemodelsarepromotedbydevelopmentpolicies(CITES).

Inthehealthsector,thestateplaysaminimalroleintheactualdeliveryofhealth

servicesbutprovidesextensivefinancingfornon-stateproviders.Themajorsectarianparties

andmovementsholdgreatswayinpublicinstitutionsthroughthesectarianpower-sharing

system,perpetuatingweakstatecapacityandeffectivelyinhibitingreform.Asaresult,state

effortstobuildamorerobustpublicwelfareinfrastructureandtoexertmoreregulatory

controloverprivateandnon-stateactorsinthewelfareregimehasmetstiffresistance,

althoughtheMinistryofPublicHealthhasincreaseditsstewardshipofthehealthsectorin

recentyears.Inthissystem,stateagenciesandsocialprogramsarelucrativesourcesof

patronageforparties,politicalmovements,andlocalpoliticians,creatingentrenchedinterests

inthestatusquo(Cammett2014,ch.2).

AlthoughthemajorityofhealthcareprovidersinLebanonworkinthefor-profit

privatesector,thecharitablesector,whichcaterstopoorandlowermiddleclasspeople,isa

vitalandgrowingcomponentofthehealthsystemandisanimportantpartnerinthe

LebaneseMinistryofPublicHealth’s(MOPH)plantoofferuniversalcoveragetothe

population.TheMOPHnetworkofhealthcenters,whichisthefocusofthispaper,features

10

bothpublicsectorandnon-stateproviders.Inexchangeforprovidingheavilysubsidized

medicalservices,theMOPHprovidesnon-financialresourcesandaccesstofreeorheavily

subsidizedmedicationstocentersthatmeetminimumstandards.Religiouscharitiesand

sectarianpoliticalpartiesrunabouttwo-thirdsofprimaryhealthfacilitiesinthenetwork.Of

theremainingone-thirdofcharitablecenters,about60percentarerunbyseculargroups

(Cammett 2014, 53-54).5Invirtuallyallcharitablehealthcenters,doctorsworkonapart-time

basis,earningastandard,minimalfeecalculatedonaperpatientbasis,whiledevotingmost

oftheirtimetotheirownorotherprivate,for-profitpractices.Asaresult,thereislimited

variationintherateandstructureofcompensationfordoctorsworkinginfacilitiesrunby

differenttypesofproviders.

Sample

Thesampledesignforthepilotstudyfollowedthefollowingprocedures.First,allcentersin

thesamplearepartoftheMOPHcharitablenetwork.Second,allfacilitiesinthesample

operateonanot-for-profitbasisandprimarilyservepoorandlow-incomefamilies.Third,the

sampledfacilitiesaredrawnfromtheuniverseofcenterslocatedinGreaterBeirut,which

containsthehighestpopulationconcentrationinthecountryandfeatureshealthcentersrun

byallprovidertypes.ItisalsoimportanttonotethatmostcentersintheMOPHnetworkare

runbyaparentorganizationsuchasareligiouscharity,politicalpartyorNGO,whichhas

multiplefacilitiesacrossthecountry.GiventhatallsampledcenterswereintheMOPH

networkandarelocatedinthecapital,weexpectthesampletobesomewhatbiasedtowards

higherqualityservices.

Thedatacollectionteamwasabletocollectrelativelycompletedataon27ofthe36

centerslocatedinGreaterBeirutintheMOPHprimaryhealthcarenetwork.Table1

summarizesthedistributionofPHCsinthesampleacrossdifferenttypesofprovider

5Thesedataarefrom2008.

11

organizations,thekeyvariableofinterestinthispaper,andsamplesizesforeachdata

collectioninstrument.

12

Table1:Institutionaltypesofprimaryhealthcentersinthesample

Typeoffacility Numberoffacilitiesinthestudy

Samplesizeofchiefmedicalofficersurvey

Samplesizeofdirectobservations

Samplesizeofpatientexitinterviews

Samplesizeofmedicalvignettes/doctorsurveys

Publicinstitutions 4 4 15 16 5SecularNGOs 5 5 15 15 5Religiouscharities 11 11 63 64 20Politicalcharities 7 7 42 42 13Total 27 27 135 137 43

DataCollectionProcedures

Thedatacollectionforthisstudyentailedthedesignandimplementationofmultipleoriginal

surveys.6Cammetttrainedateamofenumeratorswhothencarriedoutthefollowingsurveys

intheselectedhealthcarefacilities:(1)surveyinterviewswiththechiefmedicalofficerand

medicalstafftoobtaininformationontheservicesandinfrastructureavailableatthefacility

andonmanagementandtrainingprocedures,amongotherissues;(2)directobservationof

clinicalexaminations;(3)exitinterviewswithpatientsattheselectedfacilities;and(4)

medicalvignettesadministeredtogeneralpractitionersateachfacilitytoassesstheirmedical

knowledgeandadvice.Severalmonthsafterdatacollectionwascomplete,Cammettthen

conductedin-depthinterviewswiththedirectorsofthehealthnetworksrepresentedinthe

sample.

Thechiefmedicalofficersurveyprovidescrucialbaselineinformationoneachhealth

center.Thequestionnairegathersdataonthenumber,educationalbackground,experience

andcompensationstructureofeachemployeeaswellastheoperatingbudgetofthefacility;7

theaveragepatientloadduringthepastyearandepidemiologicalprofilesofthepatients;

6SeveraloftheinstrumentswereadaptedfromtheworkofJishnuDasandhiscollaborators(Das,2011;Das,

HammerandLeonard,2008).7Mostintervieweesdeclinedtoprovideinformationonthefinancesandbudgetsoftheirrespectivecentersin

thesurvey,however,follow-upinterviewswiththeheadsofhealthnetworkssuccessfullygathereddataonstaff

compensationratesformanycentersinthesample.

13

availableinfrastructureatthefacilityrelatedtotheworkenvironmentandtomedical

proceduresandexaminations;andinternalproceduresformonitoringtheperformanceof

doctorsandnursesatthefacilityand,moregenerally,forhumanresourcemanagement.The

chiefmedicalofficersurveythereforeprovidesdataoninfrastructuralqualityandonsome

dimensionsofprocessqualityatthefacilitylevel.

Asecondmethodofdatacollectionprovidesinformationonthenatureofinteractions

betweendoctorsandpatientsbasedondirectobservationbythetrainedenumeratorsof

clinicalexaminations.Thedatacollectedincludeinformationaboutthepatient,suchasher

symptoms,age,gender;informationaboutthedoctor’sinteractionswiththepatient,notably

thenumberofquestionsaskedbythedoctorandthetypesofexaminationsandtreatments

given;andthepriceschargedfortheservicesrendered.Thesedataproviderelatively

objectiveinformationonthenatureofdoctorattentivenesstothepatient.Althoughthe

findingsaresubjecttoHawthorneeffects,8thissourceofbiasmaydeclinewiththetimespent

observing(LeonardandMasatu,2006).Furthermore,thebiasduetoHawthorneeffects

shouldbeconsistentacrossallcenters,enablingcomparativeanalysesofthedatacollected.

