Economics and Human Biology

(1)

Do fast food restaurants surrounding schools affect childhood obesity?

Jebaraj Asirvatham

^a,

*, Michael R. Thomsen

^b

, Rodolfo M. Nayga Jr.

^b

, Anthony Goudie

^c

aDepartmentofAgribusinessEconomics,SouthernIllinoisUniversityCarbondale,IL,UnitedStates

bDepartmentofAgriculturalEconomicsandAgribusiness,UniversityofArkansas,Fayetteville,AR,UnitedStates

cArkansasCenterforHealthImprovementArkansasCenterforHealthImprovement(ACHI),LittleRock,AR,UnitedStates

ARTICLE INFO Articlehistory:

Received23July2018

Receivedinrevisedform3December2018 Accepted30January2019

Availableonline16February2019

JELclassiﬁcation:

I10 I12 J13 L83 Keywords:

Fastfood Childhoodobesity Bodymassindex Schoolfoodenvironment Instrumentalvariableestimation

ABSTRACT

Inthisstudy,weestimatetheeffectoffastfoodenvironmentsurroundingschoolsonchildhoodbody mass index (BMI). We use two methods that arrive at a similar conclusion, but with different implications.Using schooldistancefrom thenearestfederalhighway toinstrumentforrestaurant location,wefindthesurroundingrestaurantstoonlymarginallyaffectastudent’sBMImeasure.The effectsizealsodecreaseswithincreasingradialdistancesfromschool,0.016standarddeviationsatone- thirdofamileand0.0032standarddeviationsatamileradialdistance.Thisindicatesthedecreasing influenceofrestaurantsonachild’sBMIasitsdistancefromschoolincreases.Onasubsetofstudentswho wereexogenouslyassignedtodifferentschoolfoodenvironment,wefind noeffectofthefastfood restaurants.Animportantcontextualaspectisthatnearlyallschoolsinthissampleobservedclosed campuspolicy,whichdoesnotallowstudentstoleavecampusduringlunchhours.

1.Introduction

Rising ratesinchildhood obesityhavebecomeanimportant worldwidehealthandpublicpolicyissue.Eventhoughthis has receivedmoreattentionintheUnitedStates,childhoodobesityisa growingprobleminothercountries,includingthoseinEuropeand Asia.Becausethemajorityofchildrenattendpublicschools,there ismuchinterestinregulatingthefoodenvironmentinschoolsto promotehealthobjectives(Storyetal.,2009;Sharmaetal.,2009).

France,forexample,bannedfoodpromotioninschools,whereas MexicoandIndiahavebannedsalesofsodasandcertainunhealthy foods in schools (Villanueva, 2011; Barquera et al., 2013;

KhandelwalandReddy, 2013).IntheUS, regulationhasmostly focusedonschoolmealsandthosefoodsstockedinschoolvending machines¹ . Addressingchildhood obesityhas becomea policy priority,especiallybecauseobesityduringchildhoodmaycontinue intoadulthood(Serdulaetal.,1993).

Mostpublicpolicyeffortshavefocusedonschoollunchesand otherfoodproductssoldwithin theschool.There isa growing concern,however,thatfastfoodrestaurantsaroundschoolsmight increase caloric intake among children and thereby increase obesity(DavisandCarpenter,2009).Thereareincentivesforfast- foodrestaurants,especiallynationalchains,totargetchildrenfor increasingcurrentsalesandcreatingbrandloyalty.Thisismore apparent among hamburger fast food restaurants, sandwich places,and pizzerias(Austin et al., 2005).Collectively, we will refertothesethreerestauranttypesasfastfoodrestaurants.

Animportantquestionofpublicpolicyrelevanceis toknow whetherthepresenceoffastfoodrestaurantsaroundschoolshave anyeffectonchildhoodobesity.Weseektoidentifytheeffectof fast food restaurants surrounding schools on a child’s BMI.

Estimatingthecausaleffectoffastfoodrestaurantssurrounding schools,however,ischallengingdueto:a)self-selectionoffamilies intoneighborhoods;b)omittedvariablebias;andc)selectionof fastfoodrestaurantsinintoneighborhoodsaccordingtofast-food demand.Self-selectionofparentsduetooccupation,educationor otherreasonscouldcreateneighborhoodsthatcouldbesimilarin their attitude towards health and well-beingof theirchildren.

Omitted variables in the determination of a child’s BMI could includefeaturesoftheschoolfoodenvironment.

We address biases in the estimation using an instrumental variablesapproach,wheretheinstrumentistheproximityofthe Abbreviations:ACHI,arkansascenterforhealthimprovement;BMI,bodymass

index;CDC,centersfordiseasecontrolandprevention;GIS,geographicinformation system;IV,instrumentalvariable(s).

*Corresponding authorat: Jebaraj Asirvatham, SouthernIllinois University, Carbondale,IL62901,USA.Tel.:+6184531709.

E-mailaddress:jebaraj@siu.edu(J.Asirvatham).

1‘Healthy,Hunger-FreeKidsActof2010’[42USC1751].

https://doi.org/10.1016/j.ehb.2019.01.011

ContentslistsavailableatScienceDirect

Economics and Human Biology

j o u r n a l h o m e p a g e : w w w . e l s e vi e r . c o m / l o c a t e/ e h b

(2)

school to US highways that create a demand for fast food restaurantstoservehighwaytravelers.Thisinstrumenthasbeen usedinseveralstudies(forexample,Dunn,2010;Andersonand Matsa,2011).Inadditiontothismethod,weestimateindividual- levelﬁxedeffectsforasubsetofstudentswhowereexogenously reassignedtoanotherschool.Thisstrategyfollowsthatusedby Asirvathametal.(2018)intheiranalysisofpeereffects.Boththese methods arrive at a similar conclusion– fast food restaurants surroundingschoolshaveanegligibleeffectonBMIoutcomesof schoolchildren.Theanalyticsample isbasedonmeasuredBMI records from an ongoing BMI screening program within the Arkansaspublicschoolsystem.

