Sanitation, Disease, and Anemia: ∗ Evidence From Nepal DianeCoffey† MichaelGeruso‡ July13,2015 Abstract Anemia is a health problem with significant economic consequences. In children, it impairs physical and cognitive development and reduces human capital accumulation. In adults, it re- ducesproductivity. Globally,morethan40%ofallchildrenunderfivehavehemoglobinlevelsbe- lowthethresholdforanemia.Previousliteraturehasfocusedontheroleofpoornutritionalintake in causing anemia. This paper is the first to propose the hypothesis that poor sanitation, a pub- licgoodwithotherwell-documentedhealthexternalities,significantlycontributestohemoglobin deficiencyviaitsroleincreatingapoordiseaseenvironment. Weinvestigateacausalrelationship between sanitation and hemoglobin by exploiting rapid differential improvement in sanitation acrossregionsofNepalbetween2006and2011. Weshowthatwithinregionsovertime, cohorts ofchildrenexposedtoworsecommunitysanitationdevelopedlowerhemoglobinlevelsanddis- playedhigheranemiaincidence.Theresultscontributetothebasicscienceofanemia’scausesand suggestthepossibilityofnewpolicytoolsforreducinganemiainthedevelopingworld. ∗WeappreciatecommentsandsuggestionsfromHaroldAlderman,AnjaliChikersal,OliverCummings,AngusDeaton, JeffreyHammer, H.P.SSachdev,DeanSpears,andBelénTorondel. WethankparticipantsoftheSanitationandStunting Conference,heldAugust,2013attheDelhiSchoolofEconomics,andattheNationalFoodSecurityActconference,held March, 2014 at Allahabad University, for helpful feedback. Research reported in this publication was supported by the EuniceKennedyShriverNationalInstituteOfChildHealth&HumanDevelopmentoftheNationalInstitutesofHealth under Award Number R03HD081209. The content is solely the responsibility of the authors and does not necessarily representtheofficialviewsoftheNationalInstitutesofHealth. †WoodrowWilsonSchool,PrincetonUniversity;IndianStatisticalInstitute,Delhi;r.i.c.e..Email:[email protected] ‡UniversityofTexasatAustinandNBER.Email:[email protected] 1 Introduction Anemia is a widespread problem with serious health and economic consequences. Defined by low counts of red blood cells or low levels of hemoglobin in the bloodstream, anemia implies a reduced capacityforthebloodtocarryoxygen. Inadults,itreducesproductivity(Thomasetal.,2004)andis associatedwithhighermaternalmortality(Rush,2000). Inchildren,itimpairsphysicalandcognitive development directly (Grantham-McGregor and Ani, 2001; Ozier, 2015) and affects human capital accumulation via impacts on behaviors like school attendance (Bobonis, Miguel and Puri-Sharma, 2006). Globally,morethan40%ofchildrenhavehemoglobinlevelsbelowthethresholdforanemia.1 The problem is particularly severe in the developing world, as anemia is closely associated with inadequatenutrition. Because of its damaging effects on human capital formation and productivity, anemia has at- tracted significant research and policy attention. Economic research in the area of preventing or reducing anemia has generally focused on (i) poor nutrition, and in particular iron deficiency (e.g., Bhattacharya, Currie and Haider, 2004; Thomas et al., 2004), and (ii) malaria (Sachs and Malaney, 2002), which is a parasitic infection that attacks the red blood cells.2 Nonetheless, there are reasons to believe that poor nutrition and malarial disease are not the only important causes of anemia. For one,internationalvariationinanemiaratesisnotwellexplainedbyinternationalvariationinincome (Alderman and Linnemayr, 2009). To the extent that income is a reasonable proxy for basic nutri- tion,thisposessomewhatofapuzzle. Second,althoughitiswellknownthatinsub-SaharanAfrica malariaisa majorcauseofanemia, anemiaratesarehighestin SouthAsia, wheremalaria isfarless prevalent. In this paper, we propose a third broad cause of anemia that operates in addition to nutritional intakeandmalaria,andwhichhasimportant(anddifferent)implicationsforpolicy. Weproposethat poorlocalsanitation—andspecifically, opendefecation—causeslowerhemoglobinandhigherrates of anemia in children. Open defecation means defecation outside on the open ground, without the use of a toilet or latrine. Whereas nutritional intake is a behavior with purely private benefits, poor sanitation primarily constitutes an external harm: open defecation spreads intestinal worms and other fecal pathogens across individuals, since these are transmitted by skin contact with the fecal 1See(Stevensetal.,2013)forworldwideanemiarates. 