The essay of ECo

shazuanzhe

8.WhenitRainsitPours.pdf

Home >Business & Finance homework help >Economics homework help >The essay of ECo

NBER WORKING PAPER SERIES

WHEN IT RAINS IT POURS: THE LONG-RUN ECONOMIC IMPACTS OF SALT IODIZATION IN THE UNITED STATES

Achyuta Adhvaryu Steven Bednar

Anant Nyshadham Teresa Molina Quynh Nguyen

Working Paper 24847 http://www.nber.org/papers/w24847

NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue

Cambridge, MA 02138 July 2018

Thanks to Martha Bailey, Prashant Bharadwaj, Mark Duggan, Jeanne Lafortune, Claudia Olivetti, Dimitra Politi, Paul Rhode, John Shea, Atheen Venkataramani, and David Weil for helpful conversations and seminar audiences at the NBER (CS; CH), Stanford SITE, Maryland, Michigan, Michigan State, Appalachian State, Indian School of Business, NEUDC, and SOLE for useful comments. Adhvaryu gratefully acknowledges funding from the NIH/NICHD (5K01HD071949). Molina gratefully acknowledges funding from the USC Provost’s Ph.D. Fellowship, the USC Dornsife INET graduate student fellowship, and the Oakley Endowed Fellowship. Thanks to David Carel for excellent research assistance. All errors are our own. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.

© 2018 by Achyuta Adhvaryu, Steven Bednar, Anant Nyshadham, Teresa Molina, and Quynh Nguyen. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

When It Rains It Pours: The Long-run Economic Impacts of Salt Iodization in the United States Achyuta Adhvaryu, Steven Bednar, Anant Nyshadham, Teresa Molina, and Quynh Nguyen NBER Working Paper No. 24847 July 2018 JEL No. I15,I18,J24,N32

ABSTRACT

In 1924, The Morton Salt Company began nationwide distribution of iodine-fortified salt. Ac- cess to iodine, a key determinant of cognitive ability, rose sharply. We compare outcomes for cohorts exposed in utero with those of slightly older, unexposed cohorts, across states with high versus low baseline iodine deficiency. Income increased by 11%; labor force participation rose 0.68 percentage points; and full-time work went up 0.9 percentage points due to increased iodine availability. These impacts were largely driven by changes in the economic outcomes of young women. In later adulthood, both men and women had higher family incomes due to iodization.

Achyuta Adhvaryu Ross School of Business University of Michigan 701 Tappan Street Ann Arbor, MI 48109 and NBER [email protected]

Steven Bednar Elon University Koury Business Center 124 2075 Campus Box Elon [email protected]

Anant Nyshadham Department of Economics Boston College Maloney Hall, 324 Chestnut Hill, MA 02467 and NBER [email protected]

Teresa Molina University of Hawaii at Manoa Saunders Hall 515A 2424 Maile Way Honolulu, HI 96822 [email protected]

Quynh Nguyen University of Maryland [email protected]

1 Introduction

Inadequate access to essential micronutrients such as iron, vitamin A, iodine, and zinc has stagger-

ing costs in terms of mortality, poor health, and lost productivity in low-income countries (Black

et al., 2013). The benefits of improving micronutrient availability in the short term, especially for

young children, are clear (Bhutta et al., 2013). But less is known about long-run impacts, particu-

larly of large-scale supplementation policies. How long do the effects of improved access to vital

micronutrients last? Do health effects spill over onto socioeconomic outcomes? Which individuals

are most affected by blanket campaigns? With notable recent exceptions (Clay et al., 2015; Feyrer

et al., 2017; Niemesh, 2015; Politi, 2010, 2014), these questions have received scarce attention,

and are the focus of the present study.

We draw lessons from the historical experience of the United States, where until the mid-1920’s,

natural access to iodine was limited in some areas of the country compared to others. An essential

micronutrient, iodine regulates thyroid hormone availability, which determines the density of fetal

neural networks (Lamberg, 1991). Physiological studies suggest that iodine deficiency negatively

affects cognitive function at all ages, but is particularly detrimental during gestation, when even

mild deficiency can greatly hamper cognitive development (Cao et al., 1994). Moreover, the effects

of fetal iodine deficiency disorder (IDD) are irreversible: an inadequate supply of iodine in the

first trimester of gestation permanently reduces intelligence quotient (IQ), regardless of subsequent

supplementation (Hetzel & Mano, 1989; Pharoah & Connolly, 1987; Zimmermann, 2009).1

We study the economic impacts of rapid, large-scale salt iodization in the twentieth century

US. The Morton Salt Company, the largest salt producer in the US, initiated nationwide iodized

salt distribution shortly after the invention of iodine-fortified salt in the mid-1920s. In less than

half a decade, the US went from zero to nearly universal availability of iodized salt (Markel, 1987).

Iodine deficiency rates plummeted in the following decade, most markedly in areas that were highly 1It bears mention that studies from the medical literature are either correlational or based on animal studies: causal

evidence on the effects of iodine exposure in humans is limited. The study by Feyrer et al. (2017), which we discuss below, is important in this sense, because it provides the most rigorous evidence to date of the impact of fetal iodine access on adult cognitive performance.

iodine deficient prior to the introduction of iodized salt (Brush & Atland, 1952; Hamwi et al., 1952;

Schiel & Wepfer, 1976).

Using a difference-in-differences strategy similar to the one used in this paper, Feyrer et al.

(2017) show that, among men who enlisted in the Army during World War II, exposure to iodized

salt increased the likelihood of being assigned to the Air Force (an indication of a high score on the

Army General Classification Test). They estimate that the introduction of iodized salt increased IQ

by approximately 15 points for those most deficient in iodine prior to this intervention. Building on

this importantwork, whichprovidesbothafirststageandmotivationforourstudy, wetackleseveral

important questions about the economic consequences of this natural experiment. What happened

to the labor market outcomes of those whose in utero access to iodine improved? Did increased IQ

affect incomes, as was true in Switzerland (Politi, 2014), and labor supply? Importantly, because

the sample in Feyrer et al. (2017) was exclusively male, we seek to estimate labor market effects

for women. Did men and women benefit differentially? How did these impacts evolve over the

life cycle? Did related outcomes like educational attainment and marriage respond differentially?

To answer these questions, we use a similar strategy to that of Feyrer et al. (2017), but look

to the U.S. census for our comprehensive set of economic outcomes. We compare outcomes for

cohorts born just before iodization (1920-1923) to those born during (1924-1927) and after (1928-

1931), across areas with varying pre-iodization deficiency rates. For the latter source of variation,

we use pre-iodization rates of goiter, the main physical manifestation of IDD (Love & Davenport,

1920; Olesen, 1929). We follow these cohorts through their productive lives, using data from the

1950-1980 censuses, during which these individuals were 19 to 60 years old. We present estimates

of early career outcomes, pooling the 1950 and 1960 censuses, and later career outcomes, using the

1970 and 1980 censuses.

Individuals affected by salt iodization saw improved economic outcomes. Labor force partici-

pation rose by about 0.68 percentage points and total income increased by 11% in the pooled sample

of individuals aged 19-60 in the 1950-1980 censuses. Women experienced the largest changes in

employment (greater than 1 percentage point) and income (15 percent), though men also saw small

increases in their income conditional on working (1 percent). The much larger labor supply effects

for women likely reflect the fact that women exhibited much lower employment rates than men

during this time period.2

These large increases in female labor supply were concentrated among women early in their

careers (under age 40): women affected by salt iodization were not significantly more likely to be

in the labor force in the 1970 and 1980 censuses. Impacts on female incomes do persist at later

ages, but are smaller and less precisely estimated. This pattern is consistent with impacted women

transitioning out of the labor force at later ages due to a negative income effect generated by their

higher accumulated lifetime income.3 Supporting this idea, we find that impacted women married

at later ages (nearly a quarter of a year), married more educated and higher-income men, and had

higher family income in their forties and fifties.

Like Politi (2010), which focuses on Switzerland’s historical experience with salt iodization, we

find a small but precisely estimated positive effect on educational attainment. Theoretical effects

of increased cognition on schooling are ambiguous, as the direct wage returns to ability in entry-

level labor opportunities may counteract any impacts of ability on schooling returns at early ages.

Because the magnitude of our estimated effect is very small (equivalent to about two weeks of

school), we interpret this result as evidence that both of these opposing effects are in play.4

Our study aims to contribute to three literatures. First, we add to the literature on the long-term

effects of early life conditions.5 Much of this “fetal origins” work has focused on demonstrating the

impacts of traumatic experiences (disease, natural disasters, environmental factors, etc.) in early

life. Fewer studies have estimated the gains to exposure to the purposeful large-scale distribution of

resources. The distinction between the two types of studies is important because the latter “shock”

can yield actionable information: policies with demonstrated positive impacts can be advocated 2This result is also consistent with medical evidence that female fetuses are more sensitive to maternal thyroid

deficiency than male fetuses (Field et al., 2009; Friedhoff et al., 2000), though we argue in section 5.1.1 that this biological explanation is likely secondary.

3This pattern has been documented before, specifically in the “career then family” cohorts discussed in Goldin et al. (1997); Goldin & Katz (2002)

4This is consistent with Bleakley (2010), who finds mixed results on the effect of malaria eradication on schooling in four different countries.

5See Heckman (2006), Almond & Currie (2011), and Currie & Vogl (2013) for useful syntheses.

for and reproduced. A growing set of studies–including Hoynes et al. (2016), Bleakley (2007),

Bleakley (2010), Field et al. (2009), Almond et al. (2010), Politi (2010), Bhalotra & Venkataramani

(2015), and Feyrer et al. (2017)–have recently made strides in this direction. We build on this

evidence base: the results of these studies and ours offer lessons from historical policy experiments

from which present-day policymakers, particularly in developing countries, might profitably draw.

Second, we contribute to the understanding of women’s decisions regarding the labor force in

the historical United States. We find that labor market effects of iodine are particularly pronounced

for women, consistent with evidence from recent studies on the effects of other early life interven-

tions (Bleakley, 2007; Field et al., 2009; Hoynes et al., 2016; Maccini & Yang, 2009). Moreover,

this pattern relates to important previous work on the drivers of the marked rise in labor force par-

ticipation of women over the 20th century. Goldin (1991) and Goldin & Olivetti (2013) estimate

that WWII led to a roughly 20 percent rise in female participation for higher-educated women in

cohorts born between 1915 and 1924. Bailey (2006) and Goldin & Katz (2002) show that increased

access to oral contraceptives led to later marriage, higher likelihood of professional and graduate

training in high skill occupations, and increased rate and duration of labor force participation among

cohorts born after 1940. We present complementary evidence documenting that salt iodization also

contributed to the rise in female labor force participation, generating a 3 percent increase among

females born in between the two sets of cohorts studied in these papers.

Finally, we add evidence on the long-run effects of micronutrient fortification campaigns and in

particular mass salt iodization as a means of eradicating iodine deficiency. Nearly 2 billion people

worldwide–a third of the world’s population–do not have adequate access to iodine (De Benoist

et al., 2004). Recent estimates from the economics literature suggest that the incidence of iodine

deficiency, and thus the returns to reducing IDD, may be very large (Feyrer et al., 2017; Field et al.,

2009; Politi, 2010). Policymakers in IDD-endemic countries, as well as the WHO, UNICEF, and

other international organizations, have made increasing access to iodine a high priority. Mass salt

iodization to prevent IDD is, far and away, the preferred policy: iodizing salt is much cheaper

than continuous supplementation in populations with iodine-deficient diets, and, taken with other

micronutrients such as iron, is highly cost-effective in terms of fetal and infant deaths averted

(Horton et al., 2008).

The rest of the paper is organized as follows. Section 2 discusses iodine deficiency and its

prevalence in the early twentieth century, as well as the history of salt iodization in the US. Section

3 discusses our data sources, section 4 our empirical strategy, and section 5 the results. Section 6

concludes with a discussion of the size of the economic benefits of iodization.

2 Background

2.1 Iodine Deficiency and its Consequences

Iodine is crucial to the functioning of every body cell. The thyroid gland in the lower part of the

neck uses iodine from foods to produce thyroid hormones, which are released into the blood stream

to control metabolism (the conversion of oxygen and calories to energy). Optimal iodine intake as

recommended by the WHO is very small: a daily dose of 90 µg for children of 0-59 months, 120

µg for ages 6 to 12, 150 µg for older ages (which is the amount found in half a teaspoon of iodized

salt), and 200 µg for pregnant and lactating women (Clar et al., 2002). Foods with high iodine

content include some milks, leafy vegetables, and seafood, but individuals in areas without natural

access to iodine – far from the ocean or in mountain regions susceptible to erosion (Hetzel, 1989)

– are at risk of not meeting these recommended levels of iodine intake.

At any point from the fetal stage to adulthood, insufficient iodine intake can cause a number of

functional and developmental abnormalities, often referred to as iodine deficiency disorders (IDD).

The main physical manifestation of IDD is goiter (the enlargement of the thyroid gland),6 but other

IDDs include hypothyroidism (which results in fatigue, lethargy, slow speech, and thought), im-

paired mental function, retarded physical development, and increased susceptibility of the thyroid

gland to nuclear radiation (De Benoist et al., 2004). This paper focuses on the availability of io- 6Goiter may not be visible if iodine deficiency is minimal. On the other hand, iodine deficiency is the primary, but

not exclusive, cause of goiter. Goiter, when sufficiently large, may cause complications such as respiratory difficulty.

Figure 1: State-Level Goiter Rates from World War I Draft Examinations

Notes: State-level goiter rates, calculated from Love & Davenport (1920), are summarized in Table A1. Each shade represents a different quintile of the goiter distribution, with the darkest gray representing the highest quintile. Thick lines denote census division boundaries.

dine in utero because fetal iodine levels are particularly crucial for brain development: insufficient

iodine intake during gestation can cause irreversible cognitive damage (De Escobar et al., 2004;

Zimmermann, 2009).