Third,patientexitsurveyswerecarriedoutatthehealthcenterstoassesspatient

perceptionsofthecaretheyhavereceived.Thesurveycollectsbasicinformationonpatient

characteristicssuchaseducation,wealthandage;self-reportedhealthstatus;aspectsofthe

doctor-patientinteraction;andpatientsatisfaction.Theseresponsesprovideasubjective

measureofthequalityofcarebydiversetypesofproviders.

Afinalsurveyentailedtheadministrationofmedicalvignettestodoctorsatthehealth

centersinordertoassesstheirmedicalknowledge.Twotrainedresearchersconductedthe

interviewwiththedoctor,withoneservingasa“patient”andtheotherasthe“recorder.”

8Hawthorneeffectsrefertothetendencyofintervieweesorthesubjectsofastudytoimprovetheirbehavioror

productivitywhentheyareconsciousofbeingobserved.

14

Theypresentedfourcasesofhealthconditionsorillnesses,whichwereadaptedtothe

Lebaneseepidemiologicalprofile,experiencedbydistincthypotheticalpatientswhovaryby

ageandgender.9Eachvignettebeganwiththepatientpresentinghersymptomsandthe

recorderinvitingthedoctortoproceedexactlyasshewouldforanormalpatient.Inresponse

toeveryhistoryquestion,thepatientprovidedastandardizedresponsethatwascarefully

rehearsedinadvance.Similarly,anyphysicalexaminationrequestedbythedoctorwas

followedbyastandardizedanswerofferedbytherecorder.Afterthedoctorgavethe

diagnosisandtreatmentplan,thepairofenumeratorsadministeredthenexthypothetical

case.Theinformationgatheredfromclinicianresponsesisusedtoconstructanindexof

medicalknowledgeandadviceofthemedicalstafffromdifferenttypesofproviders,

generatingarelativelyobjectivemeasureofprocessqualityand,morespecifically,ofdoctor

competence.

Finally,Cammettcarriedoutin-depthinterviewswithMOPHofficialsandthedirectors

ofthehealthcentersandnetworksincludedinthesample.Theseinterviewsgathered

informationonthehistoryofthehealthprogramsrunbydifferentinstitutions;the

organizationalmissionsoftheparentnetworks;staffselection,trainingandmanagement

procedures;thefinancesandbudgetsofthehealthnetworksandindividualfacilities;and

otherrelevantinformation.Thedatafromtheseinterviewsfillinsomegapsinthesurvey

data,particularlyrelatedtofinancesanddoctorcompensationschemesandtotheroleof

organizationalmissioninshapingthehealthprogramsofdiversenon-stateinstitutional

networks.

9BecausesectissopoliticizedinLebanon,thenamesofthehypotheticalpatientsweredeliberatelychosentobe

neutralwithrespecttoreligiousidentity.Forexample,namesthattendtobeusedintheShi’acommunity,such

asHussein,orintheChristiancommunity,suchasTony,werepurposefullyavoided.

15

DescriptiveAnalyses:IndicatorsofHealthCareQualityandVariationacrossProviderTypesAsexplainedabove,healthcarequalityincludesthreecomponentsrelatedtothestructure,

processandoutcomeofthedeliveryofhealthservices.Table2providessummarystatistics

fortheselectedmeasuresofquality,andmeansandstandarddeviationsforeachprovider

type.

14

Table2:Summarystatisticsofselectedqualityindicators

Qualityindicator Source N Mean St.Dev.

Min Max Publicmean(St.Dev.)

SecularNGOmean

(St.Dev.)

Religiouscharities

mean(St.Dev.)

Politicalcharities

mean(St.Dev.)

Workplaceequipment Chiefmedicalofficersurvey 27 0.943 0.091 0.636 1 0.89(0.17)

0.95(0.05)

0.94(0.08)

0.97(0.07)

Healthequipment Chiefmedicalofficersurvey 27 0.772 0.159 0.286 0.929 0.79(0.13)

0.81(0.1)

0.78(0.16)

0.72(0.22)

Organizationalmonitoring Chiefmedicalofficersurvey 27 1.667 1.177 0 3 1.75(1.5)

1.6(1.14)

1.82(1.17)

1.43(1.27)

Goodgovernance Chiefmedicalofficersurvey 27 0.578 0.279 0.056 1 0.51(0.4)

0.5(0.21)

0.64(0.29)

0.58(0.28)

Numberofphysicalexaminationsbydoctor

Directobservation 135 2.733 1.565 0 6 2.8(1.42)

2.87(1.41)

2.62(1.66)

2.83(1.56)

Doctormedicalknowledge Medicalvignettes 45 1.211 0.727 0 4 1(0)

2.3(1.1)

1.18(0.47)

1.12(0.65)

PatientsatisfactionwiththePHC

Patientexitsurvey 134 3.761 0.685 3 5 3.67(0.62)

4.27(0.7)

3.68(0.59)

3.73(0.78)

Patientsatisfactionwiththedoctor

Patientexitsurvey 134 3.791 0.684 3 5 3.67(0.49)

4.4(0.74)

3.7(0.61)

3.76(0.73)

15

Thefirstmeasure,“workplaceequipment”,whichrelatestoinfrastructuralquality,

capturestheavailabilityofmaterialsandequipmentessentialtorunacleanandfunctional

workingenvironmentforthedeliveryofprimaryhealthservices.Thisvariableisacomposite

indexbasedonachecklistofitemsavailableintheclinic.Thesecondvariable,“health

equipment,”isalsoacompositeindexmeasuringtheavailabilityofmaterialandequipment

usedinmedicaldiagnosesandtreatment.10

ThemeansinTable2indicatethattheavailabilityofinfrastructure,whetherrelatedto

theadministrativefunctioningofthecentersortomedicalequipment,isroughlysimilar

acrossalltypesofnon-profitproviders.Theaveragescoresfortheavailabilityofmedical

equipmentaresomewhatlowerbutalsorelativelyhigh,andthevaluesdonotvarywidely

acrossthedifferenttypesofhealthnetworks.T-testscomparingthemeanlevelsofthese

variablesindicatethatmeasuresofinfrastructuralqualitydonotdiffersignificantlyacrossall

providertypes.ThisisnotsurprisingthatmembershipintheMOPHprimarycarenetwork

requiresthatfacilitiesmeetbaselinestandardsfortheavailabilityandmaintenanceof

equipmentandsupplies.