2Malariahasattractedattentionineconomicsforitshumancapitaleffectsbelievedtooperatethroughanemia(see,e.g., CohenandDupas,2010). 1 matterthatisleftintheopen. Aswediscussbelow,thereissignificantepidemiologicalevidencesug- gestingthatsanitationcouldplayanimportantroleindetermininganemia. Intestinalparasitesand other infections are spread by open defecation; these diseases can affect the intestinal wall in ways thatleadtodecreasedabsorptionofnutrients,includingiron,vitaminB12andfolicacid(Rosenberg andBowman,1982;Nath,2005),whicharecriticalfortheproductionofhemoglobin. Ifopendefeca- tiondidindeedaffectanemiarates,itspotentialroleintheworldwide,aggregatepatternsofanemia would be staggering: more than a billion people (about 14% of the world’s population) defecate in theopen. Ours is the first paper to investigate a link between open defecation and anemia. Nevertheless, thepossibilityofsuchalinkissuggestedbyasmallbodyofpriorwork. Withrespecttotheconnec- tion between open defecation and intestinal parasites, a randomized control trial in Indonesia that includedatoiletconstructionandanti-opendefectionbehaviorchangeinterventions, foundthatre- duced rates of open defecation were associated with reduced parasitic infections (Cameron, Olivia andShah,2013). Withrespecttotheconnectionbetweenparasiticinfectionsandanemia,arandom- ized control trial among Kenyan children found that a single dose of deworming medicine was as effective in improving hemoglobin levels as a daily supplement of 13 micronutrients including iron taken for eight months (Friis et al., 2003). In another randomized control trial in India, Bobonis, Miguel and Puri-Sharma (2006) found that a combination of nutrient supplementation (iron and vi- tamin A) and deworming medications had a positive—though statistically insignificant—effect on hemoglobin levels.3 Taken together, these studies suggest pathways through which poor sanitation could affect anemia, since contact with human fecal matter is a primary vector for many intestinal worm infections.4 Despite this suggestive evidence, no prior study has shown a link between open defecationandanemia,whichistheaimofthispaper. We examine the impact of sanitation on children’s hemoglobin levels in the context of Nepal. Nepalisanidealempiricalsettingfortworeasons. First,Nepalhasverylittlemalaria,whichinother 3Bobonis,MiguelandPuri-Sharma(2006)wasnotdesignedtoseparatelyidentifytheeffectsofironsupplementation anddewormingmedicine. 4Bleakley (2007) provides evidence more directly connected to sanitation, but without a direct connection to anemia orhemoglobin. HestudieseffortstoeradicatehookworminfectionsintheUSSouthattheturnofthetwentiethcentury. Hookwormsarespreadbycontactwithhumanfeces. Theeradicationeffortsfeaturedadewormingcampaign,andlater, technical assistance in constructing sanitary latrines to arrest the cycle of reinfection. The study showed that the inter- ventionshadlargeeffectsonschoolattendanceandliteracy,aswellaslater-lifeearnings. Ahypothesizedchannelforthe humancapitaleffectsinBleakley(2007)wasthathookwormscausedanemia,thoughthehistoricaldatausedinthatstudy didnotallowforanydirectevidenceonhemoglobinlevels. 2 developing country contexts would be an important confounder when examining anemia. Second, Nepal has had relatively high rates of open defecation historically, but also rapid improvement in sanitationintherecentpast. In2006,about50%ofNepalesehouseholdsdefecatedintheopen—that is,theyreportedusingabush,field,ornofacility. Bythismeasure,Nepalesehouseholdshavefaced among the worst sanitation environments in the world. The rates of open defecation in Nepal were worse, for example, than most countries in sub-Saharan Africa at the time. However, beginning in 2004/2005,thegovernmentofNepallaunchednewnationalinitiativesaimedatreducingopendefe- cation. Since that time there has been a rapid improvement in latrine and toilet use. By 2011, the fraction of households defecating in the open had decreased to a