Child care quality: centers and home settings that serve poor families

tmostfromqualitycentersisoften explainedfromade citperspective:thesechildrenenterwithlowercognitiveandlanguagepro ciencies, comparedtomiddle-classchildrenwhobene tfromstrongerhomepractices(forreview, Burchinal,1999 ). Butthispatternwouldalsobeevidentifchildrenfrompoorfamiliesbene tfromrelativelyhighquality centers,comparedtolowerqualitycentersthatmaybedisproportionatelysituatedinmiddle-classand blue-collarneighborhoods( Fuller&Strath,2001 ).<br><br> Otherstudies,incontrast,showthatlow-incomeandnon-Englishspeakingparentsaremorelikelyto selectlowerqualitycenters(alongstructuralfeaturesorusingtheEarlyChildhoodEnvironmentRating Scale,ECERS),ortheseparentsrelymoreonhome-basedproviderswhoarelesseducated,onaverage, andlesslikelytocreateeducationallyrichsettingsforyoungchildren( CQO,1995 ; Goelman&Pence, 1987 ; NICHD,1997 ). Unevenqualitylevelsamongsubsidizedcentersmaybeexplainedbyinstitutionalforcesoperating atstateandlocallevels.Forexample, Fulleretal.(2003) foundthatcentersreceivingalargershareof theirbudgetfromsubsidiesandfallingunderCalifornia 9stighterregulationsdisplayedlowerchild 3staff ratiosandfewerchildrenperclassroom,comparedtofee-supportedcentersfallingunderlessstringent standards.WhenFloridaimprovedrequiredstaf ngratiosandteachereducationlevels,offeringanatural experiment,theobservedqualityofcentersandchildoutcomesrosemarkedly( Howesetal.,1998 ). 508 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 CenterslocatedinstateswithstrongerqualityregulationshavegenerallyscoredhigherontheECERS thancentersservingsimilarpopulationsinstateswithweakerstandards( CQOStudyTeam,1995 ; Holloway,Kagan,Fuller,Tsou,&Carroll,2001 ).TheNICHDstudyofearlychildcarefoundthatstate qualitystandardslikelymatter,showingthatchildrenscoredhigheronschoolreadinessandlanguage comprehensionat36monthsofagewhentheyattendedcentersmeetingseveralqualitybenchmarks, includingstaf ngratiosandprovidereducationlevels,althoughfewpriorselectionfactors(forchildren orteachers)wereavailableonwhichtocontrol( NICHD,1999 ).<br><br> Theregulationofprovidereducationlevels,orstrongercentersubsidiestosustainbettertrainedstaff, mayhelptoexplainqualityvariation.Forexample,centerstaffwithtwoorfour-yearcollegedegrees maydisplayhigherlevelsofsensitivityandresponsivenesswithpreschool-agechildren( Howes,1997 ; Howes,Whitebook,&Phillips,1992 ; Hamre&Bridges,2004 ).Butwestilldonotknowthemagnitude withwhichgradationsofeducationorchilddevelopmenttrainingyieldsigni cantchild-leveleffects,due toinsuf cientselectioncontrolsinearlystudies. Despitethesedifferingaccountsforwhyqualityvariesacrosscommunities,westillknowlittleabout thelevelanddistributionofqualityamonglow-incomeneighborhoods.Withsuchscarceobservational datafrompoorneighborhoods 3forcentersandhome-basedprograms 3evidenceisequallythinregarding howmultiplegaugesofqualitymaybeinterrelatedornot. 1.2.Comparingthequalityofcenterandhome-basedcare Mostyoungchildrenwithworkingmothersstillreceivecareinvarioustypesofhome-basedsettings.<br><br> In1997,forexample,amongthenation 9s10.1millionchildrenunder5yearsoldwithan employed mother,justover44%werecaredforbyagrandparentorkinmemberotherthanthefather,15%by unrelatedindividuals,and13%inregulatedFCCHs( Smith,2002 ).Amongtheadditional9.1million childrenwith stay-at-home mothers,20%receivedregularcarefromkin,5%fromunrelatedindividuals, andanegligiblenumberfromFCCHproviders. Kithandkinprovidersaresubjectedtocriminalbackgroundchecksinsomestatesbutrarelyface otherregulatoryrequirements.Theyareincreasinglyreimbursedthroughfederallyfundedvouchers, particularlyafter1996whenwelfare-to-workreformsspurredsubstantialincreasesinchildcarespending. Justoverafourthofallfamiliessupportedthroughthefederalchildcareblock-grant,forinstance,are usingunregulatedkithorkinmembersascaregivers,about700,000familiesmonthly( U.S.Department, 2002 ).InCalifornia,justoverhalfofallfamiliesinthestatewelfare-to-workprogramwhodrewchild caresubsidiesin2001wereusingunregulatedkithorkin( Hirshberg,2003 ).<br><br> Government 9srisingsupportofinformalarrangementshasunfoldeddespitelittleevidenceonthe qualityofthesesettings. Kontosetal.(1994) observed226home-basedsettings,includinglicensedand unlicensedFCCHsandindividualcaregivers,spreadacrossthreecities,revealinglowlevelsofquality, onaverage.Ontheseven-pointFamilyDayCareRatingScale(FDCRS),just9%scoredinthe cgood quality drange,de nedasscalescoresofbetween5and7points.Overathirdscoredinthe 8poorquality 9 range,withaveragescalescoresofjust1or2points,similarto vemoremodestevaluationsreviewed bytheKontosteam. 1 1 Thisstudysufferedfromalowresponserate.Oftheoriginal947eligibleparentsdrawn,739agreedtoparticipate.Butonly 145oftheirprovidersagreedtobeobserved.Asecondsamplewasdrawn,yieldinganadditional81providers.Anyresulting biasmaybeinthedirectionofin atingtheobservedqualityofthesesettings.Thatis,iftheresearchteamhadgainedaccessto amorerandomsampleofhome-basedproviders,averagequalitymayhavebeenlower.<br><br> B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 509 TheKontosstudyalsofoundthatyoungchildrenwerenotmoresecurelyattachedtokinmembers, comparedtounrelatedcaregivers(two-thirdsofallindividualcaregiverslivedbelowthepovertyline). Thisisconsistentwith Fosburg 9s(1982) nding,alsobasedonobservations,thatkinproviderstendedto interactlesswithyoungchildren,comparedtounrelatedcaregivers.Forty-sixpercentofthekinproviders hadnotcompletedhighschool,comparedtojust9%oftheFCCHproviders.RegulatedFCCHswere disproportionatelyusedbymiddle-classfamilies,comparedtolow-incomefamilieswhoweremorelikely toselectkinproviders. TheKontosteamalsofoundthatsomeindicatorsofqualitywereinterrelated.Withinhome-based settings,FCCHsdisplayedthehighestqualitylevelsasgaugedbytheFDCRSandtheArnettScaleof CaregiverSensitivity;theseprovidersexpressedastrongerintenttostayintheearlychildhood eld, comparedtokinproviders.FCCHproviders,whetherregulatedornot,tendedtocareformorechildren duringtheday,comparedtokinproviders,indicatingthatlargersizemaybeapositivecorrelateofother qualitymeasures.<br><br> Hollowayetal.(2001) observedsomewhathigherqualityamongFCCHsinCaliforniaandConnecti- cutintermsofmeanFDCRSscores,comparedtotheKontos ndings,forsampledmiddle-classand low-incomefamilies.TheKontosteamreportedFDCRSscoresaveraging3.4(outof7maximum,av- eragingacrosssubscales),comparedtoameanof4.3amongtheFCCHsobservedintheHolloway study.ThelatterteamalsoreportedwidevariabilityinscoresacrossindividualFDCRSscales,ranging fromameanof1.5onchildren 9sengagementwithactivitycorners,to3.0ontheuseofworksheets, and5.7ontheavailabilityofsandorwaterplaymaterials.FDCRSitemsrelatedtoplayandlearn- ingmaterialsweremoderatelycorrelatedtothecaregivers 9educationlevels( r =.32, p <.01)butnot tosocial-interactionitems.FCCHsenrollingmorechildrendisplayedhigherscoresonbothdimen- sionsofqualityembeddedintheFDCRS,comparedtohomeswithfewerchildren(similartoKontos 9 ndings). 