Abstract
We use data from three waves of India National Family Health Survey to explore the relationship between the month of birth and the health outcomes of young children in India. We find that children born during the monsoon months have lower anthropometric scores compared to children born during the fall-winter months. We propose and test hypotheses that could explain such a correlation. Our results emphasize the importance of seasonal variations in environmental conditions at the time of birth in determining health outcomes of young children in India. Policy interventions that affect these conditions could effectively impact the health and achievements of these children, in a manner similar to nutrition and micronutrient supplementation programs.
I. Introduction
A large literature in economics, human biology, and medicine has been devoted to understanding the effects of early childhood conditions on outcomes later in life (see, for example, Alderman, Hoddinott and Kinsey 2001 and 2006; Martorell 1999; Beaton 1993). Research has consistently shown that the month of birth is an important predictor of health outcomes, morbidity, and mortality. But, so far, no convincing theories have been proposed to explain this association.
Most of the studies showing correlations between the month of birth and children’s health outcomes have been restricted to high-income settings. The correlation between an individual’s height and his/her month or season of birth was documented by Weber et al. (1998), for Australia; Kihlbom and Johansson (2004) for Sweden; Shephard et al.(1979) and Koscinski et al.(2004) for Poland. Early exposure to cold conditions was reported to be associated with higher weight during adulthood in England (Phillips and Young 2000). Van Hanswijck de Jonge et al.(2003), using data from the United States, found that birth weight and early infancy weight gains varied by season and were modified by ethnicity. Tanaka et al. (2007) showed that in Japan the height and weight of schoolchildren were influenced by their months and seasons of birth.
In this paper we use data from three rounds of the India National Family Health Survey (NFHS) to (i) quantify the effect of the month of birth on children’s anthropometrics, and (ii) test the hypotheses that might explain the relationship between the months of birth and children’s health outcomes. We find that Indian children born during the monsoon months have worse health outcomes than children born during fall-winter months. The “month-of-birth” effect is shown to be present in all three waves of the NFHS and persists after controlling for a wide set of observable and unobservable characteristics. The effect is large: The differences in height among children born during the monsoons and other children are comparable to the differences in health outcomes between children born to illiterate mothers and to mothers with completed primary education.
The empirical tests conclude that one of the likely explanations for the observed pattern of changes in children’s health by the month of birth could be a higher prevalence of malnutrition and a wider exposure to diseases in the lean monsoon season. Our results show that seasonal climatic variations affect the environmental conditions at the time of birth and determine health outcomes for young children in India. Policy interventions that influence such conditions might be as effective in improving children’s health as nutrition and micronutrient supplementation programs and could drive long-term economic growth by leading to healthier and more productive adults.1
This paper contributes to the research on child health and nutrition in several ways. First, in contrast to the existing medical literature on the effect of the month of birth on individual health, we try to present a comprehensive causal analysis of this phenomenon, taking into account both socio-economic and environmental factors that might affect children’s health outcomes. Second, we analyze the impact of the months of birth on children’s health in a context of a poor developing country, as compared to most of the literature on this topic, which focuses on developed countries. Third, we demonstrate the “month-of-birth” effect for children younger than three years of age; the majority of studies analyze the consequences of seasonality of birth on outcomes in adult life. Our results prove to be useful in allowing for greater precision in the use of climatic data as explanatory variables in an analysis of health outcomes.
The following section describes the setting of our data and shows some descriptive results. Section III sets up the main question of the paper. Section IV outlines and tests the main hypotheses explaining the association between the month of birth and children’s health outcomes. Section V discusses policy implications and offers some conclusions.
II. Data and Descriptive Statistics
This analysis uses data from three waves of the India NFHS (1992/ 1993, 1998/1999, and 2005/2006). The NFHS is a survey of representative households covering states and territories of India containing approximately 99 percent of its population. The survey structure corresponds to the typical structure of demographic and health surveys (DHS) conducted in several other countries. Our main sample contains information on 45,279 children from the 1992/1993 round; 30,984 children from 1998/1999 round; and 48,679 children from 2005/2006 round of the India NFHS.2 These children were residing correspondingly in 33,032, 26,056, and 33,968 households.
The NFHS uses three types of questionnaires: The household questionnaire collects information on the family structure and background, and on various characteristics of household members. The woman’s questionnaire is administered to women aged 15 to 49 and covers dates and survival status of all births, current pregnancy status, childbearing intentions, and childcare practices. The village questionnaire gathers information on the village area and population. We use the constructed household wealth index as a measure of household welfare. The descriptive statistics for the main variables used in this paper are shown in Table 1.
We focused our analysis on the age-adjusted measure of height-for-age (HAZ), which reflects children’s development relative to a reference population of well-nourished children (World Health Organization 2006).3 The height-for-age (stunting) is an indicator of the long-term effect of malnutrition (Dibley et al. 1987). For comparability between the rounds of NFHS, we restricted our sample to children less than 36 months of age.
