Household dairy production and child growth: Evidence from Bangladesh

(1)

Household dairy production and child growth: Evidence from Bangladesh

Samira Choudhury

^a

, Derek D. Headey

^b,

*

aSOAS,UniversityofLondon,UnitedKingdom,UnitedStates

bPoverty,Health&NutritionDivision,TheInternationalFoodPolicyResearchInstitute(IFPRI),UnitedStates

ARTICLE INFO

Articlehistory:

Received11January2018

Receivedinrevisedform25June2018 Accepted11July2018

Availableonline17July2018

JELclassiﬁcations:

O130 Q120 I150

Keywords:

Livestock Dairyproduction Animal-sourcedfoods Stunting

ABSTRACT

Researchfromrichercountriesfindsthatdairyconsumptionhasstrongpositiveassociationswithlinear growthinchildren,butsurprisinglylittleevidenceexistsfordevelopingcountrieswheredietsarefarless diversified.Oneexceptionisarecenteconomicsliteratureusingthenotionofincompletemarketsto estimatetheimpactsofcattleownershiponchildren’smilkconsumptionandgrowthoutcomesin EasternAfrica.Inadditiontoexternalvalidityconcerns,anobviousinternalvalidityconcernisthatdairy producersmaysystematicallydifferfromnon-dairyhouseholds,particularlyintermsoflatentwealthor nutritionalknowledge.Were-examinetheseconcernsbyapplyinganoveldoubledifferencemodelto datafromruralBangladesh,acountrywithrelativelylowlevelsofmilkconsumptionandhighratesof stunting.Weexploitthefactthatacow’slactationcyclesprovideanexogenoussourceofvariationin householdmilksupply,whichallowsustodistinguishbetweenacontrolgroupofhouseholdsthatdonot owncows,atreatmentgroupthatowncowsthathaveproducedmilk,andaplacebogroupofcow- owninghouseholdsthathavenotproducedmilkinthepast12months.Wefindthathouseholddairy productionincreasesheight-for-ageZscoresby0.52standarddeviationsinthecritical6–23month growthwindow,thoughinthefirstyearoflifewefindthathouseholddairysupplyisassociatedwitha 21.7pointdeclineintherateofbreastfeeding.Theresultsthereforesuggestthatincreasingaccesstodairy productscanbeextremelybeneficialtochildren’snutrition,butmayneedtobeaccompaniedbyefforts toimprovenutritionalknowledgeandappropriatebreastfeedingpractices.

1.Introduction

Worldwide,childundernutritionisincreasinglyrecognizedasa signiﬁcant global health problem and a major constraint to economicdevelopment. Childundernutrition is associated with almost 3.1 million child deaths (Black et al., 2013), impaired cognitive developmentin early childhood (Walker et al., 2011;

Grantham-McGregoretal.,2007),reducedschoolattainmentin childhood,andlowerlabourproductivityandwagesinadulthood (Shekaretal.,2006;Victoraetal.,2008;Hoddinottetal.,2008).

Nutritionists,moreover,have increasinglyemphasizedthat it is goodnutritioninearlychildhood–inuteroandtheﬁrst24months after birth – that is truly critical for ensuring healthy growth (Shrimptonetal.,2001;Victoraetal.,2010).

Aparticularlystrikingnutritionalfeatureofdevelopingcountry populations is that growthfaltering appears to be particularly

pronouncedfromroughly6monthsofageto20monthsofage,a period that coincides with theintroduction of complementary foodsthatareoftenlowinhighqualityproteinandmicronutrients, suchasrice,wheat,maizeorstarchyrootsandtubers. Previous researchhasfoundthatcalorieintakealoneisnotalwaysastrong predictorofchildgrowthindevelopingcountriessettings(Griffen, 2016), perhaps because calorie requirements for infants are relatively modest. Instead, many researchers point to low consumption of animal-sourced foods (ASFs) as a critical constraint (Allen, 2003; Brown, 2003; Demment et al., 2003;

HeadeyandHoddinott,2016;Neumannetal.,2002;Puentesetal., 2016;Randolphetal.,2007).Indeed,in theabsenceoffortiﬁed foods, young children cannot meet their micronutrient needs withoutdailyintakeofASFs(PAHO/WHO,2003).

DairyconstitutesaparticularlyimportantcomplementaryASF for young children because of the familiarity of its taste to exclusively breastfed children, and because of its nutritional proﬁle.Dairyishighinallthreemacronutrients(energy,fatand protein),aswellasimportantmicronutrientssuchasvitaminA, vitaminB12,andcalcium(MurphyandAllen,2003).Moreover,like

*Correspondingauthor.

E-mailaddress:d.headey@cgiar.org(D.D.Headey).

https://doi.org/10.1016/j.ehb.2018.07.001

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)

otherASFs,dairy contains several essential fattyacidsthat are hypothesizedtobecritical forprocesses ofcellulargrowth and bone formation (Semba et al., 2016). Dairy has a protein digestibilitycorrectedaminoacidscore(PDCAAS)ofabout120%.

Many studies also suggest dairy intake affects child growth throughastimulatingeffectonplasmainsulin-likegrowthfactor1 (IGF-1).Milkalsocontains mineralssuchaspotassium,magne- sium, and phosphorus, that could also be a factor stimulating growth,aswellaslactose.

Consistentwiththisbiologicalevidence,arangeofresearchhas linkedlineargrowthtochildhooddairyconsumption,albeitmostly indevelopedcountrysamples(Iannottietal.,2013;deBeer,2012;

DrorandAllen,2011;Wiley,2005,2009;Sadler&Catley2009).In developing countries there have been remarkably few efﬁcacy trials of dairy supplementation ongrowth in infants or young children, though several dairy consumption programs have demonstratedsomeimpactonlineargrowthatolderages(Iannotti etal.,2013).¹

Becauseof the limitations of experimental evidence onthis subject, economists have increasingly utilized observational or quasi-experimentalanalysestoexploretheassociationsbetween dairyproductionandchildnutritionoutcomesinlessdeveloped settings.In economichistorystudies,Baten(2009,2014)testsa

“protein proximity” hypothesis with 19^th Century European militaryrecruitmentdatafromCentralEurope.Utilizingtheidea thatfreshmilkintheseeconomiescouldnotbetradedoverlarge distances, heﬁnds that adult menin closerproximity todairy productionweresubstantiallylesslikelytobetooshortformilitary recruitment.Still otherstudies hypothesizethat trends inmilk consumptionexplainlongertermsecularimprovementsinheights at later stages of economic development, suchas 20^th Century Japan(Takahashi,1984)andIndia(Mamidietal.,2011).Arecent paperalsoexaminedadultheightsin42Europeancountrieswith varyinglevelsofdevelopment.Evenaftercontrollingforgenetic factors,theyfoundthatthenationalsupplyofproteinfromdairy products was the single strongest predictor of adult stature ( Grasgruberet al., 2014).A related studyof 105countries from different continents also found strong associations between averagemilkconsumption levelsandadultmaleheights (Gras- gruberetal.,2016).