2.2 The Geography of Iodine Deficiency in the US

The map in Figure 1 illustrates the geographic distribution of goiter incidence across the U.S.,

based on data from the 1917 WWI draft examinations. To our knowledge, this is the first nation-

wide goiter survey in the US. Evidence from other sources suggest that the draft statistics offer a

good representation of the geographic pattern of iodine availability for the general population. For

example, Figure B1 in the online appendix highlights areas with low iodine content in drinking

water and shows a similar “goiter belt” in the northern states.7 Similarly, Olesen (1929) concludes 7We do not use the water iodine content data because the author presents numerical information for only 27 states,

many of which use data from only one location.

that independent thyroid survey data from elementary school to college aged samples across the

nation were consistent with the geographic variation in in goiter incidence from the draft statistics,

in areas where he was able to collect data.

As can be seen in Figure 1 (as well as the corresponding Table A1), the draft statistics show

considerable variation in goiter prevalence across states even within areas defined by the nine Cen-

sus divisions. This variation is crucial as it allows us to control for division-level time effects to

remove any systematic coincidence between the goiter distribution and geographic differences in

economic development over time, such as the North-South divide.

2.3 Introduction of Iodized Salt in the US in 1924

The hypothesis that iodine can help prevent goiter has existed since the mid-1800’s (Zimmermann,

2008). It was not until 1895, however, that iodine was first discovered in the thyroid gland (Bau-

mann, 1896). After this, experiments were conducted to study the impact of changes in iodine

content on the incidence of goiter in different kinds of animals (Marine & Feiss, 1915; Marine &

Lenhart, 1909; Smith, 1917). In the United States, Hall (1914) and Olesen (1915) were the first

to record goiter rates in humans in an organized manner. An experiment started in 1917 by scien-

tist David Marine and colleagues, which involved providing iodated syrup to school girls in Ohio,

offered the first direct evidence that iodine supplementation could control and prevent goiter in hu-

mans (Marine & Kimball, 1917, 1920). This evidence inspired Switzerland to set up prophylactic

programs, one of which used salt as an iodization vehicle.

Around the same time as the Ohio experiment, a few other factors placed focus on goiter in hu-

mans as a health problem in America (Annegers & Mickelsen, 1973). These included (a) decline of

other childhood diseases, which allowed more attention to shift to goiter, (b) McClendon’s discov-

ery of the relationship between goiter and the iodine content of drinking water, (c) the WWI draft

examinations, which revealed the nationwide extent of goiter prevalence, and (d) a large incidence

of goiter in Michigan (Levin, 1919).

According to Levin (1919), roughly 24 percent of army recruits from Houghton County, Michi-

gan had simple goiter and 2.4 percent of the men were disqualified for the army based on the size

or cystic type of goiter.8 Dr. David Cowie, of the University of Michigan, became interested

in eliminating widespread simple goiter in his home state. Cowie organized a symposium of the

Michigan State Medical Society in June 1922, which created the Iodized Salt Committee to explore

the feasibility of adding iodine to salt in Michigan. The committee initially entertained the idea of

proposing a law prohibiting non-iodine fortified salt (Markel, 1987), but they were swayed to allow

the profit motives of salt manufacturers to lead them to distribute a product endorsed by medical

experts. Introducing iodine to salt was relatively costless as it could be added at the same time as

magnesium, which is included to regulate salt flow from the container (Markel, 1987).

Cowie worked with several small salt manufacturers based in Michigan, who were eager to

supply a product that they perceived would have a large demand with little extra cost to produce.

The Executive Council of the Michigan State Medical Society officially endorsed salt on March

12, 1924 and iodized salt appeared in Michigan groceries on May 1, 1924. The product proved so

popular that Morton Salt Company, who initially resisted because of concerns about the high cost

of separating a different version of salt for Michigan retailers, began nationwide distribution a few

months later. With educational efforts of the Michigan State Medical Society, which gave lectures

on the medical benefits of iodized salt and zealous advertisements by salt producers,9 iodized salt

rapidly grew popular. By 1930, iodized salt sales were eight times plain salt sales (Markel, 1987).

Many later surveys found marked decreases in thyroid enlargement, especially among contin-

uous users of iodized salts. Interestingly, Schiel & Wepfer (1976) found that, among Michigan

school children in 1924-51, there was a decline in goiter rates among non-users. Cowie attributed

this to ingestion of iodized salt without realizing it, such as in school canteens and restaurants,

which seems plausible given that iodized salt made up 90% of salt sales in Michigan at the time 8“If these men were considered unacceptable to the US Army because of decreased mentation, inability to complete

average mental or physical labor, dyspnea, cardiac, and metabolic abnormalities seen in patients with severe colloid goiter, how useful were they considered in daily functions and work productivity in civilian life?” (Markel, 1987, pg 221)

9Advertisements would often contain phrases such as “Remember, too, that Morton’s Iodized Salt protects children against simple goiter - that often unnoticed nutritional disease which is frequently accompanied by underdevelopment, irritability and backwardness at school.” (Women’s Home Companion, June 1934)

(Markel, 1987). This observation alleviates concerns about self-selection into using iodized salt

and supports the approximate universality of the intervention. The shaded areas in Figure B2 (in

the online appendix) map the areas with endemic goiter as determined by goiter surveys compiled

by the Chilean Iodine Educational Bureau in 1950 and the American Geographical Society in 1953.

While it is unclear what definition of endemic goiter was used for these maps, prevalence decreased

considerably between the WWI draft era (Figure B1) and 1950, and the diminishing trend continued

to 1953.

2.4 Goiter and Confounding Factors

Marine’s 1917-19 experiment was the first to inform the US public that iodine supplementation

could prevent and treat goiter; hence there is little reason to suspect a direct role of iodine in res-

idential selection or selection into iodine-rich diets. Supporting this claim, Figure B1 shows that

goiter incidence was concentrated in the northern states, which were socioeconomically better off

compared to the southern states.

However, one might worry that there may have been other important changes in the US diet,

concurrent with the roll out of iodized salt. In fact, food fortification in the US began with salt

iodization in 1924, with discoveries of the role of vitamin and mineral deficiencies in many diseases

and sicknesses (Backstrand, 2002). However, the knowledge remained mostly in the laboratory

until May 1941, when President Roosevelt called a National Nutrition Conference for Defense.

Figure B3 in the online appendix shows the change in the per capita riboflavin, iron, niacin, and

thiamin contents of American food between 1909 and 1994, which does not coincide with the

timing of salt iodization.

Similarly, one might still suspect that high goiter areas prior to iodization were also more likely

to have high incidence of other nutrient deficiencies or health issues such as malaria or hookworm.

In order to address concerns about the contemporaneous eradication of these infectious diseases,

we check the robustness of our main results to controlling for baseline geographic variation in

the prevalence of malaria and hookworm, interacted with post-iodization dummies. The robust-

ness of our results alleviates concerns about the geographic pattern of goiter incidence, and hence

the pattern of expected benefits from iodized salt, coinciding with the geographic pattern of other

health-related improvements.

3 Data

3.1 Goiter Data

Our information on the geographic distribution of goiter before 1924 comes from the data used

to create Figure 1: medical examinations of over two and a half million drafted men aged 18

to 30 before World War I. Conducted on a large sample of men from all over the United States

within a short period of time between 1917 and 1918, these examinations offer a snapshot of the

geographic distribution of various mental and physical defects prior to the iodization of salt.10 Love

& Davenport (1920) documents prevalence rates in this sample for over 200 medical conditions,

including goiter. In this paper, we use state-level prevalence rates, although rates for smaller regions

(collections of counties known as sections) were recorded as well.11 Table A1 reports for each state

the goiter rate recorded in Love & Davenport (1920). The median state-level goiter rate is 0.214%

and the maximum is 2.69%.

3.2 Census Data

We also use data from the United States Decennial Census (Ruggles et al., 2015), restricting to

individuals born in the twelve-year period spanning 1920 to 1931, which includes the years before,

during, and after the nationwide spread of iodized salt. We are interested in labor, education, and 10Because this paper focuses on in utero exposure to iodine, the ideal dataset would consist of goiter rates from a

representative sample of women of childbearing age, instead of men. In the data collected by Olesen (1929), there is a high correlation (0.87) between goiter rates among school-aged girls and boys, which suggests that the distribution of female goiter rates across states should be similar to what is captured by the goiter rates in Love & Davenport (1920), calculated from men of the relevant age range. We do not use the Olesen (1929) goiter rates as our measure of iodine availability because the samples are not representative of each state and numerical rates are only available for 37 states.

11Unfortunately, we cannot use the section-level goiter data because we only have state of birth, not county of birth, for the individuals in our sample.

marriage outcomes for this cohort in the 1950 to 1980 censuses, during which they were working-

aged adults aged 19 to 60. We use the 1% samples for 1950 to 1970 and the 5% sample for 1980.12

Individuals are assigned to the goiter rate in their state of birth, which proxies for their risk

of being born to an iodine deficient mother. Individuals are also grouped according to their birth

year. Those born in the years 1920 to 1923 are marked as “pre-iodization,” those born in 1924 to

1927 are classified as born “during iodization,” and those born in 1928 to 1931 are considered “post-

iodization.” Iodized salt first appeared in grocery stores in 1924 and was reported to have generated

eight times more sales than regular salt by 1930 (Markel, 1987). In creating the “during” category,

we allow four years of leeway following the initial introduction of iodized salt to ensure that the

after cohort was exposed to an environment with sufficiently widespread iodized salt availability.

3.2.1 Outcome and Control Variables

We are primarily interested in labor market outcomes. To represent employment, we use two vari-

ables: a dummy for labor force participation (which includes job-seekers as participants) and a

dummy for employment (which is set to zero for job-seekers). Individuals who report working in

the last year also report the number of weeks they worked (in intervals). We create an indicator

variable to represent individuals who worked at least 40 weeks (conditional on having worked in

the past year) to study the intensive margin of labor supply. We also look at total income, which

includes income from all sources, including wages and self-employment income. We transform

total income using the inverse hyperbolic sine function.13 For all of these labor market variables,

we pool individuals from the 1950 to 1980 censuses, during which our sample was aged 19 to 60.

In addition to these labor market outcomes, we also study years of educational attainment,

whether an individual has ever been married, and age at first marriage. Because these are all stock

variables that are unlikely to be determined in the teenage years or twenties, we only look at indi-

viduals in the 1970 and 1980 censuses (when our sample is aged between 39 and 60). To further 12Summary statistics by census year are reported in Table B1 in the online appendix. 13sinh−1( ̂Income) = ln(Income+(Income2 +1)1/2). The income variable is therefore not conditional on working,

and includes zeros for those who do not work.

investigate marriage quality, we also study spousal income and education for individuals currently

married (to a spouse currently living in the same the household), along with total family income.

Additional variables taken from the census include gender and race. In addition to including

female and black indicator variables as controls, we also control for pre-iodization demographic

conditions in the individual’s state of birth. This is done by calculating the black and female pro-

portions from the 1920 census in the individual’s state of birth. For more details on the construction

of variables, see the Data Appendix (in the online supplement).

3.2.2 Summary Statistics

Table 1 reports summary statistics for our sample population (individuals born between 1920 and

1931). The first column summarizes our outcome variables for the entire sample, the second re-

stricts to females, and the third focuses on males. Women have much lower labor force partic-

ipation, labor supply, and income than men. Marriage rates, age at first marriage, and years of

schooling are more similar across both genders, although women are more likely to be married and

get married slightly earlier.

3.3 Geographic Controls

Because of the clear regional patterns in the distribution of goiter, an important part of our strategy

involves allowing for differential birth cohort trends by region. We use the nine Census Bureau

divisions, listed in Table A1, to categorize states into regions. Another way we control for regional

patterns is by using the average latitude of each individual’s state of birth.14

3.4 Robustness Check Controls

In our robustness checks, we control for the pre-iodization rates of two other diseases: malaria

and hookworm. For malaria, we use the malaria mortality rates from the 1890 Census, used in 14These numbers were obtained from the online database, MaxMind: http://dev.maxmind.com/geoip/

legacy/codes/state_latlon/

Table 1: Summary Statistics

(1) (2) (3) Whole Sample

Females Males

1950-1980 Censuses

1(Employed) 0.644 0.429 0.872

(0.479) (0.495) (0.334)

1(Participated in Labor Force) 0.670 0.448 0.905 (0.470) (0.497) (0.293)

1(Worked at least 40 weeks) 0.790 0.659 0.867 conditional on working (0.408) (0.474) (0.339)

Total Income 19551.4 8690.0 31193.1 (20266.0) (12478.3) (20544.1)

Number of Observations 2383143 1236420 1146723

1970-1980 Censuses Years of Schooling 11.38 11.31 11.45

(3.137) (2.841) (3.426)

1(Ever Married) 0.945 0.951 0.937 (0.229) (0.216) (0.242)

Age at First Marriage 22.74 21.43 24.16 (5.345) (5.013) (5.331)

Spouse's Schooling 11.45 11.31 11.59 (3.007) (3.391) (2.568)

Spouse's Total Income 25498.3 41443.6 9934.9 (23175.8) (20015.6) (13414.9)

Total Family Income 48666.0 46772.0 50716.1 (20703.5) (21436.8) (19674.9)

Number of Observations 1789907 930333 859574

Notes: Sample includes all individuals born 1920 to 1931. Statistics are calculated using person-level weights provided by the census. Total income in 1999 dollars.

Bleakley (2010). Hookworm rates are also taken from Bleakley (2010) and are the same data

used by Bleakley (2007). These rates were drawn from around 20,000 recruits in 1917 and 1918, a

smaller and separate sample from the recruits tested in Love & Davenport (1920) (Kofoid & Tucker,

1921). Each individual was assigned to the relevant prevalence rates from their state of birth.

