## Abstract

This article uses the Bangladesh famine of 1974 as a natural experiment to estimate the impact of intrauterine malnutrition on sex of the child and infant mortality. In addition, we estimate the impact of malnutrition on post-famine pregnancy outcomes. Using the 1996 Matlab Health and Socioeconomic Survey (MHSS), we find that women who were pregnant during the famine were less likely to have male children. Moreover, children who were in utero during the most severe period of the Bangladesh famine were 32 % more likely to die within one month of birth compared with their siblings who were not in utero during the famine. Finally, we estimate the impacts of the famine on subsequent pregnancy outcomes. Controlling for pre-famine fertility, we find that women who were pregnant during the famine experienced a higher number of stillbirths in the post-famine years. This increase appears to be driven by an excess number of male stillbirths.

## Introduction

Famines have been associated with increased mortality in several regions of the world and in many different periods. Although this association has been documented for all age groups, increased rates of mortality have been especially striking among infants. For instance, Lindeboom et al. (2010) estimated that approximately 25 % of children younger than 1 died during the Dutch Potato Famine of 1847–1864. Infant mortality rates also increased substantially during the Finnish famine of 1866–1868, the 1941–1944 siege of Leningrad, and the World War II Dutch famine (Hart 1993; Kannisto et al. 1997).

Large-scale famine events have also offered researchers the opportunity to study the impact of intrauterine nutrition on the health and survival probabilities of infants (Almond et al. 2008; Lumey and Stein 1997; Razzaque et al. 1990; Stein and Susser 1975). This is particularly important in light of Barker’s (1990) “fetal origins” hypothesis, which argues that in utero malnutrition adversely impacts the health of the fetus and leads to increased risk for future diseases (Almond and Currie 2011). Although serious famines are relatively rare, they enable researchers to circumvent factors such as household environment and socioeconomic background that might confound the results of studies relying on self-reports of maternal nutrition. Thus, they present a unique method of analyzing the impacts of fetal nutrition on child and adult outcomes.

Beyond their effects on health and economic outcomes, famines have also been hypothesized to increase the ratio of female-to-male births. Such a link can be interpreted in light of the Trivers-Willard (Trivers and Willard 1973) hypothesis that male infants will fare worse when parental conditions, such as health, are bad. From an evolutionary perspective, the reproductive advantage that females have during bad conditions enables their mothers to maximize their number of grandchildren (Rosenfeld and Roberts 2004). If mothers in good conditions are able to have high-quality sons, then they can obtain more grandchildren through sons rather than daughters given that those sons would be able to acquire multiple mates. In contrast, mothers in bad conditions will obtain more grandchildren through daughters than through sons (Almond and Edlund 2007; Cameron 2004).

Although this pattern has been observed among some nonhuman mammals (Huck et al. 1986; Meikle and Drickamer 1986; Rosenfeld and Roberts 2004), its existence among human populations is controversial, with studies continuing to provide mixed evidence (Cameron 2004). For example, Almond and Edlund (2007) found that married mothers in the United States are more likely to give birth to a son compared with unmarried mothers, and also that poorly educated women are more likely to give birth to daughters than more highly educated women. Further supporting this hypothesis, Almond et al. (2008) found an excess number of females among middle-aged Chinese cohorts who had been exposed in utero to the Chinese famine. In contrast, neither Stein et al. (2004) nor Anderson and Bergstrom (1998) found an association between intrauterine exposure to malnutrition and an excess of female births.

Even when researchers have found an impact on the sex ratio, it has been difficult to pin down the mechanism for this adjustment (Cameron 2004). One possibility is that prenatal determination of an infant’s sex could lead to selective abortions, thus changing the sex ratio at birth. However, without the ability to determine the sex of the infant, higher rates of infant male mortality is another mechanism that may lead to an imbalance in the sex ratio. Consistent with this argument, Almond and Edlund (2007), using U.S. data, found that being married lowered the probability that a deceased infant was male. It is not clear, however, whether similar patterns emerge in the context of a developing country, particularly in one where preference for sons is pronounced.

Intrauterine malnutrition has also been shown to impact long-term reproductive outcomes. Despite the fact that no association was found between intrauterine malnutrition and completed fertility, age at first pregnancy, or child spacing, Lumey and Stein (1997) provided evidence that women who were in utero during the Dutch famine had a higher risk of experiencing stillbirths and perinatal deaths later in life, especially those who were affected in the third trimester. More recently, Almond et al. (2008) provided evidence that intrauterine nutritional deprivation is linked to the reproductive outcomes of the next generation. Specifically, they found that children of those mothers who were exposed in utero to the Chinese famine were less likely to be male. Existing studies, however, have not examined whether exposure to malnutrition during pregnancy will also impact the future outcomes of the mother herself in addition to those of her exposed offspring.

Although this specific hypothesis has not been previously investigated, studies have documented a negative link between episodes of eating disorders and future pregnancy outcomes. Bulik et al. (1999), for instance, found that women with a history of anorexia nervosa had significantly more miscarriages and were more likely to deliver premature babies compared with a control group of women with no previous history of eating disorders. These findings are consistent with the findings of retrospective studies that relied on women’s own reporting of nutrition and pregnancy outcomes (Katz and Vollenhoven 2000).

