The Cognitive Link between in Utero Nutrition and Schooling:

Experimental Evidence from Tanzania∗

Plamen Nikolov†

Harvard University

March 16, 2012

Preliminary - please do not quote or cite without permission

Abstract

Because of the high returns of schooling in developing countries, policy-makers have placed consid-

erable attention to increasing school access. However, an important mechanism through which brain

development in utero can a�ect demand for education exists. Cognitive development in utero, due to

maternal de�ciency in folic acid, can biologically constrain children's demand for education. Using a

more scienti�cally credible research designed to detect causal e�ects than has been used in previous

research, we examine how reductions in micronutrient de�ciency in utero impact subsequent child

schooling attainment in Tanzania. We also examine to what extent parents allocate resources so as

to compensate for or to reinforce inequalities across children in cognitive endowments. Capturing the

behavioral response to the biological intervention allows us to disentangle the biological e�ect from the

household response to the original randomized intervention. To execute this strategy, we follow up on

a randomized control trial with micronutrient supplements o�ered to HIV-negative pregnant women

in Dar es Salaam, Tanzania between 2001-2003. While data collection is still ongoing, very prelimi-

nary results show some cognitive improvements among children born to mothers in the micronutrient

supplementation group from the original RCT.

∗I thank Martin Abel, Anna Aizer, Jean-Marie Baland, Oriana Bandiera, Raj Chetty, Lawrence Katz, Gary Becker,Amitabh Chandra, David Cutler, William Dow, Erica Field, Selim Gulesci, Darcy Hango, Patrick Moynihan, SendhilMullainathan, Joe Amick, Nagraj Rao, seminar participants at Harvard/HMS Health Economics Workshop, LSE EconomicsDepartment, FUNDP(Namur)/BREAD Summer Development Economics School, University of Dar es Salaam PhD StudentSeminar, and six anonymous NSF reviewers for helpful comments to this paper. I wish to thank Bernard Matungwa, MakariosKiyonga, Abbakari Msa�ri, A�sa Ramadhani, Sophia Linda, Albert Benson and Hasan Farijala for outstanding assistancewith in-depth tracking, surveys and qualitative interviews with survey respondents. I thank the AMID CEPR Marie CurieInitial Training Grant, the Hewlett Foundation/IIE Dissertation Fellowship, the Harvard Institute for Quantitative SocialStudies for funding various parts of this project.†National Bureau of Economic Research, 1050 Massachusetts Avenue, Third Floor-392, Cambridge, MA 02138. Tel

+1-617-855-9668. E-Mail: pnikolov@fas.harvard.edu.

1

1 Introduction

Over the past several decades, there has been considerable attention placed on increasing schooling indeveloping countries. Psacharopoulos (1994) estimates microeconomic returns to education as high as42% per annum in Botswana (for primary education) and 47.6% per annum in Zimbabwe (for secondaryeducation).1These large returns have stimulated a concerted e�ort in investing in education to stimulateeconomic outcomes (UNESCO 2007). However, an important mechanism through which in utero environ-ment can a�ect education and economic outcomes exists. Poor in utero health conditions can biologicallyconstrain cognitive development.

While most micronutrient de�ciencies are likely to be resolved with improvements in economic out-comes by way of rising caloric intake, de�ciency in utero for four nutrients in particular (B6, B9, B12,and iodine) has been biologically linked to irreversible and continuous damage to cognitive developmentthroughout an individual's lifetime (Bottiglieri et al., 1995; Bryan et al., 2002; Guilarte 1993; Fioravantiet al., 1997; Hankey 1999; Alpert and Fava, 1997; Schneede 1994; Cao et al., 1994; Hetzel and Mano,1989; Pharoah and Connolly, 1987). Because of this persistent e�ect on learning over the lifespan, theproposed research project will examine the e�ect of three of these nutrients, B6, B9 and B12, on subse-quent cognitive development, school enrollment, attendance rates, educational attainment and parentalbehavior responses. The resulting loss in cognitive capacity could have important consequences for aggre-gate human capital accumulation in a�icted settings, leading to a lower fraction of children enrolling in orattending school, slower rates of grade attainment for age, and fewer students progressing to institutionsof higher education.

Using data from an experimental intervention in Tanzania and building on a pilot phase we alreadyconducted in 2011, we address: (1) How much do biological setbacks - due to pre-natal maternal mal-nutrition - in brain development in utero, in�uence subsequent child schooling attainment? (2) To whatextent do parents allocate resources so as to compensate for or to reinforce inequalities across childrenin endowments? We follow up on an individual-level randomized trial of nutritional supplements o�eredto HIV− pregnant women in Dar in Tanzania conducted between 2001 and 2004. We started trackingactivities in July 2011. So far, we have maintained an e�ective tracking rate of more than 90%.

We focus on Tanzania due to the high general micronutrient de�ciency and particularly de�ciency formicronutrients on which we intend to focus (Kavishe, 1987; Kavishe, 1991).2 In Tanzania, as in otherSub-Saharan African countries, dietary intake of nutrients in pregnant women is marginal or lower thanthe recommended intakes and therefore these women are at high risk for de�ciencies (Mulokozi 2003; FAO,1995; FAO, 1996a; FAO, 1996b; WHO, 1994; WHO/UNICEF, 1995). B9 (folic acid) and B12 de�cienciesoccur mainly due to the increased requirements during pregnancy which are not balanced by adequatedietary intake, and the high prevalence of malaria (Kavishe, 1987; Kavishe, 1991). The prevalence ofde�ciency in B9 (<5ng/mL) and de�ciency in B12 (<200pg/mL) is respectively as high as 80% (Fleming,1989; Baker, 1981; Massawe et al., 1996) and 60% (DeMayer, 1985; Van den Broek, 2000) for women insome Sub-Saharan African countries.

Our study improves previous economics research in six major ways. First, the original RCT in-cluded folic acid and induced variation in a very speci�c dimension of the child's endowment�cognitiveability�whereas previous studies3 focus on measure on non-speci�c proxies which may re�ect many dimen-

1Because it is di�cult to �nd exogenous sources of education in developing countries, most economics studies rely on Mincerian

estimations. Behrman (1990) argues that estimates of returns to education are likely overstating the returns. Strauss and Thomas

(1995) suggest that the evidence is inconclusive and deserves further study.2Recent studies with blood analyses among pregnant women and non-pregnant adults in Dar es Salaam revealed a 70% prevalent

de�ciency of B micronutrients including thiamine, riboavin, and micronutrient B12 (Fawzi 1999; Mulokozi 2003).3See Almond (2006), Rosenzweig and Zhang (2009), Rosenzweig and Schultz (1983), Rosenzweig and Wolpin (1988), Pitt et al.

(1990).

2

sions of the child's endowment or many mechanisms through which the in-utero e�ect operates.4Second,previous studies rely on sibling �xed e�ects models for identi�cation and are likely mis-speci�ed. Evenwithin siblings, unobserved early life child investments likely correlate with endowment measures and alsowith the extent of parental investment during the remainder of childhood.5Third, we capture not onlyneonatal and early-life investments but also parental allocation behaviors in later childhood. Fourth, weconduct actual tests with children to proxy cognitive and non-cognitive development. Fifth, we mea-sure parental post-natal investment behavior on various margins including parental months breastfeeding,care-taking, childhood environment, psychological care, household and parental expenditures for chil-dren and self-reported warmth. Sixth, and perhaps most importantly, previous work in this area hasrestricted attention to within-sibling estimates and have have therefore been unable to separately identifyown endowment vis-à-vis sibling endowment e�ects. Our identi�cation strategy can disentangle these twoe�ects.

2 Conceptual Framework: Mechanisms by which Health Determines

Education Outcomes

Prenatal and early-life health interventions in�uence educational outcomes through four general directand indirect channels.

The �rst mechanism is a biological one and works through permanent limitations of intellectual ability.Unlike nutritional shortages during childhood or adulthood, fetal de�ciency for the nutrients chosen inthis study (B6 and B12 and folic acid) is believed to permanently limit intellectual ability, so its impactis likely to be particularly acute and persistent. Micronutrient B6, comprising three chemically distinctcompounds, pyridoxal, pyridoxamine, and pyridoxine, is involved in the regulation of mental function andmood. B6 is also an essential homocysteine re-methylation co-factor, and de�ciency is associated with anincrease in blood homocysteine levels. Homocysteine is a risk factor for cerebrovascular disease and mayalso have directly toxic e�ects on neurons of the central nervous system (Guilarte 1993). Maternal bloodlevels of folic acid (B9) and B12 have also been causally linked to neural tube defects (Botto et al., 1999;Christensen et al., 1999ab; Czeizel and Dudás, 1992; Laurence et al. 1981; MRC, 1991).6

The second mechanism, also a biological one, works through improving the immune system and im-proving general physical health, both of which can lead to improved school attendance. Previous researchdocuments that better physical health causally improves attendance for the US and developing countries(Miguel and Kremer, 2004; Edwards and Grossman, 1979; Shakotko et al., 1981; Shakotko and Grossman,1982; Perri, 1984; Wolfe, 1985; Berger and Leigh, 1989).7 Studies have also associated B micronutrientswith improved immune functioning (Bendich and Cohen, 1988; Ramakrishnan et al. 1999). B9 has hema-tological bene�ts during pregnancy (Fleming et al., 1986). Lack of pyridoxine (B6) appears consistentlyto inhibit cell-mediated immune function as well as humoral responsiveness to a variety of test antigens.

4For example, Maccini and Yang (2009) �nd long-run e�ects of early-life variation in rainfall in rural Indonesia. While well-identi�ed,

a drawback of the study is that the operative channels are not well understood. The e�ects are hypothesized to operate via household

income (which �uctuates with rainfall in agricultural households), but the authors cannot be sure of this it is possible, for example that

a direct health channel might be operative.5See Rosenzweig and Zhang (2009), Rosenzweig and Schultz (1983), Rosenzweig and Wolpin (1988), Pitt et al. (1990).6

Two observational studies provide evidence that B12 de�ciency among children leads to lower cognitive development (Schneede, 1994;

Louwman, 2000; Allen, 1999; Penland, 2000).

7Neuropsychiatric disorders including seizures, migraine, chronic pain and depression have been linked to micronutrient B6 de�ciency

(Malouf and Grimley, 2008).