In order to allow for differential trends across states that may have been affected differently

by the Great Depression, we use state-level unemployment rates calculated from the 1930 census.

Similarly, in order to allow for differential trends across states affected differently by the mid-

century “Great Migration” of African Americans out of the rural South, we calculate the state-level

change, between 1920 and 1940, in the black population share, as well as in the total share of the

U.S. population made up by each state, to capture race composition shifts and overall population

growth.

We also include years of compulsory schooling as an additional control. We obtained this

variable from state-level data used in Lleras-Muney (2002), compiled from multiple sources. This

data reports the number of years of required schooling (either by compulsory attendance laws or

implied by child labor laws) in each state in each year from 1915 to 1939. Using the same strategy

as Lleras-Muney (2002), we match each individual to the law in place in their state of birth in the

year they turned 14 (the minimum leaving age across all states and years), as this is arguably the

most relevant to their schooling continuation choices.

Finally, we also use WWII state mobilization rates, obtained from Acemoglu et al. (2004). If

WWII mobilization had any impact on these individuals, it should have affected them during their

young adult or adult life: we therefore use mobilization rates in the individual’s state of residence

rather than their state of birth.

4 Empirical Strategy

4.1 Overview of strategy

As described in section 2, once Morton Salt Co.’s decision to iodize its supply was made, the spread

of iodized salt was wide scale and fairly rapid. Since iodization happened nationwide, incidentally

there was no true exclusion from exposure. In the spirit of Bleakley (2010), Hornbeck (2012),

and others, our basic strategy is to compare trends in economic outcomes among individuals born

in states with different levels of pre-iodization iodine deficiency rates. Feyrer et al. (2017) uses

similar strategy to identify the impacts of iodization on recruits’ placement into the Army v. the

Air Force.

We use the spatial distribution of goiter in 1924 in the continental US to identify differences in

pre-iodization deficiency rates. As described in section 3, we use data from the Love & Davenport

(1920) survey of military recruits. We link each individual in the census to a goiter rate using their

state of birth. We use state of birth to draw focus to the effects of in utero exposure to iodine rather

than exposure through one’s life.

We interpret the goiter value as a proxy for the extent of iodine deficiency in one’s state of birth

during early life. This proxy will, of course, not fully reflect actual iodine exposure. Nevertheless,

as shown in the previous sections, as well as in Feyrer et al. (2017), the spatial distribution of goiter

generally mirrors the distribution of iodine content in water sources. While admittedly an imperfect

proxy, the distribution allows a rough ordering of individuals according to their exposure to iodine

in utero.

In our main results, we consider the outcomes of three cohorts: those born before (1920-1923),

during (1924-1927), and after (1928-1931) salt iodization. We restrict to a fairly small window

of birth cohorts to ensure that we are comparing cohorts of relatively similar ages in each census

wave, and to avoid other important historical events around the same time (for example, the Spanish

influenza outbreak in 1919). In our main specification, we do not go back further than 1920 but in

alternative specifications, we explore longer “before” periods. We consider the middle (“during”)

group because, while the proliferation of iodized salt across the US was rapid, we do not have

data on the geographic pattern of this nationwide spread. During the proliferation period, it would

be possible to find muted effects simply because iodized salt had not yet reached some markets.

To allow for this, we separate the “during” and “after” iodization periods. We also show that our

results are robust to the use of a more flexible specification (an event-study analysis) that does not

rely on this somewhat arbitrary assignment of cohort dummies (see Figure 2).

We interpret differences in trends in economic outcomes coincident with the proliferation of

iodized salt across individuals born in states with varying pre-iodization levels of goiter as causally

related to salt iodization. Because all three cohorts were eventually exposed to iodized salt by late

childhood and for the remainder of their lives, we are identifying the impact of differential exposure

to iodine specifically in utero, which is our primary interest because of the irreversible nature of

the cognitive damage that can be caused by lack of iodine during the fetal period. Our estimates are

therefore somewhat conservative because they do not consider the potential benefits of increased

iodine availability later in life, which all of our cohorts (including our control cohort) may have

experienced.

4.2 Specification

The basic difference in differences strategy, then, is to compare the outcomes of cohorts born before

to those born during and after iodization, across individuals born in states of varying levels of iodine

deficiency. We estimate the following specification, for individual i born in year t in state s (census

division d), for outcome y recorded in census year c, where Gs is the continuous goiter rate, Dt

is a dummy for belonging to the “during” cohort, and At is a dummy for belonging to the “after”

cohort:

yistc = β1GsAt + β2GsDt + µs + ζdt + λct + ηXistc + εistc. (1)

Here, β1 and β2 are the main coefficients of interest, measuring the difference in birth cohort

trends in outcome y across individuals living in states with different levels of iodine deficiency.

The specification includes state of birth fixed effects (µs) and year of birth fixed effects (which

are interacted with census waves, in λct, as well as census divisions, in ζdt) that absorb the main

effects of Gs, Dt, and At. The census division of birth by birth year interactions (ζdt) are crucial

because they control for any regional trends over time that may coincide with the national goiter

distribution.15 By including division-by-birth-year fixed effects, we ensure that we are comparing

outcome variable trends (by birth cohort) across high and low goiter states of birth in their deviations

from each Census division’s average non-linear trend.16 Census wave by birth year interactions

(λct) are included to account for differential cohort trends in the outcome variables as the cohorts

age (from one census wave to the next). Included in Xist are individual controls for race and gender,

as well as controls for the proportion of the population that is female and that is black (measured in

1920) in the individual’s state of birth, interacted with the during and after dummies. Finally, we

also include average latitude (of the state of birth) interacted with during and after dummies in order

to alleviate concerns about differential trends for Northern and Southern states confounding our

estimates. Standard errors are clustered at the state of birth level to allow for arbitrary correlation

of the errors for individuals born in the same state.

We conduct this analysis on all individuals born between 1920 and 1931,17 using the 1950 to 15Evidence of differential regional trends potentially correlated with goiter rates give us reason to be believe the

inclusion of these interactions is crucial. For example, we compare the division with the highest average goiter rate (Pacific) to the division with the lowest average goiter rate (East South Central). In the East South Central division, female labor force participation and income were growing faster than in the Pacific division in the years leading up to the iodization of salt (1914 to 1923). In addition, the female probability of working at least 40 hours a week was decreasing for the East South Central division but increasing for the Pacific division over this same time period. It is difficult to translate these trends into predictions about how the failure to account for division trends should change our coefficient estimates (because it depends on whether we expect these trends to diverge or converge in the absence of iodization). What is clear, however, is that divisions with different average levels of goiter were trending differently prior to salt iodization and it is important that we control for these division-by-cohort interactions in order to avoid picking up division-specific trends in our goiter coefficient of interest.

16There are many types of differential cohort trends that we could in theory control for, but given the geographic distribution of goiter that we observe, our major concern is in broad regional trends that may be non-linear (rather than linear trends at the state-level, for example). Once we have controlled for these division-by-birth-year interactions, we argue that controlling additionally for state-specific linear trends is less important, given that our analysis is now within division (and divisions are relatively small).

17As we explain in section 4.1, we use a relatively short window of birth years to ensure that the cohorts we are comparing are relatively similar in age and to avoid picking up the effects of other important historical events, but we show robustness to extending this period back to 1914.

1980 censuses. We then look at men and women separately. In order to trace out the effects of salt

iodization on labor market outcomes as our cohorts age, we also run these by-gender regressions

separately for the 1950-1960 censuses (when our sample was aged 19 to 40) and the 1970-1980

censuses (when they were aged 39 to 60). We test the robustness of our results to the inclusion

of controls for contemporaneous disease eradication programs (related to hookworm and malaria),

unemployment rates in 1930, demographic changes from 1920 to 1940, compulsory schooling laws,

and WWII mobilization rates in an individual’s state of residence.

In order to rule out the existence of differential pre-trends across high and low goiter states

(which would indicate a potential violation of our difference-in-difference assumptions), we expand

our sample to include individuals born in 1916-1919 and run the following regression:

yist = β1GsAt + β2GsDt + β3GsPt + µs + ζdt + λct + ηXist + εist. (2)

Here, Pt represents an indicator variable for those born in the pre-1920 “pre-trend” period. If

there were no differential cohort trends across the goiter distribution prior to the introduction of

iodized salt, we should be unable to reject the null that β3 is equal to zero.

5 Results

5.1 Labor Supply and Income

In all of the regressions discussed in this section, our coefficients of interest are the after-by-goiter

rate interaction and the during-by-goiter rate interaction: these represent the effect of salt iodization

on our outcomes of interest. Although the following tables only report these two coefficients, all

specifications also include state of birth fixed effects, year of birth by census wave interactions,

census division of birth by birth year interactions, a female dummy, a black dummy, and after and

during dummies interacted with average state latitude and 1920 state-level female and black pro-

portions. We multiply each relevant coefficient by the inter-quartile range of the goiter distribution

(0.709) to obtain a value that can be interpreted as the effect of moving from a relatively low goi-

ter state (at the 25th percentile) to a high goiter state (at the 75th percentile) when discussing the

results.

Table 2: Effects of Salt Iodization on Labor and Income Outcomes

(1) (2) (3) (4)

After x Goiter Rate 0.00707*** 0.00680** 0.00877** 0.105***

(0.00248) (0.00284) (0.00371) (0.0290)

During x Goiter Rate 0.00355 0.00323 0.00848*** 0.0267

(0.00231) (0.00233) (0.00268) (0.0259)

Observations 2383143 2383143 1537003 2135396

Mean of Dep. Var. 0.644 0.670 0.790 7.902

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1920-1931. All regressions include state of birth fixed effects, year of birth x census year

dummies, census division of birth x birth year dummies, gender, race, and During and After dummies interacted

with average state latitude and 1920 state-level female and black proportions. 1(Worked at least 40 weeks) is conditional on having worked in the past year. sinh-1(Income) takes the inverse hyperbolic sine of total incomeweeks, including zeros for those not working.

1(Participated in

the Labor Force) 1(Employed)

1(Worked at least

40 weeks) sinh-1(Income)

Table 2 reports the full-sample regression results for our labor outcomes of interest. The effects

of salt iodization on the probability of being employed (column 1) and labor force participation

(column 2) are both positive and significant, with effect sizes around 0.7 percentage points for the

after-by-goiter interactions. The during-by-goiter interactions in these regressions are also positive,

but smaller and statistically insignificant. These smaller during coefficients might be an indication

that it took time for the take-up of iodized salt to spread nationwide, but we discuss evidence later

(in Figure 2 and Table 5) that the effects of salt iodization do show up relatively quickly – just not

immediately.

In Table 2, we also find that salt iodization increased the likelihood of working at least 40 weeks

in the year, conditional on having worked in the last year, for both the during and after cohorts. In

addition, we find an 11% increase in total income (for the after cohort).

Table 3: Pre-Trends in Labor and Income Outcomes

(1) (2) (3) (4)

After x Goiter Rate 0.00707*** 0.00680** 0.00880** 0.107*** (0.00248) (0.00284) (0.00370) (0.0288)

During x Goiter Rate 0.00356 0.00323 0.00854*** 0.0274

(0.00231) (0.00233) (0.00269) (0.0259)

Pre-1920 x Goiter Rate -0.00317 -0.00264 -0.00422 0.0107

(0.00210) (0.00183) (0.00370) (0.0253)

Observations 3114884 3114884 1940335 2784412

Mean of Dep. Var. 0.635 0.660 0.794 7.930

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). Pre-1920 is a dummy equal to 1 for those born before 1920.

After is a dummy equal to 1 for those born 1928-1931. During is a dummy equal to 1 for those born 1924-1927. These

regressions use the 1950 to 1980 censuses, restricting to individuals born in 1916-1931. All regressions include state

of birth fixed effects, year of birth x census year dummies, census division of birth x birth year dummies, gender,

race, and Pre-1920, During, and After dummies interacted with average state latitude and 1920 state-level female

and black proportions. 1(Worked at least 40 weeks) is conditional on having worked in the past year. sinh-1(Income) takes the inverse hyperbolic sine of total income, including zeros for those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

Our interpretation of these coefficients relies on attributing the change in trends after 1924 to

the introduction of iodized salt. If, however, high and low goiter states were trending differently

before 1924, this would suggest that the difference in trends after 1924 may not be due to salt iodiza-

tion. In order to test for the existence of differential pre-trends, we run the regression specified in

equation 2, including cohorts born in an even earlier period: 1916 to 1919. Results are reported

in Table 3, where the pre-1920-by-goiter coefficient estimates the difference across the goiter dis-

tribution in cohort trends prior to the introduction of iodized salt. Across all specifications, these

pre-1920 coefficients are not significantly different from zero. This alleviates concerns that states

were experiencing different cohort trends – before 1924 – systematically correlated with the goiter

distribution. Moreover, the during-by-goiter and after-by-goiter coefficient estimates are almost

identical to those in the previous table.

In Figure 2, we present further graphical evidence using an event study analysis. Here, we

employ a more flexible specification, modifying equation (2) by replacing the before, during, and

after interactions with birth year dummies interacted with goiter rate. We extend our study period

to include the 1914 birth cohort. We let 1923 serve as the omitted category because it is the last

cohort with no exposure to iodized salt during the in utero period.18 Figure 2 plots the coefficients

and 90% confidence intervals for the birth-year by goiter interactions, for each of our labor market

outcomes of interest.

These results are consistent with our previous findings. Across all outcomes, we see the co-

efficients shift upward starting in either 1924 or 1925, and coefficients remain higher than zero

throughout the post-iodization period (with only two exceptions – 1927 in the first two panels).