Malnutrition is highly prevalent in India. According to NFHS data, about 80 percent of children under the age of three were underweight in 1992, with minimal changes in 1998 and 2005. In 1992, 72 percent of boys and 70 percent of girls were stunted.4 The proportion of stunted children had decreased to 65 percent by 2005. The overall averages of HAZ rose over the years for both boys and girls. The average HAZ had risen for boys from −1.9 in 1992 to −1.5 in 2005. Girls experienced similar improvements in health outcomes between 1992 and 2005.
The distributions of births by calendar months by gender and for urban and rural samples are presented in Figure 1.5 The proportions of boys and girls born in each month are similar. The highest birth rates are registered in August, September, and October—these children were conceived in winter. The fewest children were born during the winter months of December, January and February—these children were conceived in spring. The wedding season in India, which falls in the months from November to February could partially explain this seasonality of birth (Medora 2003). Figure 2 shows the proportion of children, among all children born in a particular month, who died before the age of three years. The incidence of mortality appears to be unrelated to the month of birth of boys and girls.
In addition to the data from NFHS, we use historical rainfall data from TS2.1 database supported by the Tyndall Centre for Climate Change Research at University of East Anglia in Norwich, U. K. This database contains monthly meteorological data for the period from 1901 to 2002 at the nodes of a global grid spaced at 0.25 degree latitude and longitude width (Mitchell and Jones 2005). We assigned the monthly rainfall data to the districts of NFHS covered in 1992 and 1998 rounds. The effect of the rainfall on children’s health is identified through the variation in the rainfall over the three years for which we observe children in each round of NFHS.
III. Variations of child health outcomes by month of birth
Figure 3 shows the changes in height-for-age Z–scores by the month of birth for boys and girls in the pooled sample of three rounds of NFHS.6 Anthropometric measures appeared to be the lowest for children born in summer and improving for children born in fall and early winter. This relationship held for both girls and boys. For example, if the average HAZ for boys born in urban areas of India in June was about −1.61 (standard error of 0.05), HAZ for boys born in December was −1.17 (SE 0.06). The average HAZ increased from −1.78 for girls born in urban areas in June to −1.21 for girls born in December.
Figure 4 shows the average HAZ by months-of-birth normalized to the beginning of the monsoon season that starts in late May/early June in the southern states of India and in late July in Northern India. According to this normalization, children born in June in the southern states of India would have a normalized month-of-birth of 0, and children born in November would have a normalized month-of-birth equal to six. The trends in health outcomes depicted in Figure 4 are similar to those of Figure 3: Children born during the monsoon months are more likely to be stunted than the children born during the six months after the start of the monsoon.7
Descriptive Statistics for the Main Explanatory Variables.*
Percent of Total Number of Births by Month of Birth
Would the effect of the month of birth on a child’s health be significant after controlling for the characteristics of a household, a mother, and a child? To find the answer to this question we rely on a standard theoretical framework of household utility maximization that incorporates the production function determining child’s health (Behrman and Deolalikar 1988). According to that theory, a household utility is a function of consumption and leisure of household members, as well as the quality (health) and quantity of their children. A household maximizes its utility subject to standard budget constraints and the restrictions imposed by the health production function. The household demand for child i’s health Zi depends on a set of exogenous characteristics of a child Xi, household characteristics Xh, characteristics of its mother Xm, community characteristics Xc, and some unobserved factors captured by random error εi (Thomas, Strauss, and Henriques 1991). This relation can be expressed as:
Proportion of Children that Died Before the Age of 36 Months Among All Children Born in a Particular Month
The child’s characteristics include, among others, its month of birth. In the linearized form, the production function of child’s health (Equation 1) can be expressed as:
where vector 
combines the child’s, mother’s, household’s and community’s characteristics, Mik is a dummy variable equal to 1 if child i is born in month k, and πk and α are parameters. The top panel in Table 2 shows the coefficients of the linear regressions of HAZ for children younger than 36 months. The “month-of-birth” effect persists when we control for a wide set of exogenous characteristics in Equation 2. Relative to December, children born in other months have worse health outcomes, with the largest negative difference in May, June, and July. The middle panel in Table 2 shows the regression coefficients for the specifications with dummy variables corresponding to months-of-birth normalized to the beginning of the monsoon season. The estimates based on this specification demonstrate that children born in the months close to the start of the monsoon season have worse health outcomes compared to children born during winter months.
Health Outcomes (HAZ) by Month of Birth and Gender
The coefficients on the control variables in the regressions used in Table 2 reveal the expected relationship between child health outcomes and the characteristics of a household, a mother and a child. A higher birth order has a negative impact on the health outcomes for both boys and girls (for example, Horton 1988; Angrist and Evans 1998). Consistent with the findings in the literature (for example, Holmes 2006) health outcomes deteriorate with the age of a child. Children living in wealthier households and with better-educated mothers are less likely to be stunted. The relationship between a child’s health and his/her mother’s age has an inverted Ushape: HAZ improves with the mother’s age till the age of about 40 and then declines for children of older mothers.
The potential endogeneity of the month-of-birth could have strong implications for the findings of this paper. Suppose parents believe that children born during certain months of the year are more likely to get sick. Then parents would try to plan their pregnancies to give birth in months most favorable for their children’s health. Or, parents might try to compensate the perceived negative effects by providing better care for children born in “bad” months. In the presence of an unobserved heterogeneity in the parental inputs to the production function of a child’s health, the variation in health outcomes by the month of birth could be partially attributed to differences in the parental behavior (for example, Rosenzweig and Shultz 1982).