In contemporary developing countries several studies have examined associations between household livestock ownership andchildgrowthoutcomes,thoughnotallstudiesfocusonmilk- producing animals specifically. Like Baten (2009, 2014) these studiesassume(oftenimplicitly)thatfreshmilkisgenerallynon- tradableandnotaperfectsubstituteforpowderedmilk.Hoddinott etal.(2015) usetwo largesurveys fromEthiopiatospecifically exploretheassociationbetweencattleownership,dairyconsump- tionandHAZscores.Theycitethefactthat90%ofmilkproducedin rural Ethiopia is consumed by the household producing it, implyingthatcattleownershipoughttobeaverystrongpredictor ofregulardairyintake.Consistentwiththatconjecturetheyfind strongpositiveassociationsbetweencattleownershipandHAZ(as highas0.47standarddeviationsinthe12–23monthage-range).

Theyalsoimplement placebo teststoexplore theconcernthat cattle ownership proxies for generic wealth effects on child nutrition.

Rawlinsetal.(2014)evaluateHeiferInternational’sdairycow and goat ownership programs in Rwanda, albeit in a non- randomized quasi-experimental design with a small sample of 217 children aged 0–59 months (precluding the possibility of detailedagedisaggregation).Theyﬁndthatchildrenfromhouse- holdswhoreceivedagoat12monthspriortothetimeofthesurvey saw no growth differentialover controls, whereas transfers of pregnant cows (high-productivity foreign breeds) improved height-for-age Zscoresby0.57standard deviations,a largebut impreciselyestimatedeffect.Similarly,Kabungaetal.(2017)use matchingmethodstogaugetheimpactsofadoptionofimproved dairycowvarietiesonHAZofchildrenaged6–59months.They ﬁndHAZimpacts of0.48-0.49 standarddeviations, thoughalso someevidenceoflargerimpactsforhouseholdwithgreaterherd sizesorlargeracreage.²

Overall, there is fairly consistent evidence that dairy cow ownershipisassociatedwithchildgrowthinpoorerpopulations, althoughthereareseverallimitationsandcaveatssurroundingthis evidence.First,theevidenceisconfinedtoEastAfricanlocalities wherecattleownershipisrelativelycommon,soexternalvalidity isaconcern.Second,thisliteraturepotentiallysuffersfromseveral internalvalidityissues,includingtheconfoundingroleoflivestock asa sourceofimperfectlymeasuredruralwealth,andpotential concerns over associations between livestock ownership and ethnicity.³ Another outstanding concern not addressed in the previous literatureisthat theavailabilityofcow’smilkleadsto premature cessation of breastfeeding by mothers. Exclusive breastfeedingisstronglyrecommendedforthefirst6monthsof life,especiallyindevelopingcountrysettings,becauseofitscritical roleinpreventingdiarrheaandrespiratoryinfections(Hortaand Victora, 2013), and becausecow’s milk can stress a newborn’s immaturekidneysandirritatetheliningofthestomachandsmall intestine,leadingtobloodlossandiron-deficiencyanemia(FAO, 2013).

Inlightoftheselimitations,thispaperutilizesauniquedataset toattemptamorecomprehensiveassessment ofthenutritional implicationsofdairyproductionandconsumptioninBangladesh.

Bangladeshisaparticularlyimportantcasestudyinthecontextof dairy production.Inadditiontoitshighratesofstunting(36%), HeadeyandHoddinott(2016)emphasizethatBangladeshhasan under-diversiﬁedfoodsupply,withFAOdatasuggestingthatASFs account forless than5% oftotal caloriessupplied (Headeyand Hoddinott, 2016).Thissituationpartlystemsfromexceptionally lowlevelsofmilkconsumption,whichinpercapitatermsisless thanhalfthatofneighbouringIndia(HeadeyandHoddinott,2016).

Alikelyexplanationofthisisthecountry’sexceptionallysevere landconstraints(andhencefeedconstraints),withaveragefarm sizes in Bangladesh averaging just half a hectare, and rural landlessness widespread. It may alsobe that cultural norms – historical unavailability of milk – has kept demand for milk relativelylow.

InthispaperweusethenationallyrepresentativeBangladesh Integrated Household Survey (BIHS) of rural areas, which was conductedovertworoundsin2011/2012and2015.Uniquelyfor

1 Observationalevidenceonthelinkagesbetweendairyconsumptionandchild growthisalsolimitedbythepaucityofhighqualitydataon“usualdiets”in developingcountries.ThewidelyusedDemographicHealthSurveys(DHS)now24- hourrecallindicatorsoffoodconsumption,but thisislikelyarelativelypoor indicatorofregularconsumptionofdairyproductsinmanysettings.Somechildren whodidconsumemilkinthelast24hoursmaynotberegularconsumersofmilk, andviceversa.Thismisclassiﬁcationmayleadtoattenuationbiaseswhentryingto estimatetheimpactsofmilkconsumptiononchildgrowthwithobservationaldata.

2Inadditiontothesestudies,otherstudieslookatlivestockownershipandHAZ without specificallydistinguishing livestock breeds nordairycow ownership specifically.Mositesetal.(2015)findsignificantnegativeassociationsbetweentotal livestockownershipandstuntinginEthiopiaandUganda,butnotinKenya.Azzarri etal.(2015)applyaninstrumentalvariable(IV)approachtoasmallerUgandan householdsurveyandfindnoimpactoflargeruminantsonstuntingoutcomes,but somenegativeassociationswithunderweightstatus.

3Hereweonlyreviewpublishedstudiesoncattleownershipandchildgrowth.

Rawlinsetal.(2014)reviewseveralmuchearlierunpublishedstudiesonthistopic, thoughallinvolveverysmallsamplesizes,andallstillpertaintoEastAfrica(Kenya, Malawi,Rwanda).