3

The third mechanism, an indirect one, works through the nutrients' e�ect on increasing prospectivelifespan (Ozanne and Hales, 2004; Case, Fertig and Paxson, 2005; Lindeboom, Portrait and Van den Berg,2010; Leon D.E. et al., 1998). Higher life expectancy increases individual incentives to invest in humancapital (Ben-Porath, 1967; Kalemli-Ozcan, Ryder and Weil, 2000; Jaychandran and Lleras-Muney, 2009;Fortson, 2011).

Final mechanism relates to parental responses to children's initial biological endowments. Investmentsin children, as outcomes of the household's allocation problem, are in general jointly determined byparental preferences for each child, the production function for child quality, and the intra-householddistribution of initial endowments. Depending on the parameters of this allocation problem, parentsmay choose to reinforce or compensate for initial endowments (Leibowitz, 1974). Second, exogenousshocks to children's health could in�uence subsequent parental labor supply decisions related to increasedtime for child care-taking (Ruhm, 2004). The sign and magnitude of these behavioral responses indicatewhether � and by how much � parental investments magnify or dampen the biological e�ects of early-lifeinterventions.

3 Micronutrient de�ciency, cognitive outcomes and parental responses:

a simple model

Schooling choice

We start with a simple model based on Ben-Porath's human capital acquisition model (Ben-Porath, 1967;Willis, 1986). Abstracting from the within-household allocation issue for the time being, we start witha simple set-up in which individuals choose how much education (denoted e) to obtain to maximizediscounted lifetime earnings, y, and examine how these schooling investments change as a function ofchild quality (denoted q). We posit that child quality comprises both child's health status and cognition.We denote discounted future bene�ts to education with b(e, q) and discounted future costs (includingboth direct and indirect ones) with c(e, q). Both bene�ts and costs are increasing in education and child

quality. Marginal bene�t to schooling declines with education ( d2bde2

< 0) and the cost function is convex

( d2cde2

> 0). Both bene�ts and costs increase mechanically with child quality8, thus expanding the e�ectivetime constraint. The F.O.C. for the individual maximization problem is:

ye(e?, q) = 0

From be(e?, q) = ce(e

?, q), we can show:

de?

dq= −beq − ceq

bee − cee(1)

The denominator of (1) is negative, but the numerator's sign is ambiguous. There is no obvious signon the di�erence ex ante. Based on (1), we present the �rst testable proposition below.

Household Allocation

In addition to initial endowments, household allocation can also in�uence cognitive development andschooling outcomes. Parental allocation of resources among household children can occur according to atleast two alternative types of decision rules. One allocation rule would be based on equality: either equalityof inputs (all children receive the same) or equality of outcomes (entailing compensatory distribution, withthose who start with less in the way of initial endowments receiving more). An alternative allocation rule

8This could be due to an increase in more productive time due less time an individual is sick.

4

would base allocation decisions on e�ciency. In this scenario, parental resources go to children whoo�er the best prospects of gaining the most or providing the greatest return to parents for the resourceinvestment. This entails reinforcing distribution.

To add the parental response and siblings to the model, we consider a household with two childrenindexed i = 1, 2. As before, each child is born with an exogenous quality endowment qi. Sibling endow-ments determine the within-family price distribution for parental investment in additional child quality,hi. We denote the price of quality for child i as pi ≡ p(qi). p is a decreasing function of q.9

Parents' utility10 is given by:

u(h1, h2) =

(αh

γ−1γ

1 + (1− α)hγ−1γ

2

) γγ−1

(2)

α ∈ (0, 1) represents the relative utility weight given to the quality of child 1, and γ ∈ (0,+∞) is theelasticity of substitution.11

The household maximizes:

maxh1,h2

=

(αh

γ−1γ

1 + (1− α)hγ−1γ

2

) γγ−1

(3)

s.t. p1h1 + p2h2 ≤W (4)

pi = p(qi), i = 1, 2 (5)

From setting up the FOC s and simplifying, we get:

h1 =

(αp2

(1− α)p1

)γh2 (6)

The demand functions for quality as a function of price and income are:

h1 =W{(

αp2(1−α)p1

)γ}{(αp2

(1−α)p1

)γ}p1 + p2

(7)

h2 =W{(

αp2(1−α)p1

)γ}p1 + p2

(8)

3.1 Comparative Statics

We examine how parental investments change in response to changes in p1. Di�erentiating the demandfunctions (7) and (8) with respect to p1 and rearranging terms, we get:

∂h1∂p1

=W({(

αp2(1−α)p1

)γ}p1 + p2

)2p2∂

{(αp2

(1−α)p1

)γ}∂p1

−{(

αp2(1− α)p1

)γ}2 (9)

9This captures the fact that a dollar of investment in quality will yield larger returns for the child with a relatively higher endowment.10This is based on Becker's common preference model [Becker (1964, 1974, 1981)]. For tractability, we adopt a CES utility function,

which parametrizes the extent to which children's qualities are complements or substitutes.11Child quality is complementary when γ < 1, and substitutable when γ > 1.

5

∂h2∂p1

= −W

{(αp2

(1−α)p1

)γ}+ p1

∂{(

αp2(1−α)p1

)γ}∂p1({(

αp2(1−α)p1

)γ}p1 + p2

)2 . (10)

The signs of the own- and cross-price elasticities thus depend in part on∂{(

αp2(1−α)p1

)γ}∂p1

= −{(

αp2(1−α)p1

)γ}( γp1 ) <

0. From this expression, we get ∂h1∂p1

< 0. The sign of ∂h2∂p1depends on

{(αp2

(1−α)p1

)γ}+ p1

∂{(

αp2(1−α)p1

)γ}∂p1

.

From above, ∂h2∂p1

< 0 i�{(

αp2(1−α)p1

)γ}+ p1

∂{(

αp2(1−α)p1

)γ}∂p1

> 0. This is equivalent to ∂h2∂p1

< 0 i�

0 < γ < 1.

3.2 Testable Hypotheses

Proposition 1. If − beq−ceqbee−cee > 0, then as q increases, e? increases. In other words, as child quality

increases, optimal schooling increases.

Proposition 2. ∂h2∂p1

< 0 i� 0 < γ < 1. In other words, an increase in child's quality generates an

increases in parental investments for this child. In addition, an increase in a sibling's quality

increases own investments i� child quality is complementary in the parental utility function,

and decreases own investments if child quality is substitutable in the parental utility function.

4 Background: Context and Study Design

4.1 Study Design

Original Randomized Control Trial

We collect follow-up data on a recent randomized control trial (RCT) of pregnant women. The double-blind trial, assigning 8468 pregnant women to receive a daily oral dose of either a micronutrient supplementor placebo, examined the e�ect of nutrient supplementation on low birth weight (< 2500 g), prematurity,and fetal death. In 2001-2004, Fawzi (2007) conducted a randomized trial with pregnant women in Dares Salaam, Tanzania to account for di�erences in physical health outcomes at birth for children. Pregnantwomen who attended antenatal clinics12 in Dar es Salaam, Tanzania, between August 2001 and July 2004were invited to participate in the trial. Simple random sampling was used. Requirements for eligibilityincluded a negative test for HIV infection, a plan to stay in the city until delivery and for 1 year thereafter,and an estimated gestational age between 6 and 27 weeks according to the date of the last menstrual period.A list was prepared according to a randomization sequence in blocks of 20; at enrollment, each eligiblewoman was assigned to the next numbered bottle. The treatment tablet included 20 mg of micronutrientB2 , 25 mg of micronutrient B6 , 50 mg of micronutrient B12, and 0.8 mg of B9 (folic acid).13Averagetake up rate of treatment in the original RCT was 0.88. As part of the original trial, the research team

12According to a DHS 1996 (Bureau of Statistics, 1996), 97% of pregnant women attend antenatal care (ANC), and 70% do so at

least four times.13

The active tablets and placebo were similar in shape, size, and color and were packaged in identical coded bottles.

6

collected information on socio-economic status of study participants, information about family members,bio-marker samples at baseline and after six months post birth. Table 1 (Appendix A) presents descriptionof the study sample for several socio-economic variables against Demographic and Health Survey (DHS)data from the 2003/4 Wave for women who ever gave birth: no signi�cant di�erences exist between theRCT sample and the 2003/2004 DHS sample from Dar es Salaam.14

Original RCT Results and Tracking Information

Of the 8379 women with known birth outcomes in the original randomized trial study (Fawzi, 2007), 8137gave birth to live babies and were eligible for the analyses of birth weight and prematurity outcomes. TheIntent-to-Treat (ITT ) estimate on birth weight di�erence between treatment and controls groups is 67 g

(P < 0.001). Adjusting for compliance of 0.88, the TOT estimate is 76 g. In addition, treatment had nosigni�cant e�ects on prematurity or fetal death.

The original research team collected detailed map cue information in 2004 enabling the original researchteam to locate study participants at their homes and to conduct follow-up interviews capturing neonataloutcomes.

Pilot Phase Follow-up

With logistical support of tracking enumerators and enumerator personnel from a team for the KageraHealth and Development Survey (KHDS), a study on the long-run wealth dynamics of households andindividuals within North West Tanzania, we started pilot phase activities in July 2011. The KHDS survey,and the same team we use for this project's tracking, had maintained a highly successful tracking rate inEastern Africa: in 2010, 88% of the original 6353 respondents15 had either been located and interviewed,or, if deceased, su�cient information regarding the circumstances of their death collected.

For each household selected for the pilot phase in this project, we followed up with a parent andat least two children. Because the original RCT recruited study participants mostly between 2001 and2002, the average age of the index child was 9.2 years in July 2011. For data collection, we selectedthe index child in utero at the time of the original trial and all kids born before the original RCT. Weselected children born before the original RCT to avoid endogeneity of parental behavior responses withthe respect to child quality.

5 Data and Estimation Strategy

5.1 Empirical Speci�cations

Throughout the following speci�cations, we use the following variable de�nitions: Tif is the binary variablethat child i of mother f was treated with micronutrient supplement in the original medical trial, A is avector of birth-month dummies, and X includes binary controls for gender, sex-speci�c birth order andother socio-economic variables.

