1 Introduction

The association of birth months with educational performance and long-term cognitive achievements has been noted for over half a century in the western countries. Specifically, various studies have shown that children born in the summer months tend to be slow learners and have lower achievement test scores (e.g., Freyman, 1965; Pidgeon, 1965). More recent studies also found that youngest students born in summer months in grade schools in the UK perform less well than the children born earlier (Bell, 1990; Sharp, 1995) and tend to have more learning difficulties (Martin et al., 2004). Besides the differences in cognitive and academic achievements, summer born children are also found to be more likely to have behavior problems (Goodman et al., 2003). Up to date, two perspectives, the age position effect and fetal origins, have been proposed to explain the associations between birth months and educational performance and long-term cognitive achievements.

The “age position” effect induced by school entry policies has been invoked by education researchers for the association of birth season with educational performance and behavior problems in western and developing countries (Dhuey et al., 2017; Melkonian & Areepattamannil, 2018). Because children usually start formal schooling in the fall month of September, those born in the summer before September are usually the youngest in the grade, thus putting them at a disadvantaged due to immaturity. Although early studies indicate that advantage of older students than their peers may not persist beyond primary school (e.g., Shepard & Smith, 1986), recent studies based on longitudinal data and advanced methods have demonstrated long-term beneficial effect of relatively older age positions on educational attainment, cognitive performance and other social outcomes. One study (Dhuey et al., 2017) used large-scale population level birth and school data from Florida to compare the cognitive development of children born in August and September from age 6 to 15. Using a regression discontinuity design, the study reveals that older children in a grade (born in September) tend to perform better in both reading and mathematical test scores across a range of heterogeneous groups. They also found that older age at school entry increases college entry and reducing likelihood of arrests for criminal behavior. Similarly, in a study of children in Scotland, Lawlor et al. (2006) found some seasonal pattern in children’s school performance with children born in autumn and early winter having the highest scores in reading ability at age 9 and arithmetic ability at age 11. However, after adjusting for age at school entry and relative age to class peers, the association between season of birth and intelligence become attenuated and insignificant. Therefore, the authors concluded that the apparent season of birth patterning was largely caused by the school entry policy that leads to the age position effect. Similar findings about the abrupt difference between students born in two adjacent months that flank the start of the school year were also documented in Britain (Day et al., 2015) and other countries (Melkonian et al., 2018). According to this line of reasoning, the association of birth season with school performance is better explained by the relative age position of the student in a grade instead of the seasonal variation in biological risks during prenatal and perinatal periods.

The effect of age position that favors older over younger kids in a peer group is theorized to be caused by their psychological maturity and resilience, their competitiveness and confidence, and teachers’ perceptions and extra attention (Goodman et al., 2003). The age position effect is also shown to affect child mental development with youngest children of the school year being more likely to treated for attention deficit hyperactivity disorder (ADHD) Chen et al., 2016a, b; Elder et al., 2010; Whitely et al., 2017).

However, the age position effect due to school entry policies may not be the only explanation for the different school performance for children born in different months. A study of primary school pupils in Northern Ireland reveals robust birth season effect in several cognitive measures such as reading and spelling, particularly in early years of schooling before age position effect at school takes hold (McPhillips & Jordan-Black, 2009). Therefore, whereas age position in a grade cohort may serve as a proximal cause for certain differences in student educational and cognitive performance, there should be distal and more fundamental causes at work.

An alternative hypothesis attributes the differences in student and adult intelligence levels to seasonal exposures such as ambient temperature, maternal nutrition, or infection during pregnancy that affects fetal brain development during the critical periods. The “fatal origins” of health theory has been posited and examined as early as the 1990s. Baker (1995), among others, conducted a series of epidemiological studies in England and confirmed the fatal and infant origins of adult diseases due to low birthweight as a result of unfavorable maternal environment such as fetal malnutrition (Scrimshaw, 1997). Later longitudinal studies have found significant long-term impact of in utero fetal life shocks such as influenza exposure and prenatal nutritional deprivation on later education, health and earnings (Almond, 2006). Nutritional shocks such as Ramadan fasting for the Muslims have negative effect on birth weight and long-term adult outcomes such as educational attainment, mental health and income (e.g., Greve et al., 2015). The prenatal exposure to the severe Dutch famine during the winter of 1944 are found to significantly increase adult hypertension and mortality (Ekamper et al., 2014, 2015; Stein et al., 2006). On the other hand, nutritional supplementation program during pregnancy has significantly positive effect on the weight of toddlers, while nutritional supplementation for the babies after birth have little effect, indicating the uniqueness of the prenatal period (Linnemayr & Alderman, 2011).