(3)

suchalargesurvey,thisdatasetcontainsrichinformationbothon nutrition outcomes, individual food consumption, agricultural assetsandproduction,andarangeofotherpotentialdeterminants ofnutrition.Methodologically,weproposeanoveldifference-in- difference approach to assessing the impact of dairy cow ownershiponchildnutritionoutcomes,bydistinguishingbetween householdswithlactatingdairycowsthathaveproducedmilkover thepast12months(treatment),householdswithcowsthathave notproducedmilkinthepast12months(placebo),andhouseholds thatdonotownanydairycows(control).Wenotethatthisisnota placeboin themedical deﬁnition(according towhich a person consumesatreatmentofnointendedtherapeuticvalue),butinthe sense that non-lactating cows might have a similar long run economic value any direct milk supply tothe household. This distinctionbetweenthetreatmentandplaceboemergesfromthe factthatsmallholderdairyproducersinBangladeshtypicallyonly ownafewcowsbecauseoftheextremelandandfeedconstraints mentionedabove.Speciﬁcally,80%ofBangladeshifarmersinour nationally representative sample own just 1–2 cows and no farmersinoursampleownmorethan4animals.Giventhatatany giventimeallorsomeofthesecowswillnotbelactating–since thereisaminimum12-monthinter-calvingcycleforeachanimal even among the most technologically sophisticated dairy pro- ducers–thereisanon-trivialproportionofdairycowownersin Bangladeshwhowouldbeunabletoproducemilkonacontinuous basisforexogenousbiologicalreasons.⁴

Ineffect,then,thecombinationofsmallherdsandabiologically determinedcomponentofthelactationcyclepotentiallycreatesa validplacebo groupofchildrenwhoaretreatedwithcowsthat havenotproducedanymilk.Wethereforetestthreehypotheses:

(i)Children in treatmentgroup willbe taller than childrenin control;

(ii)Children in the placebo group will not be taller than the control;and

(iii)Childrenintreatmentgroupwillbetallerthanplacebogroup children.

In additiontothesetestswealsoexaminewhetherlivestock ownershipormilkproductionisassociatedwithotherobservable potentially confounding factors, such as maternal nutritional knowledge and empowerment, and overall child diversity, exclusiveof milk.And unlikepreviousstudiesin this literature weexplorethepolicy-relevantquestionof whetheraccess toa stablehouseholdlevelsupplyofdairyproductsleadstosubstitu- tionbetweenbreastfeedinganddairymilkintake.

Wefindthatmilkproductionisstronglyassociatedwithlinear growth,butonlyforchildreninthecrucialfirst1000daysoflife (particularlythe12–23monthrange).Theeffectsweobserveare verycloseinmagnitudetothoseobservedintheaforementioned quasi-experimentalstudybyRawlinsetal.(2014)forRwandaand Kabunga et al. (2017) for Uganda, but larger than the more observationalstudybyHoddinottetal. (2015)whoanalysethe impacts of owning any cow, rather than milk-producing cows specifically(rendering theirresultsmore likean intent-to-treat analysis).Nullresultsfortheplacebogroupalsolendcredenceto the identification assumptions underlying our approach, as do additionalplacebotestswhichruleoutsystematicdifferencesin nutritionalknowledgeandwomen’sempowerment.However,we

dofindsomeevidenceofpotentiallyharmfuleffectsofhousehold dairy availability on breastfeeding in the first year of life, suggestingdairy-orientednutritionstrategiesneedtoproactively promoteexclusivebreastfeedinginthefirstsixmonthstoprevent prematuresubstitutionintodairy.

Theremainderofthispaperisorganizedasfollows.Section2 describesthedataandthemethodsusedtoanalysethem.Section3 testsassociationsbetweendifferentASFproductionandvarious nutritionoutcomes.Section4providessomeimportantsensitivity testsandextensions,andSection5concludeswithadiscussionof theimplicationsoftheseﬁndingsforprogramsandpolicies,aswell asforfutureresearch.

2.Conceptualmodel,dataandmethods

As outlined above,ourobjective in this paper is totest for significant differences in milk consumption and child growth betweenhouseholdgroupsthataredefinedbydairyproduction andcowownership.Previouspapersinthisliteraturehavetended to focus on a comparison between a “treatment group” of households that own any dairy cow and a “control group” of householdsthatdonotownanydairycows.Inourdataweinstead narrowthedefinitionoftreatmenthouseholdstothosethatowned cowsthatactuallyproducedmilkinthepast12months(hereafter treatment).Wethendefinewhatcanbethoughtofasa“placebo group”ofchildrenexposedtocowsthathadnotproducedanymilk inthepast12months(notethatwethinkofthisgroupasaplacebo becausethetreatmentisnotmilkperse-inwhichcasetheplacebo wouldbeamilksubstitute-butmilk-producingcows).Inanideal experimentaldesignchildrenwouldberandomlyassignedacross groups,butinobservationalsettingsasignificantconcernisthat theremaybesystematicnutrition-relevant differencesbetween treatedandnon-treatedchildren(e.g.wealth,nutritionalknowledge, women’s empowerment). Achieving more experimental conditionsmightthereforerequireextensivecontrolforpotential confoundingfactors.

TheconceptualmodeldescribedinHoddinottetal.(2015)isa usefulstarting pointfor thinkingaboutthevariousfactorsthat mightinﬂuencehouseholddecisionmakingprocesseswithrespect todairyproduction,dairyconsumptionandchildnutrition.They posit a householdutilitymodel inwhich child nutrition isone argument.Nutritionalstatusisitselfafunctionofnutrient(food) intake, as well as nutritional knowledge, culture, healthcare, geneticendowments,and locationalcharacteristics(suchasthe prevalenceofdisease;accesstoinformationaboutgoodchildcare practices).In aworldofperfectlyfunctioning markets,nutrient intakewouldbeprimarilyinﬂuencedbyincome,andhouseholds could sequentiallymaximize farm and nonfarm income before decidinghowtospendthatincomesoastomaximizenutrition outcomes subject to other arguments in the utility function.

However, the perishability of milk in poorly developed value chainsrendershouseholdproductionandconsumptiondecisions non-separable. Inother words,if households struggle toaccess affordablemilkviamarkets,theycouldopttoowndairycows.This impliesthatthedecisionstoowndairycowsand/orproducemilk maybeendogenous,inﬂuencedasitisbynutritionknowledgeand farm production parameters such as the availability of capital (income,savings,wealth),accesstoland(feed),accesstoinputand outputmarketstoobtainfeedandsellproduce,householdlabour supply, farm management skills, and the role of women in household decisionmaking, including dairy production and feedingpractices.