national mean of 35%, with signif- icant variation in improvements across regions. Consistent with the notion that national efforts to reduceopendefecationwouldhavedifferentscopeforimprovementaccordingtothebaselinelevel ofopendefecation,weshowthattheprimarydeterminantofreductionacrossregionsafter2006was the regional level of open defecation in 2006. For example, regions that were already open defeca- tion free by 2006 could experience no further improvements, while regions with the highest open defecation rates (as high as 70%) experienced the largest level changes (in excess of 30 percentage points). Weexploitthegeographicallyheterogeneoussanitationimprovementsfrom2006to2011toiden- tify impacts of poor sanitation on hemoglobin in difference-in-differences regressions. We find that, controlling for own defecation practice, a 10 percentage point decrease in the fraction of neighbors whodefecateintheopenisassociatedwitha0.20g/dLincreaseinhemoglobinlevels,orabout14% of a standard deviation. To put this effect size in context, interventions in the experimental nutri- tionliterature,suchasmicronutrientsupplementation(Friisetal.,2003),ironsupplementation(Lind etal.,2003),andironfortificationinfoods(VanStuijvenbergetal.,1999),haveeffectsizesthatrange from 0.20 g/dL to 0.41 g/dL. The effect sizes we estimate are consistent with the experimental evi- denceontheefficacyofdeworminginterventionsbyFriisetal.(2003), whichincreasedhemoglobin by 0.21 gm/dL. As we describe in greater detail below, the spread of intestinal worms via contact withfecalmattercomprisesoneofthetwochannelsbywhichtheeffectswefindmayoperate. The identifying assumption underlying our difference-in-differences analysis is that improve- ments in sanitation were not correlated with other changes within regions that independently af- fectedhemoglobinlevels. Inaseriesofplacebotests,weshowthatimprovementsinphysicalinfras- 3 tructureotherthantoilet/latrineuse,suchashouseholdelectricityanduseofsaferwatersources,did notpredictchangesinchildren’shemoglobin. Thusourfindingsarenotmerelyreflectingeffectsofa generalpatternoflocaldevelopment. Wealsoshowthathealthbehaviorsthatwouldindependently affectanemiabutwouldnotbeexpectedtobeinfluencedbychangingratesofopendefecation,such asdietandtheconsumptionofironsupplements,donotrespondtothevariationinopendefecation thatidentifiesourestimates. Themaincontributionofourstudyistodevelopandtestanovelhypothesisaboutthecausesof anemia. Inthisway,ourworkcomplementsalargeexistingbodyofresearchonanemiathathasfo- cusedontherolenutritionalsupplementsincombattinganemiaindevelopingcountries. Ourstudy isdistinctfromthepriorworkinthatpaststudieshavefocusedonprivatebehaviorsorindividually- administered health inputs. For example, past experimental studies of interventions targeting ane- mia have randomized whether a child received an iron supplement, fortified food, or deworming medicine. Incontrast,ourstudyfocusesonacauseofanemiathatisexternaltothechild’sbehaviors andtothehealthinputschosenbythechild’sparents. Oneimplicationoftheexternalnatureofopendefecationisthatinvestigatingimpactsonanemia requiresvariationthatarisesatthelocalgeographiclevel(asitdoesinourempiricalanalysis),rather than at the person-level. This suggests that the kind of cluster-randomized trials that have recently beenfieldedtoexamineotheraspectsoflocalsanitation(see,e.g. Guiteras,LevinsohnandMobarak, 2015andClasenetal.,2015)arealsotherightapproachforfutureexperimentalworkthatbuildson ourfindingsaboutanemia. Beyond advancing the basic science of the causes of anemia, our study informs the potential policy responses. Reducing anemia in children can in part be accomplished by changing the health behaviors of community members who are neither children nor parents. With open defecation, it is the behaviorofneighborsthatmatters,andthissuggestsnewavenuesforanemiaprevention. Sanitation improvementraisesitsownsetofdifficultiesexactlybecausesanitationisapublicgoodandtherefore subjecttoinadequateprivateinvestment(Guiterasetal.,2014),suggestingawelfare-improvingrole forgovernment. Finally,thispapercontributestoagrowingliteratureconcernedwiththeadversehealthandhu- mancapitalconsequenceofopendefecation. Sanitationhasattractedsignificantpolicyattentionand