Thereareseveralwaysthatfastfoodrestaurantssurrounding schoolscouldinﬂuenceBMIoutcomes.First,fastfoodrestaurants andpizzeriasoffercalorie-densefoodsthatplaya directrolein obesity.Eventhoughmostschools in thisstudyobserve closed campuses, a high density of restaurants around schools will generally increase thelikelihood that students visit restaurants beforeor after school hourseither unsupervisedwithfriends orwith aparentorcaregiver.Second,restaurantslocatednearbyreduces travel cost to obtain food. It could also be that proximity disincentivizes traveling longer distances to purchase healthier foods.Third,exposuretothesightofrestaurantsmightincreasethe likelihoodthatchildren wouldchoosefastfoodsafterbeingexposed tofastfoodrestaurantsenroutefromschooltohome.Fastfood signagecouldreinforcemarketingmessagesaimedatchildrenand increaselikelihoodoffastfoodrequestsonfood-away-from-home mealoccasions.Schoolmealscanbeservedasearlyas10:30amand manychildrenwillbehungryattheend oftheschoolday.The presenceoffastfoodsinschoolenvironmentcouldthusincrease desireforfastfoodsonotherdiningoccasionsregardlessofwhether thechildconsumesfastfoodonthewaytoorfromschool.Finally, teacherswholeavecampusforlunchcouldbringinitemsfrom restaurants(e.g.,drink cups)and therebymodel fast foodas a reasonable meal choice. This is similar to the argument that advertisementstargetingchildrenmight increasefastfoodcon- sumption(Jashinskyetal.,2017).StudyparticipantsinCambridge- shire,UK,whofacedagreaterexposuretofastfoodoutletsshowed higherintakeoftakeawayfood(Burgoineetal.,2014).

2.Contributiontothecurrentliteratureonfastfoodand obesity

Much of the work in this line of research provides some evidenceofapositiveassociationorcorrelationofobesityratesor BMI with restaurant availability or proximity (Williams et al., 2014).Most of thestudies are correlational and therefore it is difﬁculttoimplyanycausality.Inthissection,wediscussfewofthe studiesthathaveconductedamorerigorousanalysisgoingbeyond correlationorassociation.Thesestudies ingeneral ﬁnd asmall effect.

Threestudiesusedaninstrumentalvariablesstrategybasedon proximitytohighways(orhighwayon/offramps),toestimatethe impact of fast food density on body weight outcomes among adults.Inacounty-levelstudy,Dunn(2010)usedthenumberof exitsonahighwaywithinacountytoinstrumentforthenumberof restaurantsinacountyandﬁndsrestaurantavailabilitytoeffect onlyfemalesand non-whitesinmedium-densitycounties in11 states.Sucharelationshipwasnotfoundinruralorlow-density counties.InasimilarstudyonresidentsofcentralTexas,Dunnetal.

(2012)reportedonlynon-whitestoexhibithigherobesityratesin responsetofastfoodexposure.Theseauthorsalsouseddistanceto the nearest major highway as an instrument for the fast-food restaurant.AndersonandMatsa(2011)ﬁndnoeffectoffastfood restaurants.Theyalsoexploitthevariationintravelcostsoflocal residents to restaurants along interstate highways, which

primarilyservestravelers.Usingrelateddata,theyobservedthat obese individuals complement eating outside by consuming nutritionallydeﬁcientor“junkfoods.”Theyconcludethatpolicies focusingsolelyonregulatingfastfoodrestaurantsmaynotachieve anysigniﬁcantreductioninBMI.

The above studies examine adult populations. Two studies analyzefastfoodrestaurantsaround schoolsand findfastfood restaurantstosignificantlyaffectchildhoodobesityrates.Currie etal.(2010)usesacross-sectionalsampleofstudentsfromgrade9 inpublicschoolsinCalifornia.Intheirstudy,theeffectisidentified by comparing groups of individuals at only slightly different distancestoa restaurant(i.e.,ofonlyone-tenthofa mile).They measurechangesinexposure,withobesitybeingmeasuredasthe fractionofobesestudentsatthegrade-levelinaschool.Intheir study,theyusechangesinobesityprevalenceinthesamegrade (grade9)butacrossdifferentyears.Alviolaetal.(2014)studythe publicschoolsampleinArkansas.Followingtheearlierstudieson adultpopulations,theirinstrumentalvariablewasthedistanceof school from the nearest highway. Using school-level cross- sectionaldata,theyreportcoefficientsthatmeasurethedifference inobesityratesacrossschoolsthatfacedifferentrestaurantcounts.

Theirstudyestimatesa1.23percentagepointincreaseinschool obesityratesinresponsetoanadditionalrestaurantwithinaone- mileradiusfromaschool.Inthisarticle,wealsofocusonArkansas public schoolchildren. However, unlike Alviola et al.’s (2014) school-levelanalysis, thisstudyusesindividual-levelanalysisof BMI z-scores overtime; i.e, weuse student-level paneldata in contrast to the school-level cross-sectional data they used. In additiontoanidentiﬁcationstrategy basedonanIVestimation usingdistanceofschoolstothenearesthighwayasaninstrument, wealsoemployanadditionalidentiﬁcation strategybasedona plausiblyexogenousreorganizationofpublicschoolsthatcaused somechildrentobereassignedtoanewschoolzone.Hence,these re-assignedstudentsfaceanewfoodenvironmentintheirnewly assignedschool.

Thepresentarticleextendsthesestudiesinseveralways.One,we haveauniquepaneldataset,whichgivesustheabilitytomeasure changeswithinastudentandestimatefixedeffectsattheindividual level.Severalofthepreviousstudiesuseobesityrates(proportionof obeseadultsorchildren)orobesitystatus(abinaryvariable).Weare abletoestimatetheimpactonprecisechangesinBMIintermsof standarddeviationsofthechild’sBMIz-score.Two,incontrastto severalpaststudies,theBMIdatausedherearemeasuredbytrained personnel,asopposedtobeingself-reported.Three,wecontrolfor thecommercial food environmentnear a child’sresidence with precisegeographicinformation.Itisimportanttocontrolforother sourcesof caloriesbecause this could influencethequality and quantity offood consumed.Four,as mentionedabove, this study uses twomethodstoidentifytheeffectoffastfoodavailabilityaround schools onchildren’sBMI z-scores:IVandtheexogenous assignment of students to nearby schools as a resultof a court-mandated restructuringprogram.Thesearediscussedindetailinthesection IdentificationStrategies.

3.Methods

Theprimaryaimofthisstudyistoestimatetheeffectoffast foodrestaurantssurrounding schoolsonchildhood obesity.The variationinastudent’sexposuretofastfoodrestaurantscomesin two ways.One,changeinthecountofrestaurantssurrounding schools.Two,studentsmovingtodifferentschoolseitherbecause ofanaturalprogressionthroughthepublicschoolsystemorbyjust relocatingwithinthestateand,thereby,toanotherschool.Inthis dataset,about42percentofthestudentsrelocatedbutremainedin thepublicschoolsinArkansas.Achild’sBMIcouldbeinﬂuencedby several environmentalfactorsinsideschools,outsideof schools,

(3)

withinthefamily,andwithinthecommunity,someofwhichcould berelatedtothedensityoffastfoodrestaurants.Ourdataallowus tocontrolforseveralobservablefactors.However,notethatthe coefficient on the number of fast food restaurants takes into accountonlythecovariationbetweentherestaurantcountaround theschoolandBMIz-score.Foodintakefromhomemightbiasthe coefficientofinterest,butweincludevariablesthatcapturethe commercial food environment around a student’s residence, includingdistanceto nearestgrocery, dollar store, convenience store, fast food, pizzeria and sandwich place. To the extent householdbehaviorstowardsfoodconsumptionpersistwithtime, theinclusionofstudent-levelfixedeffectscanbeadvantageousas theydifferenceoutunobservedtime-invariantfactorsthatmight becorrelatedwiththevariableofinterest.