Sinceomissionofthesekindsoffactorscouldleadtobiased coefﬁcientsontheimpactsofcattleownershipormilkproduction onchildgrowth,ourempiricalmodelsneedtocontrolforthese factors as extensively as possible. Fortunately, the Bangladesh

4Other potentiallyendogenousdeterminants ofthe lactation cycleinclude seasonaldiseasesandheatstress,landaccess,poormanagementpracticesrelated tooestrusdetection,pooranimalnutrition,andpooraccesstomalecattleor artiﬁcialinsemination services(Shamsuddinetal.,2007;Kamal,2010).Kamal (2010)writes:

(4)

Integrated HouseholdSurvey (BIHS) not onlycontains detailed dataonchildren’sfoodintakeandnutritionoutcomes,butalsoan exceptionally richarrayof dataonincome, wealth, agricultural productionandassets,accesstomarkets,women’sempowerment andwomen’snutritionknowledge(IFPRI,2016).BIHSisalsoalarge surveyrepresentative of ruralBangladeshthat hasbeenimple- mentedin two rounds (2011/2012and 2015) and constitutes a panelforthemajorityof households.However,becauseweare interestedinchildgrowthintheﬁrst5yearsoflife–particularly the12–23monthperiod–wetreatbothroundsasrepeatedcross- sectionsrather thana panel.⁵ Thecombinedroundsmake upa sample of 11,796 households (some surveyed twice), which includes4268pre-schoolchildrenaged0–59months.

Height-for-age Z scores (HAZ), using the World Health Organization’s global child growth reference standards (WHO, 2006), constitutes our primary outcome of interest. As noted above,from6monthstoaround24monthsgrowthfalteringtends tobeparticularlypronouncedindevelopingcountrypopulations duetoprolongednutritionaldeﬁcienciesassociatedwithinappro- priatecomplementaryfeedingandrepeatedorchronicinfections (Victora et al., 2010). It is also common to deﬁne children as stuntedifHAZfallsbelow-2,thoughstatisticalepidemiologists have strongly argued against using dichotomous dependent variables,as it unnecessarily discardsvaluable information and reduces precision (Royston et al., 2006). However, we report stuntingresultsasanextensiontoourmainHAZresults.

In this paper our interest is in dairy production-dairy consumption pathways, rather than dairy production-income/

wealth pathways (in principle, income from any source could improve diets). Our regression models therefore control for householdexpenditure and wealth, butourdatasetalso allows ustoexaminewhetherdairyconsumptionislikelytobethemain mechanism linking cow ownership to child growth by using additionaldataonchildren’sconsumptionofvariousfoodsaswell ashouseholddataonhowdifferentfoodswereobtained.Interms of theformer weprimarilyfocus onchildren’s consumption of dairyproductsinthepast24h,deﬁnedasadichotomousindicator.

Tohelprulemoregenericincome-basedpathwayswealsousea dietarydiversityscore(0–6foodgroups)thatexcludesdairy,as wellasestimatesofchildren’stotalcalorieconsumption(exclud- ing breastmilk). Our expectation is that dairy production inﬂuencesdairyconsumption,butnotnon-dairydietarydiversiﬁ- cationortotalcalorieintake.Wecanalsoexplorehowhouseholds sourced different foods since the BIHS asks respondents to estimatetheproportion of each foodprovided throughmarket purchases, provided by other sources, or provided by home production.Wealsonotethat,inprinciple,theseconsumptiondata mightalsobeusedtoexaminetheimpactsofdairyconsumption on child growth. However, a critically important limitation of consumptiondataisthattheyarebasedonshortrecall periods (24-hourorweeklyrecall),meaningthattheyarepotentiallyquite poorindicators of regular consumption of milk in thepast 12 monthsormore(Thorne-Lymanetal.,2014).

This measurement problem with short-recall consumption suggeststhatlongerrecallquestionsonmilkproductionmaybea much better indicator of regular access to dairy products in settings where markets for perishable products are highly

imperfect.However,sincelong-recallproductionquantityindica- torsalsosufferfrombiasweuseasimplerdichotomousindicator ofwhetherornotmilkwasproducedinthelast12months–along withcowownership-todeﬁneourtreatment,placeboandcontrol groups.

Clearlythesegroupsarenottheresultofrandomassignment, althoughwecanusemultivariateregressionstoreducethebiases ofconfoundingfactorsthatinfluencecowownershiporlactation decisions. Wefirst assess thedeterminants ofmilk production, withtheexpectationthatcattleherdsize(femaleandmales)isa keyobservabledriverthatwecansubsequentlycontrolforinour main HAZ regressions.We then usemultivariate reduced form regressionstocontrolforabroaderrangeofpotentialconfounding factors.Inadditiontodairyherdsize,wewerealsoconcernedthat cattleownershipmaysimplyreflectmoregenericlivestockwealth, sowe extensivelycontrol forother forms oflivestock (bullock/

buffalo,goat,sheep,chicken,duckandotherbirds)andaggregate livestockintoanindexofTropicalLivestockUnits(TLU),whichcan be thoughtof as a measure of aggregate livestock wealth. The remainingcontrolvariablesaremorecommontomostnutrition specifications,andtoestimationofhealthproductionfunctions, suchasToddandWolpin(2007)andHoddinottetal.(2015).This includes child characteristics (sex, age, breastfeeding status), parentalcharacteristics(ageandschooling),householdcharacter- istics(percapitamonthlyexpenditure,theaggregatevalueof26 householdassets, hectares ofcultivableland owned,household toilet and water access, access to electricity, exposure to NGO services)andseveralcommunitycharacteristics(distancestothe nearestweekly/periodicoutdoormarket,andtothenearesttown andtothenearesthealthcentre).Ourregressionsalsoincludefixed effectsforall65districtsinwhichtheBIHSwasconducted.Clearly themaincoefficientofinterestisthatpertainingtothetreatment group,whichweinterpretastheeffectofdairyavailabilityonchild growthnetofanyimpactsofdairyproductiononotherinputsinto the health production function, suchas income, or changes in breastfeeding. However, we also test for significant differences betweenthecoefficientsfortreatmentandplacebo,andwhether thecoefficientforplaceboissignificantlydifferentfromzero(i.e.

from the control, the omitted control group). A significant coefficient on placebo would suggest that cattle ownership influenceHAZthroughchannelsotherthandairyconsumption.