Despite these findings, the fetal origins hypothesis has not spell out clear mechanisms for the season of birth effect on birth outcomes. It involves the interaction of biological and environmental factors such as the seasonal patterns of nutritional deficiency and variant exposure to sunlight during pregnancy, or adverse effect of seasonal influenza on pregnant mothers. Most recent literature has just begun to identify clear seasonal patterns in birth outcomes and potential mechanisms for the seasonality. Based on analysis of nearly half a million people in the UK, one study revealed the season of birth effect on birth weight and adult short stature, and even educational attainment (Day et al., 2015). The season of birth effect is attributed to the different levels of exposure to sunshine during pregnancy, because sunshine exposure is directly related to vitamin D levels vital for feudal development. Another study of birth records in several US states comprising over 3 million children reveals significantly shortened gestation periods for babies conceived in the first five months of the year, especially for May conceptions (Currie & Schwandt, 2013). More important, the authors also find that the seasonal pattern of gestation length correlates well with expecting mothers’ exposures to influenza season in the winter and spring months peaking in January and February. As a result, babies conceived in May who are expected in mid-February are more likely to have a premature birth in late January and early February due to exposure to peak influenza season in last trimester of the pregnancy. However, even after controlling for gestation length and family social economic conditions, birth weight is still found to have clear seasonal patterns. Summer conceptions in the months of June to August (with births delivered in March, April and May) have significantly higher birth weight; they also correspond to significant maternal weight gain during pregnancy, signifying better nutrition status for these mothers. However, the authors do not dwell on the causes of nutritional advantage of these summer conceptions. A recent study used US-wide birth data and county-level influenza prevalence to examine the effect of influenza exposure during different trimesters of gestation on fetal development including birth weight, gestation age, and infant mortality (Dorélien, 2019). Besides documenting the seasonality of birth outcomes, the study reveals the early and late trimester exposure to influenza are associated with higher infant mortality rates and premature birth and low birth weights. As various studies have established that preterm birth and low birthweight are associated with later developmental delays, learning difficulties and poor school outcomes (Kirkegaard et al., 2006; Hochstedler et al., 2020), it is plausible that seasonal variation in birth outcomes is an essential link to the different adult educational attainment and health outcomes.

In China, there are very few studies on the relationship of the season of birth with short- and long-term health and cognitive outcomes, although ample research has examined the effect of maternal nutrition and exposure to nutritional shocks on birth outcomes and long-term health. Several studies have revealed that fetal and early life exposure to the severe famine from 1959 to 1960 are associated with adverse adult health outcomes such as anemia, dyslipidemia, and hypertension (Shi et al., 2013; Wang, 2017), although a review article suggests that health disparities between famine and post-famine cohorts may be largely attributed to age differences (Li & Lumey, 2017). On the effect of early life famine exposure on adult cognition, limited evidence shows a cognitive deficit for the birth cohort with full-term prenatal exposure to famine (Xu et al., 2018) compared to both the early and later cohorts. One study that specifically addressed the relationship between birth season and child development in China. The study of a sample of infants from a poor mountainous region in northwest China revealed significant advantage for infants born in winter months versus those in spring months in both cognitive functioning and physical motor skills when they were six months old (Bai et al., 2018). Instead of considering the in-utero origins of infant outcomes such as maternal nutrition, the authors argued that winter-born infants benefited from their more frequent outdoor activities in their early infancy of spring and summer months. Using a national sample (CFPS), Maitra et al. (2022) recently founded that child’s season of birth significantly affected the height of girls aged less than 5 years in agricultural households of China, but did not affect their weight. And such initial disadvantages had little effect of season of birth on the cognitive and non-cognitive ability of children in adolescence. Liu et al. (2020) found that learning capacity for children born in April–May–June was better than those born in 1st and 4th quarters (January–February–March and October–November–December) due to seasonal variation in natural ventilation and indoor-outdoor temperature, based on a cohort study in ten schools from tow northeast China cities.