Assume the error terms in Equation 2 can be decomposed into the child-specific IID error term μij and the household-specific unobserved factor νj, capturing the heterogeneity in parental inputs to the production function of children’s health. νj can be correlated with the month in which a child is born. Then, Equation 2 will have a following form:
Health outcomes (HAZ) by gender and the month of birth normalized relative to the beginning of monsoon season
Under the assumption that the unobserved heterogeneity in parental inputs (for example, far-sightedness of the parents) is constant over time (siblings), we can account for the endogeneity of the month-of-birth by estimating the fixed-effect (FE) regression on the sample of siblings, thus removing the household-specific component νj. The household FE would also absorb a host of other time-invariant geographical differences that might affect child health outcomes. This regression includes explanatory variables that differ among the siblings living in the same household: dummies for the month of birth, age of a child and the child’s birth order. The bottom panel of Table 2 shows that most of the FE coefficients on the 11 dummies for the months of birth are statistically significant.8 The seasonal patterns in child health are similar to patterns revealed by OLS estimations: children born in fall-winter months are healthier compared to children born in summer.
Does child’s Height-for-age Z-score Depend on the Month of birth? (Regression on the Pooled Sample of Three rounds of NFHS. Coefficients on the month-of-birth dummies).
The FE estimates could be criticized on the basis that the subsample of women who have two births in the three-year period prior the survey might not be representative for all women. Their births are more likely to be small and to experience growth shortfalls. However, the comparison of estimates with and without fixed effects for the sample of children used in the fixed effect regression—that is, the sample that excludes children who have no siblings in the relevant age range, also show that the FE coefficients on the month-of-birth dummies are larger than the OLS coefficients. The OLS coefficients estimated on the restricted sample are close in magnitude to the OLS coefficients estimated on the general sample (these results are available from the authors). This makes us believe that the selection bias might be not too strong and the FE results can be extrapolated for the whole population.
The coefficients on month-of-birth dummies exhibit similar patterns in regressions for boys and girls in all three specifications shown in Table 2. The Chow test (Chow 1960) of the equality of these coefficients confirms this observation—the month-ofbirth coefficients for boys are statistically different from the coefficients for girls only in three out of 42 pair comparisons. Boys and girls might be subjected to different patterns of selection and post-natal treatment and the similar effects of month of birth on children health indicates that seasonality in outcomes cannot be explained by gender differences in selection and treatments. The further results in the paper are based on the pooled sample of boys and girls.
The average HAZ of children born during the monsoon season is about 0.5 standard deviations (SD) lower than the average HAZ of children born in the fall-winter months (Figure 4). After controlling for the characteristics of the child, the mother and the household they live in, the “month-of-birth” effects ranged from 0.4 to 0.8 SD (Table 2).9 The magnitude of this effect is similar to the differences in Z-scores between children of illiterate mothers and mothers with incomplete secondary education, as observed in our data. Alderman, Hoogeveen, and Rossi (2005) report that the anthropometric measures of children born during the “lean” season in Tanzania are 0.2 to 0.4 SD lower than the measures of children born in other months of the year. These effects are comparable with the effects of nutrition programs and the estimated elasticities of the changes in household’s and mother’s characteristics. For example, in Bangladesh the average HAZ of children with illiterate mothers is about 0.4 SD lower than the HAZ of children whose mothers hold a university degree (Moestue and Huttly 2008). Similar effects of maternal education on children’s anthropometrics are found by Alderman and Garcia (1994) for Pakistan and Kassouf and Senauer (1996) for Brazil. The provision of the micronutrient supplements resulted in a 0.2 SD increase in HAZ for children in Vietnam and Mexico (Thu et al. 1999; Rivera et al. 2001). In Pakistan, a vaccination against upper respiratory illnesses improves HAZ by more than 1 SD according to Alderman and Garcia (1994).
The above estimations demonstrate the consistently strong correlations between the month-of-birth and children’s health outcomes for boys and girls, across different regression specifications, years, and for different samples. In the next section we try to establish the causality of this correlation.
IV. Explaining the Correlation between the Month of Birth and Health Outcomes
The standard theories attribute variations in children’s health outcomes by month-of-birth to seasonal differences in child’s prenatal and postnatal nutrition and disease environment (Doblhammer 2004; Doblhammer and Vaupel 2001; Almond 2006; Maccini and Yang 2009; Jayachandran 2009). Both diet and the absence of disease are crucial for the adequate growth of children and may work in synergy (Alderman and Garcia 1994). Changes in food supply and food quality affect intrauterine growth. In India, mothers who gave birth in the fall or an early winter season had access to better-quality food and fresh fruits and vegetables during most of their pregnancies. Occurrences of infectious diseases that impact the mother, fetus, and newborn child are correlated with seasonal climatic changes because of the interaction between climate and the vectors of disease, and the interaction between the nutritional status and immune functions of a mother and her child. Other seasonal factors that might affect health of children include the effect of exposure to sunlight on the children’s and mother’s metabolisms, and the seasonal variations in availability and rates of vitamin absorption. Artadi (2005) demonstrates that infant mortality rates vary significantly by the month of birth due to the variation in an incidence of diseases mostly caused by changes in rainfall.