To test our �rst prediction, we estimate:

cognitive scoreif = α+ β1(Tif ) + β2(Aif ) + β3(Xif ) + µf + δif (11)

Using educational outcome proxies described in section 7.1, we can estimate:

14At every monthly visit, a new bottle was given to each woman, and the pills remaining in the used bottles were counted.15Baseline for KHDS was collected between 1991-1994; For more information on the KHDS tracking activities and tracking rates,

see http://www.edi-africa.com/research/khds/tracking.htm

7

educational outcomeif = α+ β1(Tif ) + β2(Aif ) + β3(Xif ) + µf + δif (12)

We outline in Section 5.2 under Primary Study Outcomes empirical proxies that we plan on usingto estimate (11) and (12). β1 in the equation above is the key parameter of interest and it capturesthe educational attainment e�ect of mother's micronutrient endowment during pregnancy. To examinewhether the fetal e�ects of micronutrients are stronger for females, we will also run the above regressionseparately by gender.

We anticipate a threshold level of micronutrient de�ciency below which rates are too low to observe asigni�cant treatment e�ect, and a second threshold (e.g. 400 mcg of folic acid) above which the treatmentwill be insu�cient to protect against adverse e�ects. The treatment impact among individuals in the lowertercile of the treated population is likely to be larger than it is for individuals with the highest baselinerates of micronutrient de�ciency. These predictions can be tested by studying variation in program e�ectby level of consumption of foods rich in naturally occurring folate (or naturally occurring B6 and B12).Both participants' blood sample information at baseline and food consumption for foods rich in naturallyoccurring folate will provide us with information to identify heterogeneous e�ects.

To test our second prediction, we exploit within- and between- family sibling variation. We use acombination of the randomization and a di�erence-in-di�erence technique to use the siblings, born beforethe index RCT child, as a control group (∆ = Sif t−Sif t−1). We restrict our speci�cation to the following:

parental investmentif = α+ β1(Tif ) + β2∆ + β3 ∗ T ∗∆ +Xif + δif (13)

4 is a di�erence of parental investment between siblings within the same family, and X includesbinary controls for gender, sex-speci�c birth order and other socio-economic variables. β3 in the equationabove is the key parameter of interest for the di�erence-in-di�erence estimation and it captures theparental investment response within the family in response to the micronutrient endowment increaseduring pregnancy. We outline in Section 5.2 under Parental Behavior Responses empirical proxies thatwe plan on using to estimate (13) and (14).

Closely related to the second prediction, we also test for parental responses across treatment andplacebo groups upon un-blinding of the study with:

parental investmentif = α+ β1(Tif ) + β2(Aif ) + β3(Xif ) + µf + δif (14)

parental investment in the speci�cation above will focus on post-intervention outcomes that a�ectinfant health after birth other than the micronutrient supplements. β1 in the equation above is thekey parameter of interest and it captures the di�erence between treatment and control groups in post-intervention health investments a�ecting infant health. Based on Peltzman (1975), the idea we subjectto testing is that the treatment group might in theory become more complacent regarding engaging inhealth investments into infant health than the control group, once the trial's results were �un-blinded�.

5.2 Primary Study Outcomes

The majority of the children in the original RCT were born in 2001-2002. Therefore, most of the childrenin the sample are ages 9-10. To proxy outcomes described in the empirical speci�cations, we collect dataon:

8

Cognitive Development and Non-Cognitive Skills

Because abilities are multiple in nature, we administer a battery of cognitive tests and non-cognitive teststo children for at least 2 children in both treatment and control households. Broadly, we test memory,access to information tasks, speed of processing, verbal acquisition and impatience. All cognitive testsdescribed below were used in the Makwami study, conducted in Bagamoyo, Tanzania and designed by ateam of development psychologists at the Harvard Graduate School of Education.16

To test memory, we administer a Digit Span Test, assessing short-term memory for strings of orallypresented digits (in order of presentation and then, in a separate test, in reverse order), a Categorical

Fluency Test, assessing the number of animals and food types children can name in two one-minutesessions, and a Corsi Block Test, a test designed to test spatial memory. To test verbal ability, weadminister the Peabody Picture Vocabulary Test (PPVT) with questions equivalent in Swahili.17We testpsycho-motor skills with a Pegboard Task, in which the child has to insert a number of awkwardly shapedpegs into a specially made board.

We also administer tests for various non-cognitive character traits such as determination, dependability,persistence, self-esteem, optimism, and time preference based on Sternberg (1985), Goodman (1999)and Goodman and Scott (1999). For example, to test impatience, we conduct a cookie version of theMarshmallow Test based on Shoda et al. (1990). The cookie test is conducted by presenting a cookieto the child. The child is given an option to eat the cookie now or asked to wait for �ve minutes by noteating the cookie for a �nal outcome of two cookies. If the child decides not to wait, he or she could eatthe �rst cookie within the �rst �ve minutes. In the case the child eating the cookie within the �rst �veminutes, we record the number of seconds it takes the child until he starts eating the cookie.

Enrollment and Attendance Status

We collect an array of current and past year's schooling status of the children in the interviewedhouseholds. In particular, we collect information for both the treatment and control individuals forseveral key variables: (1) enrollment rate status, (2) current and past measures of school attendance, (3)current and past school passing rates, (4) current and past schooling examination results.

Parental Behavior Responses

Parental time allocation, monetary expenditures on children's skills, health, learning, motivation,development of �credentials� are all important forms of parental investment in the human capital ofchildren (Becker and Tomes 1986, p. S5) and ultimately could in�uence both cognitive and non-cognitivedevelopment and schooling outcomes.18 We collect current and recall data on a number of measures:parental perceptions of their children's development throughout various stages in childhood, parentaleducational expenditures, parental home investment (e.g. hours per week read to own children or playedwith own children, self-reported warmth, parental time for emotional support and months breastfed),child vaccinations, post-natal parental behavior and non-cognitive parental assessment based on Goodman(1999) and Goodman and Scott (1999).

16See Sternberg et al. (2002).17All measures were piloted and tested for validity and test-retest reliability. All testing was done in the child's language of preference.18

Surveying past parental investment raises telescoping and memory lapse problems, which could lead to recall biases (Chen, Mu and

Ravallion, 2006).

9

Secondary Outcomes Measures

In the survey instrument, we also collect information on other health and non-health outcomes: (1)child's past and current physical health based on clinical records and self-reported data (BMI, variousillness episodes, motor functioning, vision, hearing, pain, sleep and speaking), (2) current household andindividual wealth, asset and expenditure information, (3) current and past household demographic, fertilityand educational history, (4) current and past household income generating activities, (5) current and pasthousehold member clinical and self-reported illness information, (6) location and migration history.

5.3 Potential Threats to Internal Validity

5.3.1 Sample Attrition

Despite the experimental design of the study, causal inference could fail in the presence of high di�er-ential attrition across treatment groups. If key explanatory variables, and most importantly micronutrienttreatment assignment, were strongly related to attrition, then resulting estimates will su�er from bias.In what follows, we report preliminary attrition results based on our pilot phase activities conducted in2011.

Tracking of households proceeds in two phases: regular tracking and intensive tracking. We reporthere e�ective tracking rates (ETR), calculated as a fraction of those found, or not found but searched forduring intensive tracking, with weights adjusted properly. The ETR is a function of the regular phase

tracking rate (RTR) and intensive phase tracking rate (ITR) as follows:

ETR = RTR+ (1�RTR) ∗ ITR

This is closely related to the tracking approach employed in the Moving to Opportunity project (Klinget al. 2007, Orr et al. 2003). Based on our pilot e�ort so far, the RTR is 77.3% and the ITR is 63.3%,which implies an e�ective tracking rate of 91.69. More than 98% of the attrition is driven by loss ofcontact and not mortality. Table 2 (Appendix A) presents how the tracked sample compares to theoriginal RCT population.

Reassuringly, survey tracking rates are nearly identical in the treatment and control groups. Withsurveys collected so far, we ran a regression of attrition status in the follow-up wave against baselinetreatment status and various baseline socio-economic variables. No coe�cients in this estimation werestatistically signi�cant indicating no evidence of di�erential attrition (Table 3, Appendix A).

5.3.2 Estimating Treatment Externalities

The estimation of treatment e�ects in this paper could be complicated by the possibility of spillovere�ects. These e�ects could occur either for units in the control group or for non-treated siblings inhouseholds with treated children.19 The most pertinent possibility is a behavioral response: cognitiveimprovement among treated children enhances sibling rivalry and incentives among non-treated siblingsfor more parental attention or human capital accumulation. One approach is to estimate the AverageIndirect Treatment E�ect (ITE ) is:

ITE = E(C1=C0|T = 1, E = 0),

where C1 and C0 are behavioral outcomes (e.g. learning, cognitive development) for siblings withinfamilies of respectively treated and non-treated children. E = 1 denotes treatment eligibility and E = 0ineligibility. Because of the experimental design, to test for indirect e�ect, one can simply compare average

19

While spillovers lead to biased treatment estimates is provided in Angrist, Imbens, and Rubin (1996) .

10

behavior outcome (e.g. learning e�ort) of eligibles and ineligibles in treatment and control households.However, even when based on experimental data, this simple estimator of ITE above is a conservativebound of the true spillover e�ect and subject to two assumptions (VanderWeele and Tchetgen, 2011). Toincrease the precision of the estimates, it is common practice to estimate the above parameters using aregression, adding predetermined determinants of the outcome we choose to focus on (e.g. learning e�ort).For more sophisticated estimation of ITE, we plan to follow the approach of recent papers, such as Du�oand Saez (2002), Miguel and Kremer (2002), Katz, Kling, and Liebman (2001), Kremer and Levy (2001),and Sacerdote (2001) that use individual-level randomization of treatment.

5.3.3 Intent to Treat E�ects on Mediating Factors

The supplementation treatment may also create di�erential change in a nexus of several mediatingfactors which in�uence the outcome of interest through a channel other than the micronutrient supplemen-tation. For example, it may be that experimental treatment directly induces improvements in cognitivedevelopment20 which in�uences educational attainment and in addition experimental treatment has anindirect e�ect on education through a child's physical health (Miguel and Kremer, 2002); let's call thisindirect e�ect Mit.

21 The reason why we may want to disentangle the magnitude of these two e�ects hasto do with the implications of these two e�ects for long-run impact on economic outcomes. Enhancementin children's cognitive capacity have investment e�ects potentially much higher than children's health asthe bene�ts of a cognitive improvement multiply over time.