Intermsofthefeaturesofthefoodenvironmentwithinschools, allschools weresubjecttoAct1220of2003.Act1220 required nutritionstandardstobeappliedtoallfoodsandbeveragessoldor made available to maintain a healthyschool environment. For example,Act1220requires50%vendedbeveragestobeahealthy choicesuchaswater,100%fruitsjuiceandlow-fat/fat-freemilk.

Act1220was in effectacrossArkansasduring theentirestudy period.Thus, wedo notexpectwide variationinwithinschool environment across schools in terms of the food and physical education/activity environment. Other factors within schools couldplaysomerole.Forexample,peersmayinﬂuencedietand physicalactivitychoicesthatmighttheninﬂuencebodyweight.

Onechallengewithincludingpeers’variableisthatitbringswithit intothemodelahostofotherbiases.AsdescribedinAsirvatham etal.(2018),thesebiasescouldfurthercomplicatethemodel.We, therefore,donotincludepeers’weightvariablebutincludeschool- levelpercentagesofdifferentracegroupsandmealstatus.Bothof thesearecorrelatedwithobesityand,therefore,tosomeextent accountforpeersinﬂuence(Boydetal.,2011).Nevertheless,we acknowledge that the methods used here can only partially addressesthebiases createdbyunobservedchangeswithinthe schoolenvironmentduringthestudyperiod.

Therecouldalsobecommonfactorsatthecommunity-level that might drive restaurant counts. Consider a town with a population that does not demand high calorie foods. Such a populationmightalsohavelowerrestaurant countsinthenear vicinityofschools. Oneimplication ofsuchdemandis thatthe lowerincomepopulationmightdemandlower-costfoodproducts, whichareoftenprovidedbyfastfoodrestaurants.Toaddressallof these biases, we use an instrument that drives the fast food restaurants’locationchoice.

4.Identiﬁcationstrategies

We estimate the effect on two samples, one on the whole studentpopulationandtheotheronasubsampleofstudentswho wereexogenouslyreassigned toother schools in response toa court-mandated reorganization of schools and districts. This allows us to test for possible mechanisms of the effect and propose some recommendation to reduce its effect on BMI outcomes.Althoughcomplementary,theanalysesweconducted usingthesetwosamplesprovidedifferentinsights.Weutilizedthe instrumentalvariablesapproachonthelargerstudentpopulation for whom all information was available, and then exploited a naturalexperimentofexogenousschoolassignmentofasubsetof studentsin response toa court-mandatedschool restructuring.

Belowwediscussbothmethodsindetail.

4.1.Instrumentalvariable(IV)method

The instrument we use is thedistance of theschool tothe nearestinterstateorUShighway.Thishasbeenwellestablishedin

the literature. Dunn (2010); Anderson and Matsa (2011), Dunn etal.(2012),andAlviolaetal.(2014)usesomemeasureofhighway proximityasaninstrumentalvariableforfastfoodrestaurants.Fast foodrestaurantslocateinplaceswherethereisdemandforthe products they offer. By locating closer to highways, fast food restaurantsareabletoproﬁtfromthedemandfromtravelers,but thiscouldexogenouslyincreasesexposuretofast-foodrestaurants forthose livingnearmajorhighways, orinourcaseforschools locatedclosetomajorhighways.Theidentifyingassumptionisthat nearnesstohighwayisafactorin restaurantlocationdecisions.

That is, nearness to highway is a factor in restaurant location decisionsbecauseproximitytointerstateandUShighwaysmight bringinﬁrmssupplyingservicesandproductstotravelers.

UsingCountyBusinessPatternsdata,Dunn(2010)showsthat interstateexitsdrivethelocationoffastfoodestablishmentsin waythatisdifferentfromotherbusinesses. Healsoshowsthat number of interstate exits isnot correlated withother healthy behaviors,suchasfruitandvegetableconsumptionandphysical activities. Dunn (2010) discusses both aspects of demand and supplythatcouldbiasanestimate.Forexample,higherdemandfor fast food could lead to more restaurants. On the other hand, residentswithpreferenceforhealthcouldenforcerestrictionson locationandnumberoffastfoodrestaurants.Heconcludesthatthe directionofthebiasisgenerallypositive.

In our context,schools factoring in distancefrom stateand federal highways might violate the exclusion restriction. This violationwould occurbecauseaschool’sdecisionanda restaurant’sdecisionareinﬂuenced bythesameexplanatoryvariable.

Eventhoughsomestatesdostipulateschoolconstructiontobeata distancefromhighways,Arkansasonlystipulatesdistancefrom anysourceofsoundthatproduces65decibelssustainedand75 decibelspeak².Thus,wearguethatthedistanceofschoolsfrom highwayinthecontextofpoliciesinplaceinArkansasdoesnot violatetheexclusionrestriction.

Duringourstudyperiod,thelocationsofhighwaysareﬁxed.

However,studentschangeschoolsaspartofanaturalprogression through the public-school system and this creates time series variationinthedistancebetweenthechild’sschoolandhighway.

Theinstrumentusedintheliterature,ingeneral,considersonly interstate highway, but due to the sparseness of interstate highways in Arkansas, as shown by Alviola et al. (2014), we considerbothinterstateandUShighwaysforourinstrument.⁵The standardwaytoincludetheIVinanequationistouseitasasingle variable.Inthisstudy,weincludeitintwodifferentways.Firstly,as asinglecontinuousvariable.Secondly,tocontrolforanon-linear relationship,theIVissplitintosegmentsbasedondifferentradial distancesfromtheschooltothehighway.Sinceweestimatethe effectatone-third,two-thirds,andamile,theinstrumentissplit intofourparts,namelyone-thirdmile,two-thirdsmile,amile,and morethanamile.Aschoolthatisathirdofamileawayfromthe nearestfederalhighwaywillhavethespeciﬁcdistanceinthesub- variableforone-thirdmile,withzerosinothersub-variables.This istantamounttorelaxingtheassumptionthattheinstrumenthasa uniquelinearrelationshipwiththeendogenousvariable.