Though these coefficients are not precisely estimated, the patterns display an upward shift by 1925

(that continues to increase in Panels A, B, and D), which is consistent with the results of Feyrer

et al. (2017).

Importantly, prior to 1924, the trend in coefficients is fairly flat across all outcomes, though

there is a considerable amount of variation for the worked 40 weeks variable prior to 1924, and

there is a slight downward trend in the income variable prior to 1924. We confirm (see Figure B4

in the online appendix) that the positive 1914 coefficient in the income regression appears to be

a random fluctuation and that the pre-1924 trend flattens out when we extend the period back to

1912. These results are indicative of a rapid, though not instantaneous, take-up of iodized salt by

the U.S. population, and validate our definitions of the During and After cohorts. Though we lack

statistical precision in our estimates of this rigorous specification, these results are consistent with

our baseline specification, which we use for the remainder of the paper.

Figure 2: Year-by-Year Effects of Salt Iodization on Labor and Income Outcomes

Notes: Each point represents the coefficient estimate (and 90% confidence interval) for Goiter Rate interacted with the birth year indicator listed on the x-axis. These regressions use the 1950 to 1980 censuses, restricting to individuals born in 1914-1931. All regressions include state of birth fixed effects, year of birth x census year dummies, census division of birth x birth year dummies, gender, race, and During, After, and Pre-1920 dummies interacted with average state latitude and 1920 state-level female and black proportions.

Table 4: Effects of Salt Iodization on Labor and Income Outcomes, By Gender

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0108*** 0.0121*** 0.0144*** 0.149***

(0.00360) (0.00395) (0.00525) (0.0505)

During x Goiter Rate 0.00519 0.00579 0.0206*** 0.0192

(0.00372) (0.00405) (0.00466) (0.0398)

Observations 1236420 1236420 606704 1108650

Mean of Dep. Var. 0.429 0.448 0.659 5.586

Panel B: Males

After x Goiter Rate 0.00197 0.000148 0.00562 0.0288

(0.00392) (0.00364) (0.00443) (0.0173)

During x Goiter Rate 0.000934 -0.000487 0.000888 0.0116

(0.00237) (0.00247) (0.00384) (0.0240)

Observations 1146723 1146723 930299 1026746

Mean of Dep. Var. 0.872 0.905 0.867 10.38

Panel C: Female-Male Difference

After x Goiter Rate 0.00884 0.0119** 0.00874 0.120**

(0.00576) (0.00521) (0.00592) (0.0504)

During x Goiter Rate 0.00425 0.00628 0.0197*** 0.00759

(0.00464) (0.00516) (0.00658) (0.0424)

1(Worked at least

40 weeks)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1920-1931. All regressions include state of birth fixed effects, year of birth x census year

dummies, census division of birth x birth year dummies, gender, race, and During and After dummies interacted

1(Employed) 1(Participated in

the Labor Force) sinh-1(Income)

5.1.1 Gender Heterogeneity

We next ask whether this cognitive shock impacted labor market outcomes differently for men and

women. Table 4 reports the results of two separate regressions: one for women (Panel A) and one

for men (Panel B), along with the difference in our main coefficients across the two specifications

(Panel C). Stark gender differences are apparent. All of the positive effects on labor supply and

income, reported in the previous tables, are driven by women. In fact, there are no significant coef-

ficients in the male regressions, and for labor force participation and income, the after interaction

coefficients are significantly larger for women than for men. Comparing the dependent variable

means for men and women, it is clear that women have much lower labor supply than men during

this period. This implies a much larger scope for growth in female employment than male em-

ployment, which could explain this drastic heterogeneity. These results are also consistent with the

hypothesis that female fetuses are more sensitive to maternal thyroid deficiency than male fetuses

(Field et al., 2009; Friedhoff et al., 2000), but – as we discuss in the next paragraph – we suspect

this biological explanation is a secondary one.

In Appendix Table A2, we repeat our analysis (separately for each gender) for three variants

of our total income variable: income levels in dollars (including non-earners with zero income),

income levels conditional on working, and log income (conditional on working). Interestingly, we

do find positive effects on male income that are small in magnitude but significantly different from

zero. In fact, both men and women show a 1% increase in income (significant for men, insignificant

for women, but not significantly different from each other), conditional on working.

Taken together, our results suggest that the most meaningful labor market effect of salt iodiza-

tion was a large increase in female labor supply. In addition, however, this shock generated small

increases in income (conditional on working) that were similar in percentage terms for both men

and women. Given that Feyrer et al. (2017) document that iodized salt had large effects on male

cognitive ability, it appears that this cognitive improvement led to only small changes in male eco- 181924 is only a partially treated year, but our specification allows us to observe whether we begin to see any effects

in this year.

nomic outcomes (a statistically significant increase in conditional income). Women, for whom

the magnitude of the cognitive improvement generated by iodized salt is not known,19 seemed to

have been much more dramatically affected by iodization. However, it is important to note that

the substantially larger effects on female labor force participation do not necessarily imply that the

cognitive effects of iodine were larger for women. Indeed, the fact that men showed such large

cognitive effects suggest that the reason for the gender difference we find is not biological but in-

stead, market-related. As Molina (2016) shows, an early-life health shock can have vastly different

effects on men and women because of the different labor market conditions that men and women

face. In our context, almost all men in our sample were in the labor force, while most women were

not. Put differently, the marginal man affected by salt iodization was already in the labor force,

while the marginal woman was likely not, which could be an important explanation why a large

cognitive shock only affected women along this dimension.

The next table breaks our sample down even further in order to study how the effects of salt

iodization may have differed over the course of these individuals’ careers. In particular, we are

interested in comparing effects in young adulthood and prime ages to effects in later adulthood.

Focusing on women, who were the only ones significantly impacted by salt iodization, we run

our labor market outcome regressions using only the 1950 and 1960 censuses (during which our

sample individuals were aged 19 to 40) and then using only the 1970 and 1980 censuses (during

which they were 39 to 60 years old). Sample sizes are substantially larger in Panel B because we

are using a 1% sample for all census waves except 1980 (which shows up in Panel B), for which we

use the 5% sample. Table 5 reports these regressions in Panels A and B, respectively, and reveals

a clear pattern. The effects of salt iodization on labor supply seem to be entirely driven by the

large impact salt iodization had on women early in their careers. For all outcomes, the early census

coefficients are larger in magnitude than the late census coefficients. With the exception of income

– for which we see a 6% increase even in later census waves – none of the late census coefficients

are significantly different from zero. It is worth noting that for three out of the four outcomes of 19The sample in Feyrer et al. (2017) was exclusively male.

Table 5: Effects of Salt Iodization on Female Labor and Income Outcomes in Early and Late Cen- suses

(1) (2) (3) (4)

Panel A: 1950-1960 Censuses (Ages 19 to 40)

After x Goiter Rate 0.0185*** 0.0221*** 0.0288*** 0.241**

(0.00593) (0.00657) (0.0106) (0.100)

During x Goiter Rate 0.0133** 0.0135* 0.0426*** 0.0309

(0.00631) (0.00695) (0.0111) (0.0761)

Observations 306087 306087 80777 180375

Mean of Dep. Var. 0.357 0.374 0.568 4.534

Panel B: 1970-1980 Censuses (Ages 39 to 60)

After x Goiter Rate 0.00266 0.00152 0.00286 0.0604*

(0.00330) (0.00329) (0.00405) (0.0327)

During x Goiter Rate -0.00346 -0.00254 0.00218 0.00762

(0.00265) (0.00278) (0.00457) (0.0337)

Observations 930333 930333 525927 928275

Mean of Dep. Var. 0.505 0.526 0.736 6.701

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions restrict to women born in 1920-1931. All

regressions include state of birth fixed effects, year of birth x census year dummies, census division of birth x birth

year dummies, gender, race, and During and After dummies interacted with average state latitude and 1920 state-

level female and black proportions. 1(Worked at least 40 weeks) is conditional on having worked in the past year. sinh

-1 (Income) takes the inverse hyperbolic sine of total income, including zeros for those not working.

1(Participated in

the Labor Force) 1(Employed)

1(Worked at least

40 weeks) sinh

-1 (Income)

interest in Panel A, there appear to be significant effects on the during cohort, emphasizing that the

effects of iodization do appear to show up quite rapidly (as was the case in Feyrer et al. (2017)).

Salt iodization, as a positive shock to cognitive ability, made women more employable and

increased their earning potential early in their careers. Later in life, the affected women appear to

have dropped out of the labor force (as a result of higher accumulated lifetime income or higher-

earning husbands, as we discuss in section 5.3), leaving them no more likely to be employed than

their unaffected counterparts. In the appendix, Table A3 reveals no effects for men in either census

wave pair, with the exception of a small income increase of 2% in later census waves (significant

at the 10% level), much smaller than the female income effects.

5.2 Robustness

There are a number of reasons why trends in labor market outcomes across birth cohorts in the

1920’s might differ across states. In order to interpret the coefficients discussed above as causal

estimates of the effect of iodized salt specifically, we must assume that any other drivers of these

differential birth cohort trends across states are uncorrelated with the distribution of goiter. A viola-

tion of this assumption means that we are omitting important variables that are potentially correlated

with our outcomes of interest as well as our after-by-goiter and during-by-goiter interactions. To

rule out alternative explanations for the effects that we find, we control for a number of important

events or policies that could have potentially affected the state-specific trends in outcomes across

our before, during, and after cohorts.

First, we consider contemporaneous health improvements, such as the eradication or treatment

of diseases, which occurred roughly contemporaneously to the roll out of iodized salt. In partic-

ular, malaria and hookworm eradication programs, both concentrated in the South, took place in

the decades immediately before and during the spread of iodized salt. Malaria eradication pro-

grams started in the 1920’s, while the hookworm eradication campaign began around 1910. The

correlation between early 1900’s goiter prevalence and malaria and hookworm rates are weak and

negative (-0.33 and -0.35, respectively) and thus unlikely to be driving our results. We validate

Table 6: Effects of Salt Iodization on Labor and Income Outcomes, By Gender, with Additional Controls

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0165*** 0.0187*** 0.0176** 0.201***

(0.00408) (0.00432) (0.00692) (0.0607)

During x Goiter Rate 0.0109** 0.0126*** 0.0281*** 0.0914*

(0.00450) (0.00455) (0.00631) (0.0455)

Observations 1179112 1179112 574618 1052355

Mean of Dep. Var. 0.425 0.445 0.657 5.558

Panel B: Males

After x Goiter Rate -0.00428 -0.00603 0.00725 0.0228

(0.00444) (0.00414) (0.00480) (0.0264)

During x Goiter Rate -0.00239 -0.00436 -0.00333 0.00763

(0.00292) (0.00267) (0.00339) (0.0266)

Observations 1092150 1092150 879680 973193

Mean of Dep. Var. 0.871 0.904 0.865 10.36

Panel C: Female-Male Difference

After x Goiter Rate 0.0208*** 0.0247*** 0.0104 0.178***

(0.00683) (0.00613) (0.00798) (0.0622)

During x Goiter Rate 0.0133** 0.0170*** 0.0314*** 0.0838

(0.00625) (0.00626) (0.00666) (0.0520)

Additional Controls

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1920-1931. All regressions include state of birth fixed effects, year of birth x census year

dummies, census division of birth x birth year dummies, gender, race, and During and After dummies interacted

state-of-birth hookworm and malaria prevalence rates (interacted with During and

After), state-of-birth compulsory schooling requirements at age 14, state-of-residence

WWII mobilization rates, state-of-birth unemployment rates in 1930 (interacted with

During and After), state-of-birth change in black population share from 1920 to 1940

(interacted with Durng and After), state-of-birth change in the total share of US

population living in state from 1920 to 1940 (interacted with During and After)

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

this, however, by including controls for pre-iodization prevalence rates of malaria and hookworm,

interacted with the during and after dummies.

We also address the possibility that changes in compulsory schooling laws, implemented at dif-

ferent times across states, resulted in differential trends in labor outcomes, which we are attributing

to introduction of iodized salt. We use data collected by Lleras-Muney (2002), which records the

minimum years of schooling required by law in each state from 1915 to 1939. Like Lleras-Muney

(2002), we match each individual to the compulsory schooling laws in place in their state of birth

at age 14 (the lowest minimum leaving age across all states). For the analysis discussed here, we

use the number of years of school required according to compulsory attendance laws, although the

results are similar when we use the number of years required according to child labor laws.

It is also well-documented that state mobilization rates for World War II affected labor force

participation, particularly for females during this period (Acemoglu et al., 2004). To control for

this, we use the state-level mobilization rate in an individual’s state of residence from Acemoglu

et al. (2004).

Finally, all of our sample individuals either lived through the Great Depression or its immediate

aftermath, but the before, during, and after cohorts were exposed at different points in their life,

which could have had important implications for the severity of the long-term impact on each

cohort. If, in addition, the Great Depression hit some states harder than others, it becomes another

potential reason for differential birth cohort trends across states. In order to proxy for a state’s

economic conditions during the Great Depression, we calculate state-level unemployment rates

from the 1930 census.20 We match this to individuals using their state of birth and control for the

interaction with during and after dummies.

We use a similar strategy to address concerns that the Great Migration could have also resulted in

the differential cohort trends that we are attributing to the iodization of salt. During the childhood

years of our sample cohort, many of these individuals were exposed to large shares of African

Americans either moving out of or moving into their communities. To rule out the possibility that 20Unemployment rates are not available in the 1920 census.

these demographic shifts are driving our results above, we allow for differential trends across states

that experienced different racial composition changes and different levels of population growth

between 1920 and 1940, two decades of substantial migration that coincided with the childhood

years of our cohorts. Specifically, we include During and After interactions with the following two

variables: the 1920 to 1940 change in the black population share in an individual’s state of birth

and the 1920 to 1940 change in the share of the total U.S. population living in an individual’s state

of birth.