More generally, let Mi(t) denote the potential value of a mediator of interest for unit i under thetreatment status Ti = t. Yi denote outcome, Xi denote co-variates. Following Robins and Greenland(1992), we de�ne indirect e�ects, or causal mediation e�ects, for each unit i as:

δi(t) ≡ Yi {t,Mi(1)}=Yi {t,Mi(0)}

for each treatment status t = 0, 1. This causal quantity is the change in the outcome corresponding toa change in the mediator from the value that would be realized under the control condition, Mi(0), to thevalue that would be observed under the treatment condition, Mi(1), while holding the treatment statusconstant at t. We can also de�ne the direct e�ects of the treatment as :

ςi(t) ≡ Yi {1,Mi(t)}=Yi {0,Mi(t)}

for each unit i and each treatment status t = 0, 1.In practice, just as with treatment e�ects, we are interested in an average of the mediation e�ects (or

how much of the schooling e�ect is through health). This is the average causal mediation e�ect (ACME)

δ̄(t) and is de�ned as δ̄(t) ≡ E[Yi {t,Mi(1)}=Yi {t,Mi(0)}]. Similarly, the average direct e�ect (ADE) isς̄(t) ≡ E[Yi {1,Mi(1)}=Yi {0,Mi(0)}]. The �rst e�ect, δ̄(t), is the one working through health status andthe second e�ect, ς̄(t), works through cognition. Running an OLS of (12) with variables for health andcognitive development will su�er from post-treatment bias.

With an assumption of sequential ignorability (Imai, Keele, and Yamamoto, 2010b), we can estimateς̄(t) and δ̄(t) either parametrically or non-parametrically. Parametrically, we can implement a linearstructural estimation model (LSEM ) by essentially estimating two equations:

20We assume that the direct e�ect on education is coming only through improved cognition.21

Another plausible indirect mechanism can work through kid's health status in�uence on subsequent parental labor force participation

or parental hours worked: poor children's health requires more parental time leading to decreased parental labor force participation

and lower earnings; this, in turn, a�ects parental investment in their children's human capital. However, for simplicity and without

loss of generality, we illustrate how to handle mediating factors only with one channel here.

11

Mi = α2 + β2Ti + ε2i,

Yi = α3 + β3Ti +Mi + ε3i.

We can follow Baron and Kenny (1986) to estimate the last two equations by �tting two separatelinear regressions.22

Non-parametrically, we can follow Imai, Keele, and Tingley (2010a). We can non-parametrically modelmtm(x) ≡ E(Yi|Ti = t,Mi = m,Xi = x) and yt(x) = p(Mi|Ti = t,Xi = x). Then, we can use the following

estimator: δ̂(t) = 1nK

n∑i=1

K∑k=1

{µ̂tm̃

(k)1i

(Xi)− µ̂tm̃(k)1i

(Xi)},

where m̃(k)1i is the k-th Monte Carlo draw of the mediator Mi from its predicted distribution based on

the �tted model�ψt(Xi).

6 Sample Design and Tracking Activities

6.1 Power Calculations

In choosing the number of participants to follow up with, we conducted power calculations to ensure asample size with adequate statistical power that a false null hypothesis will be correctly rejected (Cohen1988). Using the statistical package Optimal Design, we computed this probability under three di�erentscenarios related to the expected e�ect size.23 For the medium scenario at a conventional 0.8 level24, weneed to follow up with slightly over 1000 households.

6.2 Two-stage Tracking

Tracking of children whose mothers participated in the RCT proceeds in two phases: regular trackingand intensive tracking. Regular tracking is based on a random sub-sample (approximately 3000) of theoriginal study population (8379 births). After 3 months, we found and successfully interviewed almost80% of children and parents comprising our sub-sample. Of the remaining unfound respondents of thissub-sample, we selected a quarter for more intensive tracking. The intent of the intensive tracking is toobserve when regular tracking begins to wind down, to choose one quarter of the remaining unfound focusrespondents to be sought intensively for another 6 months, and then to multiply the sampling weightsof these individuals by four to account for the unfound focus respondents who were not tracked duringthe �nal months of �eld interview enumeration. Following this approach, we maintain a representativesample by merely re-weighting individuals25 of this intensive sub-sample in the �nal analysis, so that eachrespondent in the intensive sample represents the other unfound individuals who were not intensivelytracked.

22The original Baron and Kenny (1986) method does not fully accommodate settings in which the exposure and the mediator

interact in their e�ects on the outcome. We adopt an approach based on VanderWeele and Vansteelandt (2009, 2010a) to deal with

these settings.23The three scenarios are based on the magnitude of the e�ect of iodine treatment program on schooling in Tanzania based on Field

et al . (2008).24See http://www.statsoft.com/textbook/power-analysis/25The probability weights applied to individuals in this intensive tracking sample will be adjusted in the �nal data set. Sampling

will be performed in STATA 12 and, in general, proceeds as follows: within each stratum, children are assigned a uniform (0,1] random

variable, children are sorted by this variable, and the �rst n children in that stratum are assigned to the sample. Creation of the

intensive tracking sample is performed in STATA using the �sample� command.

12

For each household tracked, we conduct an interview with a parent and at least two children: onechild will be the index child in utero at the time of the original trial and all kids born before the originalRCT.26

7 Results

As of December 1, 2011, we attempted tracking 519 households from original RCT (both regular andintensive tracking):

Treatment Group Control Group

Tracked 205 196

Attritors 59 59

Tables 5-7 (Appendix A) present very preliminary results on cognitive and health outcomes forspeci�cation (11) from Sub-section 5.1. So far, with a little less than half of the necessary target sample,we �nd some evidence of impact of micronutrients on cognitive outcomes. So far, we �nd no evidence onhealth outcomes although for the height outcome speci�cations, the results are very close to the 10% levelof signi�cance. Data collection is still ongoing but we end data collection activities in April 2012.

At the CEPR conference, I plan to present preliminary results for all speci�cations (11) - (14) withstrong emphasis on the behavioral response speci�cations.

8 Research Contribution and Policy Implications

In utero exposure to four speci�c micronutrients can biologically, irreversibly, and permanently a�ectsubsequent cognitive development, educational attainment and ultimately economic outcomes. Our storyis quite simple: the proposed study aims to contribute to the economics literature by calibrating theimportance of in utero micronutrient de�ciency for human capital formation, and improves on previousresearch in several key dimensions. First, the study will be based on a randomized trial, a technique entirelyremoving any selection bias. Second, in contrast to previous studies - because the initial randomization wasat the individual level and because of a much larger sample size (8469 children) - our ability to detect e�ectsis signi�cantly greater.27Third, we improve on previous studies' rather narrow focus on academic andcognitive tests not only by administering a wider set of cognitive tests but also by using the Primary SchoolLeaving Examination (PSLE) pass rates by gender, available from the Tanzania Ministry of Education.28Afourth area of scienti�c contribution will cast light on recent �eld evidence that female fetuses are moresensitive than male ones to in utero folic acid exposure.29Due to the individual randomization and the

26As already explained in the pilot phase section, this selection of children born before the RCT is in e�ort to avoid endogeneity of

parental behavior responses with the respect to child quality.27For instance, in the well-known INCAP study, Martorell et al. (1995) provided di�erent nutritional supplements to Guatemalan

children and later �nd signi�cant impacts on their cognitive skills during adolescence. However, that study randomly assigned children

to the treatment and comparison groups at the village level, and thus has an e�ective sample size of only four villages.28The current study will use data from the Tanzania Standard 1-4 Examinations.29See Berry et al. (1999).

13

follow-up information, we will be able to rule out30 the possibility that gender di�erences are driven bysex-speci�c household responses to improvements in cognition rather than disproportionate increases infemale cognitive capacity. Finally, we pin down a speci�c mechanism through which health can a�ectincome via education by even more narrowly disentangling the individual importance of physical health,cognitive development and increases in longevity.

The project's results, combined with previous positive �ndings on the long-term health e�ects of theprenatal period (Almond 2006), help explain the gradient between adult health and economic outcomes.That fetal health may be at the fulcrum of this relationship also su�ers no shortage of policy implications.First, the large gap in the research literature on in utero health may be causing us to miscalculatethe bene�ts of nutrition programs precisely because they do not account for the cognitive developmentchannel, which we focus on in this proposal. Even holding schooling attainment constant, small di�erencesin average IQ at the group level could have large e�ects on social and economic outcomes. Second, ourresults will provide answer to the practical challenge of identifying the e�ects of a particular improvementto fetal health and whether public policies that achieve these improvements are cost-e�ective. Finally,results from this study could help policy makers accurately prioritize nutrition interventions and moreclearly understand how to improve education in developing countries. The existence of linkages betweenfetal health and economic outcomes imply that resources are not being allocated optimally across the lifecycle: individual investments and public policies that bene�t maternal and fetal health have been under-funded if fetal origins e�ects have not been accounted for in expenditure decisions, as they presumablyhave not. Therefore, social welfare can be substantially improved.

30Field, Robles and Torero (2008) can not distinguish between the biologic mechanism and sex-speci�c household responses due to

natural experiment design at the district level and dataset limitations.

14

References

[1] Adams P., M.D. Hurd, D.L. McFadden, A. Merrill and T. Ribeiro, 2003., Healthy, wealthy, and wise?Tests for direct causal paths between health and socioeconomic status, Journal of Econometrics 112:3�56.

[2] Aizer, Anna and Flavio Cunha. 2010. �Child Endowments, Parental Investments andthe Development of Human capital: Evidence from Siblings.� working paper URL:http://www.bu.edu/econ/�les/2011/03/endowments-investments-development-human-capital.pdf

[3] Allen, L. H., Penland, J. G., Boy, E., DeBaessa, Y. & Rogers, L. M. 1999. Cognitive and neuromotorperformance of Guatemalan schoolers with de�cient, marginal and normal plasma B-12. FASEB J.13: A544.

[4] Almond, Douglas 2006. �Is the 1918 In�uenza Pandemic Over? Long-Term E�ects of In UteroIn�uenza Exposure in the Post-1940 U.S. Population,� Journal of Political Economy, v. 114, pp.672�712.

[5] Alpert, J. E. & Fava, M. (1997) Nutrition and depression: the role of folate. Nutr. Rev. 55:145-149.