Despite this growing literature, some important questions remain unanswered, and others need to be reexamined. First, do women exposed to malnutrition during pregnancy have a higher probability of a female live birth? Second, conditional on being a live birth, do male infants exposed to fetal malnutrition have a higher probability of early mortality? We use the Bangladesh famine of 1974 (henceforth, “the famine”) as a natural experiment to answer these questions. Although Razzaque et al. (1990) also analyzed mortality outcomes in the context of the Bangladesh famine, our analysis compares the outcomes of siblings, only one of whom was exposed to the famine. This within-family analysis is an important methodological contribution because it controls for all time-invariant confounding factors at the family level; and unlike existing studies, it does not rely on geographical variation in the intensity of famines or on comparing children exposed at different stages of gestation.1 We also examine a third question: Does severe undernutrition during pregnancy impact the reproductive outcomes of exposed women in future pregnancies after the famine event is over? In particular, are women who experience a famine while pregnant more likely to have future miscarriages and stillbirths? Do these reproductive outcomes vary by the sex of the child? Despite the importance of understanding whether nutritional deprivation during pregnancy impacts future reproductive outcomes, this is the first study that uses a natural experiment to link maternal malnutrition with subsequent fertility outcomes.

## Famine Background

Bangladesh, formerly a part of India (known as East Bengal, and then East Pakistan), became independent in 1971. While part of the Indian colony, Bangladesh experienced massive death and suffering during the Great Bengal Famine of 1943. Both the India Famine Inquiry Commission (1976) and Sen (1981) studied the causes and severity of this famine. Bangladesh suffered another significant famine in 1974, following a smaller one in 1971. Toward the middle of 1974, severe flooding led to a sharp rise in unemployment, particularly among rural farmers and laborers (Sen 1981). Although food availability remained unaffected until the harvest period later in the year, food prices began to rise immediately, eventually increasing by 500 % (Dyson 1991). Although the precise cause of these increased prices remains unclear, the most likely factors appear to be macroeconomic and include a combination of inflation and speculation about future price increases (Sen 1981). Regardless, the unemployment and dramatic price increases placed a severe burden on rural areas.

The famine began in March of 1974, reaching its peak between July and October. It began to subside by the end of that year, although the impact of high food prices and increased mortality lasted well into 1975 (Alamgir 1980). Figure 1 depicts the price of medium rice in Bangladesh between July 1972 and June 1976 (Alamgir and Salimullah 1977). The nationwide price of rice did not peak until February 1975, returning to pre-famine levels by the end of that year.

The famine and high food prices led to increased mortality rates that persisted through 1976 (Razzaque et al. 1990). Estimates of fatalities range from 450,000 to well over 1 million (Alamgir 1980), which accounts for approximately 0.6 % to 1.3 % of the estimated national population (United Nations Secretariat 2003). Dyson (1991:287) used data from Matlab, the region of particular interest in this study, to plot mortality rates in Bangladesh before, during, and after the famine, showing a peak death rate in early 1975 that remained high well after the famine ended because many remained weakened and sick. In fact, the death rate for both infants and the population overall was higher in 1975 than in 1974.

Although child mortality rates tend to be higher for females than for males in Matlab during normal years (Choe and Razzaque 1990; Fauveau and Chakraborty 1994; Langsten 1981), the famine resulted in significant but similar increases in child mortality rates among male and female children (Bairagi 1986; Koenig and D’Souza 1986). Moreover, fertility rates declined by about 34 % during 1974–1975, before increasing by 17 % in the post-famine years, thereby partially offsetting the famine’s effect (Razzaque 1988).

## Data

The 1996 Matlab Health and Socioeconomic Survey is a cross-sectional data set sampling 4,364 households in Matlab district, a poor rural and agricultural area of Bangladesh approximately 55 km southeast of Dhaka (Rahman et al. 1999). These households cover 2,687 baris (groups of households living and working together, sharing a common outdoor space), which account for a one-third random sample of all Matlab baris. All women aged 15 and older in the sample were asked about their fertility history, including any subsequent mortality outcomes for their children. From this information, we are able to gather the date, sex, birth outcome (miscarriage, stillbirth, or live birth), and neonatal and infant mortality outcomes for 24,916 children born between 1919 and 1996. We then supplement this information on births with demographic characteristics of the mother, including her age at the pregnancy outcome, age at first marriage, years of education (in years), adult height (in centimeters), and the number of male and female children she had prior to 1974. We also use information about the relative size of the infant at birth compared with other infants (i.e., much bigger, bigger, smaller, or the same size); the number of prenatal visits that the mother had during a given pregnancy; whether the infant was breast-fed; whether the child was born in a hospital, a clinic, or at home; and whether there was a skilled care provider present at the time of birth. These variables are important to account for in the empirical analysis because they could have an impact on the health and survival rates of the infants and may be correlated with famine exposure.

The reliability of the women’s fertility histories depends on their ability to recall every fertility event, but the events are cross-checked against birth records in the vital events database of Matlab. Thus, the mortality records should be unaffected by any recall bias. However, in terms of prenatal outcomes that would not be recorded in the vital events database (such as miscarriages or stillbirths), a small possibility remains that women are intentionally biased in the recall of their fertility history toward (or against) remembering stillborn male children compared with stillborn females, or they may have better recall for instances that happen near dramatic events, such as the famine. We discuss the potential implications of any such recall bias later in our Results section.

Our goal is to examine the effects of the 1974 famine on birth outcomes and subsequent fertility outcomes. Examining the prices of rice in Fig. 1 suggests that the most severe period of the famine occurred between August 1974 and October 1975, when prices were more than 50 % higher than pre-famine levels. Thus, we first compare the birth outcomes for all live births occurring between September 1974 and December 1975 to live births occurring in other periods between 1970 and 1980.2 Children born within the treatment window of September 1974–December 1975 were in utero during the most severe part of the famine for at least one full month of the third trimester. A child born in September 1974 would have experienced the famine for the full month of August 1974. Assuming an average gestation length of 38 weeks (266 days), a child born in December 1975 would have experienced the famine for at least one full month of the third trimester. Thus, we exclude from this initial treatment group those infants who were exposed to the famine only in the first trimester as well as those infants exposed only in the first and second trimesters. This definition of the treatment is guided by the famine literature that has provided evidence that malnutrition may be particularly harmful to fetuses during the last trimester (Lumey and Stein 1997). The timeline in Fig. 2 shows the span of this treatment as well as alternate treatment windows, which we discuss in detail in The Empirical Model section.