Table 6 reports the results of regressions that control for all of the potential confounders just

discussed. The first panel reports the results for women and the second panel reports the results

for men. Across all outcomes and samples, we find very similar results: positive and significant

effects on female labor supply and income, but no effects on male outcomes.

The online appendix contains additional robustness checks, where we (1) account for mean

reversion (Table B3), (2) show that the Dust Bowl was not an important confounder (Table B4),

(3) show that our results are robust to dropping states below the Mason-Dixon line (Table B5), (4)

show that it is indeed goiter in the state of birth (rather than the state of residence) that is driving

our results (Table B6), and (5) show that our results our robust to a specification that compares

individuals at more similar ages (Table B7).

We also show that increased access to iodine did not affect the mortality rates of our cohorts,

which alleviates concerns about our results being driven by a changing sample composition induced

by differential mortality. Table A4 shows no differential trends across the goiter distribution in

terms of cohort size or cohort gender composition.

5.3 Additional Outcomes

Having established that improved access to iodine substantially improved labor market outcomes,

particularly for females, we next study the effects of this cognitive shock on other dimensions of

life. Table 7 reports the results of our main regressions on educational attainment and marriage

outcomes, restricting to the 1970-1980 census waves in order to focus on a sample of individuals

Table 7: Effects of Salt Iodization on Education and Marital Outcomes

(1) (2) (3) (4) (5) (6) (7)

Panel A: Females

After x Goiter Rate 0.0712* 0.000541 0.232*** 0.0888** 0.00755 0.0177* -0.00488**

(0.0379) (0.00182) (0.0478) (0.0402) (0.0123) (0.00936) (0.00237)

During x Goiter Rate 0.0359 0.000702 0.0394 0.0727*** 0.0277** 0.0113 0.000885

(0.0274) (0.00191) (0.0543) (0.0264) (0.0112) (0.00812) (0.00236)

Age 0.0276***

(0.000414)

Observations 930333 930333 761885 695464 692713 922124 25581376

Mean of Dep. Var. 11.31 0.951 21.43 11.31 11.05 11.18 0.731

Panel B: Males

After x Goiter Rate 0.0313 0.00288 -0.0405 0.0189 0.0178 0.0168** 0.00388**

(0.0464) (0.00235) (0.0454) (0.0288) (0.0448) (0.00793) (0.00168)

During x Goiter Rate 0.0990** 0.00220 -0.0355 0.0249 -0.00546 0.0202*** 0.00321

(0.0396) (0.00187) (0.0594) (0.0259) (0.0511) (0.00641) (0.00250)

Age 0.0334***

(0.000207)

Observations 859574 859574 692623 716526 714625 846971 23566464

Mean of Dep. Var. 11.45 0.937 24.16 11.59 5.833 11.35 0.641

Panel C: Female-Male Difference

After x Goiter Rate 0.0399 -0.00234 0.273*** 0.0699** -0.0102 0.000901 -0.00875***

(0.0369) (0.00355) (0.0575) (0.0262) (0.0499) (0.0103) (0.00294)

During x Goiter Rate -0.0631** -0.00150 0.0749 0.0478* 0.0331 -0.00885 -0.00232

(0.0302) (0.00267) (0.0716) (0.0251) (0.0571) (0.00834) (0.00282)

sinh -1

(Spouse's

Income)

sinh -1

(Family

Income)

1(Ever

Married)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the goiter rate in the individual's state of birth

from Love and Davenport (1920), scaled by the difference between the 75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for

those born 1928-1931. During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1970 to 1980 censuses, restricting to individuals born in

1920-1931. All regressions include state of birth fixed effects, year of birth x census year dummies, census division of birth x birth year dummies, gender, race,

and During and After dummies interacted with average state latitude and 1920 state-level female and black proportions. 1(Worked at least 40 weeks) is

conditional on having worked in the past year. sinh -1

(income variable) takes the inverse hyperbolic sine of the relevant income variable (spouse's total income or

total family income), including zeros for those not working. Column 7 uses a panel dataset, where each observation represents an individual-age for each age

from 14 to 45 (the 1st and 99th percentiles of age at first marriage).

Years of

Schooling

1(Ever

Married)

Age at First

Marriage

Spouse's Years

of Schooling

old enough to have completed their schooling and made their first marriage decisions.

First, we ask whether educational attainment was an important mechanism behind the positive

labor market effects of improved cognitive ability. Column 1 of Table 7 suggests that it was not.

Although the after-by-goiter coefficient is positive and statistically significant for women (and the

during-by-goiter coefficient is positive and significant for men), the magnitudes of these coeffi-

cients are small, translating to about 2 weeks of school, much too small to be generating the large

effects on income reported in Table 5. It should be noted that in a standard model of educational at-

tainment (Card, 2001), a positive shock to the ability endowment can lead to either an increase or a

decrease in educational attainment because it can raise the returns to education as well as the initial

wage earned (without any education).21 In this case, these two opposing effects appear to almost

cancel each other out, leading to a small but significant increase in average educational attainment.

Next, we ask whether changes in labor market outcomes were accompanied by changes in mar-

ital decisions. Increased iodine availability does not appear to have affected overall ever-married

rates, which is unsurprising given that over 90% of individuals have been married at least once.

However, this cognitive shock does appear to have resulted in delays in marriage for women: we

estimate a small but statistically significant increase in the age at first marriage (conditional on

having ever married) of approximately a quarter of a year. Because age of marriage is a censored

variable, we verify that our results hold when we conduct this analysis at the individual-age level,

where each observation represents an individual at a particular age (from 14 to 45, the 1st and 99th

percentiles of the age at first marriage variable). In column 7, we regress an indicator for whether

the individual has ever been married by that age on our usual specification, controlling additionally

for age. Consistent with column 3, column 7 reveals that salt iodization reduces the likelihood of a

woman being ever-married at any given age. On the other hand, we see the opposite effect on men,

who are more likely to be ever-married at any given age as a result of iodization.

Interestingly, the increased availability of iodine also appears to have affected spousal quality 21If the ability endowment shock increases the initial wage regardless of education and the return to education is

relatively low, then lifetime income can increase with lower schooling. If there is disutility associated with schooling then there is further downward pressure on educational attainment.

for women, where spousal characteristics are measured for individuals living in the same household

as their spouse. In columns 4 and 5, we see that women affected by salt iodization marry more

educated and higher income spouses. We do not see the same effects on spousal quality for men.

Consistent with the findings that exposure to iodine resulted in higher-income spouses as well as

higher individual income for women (column 4 of Panel B of Table 5), increased access to iodine

led to significantly higher family income for females. This result is also true for men (and the effect

sizes are the same across genders).

These results – in particular, the spousal quality effects for women – help shed light on our

findings in Table 5, which revealed that the positive labor force participation effects for women

were largest in the early census waves and faded out later in their careers. Greater access to iodine

increased female income early in their careers and also resulted in marriages to more educated

and higher-income men. Both of these factors likely led to higher accumulated wealth for these

women later in life (which is consistent with, though not fully captured by, the positive effects on

total family income in Table 7). The fade-out of the female labor force participation effects can

therefore be explained by a simple income effect (wealthier women demanding more leisure).22

We explore this hypothesis further in Appendix Table A5, where we look again at our main labor

market outcomes for females in the late censuses, but allow for heterogeneity in the during and after

interactions across women with above-median and below-median family income. We acknowledge

that the endogeneity of the family income variable makes a definitive causal interpretation of these

results impossible, but what we find does offer suggestive evidence for the hypothesis discussed

above. In particular, for women with below-median family income, the positive labor force effects

of iodization do persist into the late census waves. It is the women with above-median family

income, for whom the negative income effect is likely to be more relevant, who are responsible for

the overall null effects on labor force participation later in their careers.

In sum, increased availability of iodine in utero affected female marriage outcomes in addition 22An alternative explanation for the fade-out of female labor force participation effects is that women who were not

affected by iodization eventually caught up to their affected counterparts, joining the labor force later in life. We are unable to provide any evidence that this was the case, though it is certainly possible that both of these explanations played a role.

to their labor market outcomes, which sheds some light on why the female labor force participation

effects did not persist into the late census waves. Importantly, these results reveal several other

outcomes (spousal quality for women and total family income for both genders) that, unlike female

labor force participation, appear to have been persistently affected by the increased access to iodine.

6 Conclusion

In this study, we document the effects of the rapid nationwide iodization of salt in the United States.

We estimate substantial impacts on employment and labor force participation for women early in

their careers. There is evidence of smaller income effects for both genders that persist into their

forties and fifties. Additional results show that impacted women marry more educated and higher-

income spouses at later ages, consistent with treated new labor force entrants transitioning out of

the work force at later ages due to a negative income effect.

Our results contribute to several strands of literature and current policy debates. First, this study

contributes to the growing literature on the long-term effects of early-life conditions, particularly

the smaller set of recent work, estimating gains to purposeful and beneficial large-scale policy inter-

ventions like fortification schemes. These results differ from earlier studies of early-life “shocks”

in that they validate the impacts of actionable policies which can then be reproduced elsewhere. In

this way, the study of historic successes, and failures for that matter, in the US and other developed

settings, can potentially provide important predictions for academic researchers and policy-makers

faced with similar issues in developing countries today.

Second, while previous studies have estimated the roles of historical events, such as World War

II and the staged rise in access to contraception, in explaining increases in labor force participa-

tion among women (e.g. Goldin (1991), Goldin & Olivetti (2013), Goldin & Katz (2002), and

Bailey (2006)), we contribute complementary evidence that salt iodization explains a rise of 2.21

percentage points in early censuses (roughly 6 percent of the total rise from 1950 to 1990). Our ev-

idence pertains to cohorts born after those most affected by the war, but before those most affected

by increased access to oral contraceptives.23 Unlike for these previously studied events, impacts

on participation of salt iodization are not focused on higher-educated women but prevail despite

negligible impacts on schooling completion.

Additionally, our study provides evidence of the magnitude of benefits from eradication of

deficiencies in essential micronutrients such as iodine. Many developing country populations face

myriad nutritional constraints, which have long-lasting impacts on health, economic livelihoods,

and general welfare. Our estimates show that salt iodization led to a roughly 1.21 percentage point

rise in female labor force participation. From a base labor force of 13 million women in 1940

(Durand & Goldfield, 1944), this amounts to almost 2 billion USD in additional income using the

mean income for the female sample inflated to 2016 dollars (13 million x 0.0121 x 12,514 USD =

1.97 billion USD).24

Lastly, it should be noted that the “intervention” cost the taxpayer nothing, in that the roll-out

of iodized salt was completely undertaken by the private sector. That is, the cost of salt iodization

was fully borne by the salt producer,25 while the cognitive benefit was realized by the general

population. We conjecture that the rapid rise in both supply and demand might be attributable to

the efficiency and underlying profit motive of the private firm that undertook the intervention.

23Indeed, we estimate our rise in labor force participation due to salt iodization relative to the cohort affected by the war.

24This calculation does not take into account the value of home production, which could have decreased with more women entering the labor force.

25“The producers and the wholesale grocers each bore one half of the added expense so that the iodized salt would not cost the consumer one cent more.” (Kimball, 1937, pg. 32)

References

Acemoglu, Daron, David Lyle et al. 2004. Women, war, and wages: The effect of female labor

supply on the wage structure at midcentury. Journal of Political Economy 112(3). 497–551.

Almond, Douglas & Janet Currie. 2011. Killing me softly: The fetal origins hypothesis. Journal

of Economic Perspectives 25(3). 153–172.

Almond, Douglas, Joseph Doyle, Amanda Kowalski & Heidi Williams. 2010. Estimating marginal

returns to medical care: Evidence from at-risk newborns. Quarterly Journal of Economics

125(2). 591�634.

Annegers, J & O Mickelsen. 1973. Goiter and iodized salt, a historical review and examination of

the present role of iodized salt. Unpublished manuscript .

Backstrand, J. 2002. The history and future of food fortification in the United States: A public

health perspective. Nutrition Reviews 60. 15–26.

Bailey, Martha J. 2006. More power to the pill: the impact of contraceptive freedom on women’s

life cycle labor supply. The Quarterly Journal of Economics 121(1). 289–320.

Baumann, F. 1896. Ueber das normale vorkommen von jod im thierkorper. Physiol Chem 21.

319–330.

Bhalotra, Sonia R & Atheendar Venkataramani. 2015. Shadows of the captain of the men of death:

Early life health interventions, human capital investments, and institutions .

Bhutta, Zulfiqar A, Jai K Das, Arjumand Rizvi, Michelle F Gaffey, Neff Walker, Susan Horton,

Patrick Webb, Anna Lartey, Robert E Black, The Lancet Nutrition Interventions Review Group

et al. 2013. Evidence-based interventions for improvement of maternal and child nutrition: what

can be done and at what cost? The Lancet 382(9890). 452–477.

Black, Robert E, Cesar G Victora, Susan P Walker, Zulfiqar A Bhutta, Parul Christian, Mercedes

De Onis, Majid Ezzati, Sally Grantham-McGregor, Joanne Katz, Reynaldo Martorell et al. 2013.

Maternal and child undernutrition and overweight in low-income and middle-income countries.

The lancet 382(9890). 427–451.

Bleakley, Hoyt. 2007. Disease and development: evidence from hookworm eradication in the

American South. The Quarterly Journal of Economics 122(1). 73–117.

Bleakley, Hoyt. 2010. Malaria eradication in the americas: A retrospective analysis of childhood

exposure. American Economic Journal: Applied Economics 2(2). 1–45.

Brush, Brock E. & J.K. Atland. 1952. Goiter prevention with iodized salt: Results of a thirty-year

study. The Journal of Clinical Endocrinology And Metabolism 12(10). 1380–1388.