1.3.Researchquestions Movingfromthesequalityissuesandtheextantliterature,webeganourstudyfocusingonapairof interrelatedquestions:Whattypesofchildcaredolow-incomeparentsselect,andwhatistheobserved qualityofthatcare?Theextenttowhichmaternalemploymentshapesyoungchildren 9sdevelopmentis likelymediated,inpart,bythesechildcareselectionpatterns.Givenscarceknowledgeofhowquality mayvaryacrosspoorneighborhoods,wesampledfamiliesandfollowedchildrenintotheirchildcare providersin vecitiessituatedacrossthreestates:California,Connecticut,andFlorida. Second,wecomparedthequalityofcenterandhome-basedcare,giventhepropensityofmanypoor familiestochoosethelatterandthelackofobservationalevidencefromthesesettings.Weadminis- teredmeasuresofquality,especiallygaugesofchild 3caregiverinteractionsthatallowedustoarriveat comparativeclaimsaboutthequalitiesofcenterversushome-basedproviders.<br><br> Third,werevisitedthequestionofwhethervariousdimensionsofqualityareinterrelated,orwhether certainstructuralorsocialprocessmeasuresmaymoveindependentlyamongcenterandhome-based settingsinlow-incomecommunities. Finally,weexaminethepredictivevalidityofcontextualandfamilyfactorsinestimatingthequal- ityofcarethatmothersselected.Thisallowedustoilluminatetherelativeef cacyofinstitutionalor contextualfactors,vis- ` a-visfamily-leveldeterminants.Inturn,wethenestimatedthedegreetowhich non-participationofsomecentersorhomeprovidersbiasedourestimatedlevelsofquality. 510 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 Table1 Attributesofallsampledmothersandchildren,1998(meansandS.D.sreported; n =927families) ResearchsiteMother 9sage (year) Focalchild 9s age(month) Ethnicity(%)Nohigh school diploma(%) Motherliveswith atleastoneother adult(%) BlackLatinoVietnamese SanFrancisco ( n =195) 292857(111)18(35)0(0)59(115)66(129) SanJose( n =219)29297(15)51(112)26(57)65(142)83(182) Manchester ( n =73) 262420(15)19(14)0(0)35(26)42(31) NewHaven ( n =238) 252544(105)21(50)0(0)36(86)36(86) Tampa( n =202)322947(95)14(28)1(2)50(101)52(105) 2.Procedures 2.1.Sample 4familiesandchildcaresettings Werandomlysampledsinglemotherswithatleastonechild,12 342monthsofage,duringthesecond halfof1998.Allwomenhadrecentlyenteredtheirstate 9snewwelfare-to-workprogram,placingimme- diatepressureonthemto ndpaidworkoutsidethehome.InourFlorida(Tampa)andtwoCalifornia (SanFranciscoandSanJose)sites,weconductedamaternalinterview,lasting90 3120min,atthelocal welfareof ceorbyphone.InConnecticut(ManchesterandNewHaven),participatingmotherswere interviewedbyphone.<br><br> 2 Amongthewomenmeetingthescreeningcriteria 3unmarriedmothersentering newwelfareprogramswithayoungchild 389%agreedtositforthe rstinterview.Thisresultedina sampleof927initialparticipants,spreadacrossthe5sites.Ifmorethanonechildfellintoourtargetage range,weselectedtheeldestasthefocalchild. Afollow-upinterviewonavarietyofchildcareissueswasthenconductedbetweenfourand18months aftertheinitialinterview,aftermostwomenhadselectedachildcareprovider. 3 Thedataforthepresent paperstemfromthesubsetwhoreportedusinganonparentalchildcareproviderforatleast10hperweek, equaling568oftheoriginalmothers.TheConnecticutfamiliesenteredarandom-assignmentexperiment, whereasnoassignmenttoacontrolgroupwaspossibleinCaliforniaorFlorida.Thepresentanalysisis notaffectedbytheexperimentaldesignoftheConnecticutsubstudy.<br><br> Table1 reportsbasiccharacteristicsfor all sampledmothersandchildren,whetherornottheyselected anonparentalchildcareprovider.Mostmotherswereintheirmidtolate20sandhadmodestlevelsof schoolattainment,withsigni cantvariabilityacrossthecitysites.Themeanfocalchild 9sageranged 2 TheConnecticutfamilieswerepartofarandom-assignmentexperimentwhichincludedsplittingthemothersintoaprogram orcontrolgroup.Thelatterhalfofthesampledidnotfacepressuretoworkoutsidethehome,nordidtheyfacetimelimitson receivingcashaid.Inclusioninthecontrolgroup,however,hadnosigni canteffectonqualityofcareselectedwhentestedin ourselectionmodels.Allotherdesignfeatures,maternalinterviewdata,andobservationalmethodswereidenticalacrossall ve researchsites.ForadditionaldetailsontheConnecticutsubsample,see Fulleretal.(2002) . 3 InterviewsofConnecticutmothersoccurredabout18monthsaftertheyenteredthestate 9snewwelfareexperiment.During theinitialinterviewswediscoveredthatmanywomenalreadywereusingachildcareprovider,sincetheywereworkingpart-time eventhougheligibleforcashaid,ortheywereunemployedbuthadenrolledtheirchildinacenterorFCCHprogram. B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 511 Table2 Attributesofsamplemothersandchildrenwithobservationaldatabytypeofchildcareselected,1998(meansandS.D.sreported) ChildcaretypeMother 9sage (year) Focalchild 9s age(month) Ethnicity(%)Nohigh school diploma(%) Motherliveswith atleastoneother adult(%) BlackLatinoVietnamese Center( n =166)273137(61)26(43)7(12)42(70)70(116) Familychildcare home( n =69) 262833(23)24(17)14(10)46(32)79(55) Kithorkin ( n =118) 272739(46)27(32)9(11)44(52)78(92) F orchi-square statistic .656.94 ** .57.242.77.423.30 7 7 p <.01.<br><br> from24monthsinManchesterto29monthsinSanJoseandTampa.Eachlocalsamplecloselymatched attributesoftheirrespectivecounty 9suniverseofwelfarefamiliescomprisedofasinglemotherandat leastoneyoungchild( Fulleretal.,2002 ).Thesesamplesarenotnecessarilynationallyrepresentativeof low-incomewelfareentrants. Weaskedeachmotherforherconsenttovisittheprimarychildcareprovider.Wegainedaccessto378 ofthechildcaresettings,equaling67%ofthesettingsreportedbythe568mothersusingnonparental care(atleast10hperweek).Theshareofcenterteacherswhoconsentedtotheobservationalsessionwas considerablyhigher,81%,comparedtotheparticipationrateofhome-basedproviders,60%(combining FCCHsandkithorkinproviders).Completedatanecessaryforthepresentanalyseswereavailablefor 353childcaresettings.including166observedcenters,69FCCHs,and118kithorkinarrangements. Table2 providesthesamedescriptivedatabutonlyforthesubsetoffamiliesforwhomwegained accessto,andobserved,theirprimarychildcareprovider.Childrenenrolledincentersweresigni - cantlyolder(31monthsofageonaverage),comparedtochildreninFCCHsorkithandkinarrange- ments(28and27months,respectively).Childrenincentersalsotendedtobefromhomeswherethe motherwastheonlyresidentadult,comparedtochildrenplacedinhomesettingswholivedwith atleastoneotheradult,perhapsofferingmoresocialsupportorindicatingamoreextendedkinship system.<br><br> Thetypeofcareselectedbymothersvariedsigni cantlyacrossthethreestates.Forexample,among thewomeninTampa,fully70%hadselectedacenterprogramafterenteringthestudy;5%hadselected anFCCH;and25%akithorkincaregiver.Incontrast,just13%oftheConnecticutmothersselected acenter;10%usedanFCCH;and77%wereinakithorkinarrangement.Californiafellinbetween: 29%selectedacenter;17%anFCCH;and54%akithorkinprovider.Thesebetween-stateselection differencesgenerallymappedagainsttheneighborhoodsupplyofcenters,asearlierdetailed( Fulleret al.,2002 ). 