Our outcomes of interest include whether the child is male, whether the child died during the first 29 days of life (neonatal mortality), and whether the child died during the 1–12 months after birth (postneonatal mortality). Table 1 presents the mean values of these key outcomes for children who were in utero during the famine, the means for children who were in utero in other periods, and the difference between them. Compared with nonaffected children, children in utero during the famine had significantly higher mortality rates in the first month after birth but had similar postneonatal mortality rates compared with non-affected children. Children exposed in utero are less likely to be male, although this difference is not statistically significant. The rates of miscarriages and stillbirths, however, are not statistically different between children who were in utero during the famine and children who were not. There is no other statistically significant difference between children on any other observable characteristic, such as mother’s age at first marriage or mother’s age at birth.

To determine whether the famine had a lasting impact on subsequent fertility, Table 2 compares key post-famine fertility outcomes of women who became pregnant after a long period of exposure to the famine (had a live birth, stillbirth, or miscarriage between January 1976 and July 1976) with all other women aged 15 and older who were not pregnant during the famine. Women who experienced a pregnancy outcome between January 1976 and July 1976 would have become pregnant after at least nine months of exposure to the famine but before the famine was over. Thus, in contrast to the immediate effects we expect to see on children in utero, we hypothesize that malnutrition may take longer to affect the adult mothers themselves.

After separating women who were fertile during the famine according to whether they became pregnant before the famine ended but still following at least nine months of famine exposure, we then examine their post-famine fertility outcomes for every pregnancy outcome that occurred after 1977. Limiting the post-famine pregnancy outcomes to only those occurring after 1977 ensures that these outcomes are not affected by any direct impacts of the famine itself, which had been over for one year by that time. Specifically, we are interested in whether the post-famine pregnancies resulted in miscarriages or stillbirths. As shown in Table 2, we find little evidence that women who became pregnant after exposure to the famine were more likely to have a post-famine miscarriage. In contrast, the simple mean comparison suggests that women who were pregnant during the famine were more likely to have a stillbirth in a future, post-famine pregnancy and particularly more likely to have a male stillbirth.

The descriptive statistics also suggest that women who were pregnant during the famine are positively selected on observable characteristics: they have higher educational levels, are married to more-educated husbands, and married at a younger age. Moreover, they had a larger number of sons before the famine. Although our empirical specifications control for these factors, we present results that decompose the sample by spouse’s education and landholdings. This analysis allows us to examine whether the results vary by socioeconomic background.

## The Empirical Model

### Children Specifications

We start by estimating the following probit regression for children born between 1970 and 1980:3
$Ci=α+β1Faminei+β2YOBi+Xiδ+εi,$
(1)
where Ci is the outcome of infant i. To investigate the impact of exposure to the famine on the sex of the child, Ci takes the value of 1 if the child is male and 0 otherwise. For neonatal mortality, Ci takes the value of 1 if the child died within the first 29 days after birth and 0 otherwise; and for postneonatal mortality, it takes the value of 1 if the child died during the first 1–12 months of life and 0 otherwise. Faminei is an indicator variable that equals 1 if the child was exposed to the famine while in utero (i.e., was born between September 1974 and December 1975).

The vector X includes demographic characteristics of the mother, as listed in the Data section. It also contains variables controlling for the relative birth weight of the infant and measures of prenatal and postnatal care, again as described in the Data section.4 In addition, we include in vector X controls for season of birth and whether the mother’s village is part of the treatment group of a Maternal and Child Health and Family Planning services program operating in the area.5 Furthermore, including a linear year of birth trend, YOBi, ensures that β1 measures the difference in outcomes for children born in the treatment window separate from the cohort trend (Almond 2006). Standard errors are clustered at the bari level to account for any serial correlation.

Women who were pregnant during the famine, however, are likely to be different from women who did not get pregnant during the famine on some important unobservable dimensions. Because many of the women in our sample gave birth to more than one child during the period 1970–1980, we are able to include mother fixed effects, mj, in Eq. (1) and estimate the following OLS regression:
$Cij=α+β1Famineij+β2YOBij+Xijδ+mj+ηij,$
(2)
where Cij is the outcome of infant i born to woman j. The vector X in Eq. (2) includes similar variables included in Eq. (1), with the exception of mother’s education, age at first marriage, and height. To ensure that our results are not spurious, we also report results from estimating Eqs. (1) and (2) using births in 1960–1970 and in 1980–1990; we assign 1964–1965 and 1984–1985, respectively, as placebo treatment years.

### Maternal Specification

The impact of exposure to the famine on future fertility outcomes is estimated from the following probit regression:
$Mij=α+β1FamMotherj+β2YOBi+Zjγ+Xijδ+εij,$
(3)
where M is an indicator variable that equals 1 if a post-famine pregnancy i resulted in a miscarriage or stillbirth for woman j, and 0 otherwise. Our main treatment group in this specification differs from the one that we considered for the child’s outcomes. Specifically, the indicator variable FamMother takes the value of 1 if the mother had any pregnancy outcome between January 1976 and July 1976 (miscarriage, stillbirth, or live birth). This birth window implies that the mother conceived between April 1975 and October 1975 and would have been exposed to at least nine months of the famine before becoming pregnant, assuming an average gestation length of 38 weeks. Thus, we compare the likelihood of a miscarriage and stillbirth for women who became pregnant after a lengthy period of exposure to the famine with women who did not become pregnant after their exposure to malnutrition. The vector X includes controls similar to those included in Eq. (1), with the exception of prenatal and postnatal healthcare indicators, which cannot be included in a sample of pregnancy outcomes that are not all carried to term. In addition to the total number of pre-famine (pre-1974) live male and female births for woman j included in X, vector Z includes post-famine live births by sex (post-1976) and spouse’s education. Later in the article, we reproduce the results using different periods of exposure to the famine. Because FamMother does not vary by pregnancy outcome, we cannot include mother fixed effects in this specification.