4.2.Exogenousassignmenttoschools

Inadditiontotheinstrumentalvariablemethod,weanalyzea subsampleofstudentswhowereexogenouslyassignedtodifferent schools.Thissubsampleof2739observationsconstitutesasmall f-

r- a- c- t-

2Ontransportationroutes,Arkansasrequiresnoisefromairandmotorvehicle trafﬁctonotexceedsounddecibellevelof65dbsustainedand75dbpeak-Arkansas Rule and Regulations Governing the Minimum Schoolhouse Construction Standards;ArkansasCode6-20-1406.

(4)

ionofthetotal1,352,696observationsinthemainanalysissample, but provides a complementary analysis to the IV methods explainedabove.

Thestrategy is similar tothat pursuedby Asirvathamet al.

(2018)intheiranalysisofpeereffectsusingtheArkansasBMIdata.

Asexplainedinthisearlierstudy,theexogenousreassignmentwas createdbya courtmandated restructuringof publicschools. In 1992,theLakeViewSchoolDistrictand otherplaintiffsclaimed thattheschoolfundingsystemwasunconstitutional.IntheLake ViewSchoolDistrictNo.25v.Huckabeecase,theArkansasSupreme Courtruled that thestateeducational fundingwas unconstitu- tional.To meet the courtmandate, theState passedthePublic EducationReorganizationAct,Act60,duringtheSecondExtraor- dinarySessionof2003.

School reorganization thus occurred to overhaul the public schoolfundingsystem.Sincetheprimarymotivationwasnotto restructuretheschoolstoimprovestudents’healthandbecause the legislation was passed in a special session of the state legislature in response to the court ruling, we posit that this restructuring created an exogenous school assignment for the studentsitaffected.

ThereweretwowaystheStatesoughttocomplywiththecourt decision: either through consolidation or annexation for all districts with fewer than 350 students. Consolidation involved cuttingadministrativeoverheadbybringingschoolsand/orschool districtsunderfewermanagementpersonnel.Annexationrequired physicallyclosingschoollocationsandhavethechildrenattenda nearbyschoolthatmetthecriterionsetbythelegislature.Thus, annexation created a change in school of attendance that was plausiblyexogenous.

An exogenous reassignment to other schools more directly addressestheissueofself-selection intoschools.Thisreassign- ment, we argue, is largely uncorrelated with observables and unobservables that determineBMI outcomes.This is especially valid because the students are not in their original school attendancezone, which then largelyrules outcommon factors orlinkagesbetweenstudentBMIoutcomesandrestaurantlocation aroundtheirre-assignedschool.Intheresultssection,wediscuss differencesbetweenannexedschoolsandreassignedschools.

5.Data

In an effort to combat high rates of childhood obesity, the ArkansasGeneralAssemblypassedtheAct1220of2003.Among otherthings,thislegislationmandatedthatpublicschoolchildren beassessedforBMIbeginninginthe2003–2004schoolyear.BMI screeningshavebeenongoingsincethattime.WeusetheseBMI screeningsasthemain datasourceinthisstudy.TheArkansas CenterforHealthImprovement(ACHI)ledthedevelopmentand implementation of the state-wide BMI assessment process.¹³ ACHIdevelopedastatewideprotocolforstandardizedmeasure- ments across the state. Height and weight measurements are measuredby trainedpersonnel in schoolsand are reported to ACHI.ThedatasetweuseincludesBMIz-scores,race,genderand participationinfreeorreducedlunches.Thesedataalsoinclude theweightstatusofeachchild,whichbasedonreferencegrowth chartsfromtheCentersforDiseaseControlandPrevention(CDC).

Only those students with at least two BMI observations are includedin ouranalysissample.Medicalprofessionals andthe CentersforDiseaseControl&Prevention(CDC)useBMIz-scores as a general indicator to track body weight progress among children. As a child grows, so does his/her body weight and height. The BMI z-score takes into account this growth and development,whichalsodiffersbysex.TheBMIz-scoreprovides a uniform measure of body weight across our sample that containschildrenthatdifferbyageandsex.

Ourempiricsarebasedonapaneldatasetcoveringtheyears 2004–2010.Oneproblemweconfrontedinassemblingthedataset isthatstatepolicyrelatingtothefrequencyofBMImeasurement changedduringourstudyperiod.From2004to2007,theBMIof schoolchildrenwasmeasuredannuallyforallgrades.Thereafter, BMI was measured and reported only for children in even- numbered gradesinclusiveof kindergarten.Thus, we haveBMI prevalenceratesforstudentsfromallgradesfrom2004through 2007,butonlyforstudentsin evengradesafter2007.Thenon- reportingofobesityprevalenceinoddgradesafter2007shouldnot biasourestimates,sincethedecisiontostopmeasuringtheBMIof childreninodd-numberedgradeswasexogenousinthatitwasnot madebythechild,thechild’sfamily,orthechild’sschool.However, thischangeinreportingdoesaffectourabilitytotakeintoaccount BMIchangesinaconsistentfashionovertime.

Anothersourceofdatawebringintotheanalysisislocationof foodbusinesses.ThesearebasedonDun&Bradstreetbusinesslists andincluderestaurants,grocers,andotherfoodstores.Detailson the construction of the food environment are provided in the Appendix.Using GeographicInformationSystem (GIS)software, wecreatemeasuresofthecommercialfoodenvironmentaround schoolsandresidences.Thesevariablesmeasurethenumberand typeofrestaurantsatvaryingradialdistancesfromschoolsinthe increments of a third of a mile up to a mile. Speciﬁcally, we measurethenumberoffastfoodrestaurants,sandwichplacesand pizzerias.Measuresof thefoodenvironmentarounda student’s residence include distance to the nearest grocery store, dollar store, convenience store, fast food restaurant, pizzeria, and sandwichplace.

Fastfoodrestaurantcountsmayreﬂectlocaldemandforfast food.Tothebestoftheauthors’knowledge,allbutoneschoolhada

“closedcampus” policyduringthestudyperiod. Seniorsinthis highschoolhadtheoptionofeatingoutduringthelunchperiod with school permission. Students in the sample would not generallyhavesanctionedaccesstofastfoodsduringtheschool daysuggestingthat accesswould beprimarilybeforeand after schoolhours.Wealsorestrictedthesampletothoselessthan18 yearsold.

5.1.Annexedsample

Table1presentsthesummarystatisticsoftheentirepopulation ofstudentsincludedintheanalysis(calledthegeneralsample)and alsocomparesittothestudentswhowerereassignedtoadifferent school(calledtheannexedsample).Studentsinthegeneralsample differ from those students who were reassigned to a different schoolincertaincharacteristics.TheBMIz-scoresintheannexed sampleare0.05standarddeviationslargerthantheoverallsample.