The monsoon is a climatic event that separates seasons and is essential for agriculture in India. The deprivation of nutrients and other health-related intakes during the monsoon was shown to have a considerable effect on the health status of women and children in India (for example, Chambers 1982; Sahn 1987). The monsoon is associated with heavy rains that bring a multitude of diseases (see Tables A1a and A1b). Contaminated drinking water results in a high incidence of water-borne gastrointestinal infections (Khatiwada and Rimal 2007) and cholera (Saraswathi and Deodhar 1990). Excess humidity aggravates skin conditions, asthma and eclampsia (Subramaniam 2007). Mosquito- and rat-borne diseases such as dengue (Bharaj et al. 2008) and malaria (Kabanywanyi et al. 2008) are transmitted faster in the wet and humid environment created by the monsoon. Awasthi and Pande (1997) and Bamji et al. (2004) show that the combined morbidity is the highest in the monsoon compared to winter months. The National Nutrition Policy developed by Department of Women and Child Development in India (GoI 1993) indicates that children of poor households are at risk of malnutrition during the monsoon season and recommends giving special rations to the seasonally at-risk population.
The NFHS provides no data on the nutritional intake of mothers during pregnancy and histories of illnesses for the children during the first months of their lives. In the absence of such information we use household wealth and the education of the mother as proxies for the nutritional status and disease environment into which the children were born. We argue that wealthier households have better means to smooth out consumption of their children during the lean season and can obtain better healthcare if their children become sick. Maternal education has a strong positive impact on children’s nutrition and health outcomes via modern attitudes toward healthcare and reproductive behavior (Caldwell 1979; Thomas Strauss and Henriques 1991; Glewwe 1999).
To test the effect of nutrition and disease environment at birth on children health outcomes we regress HAZ of a child on a set of explanatory variables that include the socio-demographic characteristics of a household, a household wealth index, characteristics of the mother (including her educational attainments), location dummies, the month-specific level of precipitation in a district, month-of-birth dummies, and interactions of the month-of-birth with wealth index Ii and years of mother’s education Ei. This regression can be expressed as:
A significance of coefficients on month-wealth interactions (βk) or month-education interactions (γk) would point to the presence of seasonal differences in the effect of wealth and maternal education (as a proxy for nutrition and healthcare) on children’s health outcomes.
Table 3 shows the estimated coefficients on interactions of month relative to the start of the monsoon and household wealth and maternal education.10 The table is based on two econometric specifications: OLS regression that includes dummy variables for the month of birth relative to the start of the monsoon; and a fixed effect (FE) regression estimated on the sample of siblings. In FE regression we tried to address concerns about the potential endogeneity of the month of birth (discussed in the previous section) and use the FE approach to control for time-invariant unobservable characteristics that might be correlated with the month of birth.
Estimations in Table 3 demonstrate that household wealth and the education of mothers have significant effects on children’s health outcomes by month-of-birth. Wealthier families seem to be able to compensate for the negative impact of the monsoon on their children’s health. The seasonal variations in HAZ for children living in wealthier households and with better-educated mothers were smaller compared to the variation in health outcomes of other children (βk > 0 and γk > 0). For example, in urban areas of India the difference in HAZ of children born during and six months after the monsoon is about 27 percent smaller for the children from wealthiest decile relative to children from the poorest 10 percent of wealth distribution. For children from rural households the percent of the month-of-birth effect offset by the interaction term is about 10 percent. The results of the F–test (χ2 test for the fixed effect model) on the joint significance of interaction coefficients are shown in the last line of Table 3. The tests strongly reject the null-hypotheses of no seasonal heterogeneity by maternal education and household wealth.11
Does the Effect of the Month of Birth on Child’s Health Depend on Household Wealth and Maternal Education? (Coefficients on the Interactions of Household Wealth and Maternal Education with the Month of Birth relative to the start of the monsoon. F-test for OLS regression and χ2 − test for FE Regression on the Joint Significance of the coefficients on the interaction terms.)
NFHS collects information on the diseases experienced by a child before the survey interview. These data can be used to support the findings discussed above. Table 4 shows the estimates of the probability of a child younger than 12 months to have a diarrhea during two weeks prior the interview. The estimation controls for a wide range of characteristics of a child, its mother and the household it lives in and demonstrates that young children living in rural areas of India are more likely to experience diarrhea during the summer months or months around the start of the monsoon. We find no such effect for children living in urban areas of India whose families have better access to clean water and sanitation. These results provide some evidence to the hypothesis about the importance of the effect of postnatal disease environment and nutrition on children health outcomes later in life.