## Results

### Children Findings

Table 3 presents the estimation results for Eqs. (1) and (2). Looking at the implied marginal effects, children born between September 1974 and December 1975 are 2 percentage points less likely to be male (column 2), although this estimate is not significant at conventional levels. As one would expect, because the sex of a child can be considered random, the coefficient estimates for most of the controls listed in the previous section (not reported in Table 3) are small and not statistically significant.

Columns 4 and 7 of Table 3 show estimation results of Eq. (1) using infant mortality at one month, and between one month and one year as the outcomes. Marginal effects evaluated at the mean are shown in columns 5 and 8. Children born during the treatment period are 2 percentage points more likely to die within their first month after birth (a 32 % increase over the mean) but have no difference in life expectancy between one month and one year. Consistent with the findings of Razzaque et al. (1990), exposure to the famine in utero increased the probability of neonatal mortality. Although not presented in this article, the controls in the regressions have expected signs. For instance, mother’s age at birth is negatively correlated with mortality within one month of birth, and children who were reported as being smaller than average at birth have a higher likelihood of mortality.

Perhaps mothers who opt to get pregnant during famine periods are unobservably different from mothers who do not. For instance, maybe more cautious mothers who are concerned about their potential child’s health avoid becoming pregnant during food shortages.6 To control for such traits, we include mother fixed effects in the regressions in columns 3, 6, and 9, thereby limiting the sample to children whose mothers had both a famine-affected birth and a birth that was not affected within the sample period. The infant mortality results remain similar in both magnitude and significance to our original probit models when mother fixed effects are not included, which gives us confidence that the natural experiment is indeed valid. However, in contrast to the findings of Razzaque et al. (1990), the fixed-effects specification in column 3 indicates a clear and significant decrease in the probability of male birth. Specifically, women who were pregnant during the famine are 4 % less likely to have a male birth compared with women who were pregnant during the surrounding years (significant at the 5 % level). Thus, our results provide evidence in favor of the Trivers-Willard hypothesis that women are less likely to give birth to sons during lean times.

Although the mother fixed effects capture relevant time-invariant characteristics of the mother, one lingering concern may be the possibility of nonrandom migration by males during the famine. We are unable to observe the presence of the father or to account for the pregnancies and births that did not take place because the father was absent. However, Kuhn (2005) found that single men in Matlab were most likely to migrate for work compared with married men and men who own less land. We would be most concerned about this type of selection bias if our data showed larger famine effects on children from families with fewer landholdings. In a later section of this article, we show that wealthier families faced the largest impacts during the famine, which leads us to conclude that selective migration is likely not a significant mechanism driving our results. Another concern could be that the famine changed access or availability of family planning methods. Although possible, it is unlikely in the Bangladeshi context. Access to family planning clinics and availability of contraceptives was extremely limited in Matlab prior to 1978 (Janowitz et al. 1997), and the government-run clinics providing counseling on family planning were mostly run by male workers and were often dirty and unsterile (Joshi and Schultz 2007).

To test whether our results are spurious, we conduct two sets of placebo tests. We estimate Eqs. (1) and (2) using the same outcomes as the previous regressions but redefining the treatment and sample around the years 1964 and 1984. Neither 1964 nor 1984 were affected by famine, war, monsoon, or other catastrophic events. The sample for the 1964 placebo test includes births between 1960 and 1970. For the 1984 placebo test, the sample includes births between 1980 and 1990. The results of these analyses are presented in Table 3. Unlike our results for the famine cohort, we find no significant difference between the likelihood of a male birth in the placebo treatment year and the other birth years in the samples. In addition, estimates from the infant mortality regressions remain small and insignificant, further supporting our findings that the 1974–1975 famine is the driving factor behind the increased infant mortality observed in our main results. We do not believe that recall bias is a concern given that our outcomes of interest are verifiable using the vital statistics registries.

As an alternative to placebo years, we also estimate each of our main specifications after incorporating decade-of-birth dummy variables interacted with the famine treatment window, using the full record of births covering 1919–1996. Results are qualitatively similar to our findings for the placebo years and are available upon request. We also estimate a regression that limits the sample to births between 1974 and 1980 given that Bangladesh suffered a war and a lesser famine from 1970–1973. Results from those regressions are nearly identical to those presented here and are also available upon request. In other results not presented, we estimate whether the famine increased the likelihood of miscarriage or stillbirth and find a small (2 %) increase in the likelihood of miscarriage but no relationship with stillbirths.

Table 4 presents infant mortality results separately for male and female infants. The coefficient estimates suggest that the results from Table 3 are in fact driven by male mortality. The results given in columns 2 and 3 imply that male infants who were exposed to the famine in utero were 3 to 4 percentage points more likely to die within the first month after birth. Female infants, on the other hand, do not have significantly different survival rates associated with famine exposure. The estimates using 1964 or 1984 as placebo famine years do not indicate any positive relationship with infant mortality rates.