Thereisanegligibledifferenceacrossgenderproportion.Annexed schools,however,had27percentagepointsmoreAfricanAmerican students and21percentage pointslessCaucasianstudents. The annexed sample also had a higher percentage of students qualifying for free lunches, 65% compared to 43% among all schools.Intermsofthecommercialfoodenvironment,studentsin annexedschoolswerefurtherawayfromallofthemeasuredfood establishments.Thissamplealsohadslightlymoreruralstudents.

The summary statistics generally reﬂect the communities that wereaffectedbythis schoolrestructuring process.Forinstance, thisreformclosedsmallerschoolsandintegratedthemintolarger schools. Smaller schools are predominantly in the rural areas.

Overall, differences in school-level demographic characteristics betweenthegeneralsampleandtheannexedschoolsaremarkedly differentin terms of race, householdincome, urbanization and accesstosomeofthemostimportantfoodestablishments.

Oneconcernin therestructuringprocessis ifstudentswere reassigned to schools with similarcharacteristics or in similar

(5)

neighborhoods.Asirvathametal.(2018)assesseddifferencesinthe schoolsthatclosedand theschools towhich studentsfromthe closedschoolswerereassigned.Dataonchildrenfromthesending schools, i.e., annexed schools, constitute the second analytic sample we used in our analysis. This exogenous assignment addressestheissueofself-selectionbecause,haditnotbeenforthe annexation,thesestudentswould nototherwisebeattending a schooloutsideoftheiroriginalattendancezone.Asirvathametal.

(2018)foundsigniﬁcantdifferencesinpercentfree andreduced lunchbetweensendingschoolsandreceivingschools indicating that the schools differed in socioeconomic characteristics.

However,bythenatureoftheschoolrestructuring processand differenceinstudentpopulationcharacteristics,wecanafﬁrmthat a student’s current BMI is unlikely to be correlated with unobservedfactorsatthenewschool.

6.Econometricmodel

Ourbasicmodeltoestimatetheeffectoffastfoodrestaurants surroundingschoolsonstudentBMIis:

Yit=β0+β1Rk(i)t+β2Xit+β3Xk(i)t+β4Fit+Uikt, (1) whereYitistheBMIz-scoreofthei^thstudentintimet;Rk(i)tisthe restaurantcountsaroundthek^thschoolofthei^thstudentattimet;

Xitisavectorofstudenti’scharacteristics;Xk(i)tisavectordepicting demographicstudentcharacteristicsattheschool-levelinschoolk where student i attends, including proportion of students of differentracecategoriesandgenderandpercentqualifyingforfree and reduced lunch; Fit is the vector of commercial food environment near the residence of student i which includes distancetonearestgrocery,dollar store,convenience store,fast food,pizzeriaandsandwichplace;andUiistheerrortermwhich equals

m

i +

e

it, where

m

i is the unobserved time invariant componentand

e

îtîs^the^sphericalêrror^term.^Variables^measuring

different aspects of the commercial food environment around student’s school and residence were constructed using GIS software and geolocation of the schools, food establishments andstudentresidence.Toteaseouttheeffectofdistanceofthefast food restaurants from schools, we developed three variables

measuringthenumberof fastfoodrestaurantswithindifferent radial distances fromschool, namelyone-third of a mile, two- thirdsofamile,andamile.Besidesthenumberofrestaurantsat differentradialdistances, wealso createdtwo sets ofvariables withdifferentcombinationsofrestaurants.Oneincludedonlyfast foodrestaurantsandtheotherincludedfastfood,sandwichesand pizzerias.

Thepanelnatureofthedataandtheamountofinformationon students, schoolsand foodenvironmentallow ustocontrolfor individualandschool-levelcharacteristics,andalsotousestudent- levelfixedeffectsthatfurtherreducestheendogeneitybiasdueto omittedvariablesintheestimate.Sincetherecouldbeyear-to-year changesinastudent’sBMIthatifnotaccountedformightbiasthe estimates,wealsoestimateatwo-wayFEmodelbyaddingbinary variablesfordifferentyearsinthedataperiod.Thus,thetwo-way FEmodelsincludebothstudent-levelfixedeffectsandyearfixed effects.Notethattheobservationsareannualsoyearfixedeffects wouldcapturetime-invariantyeartoyearchangesintheBMIthat arenotcapturedbytheincludedright-handsidevariables.

7.Resultsanddiscussion

Asdiscussedabove,ourresearchobjectiveistoestimatethe impactoffastfoodrestaurantssurroundingtheschoolsonachild’s BMIz-score,andcheckifthereareanydifferencesintheeffectsby radialdistanceofrestaurantsfromschool.Inthissection,wefirst present results from the pooled OLS (top panel, Table 2) and studentfixedeffectsmodels(bottompanel,Table2).Thepooled OLSshowsthattheeffect,thoughlikelybiased,isverysmallorzero inmagnitudeandthatthesignofthecoefficientisalwaysnegative, which is inconsistentwith ourprior expectation that fastfood exposureshouldbepositivelyrelatedtoBMI.

Importantly, healthier parents choosing healthier neighbor- hoodsmightcreatepositivebias.Onthesupplyside,restaurants locating in neighborhoods with a larger proportion of lower incomefamiliesmightatleastshowsomerestauranteffect.Pooled OLS,however,assumesallobservationsareindependent.Ourdata havemultipleobservationsoneachstudentandthereforeweare abletoobtainwithinestimatesbycontrollingforthestudent-level ﬁxedeffects. These results are in thebottom panelof Table 2.

Someoftheestimatesarepositiveinsignbutareverysmallin magnitude. Moreover, results are robust to inclusion of grade effects or additional controls for food stores around children’s residences. Controlling for the commercial food environment surroundingtheresidenceofeachstudenthadlittleimpactonthe estimates once student-level time-invariant unobserved factors areaccountedfor.

Eventhoughwecontrolforunobservedtime-invariantfactors atthestudent-levelbeyondtheresidentialfoodenvironment,we cannot entirely rule out the self-selection bias. This could potentiallybiastheestimatebecauseparentschoosewherethey live.Forexample,anoccupationrequiringfrequentlong-distance travelmightincreasethelikelihoodofchoosingaresidencenear highways.

Aspreviouslydiscussed,ourfirst identificationstrategy isto instrumentfortherestaurantcounts.ResultsfromfixedeffectsIV methodarepresentedinTables3and4.Theheteroskedasticity- robustfirststageregressionresultsinTable3indicatethataverage distanceofschools fromfederalhighways (theinstrument)isa very significant and strong predictor of restaurant counts surroundingschools. Thenegativecoefficientindicatesthat the numberofrestaurantsdecreaseasthedistanceoftheschoolfrom the highway increases. Large F-statistics are indicative of high correlationbetweenrestaurantscountsandtheinstrument.