Twocompositeindicatorsmeasuregovernanceatthelevelofthefacilitybasedon

questionsinthechiefmedicalofficersurvey.Thefirst,“organizationalmonitoring,”isan

indextogaugeoversightpoliciesandpracticeswithinthenetworkandfacilityitself.The

variableisanadditiveindextoassesswhethertheadministrationemploysoneormore

methodsofmonitoringthehealthcenter,includingvisitsbyrepresentativesfromtheparent

organization,theimplementationofpersonnelsurveystoobtainfeedbackonstaffconcerns,

andthefieldingofpatientsatisfactionsurveys.Asecondindicator,“goodgovernance,”isa

morecomprehensiveindexoffacility-levelsupervisionandmanagementandincludes

variablesrelatedtoexternalmonitoringbytheMOPHandinternaloversightbythe

10SeetheSupplementalOnlineAppendixPartA1foritemsincludedintheconstructionofthesetwoindicators.

16

administrationofthefacility.Theindexisbasedonanaverageofsixindicators,eachofwhich

rangesfrom0to1,includingregularvisitsbygovernmenthealthinspectorstothefacility,

regularvisitsbytheparentorganizationtoinspectthefacility,theadministrationofpatient

satisfactionsurveys,thecollectionofstaffsurveys,regularstaffmeetings,and

institutionalizedchannelsofcommunicationbetweenstaffmembersandthemanagementof

thecenter.

AsseeninTable2,acrossthefourtypesofproviders,nomajordifferencesareevident

intheextenttowhichorganizationsmonitortheirfacilitiesorpromotefeedbackanddialogue

withstaffandpatients.T-testscomparingthemeanlevelsofthesevariablesconfirmthat

levelsofinternalmonitoringandgovernancedonotdiffersignificantlyacrossallprovider

types.Again,thislackofvariationmayreflecttheneedtocomplywithasetofbasic

managementpracticesinordertomeettheconditionsformembershipintheMOPHprimary

healthcarenetwork.Giventherealandperceiveddeficiencyofpublicserviceprovisionin

academicresearch(CITES)andintheLebanesecontext[ARABBAROMETER/WVSSURVEY

DATA],thelackofvariationincertaindimensionsofhealthcarequalityacrossprovidertypes

–includingthepublicsector-isanimportantandcounterintuitivefindingworthyoffurther

research.

Anothermeasureofprocess-relatedquality,doctoreffort,isderivedfromdirect

observationsofclinicalexaminations.Onemeasureofdoctoreffortrecordsthenumberof

physicalexaminationsofthepatientbythedoctor(Das,HammerandLeonard,2008).11As

seeninTable2,thisvariablealsodoesnotsuggestmeaningfulvariationacrossprovider

types,afindingconfirmedbytheresultsofat-test.Infact,themeansandstandarddeviations

foreachprovidertypearequiteclosetoeachother.Thisfindingismoresurprisingvis-à-vis

11Thespecificphysicalexaminationsinthisstudyincludetheuseofastethoscope,bloodpressuremeasurement,gaugingbodytemperature,palpitation,checkingthepulse,andotherphysicalexaminationsrecordedbytheobserver.

17

sometheoriespresentedabove,whichimplythatstaffmembersatreligiousfacilities–and

especiallyatfacilitiesrunbyreligiousordersthatmakegreatdemandsontheiradherents–

wouldexertmoreefforttotheirworkininstitutionsrunbythereligiousorder.

Themedicalvignettesprovideawealthofinformationrelatedtoprocessquality,

focusinginparticularondoctors’medicalknowledge.Basedonfourvignettesofdifferent

healthconditionscommonlyfoundinLebanon,weconstructanindicatorof“doctorobjective

knowledge,”whichgaugesthenumberofvignettesdiagnosedcorrectlybythedoctorand

rangesfrom0to4.ThismeasurepointstoapotentialwelfareadvantagebysecularNGOs.As

seeninTable2,doctorsinNGOsearnthehighestaveragescorewith2.3conditionscorrectly

diagnosed,whereastheaveragescoresforotherorganizationaltypesareallapproximately

onecorrectdiagnosisoutoffour.At-testcomparingtheaveragenumberofcorrectdiagnoses

ofdoctorsintheNGOtypewiththemeanofallotherprovidertypesalsosuggeststhatthe

differenceisstatisticallysignificantatthe10percentlevel(t=-2.3527,df=4.224,p-value=

0.07483).ThisfindingprovidessuggestiveevidencethatNGOssomehowrecruitmore

competentdoctors.

Finally,theindicatorsweuseforoutcome-relatedqualityaresubjectivemeasuresof

satisfactionreportedbypatientsforthereasonswenoteearlierinthepaper.Intwodifferent

questions,patientsreporttheirlevelsofsatisfactionwiththehealthcenterandwiththeir

doctor,respectively.Patientsatisfactionalsoexhibitsmeaningfulvariationacrossprovider

types,againwithsecularNGOsdisplayingthehighestoverallvaluesonrelatedmeasures.The

averagescoreofpatientsatisfactionwiththecenterisalmost4.3forsecularNGOs,whereasit

isaround3.7forothertypes.Similarly,theaveragescoreofpatientsatisfactionwiththe

doctoris4.4forsecularNGOs,whereasitisaround3.7forotherorganizationaltypes.At-test

comparingthemeanvaluesofpatientsatisfactionwiththedoctorinsecularNGOsandin

othertypesofprovidersalsoindicatesthatthedifferenceisstatisticallysignificantatthe1

percentlevel(t=-3.4446,df=16.769,p-value=0.003146).Thisresultindicatesthatpatients

18

havemorefavorableperceptionsofdoctorsatNGOsthanatothertypesoffacilities,

regardlessofwhetherornotthequalityofcarewassuperiorbymoreobjectivemeasures.

Insum,descriptiveanalysesindicatethatmeasuresofqualityinprimaryhealthcare

aresimilaracrossprovidertypesforstructuralindicators,suchastheavailabilityof

administrativeandmedicalinfrastructure,andsomedimensionsofprocess-related

indicators,notablygovernanceproceduresandprovidereffortatthefacilitylevel.Measures

ofdoctorknowledgeandpatientsatisfaction,however,varyacrossprovidertypes,with

secularNGOsexhibitingadistinctadvantageinbothareas.Inthenextsection,weexplore

thesedescriptivefindingsinmoredetailtoseeiftheapparentsecularwelfareadvantagestill

holdsaftercontrollingforpotentialconfounders,andifso,whichcharacteristicsofNGO-run

healthcentersandofdoctorsatthesefacilitiesmightcontributetoexplainingthisvariation.