2.2.Measuresofcenterandhome-basedproviderquality Weassessedstructuralfeaturesofchildcaresettings, rstthrougha45-mininterviewwiththecenter teacher,FCCHoperator,orkithorkincaregiver.Thisprovidedinformationonthenumberofchildrenin thesetting,presenceofassistants,educationandexperiencelevelsoftheprovider,andtrainingspeci c 512 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 tochilddevelopment.Thechild 3staffratioandgroupsizewereveri edduringourobservationsbutonly forthedaywevisited. Duringtheobservation,beitacenterorhome,weadministeredtheArnettScaleofCaregiverBehavior ( Arnett,1989 ).Itfocusesonthecharacterofsocialinteractionbetweenchildandcaregiver,including thecaregiver 9sattentivenessandresponsiveness,propensitytoexplainmisbehaviorandreasonwiththe child,aswellaswarmthandaffect.Itemscodedbythe eldresearcherincluded, c(caregiver)speaks warmlytothechildren dand cpayspositiveattentiontothechildrenasindividuals. dTheArnettScalealso measurestheextenttowhichcaregiversexplainmisbehaviorormorecooperativebehavior,ratherthan discipliningthroughdirectivesabsentexplanation.Fieldstaffweretrainedcentrallyonallobservations measuresandrequiredtoreacha90%levelofinter-rateragreementonindividualscalesfortheArnett andfortheotherobservationmeasuresdiscussedbelow.<br><br> WealsoemployedtheChild 3CaregiverObservationSystem(C-COS)developedbyMathematica PolicyResearch,Inc.forthenationalevaluationofEarlyHeadStart.Thisisalow-inferencerecordingof actionsinwhichthefocalchildand/orproviderisengagedover40timedsnapshots.Fieldstaff,during each30-ssnapshot,check-offpossiblebehaviors,includingverbalinteractionbetweentheproviderand child,eitheractorinitiatingaquestion,childisworkingwithmaterials,thechild 9semotionalresponsesas aresultofactivityorinteraction,watchingtelevisionorvideo,andoverallratingsregardingthewarmth andresponsivenessmanifestinthechild 3caregiverrelationshipoverthe40snapshots.Duringtraining sessions,beforetheycouldbegin eldwork,eachstaffmemberreached90%inter-rateragreementamong the22possiblebehaviorsoremotionalreactionsobservedacross10snapshots. Attheendoftheobservationperiod eldstaffcompletedtheECERSforcentersandtheFDCRSforall home-basedsettings.Eachinstrumentgaugesavarietyofstructuralandphysicalaspectsoforganizations, suchasfacilitiesquality,availabilityofdevelopmentallyappropriatelearningandplaymaterials,the arrangementofchildactivities,andthenatureofchild 3caregiverinteraction( Harms,Clifford,&Cryer, 1997 ). 4 WehadreservationsaboutusingtheFDCRSinkithandkinsettings,sinceitawardshigher scorestohomesettingsthatareformallyarrangedforchildrenandinwaysthatresemblecenters.But thesescoresweresupplementedbythesocial-interactionconstructsgaugedbytheArnettandC-COS instruments.ThemajorityofmothersselectingFCCHsresidedinSanJose,alimitationdiscussedbelow.<br><br> 2.3.Predictorsofchildcarequalityselectedbymothers Attentiontotheselectionofchildcareofvaryingqualityinformstwoquestions.First,weknowlittle aboutwhetherfamily-levelfactorsshapethequalityofcarethatparents 8choose, 9beitacenterorhome- basedprovider.Giventheinstitutionalforcessketchedabove,family-levelfactorsmayplaylessofrole thansubsidyandregulatorypolicies,atleastinlow-incomecommunities.Second,earlierobservational studiesofchildcarequalityhavesufferedfromwidelyvaryingresponserates.Gainingaccesstohome settings,oftenselectedbylow-incomeparents,hasbeendif cultforresearchteams.Atypicaldesignisto focusexclusivelyoncenters,samplingorganizationsdirectlyratherthanfamiliesthatsortintodiffering typesofcare.Thisservestoboostresponseratesbutalsonarrowsvariabilityamongfamilies.Center- 4 WedecidedtousetheECERS,ratherthantheITERS,assumingthatmostchildrenwouldbeattheupperendofourage distribution.Thisprovedtobetrue:two-thirdsoftheparticipatingchildrenenrolledincenterswere24monthsorolder.And sinceoursamplewassplitbetweenchildrenincenterorhome-basedcare,wedidnotwanttointroduceanotherqualitymeasure forthissmallgroupofyoungtoddlers. B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 513 focuseddesignsobviouslyignorehome-basedsettings.Inaddition,thefactorsdrivingselectionintochild caretypes,orsettingsofvariablequality,cannotbestudied. Bystartingwithfamilies,ratherthanchildcaresettings,wecanmodelpredictorsofselectedquality, thenusethisinformationtoadjustforpossiblebiasduetolimited,non-randomaccesstocenterorhome settings.Westudiedseveralfactors,throughmultivariateregression,thathelpedtoexplainthequality ofcenterorhome-basedcare.Afteridentifyingtheseselectionfactors,wethenestimatethequalityof centerandhomecareforourentirefamilysample,andcomparethesepredictedqualitylevelsagainst observedlevels.<br><br> Maternal-levelfactorsincludemother 9sage,ethnicity,schoolattainment(dichotomouslycodedas receivingahighschooldiplomaornot),mother 9sPeabodyPictureVocabularyTest(PPVT)score(per- centilescore, Dunn&Dunn,1997 ),employmentstatus(currentlyworking,regularhoursorodd-hours), andfrequencywithwhichthemotherreadswiththechild,ameasurefromtheHOMEInventory ( Bradley,1993 ). 5 Thesedatawereobtainedduringourinitialinterviewin1998,withtheexception ofthePPVTadministrationwhichoccurredduringourhomevisitduringwave2datacollectionin 2000. Dataonseveralcontextualfactorswereavailabletomakeourselectionmodelmorecomplete,including theresearchsite(codedasdummyvariableswithTampaasthebase)andthenumberofcenter-based enrollmentslotsoperatingineachmother 9scensustractin1998.Thesedatawerecollectedfromstateand localagenciesthatlicenseorregistercenters.Wealsoexaminedthepossiblein uenceoffamilyincome, ethniccomposition,andmeanschoolattainmentlevelswithinsampledmothers 9respectivecensustracts, butthesedemographicsfeatureswerenotrelatedtoqualityofcareselected.<br><br> 2.4.Sharpeningqualitymeasures 4principalcomponentsanalysis Wefactoranalyzedthestandardinstruments 3includingECERS,FDCRS,ArnettandC-COSscales 3 toassessunderlyingconstructswhichcouldbeidenti ed.Forexample,theprincipalcomponentsanalysis identi edtwodimensionsacrosstheArnettscalesforoursampleofproviders(withEigenvaluesgreater than1).The rstdimensionappearedtocapturetheprovider 9spositiveaffectandresponsivenessto children(11items,alpha=.89).Theseconddimension,comprisedofthreeitems,wasrelatedtothe provider 9sexplanationsandtalkwiththechild,forinstance,explainingtotheyoungsterwhyhisorher misbehaviorwasnotappropriate(5items,alpha=.74).Thisconstructprovedtoeffectivelydiscriminate differentchildcaresettings. TheprincipalcomponentsanalysisoftheECERSscalesrevealedtwodistinctfactors:eightitemsrelated tostructuredlearningorplayactivities(alpha=.98),andseveralremainingitemsrelatedtofacilitiesand furnishingsintheclassroomoroutdoorplayarea(7items,alpha=.97).Theformercompositeprovedto bemorerobustwhenestimatingqualitylevelsofcentersselectedbymothers.Similarly,theFDCRSscales splitintothestructured-activitiesdimension(11items,alpha=.98)andintoitemsrelatedtodisciplineand socialization(5items,alpha=.94).Again,theformercomposite(alpha=.98)displayedhighinter-item 5 Otherhomepractices,includingpro-developmentalactivitiesfromtheHOMEinventory,suchasreadingwithone 9schild, werenotrelatedtoqualityofcareselected.Avarietyofsocialsupportmeasuresalsoheldnosigni cantrelationshiptothe qualityofcareselected.Weexaminedthetimespentinthesocialwelfaresystem(monthsoncashaid)andinvolvementwith otherfamilysupportprograms,butthesefactorswerenotrelatedtoqualityselection.Theyweredroppedfrom nalselection models,detailedbelow. 514 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 reliabilityandpredictivevalidityinestimatingqualitylevelsselectedbymothers.Themostinternally reliableC-COSitemsrelatedtothefrequencyofinteractionbetweenproviderandchild,includingthe provider 9spropensitytoinvitethechild 9sverbalization,andengagementofthechildingroupactivities, asopposedtowanderingaboutunengagedinanyidenti ableactivityorsocialinteraction(7itemsin total,alpha=.90).