In Table 5, we vary the window of treatment to better understand the impact of exposure to malnutrition through the different stages of pregnancy. The first row repeats the main results from Table 3, and the second and third rows alter the length of the treatment, effectively increasing the amount of time that the infant is exposed to the famine while in utero. The treatment window of births in the next row—between December 1, 1974, and October 31, 1975—means that the child would have been exposed to the famine for the full third trimester and at least one month of the second trimester. The treatment window of April 1, 1975, through October 31, 1975, means that the child would have been exposed to the famine during the entire nine months in utero. The results from Table 5 suggest that as the time of exposure to famine lengthens, the adverse impacts of malnutrition become worse but possibly at a decreasing rate. The marginal effects for infant mortality at one month become slightly larger in magnitude when we include possible second-trimester exposure but become statistically insignificant when we examine the impact of being exposed for the full nine months of pregnancy. Moreover, the likelihood of a male birth is consistently negative and statistically significant across the different exposure windows but increases (in absolute value) only when we include possible second-trimester exposure. Thus, the results suggest that male fetuses are especially vulnerable to malnutrition during their first and second trimesters in utero.

Taken together, the results of Tables 3, 4, and 5 suggest that famine exposure significantly decreases the likelihood of a male birth while increasing the likelihood of infant mortality by one month, particularly for male infants. However, the adverse effects of the famine appear to fade as the child reaches age 1: the famine coefficients on the likelihood of death between one month and one year are statistically insignificant.

### Maternal Findings

The previous estimations focused on the effect of the famine on children in utero. However, a famine pregnancy could also affect the mother. The next regressions estimate how experiencing a pregnancy after famine exposure is related to women’s long-term fertility outcomes. The results from estimating Eq. (3) are reported in Table 6. We find no evidence of a relationship between exposure to the famine and future miscarriages. However, the results suggest that becoming pregnant after exposure to famine increases the likelihood of a stillbirth by about 2 percentage points (a 61 % increase over the mean; see column 4). Because it is possible to identify the sex of a stillborn child, we estimate the number of stillbirths separately for male and female stillbirths. Although future female stillbirths are more common among women who became pregnant during exposure to the famine, this difference is not significant. In contrast, male stillbirths (which are always more common than female stillbirths) become significantly even more common among these women.

In a culture with some preference toward males, perhaps this difference is driven by a bias in recall: male stillbirths may be more often remembered than female ones. To alleviate this concern, we estimate the same regression using placebo famines in 1964 and 1984, finding no evidence of an increased likelihood of male stillbirth among these cohorts. Furthermore, we do not believe that any recall bias in the timing of the stillbirths would be correlated with the famine because all these stillbirths took place after the famine. There may be some recall bias on whether the mother experienced a post-famine pregnancy; however, because any such bias would misassign some treated women to the comparison group, it would place only a downward bias on our results.

To check the sensitivity of the results to the treatment window chosen in Table 6, Table 7 presents estimates of Eq. (3), where we vary the definition of exposure to the famine. The first row repeats the original specification from Table 6, where treated women are those who had a pregnancy outcome between January 1, 1976, and July 31, 1976. The next two rows follow the same rule that the woman must become pregnant before the famine is over but reduces famine exposure to six months (pregnancy outcome between October 1, 1975, and July 31, 1976). The third row reduces time of exposure to the famine to three months (pregnancy outcome between July 1, 1975, and July 31, 1976). The effect of the famine on future stillbirths is consistently significant at the 1 % or 5 % level, with a marginal effect around 2 percentage points. Moreover, this effect becomes stronger when the fetus is male. For purposes of comparison, the last row of Table 7 restricts the treatment to the same window used for infants in Table 3. This window does not generate significant results, which suggests that malnutrition resulting from famine may take longer to affect adults than children. This result is consistent with the medical literature on malnutrition among adults (Collins 1995; Davis 1996).

### The Role of Demographic Characteristics

Finally, we investigate whether the results vary by family’s wealth or education level. As reported earlier, the descriptive statistics suggest that women who became pregnant during the famine may have been positively selected on observable characteristics. Unfortunately, our data set does not provide direct information on wealth or income at the time of birth. However, our data include information on landholdings by household in 1996 (the year of the survey). Because land is the largest household asset in Matlab, it provides a good proxy for wealth. Moreover, the inactive nature of land exchanges in the South Asian land market (at least during the period of our data) enables us to make the reasonable assumption that households that reported owning land at the time of the survey were also landholders at the time of their child’s birth (for more detailed descriptions of the land market, see Binswinger and Rosenzweig (1986), Pitt and Khandker (1998) and Rosenzweig and Wolpin (1985)). In fact, several studies have found turnover in landownership to be so low that they have suggested that land may be used as an exogenous variable (e.g., Pitt and Khandker 1998).

Table 8 reports results on the impact of exposure to the famine in utero by landholdings and years of schooling of the husband, another proxy for socioeconomic status. To facilitate the comparison, the first row repeats the results from Table 3. The second and third rows report the results for the top and bottom tercile in landholdings (measured in decimals), respectively. Although the results in column 3 suggest a negative association between exposure to in utero malnutrition and the likelihood of a male birth across the different socioeconomic groups, the results are statistically insignificant, perhaps because of the significantly reduced sample size. The results for infant mortality, however, do vary by landholdings. Specifically, we find that children born to mothers in the top tercile of landholdings are more likely to die within one month of birth compared with children born to mothers in the bottom tercile of the landholdings distribution. This pattern of results is consistent with the finding reported earlier that women who became pregnant during the famine are more educated (when surveyed in 1996) and are married to more-educated husbands compared with women who avoid getting pregnant during the famine. The results for infant mortality by husband’s years of schooling are generally not statistically significant, but the magnitude of the results suggests that children of women with more-educated spouses faced a higher likelihood of mortality, further supporting the idea that women who became pregnant during the famine are positively selected in terms of wealth and education.

We also repeat the analysis of post-famine pregnancy outcomes by landholdings and education. The results, shown in Table 9, suggest that women from households in both parts of the landholdings distribution are more likely to have post-famine stillbirths. Similar to the results in the full sample, the excess stillbirths are mainly driven by males. The magnitude of the results is slightly larger for women in the bottom tercile of the landholdings distribution, but the estimates between the two terciles are not statistically different from each other. Similar patterns are found when we analyze the results by terciles of the education distribution.