[6] Angrist, J., G.W. Imbens, and D. Rubin. 1996. �Identi�cation of Causal E�ects Using InstrumentalVariables,� Journal of the American Statistical Association, 91434., 444-472.

[7] Auster, R., Leveson, I. and Sarachek, D. 1969. The production of health: an exploratory study.Journal of Human Resources 4, 411�436.

[8] Baird, Sarah, Joan Hamory and Edward Miguel 2007. �Lessons from the Field: Attrition, Migrationand Data Quality in the Kenya Life Panel Survey.� Working Paper.

[9] Baker, S. J., Nutritional Anemias. Part 2: Tropical Asia., Clinical Haemat., 10, 843, 1981.

[10] Baron, R. M. and Kenny, D. A. (1986). The moderator�mediator variable distinction in socialpsychological research: Conceptual, strategic, and statistical considerations. Journal of Personalityand Social Psychology 51 1173�1182.

[11] Barro, R. 1997., Determinants of Economic Growth: a Cross-Country Study, MIT Press, Cambridge,U.S.A.

[12] Barro, R. & Lee, J.-W. 1993., `International Comparisons of Educational Attainment', Journal ofMonetary Economics 32, 363�394.

[13] Basta SS, Soekirman Karyadi D, Scrimsha NS. 1979. Iron de�ciency anemia and the productivityof adult males in Indonesia. American Journal of Clinical Nutrition 32: 916-25.

[14] Baydar, N. and J. Brooks-Gunn. 1991. �E�ects of Maternal Employment and Child-Care Arrange-ments on Preschoolers' Cognitive and Behavioral Outcomes: Evidence from the Children of theNational Longitudinal Survey.� Development Psychology 27(6): 932-45.

[15] Beaton GH, Martorell R, L'Abbe KA, Edmonston B, McCabe G, Ross AC and Harvey B. 1992.E�ectiveness of Vitamin A Supplementation in the Control of Young Child Morbidity and Mortalityin Developing Countries. SCN State-of-the-Art Nutrition Policy Discussion Paper No. 13.

15

[16] Becker, Gary S., 1964. Human Capital. New York: Columbia University Press.

[17] Becker, Gary S. 1974. �A Theory of Social Interactions.� Journal of Political Economy 87: 1063-93.

[18] Becker, Gary S., A Treatise on the Family (Cambridge: Harvard University Press, 1981).

[19] Becker, Garty S. and Tomes, Nigel. "Human Capital and the Rise and Fall of Families," J. Lab.Econ., July 1986, 4(3, Part 2), pp. S1-39.

[20] Behrman, Jere R. 1997. Intrahousehold Distribution and the Family in Rosenzweig, Mark R. andOded Stark (Eds.) Handbook of Population and Family Economics. Amsterdam: Elsevier.

[21] Behrman, Jere R., Robert A. Pollak, and Paul Taubman, 1982, �Parental Preferences and theProvision of Progeny,� Journal of Political Economy 90, No. 1, 52-73.

[22] Behrman, J. and Wolfe, B. 1989. Does more schooling make women better nourished and healthier?:adult sibling random and �xed e�ects estimates for Nicaragua. Journal of Human Resources 24,644�663.

[23] Behrman, Jere R., �The Action of Human Resources and Poverty on One Another,� Working Paper74, Living Standards Measurement Studies 1990.

[24] Behrman, Jere R., Robert A. Pollak. and Paul Taubman, 1995, �The Wealth Model: E�ciency andDistribution in the Family,� in Jere R. Behramn, Robert A. Pollak. and Paul Taubman, eds., FromParent to Child, pp.113-38. Chicago: University of Chicago Press.

[25] Behrman JR, Hoddinott J, Maluccio J, Martorell R, Quisumbing A, Stein A. 2003. The Impactof Experimental Nutritional Intervention in Childhood on Education Among Guatemalan Adults.International Food Policy Research Institute: Brief Discussion Paper 207.

[26] Bendich, A. Antioxidant vitamins and immune responses. In: Nutrition and Immunology R. Liss,New York 1988.

[27] Bendich, A. and M. Cohen, B vitamins: e�ects on speci�c and non-speci�c immune responses. In:R.R. Chandra Editor, Nutrition and Immunology Liss, New York 1988.

[28] Ben-Porath, Yoram, �The Production of Human Capital and the Life Cycle of Earnings,� Journalof Political Economy, 75 (1967), 352�365.

[29] Berger, M. and Leigh, J. P. 1989. Schooling, self-selection, and health. Journal of Political Economy24, 433�455.

[30] Berkman DS, Lescano AG, Gilman RH, Lopez S, Black MM. 2002. E�ects of stunting, diarrhoealdisease, and parasitic infection during infancy on cognition in late childhood: A follow-up study.Lancet 359: 564�571.

[31] Berry RJ, Li Z, Erickson JD, Li S, Moore CA, Wang H, Mulinare J, Zhao P, Wong LY, Gindler J,Hong SX, Correa A. Prevention of neural-tube defects with folic acid in China. China-U.S. Collab-orative Project for Neural Tube Defect Prevention. N Engl J Med. 1999;341:1485�1490.

[32] Bils, Mark and Peter Klenow, Does Schooling Cause Growth?, American Economic Review, Decem-ber 2000, 90, 11601183.

16

[33] Blau, F. D., and A. J. Grossberg. 1992. �Maternal Labor Supply and Children's Cognitive Develop-ment.� Review of Economics and Statistics 74(3): 474-81.

[34] Bottiglieri, T., Crellin, R. F. & Reynolds, E. H. (1995) Folate and neuropsychiatry. Bailey, L. B.eds. Folate in Health and Disease 1995:435-462 Marcel Dekker New York, NY.

[35] Botto LD, Moore CA, Khoury MJ, Erickson JD. Neural-tube defects. N Engl J Med 1999;341:1509-1519.

[36] Bryan J, Calvaresi E, Hughes D. Short-term folate, vitamin B-12 or vitamin B-6 supplemen-tation slightly a�ects memory performance but not mood in women of various ages. J Nutr.2002;132:1345�1356

[37] Bureau of Statistics. Tanzania Demographic Health Survey 1996: Dar Es Salaam. Calvaton, MD:Bureau of Statistics, 1996.

[38] Cao XY, Jiang XM, Dou ZH (1994). Timing of vulnerability of the brain to iodine de�ciency inendemic cretinism. New England Journal of Medicine 331: 1739�44.

[39] Carneiro, Pedro, Flavio Cunha, and James J. Heckman. 2003. �Interpreting the Evidence of FamilyIn�uence on Child Development.� In The Economics of Early Childhood Development: Lessons forEconomic Policy. Minneapolis, Minnesota: The Federal Reserve Bank. Presented at �The Economicsof Early Childhood Development: Lessons for Economic Policy Conference,� Minneapolis FederalReserve Bank, Minneapolis, MN. October 17, 2003.

[40] Case, Anne, Angela Fertig and Christina Paxson. "The Lasting Impact Of Childhood Health AndCircumstances," Journal of Health Economics, 2005, v24(2,Mar), 365-389.

[41] Cattell RB. 1943. The measurement of adult intelligence. Psychol Bull 3: 618�648.

[42] Chang SM, Walker SP, Grantham-McGregor S, Powell CA. 2002. Early childhood stunting and laterbehaviour and school achievement. J Child Psychol Psychiatry 43: 775�783.

[43] Chandra, R.K and D. Vyas, Vitamin A, immunocompetence, and infection. Food Nutr Bull 11 1989.,pp. 12�19.

[44] Chen, S., Mu, R. and Ravallion, M., Are there Lasting Impacts of a Poor-Area Development Pro-gram?, Mimeo, Development Research Group, World Bank, October 2006.

[45] Christensen B, Yang H, Gravel RA, Rozen R. A common variant in methionine synthase reduc-tase combined with low cobalamin (vitamin B12) increases risk for spina bi�da. Mol Genet Metab1999a;67:317-23.

[46] Christensen B, Arbour L, Tran P, et al. Genetic polymorphisms in methylenetetrahydrofolate re-ductase and methionine synthase, folate levels in red blood cells, and risk of neural tube defects.Am J Med Genet 1999b; 84:151-7.

[47] Cohen, Jacob. 1988. Statistical Power Analysis for the Behavioral Sciences. Hillsdale, N.J.:LawrenceErlbaum Associates.

[48] Corman, H., T. J. Joyce and M. Grossman. 1987. "Birth Outcome Production Functions in theU.S." Journal of Human Resources, 223.: 339-360.

17

[49] Czeizel AE, Dudás I. Prevention of the �rst occurrence of neural-tube defects by periconceptionalvitamin supplementation. N Engl J Med 1992; 327:1832-5.

[50] Dasgupta, Partha. 1993. Altruism and the Allocation of Resources. Social Service Review 67.3:374-87.

[51] Datar, Ashlesha, M. Rebecca Kilburn and David S. Loughran. 2006. Health Endowments andParental Investments in Infancy and Early Childhood. Labor and Population Working Paper WR-367. Santa Monica: RAND Corporation.

[52] DeMaeyer, E. and Adiels-Tegman, M., The prevalence of anaemia in the world, World Health Stat.

Q., 38-302, 1985

[53] Du�o, E, and Emmanuel Saez 2002.: �The Role of Information and Social Interactions in RetirementPlan Decisions: Evidence from a Randomized Experiment,� unpublished working paper, MIT andUniversity of California, Berkeley.

[54] Edwards, L. and Grossman, M. 1979. The Relationship Between Children's Health and IntellectualDevelopment. In Health: What Is It Worth? Edited by Mushkin, S. and Dunlop, D.., New York:Pergamon Press.

[55] FAO. 1995. FAOSTAT database.

[56] FAO. 1996a. The Sixth World Food Survey. Rome.

[57] FAO. 1996b. Food, agriculture and food security: developments since the World Food Conferenceand prospects. World Food Summit technical background document 1. World Food Summit technicalbackground documents, Vol. 1. Rome.

[58] Farrell, P. and Fuchs, V. 1982. Schooling and Health: The Cigarette Connection. Journal of HealthEconomics 1, 217�230.

[59] Fawzi WW et al. Randomised trial of e�ects of vitamin supplements on pregnancy outcomes andT-cell counts in HIV-1 infected women in Tanzania. Lancet 351: 1477-1482, 1998.