StatisticalAnalysesofSubjectiveHealthCareQuality

Controllingforpotentialconfounders

Thedescriptiveanalysissuggeststhatpatientsatisfactionlevelsarehigherathealthfacilities

runbysecularNGOs.Alinearregressionmodelthatusesdummiesforeachprovidertype

withpublicinstitutionsasthebenchmarkcategoryshowsthatthisassociationisalso

statisticallysignificantatthe5percentlevel(Table3Column1).Column2addstothemodel

anumberofpatientcharacteristicsthatcouldaffectboththechoiceofproviderandpatient

satisfaction.Thesepotentialconfoundersincludepersonalcharacteristics,generalhealth

conditions,anddistancetraveledtothefacility.12ThecoefficientonthetypeNGOvariable

remainspositiveandstatisticallysignificant.InColumn3,wethenshowthatpatient

satisfactionwiththefacilityisalmostperfectlypredictedbypatientsatisfactionwiththe

doctor.

12Formoreinformationonanddescriptivestatisticsofthesecontrolvariables,seetheSupplementalOnlineAppendixPartA2.

19

Takingsatisfactionwiththedoctorasthedependentvariable,Columns4and5suggest

thatpatientsaremoresatisfiedwithdoctorsinfacilitiesrunbysecularNGOs,evenafter

controllingforthesamebatteryofpotentialconfounders.Therelationshipisstatistically

significantattheconventional5percentlevel.Forourkeyvariableofinterest–thevariable

indicatingNGOtype–wealsoreportblock-bootstrappedstandarderrorstoovercomethe

potentialproblemofclusteringatthePHClevel.13Eventhoughthelargerstandarderrors

showthattheuncertaintyaroundtheestimatedeffectofNGOtypeincreaseswithblock-

bootstrapping,theeffectsarestillarguablyrobust,especiallygiventhesmallsamplesize.

TheestimationsinTable3suggestthatpatientsatisfactionwithdoctorsinNGO-run

facilitiesisalmostonestandarddeviationhigherthanthatofpatientsinpublicinstitutions.In

otherwords,therelationshipissubstantivelyimportant,callingforfurtherexploration.14

Becausepatientsatisfactionwiththedoctoralmostperfectlypredictssatisfactionwiththe

facility,wefocusonsatisfactionwithdoctorsasthedependentvariableintheremainderof

theanalyses.

13Block-bootstrappingisatechniqueofestimatinguncertaintywhenthereisalegitimateconcernaboutcorrelatederrortermsinamodel(orwithin-groupdependence),butthenumberofclustersissmallforcalculatingcluster-robuststandarderrors(Cameron,GelbachandMiller,2008).Inourcase,weuseblock-bootstrappingasourdataisclusteredatthePHClevel.14HierarchicallinearmodelswithvaryinginterceptsatthedoctorlevelorthePHClevel,andanorderedprobitmodelgenerateverysimilarresultsintermsofsubstantialandstatisticalsignificancetotheresultsofthelinearregressionmodelwereportinthemainbodyoftext.SeeOnlineAppendixPartA3foralternativespecificationsoftheoutcomemodel.Furthermore,weprovidetheresultsofmatchinginOnlineAppendixPartA4,whichleadtoverysimilarresults.

20

Table3:Regressionresultsregardingprovidertypeandpatientsatisfactionwiththehealthcenterandwiththedoctor

========================================================================================================================================================= Dependent variable: ------------------------------------------------------------------------------------------------------------------ Patient's satisfaction with the PHC Patient's satisfaction with the doctor (1) (2) (3) (4) (5) --------------------------------------------------------------------------------------------------------------------------------------------------------- Provider: NGO 0.600 0.512 0.048 0.733 0.568 (0.244)** (0.270)* (0.165) (0.239)*** (0.265)** [0.373]+ [0.412] [0.179] [0.342]** [0.380]+ Provider: Religious 0.016 0.290 0.110 0.032 0.221 (0.192) (0.212) (0.128) (0.188) (0.208) Provider: Political 0.065 0.316 0.126 0.089 0.234 (0.201) (0.223) (0.135) (0.198) (0.219) Gender: Female 0.199 0.004 0.239* (0.130) (0.079) (0.127) Age -0.003 -0.001 -0.002 (0.005) (0.003) (0.005) Socioeconomic status -0.187*** -0.009 -0.218*** (0.070) (0.044) (0.068) Minutes of transport to center -0.007 -0.009 0.002 (0.014) (0.009) (0.014) Vehicle used in transport (dummy) 0.032 0.067 -0.043 (0.163) (0.098) (0.160) Days of sickness before visit 0.003 0.002 0.002 (0.005) (0.003) (0.005) Self-reported health status 0.097 -0.009 0.130* (0.071) (0.043) (0.070) Previous visit to center (dummy) 0.173 0.139* 0.042 (0.136) (0.082) (0.133) Patient satisfaction with doctor 0.817*** (0.059) Constant 3.667*** 3.591*** 0.645** 3.667*** 3.606*** (0.172) (0.411) (0.325) (0.169) (0.404) --------------------------------------------------------------------------------------------------------------------------------------------------------- Observations 134 121 121 134 121 R2 0.070 0.155 0.698 0.103 0.196 Adjusted R2 0.049 0.070 0.664 0.082 0.115 Residual Std. Error 0.668 (df = 130) 0.662 (df = 109) 0.398 (df = 108) 0.655 (df = 130) 0.650 (df = 109) F Statistic 3.283** (df = 3; 130) 1.815* (df = 11; 109) 20.789*** (df = 12; 108) 4.962*** (df = 3; 130) 2.412** (df = 11; 109) ========================================================================================================================================================= Note: Normal standard errors are in parantheses, and block-bootstrapped standard errors (10,000 resampling) are in brackets. *p<0.1; **p<0.05; ***p<0.01; +p<0.15.

21

Potentialmediators

Whatfactorsmightmediatebetweenprovidertype,i.e.theapparentNGOadvantage,and

patientsatisfaction?Wefocusondoctor-levelvariablesaspotentialmediators,sincepatient

satisfactionistoaveryhighdegreedeterminedbysatisfactionwiththedoctor.Potential

mediatorsatthelevelofthedoctor,whichcanbothbeaffectedbyprovidertypeandaffect

patientsatisfaction,include:15

• Medicalknowledge:Patientsmaybemoresatisfiedwithdoctorswhoaremorecompetent,

asmeasuredbytheirmedicalknowledge.

• Jobsatisfaction:Thevignettessurveyasksdoctorshowsatisfiedtheyarewith

organizationwheretheywork.IfdoctorsinNGO-runfacilitiesexpressgreatersatisfaction

withtheirjobs,thenpatientsmayratethemmorefavorably.

• Professionalexperience:Patientsmaybemoresatisfiedwithdoctorswithmore(orless)

experience,andlevelsofexperiencemightalsobecorrelatedwithprovidertype.