We also use rainfall data to determine the effect of the postnatal environment on health. Table 4 shows the results of height-for-age regressions of Equation 2 where, in addition to the variables used in the estimations presented in Table 2, we added levels of rainfall during the month of a child’s birth at a district level. The coefficients on the rainfall variables are significant only for girls residing in urban areas of India, which is different from finding by Manccini and Yong (2009). The inclusion of the rainfall attenuates the coefficients of the month of birth dummies for children in urban areas, especially for girls. Rainfall has little effect on the month-of-birth coefficients for rural areas.12
A. Understanding the nature of the potential sample selection
Several types of sample selections on different stages of child’s life could potentially affect our results. The seasonal fertility patterns can be different for rich (bettereducated) and poor (less-educated) families; selective survival can affect the composition of children born in a particular month; and child outcomes may vary with the month of birth because of the differences in the parental effort. In this section we try to address these problems in turn.
If, during certain seasons, more children are born in better-educated and/or wealthier families, the correlation between children’s health outcomes and their monthsof- birth can be attributed to the difference in resources available to the children (for example, Bronson 1995). Buckles and Hungerman (2008) explain the effect of season of birth on later health and professional outcomes by changes in the characteristics of women giving birth throughout the year in the United States. Dehejia and Lleras-Muney (2004) show that the changes in parental behavior and the differential fertility may result in difference in the health of children over the business cycle and also seasonally. In developing countries, women’s involvement in agricultural activities, food availability, the seasonality of marriages, and male migration are more important determinants of the seasonality of birth. Panter-Brick (1996) dem- onstrates that, in Nepal, seasonal rates of pregnancies are determined, among other things, by seasonality of marriage (which, in turn, is determined by agricultural cycles), and marital disruptions related to out-migration of males and agricultural activities; the peaks of conception are observed in the beginning of the monsoon season of June-July and rice harvesting in December. Rajagopalan, Kymal, and Pei (1981) documented the strong effect of agricultural cycles on births in Tamil Nadu in India, emphasizing large differences in the seasonality of birth between urban and rural areas. Agricultural cycles are shown to influence timing of birth and infant mortality in Sub-Saharan Africa; in rural families fewer children are born during the months of high demand for female labor even though children born in these months have higher chances of survival (Atradi 2005).
Left panel: Probability of a child younger than 12 months to have a diarrhea during two weeks prior to survey interview. Right panel: Does Child’s Height-for-Age Z-score Depends on the Level of Rainfall During the Month of Child’s Birth? (Coefficients on the dummies for the interview months relative to the start of the monsoon. Pooled sample of 1992 and 1998 rounds of NFHS.)
To evaluate the differences in the fertility patterns across socioeconomic groups, we estimate the relationships between the month of birth, household wealth, and maternal education, controlling for the characteristics of a household and a mother. This relationship can be expressed as:
where Prob(Mik = 1) is the probability of child i to be born in month k. Given an unordered structure of the month-of-birth variable and assuming that εik’s are independent and identically Gumbel distributed, we applied the multinomial logit specification for this estimation.13 A significance of the coefficients βk and γk would indicate that household wealth and education of the mother affect the probability of a child to be born in a certain month of the year.
Table 5 shows the multinomial logit estimates of the coefficients on the wealth index and maternal education for 11 month-of-birth relative to the start of the monsoon categories for boys and girls using the pooled sample of three waves of NFHS. For children residing in urban areas wealth and maternal education have no significant impact on the seasonality of their births. The effects of wealth and mothers’ education on the month of birth are significant for rural children born during summer months. But the pattern of this significance differs from the patterns we would expect to observe based on Figure 4. For example, better-off rural households are more likely to have their children born in the summer months. But months close to the beginning of monsoon are the “bad” months to be born in, in terms of health outcomes. We find no effect of wealth and maternal education on timing of birth of urban children. The results of likelihood ratio tests of the significance of household wealth index and maternal education in determining a child’s month of birth are shown in the bottom part of Table 5. These tests confirm that both wealth index and maternal education contribute little to determining the month of year in which a child will be born and thus our empirical results should not be affected by this type of selection bias.
Does the Month of Birth depend on Household Wealth, Education of the Mother, or whether the Child Was Wanted. (Multinomial Logit and SML Coefficients on household wealth, maternal education, and child’s “desirability” dummy.)
B. The “unplanned pregnancy” selection
Suppose parents believe that certain months are “bad” for their children to be born in. Then, in order to improve their children’s health outcomes, parents would plan their pregnancies to give birth during the “good” months. Under this assumption, children born during “bad” months would more likely be a result of unplanned pregnancies and thus to have disadvantaged health status (Kost et al. 1998).14 If this assumption is true, the observed variation in children’s anthropometrics across the months of the year can be explained by the higher proportion of unplanned births during the monsoon season.
The NFHS questionnaire asked mothers whether their pregnancy with a particular child was desirable—in a form of the following question:
In the time you became pregnant with (Name) did you want (1) to become pregnant then, (2) to wait until later, (3) no more children at all?
The birth of about 20 percent of children could be classified as unplanned—that is, the mothers of these children either wanted to have children later or did not want to have more children at all. There are no significant differences in the shares of desirable pregnancies of boys and girls because parents cannot select the sex of the child at the time of conception.15
We use these data to evaluate the importance of “unplanned pregnancy” bias for our main results. We estimate the probability of a child being born in a certain month of the year as a function of the “desirability” of a child and a wide set of controls.
where Di is the “child desirability” dummy, which is equal to one if a mother wanted to have that child when she became pregnant, and zero otherwise. A significance of coefficients on the “child desirability” dummies (ηk’s) would indicate the concentration of unwanted pregnancies in some months of the year. The right panel of Table 5 with estimates of Equation 6 shows that for urban areas of India, “child desirability” had virtually no effect on the probability of children being born in a certain month of the year. In rural areas, children who were wanted by their parents were more likely to be born four months prior to monsoon and in the month of the monsoon. These patterns of births would result in a variation in health outcomes, very different from those observed.