NGOinvestmentinrecentyearsforreasonsunrelatedtoanemia. Ourfindingsonanemiastrengthen 4 the rationale for such investments, and may play a role in explaining some of the recent findings in the literature. Lawson and Spears (2014) show that exposure to open defecation is associated with lowerlater-lifeproductivity,andSpearsandLamba(2013)shownegativeimpactsonchildcognitive function. Bothareoutcomesknowntobeaffectedbyanemia.5 The remainder of the paper is organized as follows. Section 2 discusses the known causes of anemiaandreviewstheexistingepidemiologicalevidenceofachannelfrompoorsanitationtolower hemoglobin. Section 3 presents some new stylized facts from international comparisons that are intended to motivate our main analysis. Section 4 describes our data, identifying variation, and empirical strategy. Section 5 reports results, and section 6 traces out the significance and policy relevanceofourfindings. Section7concludes. 2 Background on Anemia and Sanitation Hemoglobinisaproteinwhichresidesinredbloodcells,andwhichbindstoironinordertoattract oxygen and carry it throughout the body. Iron deficiency anemia is defined by hemoglobin below a threshold level. There are several known causes of low hemoglobin. These involve either too little productionortoomuchdestructionofhemoglobin. Poordiets,particularlyamongyoungchildren,areanoften-citedcauseofanemiaindeveloping countries (Yip and Ramakrishnan, 2002; Tolentino and Friedman, 2007). Although a major cause of low hemoglobin production is iron deficiency in the diet, low hemoglobin can can also be caused by lack of vitamin B12 and folic acid, two nutrients necessary for the production of red blood cells. The late introduction of solid foods in infants, and diets containing inadequate amounts of these essential nutrients are both important contributors to low hemoglobin in South Asia, the region of thedevelopingworldwestudyhere(Menon,2012).6 Malaria is another important cause of anemia, particularly in sub-Saharan Africa. The disease is transmitted by a mosquito bite, during which a parasitic protozoa carried by the mosquito enters the person’s bloodstream. (The protozoan malaria parasite is significantly different in form, life cy- cle, and symptomatic effects from the intestinal worms we discuss below, which are transmitted by 5More broadly, our study connects to a wide literature on the role of water, sanitation, and disease environment in drivinghealthandhumancapitalaccumulationinthedevelopingworldandthehistoricalUS.See, forexample, Cutler andMiller(2005);Watson(2006);Bleakley(2007);Spears(2012);AlsanandGoldin(2015)). 6Astrictvegetariandiet,containingnoanimalproteins,wouldbelackinginvitaminB12. 5 contactwithhumanexcreta.) Themalariaparasiteattacksredbloodcells,whichareinturnattacked bythehost’simmunesystem. Thisdestructionofredbloodcellsleadstoanemia. Howcouldpoorsanitationaffectanemia? Therearetwoplausiblechannels. Thefirstisrelatedto intestinal parasites and the second is a condition known as environmental enteropathy.7 Our study doesnotattempttodistinguishbetweenthesetwoepidemiologicalpathways,asbothareconsistent withanimpactofpoorsanitation,andbotharelikelytobeoperatingsimultaneously. Intheepidemiologicalliterature,intestinalparasitesareknowntocauseanemiabycausingblood lossinthestool,lackofappetite,increasedmotilityoffoodthroughtheintestine,andcompetitionfor nutrients. Intestinalparasitesalsocausedamagetotheintestinalwallthatleadstodecreasedabsorp- tion of nutrients, including iron, vitamin B12 and folic acid (Rosenberg and Bowman, 1982). It has long been known that open defecation spreads intestinal parasites; Cairncross (2003) cites research from the 1930s that describes how variation in community latrine use in the southern United States predictedparasiteinfectionsinchildrenlivingindifferentplaces.8 Thesecondpathwayfromopendefecationtohemoglobinisenvironmentalenteropathy,known as tropical sprue in an older medical literature. It is a disease which alters the lining of the intes- tine and inhibits absorption of calories and nutrients. It is believed to be caused by the ingestion of large quantities of fecal pathogens (Walker, 2003; Humphrey, 2009; Lin et al., 