<br><br> 6 3.Findings 3.1.Dataanalysisplan Ouranalysisinvolvedthreesteps.Weidenti eddistinctdimensionsofqualitygaugedbyseveralscales, andstudiedtheextenttowhichthesemultiplemeasureswereinterrelated.Next,weexaminedmean differencesqualitylevelsobservedincentersversushome-basedsettingsacrossthemultipleindicators. Wehaveseparatedhome-basedsettingsbetweenFCCHsandkithorkinarrangementsinallbutonecase, asdetailedbelow.Finally,weanalyzedhowcontextualfactorsandfamily-levelfactorswererelatedto thequalityofcarethatmothersselected. 3.2.Interrelationshipsamongthequalitymeasures Thequalitymeasureswereinterrelatedinseveralinstances.Detailsofthiscorrelationalanalysisappear in AppendixA .Whenobservinginsidecenters,theECERSwasrelatedtobasicindicatorsofsocial interaction:thepropensityoftheprovidertoengagethechildintalkandlessdisengagementbythefocal childinclassroomactivities(bothfromtheC-COS).Inaddition,focalchildrenweremoreconsistently engagedwhenteachersdisplayedahigherpropensitytoreasonandexplainactionstochildren.Teacher reasoningwasnegativelyrelatedtothesimplecountofteacher 3childinteractions,suggestingthatless adulttalkwasrequiredforgroupmanagementordiscipline.Notethatreportedassociationsaresigni cant at p <.05orstronger.<br><br> Social-interactionmeasuresembeddedintheC-COSbehaveddifferentlyrelativetotheotherquality measuresusedwithinFCCHandkithorkinsettings.DespitethelowercountofFCCHsselectedby mothers,comparedtocenterorkithandkinproviders,thefrequencyofverbalinteractionbetweenthe providerandfocalchild,aswellasthepropensityoftheprovidertoaskquestions,wassigni cantly relatedtothesensitivityandresponsivenessdimensionoftheArnett.Withinkithandkinsettings,the measurestendedtobemorehighlyinterrelated:thefrequencyofprovider 3childinteraction(C-COS) wassigni cantlycorrelatedwiththeprovider 9seducationlevel;thereasoningdimensionoftheArnett wasnegativelyrelatedtopropensityofthefocalchildtobewatchingtelevision;frequencyofprovider questionstothechildwasassociatedwiththeoverallFDCRSscore(again,eachcorrelationwith p <.05). Overall,weobservedsomeassociationsbetweenselectedmeasuresofquality.Yetwiththeex- ceptionofkithandkinsettings,theArnett,ECERSandFDCRS,andC-COSwerenotalways interrelated. AppendixB providescorrelationsforallmeasures,splitforcenterandhome-based providers.<br><br> 6 Detailsontheprincipalcomponentsanalysesandexactscalescontributingtothecompositesareavailablefromtheauthors. B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 515 Table3 Observedchildcarequalityforcenterandhome-basedsettings(meansandS.D.sreported a ) SanFranciscoSanJoseTampaManchester andNew Haven F -value Centers( n =166) n =28 n =30 n =97 n =11 EarlyChildhoodEnvironmentRating Scale(ECERS) 4.6(1.1)5.8(.8)3.2(1.4)2.3(1.2)40 . 90 *** ArnettSocialInteractionScale3.0(.8)3.4(.4)2.7(.7)2.6(.7)9 .<br><br> 81 *** Caregivercompletedhighschool(%)97(18)94(25)88(32)73(47)1 . 97 CaregiverhadcollegeunitsinECE(%)57(50)64(49)13(34)56(53)17 . 62 *** Childreninsetting(groupsize)15.6(7.3)14.5(6.1)11.8(7.2) 34 .<br><br> 18 * Ratioofchildrenperadultcaregiver4.5(2.1)5.9(3.6)7.9(5.1) 37 . 38 *** FamilyChildCareHome( n =69) n =13 n =42 n =9 n =5 FamilyDayCareEnvironmentRating Scale(FDCRS) 3.0(1.7)2.8(1.3)3.8(1.9) 31 . 82 ArnettSocialInteractionScale2.7(.9)3.0(.7)3.1(.6) 3 .<br><br> 63 Caregivercompletedhighschool(%)86(36)67(48)50(53) 31 . 79 CaregiverhadcollegeunitsinECE(%)38(51)11(32)11(33) 32 . 88 + Childreninsetting(groupsize)6.5(3.7)3.6(2.8)4.1(3.4) 34 .<br><br> 82 * Ratioofchildrenperadultcaregiver2.3(1.4)2.1(1.5)2.4(1.5) 3 . 48 KithandKin( n =118) n =24 n =35 n =15 n =44 FamilyDayCareEnvironmentRating Scale(FDCRS) 2.6(1.0)2.8(1.1)2.5(1.1)2.5(1.0) . 66 ArnettSocialInteractionScale2.9(.7)3.0(.6)2.2(.7)3.0(.6)6 .<br><br> 50 ** Caregivercompletedhighschool(%)73(45)41(50)47(51)71(46)3 . 90 * CaregiverhadcollegeunitsinECE(%)0(0)6(24)7(25)4(21) . 47 Childreninsetting(groupsize)2.3(1.6)2.9(2.4)2.3(1.5) 3 .<br><br> 67 Ratioofchildrenperadultcaregiver1.6(.9)2.0(1.6)2.3(1.5)3.4(2.1)4 . 58 ** a MeanitemscoresreportedforECERS,FDCRS,andArnettmeasures. + p <.1.<br><br> 7 p <.05. 7 7 p <.01. 7 7 7 p <.001.<br><br> 3.3.Qualitydifferencesbychildcaretypeandcity Table3 reportsmeanscoresforstructuralandprocessmeasuresofqualitybysiteandprovidertype. Looking rstattheindicatorsof center quality,largebetween-sitedifferenceswereobservedforaverage ECERSscores( p <.001).Forexample,theaverageECERSscoreamongSanJosecentersequaled5.8, comparedtojust3.2inTampa(ECERSmaximumequals7).CenterqualityinbothCaliforniasites comparedfavorablytocentersobservedbytheCost,Quality,andOutcomesstudyteam,sampledfroma widervarietyofcommunities( CQO,1995 ). ForConnecticutwecombinedscoresforManchesterandNewHavenproviders.Ourdecisionrulewas notreportmeansforlessthan10settings,sincegeneralizingtositeswouldnotbevalid.Fewerthan10 participatingmothersinConnecticutselectedalicensedFCCH,sonomeansarereported.<br><br> 516 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 Arnettscoresfollowedasimilarpattern,butthedifferenceswerenotaswide,rangingfrom3.4inSan Josecentersto2.4inManchester(maximumscoreequals4, p <.001).Arnettscoresrangedlowerinthe 11Connecticutcenterstowhichwegainedaccess,butthevariabilitywasnotasgreatasthatobserved acrossECERSscores. Staf ngratiosweresigni cantlyhigherinTampacenters,averaging7.9childrenperadult,compared tojust4.5 31inSanFranciscocenters( p <.001).Again,theCaliforniacenterscomparedfavorablyto earlierobservationalstudies.Forexample,theNICHDstudyofearlychildcareobserved7.0childrenper staffmemberinclassgroupswith3-year-olds,and5.2 31forgroupswith4-year-olds( NICHD,1997 ).The AmericanAcademyofPediatricsrecommendsratiosof7to1for3-year-oldsand4to1for2-year-olds. ThesmallestclassgroupswereobservedinTampa,equaling11.8children,comparedto15.6inSan Franciscocenters( p <.05).Tampacentersreliedlessonclassroomaideswhichaccountsforwhythese centershadthehigheststaf ngratiowithsomewhatsmallerclassgroups.<br><br> Over80%ofleadteachersineachcityreportedtheyhadcompletedhighschool,exceptamongthefew centersobservedinManchester.Theshareofcenterteacherswithcollege-levelunitsinearlychildhood education(ECE)waslowerforstaffinTampacenters. Cautioniswarrantedininterpretingthe ndingsfor familychildcarehomes (FCCHs),sinceSan Josemothersdisproportionatelymadeupthegroupselectingthistypeofcare.Structuralmeasuresdid signi cantlydifferacrosssites.AveragegroupsizeinFCCHswassigni cantlyhigherinSanFrancisco, 6.5childrenonaverage,comparedtoSanJose(3.6)andTampa(4.1, p <.05).Staf ngratiosre ected thisdifferencebutwerenotstatisticallysigni cant.MeanFDCRSscoresoverallwerequitesimilarto averagelevelsobservedby Kontosetal.(1994) fortheirsampleoflowerincomefamilies. Thecharacterandqualityof kithandkin providersvariedamongsites,primarilyintermsofschool attainmentandchild 3staffratios.EducationlevelswereconsiderablyhigherinManchesterandSan Francisco,where89%and73%hadcompletedhighschool,respectively,comparedtojust41%inSanJose ( p <.05).ThispatternwasnotconsistentlyassociatedwithArnettscores,withSanFranciscocaregivers scoringrelativelyhighat2.9,alongwithSanJosecaregivers,despitetheirloweducationlevels,scoring at2.9,aswell.Incontrast,TampacaregiverswerepoorlyeducatedanddisplayedlowArnettscores, averaging2.2onthesefour-pointscales( p <.01).