Cagnacci, Angelo, A Renzi, S Arangino, C Alessandrini & A Volpe. 2004. Influences of maternal

weight on the secondary sex ratio of human offspring. Human Reproduction 19(2). 442–444.

Cao, Xue-Yi, Xin-Min Jiang, Zhi-Hong Dou, Murdon Abdul Rakeman, Ming-Li Zhang, Karen

O’Donnell, Tai Ma, Kareem Amette, Nancy DeLong & G. Robert DeLong. 1994. Timing of

vulnerability of the brain to iodine deficiency in endemic cretinism. New England Journal of

Medicine 331(26). 1739–1744.

Card, David. 2001. Estimating the return to schooling: Progress on some persistent econometric

problems. Econometrica 69(5). 1127–60.

Clar, C., T. Wu, G. Liu & P. Li. 2002. Iodized salt for iodine deficiency disorders: A systematic

review. Endocrinology and Metabolism Clinics of North America 31. 681–698.

Clay, Karen, Ethan Schmick & Werner Troesken. 2015. Pellagra and southern welfare: Evidence

from the arrival of the boll weevil, and state-level fortification laws .

Currie, Janet & Tom Vogl. 2013. Early-life health and adult circumstance in developing countries.

Annu. Rev. Econ. 5(1). 1–36.

De Benoist, Bruno, Maria Andersson, Ines Monika Egli, Takkouche El Bahi, Henrietta Allen,

World Health Organization et al. 2004. Iodine status worldwide: Who global database on iodine

deficiency .

De Escobar, G Morreale, María Jesús Obregón & F Escobar Del Rey. 2004. Role of thyroid hor-

mone during early brain development. European Journal of Endocrinology 151(Suppl 3). U25–

U37.

Durand, John Dana & Edwin D Goldfield. 1944. Sixteenth census of the united states: 1940.

population. estimates of labor force, employment, and unemployment in the united states, 1940

and 1930 .

Feyrer, James, Dimitra Politi & David N Weil. 2017. The cognitive effects of micronutrient de-

ficiency: evidence from salt iodization in the united states. Journal of the European Economic

Association 15(2). 355–387.

Field, Erica, Omar Robles & Maximo Torero. 2009. Iodine deficiency and schooling attainment in

tanzania. American Economic Journal: Applied Microeconomics 1(4). 140–169.

Friedhoff, A., J.Miller, M.Armour, J.Schweitzer&S.Mohan.2000. Roleofmaternalbiochemistry

in fetal brain development: Effect of maternal thyroidectomy on behaviour and biogenic amine

metabolism in rat progeny. International Journal of Neuropsychopharmacology 3. 89–97.

Goldin, Claudia, R Ehrenberg & F Blau. 1997. Career and family: College women look to the past.

Gender and Family Issues in the Workplace .

Goldin, Claudia & Lawrence F Katz. 2002. The power of the pill: Oral contraceptives and women’s

career and marriage decisions. Journal of political Economy 110(4). 730–770.

Goldin, Claudia & Claudia Olivetti. 2013. Shocking labor supply: A reassessment of the role of

World War II on women’s labor supply. American Economic Review: Papers and Proceedings

103(3). 257–262.

Goldin, Claudia D. 1991. The role of World War II in the rise of women’s employment. The

American Economic Review 81(4). 741–756.

Hall, D. 1914. The prevalence of goiter in the northwest, based on the examinations of 3,339

students entering the University of Washington. Northwest Medicine 6. 189–391.

Hamwi, George J., Albert W. Van Fossen, Robert E. Whetstone & Izola Williams. 1952. Endemic

goiter in ohio school children. American Journal of Public Health 45(10). 1344–1348.

Heckman, James J. 2006. Skill formation and the economics of investing in disadvantaged children.

Science 312(5782). 1900–1902.

Hetzel, Basil S. 1989. The story of iodine deficiency: an international challenge in nutrition .

Hetzel, Basil S. & Mark T. Mano. 1989. A review of experimental studies of iodine deficiency

during fetal development. Journal of Nutrition 119(2). 145–151.

Hornbeck, Richard. 2012. The enduring impact of the American Dust Bowl: Short- and long-run

adjustments to environmental catastrophe. American Economic Review 102(4). 1477–1507.

Horton, S, H Alderman & J Rivera. 2008. Copenhagen consensus 2008. Malnutrition and hunger.

Executive summary. Copenhagen: Copenhagen Business School .

Hoynes, Hilary, Diane Whitmore Schanzenbach & Douglas Almond. 2016. Long-run impacts of

childhood access to the safety net. The American Economic Review 106(4). 903–934.

Kimball, O.P. 1937. Prevention of goiter in michigan and ohio. Journal of the American Medical

Association 108(11). 860–864.

Kofoid, Charles A. & John P. Tucker. 1921. On the relationship of infection by hookworm to the

incidence of morbidity and mortality in 22,842 men in the united states army at camp bowie,

texas, from october 1917 to april 1918. American Journal of Hygiene 1(1). 79–117.

Lamberg, Bror-Axel. 1991. Endemic goitre�iodine deficiency disorders. Annals of Medicine

23(4). 367–372.

Levin, Simon. 1919. Discussion of goiter in 583 registrants. Journal of Michigan State Medical

Society 18. 98–104.

Lleras-Muney, Adriana. 2002. Were compulsory attendance and child labor laws effective? an

analysis from 1915 to 1939. Journal of Law and Economics 45(2). 401–435.

Love, Albert Gallatin & Charles Benedict Davenport. 1920. Defects found in drafted men: Sta-

tistical information compiled from the draft records showing the physical condition of the men

registered and examined in pursuance of the requirements of the selective-service act. US Gov-

ernment Printing Office.

Maccini, Sharon & Dean Yang. 2009. Under the weather: Health, schooling, and economic conse-

quences of early-life rainfall. American Economic Review 99(3). 1006–1026.

Marine, D. & H. Feiss. 1915. The absorption of potassium iodid by perfused thyroid glands and

some of the factors modifying it. Journal of Pharmacology and Experimental Therapeutics 7.

557–576.

Marine, D. & O. Kimball. 1917. The prevention of simple goiter in man. Journal of Laboratory

and Clinical Medicine 3. 40–48.

Marine, D. & C. Lenhart. 1909. Effects of the administration or withholding of iodin-containing

compounds in normal, colloid, or actively hyperplastic thyroids of dogs. Archives of Internal

Medicine 4. 253–270.

Marine, David & OP Kimball. 1920. Prevention of simple goiter in man: fourth paper. Archives of

internal medicine 25(6). 661–672.

Markel, Howard. 1987. When it rains it pours: Endemic goiter, iodized salt and David Murray

Cowie, MD. American Journal of Public Health 77(2). 219–229.

Molina, Teresa. 2016. Pollution, ability, and gender-specific investment responses to shocks.

Niemesh, Gregory T. 2015. Ironing out deficiencies: Evidence from the united states on the eco-

nomic effects of iron deficiency. Journal of Human Resources 50(4). 910–958.

Olesen, R. 1915. Goiter, its prevalence in chicago, as shown by the examination of 800 individuals.

Illinois Medical Journal 27. 16.

Olesen, R. 1929. Distribution of endemic goiter in the united states as shown by thyroid surveys.

Public Health Reports 44. 1463–87.

Pharoah, Peter & Kevin Connolly. 1987. A controlled trial of iodinated oil for the prevention of

endemic cretinism: A long-term follow-up. International Journal of Epidemiology 16(1). 68–73.

Politi, Dimitra. 2010. The impact of iodine deficiency eradication on schooling: evidence from the

introduction of iodized salt in switzerland .

Politi, Dimitra. 2014. The effects of the generalized use of iodized salt on occupational patterns in

switzerland .

Ruggles, Steven, Katie Genadek, Ronald Goeken, Josiah Grover & Matthew Sobek. 2015. Inte-

grated public use microdata series: Version 6.0 [machine-readable database]. minneapolis: Uni-

versity of minnesota: 2015.

Sanders, Nicholas J & Charles Stoecker. 2015. Where have all the young men gone? using sex

ratios to measure fetal death rates. Journal of health economics 41. 30–45.

Schiel, Joseph B. Jr. & Anita Joan Wepfer. 1976. Distributional aspects of endemic goiter in the

united states. Economic Geography 52(2). 116–126.

Smith, G. 1917. Fetal athyreosis. Journal of Biological Chemistry 29. 215–225.

Trivers, Robert L & Dan E Willard. 1973. Natural selection of parental ability to vary the sex ratio

of offspring. Science 179(4068). 90–92.

Zimmermann, Michael B. 2008. Research on iodine deficiency and goiter in the 19th and early

20th centuries. The Journal of Nutrition 138(11). 2060–2063.

Zimmermann, Michael B. 2009. Iodine deficiency in pregnancy and the effects of maternal iodine

supplementation on the offspring: a review. The American Journal of Clinical Nutrition 89(2).

668S–672S.

A Appendix

In Table A1, we list all states (excluding Hawaii) by Census division, along with their corresponding

goiter rates from Love & Davenport (1920).

Table A1: State-Level Goiter Rates from Love & Davenport (1920)

Census Division State

Love and Davenport

(1920) Goiter Rate

(ppt)

New England Connecticut 0.089

New England Maine 0.066

New England Massachusetts 0.032

New England New Hampshire 0.070

New England Rhode Island 0.055

New England Vermont 0.214

Middle Atlantic New Jersey 0.043

Middle Atlantic New York 0.119

Middle Atlantic Pennsylvania 0.410

East North Central Illinois 0.779

East North Central Indiana 0.649

East North Central Michigan 1.143

East North Central Ohio 0.559

East North Central Wisconsin 1.402

West North Central Iowa 0.668

West North Central Kansas 0.125

West North Central Minnesota 0.804

West North Central Missouri 0.399

West North Central Nebraska 0.214

West North Central North Dakota 0.873

West North Central South Dakota 0.409

South Atlantic Delaware 0.059

South Atlantic District of Columbia 0.139

South Atlantic Florida 0.025

South Atlantic Georgia 0.052

South Atlantic Maryland 0.094

South Atlantic North Carolina 0.181

South Atlantic South Carolina 0.094

South Atlantic Virginia 0.338

South Atlantic West Virginia 0.789

East South Central Alabama 0.056

East South Central Kentucky 0.141

East South Central Mississippi 0.064

East South Central Tennessee 0.196

West South Central Arkansas 0.040

West South Central Louisiana 0.062

West South Central Oklahoma 0.072

West South Central Texas 0.030

Mountain Arizona 0.121

Mountain Colorado 0.529

Mountain Idaho 2.691

Mountain Montana 2.100

Mountain Nevada 0.638

Mountain New Mexico 0.088

Mountain Utah 1.572

Mountain Wyoming 1.537

Pacific Alaska 1.314

Pacific California 0.445

Pacific Oregon 2.631

Pacific Washington 2.340

Table A2 reports regression results for three different variants of the total income variable.

Table A2: Effects of Salt Iodization on Income, By Gender

(1) (2) (3)

Panel A: Females

After x Goiter Rate 217.5*** 113.6 0.0199

(80.55) (109.1) (0.0139)

During x Goiter Rate 27.69 38.22 0.0116

(81.29) (116.5) (0.0145)

Observations 1108650 716883 716883

Mean of Dep. Var. 8690.0 15294.7 9.139

Panel B: Males

After x Goiter Rate 328.4** 293.8** 0.0142*

(136.9) (138.0) (0.00811)

During x Goiter Rate 295.5*** 297.0*** 0.0108**

(101.3) (91.80) (0.00456)

Observations 1026746 1003623 1003623

Mean of Dep. Var. 31193.1 32445.8 10.11

Panel C: Female-Male Difference

After x Goiter Rate -111.0 -180.2 0.00574

(157.5) (180.2) (0.0169)

During x Goiter Rate -267.8** -258.8* 0.000813

(125.4) (145.5) (0.0164)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

Goiter Rate is the goiter rate in the individual's state of birth from Love and Davenport (1920),

scaled by the difference between the 75th and 25th percentile of the goiter distribution (0.71).

After is a dummy equal to 1 for those born 1928-1931. During is a dummy equal to 1 for those

born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting to individuals born

in 1920-1931. All regressions include state of birth fixed effects, year of birth x census year

dummies, census division of birth x birth year dummies, gender, race, and During and After

dummies interacted with average state latitude and 1920 state-level female and black

proportions.

Income (in dollars) Conditional

Income (in dollars) Log(Income)

Table A3 reports the results of our labor regressions on men, separately for early and late census

waves.

Table A3: Effects of Salt Iodization on Male Labor and Income Outcomes in Early and Late Cen- suses

(1) (2) (3) (4)

Panel A: 1950-1960 Censuses (Ages 19 to 40)

After x Goiter Rate 0.00385 0.00377 0.00862 0.0350

(0.00660) (0.00588) (0.00816) (0.0355)

During x Goiter Rate 0.000883 0.000452 -0.00239 -0.00530

(0.00421) (0.00427) (0.00749) (0.0450)

Observations 287149 287149 162305 170601

Mean of Dep. Var. 0.874 0.912 0.818 9.870

Panel B: 1970-1980 Censuses (Ages 39 to 60)

After x Goiter Rate 0.000334 -0.00332 0.00137 0.0211

(0.00263) (0.00274) (0.00184) (0.0127)

During x Goiter Rate 0.00115 -0.00140 0.00303 0.0263*

(0.00238) (0.00231) (0.00182) (0.0140)

Observations 859574 859574 767994 856145

Mean of Dep. Var. 0.870 0.898 0.920 10.93

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions restrict to men born in 1920-1931. All

regressions include state of birth fixed effects, year of birth x census year dummies, census division of birth x birth

year dummies, gender, race, and During and After dummies interacted with average state latitude and 1920 state-

level female and black proportions. 1(Worked at least 40 weeks) is conditional on having worked in the past year. sinh

-1 (Income) takes the inverse hyperbolic sine of total income, including zeros for those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

In order to assess whether the increased availability of iodine affected mortality, we take all

individuals born between 1920 and 1931 in the 1940 census and collapse to the birth-year birth-

state level, counting the total number of individuals as well as calculating the male fraction of each

cohort. We use the 1940 census because it is the first census to record all of our birth cohorts of in-

terest. If salt iodization affected mortality, we would expect to see effects on cohort size. We might

also see effects on the male fraction because male fetuses are more vulnerable in utero (Cagnacci

et al., 2004; Sanders & Stoecker, 2015; Trivers & Willard, 1973). We therefore regress cohort size

and the male proportion on our after-by-goiter and during-by-goiter variables of interest, control-

ling for birth state and birth year fixed effects, along with after and during dummies interacted with

state latitude and 1920 female and black proportions. We find no differential changes across the

goiter distribution in cohort size or gender composition after the introduction of iodized salt. None

of the coefficients in Table A4 are significantly different from zero (in terms of magnitude, they

are at most a few hundredths of a standard deviation), suggesting that salt iodization did not affect

our sample composition.