## Discussion

Male infants who were exposed to the 1974 Bangladesh famine for at least one full month of their third trimester experienced higher rates of neonatal mortality compared with their siblings who were not exposed. This result is robust to variations in the length of exposure to the famine and is not present in alternative “placebo” famine years. Factors other than malnutrition, however, could have contributed to the increase in neonatal mortality. For instance, women who choose to become pregnant during a famine could possibly have unobserved characteristics that contribute to increased infant mortality. The siblings comparison, however, accounts for time-invariant factors related to the mother or the family that could impact the health of the infant. Moreover, the effects for infant mortality are stronger for women with more landholdings, suggesting positive selection into motherhood during the famine.

Other factors during pregnancy—such as limited access to quality care, increased maternal stress, increased burdens of infectious disease, and selective migration of spouses—also could have impacted infant mortality. However, the results with regard to infant mortality are robust to the inclusion of a set of variables that measure access to health care. Unfortunately, our data contain no information about retrospective stress measures, infectious diseases, or about the presence of the spouse during pregnancy. Thus, we remain unable to definitely rule out these alternative explanations or to document how their impact interacts with the impact of reduced access to food.

Pregnant mothers during the 1974 famine were also less likely to give birth to a son compared with other live births they had during the 1970–1980 period. This result supports the Trivers-Willard hypothesis and contributes to a growing literature about its relevance to human populations. Although we are unable to identify the exact mechanism through which intrauterine malnutrition affects the sex of the infant, the results suggest that male infants are particularly vulnerable to the in utero environment they face. Of course, the generalizability of this finding is limited because it may be driven by specific factors that are unique to Bangladesh. Thus, further research about the sensitivity of male infants to intrauterine factors as well as the mechanisms through which they operate is warranted.

Finally, exposure to the famine during pregnancy affected not only the exposed infants but also the post-famine reproductive outcomes of mothers. Controlling for pre-famine fertility, women who conceived after at least nine months of exposure to the famine experienced a higher likelihood of a future male stillbirth compared with other fertile women who did not become pregnant during the famine. As with the infant-related findings, these results are robust to variations in the length of exposure to the famine and are not present in alternative “placebo” famine years. To our knowledge, this is the first study to document a long-run impact of intrauterine exposure to malnutrition on subsequent pregnancy outcomes. These results, however, should be interpreted cautiously. In contrast to the analysis of infant outcomes, we are unable to account for time-invariant factors about the mother or the household. Instead, we rely on a comparison between women who conceived after a lengthy period of malnutrition and women who did not. Thus, we cannot rule out the possibility that at least part of the effects we document may be related to unobserved traits of the mother that impact her reproductive outcomes. Moreover, miscarriages are typically difficult to identify because they depend on the age at which a pregnancy is recognized, which could also be affected by the famine. Nonetheless, these findings are relevant to the literature on the reproductive outcomes of women with eating disorders as well as for relief agencies operating in areas facing nutritional constraints.

In summary, the results of this article suggest that malnutrition during famine is likely to have an especially adverse effect on both male infants and pregnant women. These impacts may be long-lasting: they appear to affect women not only in their current pregnancy but in their future pregnancies as well. An important area for future research may be to explore the mechanisms through which malnutrition impacts infant mortality rates and future fertility outcomes as well as how malnutrition interacts with other factors, such as access to health and maternal stress. A deeper understanding of the importance of different channels, whether biological or behavioral, would provide insight into how to mitigate the health effects of famine episodes in developing countries.

## Acknowledgments

We are grateful to Daniel Rees, Angela Dills, and Martha Bailey for useful comments on an earlier version of this article.

## Notes

1

There are several other smaller differences between our studies. Razzaque et al. (1990) defined the period of the famine-born as July 1974 through June 1975 and compared the outcomes during that period with those during a famine-conceived period of July 1975–1976 and a non-famine period between July 1976 and March 1977. Our study uses a different definition of the famine—August 1974 to October 1975—based on the months when the price of rice was more than 50 % higher than the pre-famine price. Our study also uses a wider window of time as a control group and has other minor differences, such as our use of a multivariate logistic regression instead of a univariate one.

2

When the specific month of birth was not remembered or unavailable, fieldworkers often coded the birth as occurring in January. Thus, the data include an inaccurately large number of January births. The main specifications in this article are estimated with these January births included. However, all results (available upon request) are qualitatively robust to the exclusion of January births from the sample. In addition, the years 1970–1972 were also associated with bad crops, high food prices, and political turmoil. We estimated similar regressions in which we limited the sample to the years 1974–1980, effectively eliminating the war, earlier famine, and hardship from the control group; we found similar qualitative and quantitative results, which are available upon request.

3

The different empirical specifications follow those used in the literature. See, for example, Almond (2006), Almond et al. (2008), Almond and Mazumder (2011), Camacho (2008) and Mansour and Rees (2012).

4

Birth weight and access to prenatal care are potential channels through which in utero exposure to the famine may affect infant mortality. The results presented are similar in magnitude and significance when these variables are excluded.

5

We include a dummy variable for the villages where the Maternal and Child Health Family Planning would be present even though that project did not begin until 1978 to account for potential differences between these villages even before the project began. We also estimate regressions without including these dummy variables and find nearly identical results, which are available upon request. Season-of-birth dummy variables include whether the child was born during the monsoon season (June–October) or the dry winter season (November–February), with the omitted category being the pre-monsoon hot season (March–May). Infant deaths are significantly higher during the winter season, largely because of agricultural cycles.