<br><br> 3.4.Explainingthequalityofcareselected Intotal,weregressednineseparatequalityindicatorsoncontextualandmaternal-levelfactorsthatmay contributetothequalityof centers selectedbymothers.Forcenters,thisincludedthetotalECERSscore, thetwoECERSdimensions(describedinthemethodssection),thetwoArnettdimensions,andthethree C-COSfactorsrelatedtosocial-interactionpatterns.Wealsoestimatedthechild 3staffratiofromthesame selectionfactors.Allpredictorswereenteredsimultaneously.Weexperimentedwithenteringdiffering blocksofpredictors,butthebasicmultivariatestoryisquiteclearwiththesimple,onestepprocedure. Dummyvariablesforsiteswerecreatedinastandardmanner,avalueof1indicatesthefamilylivedin thatparticularcityandazeroindicatesthatthefamilydidnotresideinthatsite.Tampaistheexcluded siteinallregressionestimations. Fromamongthesesevenregressions,fourexplainedatleast10%ofthevarianceinthequalityoutcome (adjusted r 2 ).Thesamenumberofregressionestimationswerestudiedfor home-based care,including andthenexcludingFCCHs.Fiveofnineregressionsaccountedforatleast10%ofthevarianceinthe qualityindicator.<br><br> B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 517 Table4 Estimatingthequalityofcentersselectedbymothers( ² coef cientsandunstandardizedS.E.sreported) ECERSaverage score ECERSfactor: structured learning activities Arnettfactor: provider 9s verbal explanations C-COSfactor:focal childinteraction withcaregiver NewHaven 2 . 63(.67) 2 . 79(.77) .<br><br> 31(.42)1 . 99(1.94) Manchester 2 2 . 05(.81) * 2 2 .<br><br> 71(.93) ** . 06(.52)1 . 60(2.11) SanFrancisco1 .<br><br> 43(.36) *** 1 . 68(.41) *** . 58(.23) * 4 .<br><br> 39(.88) *** SanJose2 . 36(.35) *** 2 . 32(.40) *** 1 .<br><br> 18(.24) *** 3 . 46(.90) *** Capacity4 . 3e 2 5(3.1e 2 4)1 .<br><br> 2e 2 4(3.6e 2 4) 2 1 . 3e 2 4(2.1e 2 4)4 . 5e 2 4(8.2e 2 4) Latina 2 .<br><br> 16(.32) 2 . 28(.37) 2 . 02(.21) 2 .<br><br> 65(.83) AfricanAmerican 2 . 25(.30) 2 . 45(.35) 2 .<br><br> 01(.19) 2 1 . 15(.78) AsianAmerican 2 . 03(.61) .<br><br> 09(.69) 2 . 63(.38) 2 1 . 45(1.56) Mother 9sschoolattainment: didnotcompletehigh school .<br><br> 18(.23) . 27(.26) 2 . 18(.16) 2 .<br><br> 38(.61) Mother 9sage . 01(.02) . 003(.02) 2 .<br><br> 02(.01) 2 . 03(.04) Frequencyofreadingsto child . 04(.12) 2 .<br><br> 03(.14) 2 . 10(.01) 2 . 16(.32) Mother 9sweeklyworking hours .<br><br> 003(.01) . 004(.01) . 001(.004) .<br><br> 02(.01) Mother 9sweeklyworking hours(missingdummy) 2 . 26(.76) 2 . 23(.86) .<br><br> 93(.51) + 2 . 34(2.1) Mother 9sworkingon regulardaytimeschedule . 16(.25) .<br><br> 29(.28) . 19(.16) 2 . 05(.64) Mother 9sworkingon regulardaytimeschedule (missingdummy) .<br><br> 65(.77) . 65(.88) 2 . 65(.51)1 .<br><br> 17(2.1) Mother 9sPPVTscore . 04(.02) + . 03(.24)3 .<br><br> 4e 2 4(.01) . 005(.05) Mother 9sPPVTscore (missingdummy) 3 . 11(1.82) + 3 .<br><br> 0(2.1) . 10(1.13) . 44(4.59) Intercept 2 .<br><br> 34 . 343 . 72 .<br><br> 28 F statistic6 . 17 *** 5 . 56 *** 2 .<br><br> 77 *** 2 . 89 *** N ofcases140140131145 Adjusted r 2 . 39 .<br><br> 36 . 19 . 18 + p <.10,forcorresponding t -statistics.<br><br> 7 p <.05,forcorresponding t -statistics. 7 7 p <.01,forcorresponding t -statistics. 7 7 7 p <.001,forcorresponding t -statistics.<br><br> Table4 reportstheestimationmodelsformothersselectingcenters.Similartothedescriptivestatistics above,thecityinwhichthemotherresidedmakesasizeabledifferenceinestimatingthetotalECERS score,theECERSdimensionrelatedtostructuredlearningactivities,thereasoningandexplanation dimensionoftheArnett,andfrequencyofverbalinteractionbetweenthefocalchildandprovider. MotherswithhigherPPVTscorestendedtoselectcenterswithslightlyhigherECERSscores.Mothers employedformorehoursweeklytendedtoselectcenterswheretheobservedteacherscoredhigheronthe 518 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 explanationsdimensionoftheArnett( p <.06).Overall,themother 9scityofresidenceismorestrongly relatedtocenterqualitythanmaternalorfamily-levelattributes.Thissuggeststhatinstitutionalforces maymorestronglyshapethequalityofcenterslocallyavailablethanhousehold-levelfactors,atleast amongtheselectiondeterminantsthatwemeasured. Thesesameselectionmodelsbehavedifferentlywhenestimatingthequalityofall home-based providers,includingFCCHandkithandkinproviders.Wedisplayin Table5 theregressionmodels whichincludeadummyvariableindicatingwhethertheproviderwasanFCCH(valuecoded,1)orakith orkinprovider(coded,0).FDCRSscoresrangedhigherinthetwoConnecticutsites,withtheFCCH dummyvariableinthemodel.FCCHs,onaverage,displayedhigherFDCRSscoresonthestructured learningactivitiesdimension,comparedtokithandkinarrangements.<br><br> WhenestimatingmeanArnettscores,weseethatAsianAmericanmothers(disproportion- atelyrepresentedintheSanJosesite)selectedhome-basedprovidersthatscoredone-halfpoint loweronthefour-pointArnettScale.ThismaybelinkedtolowereducationlevelsamongViet- nameseAmericanmothers,andpresumablytheirkithandkinmembers,relativetootherethnic groups. Weestimatedtheprovider 9seducationlevel 3foreitherthecenterteacherorhome-basedprovider 3 withalogisticregressionthatreportsonthelikelihoodofcompletinghighschool( nalcolumn, Table5 ). SanFranciscoprovidersweremorelikelytohavecompletedhighschool,relativetoTampahome-based providers( p <.05).Importantly,FCCHsprovidersweremorelikelytohavecompletedhighschool, comparedtokithandkinhomeproviders( p <.01).Motherswhoworkedmorehoursperweektendedto selectlesswelleducatedproviders.Itmaybethatmotherswholaborformorehoursperweekmustrely moreheavilyonkithorkin, ndingonlypart-daycenterarrangements,andtendtospendlesstimewith theirchild(suggestedby Fuller,Kagan,&Loeb,2002 ).<br><br> Finally,weranidenticalregressionestimatesofqualityselectedjustformotherswhoselectedakith orkinprovider,droppingtheFCCHcases.LatinaandBlackmothersselectedkithorkincaregivers withlowerFDCRSscoresandfewerstructuredlearningactivities,butArnettscoreswerenodifferent, comparedtoWhitemothersandtheirselectedkithorkinproviders.Arnettscoreswerelowerwithinkith andkinsettingsselectedbymotherswhowereemployedforlongerhourseachweek(allcorresponding coef cientssigni cantat p <.05). 3.5.Estimatingqualitylevelsfortheentirefamilysample Movingfromtheparametersidenti edintheseselectionmodels,wethenestimatedmeanquality levelsinasecondsetofregressionsafterincludingallmothersinterviewed,independentofwhetherwe hadgainedaccesstotheirchildcareprovider.Forexample,weobtainedcompleteinterviewdataon200 motherswhoreportedusingcentersbutobserved166ofthesecenters.Similarly,completeinterviewdata formothersselectingakithorkinproviderwereavailablefor293mothers,eventhoughweobservedjust 118oftheirproviders. Table6 summarizestheresultsfromthissecondsetofregressions,afterestimatingmeanqualitylevels forthecompletesetoffamiliesusingeitheracenterorkithorkinprovider.ThemeanECERSscale score,forinstance,equaled3.9amongthe166observedcenters.Buttheestimatedmeanisjust3.4after includingall200familiesintheselectionmodel( p <.01).Thepatternissimilarforthestructuredlearning activitiesdimensionoftheECERS,suggestingthatobservedcentersareofhigherqualitythanthebroader rangeofcentersselectedbytheentirefamilysample.However,nosigni cantdifferencesweredetected B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 519 Table5 Estimatingthequalityofhome-basedprovidersselectedbymothers,includingfamilychildcarehomes( ² coef cientsandunstandardizedS.E.sreported) FDCRSmean scalescore FDCRSfactor: structured learning activities Arnettmean scalescore Child 3adult staf ng ratio Caregiver 9s school attainment NewHaven 2 .