Fetal Iodine Deficiency and Schooling: A Replication of Field, Robles, and Torero (2009)*

• Author: Stefan Swartling Peterson,Fredrik Sävje,Niklas Bengtsson
• Published date: 01 April 2020
• DOI: http://doi.org/10.1111/sjoe.12341
• Date: 01 April 2020

Scand. J. of Economics 122(2), 582–621, 2020

DOI: 10.1111/sjoe.12341

Fetal Iodine Deficiency and Schooling: A

Replication of Field, Robles, and Torero

(2009)*

Niklas Bengtsson

Uppsala University,Uppsala, SE-752 36, Sweden

niklas.bengtsson@nek.uu.se

Fredrik S¨avje

YaleUniversity, New Haven, CT 06520, USA

fredrik.savje@yale.edu

Stefan Swartling Peterson†

Uppsala University,Uppsala, SE-752 36, Sweden

stefan.peterson@kbh.uu.se

Abstract

Scholars have theorized that congenital health endowment is an important determinant of

economic outcomes later in a person’slife. Field, Robles, and Torero(2009, American Economic

Journal: Applied Economics 1, 140–169) find large increases in educational attainment caused

by a reduction of fetal iodine deficiency following a set of iodine supplementation programs in

Tanzania. We revisit the Tanzanian iodine programs with a narrow and wide replication of the

study by Field et al. We are able to exactly replicate the original results. We find, however, that

the findings are sensitive to alternative specification choices and sample restrictions. We try to

address some of these concerns in the wide replication; we increase the sample size fourfold,

and we improve the precision of the treatment variable by incorporating new institutional and

medical insights. Despite the improvements, no effect is found. We conclude that the available

data do not provide sufficient powerto detect a possible effect, as treatment assignment cannot

be measured with sufficient precision.

Keywords: Education; fetal origins hypothesis; iodine deficiency; prenatal exposure; replication

JEL classification:I12; I21; J16; O15

*Weextend our gratitude to Douglas Almond, SebastianAxbard, Per Engstr¨om,Rita Ginja, Linna

Mart´en, Eva M¨ork,three anonymous referees, and seminar participants at Uppsala University and

EEA Gothenburg 2013 for comments and suggestions. N. Bengtsson and F. S¨avje acknowledge

financial support from the SwedishInter national DevelopmentCooperation Agency through grant

Uforsk SWE-2012-109.

†Also affiliated with UNICEF.

C

The editors of The Scandinavian Journal of Economics 2018.

N. Bengtsson, F. S¨avje, and S. Swartling Peterson 583

I. Introduction

How does improved health in utero affect educational outcomes later in

life? Field et al. (2009, hereafter FRT) seek to shed light on this question

by estimating the effect on educational attainment of iodized oil capsule

(IOC) distribution programs launched in 1986 in Tanzania. The programs

are noteworthy for their combined size. The targeted districts contain a

quarter of the Tanzanian population, and a total of six million capsules

were distributed (Peterson et al., 1999). FRT use the lagged roll-out of

the programs for identification. They find large effects. The intent-to-

treat estimates indicate that being protected from iodine deficiency in

utero increased educational attainment by 0.35 years, on average. The

estimated effect is particularly large for girls, with an estimated intent-

to-treat effect of 0.59 years. Using the coverage rates of the programs to

derive the hypothetical case of achieving full coverage, FRT calculate that

“the expected increase in grade attainment for a child protected from fetal

iodine deficiency is a minimum of 0.73 years”.

FRT conclude that countries with similar iodine supplementation

programs have experienced a 4.8 percent increase in school participation

as an effect of the supplementation. The results provide evidence of a

causal link between the geographical health environment and economic

development. In particular, the study by FRT is one of the first to establish a

link between fetal health and outcomes later in life using quasi-experimental

methods; see Almond and Currie (2011) for a review of this literature. Since

the publication of the original article, the empirical strategy and data have

been used to address other aspects of child health and household behavior

(Adhvaryu and Nyshadham, 2016).

We revisit the Tanzanian experience by replicating the FRT study. We

attempt to exactly reproduce their results using the original data. The

exercise is successful, but it highlights a series of sample restrictions and

other specification choices that warrant further investigation. We examine

the robustness of the results with respect to these choices and we find

that the large estimates rely on the exact specification used in the original

study. The estimated effects are smaller and not statistically significant at

conventional levels with alternative specifications that are well motivated

with respect to both identification and statistical efficiency.

We investigate seven aspects of the specification of FRT. However, we

are primarily concerned about four choices for which we fail to find a

clear motivation. First, FRT use within-household birth order as a control

variable, but their code does not accurately sort the respondents by birth

date. The resulting birth order is essentially random. Second, they use the

education level of the spouse of the household head as a control variable.

The variable is coded as missing in households where the head does not

C

The editors of The Scandinavian Journal of Economics 2018.

584 Fetal iodine deficiency and schooling

have a spouse. As a result, all single-parent households are dropped from

their analysis. The dropped observations constitute approximately 21 percent

of the sample. Third, FRT exclude children aged 14 from the analysis.

This cohort was targeted by the IOC programs, and FRT include this

age group in their supplementary analyses. The inclusion of children aged

14 increases the sample size by 22 percent. Fourth, treatment is defined

as whether the respondent was protected from iodine deficiency during

pregnancy. The respondent’s birth date is therefore of great importance

when deriving the treatment variable. In their main analysis (using the

Tanzanian Household Budget Survey that started surveying in 2000), FRT

derive the birth year as 2000 minus the reported age. This approach ignores

the timing of the interviews (which is reported at the monthly level), and, in

particular, it disregards the fact that a quarter of the sample was surveyed

in 2001. The results of FRT are fairly robust to each of these issues

addressed in isolation. The combination of the issues is the concern. The

estimates in FRT are 3.5 higher than those produced by a specification

that only addresses the four issues listed above. With such a specification,

the estimated effects are close to zero and are not statistically significant,

despite lower estimated standard errors.

Prior studies have documented both cognitive and health effects of fetal

iodine deficiency, and it is plausible that some of these effects car ry on to

educational outcomes. A relevant critique of the null result in our narrow

replication is that it would be possible to detect positive effects with better

data. In particular, the actually assigned treatments are never observed in

the original study, and the independent variable of FRT is the probability of

treatment as approximated by a model. Based on evidence in the medical

literature, their model assumes that the first trimester of a pregnancy is

particularly sensitive to iodine deficiency, and they exploit the fact that

the IOC programs provide variation in protection from such deficiencies.

However, the programs are poorly documented; their start dates and lengths

are not known, and no program was able to reach all targeted people.

Furthermore, the existing survey data do not provide accurate information

of when and where the respondents were born. It is therefore not known

when the respondents were in utero. Finally, exact biological and medical

details about iodine intake and depletion are needed for an accurate model

of the protection from deficiency, but such details are not known. These

uncertainties introduce measurement error that could lead to imprecision

and attenuation bias.

We attempt to mitigate these concerns in a wide replication of the

original study. We keep the fixed-effects approach from FRT but use

new medical and institutional insights to improve the model used to

derive treatment probabilities. We also extend the data with four additional

datasets. The sample is almost four times as large as that in FRT. The

C

The editors of The Scandinavian Journal of Economics 2018.