[60] Fawzi WW et al. Rationale and design of the Tanzania Vitamin and HIV Infection Trial. ControlledClinical Trials, 1999, 20:75-90.

[61] Fawzi WW, Msamanga GI, Hunter D, et al. Randomized trial of vitamin supplements in relation totransmission of HIV-1 through breastfeeding and early child mortality. AIDS 2002;16:1935-44.

[62] Fawzi, WW et al. Vitamins and Perinatal Outcomes among HIV-Negative Women in Tanzania.NEJM. 35614.: 1423-1431, 2007.

[63] Fernald, L., P. Gertler, and L. Neufeld. 2008. �Role of Cash in Conditional Cash Transfer Pro-grammes for Child Health, Growth, and Development: An Analysis of Mexico's Oportunidades.�The Lancet 371: 828-837.

[64] Field, Erica, Omar Robles, and Maximo Torero. 2009. �Iodine De�ciency and Schooling Attainmentin Tanzania,� American Economic Journal: Applied Economics. Vol. 1, no. 4: pp. 140-169.

[65] Filmer, Deon, and Lant H. Pritchett. 2001. Estimating wealth e�ects without expenditure data - orTears: An application to educational enrollments in states of India. Demography 38 (1), February:115-132.

18

[66] Fioravanti M, Ferrario E, MassaiaM, et al. Low folate levels in the cognitive decline of elderlypatients and the e�cacy of folate as a treatment for improving memory de�cits. Arch GerontolGeriatr. 1997;26:1-13.

[67] Fleming, A. F., Tropical obstetrics and gynaecology. I: Anaemia in pregnancy in Tropical Africa,Trans. Royal. Soc. Trop. Med. Hyg., 83, 441, 1989.

[68] Fortson, J. 2011. Mortality Risk and Human Capital Investment: The Impact of HIV/AIDS inSub-Saharan Africa. Review of Economics and Statistics 2011 93:1, 1-15

[69] Fuchs, V. 1982. Time Preference and Health: An Exploratory Study, In Economic Aspects of Health,ed. Fuchs, V. New York, Columbia University Press.

[70] Glasizou PP, Mackerras DEM. 1993. Vitamin A supplementation in infectious diseases: a metaanal-ysis. British Medical. Journal 306: 366-70.

[71] Goldman, N. 2001. "Social ineqalities in health: Disentangling the underlying mechanisms," in We-instein, M., A. Hermalin and M. Stoto, eds.., Population, Health and Aging: strengthening thedialogue between demography and epidemiology, New York: Annals of the New York Academy ofSciences.

[72] Goodman R (1999), The extended version of the Strengths and Di�culties Questionnaire as a guideto child psychiatric caseness and consequent burden. J Child Psychol Psychiatry 40:791-799.

[73] Goodman R, Scott S (1999), Comparing the Strengths and Di�culties Questionnaire and the ChildBehavior Checklist: is small beautiful? J Abnorm Child Psychol 27:17-24.

[74] Grantham-McGregor S, Powell C, Walker S, Chang S, Fletcher P. 1994. The long-term follow-up ofseverely malnourished children who participated in an intervention program. Child Dev 65: 428�439.

[75] Grantham-McGregor, S. & Ani, C. 1997. The role of micronutrients in psychomotor and cognitivedevelopment. Br. Med. Bull. 55: 511-527.

[76] Grossman, M. 1976. The Correlation Between Health and Schooling. In Household Production and

Consumption, ed. Terleckyj, N. E. New York, Columbia University Press.

[77] Grossman, M. and Joyce, T. J. 1987. Socioeconomic Status and Health: A Personal ResearchPerspective. Paper presented at the Kaiser Foundation Conference on Socioeconomic Status andHealth, March 26�27.

[78] Guilarte, T R. Vitamin B6 and cognitive development: recent research �ndings from human andanimal studies. Nutr-Rev. 1993 Jul; 517.: 193-8.

[79] Hankey, G. J. & Eikelboom, J. W. (1999) Homocysteine and vascular disease. Lancet 354:407-413.

[80] Hetzel BS, Mano M (1989). A review of experimental studies of iodine de�ciency during fetal de-velopment, Journal of Nutrition 119: 145-151.

[81] Hillman RS, Ault KA. Hematology in clinical practice - A guide to diagnosis and management. NewYork: McGraw-Hill Inc; 1994.

[82] Hirano, K., G.W. Imbens and G. Ridder (2000), �E�cient Estimation of Average Treatment E�ectsusing the Estimated Propensity Score�, mimeo.

19

[83] Humphrey JH, Ili� PJ, Marinda ET, et al. E�ects of a single large dose of vitamin A, given duringthe postpartum period to HIV-positive women and their infants, on child HIV infection, HIV-freesurvival, and mortality. J Infect Dis 2006;193:860-71.

[84] Hadidjaja P., E. Bonang, M. A. Suyardi, S. A. Abidin, I. S. Ismid, and S. S. Margono. 1998. Thee�ect of intervention methods on nutritional status and cognitive function of primary school childreninfected with Ascaris lumbricoides. American Journal of Tropical Medicine & Hygiene 595. 791-5.

[85] Imai, K., L. Keele, and D. Tingley. 2010a. A general approach to causal mediation analysis. Psy-chological Methods 15: 309�334.

[86] Imai, K., L. Keele, and T. Yamamoto. 2010b. Identi�cation, inference, and sensitivity analysis forcausal mediation e�ects. Statistical Science 25: 51�71.

[87] Jayachandran, Seema, and Adriana Lleras-Muney (2009). �Life Expectancy and Human CapitalInvestments: Evidence from Maternal Mortality Declines�, Quarterly Journal Economics, 124(1):349�398.

[88] Jukes MCH, Nokes CA, Alcock KJ, Lambo J, Kihamia C, et al. 2002. Heavy schistosomiasis asso-ciated with poor short-term memory and slower reaction times in Tanzanian school children. TropMed Int Health 7: 104�117.

[89] Kalemli-Ozcan, S., Ryder, H., Weil, D. N., 2000. Mortality Decline, Human Capital Investment andEconomic Growth. Journal of Development Economics 62, 1-23.

[90] Kalemli-Ozcan, Sebnem. 2002. �Does the Mortality Decline Promote Economic Growth?� Journalof Economic Growth 7 4: 411�39.

[91] Katz, L., J. Kling, and J. Liebman 2001.: �Moving to Opportunity in Boston: Early Results of aRandomized Mobility Experiment,� Quarterly Journal of Economics, 116, 607�654.

[92] Kavishe, F.P. 1987. The food and nutrition situation in Tanzania. TFNC Report No. 1215. Dar-es-Salaam, United Republic of Tanzania, Tanzania Food and Nutrition Centre TFNC.

[93] Kavishe, F.P. 1991. The control of micronutrient malnutrition: the experience of Tanzania. In End-ing hidden hunger, p. 89-115. Proceedings of a Policy Conference on Micronutrient Malnutrition,Montreal, Quebec, Canada, 10-12 October 1991. Atlanta, Georgia, USA, Task Force on Child Sur-vival and Development.

[94] Kenkel, D. S. 1991. Health Behavior, Health Knowledge, and Schooling. Journal of Political Economy99, 287�305.

[95] Kenkel, D. S. 1995. Should You Eat Breakfast? Estimates from Health Production Functions. HealthEconomics 4, 15�29.

[96] Kihara M, Carter JA, Newton C. 2006. The e�ect of Plasmodium falciparum on cognition: Asystematic review. Trop Med Int Health 11: 386�397.

[97] Kremer, Michael, and Dan Levy 2001.: �Peer E�ects from Alcohol Use Among College Students,�unpublished manuscript, Harvard University.

[98] Kvalsvig J. D., R. M. Cooppan, and K. J. Connolly. 1991. The e�ects of parasite infections oncognitive processes in children. Annals of Tropical Medicine & Parasitology 855. 551-68.

20

[99] Levav M., A. F. Mirsky, P. M. Schantz, S. Castro, and M. E. Cruz. 1995. Parasitic infection inmalnourished school children: e�ects on behaviour and EEG. Parasitology 110Pt 1. 103-11.

[100] Laurence KM, James N, Miller MH, Tennant GB, Campbell H. Double- blind randomised con-trolled trial of folate treatment before conception to prevent recurrence of neural-tube defects. BMJ1981;282:1509-11.

[101] Leibenstein, Harvey. 1957. Economic backwardness and economic growth: Studies in the theory of

economic development, New York: Wiley & Sons.

[102] Leibowitz, Arleen. 1974. �Home Investments in Children� Journal of Political Economy, 82(2, part2), s111-s131.

[103] Leon, D.A., H.O. Lithell, D. Vågerö, I. Koupilová, R. Mohsen, L. Berglund, U.B. Lithell and P.M.McKeigue (1998), Reduced fetal growth rate and increased risk of death from ischaemic heart disease:cohort study of 15000 Swedish men and women born 1915-29, British Medical Journal 317, 241-245.

[104] Lindeboom, M., Portrait, F. & Van den Berg, G. J. (2010). Long-run e�ects on longevity of anutritional shock early in life: the Dutch Potato Famine of 1846-1847, Journal of Health Economics29(5): pp. 617-629.

[105] Lin, Ming-Jen, Liu, Jin-Tan and Chou, Shin-Yi,As Low Birth Weight Babies Grow, Can 'Good'Parents Bu�er this Adverse Factor? A Research Note (January 2007). NBER Working Paper No.W12857. Available at SSRN: http://ssrn.com/abstract=959132

[106] Louwman, M.W.J., van Dusseldorp, M., et al. 2000. Signs of Impaired Cognitive Function in Ado-lescents with Marginal Cobalamin Status. Am J Clin Nutr; 72:762-9.

[107] Sianesi, Barbara and Van Reenen, John. 2003 The returns to education : macroeconomics. Journalof economic surveys, 17 2. pp. 157-200. ISSN 0950-0804

[108] Maccini, S., and D. Yang (2009). Under the weather: health, schooling, and economic consequencesof early-life rainfall. American Economic Review 99(3), 1006-1026.

[109] Malouf R, Grimley Evans J. Vitamin B6 for cognition. Cochrane Database of Systematic Reviews2008, Issue 2. Art. No.: CD004393. DOI: 10.1002/14651858.CD004393.