<br><br> 40(.38) 2 . 32(.39) . 62(.23) ** 2 .<br><br> 24(.60) *** 1 . 52(.72) * Manchester 2 . 55(.57) 2 .<br><br> 43(.58) . 47(.44) . 90(1.01) 3 SanFrancisco 2 .<br><br> 34(.36) 2 . 24(.37) . 42(.22) + 2 .<br><br> 36(.47)1 . 64(.71) * SanJose 2 . 36(.35) 2 .<br><br> 28(.35) . 42(.22) + . 07(.45) .<br><br> 27(.63) Capacity 2 2 . 1e 2 4(2.3e 2 4) 2 3 . 1e 2 4(2.3e 2 4) 2 2 .<br><br> 7e 2 4(1.4e 2 4) + 2 2 . 2e 2 4(3.2e 2 4) 2 1 . 1e 2 4(4e 2 4) Familychildcarehome .<br><br> 48(.21) * . 63(2.92) ** . 16(.13) .<br><br> 50(.28) + 1 . 13(.43) ** Latina 2 . 16(.28) 2 1 .<br><br> 17(.29) 2 . 30(.18) + . 46(.38) 2 .<br><br> 38(.54) AfricanAmerican 2 . 30(.25) 2 . 36(.25) 2 .<br><br> 19(.17) 2 . 02(.36) 2 . 12(.49) AsianAmerican 2 .<br><br> 18(.38) 2 . 00(.38) 2 . 50(.25) * 2 .<br><br> 10(.51) . 17(.78) Mother 9sschoolattainment:did notcompletehighschool . 05(.20) .<br><br> 03(.20) . 17(.12) 2 . 16(.28) .<br><br> 04(.38) Mother 9sage 2 . 02(.02) 2 . 01(.02) .<br><br> 001(.01) . 01(.02) . 03(.03) Frequencyofreadingstochild 2 .<br><br> 13(.10) 2 . 11(.10) 2 . 14(.07) * 2 .<br><br> 07(.14) 2 . 47(.81) * Mother 9sweeklyworkinghours 2 . 004(.01) 2 .<br><br> 01(.01) 2 . 01(.003) + . 003(.01) 2 .<br><br> 02(.01) * Mother 9sweeklyworkinghours (missingdummy) . 08(.38) 2 . 02(.39) .<br><br> 34(.24) 2 . 11(.60) 2 . 67(.77) Mother 9sworkingonregular daytimeschedule .<br><br> 02(.24) 2 . 002(.25) . 08(.16) 2 .<br><br> 35(.32) 2 . 26(.46) Mother 9sworkingonregular daytimeschedule(missing dummy) . 07(.40) .<br><br> 09(.40) 2 . 15(.25) 2 . 48(.61) 2 .<br><br> 39(.81) Mother 9sPPVTscore . 01(.01) . 01(.01) .<br><br> 01(.01) . 03(.02) . 02(.03) Mother 9sPPVTscore(missing dummy) .<br><br> 70(1.28) . 97(1.30) . 76(.85)1 .<br><br> 90(1.87)1 . 89(2.54) Intercept3 . 242 .<br><br> 672 . 33 2 . 44 2 .<br><br> 38 F statistic . 841 . 182 .<br><br> 05 * 2 . 20 ** 36 . 9(chi-square) ** N ofcases166166127142169 Adjusted r 2 .<br><br> 09 . 13 . 13 .<br><br> 13 . 17 + p <.10,forcorresponding t -statistics. 7 p <.05,forcorresponding t -statistics.<br><br> 7 7 p <.01,forcorresponding t -statistics. 7 7 7 p <.001,forcorresponding t -statistics. 520 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 Table6 Qualitylevelsforobservedprovidersandestimatedforallprovidersselectedbymothers(meansandS.D.sreported) QualitymeasureObservedprovidersEstimatedforallreportedproviders T -value Centers n =166 n =200 ECERS,meanscalescore3.9(1.65)3.4(1.28)2.83 ** ECERS,structuredlearningactivities4.2(1.83)3.7(1.44)2.93 *** Arnett,verbalexplanations3.3(.94)3.4(.53)n.s.<br><br> C-COSchild 3caregiverInteraction2.9(3.69)2.9(1.78)n.s. Kithandkinsettings n =118 n =293 FDCRS,meanscalescore2.6(1.05)2.5(.37)n.s. Arnett,meanscalescore2.9(.70)2.9(.39)n.s.<br><br> Child 3staffratio2.2(1.62)2.7(1.12)2.76 ** 7 7 p <.01. 7 7 7 p <.005. fortheArnettreasoningandexplanationdimension,norforthechild 3providerinteractiondimensionof theC-COS.<br><br> 7 DifferenceswerenotobservedfortheFDCRSmeanscore,norfortheArnett,amongobservedkithand kinversusestimatedlevelsfortheentirefamilysample.However,thechild 3staffratiowassigni cantly lowerintheobservedsettings,comparedtotheestimatedratioforallkithandkinsettings.Again,those providerswhoallowedour eldstafftovisittendedtobeofhigherqualitythanthoseprovidersormothers whodeniedaccess. 4.Discussion Returningtoourcorequestions,we rstemphasizethatthesemothersselectedchildcareprovidersof widelyvaryingquality.Forthosewhoselected centers ,indicatorsofqualityvariedsigni cantlyacross the vecities.ThiswasparticularlytruefortheECERSandArnettscoreswhichcorrespondedtothe teacher 9seducationlevelandtheclassroom 9schild 3staffratio.Overall,qualitylevelswerenotimpressive, exceptforcentersinSanFranciscoandSanJosewhereECERSscoresandteachers 9schoolattainment werequitehigh,atleastcomparedtoearlierstudiesconductedinlowormiddle-incomecommunities. Familychildcarehomes displayedsigni cantlyhigherqualityonmultiplemeasures,comparedto kith andkinsettings ,includingFDCRSscoresandprovidereducationlevels.Importantly,qualitylevelsof FCCHs,whileranginglowerthancenters,werelessvariableacrosssites.Whetherinstitutionalforces areweakerthanmarketforces,suchasFCCHregulatorystandardsorsubsidylevels,remainsanopen question.Cautioniswarrantedingeneralizingfromthese ndings,sincemanyofthemothersselecting FCCHsresidedinSanJose.InCalifornia,highercenterqualitymaybetheresultofhigherquality standards,reimbursementlevels,orstrongerstaffdevelopmentefforts,comparedtoConnecticutand Florida,butsuchpoliciesdid not yieldsimilarbetween-sitedifferencesinFCCHquality.<br><br> 7 Again,thesearethequalityindicatorsforwhichwecouldexplainatleast0.10ofthevariance,basedonthequalityselection models. B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 521 KithandkinprovidersscoredquitelowontheFDCRS.Wedonotknowwhetherthismeasure,designed fororganizedFCCHs,holdspredictivevaliditywhenusedintheselessformalsettings.ArnettScalescores werenotsigni cantlylowerinkithorkinsettings,comparedtocentersorFCCHs,suggestingthatsocial relationsvaryalongdifferentdimensions,comparedtolevelsofmaterials,structuredtasks,andfacilities. TheshareofkithandkinproviderswhohadcompletedhighschoolorECEunitswasconsiderablylower, comparedtocenterteachersorFCCHproviders.Thismayaccountforsomewhathighercognitivegrowth observedforchildrenattendinghigherqualityFCCHs,comparedtokithandkinarrangements( Loebet al.,2004 ).<br><br> 4.1.Institutionalcontextandfamily-levelselectionfactors Variationinthequalityofcentersselectedwasstronglyrelatedtothemother 9scityofresidence.This suggeststhatstatesorlocalesdifferintheircapacitytoadvancecenterqualityovertime,ortheirability towidenpoorfamilies 9accesstohigherqualitycenters.Mothersworkingmorehoursperweektendedto selectlowerqualityproviders,andAsianAmerican(mainlyVietnamese)mothersselectedhome-based providerswhoscoredlowerontheArnettScale,comparedtoprovidersselectedbyotherethnicgroups. Otherwise,individualandfamily-levelfactorswereweakinexplainingthequalityofcareselected, particularlyrelativetositeeffects.Thissuggeststhatforlow-incomefamilies,institutionalmechanisms thatexpandcentersupplyorregulatequalitymoreeffectivelyplayaforcefulrolerelativetofamily-level selectionfactors. Wefoundthatorganizationalmeasuresofcenters,includingmanyitemsontheECERS,were moderatelyrelatedtosocial-interactionmeasures.Childrenweremoreengagedinactivities(rather thanwanderingunoccupied,asrecordedwiththeC-COS)whenattendingcentersthatdisplayed higherECERSandArnettscores.ForFCCHproviders,theC-COSinteractionmeasures,especially theamountofverbalinteractionbetweenchildandcaregiver,weresigni cantlyrelatedtoArnett scores.