6

This type of selection implies that women who choose to become pregnant during a famine are negatively selected. Based on observable measures, the descriptive statistics from Table 2 suggest the opposite: although the magnitude of the difference is not large, women who became pregnant during the famine were more educated and married spouses who were more educated, compared with women who avoided pregnancy.

## References

Alamgir, M. (
1980
).
Famine in South Asia: Political economy of mass starvation
.
Cambridge, MA
:
Oelgeschlager, Gunn and Hain
.
Alamgir, M., & Salimullah, (
1977
).
Famine, 1974: Political economy of mass starvation in Bangladesh: A statistical annexe
.
Dacca
:
.
Almond, D. (
2006
).
Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 U.S. population
.
Journal of Political Economy
,
114
,
672
712
. 10.1086/507154
Almond, D., & Currie, J. (
2011
).
Killing me softly: The fetal origin hypothesis
.
Journal of Economic Perspectives
,
25
,
153
172
. 10.1257/jep.25.3.153
Almond, D., & Edlund, L. (
2007
).
Trivers-Willard at birth and one year: Evidence from US natality data 1983–2001
.
Proceedings of the Royal Society B
,
274
,
2491
2496
. 10.1098/rspb.2007.0524
Almond, D., Edlund, L., Li, H., & Zhang, J. (
2008
).
Long-term effects of early life development: Evidence from the 1959–1961 China famine
. In T. Ito, & A. Rose (Eds.),
The economic consequences of demographic change in East Asia
(pp.
321
345
).
Cambridge, MA
:
National Bureau of Economic Research
.
Almond, D., & Mazumder, B. (
2011
).
Health capital and the pre-natal environment: The effect of Ramadan observance during pregnancy
.
American Economic Journals: Applied Economics
,
3
(
4
),
56
85
.
Anderson, R., & Bergstrom, S. (
1998
).
Is maternal malnutrition associated with a low sex ratio at birth?
.
Human Biology
,
70
,
1101
1106
.
Bairagi, R. (
1986
).
Food crisis, nutrition, and female children in rural Bangladesh
.
Population and Development Review
,
12
,
307
315
. 10.2307/1973113
Barker, D. J. (
1990
).
The fetal and infant origins of adult disease
.
BMJ
,
301
,
1111
. 10.1136/bmj.301.6761.1111
Binswinger, H., & Rosenzweig, M. (
1986
).
Behavioural and material determinants of production relations in agriculture
.
Journal of Development Studies
,
22
,
503
539
. 10.1080/00220388608421994
Bulik, C. M., Sullivan, P. F., Fear, J. L., Pickering, A., Dawn, A., & McCullin, M. (
1999
).
Fertility and reproduction in women with anorexia nervosa: A controlled study
.
Journal of Clinical Psychiatry
,
60
,
130
135
. 10.4088/JCP.v60n0212
Camacho, A. (
2008
).
Stress and birth weight: Evidence from terrorist attack
.
American Economic Review
,
98
,
511
515
. 10.1257/aer.98.2.511
Cameron, E. Z. (
2004
).
Facultative adjustment of mammalian sex ratios in support of the Trivers-Willard hypothesis: Evidence for a mechanism
.
Proceedings of the Royal Society
,
271
,
1723
1728
. 10.1098/rspb.2004.2773
Choe, M. K., & Razzaque, A. (
1990
).
Effect of famine on child survival in Matlab, Bangladesh
.
Asia-Pacific Population Journal
,
5
(
2
),
53
72
.
Collins, S. (
1995
).
The limit of human adaptation to starvation
.
Nature Medicine
,
1
,
810
814
. 10.1038/nm0895-810
Davis, A. P. (
1996
).
Targeting the vulnerable in emergency situations: Who is vulnerable?
.
The Lancet
,
348
,
868
871
. 10.1016/S0140-6736(96)02134-4
Dyson, T. (
1991
).
On the demography of South Asian famines—Part II
.
Population Studies
,
45
,
279
297
. 10.1080/0032472031000145446
Fauveau, V., & Chakraborty, J. (
1994
).
Family planning and maternal and child health services in Matlab
. In V. Fauveau (Ed.),
Matlab: Women, children and health
(Special Publication No. 35, pp.
89
107
).
:
International Centre for Diarrhoeal Disease Research
.
Hart, N. (
1993
).
Famine, maternal nutrition and infant mortality: A re-examination of the Dutch Hunger Winter
.
Population Studies
,
47
,
27
46
. 10.1080/0032472031000146716
Huck, W. U., Labov, J. B., & Lisk, R. D. (
1986
).
Food restricting young hamsters (Mesocricetus auratus) affects sex ratio and growth of sebsequent offspring
.
Biology of Reproduction
,
35
,
592
598
. 10.1095/biolreprod35.3.592
India Famine Inquiry Commission
. (
1976
).
Report on Bengal
.
New York, NY
:
Arno Press
.
Janowitz, B., Holtman, M., Hubacher, D., & Jamil, K. (
1997
).
Can the Bangladeshi Family Planning Program meet rising needs without raising costs?
.
International Family Planning Perspectives
,
23
,
116
121
. 10.2307/2950767
Joshi, S., & Schultz, P. T. (
2007
).
Family planning as an investment in development: Evaluation of a program’s consequences in Matlab, Bangladesh
(Economic Growth Center Discussion Paper No. 951).
New Haven, CT
:
Yale University
.
Kannisto, V., Christensen, K., & Vaupel, J. W. (
1997
).
No increased mortality in later life for cohorts born during famine
.
American Journal of Epidemiology
,
145
,
987
994
. 10.1093/oxfordjournals.aje.a009067