A mother’s opinion about the “desirability” of her child at the time of conception could be a subject to recall errors. Such measurement errors would attenuate the coefficients of interest. We instrument the “desirability of birth” variable with the gender composition of older siblings in the household. Our exclusion restriction is based on the argument that the genders of the child’s siblings would affect the “desirability” of a child for its parents (for example, Thomas 1994; Duflo 2003) and would have no direct effect on the month the child is born in. This exclusion restriction relies on the assumption that parents have no control over the gender of their children. At the same time, selective abortions and other methods to select gender are known to occur in India. To test the validity of our exclusion restriction we reestimate our model on the total sample and on the sample of observations from the regions where the methods of gender controls are not practiced.16
We estimate the system of two simultaneous equations: a binary outcome equation that determines the “desirability of birth” as a function of the characteristics of the mother, household characteristics, and the gender composition of the child’s older siblings (we use “a household already has a male child” and “the previous child was a boy” as instruments for desirability of children) and the multinomial equation for the quarter of birth as a function of maternal and household characteristics and the endogenous indicator of desirability. This system is estimated by the method of Simulated Maximum Likelihood that relies on the Geweke, Hajivassiliou, and Keane (GHK) algorithm for estimating higher-dimensional cumulative normal distributions (Hajivassiliou and McFaden 1990; Hajivassiliou 1993). The last panel Table 5 shows the coefficients on the instrumented “desirability of birth” variable in the multinomial part of the estimation.17 Similar to the first specification, the instrumented “desirability” variable has no impact on the probability of being born in a particular quarter.
C. The “survival” selection
The “selective survival” of strong children at birth and during infancy can affect seasonal differences in children’s health outcomes. If mortality at birth is higher in a certain season than mortality at birth in other seasons, children who survived during the high-mortality season might be more robust and would have better outcomes later in life (Samuelson and Ludvigsson 2001). In other words, weak children born in high mortality seasons die and only the strong children survive. Alternatively, a switch from breastfeeding to supplementary milk and solid food at about six months after birth might be associated with higher child mortality (Barrera 1990; Olango and Aboud 1990), with higher chances of survival for stronger children. Doblhammer and Vaupel (2001) show that in Denmark the death rate for infants born in June is 32 percent higher than the death rate of those born in January. Similar patterns are found by Breschi and Bacci (1998) for Switzerland and Belgium. In Gambia and Bangladesh, birth during the hungry season resulted in excess mortality during the first year of life (Moore et al. 2004).
Let 
be a latent variable determining the survival of a child, which is a function of its health endowments at birth (weight at birth Wi), as well as maternal and household’s characteristics. Let Si be an observed event of child i surviving longer than 12 months. Then:
To test whether selective child mortality during the first months after birth explains the variation in health outcomes, we estimate the probability of a child surviving past age of one as a function of the characteristics of a household, a mother, and a child; month-of-birth dummies; and the interaction between the month-of-birth and weight at birth. The significance of the coefficients on the interactions between the month of birth and the weight at birth (γk’s) in Equation 7 rejects the null hypothesis that the survival probabilities of weak and robust children born in a certain month of the year are the same.
The probit estimations of Equation 7 indicate that children of a higher birth order, born in poorer households or with younger mothers, and children born with low weight are more likely to die in the first year after birth. The sample for this estimation includes all children born in all families. The left panel of Table 6 shows the coefficients on the interaction terms of the month of birth and child’s weight measured at birth (γk’s) in Equation 7. Interaction terms are significant for some months, but the pattern of this significance is inconsistent with the observed variation in children’s health outcomes. For example, in urban areas higher birth weight increases the probability of survival for children born in one month prior to monsoon, and has an opposite effect on survival probability for children born in rural areas. The estimations shown in the left panel of Table 6 provide no evidence in support to the existence of “selective survival” biases.18 However, the information on measured weight at birth could be imprecise or registered with an error. We use an alternative measure of child weight at birth to address the potential attenuation bias in the estimated coefficients in Equation 7 due to such imprecisions.
The NFHS asks mothers to categorize the weight of their children at birth as large, average, or small.19 We use this subjective assessment as a proxy for health endowments at birth. The weight of about 20 to 25 percent of children was smaller than average according to their mothers” assessments. Similarly to the results in the left panel of Table 6, the patterns of survival shown in the right panel are inconsistent with the observed monthly variation in children’s health outcomes. For example, low birth-weight children were less likely to survive if they were born in one month before (urban areas) or after (rural areas) the start of the monsoon. If present, selective survival bias would result in better health outcomes for children born during these months, which is not the case. Based on these evidences, we find no support for the existence of the “selective survival” bias in our results.