2013; Kosek et al., 2013). IthasbeendocumentedthatenteropathyaffectstheabsorptionofvitaminB12andfolicacid, two essential nutrients for the production of hemoglobin (Nath, 2005). Although the link between open defecation and enteropathy less well understood than the link between open defecation and intestinalworms,itishypothesizedthatopendefecationexposespeopletothekindsofbacteriathat, when ingested in large quantities, lead to decreased absorption of micronutrients necessary for the productionofhemoglobin(seeWalker,2003,Nath,2005andHumphrey,2009). Medicalresearchers have hypothesized a link between enteropathy and anemia as long ago as the 1920s (Baumgartner andSmith,1927). 7Itisalsopossiblethatintestinaldiseasesspreadbyopendefecationcouldaffectpregnantwomen,andleadthemto passlowironstoresontotheirbabies.SeeAllen(1997). 8Childreninfectedwithparasitespassthemthroughtheirbowelmovementstothesoil;manyparasiteeggscanlivein thesoiluntiltheycomeintocontactwithnewhosts,suchasotherchildrenwhocometothesameplacestodefecateinthe open. 6 3 Stylized Facts from International Comparisons To motivate the our econometric analysis below, we begin in Figure 1 by documenting some cross- countrysummarystatisticsrelatingsanitationtoanemia. Toourknowledge,oursisthefirststudyto documentthesepatterns,evencross-sectionally. Dataonanemiaandopendefecationdatausedinthefigurecomefrom81nationallyrepresenta- tiveDemographicandHealthSurveys,covering45countries. Observationsaremeansfromsurveys (countries × years) and we include every survey for which data on children’s hemoglobin was col- lected.9 Thesizeofmarkersisproportionaltocountrypopulation,andmanycountriesappeartwice in the scatterplot for different survey years. The data, which include countries from sub-Saharan Africa,LatinAmerica,Europe,andAsiaaredescribedinmoredetailinAppendixA.1andthecoun- try×yearsthatmakeupindividualobservationsarelistedinTableA1. Followingtherecentliterature(e.g.,Guiteras,LevinsohnandMobarak,2015;GerusoandSpears, 2015;HammerandSpears,2013),wecapturethesanitationenvironmenttowhichachildisexposed bycalculatingthefractionofhouseholdsthatdefecateintheopen. Thegreaterthefractionofhouse- holds which do not use a toilet or latrine, the greater the frequency with which a child comes into contactwithgermsorparasitestransmittedbyfeces. In Panel A of Figure 1, we plot the unconditional relationship between the average hemoglobin level of children and the mean open defecation rate in the country. We restrict attention to children aged 6 to 35 months, as this was the common age range for which hemoglobin data was recorded across the 81 DHS surveys represented in the figure. The figure also plots a regression line cor- responding to the population-weighted OLS coefficient of open defecation on hemoglobin. The plot shows a clear association in which more open defecation (i.e., worse sanitation) predicts lower hemoglobinlevelsofchildren. InPanelB,wemodifythehorizontalaxissothatitmeasuresthelogofopendefecationpersquare kilometer, following a recent literature showing that the risk of transmitting pathogens via open defecationisincreasinginpopulationdensity(Spears,2013;Hathietal.,2014). Dataontotallandarea and population, which are used to construct the measure of open defecation per square kilometer, come from the Penn World Tables.10 The clear negative relationship continues to hold. While the 9WedescribetheDHSsurveymethodology,sample,andquestionnaireinmoredetailbelow.TheDHSsurveyforNepal in2006and2011comprisesourmainanalysisdataset. 10Populationdensityatthecountrylevelisintendedasaproxyforpopulationdensitymeasuredatamoredisaggregated, 7 slopesarenotdirectlycomparablebetweenPanelsAandBsincethehorizontalaxesaredifferent,the overalldeclineinhemoglobinlevelsbetweenobservationswiththebestsanitation(leftmostpoints) andworstsanitation(rightmostpoints)aresimilaracrossthetwopanels. A natural question in this context is whether the places with worse sanitation are merely worse inotherwaysthatwouldindependentlypredictanemia. Inparticular,malariaincidence,whichim- pactsanemia,maybeworseincountrieswherethesanitationenvironmentisworse. Andnutritional intake, which is the leading known cause of anemia, would be expected to improve