• Perceiveddoctorcredentials:SomeLebaneseregarddoctorswhoreceivedtheirmedical

degreesfromformercommunistcountriesaslessqualified,andthereforepatientsmaybe

lesssatisfiedwithdoctorswiththesecredentials,irrespectiveoftheircapabilities.

Toseeifanyoftheabovefactorsactaspotentialmediators,wefirstneedto

demonstratethatthereisastatisticallysignificantrelationshipbetweenagivenvariableand

providertype,especiallyNGOtype.Tothatend,weregressthesevariablesonprovidertypes

alongwithappropriatecontrols(seeTable4).Theresultssuggestthatthereisapositiveand

significantrelationshipbetweenNGOtypeanddoctormedicalknowledgeandwithdoctorjob

satisfaction.Inotherwords,doctorswhoworkinNGOsarebothmorecompetentinwhat

theydoandmoresatisfiedwiththeircurrentjob.Thus,thesetwofactorsmaymediatethe

positiveassociationbetweensecularNGOsastheprovidertypeandpatientsatisfaction.

15SeetheOnlineAppendixPartA2fordescriptivestatisticsonthedoctor-levelpotentialmediatorvariables.

22

Table4:Regressingpotentialmediatorsonprovidertype

============================================================================================================ Dependent variable: ------------------------------------------------------------------------------------ Dr. obj. knowledge Dr. job sat. Dr. experience Dr. degree: Communist (1) (2) (3) (4) ------------------------------------------------------------------------------------------------------------ Provider: NGO 1.185 0.933 -5.368 -0.200 (0.382)*** (0.439)** (6.599) (0.317) [0.444]** [0.501]* [5.539] [0.335] Provider: Religious 0.152 -0.027 -2.247 -0.447* (0.321) (0.370) (5.595) (0.255) Provider: Political 0.087 0.022 0.418 -0.217 (0.320) (0.370) (5.575) (0.267) Doctor experience -0.014 0.006 (0.010) (0.012) Doctor degree: Communist -0.297 0.547** -6.839* (0.217) (0.256) (3.588) Constant 1.540*** 3.035*** 27.072*** 0.800*** (0.413) (0.481) (5.455) (0.224) ------------------------------------------------------------------------------------------------------------ Observations 38 37 38 39 R2 0.376 0.316 0.128 0.097 Adjusted R2 0.279 0.206 0.023 0.020 Residual Std. Error 0.595 (df = 32) 0.682 (df = 31) 10.372 (df = 33) 0.501 (df = 35) F Statistic 3.462** (df = 5; 32) 2.864** (df = 5; 31) 1.214 (df = 4; 33) 1.252 (df = 3; 35) ============================================================================================================ Note: Normal standard errors are in parantheses, and block-bootstrapped standard errors (10,000 resampling) are in brackets. *p<0.1; **p<0.05; ***p<0.01; +p<0.15.

Estimatingthemediationeffect

Wenowtestthelinkbetweenthepotentialmediatorsandthedependentvariable,i.e.patient

satisfaction.Totestbothofthehypothesizedrelationships(betweentheexplanatoryvariable

andthepotentialmediator,andbetweenthepotentialmediatorandthedependentvariable)

simultaneously,weusethemediationanalysistechniqueandthemediationpackage(Imai,

KeeleandTingley,2010;Imai,Keele,TingleyandYamamoto,2011;ImaiandYamamoto,

2013).Unlikeothercausalmediationanalysistechniques,thismethodenablesnon-

parametricidentificationofthemediationeffect,eveniflinearrelationshipsareassumed

betweentheexplanatoryvariableandthemediatorandbetweentheexplanatoryvariableand

thedependentvariable.Itproducesestimationsoftheaveragecausalmediationeffect

(ACME),whichrepresentstheportionoftheestimatedeffectoftheexplanatoryvariableon

theoutcomevariablethatgoesthroughthetestedmediator.

ToestimatetheACMEforeachpotentialmediator,themediationpackagerequires

specificationofanoutcomemodelandamediatormodel,throughwhichitthengenerates

predictionsforthemediatorandtheoutcomeandnonparametricallycomputestheACME.We

specifythesameoutcomemodelasinColumn5ofTable3,whileaddingthepotential

mediatorsandcontrolsatthedoctorlevel,asrequiredbythistechnique.Potentialmediator

variablesareatthelevelofdoctor,thustheoutcomemodelturnsintoamulti-levelmodel.For

thepotentialmediators–doctormedicalknowledgeanddoctorjobsatisfaction–themodelis

specifiedasinColumn1andColumn2ofTable4,respectively.Thus,themodelspecifications

forthemediatorandtheoutcomecanbedepictedasfollows:

!" = % + '(" + )*" + +"

,-" = ." + /0-" + 1-"

." = . + 2(" + 3!" + 4*" + 5"

inwhichVjisthevectorfordoctor-levelcovariates,Xijisthevectorforpatient-level

covariates,and�j,�ijand�jareeachnormallydistributedstochasticerrorswithzero

24

mean. TheACMEsareidentifiedwith90percentquasi-Bayesianconfidenceintervalsbased

on1,000simulations.TheresultsarepresentedinTable5.16

Table5:EstimatingtheAverageCausalMediationEffect(ACME)

Potentialmediatorvariable

Averagecausalmediationeffect

90%CIlowerlimit 90%CIupperlimit Proportionofthetotaleffectthroughthismediator

Doctormedicalknowledge

0.135 -0.213[-0.261]

0.536[0.617]

17.65%

Doctorjobsatisfaction

0.244 0.007[-0.053]

0.621[0.704]

28.17%

Note:“mediate”commandinthemediationpackageinRisusedtocalculatetheestimationsreportedinthistable.ACMEestimateandquasi-Bayesianconfidenceintervalsforeachpotentialmediatorarecalculatedwith1000simulations.Block-bootstrappedconfidenceintervallimits(individualPHCsareusedasblocks)areinbrackets.Whenblock-bootstrapping,100simulationswereusedforeachofthe500resamplings.

Table5suggeststhatdoctorjobsatisfactionisamuchmorelikelymediatorbetween

providertype(NGO,specifically)andpatientsatisfactionthandoctormedicalknowledge.

EventhoughdoctorsinNGOsusuallyhavehigherlevelsofmedicalknowledge,patients

treatedbythesedoctorsarenotnecessarilymoresatisfiedwiththecaretheyreceive,alogical

findinggiventhatnon-medicalprofessionalsarenotoftenqualifiedtoevaluatetechnical

training.ThisisrepresentedinthefirstlineofTable5,inwhichtheACMEofmedical

knowledgeisestimatedtobenotstatisticallydifferentfromzero.