Does the Probability of a Low-weight Child to Survive Past 12 Months depend on the Child’s Month of Birth? (Sample of children born no more than 36 months prior to the date of interview. Probit coefficients on the interactions of moth-of-birth dummies and the weight of a child).
D. Sensitivity analysis
We test the stability of our results using alternative econometric specifications and different samples. We also assess how our results would change if we use trimesters, quarters, and semesters of birth instead of the month of birth. In addition to calendar seasons, we define three subtropical seasons and the hungry and harvest seasons, relative to the start of the monsoon in the particular area (as in Moore et al. 2004). The three subtropical seasons are: the monsoon season (June/July-September/October), the cool-dry winter season (October/November-February/March), and the hotdry season (March/April-May/June). The hungry season starts in June/July and ends in December/January. Expectedly, the larger the time span, the weaker the impact of the season of birth on health outcomes— in other words, we find that trimestral variations in HAZ are larger compared to quarterly and half-year variations. But our main results hold when we use these different timeframes: the timing of birth appears to be an important determinant of children’s health in all these specifications.
Another concern about the stability of our results is the impact of the deterioration in health outcomes after children are switched from the breast milk to solid food, which happens 4–6 months after birth in India (for example, Barrera 1990; Olango and Aboud 1990; Adair and Guilkey 1997). To address this concern, we repeat our analysis on a sample of children older than 12 months. Again, our main conclusions remain the same: The environmental conditions around the date of birth play an important role in determining child’s health outcomes later in life.
V. Conclusions
In this paper, we use data from three waves of the India NFHS to explore the relationship between the month-of-birth and health outcomes of young children in India. We demonstrate that children’s anthropometric scores vary significantly with the month of birth: children born during the monsoon months have consistently worse health outcomes compared with children born in the fall-winter season. The “month-of-birth” effect persists after controlling for a wide set of observable and unobservable characteristics of a child, its mother, and the household where they live. The size of these effects ranges from 0.5 to 0.8 SD of the HAZ and is comparable or larger than the effects of maternal education and nutritional supplementation programs on children’s health as found in other countries.
Improving children’s health is an important development objective of many international organizations (World Bank 2002). Our results demonstrate the significance of seasonal changes in environmental conditions in explaining the variation in children’s health. Interventions to improve these conditions could have a positive impact on the health and achievements of children. Family planning campaigns can help parents optimally time births, thus improving health of their children. Current policies aimed at enhancing children’s nutritional status and health fail to incorporate measures that differentiate between children born in different months of the year (Elder, Kiess, and de Beyer 1996). It appears that low-cost modifications to the existing nutritional programs, which take into account the season of birth, may have a large impact on the health of Indian children. For instance, information campaigns could emphasize the seasonal differences in maternal and childcare practices. Supplementary feeding and infectious disease control programs should be designed to address different needs of mothers and children born in different seasons. Such programs might target children at risk of malnutrition before their health outcomes deteriorate rather than targeting children that are already stunted or underweight, which is a standard practice in most of nutritional programs.
While pointing out the importance of seasonal factors in explaining the variations in children’s health, our paper provides no information on the channels through which these factors affect children’s health. Possible next steps in this research could involve studies to differentiate between the seasonal impacts of prenatal and postnatal conditions; to understand the behavioral responses of households to offset the negative environmental conditions for children born during the “bad” season; and to analyze the channels through which the environment, at the time of birth, affects individual health. Finding answers to these questions would help design more effective programs to improve the health and achievements of Indian children over their lifetimes.
Footnotes
Michael Lokshin, mlokshin{at}worldbank.org, and, Sergiy Radyakin, sradyaking{at}worldbank.org, are both at the Development Research Group, World Bank, 1818 H Street NW, Washington, D.C., 20433. Authors thank Harold Alderman, Jishnu Das, Monica Das Gupta, Toan Q. Do, and Mark Villa for their constructive comments. The findings, interpretations, and conclusions of this paper are those of the authors and should not be attributed to the World Bank, its Executive Directors, or the countries they represent. The data used in this article can be obtained beginning June 2012 through May 2015 from the authors.
↵1. For example, Victoria et al. (2008) present evidence of the link between childhood stunning and short adult stature, less schooling, and diminished educational functioning. Hoddinott et al. (2008) show that improving nutritional status in early childhood led to higher earnings.
↵2. The number of observations in the 1998/1999 round is smaller because NFHS-2 collected height and weight information only for the last two children, under three years of age, of ever-married women who were interviewed.
↵3. The analysis of relation of month of birth and weight-for-age (WAZ) z-scores is presented in a working paper version of this paper (Lokshin and Radyakin 2009). Across rounds of NFHS, about 10 percent of eligible children were not measured, either because the children were not at home, or because their mothers refused to allow the measurements (Lokshin, Das Gupta, and Ivaschenko 2005). We find no relationship between this attrition and the seasonality of birth, gender, or type of the family (results are available from authors).
↵4. A child is considered to be mildly stunted if its HAZ is from −1 to −2 standard deviations (SD) from the mean. Children with HAZ from −2 to −3 SD are considered to be moderately stunted, and children with HAZ less than −3 are considered severely stunted.
↵5. The results shown in Figure 1 are calculated based on the sample of children younger than 36 months of age ever born in the household. The distribution of birth by month constructed based on the sample of survived children is similar to one shown in Figure 1.