A biological issue of paramount importance is the need to explore age-speciﬁc variation in the sensitivity of children’s growthtoexposuretodairyproduction,anissueemphasizedin Hoddinottetal.(2015).FortheHAZanalysisweprimarilyfocuson children6–23monthsand24–59months,aswellassmallerage intervals.ThebiologyofgrowthidentiﬁedinVictoraetal.(2010) suggeststhatmostgrowthfalteringtakesplaceinthe6–23month window,sodairyconsumptioninthisperiodoughttobecritical.

Wedoreportresultsforolderchildren(24–59),althoughitisnot clearthatour12-monthdairyproductionindicatorshouldpredict strongergrowthbecauseofmisclassiﬁcationerrors.Thatis,some 24–59monthchildrenwhomayhaveconsumeddairyinthepast 12months(accordingtoourindicator)maynothaveconsumed dairyintheircritical6–23monthwindow.

In ourextensions to the basicmodel we also examine two indicatorsthatwerenotcollectedforallhouseholdsand would thereforeentailsamplerestrictions:maternalnutritionknowledge score and a maternal empowerment score based on women’s controloverandownershipofvariousagriculturalassets.Weuse these indicators as dependent variables to test whether dairy producing households are signiﬁcantly more likely to have mothers with betternutrition knowledge or greater empowerment. Here wetest the null hypothesesthat thecoefﬁcienton treatmentisequaltothatofplaceboandcontrol.Rejectionofthis null wouldcastdoubtmightsuggestthatpartoftheestimated

5 Applyinghouseholdﬁxedeffectstolookatwithin-householddifferencesin siblings’exposuretodairyproductionwouldbepossibleinprinciple,butwould requireanevenlargersamplethanweuseinthispaper.Thisisbecausewealready splitthesampleinto12-monthagebrackets,only25%ofhouseholdsowndairy cows,andonlyaquarterofthesehavenotproducedmilkinthepast12months.

Fertility rates in Bangladesh are nowsufﬁciently low (2.9) thatrelativefew householdshavemultiplechildreninthe0-59monthagebracket,letalonethe6-24 monthwindowofinterest.

(5)

effectsof milkproduction onHAZpertainstogreater nutrition knowledgeorempowerment.Wealsoestimate alternativeHAZ speciﬁcationwhereproductionquantitiesofmilkareusedinplace of thedummy variablefor any milk produced. Thisis not our preferredindicatorbecauseofconcernsovermeasurementerror, relatedtothechallengesofaccuratelyrecallingproductionovera longperiod,butweneverthelessconsideritausefulalternative test.

3.Mainresults 3.1.Descriptivestatistics

Table1providesdescriptivestatisticsforthekeyvariablesfora sampleofchildren0–23monthsofage.Fig.1alsoreportsalocal polynomialsmoothercurve(LPOLY)ofHAZscoresagainstchildage toreveal thedynamicsof growthfalteringin ruralBangladesh.

Thereareseveralbroadinferencestobemadefromtheseresults.

First, the sample of children is highly undernourished, consistent with other nationally representative surveys of Bangladesh.MeanHAZscoresare 1.37,andonethirdofchildren arestunted(byagetwofullyhalfarestunted).However,consistent withpreviousresearch(Victoraetal.,2010),mostofthegrowth falteringin Bangladesh occurs in the 6–23 month window, as shownbytheredverticallinesinFig.1.Thisacceleratedperiodof growthfalteringcouldpartiallybeduetopoordiets.Notably,the percentageofallchildrenwhoconsumeddairyinthepast24his just22%,whichisparticularlylowgiventhatinmoredeveloped societies many children would consume milk ona daily basis.

Consistentwithlowmilkconsumption isthelowownershipof milkproducingcows(14%),whileafurther8%ownacowthathas notproducedmilkinthepast12months.

3.2.Determinantsofmilkproduction

Thehighersocioeconomicstatusoftreatmentmightimplythat any apparent benefits of dairy production partially reflect the benefits of greater socioeconomic status. This points to the importance of multivariate regression models saturated witha widearrayofcontrols,aswellastheimportanceofplacebotests.

However, we can also examine the determinants of milk productionacrossamonghouseholdsthatownatleastonecow (treatmentandplacebo)toassesstherelativeimportanceofherd sizeversusothersocioeconomicindicators.Onbiologicalgrounds onewouldexpectmilkproductiontobestronglyassociatedwith herdsize,includingthenumberofbothfemaleanimalsandmale animals.Owningmorefemaleanimalsobviouslyreducestherisk thattheherdasawholewillnothaveproducedanymilkinthepast 12months.However,withoutmaleanimals,producerswouldneed toeitherrentinbulls,oraccessartiﬁcialinseminationservices.

While the latterare common in Bangladesh, previousresearch pointstopoorfarmmanagementpracticesreducingthesuccessof artiﬁcialinseminationservices(seefootnote3).

Table2reportstheresultsforthosevariablesthatstatistically explainwhetheror nota cow-owninghouseholdhasproduced milkinthepast12months.Withtheexceptionofmaternalage,the onlysigniﬁcantpredictorsofdairyproductionstatusareindicators of herd size; coefﬁcients on the range of other indicators of

Table1

Descriptivestatisticsforchild,householdandcommunityleveldataforasampleofchildren0–23monthsofage.