ThesecondlineofTable5suggeststhatthepositiveeffectofNGOsonpatient

satisfactionmightatleastpartiallybeduetothehigherjobsatisfactionofdoctorsworkingin

facilitiesrunbysecularorganizations.TheACMEfordoctorjobsatisfactionisestimatedtobe

morethan0.2,andthevalueswithinthe90percentconfidenceintervalarealsodifferent

16Toestimatethemediationeffect,bothinthemodelpredictingtheoutcomeandinthemodelpredictingthemediatorweusesmallerversionsofthepatient-levelanddoctor-leveldatasets,because“thecurrentversionofthemediationpackagerequiresthatthemodelframesofthemediatorandoutcomemodelscontaintheexactsamesetofgroups,whichbecomesimportantwheneachmodelcontainsdifferentcovariatesandsomegroupsdropoutofthemodelframesduetomissingness.”(SeeTingley,Yamamoto,Hirose,KeeleandImai,2014.)Thus,thesmallerversionofthepatient-leveldatasetdoesnotincludethepatientswhowereexaminedbydoctoreliminatedfromthemediatormodelduetodatamissingness.Theestimationsoftheoutcomemodelbasedonthelargerdataset(n=135)andthesmallerdataset(n=97)arequalitativelythesame.

25

fromzero.ThegreaterjobsatisfactionofdoctorsinsecularNGOsexplainsonaverage28

percentoftheNGOadvantageingarneringhigherlevelsofpatientsatisfaction.

Asinthepreviousmodels,weemploytheblock-bootstrappingtechniqueinthe

mediationanalysistoovercomepotentialcorrelationinerrortermsduetotheunmeasured

effectsofindividualhealthcenters.Evenwithwiderconfidenceintervals,doctorjob

satisfactionremainsamuchmorelikelymediatorbetweenprovidertypeandpatient

satisfaction.17

Checkingthesensitivitytotheassumptionsofcausalmediation

Themediationanalysistechniqueweusedinthispaperinherentlyarguescausalityand,to

thatend,makesanimportantassumptioncalled“sequentialignorability.”Inadditiontothe

regularignorabilityofthetreatmentassumption,sequentialignorabilityassumesno

pretreatmentandposttreatmentconfoundingbetweenthemediatorandtheoutcome

variable.Totestforpretreatmentconfoundingbetweenthemediatorandtheoutcome,Imai,

KeeleandTingley(2010)offerasensitivityanalysisinwhichthesensitivityoftheACME

estimationscanbetested.Thisanalysisisbasedonthecorrelation,denotedwithρ, between

theerrortermofthemodelpredictingthemediatorandtheerrortermofthemodel

predictingtheoutcome.Ifsequentialignorabilityholds,allrelevantpretreatment

confoundershavebeenconditionedon,andthusρequalszero.Throughsimulation,itis

possibletocalculatethevaluesofρforwhichtheACMEiszerooritsconfidenceintervalis

zero.IftheestimatesoftheACMEcontainzeroatlowervaluesofρ,thisindicatesahigher

possibilitythattheremightbeunmeasuredpretreatmentconfoundersthatbothcauseboth

themediatorandtheoutcome,andthereforethesuggestedcausalpathmightbespurious.

17ThemarkedincreaseintheconfidenceintervalindicatesthatsomePHCsrunbysecularNGOsgarnerhigherlevelsofpatientsatisfactionthroughhigherlevelsofdoctorjobsatisfactionthanothers.Thisdeservesfurtherexplorationinfutureanalysesandinextensionsofthestudy.

26

Figure1reportsthesensitivityanalysis,i.e.ACMEestimatefordoctor’sjobsatisfaction

asafunctionofρ,forthecausalpathbeingarguedinthispaper.18Accordingly,ACMEturnsto

zerowhenρis0.2.Inotherwords,ifthereisapretreatmentconfounderthatleadstoa0.2

correlationbetweentheerrorterms,theACMEestimateturnsto0.Moreover,thelower

boundoftheconfidenceintervalforACMEturnszeroinverysmallamountsofcorrelation.

ThissuggestsamoderatedegreeofrobustnessoftheACMEestimatetopretreatment

confounders.Yet,theformulationofthequestionfordoctor’sjobsatisfaction19givessome

levelofconfidencefortheunconfoundedandpost-treatmentcharacteristicofthisvariable.

Furthermore,alargersamplesizewouldlikelyestablishtherobustnessofthesefindings.

Figure1:SensitivityAnalysisforDoctor’sJobSatisfactionasMediator

Sequentialignorabilityalsoassumesthatthereisnoposttreatmentconfoundingbetweenthe

mediatorandoutcomevariables.Themostimportantreasonforposttreatmentconfounding

mightbeacausalrelationshipbetweenpotentialmediators.FollowingImaiandYamamoto

18Thecurrentversionofthemediationpackagedoesnotallowforsensitivityanalyseswhenmultilevelmodelsareusedtopredicttheoutcomeandthemediator.Therefore,forthesakeofthesensitivityanalysis,weusedalinearregressionintheoutcomemodel.Thecriticalquantityofinterest,i.e.thelevelofsensitivityparameterρ,atwhichtheACMEestimateturnstozero,wouldnotdifferbetweenmodelsusingmultilevelregressionandmodelsusinglinearregression.19“Howwouldyourateyoursatisfactionwithyourjobinthishealthcenter?”(emphasisadded).

27

(2013),weregressthemediatorofinterest(inthiscase,doctorjobsatisfaction)ontheother

potentialmediator(doctormedicalknowledge)usingthetreatmentandappropriatecontrol

variables.BoththeregressionandanF-testsuggestthatthereisnosignificantrelationship

betweenthetwopotentialmediatingfactors.(See Online Appendix Part A5.) Itisimportantto

recallthatthisisabaselinecheck:Eventhoughwefailtorejectthenullhypothesisofno

conditionalassociation,wecannotfullyruleoutthepossibilityofacausalrelationship

betweenpotentialmediators.Nevertheless,thisresultgivesusmoreconfidencethatatleast

someofthepositiveeffectofNGOprovidertypeonpatientsatisfactionismediatedthrough

doctorsatisfactionwithherpositioninthehealthcenter.

Explainingthesecularwelfareadvantage?

AnalysesofdiverseindicatorsofthequalityofprimaryhealthcareinLebanonsuggestthat

doctorsatfacilitiesrunbysecularNGOsaremoresatisfiedwithandcommittedtothehealth

centerswheretheyworkandthatpatientshavemorefavorableviewsofprovidersatthese

facilities.Conversely,patientsexpressmorenegativeperceptionsofprovidersatfacilitiesrun

byreligiouscharitiesand,insomerespects,bypoliticalgroupsthanthoserunbyothertypes

ofinstitutions,whilemeasuresofinfrastructureandgovernanceproceduresshowno

meaningfulvariationacrossinstitutionaltype.Furthermore,doctorsatsecularNGOsappear

tobemorecompetentattheirprofession.