It is worth noting that Feyrer et al. (2017) document that salt iodization increased mortality

among older adults, which could have resulted in reductions in our cohort size if any of the indi-

viduals that died were of child-bearing age. If these effects were large, it would be possible for the

null effects estimated in Table A4 to be masking large decreases in mortality for children (because

of the offsetting increases in mortality for would-be mothers). Based on the estimates in Feyrer

et al. (2017), however, we argue that the number of adult deaths is much too small to be offsetting

any important decreases in mortality on the other end of the age distribution.26

26Between 1921 and 1926, Feyrer et al. (2017) estimate an additional 2,340 deaths per year resulted from iodized salt. If we assume (conservatively) that half of these deaths were from people older than child-bearing age, this leaves us with an estimate of 1000 deaths per year (of which only a small proportion would have given birth in any given year), a very small number compared to an average cohort size of around 60,000.

Table A4: Salt Iodization and Cohort Composition

(1) (2)

After x Goiter Rate 525.7 0.00602

(416.1) (0.00619)

During x Goiter Rate -146.4 -0.0330

(316.5) (0.0217)

Observations 596 596

Mean of Dep. Var. 46912.7 0.495

Standard Deviation of Dep. Var. 43529.8 0.0660

Notes: Standard errors, clustered by state, in parentheses (*** p<0.01, **

p<0.05, * p<0.1). Each observation represents a birth-state birth-year

combination. Goiter Rate is the goiter rate in the individual's state of birth

from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy

equal to 1 for birth years 1928-1931. During is a dummy equal to 1 for

birth years 1924-1927. These regressions use the 1940 census, restricting to

birth years 1920-1931. All regressions include During and After dummies

interacted with average state latitude and 1920 state-level female and

black proportions.

Cohort Size Fraction Male

Table A5 reports the results of our regressions on female labor market outcomes in late census

waves, allowing for heterogeneity in the effect of iodization across women with above-median

family income and below-median family income. For those with below-median family income, we

see that the effects of iodization on labor force participation do persist.

Table A5: Effects of Salt Iodization on Female Labor Market Outcomes, by Family Income Indi- cator

(1) (2) (3) (4)

After x Goiter Rate -0.0168*** -0.0205*** -0.0110*** -0.188***

x 1(Above Median Family Income) (0.00441) (0.00505) (0.00364) (0.0529)

During x Goiter Rate -0.0143*** -0.0168*** -0.00603* -0.181***

x 1(Above Median Family Income) (0.00341) (0.00350) (0.00348) (0.0419)

After x Goiter Rate 0.0132*** 0.0147*** 0.00978** 0.185***

(0.00421) (0.00457) (0.00457) (0.0510)

During x Goiter Rate 0.00517 0.00793* 0.00653 0.118**

(0.00392) (0.00400) (0.00518) (0.0444)

1(Above Median Family Income) 0.111*** 0.101*** 0.0649*** 0.734***

(0.00482) (0.00502) (0.00332) (0.0561)

Observations 923887 923887 524163 921829

Mean of Dep. Var. 0.506 0.528 0.736 6.707

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the 75th

and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931. During is a

dummy equal to 1 for those born 1924-1927. These regressions use the 1970 to 1980 censuses, restricting to women born in 1920-1931. All regressions include state of birth fixed effects, year of birth x census year dummies, census

division of birth x birth year dummies, gender, race, and During and After dummies interacted with average state

latitude and 1920 state-level female and black proportions. 1(Worked at least 40 weeks) is conditional on having

worked in the past year. sinh -1

(Income) takes the inverse hyperbolic sine of total income, including zeros for those not

working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at

least 40 weeks) sinh

-1 (Income)

Online Appendix

B2 Additional Tables and Figures

Table B1: Summary Statistics, by Census Year

(1) (2) (3) (4)

1950 Census 1960 Census 1970 Census 1980 Census

1(Employed) 0.588 0.631 0.705 0.656

(0.492) (0.482) (0.456) (0.475)

1(Participated in Labor Force) 0.616 0.658 0.727 0.684

(0.486) (0.474) (0.446) (0.465)

1(Worked at least 40 weeks) 0.671 0.794 0.843 0.850

conditional on working (0.470) (0.405) (0.364) (0.357)

Total Income 9135.5 17812.2 25756.6 26450.3

(10082.8) (18050.1) (22575.1) (22990.8)

Years of Schooling 11.28 11.47

(3.116) (3.156)

1(Ever Married) 0.941 0.948

(0.236) (0.222)

Age at First Marriage 22.49 22.98

(4.841) (5.789)

Spouse's Schooling 11.35 11.56

(2.982) (3.028)

Spouse's Total Income 25352.4 25653.2

(23211.6) (23136.8)

Total Family Income 49056.7 48277.9

(19756.3) (21596.4)

Number of Observations 325602 267634 509165 1280742

Notes: Sample includes all individuals born 1920 to 1931. Statistics are calculated using person-

level weights provided by the census. Total income in 1999 dollars.

Figure B1: Iodine Content of Drinking Water in the US

Simple goiter among drafted men in the US in WW I

Black areas: High goiter incidence, i.e. 6 and more

goiter cases per 1,000 drafted men

White areas: Low goiter incidence, i.e. 5 and less goiter

cases per 1,000 drafted men

Source: McClendon (1939)

Iodine content in drinking water in the US

Black areas: Iodine-poor, i.e. 22 and less parts of iodine

per hundred billion parts of water

White areas: Iodine-rich, i.e. 23 and more parts of iodine

per hundred billions parts of water

Source: McClendon and Hathaway (1924)

Figure B2: U.S. goiter distribution in the early 1950s

Source: Schiel and Wepfer (1976)

Figure B3: Change in the per capita riboflavin, iron, niacin, and thiamin content of the U.S. food supply between 1909 and 1994

Source: Backstrand (2002)

Table B2 reports the coefficient estimates for the event study specification that produced Figure

Table B2: Year-by-Year Effects of Salt Iodization on Labor and Income Outcomes

(1) (2) (3) (4)

Born 1914 x Goiter Rate 0.000920 0.00242 -0.0113* 0.136*** (0.00328) (0.00275) (0.00575) (0.0498)

Born 1915 x Goiter Rate -0.00173 -0.00180 0.00873 0.0782

(0.00330) (0.00330) (0.00579) (0.0619)

Born 1916 x Goiter Rate -0.00460 -0.00570* -0.00871 0.0307

(0.00306) (0.00333) (0.00631) (0.0388)

Born 1917 x Goiter Rate -0.000799 -0.00119 0.00509 0.0864* (0.00348) (0.00297) (0.00842) (0.0480)

Born 1918 x Goiter Rate -0.00225 -0.000649 -0.00154 0.00388

(0.00334) (0.00328) (0.00590) (0.0445)

Born 1919 x Goiter Rate -0.00200 -0.000874 0.00541 0.0347

(0.00373) (0.00321) (0.00516) (0.0543)

Born 1920 x Goiter Rate -0.00415 -0.00382 -0.00264 -0.0468 (0.00325) (0.00304) (0.00624) (0.0392)

Born 1921 x Goiter Rate -0.000311 0.0000988 0.00477 0.0608

(0.00222) (0.00210) (0.00583) (0.0382)

Born 1922 x Goiter Rate -0.00122 -0.00130 0.00643 0.0244

(0.00277) (0.00304) (0.00477) (0.0419)

Born 1924 x Goiter Rate -0.000143 0.000792 0.0115** 0.0462

(0.00321) (0.00339) (0.00435) (0.0457)

Born 1925 x Goiter Rate 0.00401 0.00311 0.00949 0.0339

(0.00364) (0.00338) (0.00575) (0.0325)

Born 1926 x Goiter Rate 0.00551* 0.00637** 0.0107* 0.0451 (0.00305) (0.00305) (0.00561) (0.0545)

Born 1927 x Goiter Rate -0.000723 -0.00233 0.0113* 0.0237

(0.00410) (0.00374) (0.00610) (0.0324)

Born 1928 x Goiter Rate 0.00663* 0.00598 0.00996 0.102**

(0.00357) (0.00410) (0.00675) (0.0385)

Born 1929 x Goiter Rate 0.00201 0.00330 0.0137* 0.132**

(0.00405) (0.00310) (0.00733) (0.0590)

Born 1930 x Goiter Rate 0.00252 0.00316 0.00971 0.102**

(0.00396) (0.00400) (0.00810) (0.0401)

Born 1931 x Goiter Rate 0.0118** 0.00999** 0.0105* 0.131***

(0.00472) (0.00483) (0.00562) (0.0447)

Observations 3463283 3463283 2103021 3090332

Mean of Dep. Var. 0.629 0.653 0.794 7.952

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1914-1931. All regressions include state of birth fixed effects, year of birth x census year dummies, census division of birth x birth year dummies, gender, race, and Pre-1920, During, and After dummies

interacted with average state latitude and 1920 state-level female and black proportions. 1(Worked at least 40 weeks)

is conditional on having worked in the past year. sinh -1

(Income) takes the inverse hyperbolic sine of total income,

including zeros for those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

Figure B4 shows that the slight negative pre-trend that appeared in Panel D of Figure 2 appears

to be a random fluctuation – when we extend birth years back to 1912, the trend flattens out.

Figure B4: Year-by-Year Effects of Salt Iodization on Total Income, Including Birth Years Back to 1912

Notes: Each point represents the coefficient estimate (and 90% confidence interval) for Goiter Rate interacted with the birth year indicator listed on the x-axis. This regression uses the 1950 to 1980 censuses, restricting to individuals born in 1912-1931. All regressions include state of birth fixed effects, year of birth x census year dummies, census division of birth x birth year dummies, gender, race, and During, After, and Pre-1920 dummies interacted with average state latitude and 1920 state-level female and black proportions.

Table B3 addresses the concern that mean reversion could be driving our results, by allowing

for differential cohort trends across areas with varying levels of the dependent variable at baseline.

To allow for these trends, the ideal strategy would involve including interactions between our Dur-

ing and After dummies and state-level averages of each of our dependent variables from a baseline

year. In our case, 1920 is the obvious choice, as this is the year of the census that immediately

preceded salt iodization. Unfortunately, the 1920 census did not collect data on almost all of our

dependent variables, with the exception of labor force participation. Therefore, for the majority of

our regressions, we use 1920 state-level averages of the closest proxy available. To proxy for em-

ployment and weeks worked, we use labor force participation. For income, we use the occupation

score, a score given to each individual based on their occupation, where higher scores represent

occupations with higher median income (based on 1950’s income and occupation data). Because

we run our regressions separately for men and women, we calculate these state-level averages by

gender and use the relevant gender-specific values in each gender-specific regression. In Table B3,

it is clear that including these mean reversion controls (1920 state-level averages interacted with

During and After dummies) has no effect on our main results.

Table B3: Effects of Salt Iodization on Labor and Income Outcomes, By Gender, Accounting for Mean Reversion

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0113*** 0.0128*** 0.0128** 0.150***

(0.00374) (0.00413) (0.00558) (0.0510)

During x Goiter Rate 0.00585 0.00662 0.0212*** 0.0240

(0.00389) (0.00421) (0.00467) (0.0392)

Observations 1236420 1236420 606704 1108650

Mean of Dep. Var. 0.429 0.448 0.659 5.586

Panel B: Males

After x Goiter Rate 0.00247 0.000325 0.00582 0.0296

(0.00411) (0.00382) (0.00459) (0.0185)

During x Goiter Rate 0.00118 -0.0000895 0.000862 0.0127

(0.00269) (0.00279) (0.00395) (0.0229)

Observations 1146723 1146723 930299 1026746

Mean of Dep. Var. 0.872 0.905 0.867 10.38

Panel C: Female-Male Difference

After x Goiter Rate 0.00884 0.0124** 0.00702 0.120**

(0.00578) (0.00518) (0.00650) (0.0514)

During x Goiter Rate 0.00467 0.00671 0.0204*** 0.0113

(0.00489) (0.00532) (0.00660) (0.0400)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1920-1931. All regressions include state of birth fixed effects, year of birth x census year

dummies, census division of birth x birth year dummies, gender, race, and During and After dummies interacted

averages of the dependent variable (or its closest available proxy) with During and After dummies.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh-1(Income)

The Dust Bowl was another event that roughly coincided with the introduction of iodized salt.