<br><br> Still,moreworkisrequiredtounderstandhowtheorganizationalstructureofsettings 3includingthear- rangementoflearningtasksandadultlanguageopportunities 3isrelatedtothecharacterofchild 3caregiver interactionandhowthesedimensionsofqualitymayplayoutdifferentlyincentersversushomesettings. Onbalance,thewaysinwhichchildrenweresociallyengagedincenterswithbettereducatedteachersap- peartobelinkedtothesupplyofmaterialsandsomedegreeofformalizationobservedwithinclassrooms (asgaugedbytheECERS). Overall,these ndingssuggestthatthe eldmightre ectonhowweconceiveofchildcareselection.<br><br> Our ndingthatmaternalandfamily-levelfactorsheldlittlepredictivepowerinestimatingthequality ofcareselected 3butcityofresidenceheldastrongrelationship 3callsoutformoreattentiontolocal context,includinghowstateandlocalagenciesshapetheavailabilityofcenterprogramsinlow-income communities.Ourownearlierworkadoptedfamily-leveldemandmodels,oftenfromeconomists,in estimatingselectionpatterns.Butiflocalsupply,regulations,andsubsidy ows(forlow-incomefamilies) areatwork,weriskoveremphasizingtheroleoffamily-levelselectionfactors. 4.2.Policyimplications Threebasicpolicyimplicationsemergefromthese ndings.First,lowqualitylevelsobservedamong manyhome-basedsettings 3includingcarebypoorlyeducatedadults 3isworrisome.Risingappropri- 522 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 ationsforchildcare,witnessedoverthepastdecade,havegreatlyexpandedtheavailabilityofparental vouchers,oftenreimbursingkithandkinmembersfortheirservices.Whileperhapsasensibleincome- supportpolicy,thisstrategyappearstolegitimateandsupportlowqualitycareinmanyinstances.The opportunitycostishigh:thesepublicdollarsaredivertedawayfromcentersandthepossibilityofstrength- eningthecenterandpreschoolinfrastructure. Onecountervailingforceisthatmanylow-incomewomenworkatnightoronweekendswhenfew centerscanaffordtoremainopen.Unlessthe nancingofcentersimproves,home-basedproviderswill remaintheonlyoptionformillionsoflow-wageworkers.<br><br> Second,theconsistent ndingthatmothersemployedforlongerhourseachweekselectedlowerquality home-basedprovidersistroublingaswell,especiallyinlightofrecurringpoliticalpressuretolengthen theworkweekforwomenreceivingpublicassistance.Earlierresearchonwelfarereformexperiments, particularlytheMinnesotaprogram,foundthatchildren 9senvironmentsanddevelopmentaloutcomes canimprovewhenmothersareallowedtoworklessthanfulltime( Zaslowetal.,2002 ).Onemediator accountingforthesepositiveeffectsmaybethatmothersarebetterableto ndhigherqualityhome-based caregiverswhoprovidelessthanfull-timecare. Third,thesharpdifferencesincenterqualityacrosscitiespromptsquestionsoverwhatspeci cin- stitutionalmechanisms 3subsidy ows,regulatorymechanisms,orprofessionaldevelopment 3areef- fectivelyraisingqualityinsomeplacesbutnotothers.Someworksuggeststhatstatemonitoringac- tivityismoreef caciousthatsimplyraisingregulatorystandards( Blau,2001 ).Lessisknownabout howlocalaction,especiallyprofessionaldevelopmentandinfrastructuregains,canadvancecenter quality. Indeed,statesandlocalgovernmentsshouldmovecarefullyonregulatoryfronts.Wefoundthatteacher orprovidereducationlevels 3atthelowend,suchasamongprovidersinTampa 3wererelatedtolower qualityonothergauges.Butwealsofoundthatpositivesocial-interactionmeasures,suchasthosegauged bytheArnett,are not necessarilyrelatedtoproviders 9formaleducation,nortostructuralmeasuresof quality(asassessedbytheECERS).Statesandcounties 3especiallyinlightofthecurrentpushfor universalpreschool 3may ndthemselvesregulatingoneasilymeasuredelementsof cquality dwhich arenotempiricallyrelatedtochildren 9sdevelopment.Ontheotherhand,social-interactionmeasuresare predictiveofdevelopmentaltrajectoriesinsomepoorcommunities(e.g., Loebetal.,2004 )butdif cultto translateintoregulations.Policymakersandprofessionalassociationsshoulddeterminewhethercertain indicatorstrulyadvancechildren 9sgrowthandhowtheseindicatorsareinterrelatedtootherfeaturesof quality.Otherwise,governmentwillimplementcostlyqualitystandardsthatholdhighsymbolicvalue butdolittletoadvancechildren 9sgrowth.<br><br> Acknowledgments TheGrowingUpinPovertyProjecthasbeenfundedbytheU.S.departmentsofeducationandhealth andhumanservices,alongwiththeCasey,Packard,MacArthur,andSpencerfoundations.Theproject iscurrentlysupportedbytheWalterandEliseHaasandHaasJr.charitablefunds.Specialthanksare dueMichaelLaracyandMarieYoungfortheirgeneroussupportofthechild-careobservationsubstudy whichinformsthepresentpaper.JudeCarrollandJanMcCarthydirectedthe eldworkinConnecticut andFlorida,respectively. B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 523 AppendixA.Interrelationshipsamongqualitymeasures Whenobservingcenter-basedprogramstwoC-COSscalesprovedtoberelatedtoscoresontheEarly ChildhoodEnvironmentRatingScale(ECERS).Thefrequencywithwhichtheproviderurgedthefocal childtotalkwasmoderatelycorrelatedwiththemeanofallscalescoresdrawnfromtheECERS( r =.24, p <.002).Frequencyissimplyreportedasthenumberofsnapshotsinwhichthebehaviorwasobserved, rangingfrom0to40instances. Inadditionthefrequencywithwhichthefocalchildwasobservedwandering,notengagedinany identi abletaskwasassociatedwiththetotalECERSscore( r = 2 .54, p <.001).Thissamewandering frequencyitemwasnegativelyrelatedtotheseconddimensionidenti edfromtheArnettsubscales, indicatingtheprovider 9spropensitytoexplainmisbehaviorandreasonwiththechildren( r = 2 .40, p <.001).AthirdC-COSitem 3thesimplecountofinteractionsbetweenthefocalchildandthe provider 3wasnegativelyassociatedwiththeprovider 9sreasoningbehaviorontheArnett( r = 2 .23, p <.006).Thismayindicatemorefrequentanddirectivedisciplineofthefocalchildwithincenter settings.<br><br> C-COSitemsbehaveddifferentlyinFCCHsettingsvis- ` a-visotherqualitymeasures.Mostnotably, thefrequencyofverbalinteractionbetweenthefocalchildandtheFCCHproviderwasmoderately relatedtothe rstidenti eddimensionoftheArnett,theitemsthattapintosensitivityandtheaffec- tivelywarmresponsivenessoftheprovider( r =.29, p <.007).ThissamedimensionoftheArnettwas moderatelyassociatedwiththefrequencywithwhichtheproviderinvitedthefocalchildtotalk( r = .22, p <.04). Forindividualkithandkinproviders,someofthesameC-COSscaleswererelatedtostruc- turaldimensionsofquality.Forinstance,thecountofobservedinteractionsbetweenthefocalchild andtheindividualproviderwasassociatedwiththecaregiver 9seducationlevel( r =.25, p <.03). ThefrequencywithwhichthecaregiverinvitedthechildtotalkwasassociatedwiththetotalFD- CRSscore( r =.30, p <.02).Thefrequencywithwhichthefocalchildwaswatchingtelevisionwas negativelyrelatedtothesecondArnettdimension,thecaregiver 9spropensitytoexplainandreason withthechild( r = 2 .30, p <.01).Insum,certainscalesontheC-COSholdfairlyconsistentcon- structvalidityintermsofbeingrelatedtotheArnettscaleand,attimes,totalECERSorFDCRS scores.<br><br> AppendixB See TablesB.1andB.2 . 