Onthesmallnessofthemagnitude,itshouldbenotedthatmore than50%oftheschoolsdidnothaverestaurantswithinthecertain Table1

Summarystatistics.

Variable FullAnalyticSample

(N=1,362,306)

AnnexationSample^a (N=2739)

BMIz-score 0.704 0.754

Race(percent)^b

Caucasian 66.45 45.3

AfricanAmerican 21.94 48.9

Hispanic 6.73 3.4

NativeAmerican 0.37 0.2

Asian 1.24 0.2

Schoolmeal(percent)

Freemeals 42.85 65.2

Reducedmeals 9.53 10.7

Nearestfoodstore(miles)

Nearestgrocery 3.09 4.7

Nearestdollar 2.92 4.9

Nearconvenientstore 1.61 2.9

Nearestfast-food 2.60 4.2

Nearestpizzeria 4.20 6.4

Nearestsandwich 2.80 5.1

Otherdemographics

Female(percent) 48.6 48.8

Age(months) 134.8 125.3

Rural(percent) 30.0 40.2

Urban(percent) 57.1 45.6

aThispopulationincludesallstudentsingrades1–9andwhohaveatleasttwo annualobservationpostannexation.

b RaceotherthanthoselistedaboveisnotincludedintheTable.

(6)

defined radii at some point during the study period. The sign suggeststhatthe fartherthe averagedistanceof aschool from nearesthighway,thelesseristhenumberofrestaurantslocating nearschools.Thereducedformfixedeffectsregressionalsoyielded significantestimatesoftheinstrument.Despitetheabsenceoffast foodrestaurantsforalargenumberofschools,themagnitudeat one-thirdradialdistancesuggestsa2.1percentincreaseinBMIz- score,0.0147standarddeviationsfromthemean0.704SD.Ata two-thirdsmileradialdistance,theestimatedecreasedbyathird andatamileradius,theestimatefurtherdecreasesbyafifth.

Theinstrumentalvariablesestimatesshowinterestingresults.

First,thegeneralresultisthatrestaurantsaroundanydefinedradii haveatleastsomeeffectonstudentBMI.Second,withinthesame groupofrestaurants,theeffectdecreasesastheradialdistance increases.Thisreflectsthefactthatasarestaurantlocatesfurther awayfromaschool,itsinfluenceonBMIbeginstowane.Thiscould indicatethereducedeffectoncalorieintakeasdistancefroma schoolincreasesorthatdistancefromschooldoesmatter–even though the effect is small. Thirdly, among different groups in general, the estimate is higher when the fast food measure excludessandwichplacesandpizzerias.Atathirdofamileradius, fastfood restaurantsalone showanincrease of0.024 standard deviations in BMI z-score compared to 0.015. This seems counterintuitive, but keep in mind that only about 50% of the schoolshad fastfoodrestaurants.Thisrelativelysmallermagni- tudecouldindicatethattheeffectofsandwichplacesandpizzerias mightbemuchlowerthanthefastfoodsitself.Thus,theIVmethod findsaverymodesteffectoftherestaurantssurroundingschools onastudent’sBMI.

Ourcomplementaryanalysisoftheannexedsampleisconsistent withthefindingsaboveinthatthereisnoevidenceofalargefast- foodeffectonBMI(Table5).Itismuchsmaller.BothOLSandfixed effectsestimatesshownosignificanteffectatanyradialdistance fromtheschool.Theobservationsonreassignedstudentsconstitut- edlessthanaquarterpercentofthetotalobservationsusedinthe analysisabove.Theregressionmodelusedtotheobtaintheresultsin Table 3include similar control variablesto the thoseused in thepanel IV models reported above. As discussed above, this sample is differentfromthegeneralsampleanalyzed inseveralrespects.Those differencescouldpartlyexplainwhy noeffectisobservedinthe annexedsamplecomparedtoonlyamarginaleffectinthegeneral sample.Comparedtothegeneralsamplestudentsintheannexed sample had to travel further away because theschool in their Table2

Estimatesofdensityoffastfoodrestaurants(FF),sandwichplaces(SW)andpizzerias(PZ)surroundingschoolsonstudent’sBMIz-score.

Model Variable(miles) BaseModel(BM) BM+gradeFE BM+gradeFE+FoodEnvironment

OLS FF,SW&PZ(1/3) 0.0011658 0.00069997 0.0011658

(0.00095281) (0.00095475) (0.0009574)

FF,SW&PZ(2/3) 0.00236391^*** 0.00246344^*** 0.0009603^**

(0.00042873) (0.00042967) (0.0004356)

FF,SW&PZ (1)

0.00201689^*** 0.002020111^*** 0.0007813^***

(0.00027862) (0.00027929) (0.0002866)

FF (1/3)

0.00092471 0.0021661 0.0018418

(0.00138714) (0.00139228) (0.0013942)

FF (2/3)

0.00384492^*** 0.0039002345^*** 0.0018201^***

(0.00071439) (0.0007154) (0.0007233)

FF (1)

0.00387424^*** 0.00394801^*** 0.0021296^***

(0.00048186) (0.00048323) (0.0004926)

FixedEffects FF,SW&PZ(1/3) 0.0007181 0.00061789 0.0005299

(0.00049631) (0.0004971) (0.0004974)

FF,SW&PZ(2/3) 0.00037238 0.00037723 0.0002952

(0.00023205) (0.00023251) (0.0002332)

FF,SW&PZ (1)

0.00027201^* 0.00028487^* 0.0002097

(0.0001574) (0.00015776) (0.0001585)

FF (1/3)

0.00089283 0.00047178 0.0003469

(0.00070364) (0.00070548) (0.0007059)

FF (2/3)

0.00002974 0.00005405 0.0000816

(0.0003881) (0.00038875) (0.0003895)

FF (1)

0.00010497 0.00005186 0.000073

(0.00026805) (0.00026894) (0.0002699)

Note:Regressorsinthebasemodel(BM)regressionmodelincludestudentage,race,gender,rural/urbanareaofresidence,participationinschoolfree/reducedlunchprogram, andyeardummyvariables.BM+gradeFEhasallthevariablesinBMplusthegradefixedeffects.ThelastcolumnreportsestimatesfromamodelthathasallvariablesasinBM, gradefixedeffectsandthecommercialfoodenvironmentaroundstudentresidence.OLSestimatesandstudentfixedeffects(FE)estimatesdifferonlyinthefactthattheFE modelsincludestudentfixedeffectsinadditiontotheotherregressorslisted.School-levelclusteredstandarderrorsareinparentheses.N=1,362,696.