[110] Marmot M. G., C. D. Ry�, L. L. Bumpass, M. Shipley, N. F. Marks. 1997. Social Inequalities inHealth- A Major Public Health Problem. Social Science and Medicine 44:901-910.

[111] Martorell R., Jean-Pierre Habicht, and Juan A. Rivera. 1995. �History and Design of the INCAPLongitudinal Study 1969-77. and its Follow-up 1988-89.�. Journal of Nutrition, 125, 1027-1041S.

[112] Massawe S, Urassa E, Lindmark G, Moller B, Nystrom L. Anemia in pregnancy: a major healthproblem with implications for maternal health care. Afr J Health Sci 1996; 3: 126�132.

[113] Micronutrient. 2008. VM De�ciency Country Report for Tanzania. Available athttp://www.micronutrient.org/VMD/CountryFiles/TanzaniaNPA.pdf

[114] Miguel, Edward, and Michael Kremer 2002.: �Why Don't People Take Their Medicine? Experimen-tal Evidence from Kenya,� mimeo, University of California, Berkeley and Harvard University.

[115] Miguel, Edward, and Michael Kremer 2004. �Worms: Identifying Impacts on Education and Healththe Presence of Treatment Externalities�, Econometrica, 721., 159-217.

21

[116] Montresor A., C. Urbani, B. Camara, A. B. Bha, M. Albonico, and L. Savioli. 1997. [Preliminarysurvey of a school health program implementation in Guinea]. [French]. Medecine Tropicale 573.294-8.

[117] MRC Vitamin Study Research Group. Prevention of neural tube defects: results of the MedicalResearch Council Vitamin Study. Lancet 1991; 338:131-7.

[118] Mulokozi G, Lietz G, Svanberg U, Mugyabuso JK, Henry JCK, Tomkins AM. Plasma levels of retinol,carotenoids, and tocopherols in relation to dietary pattern among pregnant Tanzanian women. IntJ Vitam Nutr Res. 2003;73:323�33.

[119] Murray CJL, Lopez AD eds.. The global burden of disease: a comprehensive assessment of mortalityand disability from diseases, injuries and risk factors in 1990 and projected to 2020. Cambridge,Harvard University Press Global Burden of Disease and Injury Series, Vol. 1. 1996.

[120] Mushkin Selma J. 1962. Health as an Investment. Journal of Political Economy 705. 129-57. NokesC., S. M. Grantham-McGregor, A. W. Sawyer, E. S. Cooper, and D. A. Bundy. 1992. Parasitichelminth infection and cognitive function in school children. Proceedings of the Royal Society of

London - Series B: Biological Sciences 2471319. 77-81.

[121] Mwabu, G. 2008. The Production of Child Health in Kenya: A Structural Model of Birth Weight.Journal of African Economies 17: 3, 1-49.

[122] Nokes, C, van den Bosch, C & Bundy, DAB. 1998. The E�ects of Iron De�ciency and Anemia onMental and Motor Performance, Educational Achievement, and Behavior in Children: an AnnotatedBibliography, International Nutritional Anemia Consultative Group (INACG) publication

[123] Osendarp SJ, West CE, Black RE. The need for maternal zinc supplementation in developing coun-tries: an unresolved issue. J Nutr 2003;133:817S-827S.

[124] Ozanne SE, Hales CN. Lifespan: Catch-up Growth and Obesity in Male Mice. Nature 2004;427:411-412.

[125] Paxson, C., and N. Schady. 2007. �Cognitive Development Among Young Children in Ecuador: TheRole of Health, Wealth and Parenting.� Journal of Human Resources 42(1): 49-84.

[126] Paxson, C., and N. Schady. 2008. �Does Money Matter? The E�ects of Cash Transfers on ChildHealth and Development in Rural Ecuador.� Unpublished manuscript, Princeton University andThe World Bank.

[127] Peltzman, Sam. 1975. The eects of automobile safety regulation. Journal of Political Economy, Vol.83(4): 677-726.

[128] Penland, J., Allen, L. H., Boy, E., DeBaessa, Y. & Rogers, L. M. 2000. Adaptive functioning,behavior problems and school performance of Guatemalan children with de�cient, marginal andnormal plasma vitamin B-12. FASEB J. 14: A561.

[129] Perri, T. 1984. Health Status and Schooling Decisions of Young Men. Economics of Education Review3, 207-213.

[130] Pharoah PO, Connolly KJ (1987). A controlled trial of iodinated oil for the prevention of endemiccretinism: a long-term follow-up. International Journal of Epidemiology 16:68 �73.

22

[131] Pitt, Mark M., Mark R. Rosenzweig, Md. Nazmul Hassan. 1990. �Productivity, Health, and Inequal-ity in the Intrahousehold Distribution of Food in Low-Income Countries,� The American EconomicReview. Vol. 80, no. 5, pp. 1139-1156.

[132] Pollitt E, Gorman KS, Engle PL, Rivera JA, Martorell R. 1995. Nutrition in early life and theful�llment of intellectual potential. J Nutr 125: S1111�S1118.

[133] Psacharopoulos, G. 1994., `Returns to education: a global update', World Development 22,1325�1343.

[134] Ramakrishnan U, Manjrekar R, Rivera J, Gonzales-Cossio T & Martorell R. 1999. Micronutrientsand pregnancy outcome: a review of the literature. Nutrition Research 19, 103-159.

[135] Rao, R. & Georgie�, M. K. 2000. Early nutrition and brain development. In: The E�ects of EarlyAdversity on Neurobehavioral Development. The Minnesota Symposium on Child Psychology Nel-son, C. A., ed.., pp. 1-30. Lawrence Erlbaum Associates Publishers, Mahwah, NJ.

[136] Raven JC, Styles I, Raven MA. 1998. Raven's progressive matrices: CPM parallel test booklet.Oxford: Oxford Psychologists Press.

[137] Robins, J. M., and S. Greenland. 1992. Identi�ability and exchangeability for direct and indirecte�ects. Epidemiology 3: 143�155.

[138] Rosenberg, I. H. & Miller, J. W. 1992. Nutritional factors in physical and cognitive functions ofelderly people. Am. J. Clin. Nutr. 55: 1237S-1243S.

[139] Rosenzweig, M. and T. P. Schultz. 1982. "The Behavior of Mothers as Inputs to Child Health: TheDeterminants of Birth Weight, Gestation, and the Rate of Fetal Growth." In Economic Aspects ofHealth, ed. V. Fuchs. National Bureau of Economic Research, Chicago: University of Chicago Press.

[140] Rosenzweig, M. and T. P. Schultz. 1982. "The Behavior of Mothers as Inputs to Child Health: TheDeterminants of Birth Weight, Gestation, and the Rate of Fetal Growth." In Economic Aspects ofHealth, ed. V. Fuchs. National Bureau of Economic Research, Chicago: University of Chicago Press.

[141] Rosenzweig, M. and Schultz, T. P. 1983. Estimating a Household Production Function: Heterogene-ity, the Demand for Health Inputs, and Their E�ects on Birth Weight. Journal of Political Economy91, 723�746.

[142] Rosenzweig, M., and T. P. Schultz. 1984. Market opportunities, genetic endowments, and intrafamilyresource distribution: Reply. American Economic Review 74: 521�522.

[143] Rosenzweig, Mark R. and Kenneth I. Wolpin. 1988. �Heterogeneity, Intrafamily Distribution, andChild Health,� The Journal of Human Resources. Vol. 23, No. 4: pp. 437-461

[144] Rosenzweig, M., and K. I. Wolpin. 1994. �Are there Increasing Returns to the IntergenerationalProduction of Human Capital? Maternal Schooling and Child Intellectual Achievement.� Journalof Human Resources 29(2): 670-93.

[145] Rosenzweig, Mark R. and Junsen Zhang. 2009. �Do Population Control Policies Induce More HumanCapital Investment? Twins, Birth Weight and China's �One-Child� Policy,� The Review of EconomicStudies. Vol. 76, No. 3: pp. 1149-1174.

[146] Ruhm, Christopher. 2004. The Journal of Human Resources Vol. 39, No. 1 (Winter, 2004), pp.155-192

23

[147] Sacerdote, Bruce 2001.: �Peer E�ects with Random Assignment: Results for Dartmouth Room-mates,� Quarterly Journal of Economics, 116, 681�704.

[148] Schneede, J., Dagnelie, P. C., Van Staveren, W. A., Vollset, S. E., Refsum, H. & Ueland, P. M. 1994.Methylmalonic acid and homocysteine in plasma as indicators of functional cobalamin de�ciency ininfants on macrobiotic diets. Pediatr. Res. 36: 194-201.

[149] Shakotko, R. A. and Grossman, M. 1982. Physical Disabilities and Post-Secondary EducationalChoices. In Economic Aspects of Health, ed. Fuchs, V. New York, Columbia University Press.

[150] Shakotko, R. A., Edwards, L. and Grossman, M. 1981. An Explanation of the Dynamic Relation-ship between Health and Cognitive Development in Adolescence. In Health, Economics, and Health

Economics, eds. van der Gagg, J. and Perlman, M. Amsterdam, North Holland.

[151] Shoda, Y., Mischel, W., Peake, P. K. (1990). �Predicting Adolescent Cognitive and Self-regulatoryCompetencies from Preschool Delay of Grati�cation: Identifying Diagnostic Conditions�. Develop-mental Psychology, 26(6), 978�986

[152] Sianesi, Barbara and Van Reenen, John. 2003. The returns to education : macroeconomics. Journalof economic surveys, 17 2. pp. 157-200. ISSN 0950-0804.

[153] Simeon DT, Grantham McGregor S. 1998. E�ects of missing breakfast on the cognitive functions ofschool children of di�ering nutritional status. Am J Clin Nutr 49: 646�653.

[154] Smith, J. P. 1999. "Healthy bodies and thick wallets: The dual relation between health and economicstatus," Journal of Economic Perspectives, 132.:145-66.

[155] Snedecor, G.W. and Cochran, W.G. 1980 Statistical Methods. 7th edition. Iowa State UniversityPress, Ames, Iowa.

[156] Sommer A. Tarwotjo I, Hussaini G, Susanto D, Soegiharto T. 1981. Incidence, prevalence, and scaleof blinding malnutrition. Lancet 1: 1407-8.