Variable Obs Mean Std.Dev. Min Max

Height-for-ageZscore(HAZ) 1596 1.37 1.59 5.87 5.83

Stunted 1,596 0.34 0.47 0.00 1.00

Treatment:Ownscow(s),producedmilk 1,596 0.14 0.34 0.00 1.00

Placebo:Ownscow(s),nomilkproduced 1,596 0.08 0.27 0.00 1.00

Childconsumeddairylast24hrs 1,588 0.22 0.41 0.00 1.00

Quantitymilkproduced(liters),last12m 1,312 43.74 140.42 0.00 1500.00

Numberofbullocks 1,596 0.43 0.91 0.00 8.00

Numberofcows 1,596 0.38 0.83 0.00 4.00

Ownscow,producedmilk 1,596 0.14 0.34 0.00 1.00

Ownscow,nomilk 1,596 0.08 0.27 0.00 1.00

Ownsgoat/sheep 1,596 0.15 0.35 0.00 1.00

Ownspoultry/duck/otherbirds 1,596 0.63 0.48 0.00 1.00

Owns/producesﬁsh 1,596 0.29 0.45 0.00 1.00

Totallivestockowned(TLUs) 1,596 0.68 1.20 0.00 25.80

Currentlybreastfed 1,588 0.50 0.50 0.00 1.00

Logpercapitamonthlyexpenditure 1,596 7.69 0.52 6.37 10.71

Logvalueofhouseholdassets 1,596 10.99 1.26 6.17 17.96

Landareacultivated 1,596 0.22 0.43 0.00 6.43

Accesstoelectricity 1,596 0.54 0.50 0.00 1.00

Motherprimaryeducation 1,596 0.54 0.50 0.00 1.00

Mothersecondaryeducation 1,596 0.06 0.24 0.00 1.00

Mothertertiaryeducation 1,596 0.04 0.19 0.00 1.00

Householdheadprimaryeducation 1,596 0.34 0.47 0.00 1.00

Householdheadsecondaryeducation 1,596 0.08 0.28 0.00 1.00

Householdheadtertiaryeducation 1,596 0.04 0.19 0.00 1.00

Accesstowatersupply 1,596 0.78 0.41 0.00 1.00

Accesstoimprovedtoilet 1,596 0.32 0.47 0.00 1.00

Distancetoregularbazaar(km) 1,596 1.79 1.79 0.00 25.00

Distancetohealthcentre(km) 1,596 6.10 6.34 0.00 70.00

LoanfromNGO 1,596 0.46 0.50 0.00 1.00

Malechild 1,596 0.52 0.50 0.00 1.00

Householdsize 1,596 5.55 2.29 2.00 21.00

Maternalage 1,596 25.84 5.67 16.00 51.00

Nutritionalknowledgescore 1,596 8.79 1.91 0.00 14.00

Maternalempowermentscore 1,113 0.70 0.23 0.10 1.00

Childdietdiversity(6groups,excludingdairy) 1,588 1.85 1.76 0.00 6.00

ChildCalorieIntake(kcal) 1,596 286.25 331.50 0.00 2919.28

Source:BangladeshIntegratedHouseholdSurvey2011,2015.

(6)

householdsocioeconomicstatusareallinsigniﬁcant,individually andjointly.Theresultssuggestthatmilkproductionstatusisnon- linearly relatedtoherdsize:owning 2 dairycowsor 1bullock greatlyincreasestheprobabilityofproducingmilkinthepastyear, butadditionalanimalsdonotmuchaltertheseprobabilities.

Fig.2explorestherelationshipbetweenherdsizeandannual milkproductiononthey-axisandthenumberofcowsownedon thex-axis.However,weplotacurveforhouseholdsthatownat

leastonebullock,aswellasthosethatdonot,inordertoexamine interactioneffects. Theresultsreveal theexpectedﬁnding that owningjustonecowwithnobullockresultsinverylowlevelsof milkproductionbecausethereisahighlikelihoodthatthissingle cowmaynothavebeenlactatingatanytimeinthepast12months.

Owningmorecowsgreatlyimprovesmilkproduction.Moreover, thereturnstoowningonecowandatleastonebullockarefairly high,andnotgreatlyincreasedbyowningmorecows.

Fig.1.Alocalpolynomialgraphofheight-for-ageZscoresbychildageinruralBangladesh.

Source:Authors’estimatesfromBangladeshIntegratedHouseholdSurvey2011,2015

Table2

Statisticallysigniﬁcantdeterminantsofmilkproduction(treatmentstatus)inpast12monthsamonghouseholdsthatownedatleastonecow(linearprobabilitymodel).

(1)

Producedmilkinpast12months(i.e.treatmentgroupstatus)

Owns2cows 0.349^***

(0.052)

Owns3cows 0.460^***

(0.054)

Owns4cows 0.334^***

(0.083)

Owns1bullock 0.267^***

(0.054)

Owns2bullocks 0.363^***

(0.049)

(0.081)

Owns4bullocks 0.173*

(0.100)

(0.071)

Owns6bullocks 0.162*

(0.091)

Mother’sage 0.008^***

(0.003)

Alllivestockownershipvariables? Yes

Controlsforageandgender? Yes

Othersocioeconomiccontrols? Yes

Districtﬁxedeffects? Yes

Observations 728

R-squared 0.405

Notes:Thesearelinearprobabilityestimates,withstandarderrorsareinparentheses,clusteredatvillagelevel.

***p<0.01,**p<0.05,*p<0.1.

ControlvariablesaredescribedinTable1.Source:BangladeshIntegratedHouseholdSurvey2011,2015

(7)

3.3.Associationsbetweendairyproductionanddietaryindicators

Fig.3showslocalpolynomialsmootherplotsoftherelationship between 24-hr dairy consumption and child age, with 90%

conﬁdenceintervals(CIs).Weuse90%CIsinordertoimplement aonesidedtestatthe5%levelthattreatmentstatusisassociated withhigherHAZ.Panel(i)comparestreatmenttoplacebo,while Panel(ii)comparestreatmenttocontrol.The90%CIsdonotoverlap in either panel,indicating that treatmentchildrenhave signiﬁ- cantlyhigherlevelsofdairyconsumptioncomparedtotheplacebo or control groups throughout the 0–59 month age range. The magnitudeof thedifferencebetween treatmentandplaceboand controlvariesbetween15–25percentagepointsdependingonthe ageofthechild.

Table3examinesthisrelationshipinamultivariateregression modelwithafullsetofcontrols,butalsolooksatwhethermilk productionhasanyimpactonnon-dairydietarydiversityandchild calorie intake. Results in Regression (1) suggest that milk production leads toapproximately a 14-point increase in milk consumption,althoughsomeofthedifferenceinmilkconsump- tionacrossgroupsobservedinFig.3islikelydrivenbydifferences insocioeconomicstatus(householdexpenditure,maternaleduca- tion)acrossgroups.⁶ Another strikingresult fromFig.3 is that manychildrenundertheageof12monthsconsumecow’smilk, eventhoughrecommendations(albeitbasedmoreondeveloped countrysamples)recommendmilkconsumptionbeinitiatedonly at12months(FAO,2013).Moreover,previousresearchusingthe samedatasetsuggeststhatchildrenareoftengiventhelion’sshare of a household’s milk supply in Bangladesh (Sununtnasuk and Fiedler,2017).