Robustclusteredstandarderrorsaregiveninparenthesis.

*** Signiﬁcantatthe1percentlevel.

** Signiﬁcantatthe5percentlevel.

* Signiﬁcantatthe10percentlevel.

Table3

Fixedeffectsivestimatesofdensityoffastfoodrestaurants(FF),sandwichplaces (SW)andpizzerias(PZ)surroundingschoolsonstudent’sBMIz-score.

Variable(miles) IVestimates FirstStage ReducedForm F-stat FF,SW&PZ(1/3) 0.0146856^*** 0.0000617^*** 0.000000907^*** 5872

(0.0043494) (0.000000806) (0.000000268) FF,SW&PZ(2/3) 0.004594^*** 0.0001974^*** 0.000000907^*** 8103

(0.0013601) (0.00000219) (0.000000268) FF,SW&PZ(1) 0.0029035^*** 0.0003123^*** 0.000000974^*** 9610

(0.0008596) (0.00000319) (0.000000268) FF(1/3) 0.0244387^*** 0.0000371^*** 0.000000974^*** 4856

(0.0072401) (0.000000532) (0.000000268) FF(2/3) 0.0075479^*** 0.0001201^*** 0.000000974^*** 8949

(0.002235) (0.00000127) (0.000000268) FF(1) 0.0045204^*** 0.0002006^*** 0.000000974^*** 11,225

(0.0013385) (0.00000189) (0.000000268)

Note:Regressorsineachregressionmodelincludestudentage,race,gender,rural/

urbanareaofresidence,participationinschoolfree/reducedlunchprogram,year dummyvariables,thecommercialfoodenvironmentaroundstudentresidenceand studentﬁxedeffects.School-levelclusteredstandarderrorsareinparentheses.US andinterstatehighwaysareusedasinstruments.N=1,352,696.

**Signiﬁcantatthe5percentlevel.

*Signiﬁcantatthe10percentlevel.

(7)

attendance zone closed which may have prevented access to restaurantsbeforeandafterschool-hours.TheOLSandtheﬁxed effectsestimatesshowhighstandarderrors.

The resultspresentedsofaranalyzeallstudentsandinclude several covariates that might inﬂuence a student’s BMI. In the Appendix,weranadditionalregressionstoexamineiftherestaurant

effectmightvarybyageorbygeographicareasbasedoncitycenters or populationdensities. Speciﬁcally, speciﬁcations basedonage grouppresentestimatesforelementary,middle,andhighschools (TablesA1–A3,respectively).Estimatesfromregressionsforrural andurbanschoolsarepresentedinTablesA4andA5,respectively.

Theresultsshowsomedifferencesbyagebutnothingofeconomic signiﬁcance.Themainconclusionsofthepaperappeartoholdacross thedifferentagecategories.Wealsodonotﬁndanydifferenceacross ruralandurbanareas.

8.Conclusions

Inthisstudy,weestimatetheeffectofthenumberoffastfood restaurantssurroundingschoolsonthebodymassindex(BMI)z- scoreofschoolchildren.Thestudyusesindividual-levelpaneldata fromstudentsattending publicschools inArkansas. Theresults from the two methods on two samples drawn from the same student population are similar. First, an instrumental variable strategyisemployedusinganinstrumentthathasbeenvalidated bypreviousstudies.Aschool’sdistancefromthenearesthighway isusedtoinstrumentfortheendogenousvariableofinterest,the restaurant count surrounding the school at specified radial distances.Theidentifyingassumptionisthatfastfoodrestaurants locatedbythehighwayscatertothehighwaytravelers.Weusethe sameinstrumentintwowaysforefficientestimation.Inthefirst specification,weuseasinglevariable.Intheotherspecification, we separate the instrumental variable into parts based on its distancefromthehighways.

Thesameinstrumentisusedintwodifferentspeciﬁcations,and both arrive at similar results. Second, we exploit a natural experiment that resulted when a number of Arkansas schools werereorganizedinresponsetoastateSupremeCourtdecisionon school funding. This created an exogenous reassignment of studentstoschoolsotherthantheoneintheirattendancezone.

Students in this exogenously assigned sample have different exposuretorestaurantssurroundingschools.Infact,giventheir distanceandtimetotraveltoaschoolthatislocatedfurtheraway, itisunlikelythattheywouldhavehadtheopportunitytovisita fastfoodrestaurantwhile takingtheschoolbus. Thus,thetwo estimates,althoughcomplementary,havedifferentimplications.

Table4

Fixedeffectssplit-iv(continuous)estimatesofdensityoffastfoodrestaurants(ff),sandwichplaces(SW)andpizzerias(PZ)surroundingschoolsonstudent’sBMIz-score.

Variable(miles) IVestimate IVpart^a(mile) FirstStage(estimate) ReducedForms(estimate)

FF,SW&PZ(1/3) 0.0021417 0to1/3^rd 0.001246^*** 0.00000427

(0.0014272) (0.0000267) (0.00000716)

1/3rdto2/3^rd 0.0016149^*** 0.00000642^**

(0.0000132) (0.00000337)

2/3rdto1 0.0008391^*** 0.000000186

(0.00000948) (0.000000227)

FF,SW&PZ(2/3) 0.002606^*** 0to1/3^rd 0.0010658^*** 0.00000427

(0.0007005) (0.0000584) (0.00000716)

1/3rdto2/3^rd 0.0018108^*** 0.00000642^***,*

(0.0000275) (0.00000337)

2/3rdto1 0.0016538^*** 0.000000186

(0.0,000,186) (0.000000227)

FF,SW&PZ(1) 0.0010169^** 0to1/3^rd 0.0001297 0.00000427

(0.0005787) (0.0000814) (0.00000716)

1/3rdto2/3^rd 0.0009549^*** 0.00000642^**

(0.0000401) (0.00000337)

2/3rdto1 0.0010386^*** 0.000000186

(0.0,000,268) (0.000000227)

Note:Regressorsineachregressionmodelincludestudentage,race,gender,rural/urbanareaofresidence,participationinschoolfree/reducedlunchprogram,yeardummy variables,thecommercialfoodenvironmentaroundstudentresidenceandstudentﬁxedeffects.School-levelclusteredstandarderrorsareinparentheses.USandinterstate highwaysareusedasinstruments.N=1,352,696.

aFirststageresultsfortheIVpartgreaterthan1milearenotpresentedhere.Ineverycaseisnegativeandsigniﬁcantat0.01.