with higher incomes. Since sanitation may be expected to likewise improve with higher incomes, this suggests anotherimportantpotentialconfounder. InpanelsCandDFigure1,wecontrolformalariaincidence usingnationalmalariaratesconstructedbyKorenromp(2005),andwecontrolforGDPpercapitaus- ingdatafromthePennWorldTables. ThedetailsofthedataconstructionaredescribedinAppendix A.1. To display scatter plots with these controls, we first separately regress hemoglobin on malaria ratesandGDPpercapitaandopendefecationonmalariaratesandGDPpercapita. Wethenplotthe residualsfromthoseregressionsagainsteachother. Therelationshipbetweensanitationandanemiais,infact,strongeraftertheinclusionofcontrols formalariaandGDPpercapita,withobservationsmoretightlyclusteredaroundtheregressionline. Figure 1 is intended only to provide motivation for the econometric analysis below, the patterns it revealsareconsistentwithapreviousfindingintheliteraturethatinternationalvariationinanemia rates is not well explained by international variation in income (Alderman and Linnemayr, 2009). Thepatternsarealsoconsistentwiththefactthatmanysub-SaharanAfricancountrieshaverelatively high rates of malaria but high rates of anemia (low average hemoglobin): those same sub-Saharan Africancountriesalsoexperiencelowratesofopendefecation,relativetoSouthAsiancountries. In summary, the cross-country comparison reveal an interesting and previously undocumented pattern. Poor sanitation strongly predicts low hemoglobin, both unconditionally, and controlling for income (a proxy for nutrition) and malaria incidence. The remainder of the paper investigates a causal relationship, using variation in open defecation that is plausibly exogenous to hemoglobin levels. sub-nationallevel. Subnationalmeasuresofpopulationdensitycoveringthiscountrysamplecannotbeconstructedwith anyavailabledata. 8 4 Data and Empirical Framework We investigate the hypothesized link between sanitation and anemia using data from Nepal. Nepal ranks among the worst sanitation environments in the world. As recently as 2006, half of Nepalese households disposed of excreta in the open, without the use of a toilet or latrine. But sanitation in Nepal has improved rapidly in recent years; there was a 13 percentage point decline in open defe- cationatthenationallevelbetween2006and2011. Nepal’spoorbaselinesanitation,rapidimprove- ment, and low rates of malaria (a potential confounder), make it an ideal empirical setting for our study. 4.1 Data Thedatausedinourmainanalysiscomefromthe2006and2011DemographicandHealthSurveys (DHS)ofNepal. Thesenationallyrepresentativesurveyscollectinformationonhealthbehaviorsand outcomesofhouseholdmembers,includingdataonwatersourcesandtoiletuse. Mothersalsoreport onthetypesandamountsoffoodsconsumedbytheiryoungchildren. TheDHSmeasureshemoglobinusingtheHemoCue(cid:13)R method, inwhichasurveyorintroduces a drop of blood from the respondent’s finger into a portable device which reports the respondent’s hemoglobinlevelinthefield.11 Typically,anemiaisdefinedbyhemoglobinlevelsbelowsomethresh- oldvalue. TheWorldHealthOrganization(WHO)setsthehemoglobinconcentrationthresholdat11 g/dLand7g/dLforanemiaandsevereanemia, respectively, thoughvariousresearchersandmed- ical bodies set alternative cutoffs.12 In order to maximize power and avoid sensitivity to the choice of threshold, we use the entire continuous range of hemoglobin concentration in our main results, ratherthanindicatorsforrangesofanemiaseverity. Inourdata,ifarespondenthouseholdreportsusinga“bush,field,ornofacility,”thehousehold is coded as defecating in the open. We generate variables capturing the mean open defecation at thelevelofaregionorneighborhood,dependingontheanalysis. Neighborhoodsaredefinedinthe data by primary sampling units (PSUs), which are composed of survey sampling frames of 100 to 200 households. In rural areas these frames may be whole villages. In urban areas, the frames are 11SeeMeasureDHS(2013)formoreinformationaboutcollectionofhemoglobindatainthefield. Kapooretal.(2002) comparestheHemoCue(cid:13)RmethodtoothermethodsoftestinghemoglobininIndia.Thispaperusesthehemoglobinmea- surementthatisunadjustedforaltitude. 12ThisWHOstandardappliestochildren6to60monthsofage. 9