Finally, Table 3 alsoexamineswhether thereare systematic differencesinnon-dairydietarydiversityacrossgroups,aswellas totalcalorieintake(exclusiveofbreastmilk).Wefindnosignificant associationsbetweentreatmentandthesetwodietaryindicators, although the placebo group hashigher calorie intake than the treatmentorcontrolgroups.Thelackofanyimpactonnon-dairy dietarydiversitysuggeststheresultsmaynotbeconfoundedby socioeconomicdifferences between groups (Hoddinott, Headey and Dereje 2014). The lack of a significant impact on calories

suggeststhatmilkconsumptionisnotprimarilyoperatingthrough increasingachild’soverallcalorieintakeinthiscontext.

3.4.Associationsbetweendairyproductionandchildgrowth

Table4presentsleastsquaresregressionresultswithafullset ofcontrolvariables,stratifiedby6–23months,24–59months,and then by series of overlapping 12-month age brackets used to furthercorroboratetheimportanceof milkin this6–23month window. The most striking result is the large 0.52 standard deviation(SD)differencebetweentreatmentandcontrolchildren inthe6–23monthwindow;adifferencewhichentirelydisappears inthe24–59monthwindow.Thelatterresultislikelyexplainedby thefact that theremaybelow serial correlationbetweenmilk productioninthepastyearandmilkproductioninearlieryears, preciselybecauseofvariationsin lactationcyclesamong small- scaledairyproducers.Incolumns(3)and(4)weseethattheresults areconsistentacrossthe6–17monthand12–23monthwindows, though column (4) shows a relatively large but insignificant coefficient on the placebo group coefficient, while column (5) confirms that the benefits of milk production are no longer apparent once we move above the 23 month threshold. We interpretthisasevidencethat milkconsumption hasitslargest impactinthefirst1000days;astheagerangemovesbeyond23 monthsthe12monthrecallbecomesamoreimpreciseindicatorof whether thechild actually consumed milk in the 6–23month period.Furtherconfirmationthattheresultsarestrongestinthe6– 23 month period is provided by Wald tests of significant differences between the treatment and placebo coefficients in the 6–23 month, 6–17 month and 12–23 month ranges. This suggeststhatitismilkproduction,notcattleownershipperse,that yieldssizeablebenefitsforlineargrowthinearlychildhood.

3.5.Extensions

Inadditiontotheresultsabovewealsoengagedinaseriesof extensionsdesigned toexploresomeadditionalcomplexitiesin theassociationsexaminedabove.Wefirsttestedfordifferential impactsoftreatmentonboysandgirls,butfoundnostatistically significantdifferencesinresultsfortheagerangesabove.Wealso tested for interactions between treatment status and maternal empowermentscoresandmaternalnutritionalknowledgeonthe groundsthatthesemightbemediatingfactors,butallinteractions wereinsignificant. We also includedempowerment scoresand knowledgescoresasdependentvariablestoseeifthesemightbe potential confounding factors, but treatment status had no significantimpactoneithervariable(resultsavailableonrequest).

InTable5weusedstuntingstatus(HAZ<-2)asthedependent variable.Stuntingisawidelyusedpublichealthmeasure,although usingadichotomousindicatorratherthanacontinuousindicator effectivelydiscardsinformationandislikelytoreduceprecision.

ThepatternofresultsinTable5areverysimilartothosereported inTable4,althoughtheWaldtestsnolongerreportstatistically signiﬁcant differencesacross thetreatment and controlgroups (seemingly due to the expected increase in imprecision). That caveataside,theresultsimplythatregulardairyconsumptionhas strong impacts on stunting, although treatment-control and treatment-placebo comparisons yield quite different inferences.

Among children 6–23 monthsthe model predicts a 10.4-point reductioninstuntingrelativetothecontrolgroup.However,the placebogroupalsohasalarge,negativebutstatisticallyinsigniﬁ- cantcoefﬁcientthat–interpretedliterally–wouldimplyonlya 2.4-point reduction in stunting from exposure to treatment.

Amongchildren12–23and18–29monthsthepointestimateson treatmentareevenlarger,implying14and22-pointreductionsin theriskofstuntingrelativetocontrol,and8.4-pointand11.3-point Fig.2.Meanmilkproductionasafunctionofthenumberofcowsowned,for

householdsownanddonotownbullocks.

Source:Authors’estimatesfromBangladeshIntegratedHouseholdSurvey2011, 2015

6Indeed,BIHSdataonmilksourcedfromownconsumptionshowsthatmilk- producinghouseholds acquirearound three-quartersoftheir householdmilk supplyfromownproduction,suggestingtheystillrelysubstantiallyonmarketsto supplementhouseholdconsumptionrequirements.

(8)

reductionsrelativetotheplacebogroup.Overall,then,theseresults forstuntingstatusarebroadlysimilartotheHAZresultsinTable4, althoughitisnolongerpossibletoestablishstatisticallysigniﬁcant differencesbetweentreatmentandplacebo.

Analternativetomodellingadichotomousindicatorofwhether thehouseholdproduced any milkis tospecify thehousehold’s estimateofthequantityofmilkitproducedinthepast12months, whichwemeasureasthelogoflitresperchild.OLScoefficients estimates for this indicator are reported in Table 6. These coefficientsaresignificantinthe6–23monthand12–23month brackets,andmarginallyinsignificantinthe6–17monthbracket.

Inthe6–23monthrangethecoefficientimpliesthat increasing milkproductionby10%wouldreducestuntingby0.08percentage points.Thecoefficientsareimpreciselyestimated,however,and likely suffer from attenuation bias related to the significant

challenges that respondents have in accurately answering 12- month recall questions.Overall,though,theresults arebroadly consistentwiththeresultsfromTable4.

4.Exploringtherelationshipbetweendairyproductionand breastfeeding

OneconcernwiththeresultsreportedinFig.2 isthatmany children in the treatment group consume dairy at young ages (Fig.3)whenitmaybeharmfultotheinfantdigestivesystem(FAO, 2013), or maysubstitute for breastmilk,which hasbeenlinked witharangeofdesirablehealthoutcomes(PAHO/WHO,2003).In this section we explore whether there might be substitution betweenbreastmilkandhouseholdsuppliesofdairymilk.Fig.4 plotsbreastfeedingstatusbychildagewithcomparisonsbetween Fig.3.Localpolynomialsmoothingestimatesofdairyconsumptionagainstchildagebytreatmentgroup,with90%conﬁdenceintervals.