In the 1930’s, large dust storms in the Great Plains states caused massive soil erosion and reduced

agricultural land values in highly eroded areas (Hornbeck, 2012), which could have resulted in

differential cohort trends that we might be incorrectly attributing to salt iodization. Because the

negative effects of the Dust Bowl were felt almost exclusively by a limited number of geograph-

ically concentrated states, we can check to see whether this event was a potential confounder by

excluding these states from our sample and repeating our analysis. Table B4 reports the results

of this exercise, excluding individuals born in the Great Plains and neighboring states (Arkansas,

Colorado, Iowa, Kansas, Louisiana, Minnesota, Missouri, Montana, Nebraska, New Mexico, North

Dakota, Oklahmoa, South Dakota). These results lead us to the same conclusions as before, imply-

ing that the Dust Bowl was not the reason for the results we find.

Table B4: Effects of Salt Iodization on Labor and Income Outcomes, By Gender, Excluding Dust Bowl States

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0116*** 0.0137*** 0.0163* 0.217***

(0.00402) (0.00434) (0.00807) (0.0512)

During x Goiter Rate 0.0102** 0.0105** 0.0202*** 0.0532

(0.00428) (0.00456) (0.00644) (0.0522)

Observations 938025 938025 460665 840341

Mean of Dep. Var. 0.435 0.455 0.667 5.660

Panel B: Males

After x Goiter Rate 0.00234 0.000359 0.00688 0.0207

(0.00487) (0.00474) (0.00484) (0.0147)

During x Goiter Rate 0.000571 -0.000963 0.00102 0.0104

(0.00367) (0.00383) (0.00498) (0.0312)

Observations 869965 869965 704195 778417

Mean of Dep. Var. 0.870 0.904 0.867 10.38

Panel C: Female-Male Difference

After x Goiter Rate 0.00928 0.0134** 0.00939 0.197***

(0.00612) (0.00544) (0.00891) (0.0527)

During x Goiter Rate 0.00961 0.0115* 0.0192* 0.0428

(0.00618) (0.00675) (0.00943) (0.0606)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1920-1931 and excluding individuals born in Arkansas, Colorado, Iowa, Kansas, Louisiana,

Minnesota, Missouri, Montana, Nebraska, New Mexico, North Dakota, Oklahmoa, South Dakota, Texas, and

Wyoming. All regressions include state of birth fixed effects, year of birth x census year dummies, census division of

birth x birth year dummies, gender, race, and During and After dummies interacted with average state latitude and

1920 state-level female and black proportions. 1(Worked at least 40 weeks) is conditional on having worked in the past year. sinh-1(Income) takes the inverse hyperbolic sine of total income, including zeros for those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

To address concerns that our results are being driven by differential trends across the North-

South divide, we drop states below the Mason-Dixon line (all Southern census divisions) and repeat

our analysis. As the results below show, our pattern of results remains the same.

Table B5: Effects of Salt Iodization on Labor and Income Outcomes, By Gender, Excluding States Below Mason-Dixon Line

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0103*** 0.0106** 0.0149** 0.106

(0.00368) (0.00401) (0.00653) (0.0626)

During x Goiter Rate 0.00528 0.00574 0.0184*** -0.0235

(0.00443) (0.00461) (0.00583) (0.0528)

Observations 782490 782490 390240 705656

Mean of Dep. Var. 0.433 0.452 0.667 5.619

Panel B: Males

After x Goiter Rate 0.00508 0.00180 0.00835 0.0424*

(0.00461) (0.00394) (0.00601) (0.0218)

During x Goiter Rate 0.000867 -0.00279 0.00631 0.00630

(0.00195) (0.00212) (0.00464) (0.0291)

Observations 739070 739070 613373 665189

Mean of Dep. Var. 0.882 0.915 0.876 10.52

Panel C: Female-Male Difference

After x Goiter Rate 0.00518 0.00878 0.00659 0.0634

(0.00618) (0.00540) (0.00932) (0.0566)

During x Goiter Rate 0.00441 0.00853* 0.0121 -0.0298

(0.00468) (0.00490) (0.00807) (0.0579)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions use the 1950 to 1980 censuses, restricting

to individuals born in 1920-1931, dropping all individuals born in the South Atlantic, East South Central, and West

South Central divisions. All regressions include state of birth fixed effects, year of birth x census year dummies,

census division of birth x birth year dummies, gender, race, and During and After dummies interacted with average

state latitude and 1920 state-level female and black proportions. sinh -1

(Income) takes the inverse hyperbolic sine of

total income, including zeros for those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

For our interpretation of results to be correct, it needs to be the goiter rate in an individual’s

state of birth that affects cohort trends, not the goiter rate in the state of residence. To verify that

this is the case, we run our exact same specification with two additional variables of interest: goiter

in the individual’s state of residence interacted with during and after dummies (along with state

of residence fixed effects). For those who still live in their state of birth, these variables provide

no more additional information, but for those who have migrated, these variables help us identify

whether the main results discussed above are being driven by the characteristics of an individual’s

state of residence (rather than the state of birth). As Table B6 shows, the effects are clearly being

driven by goiter in the state of birth, and not the state of residence (for which coefficients are small,

imprecisely estimated, and often of the opposite sign).

Table B6: Effects of Salt Iodization on Labor and Income Outcomes, By Gender, Controlling for State-of-Residence Goiter Rate

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0122** 0.0123** 0.0157** 0.175***

in State of Birth (0.00472) (0.00494) (0.00625) (0.0508)

During x Goiter Rate 0.00771 0.00859* 0.0232*** 0.0641

in State of Birth (0.00483) (0.00494) (0.00510) (0.0443)

After x Goiter Rate 0.000170 0.00219 0.00119 -0.00968

in State of Residence (0.00353) (0.00350) (0.00465) (0.0355)

During x Goiter Rate -0.00236 -0.00281 -0.000631 -0.0446

in State of Residence (0.00291) (0.00305) (0.00371) (0.0378)

Observations 1188488 1188488 579715 1060779

Mean of Dep. Var. 0.426 0.446 0.658 5.569

Panel B: Males

After x Goiter Rate 0.00350 0.0000201 0.00583 0.00825

in State of Birth (0.00446) (0.00411) (0.00503) (0.0222)

During x Goiter Rate 0.00181 -0.00142 0.00251 -0.0109

in State of Birth (0.00277) (0.00282) (0.00423) (0.0261)

After x Goiter Rate -0.00281 -0.000300 -0.00105 0.0227

in State of Residence (0.00243) (0.00194) (0.00339) (0.0180)

During x Goiter Rate -0.00229 0.000534 -0.00384 0.0193

in State of Residence (0.00155) (0.00159) (0.00310) (0.0136)

Observations 1101223 1101223 886967 981453

Mean of Dep. Var. 0.870 0.904 0.865 10.36

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth or state of residence (as specified) from Love and Davenport (1920), scaled by the difference between the 75th and 25th percentile of the goiter distribution (0.71). After is a dummy

equal to 1 for those born 1928-1931. During is a dummy equal to 1 for those born 1924-1927. These regressions use

the 1950 to 1980 censuses, restricting to individuals born in 1920-1931. All regressions include state of birth fixed

effects, state of residence fixed effects, year of birth x census year dummies, census division of birth x birth year

dummies, gender, race, and During and After dummies interacted with average state latitude and 1920 state-level

female and black proportions. sinh -1

(Income) takes the inverse hyperbolic sine of total income, including zeros for

those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

In our main specification, due to the use of three different birth cohorts (before, during, and

after), we end up comparing individuals of very different ages. For example, in the 1950 census,

the after cohort is aged 19-22 while the before and during cohorts are in their mid- to late-twenties.

This is one reason why we include birth year by census wave fixed effects in all of our regressions,

but we also conduct a robustness check to ensure that our results are not an artifact of the vastly

different ages of our cohorts. To do this, we use only the 1950 census for the before group (when

they are aged 27 to 30) and the 1960 census for the during and after groups (when they are aged 29

to 36), and repeat our analysis. As Table B7 shows, our conclusions remain the same: strong labor

force participation effects for women, but no effects for men.

Table B7: Effects of Salt Iodization on Labor and Income Outcomes, By Gender, Restricting to Individuals Aged 27-36

(1) (2) (3) (4)

Panel A: Females

After x Goiter Rate 0.0276*** 0.0319*** 0.0269 0.361***

(0.00733) (0.00729) (0.0186) (0.0793)

During x Goiter Rate 0.0222*** 0.0225** 0.0185 0.234**

(0.00819) (0.00885) (0.0163) (0.115)

Observations 158098 158098 47815 113298

Mean of Dep. Var. 0.324 0.340 0.564 4.185

Panel B: Males

After x Goiter Rate -0.00203 0.00559 0.0101 -0.0833

(0.00588) (0.00520) (0.00920) (0.0500)

During x Goiter Rate 0.00115 0.00310 0.0153* -0.0936*

(0.00548) (0.00452) (0.00887) (0.0533)

Observations 147207 147207 104152 107678

Mean of Dep. Var. 0.915 0.949 0.872 10.37

Panel C: Female-Male Difference

After x Goiter Rate 0.0296*** 0.0263*** 0.0168 0.444***

(0.00972) (0.00898) (0.0241) (0.0905)

During x Goiter Rate 0.0211* 0.0194* 0.00314 0.327**

(0.0112) (0.0101) (0.0197) (0.137)

Notes: Standard errors, clustered by state of birth, in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Goiter Rate is the

goiter rate in the individual's state of birth from Love and Davenport (1920), scaled by the difference between the

75th and 25th percentile of the goiter distribution (0.71). After is a dummy equal to 1 for those born 1928-1931.

During is a dummy equal to 1 for those born 1924-1927. These regressions restrict to individuals born in 1920-1931,

using the 1950 census for those born 1920-1923 and the 1960 census for those born 1924-1931. All regressions include

state of birth fixed effects, year of birth x census year dummies, census division of birth x birth year dummies,

gender, race, and During and After dummies interacted with average state latitude and 1920 state-level female and

black proportions. 1(Worked at least 40 weeks) is conditional on having worked in the past year. sinh -1

(Income)

takes the inverse hyperbolic sine of total income, including zeros for those not working.

1(Employed) 1(Participated in

the Labor Force)

1(Worked at least

40 weeks) sinh

-1 (Income)

B3 Data Appendix

B3.1 Independent Indicator Variables

• Before=1 if individual was born in 1920-1924; Before=0 otherwise

• During=1 if individual was born in 1924-1927; During=0 otherwise

• After=1 if individual was born in 1928-1931; After=0 otherwise

• Pre-1920=1 if individual was born before 1920; Pre-1920=0 otherwise

• Goiter Rate: goiter rate, from Love & Davenport (1920), in the individual’s state of birth

B3.2 Outcome Variables

B3.2.1 Basic Outcomes

• 1(Participated in Labor Force)=1 if the individual participated in the labor force (by either

working or looking for work) in the last week;

1(Participated in Labor Force)=0 if the individual did not participate in the labor force.

• 1(Employed)=1 if the individual was employed in the last week;

1(Employed)=0 if the individual did not work, irrespective of whether the individual looked

for a job.

• 1(Worked at least 40 weeks)=1 if the individual reported working at least 40 weeks in the last

year. 1(Worked at least 40 weeks)=0 if the individual worked fewer than 40 weeks in the last

year. This variable is missing for individuals who did not work in the past year.

• Total Income: This is the inverse hyperbolic sine of the total annual income earned by the

individual. All values are adjusted to 1999 prices according to Census-provided multipli-

ers. Although this variable is top-coded differently across Census years, we applied the top-

coding from the 1950 Census to all years (setting the maximum income to $70,000). The

inverse hyperbolic sine transformation was made after all of the adjustments were made.

• Years of Schooling: Years of schooling completed by the individual.

• 1(Ever Married)=1 if the individual is currently married, divorced, separated, or widowed.

1(Ever Married)=0 if the individual has never been married.

• Age at First Marriage: Individual’s age at first marriage, conditional on being ever-married.

• Spouse’s Years of Schooling: Years of schooling completed by the individual’s spouse, for

individuals who can be matched to their spouse living in the same household.

• Spouse’s Income: This is the inverse hyperbolic sine of the total annual income earned by

the individual’s spouse, for individuals who can be matched to their spouse living in the

same household. All values are adjusted to 1999 prices according to Census-provided mul-

tipliers. Although this variable is top-coded differently across Census years, we applied the

top-coding from the 1950 Census to all years (setting the maximum income to $70,000). The

inverse hyperbolic sine transformation was made after all of the adjustments were made.

• Family Income: This is the inverse hyperbolic sine of the sum of total annual income earned

by the individual’s family (living in the same household). All values are adjusted to 1999

prices according to Census-provided multipliers. Although this variable is top-coded dif-

ferently across Census years, we applied the top-coding from the 1950 Census to all years

(setting the maximum income to $70,000). The inverse hyperbolic sine transformation was

made after all of the adjustments were made.

B3.3 Other Variables

• female=1 for females; female=0 for males

• black=1 for black individuals; black=0 for all other races

• 1920 female share: The state-level proportion of the population that was female in the indi-

vidual’s state of birth, calculated from the 1920 Census.

• 1920 black share: The state-level proportion of the population that was black in the individ-

ual’s state of birth, calculated from the 1920 Census.

• Latitude: Average latitude for the individual’s state of birth

• Malaria: Pre-iodization malaria rate in state of birth according to Bleakley (2010)

• Hookworm: Pre-iodization hookworm rate in state of birth according to Bleakley (2010)

• Compulsory Schooling: Number of years of schooling required by compulsory attendance

laws in the individual’s state of birth in the year they turned 14 (which is the minimum leaving

age across all states and years), from Lleras-Muney (2002).

• Mobilization Rate: WWII mobilization rate for individual’s state of residence, from Ace-

moglu et al. (2004).

• 1930 unemployment rate: The state-level unemployment rate in the individual’s state of birth,

calculated from the 1930 census.

• 1920-1940 population growth: The state-level change, between the 1920 and 1940 census,

in the share of the U.S. population living in an individual’s state of birth.

• 1920-1940 change in black share: The state-level change, between the 1920 and 1940 census,

in the black population share in an individual’s state of birth.