Thesefindingscontradictsometheoreticalandempiricalexpectations.First,several

strandsofliteraturesuggestthatreligiouscharitiesdeliversuperiorservices,whether

becausetheircharitablemissionsservetomotivatestaffmembersorbecauseexigent

religiousorganizationalcharacteristicsattractespeciallycommittedpersonnel.Second,the

resultsmaybesurprisinginthecontextoftheLebanesewelfareregime,wherepublicand

secularprovidersarewidelyperceivedaseitherinferiorormoreunder-resourcedthan

28

religiousandpoliticalgroupswhileFBOsofvariousstripesandsectarianpartiesdominate

thepoliticalsystemandcontrolsubstantialpublicandprivateresources.

Whatmightaccountfortheostensiblesecularadvantageinservicedeliveryin

Lebanon?Paradoxically,therelativemarginalizationofsecularorganizationsinpoliticsand

thewelfareregimemayworkintheirfavor.Onthesupplyside,giventhattheylackinfluence

inthesectariansystem,secularprovidersmayattractdoctorswhoareespeciallycommitted

toasenseofprofessionalismandhavelittletogainbeyondthesatisfactionofadvancingnon-

sectarian,humanitarianprinciples,acoremissionofthesecularNGOsinthesample.20These

ideologicalcommitmentsmayserveassourcesof“intrinsicmotivation”(RyanandDeci,

2000).forstaffmembers.Furthermore,seculargroupsinLebanondonothavewell-

developedpatronageandclientelistnetworks(Cammett,2014)and,therefore,their

professionalstaffcannotfulfillreciprocalobligationsthroughserviceinthesefacilitiesnor

cantheyderivematerialbenefitsbeyondgainingprofessionalexperienceandbuildingtheir

professionalreputations,amotivationsharedbydoctorsworkinginalltypesofhealth

networks.Asaresult,onaveragedoctorswhoworkatsecularNGOsmaybemorelikelyto

selectintotheseorganizationsinordertofulfillprofessionalgoals.

Ourfindingsaboutasecularwelfareadvantageareparticularlystrongwithrespectto

subjectivemeasuresofquality.Onthedemandside,beneficiariesmayperceivesecularNGOs

tobelesscorruptsincethesegroupsareeffectivelyshutoutofnationalpoliticsandderiveno

benefitfromthesectarianpower-sharingsystem,whichiswidelydisparagedbyLebanese

(Atallah,2012).Asaresult,secularNGOs,whicharenottaintedbyassociationwiththe

corruptandineffectivepoliticalsystem,maybenefitfromthesamekindofreputational

advantagethatsomereligiousactorsenjoyinpolitieswithcorruptsecularrulers(Brooke,

2014;CammettandJonesLuong,2014;Masoud,2014;Pepinsky,LiddleandMujani,2012).

20InterviewbyCammettwithChiefMedicalOfficer,LebaneseNGO,Beirut,January19,2015;InterviewbyCammettwithDirector,LebaneseNGO,Beirut,January15,2015.

29

Furthermore,lowexpectationsofsecularNGOscouldleadtoinflatedsatisfactionratings

whenpatientsdiscoverthattheservicesrenderedarebetterthananticipated,an

interpretationthatshouldbetestedmoresystematicallywithpublicopiniondatainfuture

research.

Inshort,inLebanon,whereseculargroupsareexcludedfrompatronagenetworksand

operateonthefringesofpower,servinginaffiliatedorganizationscallsuponpersonnelto

makepersonalsacrificesbyforegoingopportunitiestobenefitfromestablishedpatronage

networksandbydevotingthemselvestogroupsthataremarginalizedinpoliticalandsocial

life.Inturn,thehighcommitmentofstaffmemberstotheprogramsandactivitiesofsecular

groups,includingintherealmofwelfare,mayresultinmorefavorableperceptionsoftheir

services.Ourtentativeexplanationthereforepointstothewaysinwhichsociopolitical

contextmediatestherealandperceivedactivitiesofserviceproviderswithdistinct

organizationalmissions.

Conclusion

BasedonfindingsfromLebanon,whichfeaturesdiversepublicandnon-stateservice

providers,thispaperexploreswhetherdifferenttypesoforganizationsexhibitawelfare

advantageinthedeliveryofbasichealthcare.InsightsfromtheliteraturesonFBOsandthe

economicsofreligionaswellasspecificcharacteristicsoftheLebanesewelfareregime

suggestthatreligiouscharitiesand,especially,sectarianpartiesshouldofferhigherquality

servicesthanothertypesofproviders,notablythepublicsectorandsecularNGOs.Wefind

insteadthatsecularNGOsexhibitanapparentwelfareadvantageonsomeobjectivemeasures

(i.e.,doctorknowledge)and,morestrongly,onsubjectivemeasures.Toexplaintheapparent

secularwelfareadvantage,wehypothesizethatseculargroupsenjoyareputationaladvantage

inLebanon,wherereligionisassociatedwiththecorruptsectarianpower-sharingsystem.

SecularNGOs,whichofferfewmaterialrewardstotheirstaff,mayalsoattractqualifiedand

30

committedpersonnel.Inshort,sociopoliticalcontextmaymediatepopularperceptionsof

distinctwelfareinstitutionsandmayevenshapeselectioneffectssothatmorequalified

professionalsopttoworkforsometypesoforganizationsoverothers.

TheseinsightsfromLebanonaremostclearlygeneralizabletootherpolitieswith

politicallysalientidentity-basedcleavagesandwherediversenon-stateprovidersplay

importantrolesinthewelfaremix.YettheLebanesemaycaseofferrelevantinsightsintothe

politicsofservicedeliveryinotherplaces,too,especiallyinlightofthegrowingimportanceof

non-stateprovision,includingincountrieswithmorestatisteconomiclegacies.Furthermore,

thefindingscallforabroaderinvestigationoftheinterplaybetweenpoliticalcontext,

organizationalmissionandthequalityofsocialserviceprovision.Inparticular,future

researchshouldexplorethewaysinwhichformalandinformalfeaturesofthepolitical

systemshapethetypesofproviderorganizationsthatattractthemostcompetentpersonnel,

whichaffectsservicequalityintangibleways,andcitizenperceptionsoftherelative

proficiencyofdistinctproviders,whichcanaffectsubjectiveevaluationsofprovidersaswell

aspatientcompliancewithmedicaladvice,amongotheroutcomes.