[157] Sommer A Tarwotjo I, Djunaedi E, West K, Loeden A, Tilden R, Mele L. 1986. Impact of vitamin Asupplementation on childhood mortality: A randomized controlled community trial. Lancet 1:1169-73.

[158] Sternberg, Robert J. Beyond IQ: A triarchic theory of human intelligence. Cambridge, U.K.: Cam-bridge University Press, 1985.

[159] Sternberg, R.J., Grigorenko, E.L., Ngorosho, D., Tuntufye, E., Mbise, A., Nokes, C.A., Jukes,M.C.H., & Bundy, D.A.P. (2002) Assessing intellectual potential in rural Tanzanian school children.Intelligence, 30, 141-162.

[160] Strauss, John and Duncan Thomas, �Human Resources: Empirical Modeling of Household and Fam-ily Decisions,� in Jere Behrman and T.N. Srinivasan, eds., Handbook of Development Economics,Vol. 3, Amsterdam: North Holland, 1995, chapter 34, pp. 1885-2023.

[161] Strauss, J. and D. Thomas, 1998, Health, Nutrition and Economic Development, Journal of Eco-nomic Literature 36 2.: 766�817.

[162] Taubman, P. and Rosen, S. 1982. Healthiness, Education, and Marital Status, In Economic Aspects

of Health, ed. Fuchs, V. New York, Columbia University Press.

24

[163] Taubman, Paul. 1996. The roles of the family in the formation of o�spring's' earnings and incomecapacity, in Paul Menchik (eds.) Household and family economics ,Kluwer Academic Publishers,Massachusetts.

[164] The Tanzanian Household Budget Survey THBS. Available online at:http://www.tanzania.go.tz/hbs/HomePage_HBS.html

[165] Thomas, Duncan, Elizabeth Frankenberg, and James P. Smith. 2001. �Lost but not Forgotten:Attrition and Follow-up in the Indonesia Family Life Survey�, Journal of Human Resources, 36 3.,556-592.

[166] Thomas, Duncan, Elizabeth Frankenberg, and James Smith 2002. �Lost but Not Forgotten: Attritionand Follow-Up in the Indonesia Family Life Survey�, Journal of Human Resources 36: 556-92.

[167] Thomas, Duncan et al 2003. �Iron de�ciency and the well-being of older adults: Early results from arandomized nutrition intervention�. Mimeo, UCLA. Thomas, D., and J. Strauss. 1997. "Health andwages: Evidence on men and women in urban Brazil," Journal of Econometrics, 77:159-85.

[168] Thomas D, Frankenberg E, Friedman J, Habicht J-P, Hakimi M, Jaswadi, Jones N, McKelvey C,Pelto G, Sikoki B, Seeman T, Smith H, Sumantri C, Suriastini W, Wilopo S. 2003. Iron De�ciencyand the Well Being of Older Adults: Early Results from a Randomized Nutrition Intervention.Mimeo, UCLA.

[169] Tomes, Nigel, 1981, �The Family, Inheritance, and the Intergenerational Transmission of Inequality,�Journal of Political Economy 89, No. 5, 928-58.

[170] Topel, R. 1999., The Labour Market and Economic Growth, in O. Ashenfelter & D. Card, eds, `TheHandbook of Labour Economics', North Holland, Amsterdam, p. Ch. 44.

[171] Umbel, V. M., B. Z. Pearson, M. C. Fernandez and D. K. Oller. 1992. �Measuring Bilingual Chil-dren's Receptive Vocabularies.� Child Development. 63: 1012-20.

[172] UNESCO (2007) `Education For All by 2015: Will We Make It?', Global Monitoring Report 2008,Paris: UNESCO.

[173] UNESCO Institute for Statistics. (January 1, 2008). Data Centre. In Beyond 20/20 WDS. RetrievedDecember 28, 2011, from http://stats.uis.unesco.org/unesco/ReportFolders/ReportFolders.aspx.

[174] Underwood, Barbara. GBD 1990 Chapter -Vitamin A De�ciency VAD. World Health Organization,Nutrition Programme, Food & Nutrition Division. January 1995.

[175] Van den Broek, N. R. & Letsky, E. A. 2000. Etiology of anemia in pregnancy in South Malawi. Am.J. Clin. Nutr. 72:247S-256S.

[176] VanderWeele, T. J. & Vansteelandt S. (2009). Conceptual issues concerning mediation, interventionsand composition. Statistics and Its Interface. Volume 2(4):457-468.

[177] VanderWeele, T. J. & Vansteelandt S. (2010). Odds Ratios for Mediation Analysis for a DichotomousOutcome. Am. J. Epidemiol. 172 (12):1339-1348. DOI: 10.1093/aje/kwq332

[178] VanderWeele, T.J. and Tchetgen Tchetgen, E.J. (2011). Bounding the infectiousness e�ect in vaccinetrials. Epidemiology, in press.

25

[179] Walker SP, Grantham-McGregor SM, Powell CA, Chang SM 2000. E�ects of growth restriction inearly childhood on growth, IQ, and cognition at age 11 to 12 years and the bene�ts of nutritionalsupplementation and psychosocial stimulation. J Pediatr 137: 36�41.

[180] Wechsler D. 1997. Wechsler adult intelligence scale-III. San Antonio Texas.: The PsychologicalCorporation.

[181] West KP, Howard JR, Sommer A. 1989. Vitamin A and Infection: Public health implications.Annual Review of Nutrition 9: 63-86.

[182] WHO. 1994. Nutrition: highlights of recent activities in the context of the World Declaration andPlan of Action for Nutrition. NUT/94.5. Geneva, Switzerland.

[183] WHO/UNICEF. 1995. Global prevalence of vitamin A de�ciency. MDIS Micronutrient De�ciencyInformation System. Working Paper No. 2. Geneva, Switzerland.

[184] Willis, R. (1986) "Wage Determinants: A Survey and Reinterpretation of Human Capital EarningsFunctions," Handbook of Labor Economics, Vol. 1, Elsevier, 1986, pp. 525-602.

[185] Wolfe, B. 1985. The In�uence of Health on School Outcomes: A Multivariate Approach. Medical

Care 23, 1127�1138.

[186] World Education Forum, Global Competitiveness Report 2001-2002. New York: Oxford UniversityPress, 2001.

[187] Zellner, A. 1962. An e�cient method of estimating seemingly unrelated regressions and tests foraggregation bias. Journal of the American Statistical Association 57: 348�368.

26

Appendix A: Preliminary Results

Table 1: Description of Study Participants

VariableDHS 2003/4(Tanzania)

DHS 2003/4(Dar EsSalaam)

RCT StudyParticipants

Age 27.81±8.26 24.82±6.13 25.15±5.07Education (# of Yrs) 6.71±1.99 6.81±3.32 7.11±2.86

Own Sofa n/a n/a 0.68±0.47Own TV 0.05±0.22 0.28±0.45 0.38±0.49Own Radio 0.55±0.49 0.76±0.43 0.80±0.40Own Fridge 0.03±0.19 0.23±0.42 0.21±0.41Own Fan n/a n/a 0.45±0.49

Marital Status (%Married)

71.40 53.49 63.19

# Children ever born 2.88±2.60 1.70±1.93 2.02±1.27BMI 22.5±4.35 25.1±5.54 24.61±3.90

# Obs 5233 247 8468

Notes: Std deviations are presented after mean values; Marital status is based on DHS 2005; BMI is based on DHS 2010.

Table 2: Randomization Check

VariableTreatmentGroup

ControlGroup

Maternal Age 25.23±5.10 25.10±5.04Education (in years) 7.12±2.88 7.12±2.84

Marital Status 63.31 63.18BMI 24.61±3.87 24.61±3.95Parity 1.03±1.28 1.03±1.26

Wealth Score 1.62±0.98 1.60±0.96TSH/pers per dayfor food ≤ 500

0.39±0.49 0.40±0.49

# Obs 4214 4214

Notes: St dev are presented after mean values

27

Table 3: Pilot Attrition Results

Variable Mean Original SamplePilot Phase

Tracked Sample

Age 25.15±5.07 27.91±4.80Education (# of Yrs) 7.11±2.86 6.98±2.46Wealth Score [0-3.23] 1.61±0.97 1.67±0.96

Marital Status(Married=1)

63.19 73.57

# Children ever born 2.02±1.27 2.20±1.25BMI 24.61±3.90 25.34±4.36

# Obs 8468 401

Notes: # Obs from the pilot phase includes tracked observations in regular and intensive tracking stages

Table 4: Di�erential Attrition

Dependent Variable: Attrition Status (1=yes; 0=no)

(OLS Estimates)

Treatment 0.02Age -0.01Education (# of Yrs) -0.01Wealth Index (1=Poorest, 5=Richest) -0.04Marital Status (Married=1) 0.04# Children ever born -0.10BMI -0.00

# Obs 519

Notes: *, **, *** indicate respectively 10%, 5%, and 1% level of signi�cance

28

Table 5: Results on Cognitive Scores

CognitiveOutcome

Measure AverageTreatmentvs ControlDi�erence

Corsi Block Span 1.84 (0.65) 0.19***Total Correct 10.26 (1.86) 1.31***Highest Level 4.87 (0.71) 0.45***

Digit Span Span (Forward) 1.61 (0.97) 0.30*Total Correct(Forward)

9.87 (2.28) 1.48*

Highest Level(Forward)

5.75 (0.88) 0.57

Span(Backward)

1.25 (0.72) 0.19*

Total Correct(Backward)

5.65 (1.97) 1.13*

Highest Level(Backward)

3.23 (0.79) 0.41*

29

Table 6: Results on Cognitive Scores

CognitiveOutcome

Measure Average Treatmentvs ControlDi�erence

Stroop

Average Time(Normal)

0.31 (1.26) -0.39

Average Error(Normal)

28.31 (8.83) -5.91*

Average Time(Reverse)

0.17 (0.46) -0.26*

Average Error(Reverse)

28.59 (8.91) -5.15*

FluencyAnimals 16.45 (9.03) 4.49*

Foods 16.56 (3.85) 2.99*

Pegboard Time taken 111.38(54.25)

-0.01

Table 7: Results on Health Outcomes

Health Outcome AverageTreatment vs

ControlDi�erence

Height (cm) 152.78 (11.31) 0.02Mid-upper arm circumference (cm) 21.91 (3.18) 0.13Skin-fold thickness 9.61 (4.13) -0.42

30