↵6. A relatively large gap between the anthropometric measures of children born in December and January persists in all rounds of NFHS. We observe similar December-January differences in health outcomes in DHS surveys from other countries. We were not able to find an explanation for that phenomenon. We have also repeated our analyses on a sample of children with the HAZ bounded in the range from −3 to +3 SD and found our results to be robust to this restriction.
↵7. States in Southern India are exposed to two monsoons, the South-West Monsoon that starts at early June and North-East monsoon that starts in October and lasts till mid-February. Kerala and Tamil Nadu are the two states that affected by the North-East monsoon the most. Interestingly, the height-by-age zscores of children living in these states exhibit a U-shape pattern very similar to the patterns found for the India as a whole.
↵8. The F-tests on the FE in all regressions in Table 2 reject the pooled-OLS specification in favor of FE specification.
↵9. Several studies point to the problems with the quality of age data in NFHS (for example, Bhat 1995; Narasimhan et al. 1997; Mishra, Lahiri, and Luther 1999). This measurement error in reported age would attenuate our results toward zero, so that the estimated effects of the month of birth on children’s health would be lower bounds (in absolute terms) of the true effects.
↵10. The complete regression is shown in Table 3A in Appendix
↵11. We tried to differentiate between the impact of prenatal and postnatal environments on the variations in children’s health outcomes. We regress the mothers’ subjective assessments of their children’s weights at birth (discussed later in this section) on the month-of-birth dummies and the standard set of controls. The correlation of weight at birth with the month of birth would indicate that the observed seasonal variations in children’s health can be explained by the variations in prenatal conditions. The coefficients on month-of-birth dummies in this regression are insignificant, providing us with no evidence for differentiating between the impacts of prenatal and postnatal conditions.
↵12. The district indicators are available only in the 1992 and 1998 rounds of NFHS. For that reason figures shown in Table 4 are based only on these two rounds of the data. In addition to the specification with the levels of rain falls at the month of birth, we also tried specifications with pre- and postnatal levels of rainfall. The prenatal rainfall has no significant effect on children health outcomes while the postnatal levels (one and two months after birth) of rainfall result in attenuation of the coefficients similar to the results shown in Table 4. These results are available from authors on request.
↵13. The Independence of Irrelevant Alternatives assumption of the Multinomial Logit is clearly violated in the case of modeling the month of birth. A multinomial probit model (MNP) would be more appropriate for estimating Equation 5. However, estimation of MNP with 12 categories appears to be computationally infeasible.
↵14. A paper by Do and Phung (2008) shows that the differences in outcomes between children born in the “good’ and “bad’ years according to the Chinese horoscope can be explained by the fact that children born in “bad’ years are more likely to be unwanted, that is to be the result of unplanned pregnancies. To confirm these results on our data we estimate IV regressions of HAZ on a predicted probability of a child’s birth to be desired and other covariates. The result of these regressions demonstrate that HAZ of children born as a result of unwanted pregnancy is significantly lower compared to HAZ of children whose birth was desired. The results of these estimations are available from authors.
↵15. Though the observed ratio of desired boys and girls could be influenced by selective abortion after conception.
↵16. Das Gupta, Chung, and Shuzhuo (2009) show that in southern and eastern states of India such as Andra Pradesh, Karnataka, Kerala, Tamil Nudu, Orissa, and West Bengal the sex ratio of children aged zero to six years is close to one, thus allowing us to conclude that the prevalence of gender-selection methods in these states was low. We use the sample of observations from these states to test our exclusion restrictions. The results of these estimations are similar to those obtained from the estimations on the whole sample.
↵17. We collapsed months into quarters to make the estimation computationally feasible. We fail to produce stable results with the FIML with 11 categories. We use the Stata routine “cmp” developed by D. Roodman (2009) for this estimation. The χ2 – tests of the significance of our instruments are shown in the last row of Table 8. Our instrument (the gender composition of the older siblings) is significant at p < 0.01 in the desirability estimation for the urban sample and is significant at p < 0.1 for rural sample. We address a concern about the robustness of our results in the presence of weak instruments by estimating the binary outcome equation by the linear probability model. In the estimations on both the urban and on the rural sample, the F-statistics on our excluded instruments are larger than the “rule-of-thumb” critical value for Stock-Yogo test (F-statistic 10) for weak instruments (Stock and Yogo 2002).
↵18. Alderman and Lokshin (Forthcoming) demonstrate that selective mortality has only a minor impact on the measured anthropometric status of children or on that status distinguished by gender.
↵19. We find a good correspondence between the measured weight and the children’ss weight at birth assessed by their mothers. For example, in 1992 round of NFHS, only about 3 percent of children who were assessed as large at birth had measured weight in the first quintile of the weight distribution. Seidman et al. (1987) shows that about 75 percent of self-reported birth weights were accurate within 100 grams. A study by Adegboye and Heitmannb (2008) that uses weight categorization similar to NFHS concludes that maternal assessment of their children’ss weight at birth “seems to be sufficiently accurate for clinical and epidemiological use.”
- Received May 2009.
- Accepted March 2011.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.