Table5

Estimates of densityof fastfood restaurants (FF), sandwich places(SW) and pizzerias(PZ)surroundingschoolsonstudent’sBMIz-scorefortheannexation sample.

Model Variable(miles) Estimate

OLS FF,SW&PZ(1/3) 0.0256623

(0.058483) FF,SW&PZ(2/3) 0.0158216

(0.0298973) FF,SW&PZ

(1)

0.0049307 (0.0262291) FF

(1/3)

0.0762247 (0.072834) FF

(2/3)

0.0300403 (0.055523) FF

(1)

0.0175953 (0.0388795)

FixedEffects FF,SW&PZ(1/3) 0.0027338

(0.0172865) FF,SW&PZ(2/3) 0.0143413 (0.0096316) FF,SW&PZ

(1)

0.0112431 (0.0089123) FF

(1/3)

0.0145423 (0.0241696) FF

(2/3)

0.0069048 (0.0141348) FF

(1)

0.006933 (0.012964) Note:Regressorsincludestudentage,race,gender,rural/urbanareaofresidence, participationinschoolfree/reducedlunchprogramandyearandgradedummies.

Basemodelalsoincludesschoolfixedeffects.Thestudentfixedeffects(FE)model includesstudentfixedeffectsinadditiontotheotherregressorslisted.School-level clusteredstandarderrorsareinparentheses.N=2739.

***Signiﬁcantatthe1percentlevel.

Robustclusteredstandarderrorsaregiveninparenthesis.

(8)

Thisstudyfindstheeffectofrestaurantssurroundingschoolson childhoodobesitytobenegligible.Ourpointestimate,whilevery small,ishighlysignificant.Thesmallerestimatemightcultivatea laissez-faire attitude towards regulating restaurants around schools,butthefindingsreportedheremustbeinterpretedwithin thecontext of the study. Most Arkansasschools to had closed campus policies during the study period that meant children wouldonlyhaveaccesstooutsiderestaurantfoodbeforeorafter schoolhours.Hence,replicatingourstudyinothercontextstotest therobustnessofourfindingswouldbewarranted.

Conﬂictofinterest

None.

Funding

ThisresearchwasfundedbytheAgricultureandFoodResearch InitiativeoftheUSDANationalInstituteofFoodandAgriculture, grant number 2011-68001-30014. This work was also partly supported by the National Research Foundation of Korea

(NRF-2014S1A3A2044459); Research Council of Norway Grant

#233800; the Arkansas Biosciences Institute (ABI); and the National Institute of General Medical Sciences of the National Institutes of Health under Award Number P20GM109096. The researchreportedinthisstudywasapprovedbytheInstitutional Review Board, University of Arkansas, Fayetteville, USA under protocols #10-11-235 and#14-07-026.Theproceduresofthis studyareinaccordancewithethicalstandardsforhumanresearch.

Acknowledgements

WethankDianaDanforthforresearchassistanceindataandin buildingthecommercialfoodenvironmentdatabase.GrantWest created the neighboring schooldatabase using ArcGIS.Stephen Lein at ACHI assisted in working with the archived BMI data.

Zhongyi Wang assisted in verifying the addresses of smaller grocerystores.WearethankfultootherstaffatACHIforassisting usinconductingtheanalysis.

AppendixA.

TableA1

Estimatesofdensityoffastfoodrestaurants(FF),sandwichplaces(SW)andpizzerias(PZ)surroundingschoolsonstudent’sBMIz-scoreamongELEMENTARYstudents.

Variable(miles) BaseModel(BM) BM+gradeFE BM+gradeFE+FoodEnvironment

FF,SW&PZ(1/3) 0.0014376 0.0014488 0.0025444

(0.002094) (0.0020926) (0.0020956)

FF,SW&PZ(2/3) 0.001168 0.0010017 0.0000156

(0.0008928) (0.0008922) (0.0,008,994)

FF,SW&PZ (1)

0.0008875 0.0007106 0.0000721

(0.0005655) (0.0005655) (0.0005736)

FF (1/3)

0.0028932 0.0028674 0.0038325

(0.0031002) (0.0030987) (0.0031031)

FF (2/3)

0.0022872 0.0019864 0.000749

(0.0014039) (0.001403) (0.001412)

FF (1)

0.0026321^*** 0.0023792^** 0.001328

(0.0009547) (0.0009543) (0.0009646)

Note:Regressorsinthebasemodel(BM)regressionmodelincludestudentage,race,gender,rural/urbanareaofresidence,participationinschoolfree/reducedlunchprogram, yeardummyvariables,andlagoftheincludedrestaurantcountinthemodel.BM+gradeFEhasallthevariablesinBMplusthegradeﬁxedeffects.Thelastcolumnreports estimatesfromamodelthathasallvariablesasinBM,gradeﬁxedeffectsandthecommercialfoodenvironmentaroundstudentresidence.School-levelclusteredstandard errorsareinparentheses.N=370,123.

TableA2

Estimatesofdensityoffastfoodrestaurants(FF),sandwichplaces(SW)andpizzerias(PZ)surroundingschoolsonstudent’sBMIz-scoreamongMIDDLEschoolstudents.

Variable(miles) BaseModel(BM) BM+gradeFE BM+gradeFE+FoodEnvironment

FF,SW&PZ(1/3) 0.0030406 0.0035474 0.0034874

(0.0030211) (0.00302) (0.0030293)

FF,SW&PZ(2/3) 0.0025396^* 0.0015756 0.0022654

(0.0014609) (0.0014658) (0.0014757)

FF,SW&PZ (1)

0.0012059 0.0005164 0.0012197

(0.0009726) (0.0009769) (0.0009895)

FF (1/3)

0.0006977 0.0011386 0.0001076

(0.0041535) (0.0041541) (0.0041603)

FF (2/3)

0.0044559^* 0.0029826 0.0039919^*

(0.0024214) (0.0024271) (0.0024382)

FF (1)

0.0,029,672^* 0.001853 0.0029658^*

(0.0016783) (0.001684) (0.0017014)

Note:Regressorsinthebasemodel(BM)regressionmodelincludestudentage,race,gender,rural/urbanareaofresidence,participationinschoolfree/reducedlunchprogram, yeardummyvariables,andlagoftheincludedrestaurantcountinthemodel.BM+gradeFEhasallthevariablesinBMplusthegradeﬁxedeffects.Thelastcolumnreports estimatesfromamodelthathasallvariablesasinBM,gradeﬁxedeffectsandthecommercialfoodenvironmentaroundstudentresidence.School-levelclusteredstandard errorsareinparentheses.N=111,464.

***Signiﬁcantatthe1percentlevel.

* Signiﬁcantatthe10percentlevel.