Katz, M., & Vollenhoven, B. (
2000
).
The reproductive endocrine consequences of anorexia nervosa
.
British Journal of Obstetrics and Gynaecology
,
107
,
707
713
. 10.1111/j.1471-0528.2000.tb13329.x
Koenig, M. A., & D’Souza, S. (
1986
).
Sex differences in childhood mortality in rural Bangladesh
.
Social Science & Medicine
,
22
,
15
22
. 10.1016/0277-9536(86)90303-5
Kuhn, R. S. (
2005
).
The determinants of family and individual migration: A case-study of rural Bangladesh
(Research Program on Population Processes Working Paper POP2005-05).
Boulder, CO
:
Institute of Behavioral Science
.
Langsten, R. (
1981
).
The effects of crises on differential mortality by sex in Bangladesh
.
,
9
(
2
),
75
96
.
Lindeboom, M., Portrait, F., & van den Berg, G. J. (
2010
).
Long-run effects on longevity of a nutritional shock early in life: The Dutch Potato Famine of 1846–1847
.
Journal of Health Economics
,
29
,
617
629
. 10.1016/j.jhealeco.2010.06.001
Lumey, L. H., & Stein, A. D. (
1997
).
Offspring birth weights after maternal intrauterine undernutrition: A comparison within sibships
.
American Journal of Epidemiology
,
146
,
810
819
. 10.1093/oxfordjournals.aje.a009198
Mansour, H., & Rees, D. I. (
2012
).
Armed conflict and birth weight: Evidence from the Al-Aqsa Intifada
.
Journal of Development Economics
,
99
,
190
199
. 10.1016/j.jdeveco.2011.12.005
Meikle, D. B., & Drickamer, L. C. (
1986
).
Food availability and secondary sex ratio variation in wild laboratory house mice (Mus musculus)
.
Journal of Reproduction and Fertility
,
78
,
587
591
. 10.1530/jrf.0.0780587
Neugebauer, R., Hoek, H. W., & Susser, E. (
1999
).
Prenatal exposure to wartime famine and development of antisocial personality disorder in early adulthood
.
Journal of the American Medical Association
,
282
,
455
462
. 10.1001/jama.282.5.455
Painter, R. C., Rosenboom, T. J., & Bleker, O. P. (
2005
).
Prenatal exposure to the Dutch famine and disease in later life: An overview
.
Reproductive Toxicology
,
20
,
345
352
. 10.1016/j.reprotox.2005.04.005
Pitt, M., & Khandker, S. (
1998
).
The impact of group-based credit programs on poor households in Bangladesh: Does the gender of participants matter?
.
Journal of Political Economy
,
106
,
958
996
. 10.1086/250037
Rahman, O., Menken, J., Foster, A., Peterson, C. E., Khan, M. N., Kuhn, R., & Gertler, P. (
1999
).
The 1996 Matlab Health and Socioeconomic Survey: Overview and user’s guide
(Unrestricted Draft Series No. DRU-2018/1-NIA).
Santa Monica, CA
:
RAND
.
Razzaque, A. (
1988
).
Effect of famine on fertility in an area of rural Bangladesh
.
Journal of Biosocial Science
,
20
,
287
294
. 10.1017/S0021932000006623
Razzaque, A., Alam, N., Wai, L., & Foster, A. (
1990
).
Sustained effects of the 1974–5 famine on infant and child mortality in a rural area of Bangladesh
.
Population Studies
,
44
,
145
154
. 10.1080/0032472031000144426
Roseboom, T. J., van der Meulen, J. H. P., Osmond, C., Barker, D. J. P., Ravelli, A. C. J., & Bleker, O. P. (
2001
).
Adult survival after prenatal exposure to Dutch famine 1944–45
.
Paediatric and Perinatal Epidemiology
,
16
,
220
225
. 10.1046/j.1365-3016.2001.00336.x
Roseboom, T. J., van der Meulen, J. H. P., Ravelli, A. C., Osmond, C., Barker, D. H., & Bleker, O. P. (
2001
).
Effects of prenatal exposure to the Dutch famine on adult disease in later life: An overview
.
Twin Research
,
4
,
293
298
. 10.1375/1369052012605
Rosenfeld, C. S., & Roberts, R. M. (
2004
).
Maternal diet and other factors affecting offspring sex ratio: A review
.
Biology of Reproduction
,
71
,
1063
1070
. 10.1095/biolreprod.104.030890
Rosenzweig, M., & Wolpin, K. (
1985
).
Specific experience, household structure, and intergenerational transfers: Farm family land and labor arrangements in developing countries
.
Quarterly Journal of Economics
,
100
(
Suppl
),
961
987
. 10.1093/qje/100.Supplement.961
Scott, S., Duncan, S. R., & Duncan, C. J. (
1995
).
Infant mortality and famine: A study in historical epidemiology in Northern England
.
Journal of Epidemiology and Community Health
,
49
,
145
252
. 10.1136/jech.49.3.245
Sen, A. (
1981
).
Ingredients of famine analysis: Availability and entitlements
.
Quarterly Journal of Economics
,
96
,
433
464
. 10.2307/1882681
Stein, A. D., Zybert, P. A., & Lumey, L. H. (
2004
).
Acute undernutrition is not associated with excess of females at birth in humans: The Dutch Hunger Winter
.
Proceedings of the Royal Society B
,
271
,
S138
S141
. 10.1098/rsbl.2003.0123
Stein, Z., & Susser, M. (
1975
).
The Dutch Famine, 1944–1945, and the reproductive process. I. Effects on six indices at birth
.
Pediatric Research
,
9
,
70
76
.
Trivers, R. L., & Willard, R. E. (
1973
).
Natural selection and the ability to vary the sex ratio of offspring
.
Science
,
179
,
90
92
. 10.1126/science.179.4068.90
United Nations Secretariat
. (
2003
).
World population prospects: The 2000 revision
.
New York, NY
:
United Nations