31

ReferencesTalalAsad,FormationsoftheSecular:Christianity,Islam,Modernity.Stanford,CA:StanfordUniversityPress,2003.SamiAtallah,ArabBarometerII:LebanonCountryReport,(AnnArbor,MI:ArabBarometer,2012).StevenBrooke,WhyDoIslamistsProvideSocialServices?InRethinkingIslamistPolitics,(Washington,D.C.:ProgramonMiddleEastPoliticalScience,2014)A.ColinCameron,JonahB.Gelbach,andDouglasLMiller,"Bootstrap-basedimprovementsforinferencewithclusterederrors,”TheReviewofEconomicsandStatistics,90(2008),414-427.Cammett,M.(2014).CompassionateCommunalism:WelfareandSectarianisminLebanon.Ithaca,NY:CornellUniversity.MelaniCammett,andPaulineJonesLuong,"IsThereanIslamistPoliticalAdvantage?",AnnualReviewofPoliticalScience,17(2014),187-206;MelaniCammett,JuliaLynchandGavrilBilev,“TheInfluenceofPrivateHealthcareFinancingonCitizenTrustinGovernment,”PerspectivesonPolitics,13(2015):938-957;MelaniCammettandLaurenMorrisMacLean.ThePoliticsofNon-StateWelfare.Ithaca,NY:CornellUniversityPress,2014.TomChristensen,andPerLægreid,"Trustingovernment:Therelativeimportanceofservicesatisfaction,politicalfactors,anddemography,”PublicPerformance&ManagementReview,28(2005),487-511.GerardClarkeandMichaelJennings,eds.,Development,CivilSocietyandFaith-BasedOrganizations:BridgingtheSacredandtheSecular(NewYork:PalgraveMacmillan,2008)RamCnaan,TheInvisibleCaringHand:AmericanCongregationsandtheProvisionofWelfare(NewYork:NewYorkUniversityPress,2002)JishnuDas,“TheQualityofMedicalCareinLow-IncomeCountries:FromProviderstoMarkets,”PLoSMedicine8(2011).JishnuDasandJeffreyHammer,“QualityofPrimaryCareinLawIncomeCountries:FactsandEconomics,”AnnualReviewOfEconomics,6(2014),525-553.JishnuDas,JeffreyHammer,andKennethLeonard,“TheQualityofMedicalAdviceinLow-IncomeCountries,”JournalofEconomicPerspectives,22(2008),93-114.AvediDonabedian,“TheQualityofCare:HowCanitbeAssessed?”JournaloftheAmericanMedicalAssociation,260(1988),1743-1748.MarkJ.DeHaven,IrbyB.Hunter,LauraWilder,JamesW.Walton,andJarettBerry,“HealthProgramsinFaith-BasedOrganizations:AreTheyEffective?”AmericanJournalofPublicHealth,94(2004),1030-1036.

32

LaurenceR.Iannaccone,"WhyStrictChurchesareStrong,"AmericanJournalofSociology,99(1994),1180-1211.KosukeImai,LukeKeele,andDustinTingley,"Ageneralapproachtocausalmediationanalysis,"PsychologicalMethods,15(2010).KosukeImai,LukeKeele,DustinTingley,andTeppeiYamamoto,"Unpackingtheblackboxofcausality:Learningaboutcausalmechanismsfromexperimentalandobservationalstudies,"AmericanPoliticalScienceReview,105(2011),765-789KosukeImai,andTeppeiYamamoto,"Identificationandsensitivityanalysisformultiplecausalmechanisms:revisitingevidencefromframingexperiments,"PoliticalAnalysis,21(2013),141-171.AzzaKaram,JulieClague,KatherineMarshall,andJillOlivier,"TheViewfromAbove:FaithandHealth,"TheLancet,1-3(2015)AnneKlassen,AntonMiler,NancyAnderson,JaneShen,VeronicaSchiariti,andMaureenO'Donnell,"PerformanceMeasurementandImprovementFrameworksinHealth,EducationandSocialServicesSystems:ASystematicReview,”InternationalJournalforQualityinHealthCare,22(2010),44-69.KennethLeonardandMelkioryC.Masatu,“OutpatientProcessQualityEvaluationandtheHawthorneEffect,”SocialScienceandMedicine63(2006):2330-2340.MichaelMarmotandRichardWilkinson(eds.),SocialDeterminantsofHealth(Oxford:OxfordUniversityPress,2005).TarekMasoud,CountingIslam:Religion,Class,andElectionsinEgypt,(NewYork:CambridgeUniversityPress,2014)ThomasB.Pepinsky,R.WilliamLiddle,andSaifulMujani,"TestingIslam’sPoliticalAdvantage:EvidencefromIndonesia",AmericanJournalofPoliticalScience,56(2012),584-600.RitvaReinikkaandJakobSvensson,"WorkingforGod?EvidenceFromaChangeinFinancingofNonprofitHealthCareProvidersinUganda,"JournaloftheEuropeanEconomicAssociation,8(2010),1159-1178.PeterC.Rockers,MargaretE.Kruk,MirianJ.Laugesen,“PerceptionsoftheHealthSystemandPublicTrustinGovernmentinLowandMiddle-IncomeCountries:EvidencefromtheWorldHealthSurveys,”JournalofHealthPolitics,PolicyandLaw,37(2012),405-437.RichardM.Ryan,andEdwardL.Deci,"IntrinsicandExtrinsicMotivations:ClassicDefinitionsandNewDirections,"ContemporaryEducationalPsychology,25(2000),54-67.Salti,NisreenandJadChaaban.“TheroleofsectarianismintheallocationofpublicexpenditureinpostwarLebanon.”InternationalJournalofMiddleEastStudies42,4(2010):637-655.

33

HenryT.,StelfoxandSharonE.Straus,"MeasuringQualityofCare:ConsideringMeasurementFrameworksandNeedsAssessmenttoGuideQualityIndicatorDevelopment,”JournalofClinicalEpidemiology,66(2013),1320-1327.WilliamSummerskill,andRichardHorton,"Faith-BasedDeliveryofScience-BasedCare,"TheLancet,1-2(2015)Thachil,Tariq,EliteParties,PoorVoters:HowSocialServicesWinVotesinIndia.NewYork:CambridgeUniversityPress,2014.DustinTingley,TeppeiYamamoto,KentaroHirose,LukeKeele,andKosukeImai,"Mediation:Rpackageforcausalmediationanalysis"JournalofStatisticalSoftware,59(2014),1-38.TranTuan,VanThiMaiDung,IngoNeu,andMichaelJ.Dibley,"ComparativeQualityofPrivateandPublicHealthServicesinRuralVietnam,"HealthPolicyandPlanning,20(2005),319-327.HeidiRolandUnruhandRonaldJ.Sider,SavingSouls,ServingSociety:UnderstandingtheFaithFactorinChurch-BasedMinistry(Oxford:OxfordUniversityPress,2005)WorldBank.MakingServicesWorkforthePoor.Washington,D.C.,WorldBank,2004.RobertWuthnow,SavingAmerica?Faith-BasedServicesandtheFutureofCivilSociety(Princeton:PrincetonUniversityPress,2004).