524 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 TableB.1 Correlationsbetweencenterqualitymeasuresandpredictorsofqualityselectedbymothers 23456789101112131415161718 1.ECERSscore.95 ** .51 ** .20 ** 2 .13 2 .24 ** .21 ** .55 ** .18 * .15 * 2 .14.04.06.10.04 2 .13.01.05 2.ECERSfactor:structuredlearn- ingactivities .45 ** .17 * 2 .14 2 .25 ** .23 ** .49 ** .19 * .14 2 .15.06.09.09.00 2 .10.04.04 3.Arnettfactor.11 2 .07 2 .13.09.43 ** .08.10 2 .03 2 .04 2 .09 2 .08.00 2 .03.05 2 .04 4.C-COS:childengaged.03 2 .01.34 ** .24 ** .26 ** .14 2 .13.09.01.04 2 .05 2 .02 2 .04.01 5.Site:NewHaven 3 3 3.26 ** 2 .03.00.04.06.01.02 2 .29 ** 2 .25 ** .03 6.Site:Manchester 3 3.02 2 .05 2 .04.17 ** .05 2 .03.01 2 .12 * 2 .09.00 7.Site:SanFrancisco 3.13.18 ** 2 .02.08.01.07.00.01 2 .03.02 8.Site:SanJose.07.10 2 .09.11.07.07 2 .06 2 .09.06.08 9.Localenrollmentcapacity.18 ** 2 .09.14.03.08 2 .07 2 .06 2 .11.03 10.Latina 2 .44 ** 2 .17 * .11 2 .05 2 .05 2 .01 2 .09.08 11.AfricanAmerican 2 .21 ** 2 .01 2 .04 2 .01.07.08 2 .04 12.Asian.19 ** .16 * 2 .30 ** 2 .10 2 .04.02 13.Nohighschooldiploma 2 .16 * 2 .17 ** 2 .08 2 .13.03 14.Mother 9sage 2 .15 * 2 .08.01 2 .05 15.Readingfrequencywithchild 2 .15 * .09.08 16.Hoursemployedperweek.45 ** 2 .08 17.Workingregulardayshift.02 18.Mother 9sPPVTscore 7 p <.05. 7 7 p <.01.<br><br> B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 525 TableB.2 Correlationsbetweenhome-basedqualitymeasuresandpredictorsofqualityselectedbymothers 234567891011121314151617181920 1.FDCRSscore.96 ** .42 ** .22 ** .13 2 .09 2 .03 2 .01.04 2 .05.15 * .03 2 .06 2 .02.02.00 2 .12 2 .04 2 .03.03 2.FDCRSfactor:structured learningactivities .38 ** .23 ** .15 2 .10 2 .04 2 .01.06 2 .07.21 ** .04 2 .10.06.02.02 2 .12 2 .06 2 .04.01 3.Arnettfactor.26 ** .04.07.07 2 .01.10 2 .11.06 2 .04 2 .13.10.14.00 2 .10 2 .18 * 2 .11.07 4.Child 3adultstaf ngratio.06.07.13.14 * 2 .14 * 2 .06.11 2 .16 * .03.10.10.06 2 .12 2 .15 * 2 .08.03 5.Caregiver 9sschoolattaitment.30 ** .02 2 .11 2 .12 2 .13.01.02.00 2 .05 2 .01 2 .05 2 .02.05 2 .03 2 .06 6.Site:NewHaven 3 3 3 2 .11 2 .19 ** 2 .23 ** .17 * .05.07 2 .06.13.09 2 .02 2 .01 7.Site:Manchester 3 3.02 2 .06.08 2 .02 2 .09 2 .06.07 2 .04 2 .22 ** 2 .13 2 .06 8.Site:SanFrancisco 3.02 2 .03 2 .02.03 2 .06.09.06.03.04 2 .04.20 ** 9.Site:SanJose.20 ** .25 ** .14 * 2 .26 ** .14 * .00 2 .03 2 .14 * 2 .16 * 2 .03.00 10.Localenrollmentcapacity.04.09 2 .16.08.03 2 .19 * .00 2 .06 2 .01.07 11.FCCH 2 .05 2 .02.05.04 2 .08.05 2 .14.00.09 12.Latina 3 3.05.15 * 2 .11 2 .06 2 .06 2 .01 13.AfricanAmerican 3 2 .10 2 .10.14 * .06.12.01 14.Asian.18 ** .13 2 .35 ** 2 .09 2 .05.00 15.Nohighschooldiploma00 2 .15 * 2 .18 * 2 .16 * .00 16.Mother 9sage 2 .19 ** 2 .04 2 .08.04 17.Readingfrequencywithchild.02.17 * .08 18.Hoursemployedperweek.44 ** 2 .03 19.Workingregulardayshift 20.Mother 9sPPVTscore 2 .02 7 p <.05. 7 7 p<.01 526 B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 References Adams,G.,&Rohacek,M.(2002).Morethanaworksupport?Issuesaroundintegratingchilddevelopmentgoalsintothechild caresubsidysystem. EarlyChildhoodResearchQuarterly , 17 ,418 3440.<br><br> Adams,G.,Snyder,K.,Sandfort,J.(2002). Gettingandretainingchildcareassistance:Howpolicyandpracticein<uence parents 9experiences (Occasionalpaperno.55).Washington,DC:UrbanInstitute. Arnett,J.(1989).Caregiversinday-carecenters:Doestrainingmatter?<br><br> JournalofAppliedDevelopmentalPsychology , 10 , 541 3552. Barnett,W.(1995).Long-termeffectsofearlychildhoodprogramsoncognitiveandschooloutcomes. FutureofChildren , 5 (3), 25 350.<br><br> Blau,D.M.(2001). Thechildcareproblem:Aneconomicanalysis .NewYork:RussellSageFoundation. Bradley,R.(1993).Children 9shomeenvironments,health,behavior,andinterventionefforts:AreviewusingtheHOMEinventory asamarkermeasure.<br><br> Genetic,Social,andGeneralPsychology , 119 ,439 3490. Burchinal,M.(1999).Childcareexperiencesanddevelopmentaloutcomes.InS.Helburn(Ed.),ThesilentcrisisinU.S.child care. AnnalsoftheAmericanAcademyofPoliticalandSocialScience , 563 ,73 397.<br><br> Clarke-Stewart,A.(1999). Characteristicsandqualityofchildcarefortoddlersandpreschoolers .Manuscript,NICHDEarly ChildCareResearchNetwork,Washington,DC. Currie,J.(2001).Earlychildhoodeducationprograms.<br><br> JournalofEconomicPerspectives , 15 ,213 3238. CQOStudyTeam.(1995). Cost,quality,andchildoutcomesinchildcarecenters .Technicalreport,UniversityofColorado, Denver.<br><br> Dunn,L.,&Dunn,L.(1997). PeabodyPictureVocabularyTest (3rded.).CirclePines,MN:AmericanGuidanceService. Fosburg,S.(1982).Familydaycare:Theroleofthesurrogatemother.InL.Laosa&I.Sigel(Eds.), Familiesaslearning environmentsforchildren (pp.223 3260).NewYork:Plenum.<br><br> Fuller,B.,Kagan,S.L.,&Loeb,S.(2002). Newlivesforpoorfamilies?Mothersandyoungchildrenmovethroughwelfare reform .Berkeley:UniversityofCaliforniaandTeachersCollege,ColumbiaUniversity. Fuller,B.,Holloway,S.D.,Bozzi,L.,Burr,E.,Cohen,N.,&Suzuki,S.(2003).Explaininglocalvariationinchildcarequality: StatefundingandregulationinCalifornia.<br><br> EarlyEducationandDevelopment , 14 ,47 366. Fuller,B.,Holloway,S.D.,&Liang,X.(1996).Familyselectionofchild-carecenters:Thein uenceofhouseholdsupport, ethnicity,andparentalpractices. ChildDevelopment , 67 (6),3320 33337.<br><br> Fuller,B.,Raudenbush,S.,Wei,L.,&Holloway,S.D.(1993).Cangovernmentraisechild-carequality?Thein uenceoffamily demand,poverty,andpolicy. EducationalEvaluationandPolicyAnalysis , 15 ,255 3278. Fuller,B.,&Strath,A.(2001).Thechild-careandpreschoolworkforce:Demographics,earnings,andunequaldistribution.<br><br> EducationalEvaluationandPolicyAnalysis , 23 ,37 355. Goelman,H.,&Pence,A.(1987).Someaspectsoftherelationshipsbetweenfamilystructureandchildlanguagedevelopmentin threetypesofdaycare.InD.Peters&S.Kontos(Eds.), Continuityanddiscontinuityinchildcare (pp.129 3146).Norwood, NewJersey:AblexPublishing. Hamre,B.,&Bridges,M.(2004).Berkeley:PolicyAnalysisforCaliforniaEducation,UniversityofCalifornia.<br><br> Harms,T.,Clifford,R.,&Cryer,D.(1997). PublishedECERSandFDCRSscales .NewYork:TeachersCollegePress. Hirshberg,D.(2003).<br><br> ChildcaredemandandsupplyunderCalWORKs:TheearlyimpactsofwelfarereformforCalifornia 9s children,1998 32000 .Berkeley:PolicyAnalysisforCaliforniaEducation,UniversityofCalifornia. Holloway,S.D.,Kagan,S.L.,Fuller,B.,Tsou,L.,&Carroll,J.(2001).Assessingchild-carequalitywithatelephoneinterview. EarlyChildhoodResearchQuarterly , 16 ,165 3189.<br><br> Howes,C.(1997).Children 9sexperiencesincenter-basedchildcareasafunctionofteacherbackgroundandadult-childratio. Merrill-PalmerQuarterly , 43 ,404 3425. Howes,C.,Galinsky,E.,Shinn,M.,Sibley,A.,Abbott-Shim,M.,&McCarthy,J.(1996).<br><br> TheFloridaChildCareQuality ImprovementStudy .NewYork:FamiliesandWorkInstitute. Howes,C.,Whitebook,M.,&Phillips,D.(1992).Teachercharacteristicsandeffectiveteachinginchildcare:Findingsfromthe NationalChildCareStaf ngStudy. ChildandYouthForum , 21 ,399 3414.<br><br> Kontos,S.,Howes,C.,Shinn,M.,&Galinsky,E.(1994). Qualityinfamilychildcareandrelativecare .NewYork:Teachers CollegePress. B.Fulleretal./EarlyChildhoodResearchQuarterly19(2004)505 3527 527 Loeb,S.,Fuller,B.,Kagan,S.L.,&Carrol,B.(2004).Childcareinpoorcommunities:Earlylearningeffectsoftype,quality, andstability.<br><br> ChildDevelopment , 75 ,47 365. NICHDEarlyChildCareResearchNetwork.(1997).Povertyandpatternsofchildcare.InG.J.Duncan(Ed.), Consequences ofgrowinguppoor (pp.100 3131).NewYork:Russe

less