Source:Authors’estimatesfromBangladeshIntegratedHouseholdSurvey2011,2015

(9)

Table3

Associationsbetweenlivestockownershipanddietaryindicatorsamongchildren6–59months(linearprobabilityandleastsquaresregressions).

(1) (2) (3)

Consumeddairy, last24hrs

Dietarydiversityscore(0-6),excludingmilk Total

calorieintake(kcal)

Treatmentgroup(vscontrol) 0.139*** 0.044 15.300

(0.042) (0.094) (25.733)

Placebogroup(vscontrol) 0.019 0.022 88.038***

(0.036) (0.084) (27.536)

Ownsbuffalo/bullock 0.015 0.018 28.835

(0.022) (0.066) (19.502)

Ownsgoat/sheep 0.021 0.071 0.896

(0.021) (0.057) (22.060)

Ownspoultry/duck/otherbirds 0.015 0.036 4.515

(0.015) (0.047) (18.027)

Owns/producesﬁsh 0.022 0.041 6.810

(0.019) (0.047) (22.133)

TotalLivestockUnits(TLU) 0.004 0.025 11.805

(0.015) (0.029) (7.631)

Allcontrols Yes Yes Yes

Districtﬁxedeffects? Yes Yes Yes

Observations 3,352 3,352 3,362

R-squared 0.172 0.362 0.495

Waldtests(p-values):

β(Treatment)>β(Control) 0.001*** 0.970 0.98

Notes:Theseareleastsquaresorlinearprobabilityestimates,withstandarderrorsareinparentheses,clusteredatthevillagelevel.***p<0.01,**p<0.05,*p<0.1.Source:

BangladeshIntegratedHouseholdSurvey2011,2015

Table4

AssociationsbetweenHAZandexposuretomilkproductionacrossdifferentagegroups(leastsquaresregressions).

(1) (2) (3) (4) (5) (6)

6-23 months

24-59 months

6-17 months

12-23 months

18-29 months

24-35 months

Treatmentgroup(vscontrol) 0.520^*** 0.040 0.548** 0.557^*** 0.473^*** 0.009

(0.165) (0.120) (0.235) (0.182) (0.137) (0.186)

Placebogroup(vscontrol) 0.162 0.173 0.028 0.094 0.371 0.106

(0.162) (0.116) (0.226) (0.257) (0.247) (0.239)

Allcontrolsvariables? Yes Yes Yes Yes Yes Yes

Districtﬁxedeffects? Yes Yes Yes Yes Yes Yes

Observations 1,154 2,384 869 788 800 830

R-squared 0.194 0.129 0.203 0.181 0.158 0.168

β(Treatment)>β(Control) 0.05** 0.17 0.02** 0.07* 0.69 0.59

Notes:Theseareleastsquaresestimates,withstandarderrorsareinparentheses,clusteredatvillagelevel.

***p<0.01,**p<0.05,*p<0.10.

“Allcontrols”includescontrolsforownershipofotherlivestockandtotalTLUs(livestockwealth),aswellasthefullsetofsocioeconomiccontrolsdescribedinTable,agender dummyandmonthlydummiesforchildage,aswellasdistrictﬁxedeffects.Source:BangladeshIntegratedHouseholdSurvey2011,2015.

Table5

Associationsbetweenstuntingstatusandexposuretomilkproductionacrossdifferentagegroups(linearprobabilitymodel).

(1) (2) (3) (4) (5) (6)

6-23 months

24-59 months

6-17 months

12-23 months

18-29 months

24-35 months

Treatmentgroup(vscontrol) 0.104** 0.049 0.034 0.136** 0.223*** 0.135

(0.046) (0.041) (0.059) (0.063) (0.058) (0.087)

Placebogroup(vscontrol) 0.080 0.048 0.028 0.052 0.110 0.091

(0.058) (0.047) (0.085) (0.085) (0.075) (0.079)

Allcontrols? Yes Yes Yes Yes Yes Yes

Observations 1,159 2,390 873 791 802 831

R-squared 0.179 0.117 0.221 0.157 0.175 0.178

β(Treatment)>β(Control) 0.72 0.99 0.95 0.39 0.17 0.64

Notes:Thesearelinearprobabilitymodelestimates,withstandarderrorsareinparentheses,clusteredatvillagelevel.***p<0.01,**p<0.05,*p<0.10.“Allcontrols”includes controlsforownershipofotherlivestockandtotalTLUs(livestockwealth),aswellasthefullsetofsocioeconomiccontrolsdescribedinTable,agenderdummyandmonthly dummiesforchildage,aswellasdistrictﬁxedeffects.Source:BangladeshIntegratedHouseholdSurvey2011,2015.

(10)

treatmentandplacebo(Paneli)andtreatmentandcontrol(Panelii).

Theresultsshowthat,frombirthtoaround8monthsofage,dairy- producing households are signiﬁcantly less likely tobreastfeed their children. Above this age range there is no signiﬁcant

difference in breastfeeding rates. This suggests that access to dairymilkmayhaveanegativespilloveronbreastfeedingpractices inthecriticallyimportant0–5monthsagerangewhenitisstrongly recommendedforinfantstobeexclusivelybreastfed.

Table6

OLSandIVestimatesoftheassociationbetweenHAZandthelogofmilkproductionperchild.

(1) (2) (3) (4) (5) (6)

6-23 months

24-59 months

6-17 months

12-23 months

18-29 months

24-35 Months

Logquantityofmilkproduced 0.084** 0.008 0.083 0.080* 0.031 0.017

(0.034) (0.024) (0.050) (0.040) (0.037) (0.043)

Allcontrols? Yes Yes Yes Yes Yes Yes

Observations 1,159 2,390 873 791 802 831

R-squared 0.192 0.124 0.201 0.179 0.151 0.158

Notes:Theseareleastsquaresestimates,withstandarderrorsareinparentheses,clusteredatvillagelevel.***p<0.01,**p<0.05,*p<0.10.Allregressionscontrolfor ownershipofotherlivestockandtotalTLUs(livestockwealth),aswellasthefullsetofsocioeconomiccontrolsdescribedinTable,agenderdummyandmonthlydummiesfor childage,aswellasdistrictﬁxedeffects.Source:BangladeshIntegratedHouseholdSurvey2011,2015.

Fig.4.Alocalpolynomialsmoothinggraphofbreastfeedingstatusbychildageforhouseholdsthathaveandhavenotproduceddairy.