In this study, we will examine the association of birth months with adult educational attainment and cognitive ability using the 1963–1975 birth cohorts in rural communities from the 2010 baseline survey data of China Family Panel Studies (CFPS). These cohorts are chosen because the prevalence of undernutrition for the rural population, especially women, during this period should affect birth outcomes and long-term health and intelligence of the children. We hypothesize that seasonal change in nutritional status such as during the Chinese New Year in January and February should affect fetal health status and thus adult educational attainment and cognition. We do not choose the earlier cohorts because the famine during 1959-61 may disrupt the normal seasonal pattern if there is any. Based on research in both western societies and developing countries, we hope to shed further light on the variation in educational attainment and cognitive skills for people born in different months and refine and extend our understanding of the in-utero mechanisms for the seasonality of child outcomes.

2 Data, measurements and methods

Launched in 2010, the CFPS is a longitudinal survey of families and their members in China. The sample households are selected using a stratified three-stage cluster sampling method with three levels: counties, communities and households, and each selected household is followed once every two years since 2010. The 2010 baseline survey has a nationally representative sample of 15,000 households from over 640 urban and rural communities throughout China (Xie & Hu, 2014). This study is based on data of over 8,500 respondents of the 1963 to 1975 birth cohorts who were born and grew up in rural areas. Although some respondents are no longer living in rural communities in 2010 due to rapid urbanization and labor migration since 1990s, they all had rural hukou registration at birth and at age 12.

The three outcome variables we examined are: (a) educational attainment, indicated by whether the respondent has obtained high school education or above, and (b) scores of two cognitive tests: mathematics test and Chinese vocabulary test. Administered by the survey interviewer, the mathematics test requires the respondents to solve 24 math problems to measure their basic math skills. The vocabulary test is a cognitive test of language ability designed by CFPS for recognition of 34 Chinese words with various levels of difficulty. The test scores are just the simple count of correct answers, with higher scores indicating high levels of cognitive ability. The predictor variable of our focus is the month of respondent’s birth. We group two adjacent months together to form six categories of birth months: January/February, March/April, May/June, July/August, September/October, November/December.

We also introduce a set of control variables in our analysis that may affect educational attainment and cognitive ability of the respondents. The year of birth ranges from 1963 to 1975, which was the last stage of Mao’s reign characterized by economic stagnation and low living standards especially in rural areas. Earlier birth cohorts are excluded due to the great famine during 1959 to 1961. Respondents’ gender and birth order are included to account for the possible impact of male gender preference and family support of older siblings for education. Mother’s age at the respondent’s birth has proved to affect birth outcomes and child health. Mother’s and father’s educational levels are also included since parental education capital at a child’s birth and school age will affect child development and educational performance.Footnote 1 Since political privilege in the party state under Mao has been shown to affect children’s chances of receiving school and college education (Liu, 1999), we identify whether the parents were communist party members and/or held administrative or cadre positions. Although CFPS collected no information about family income at the respondents’ birth and school age, all rural households in the 1960s and early 70s were generally poor and rural economic inequality was not salient.

Table 1 Descriptive statistics of rural respondents of the 1963 to 1975 birth cohorts
Full size table

Descriptive statistics of the respondents’ educational attainment, test scores, and the predictor and control variables are shown in Table 1. One thing to note is the “baby boom” in the 60’s after the Great Leap famine (1959–1961) and low number of births in the 70’s due to initial implementation of the national family planning policy. Also noteworthy is the high number of births in October. This has been attributed by Chinese researchers to the high concentration of weddings during the Chinese New Year (in late January and early February) (Liu, 1989). As shown in the table, the overall percent of all respondents with high school education or above is very low at just 12%, but still higher than their parent’s education (7% for father and 2% for mother, respectively). This is understandable because it was not until 1986 that China promulgated its first nine-year compulsory education policy. Even after completing the nine years’ primary school and middle school, only a small proportion of rural students can pass the highly competitive entrance examination for the selective-enrollment high schools. Furthermore, the tuition and fees for high school and the low acceptance rate of colleges after graduation also kept high school education out of the reach of most rural students.

This study will adopt both bivariate analysis and multiple logistic and linear regression models to examine the association of the month of birth with educational attainment and cognitive test scores. The bivariate analysis will reveal the unadjusted association. The multiple regression analysis will adjust for the respondent’s demographic characteristics, parents’ education and political status, and the home province to obtain more robust estimates of the relationship between birth month and the education and cognitive outcomes.

3 Results and findings

3.1 Bivariate analysis

First, we examine the association between birth month and educational attainment using the Pearson χ2 test. The test statistic indicates significant differences in percentage of respondents with high school education among the different birth month groups (see left panel of Table 2). Further analysis shows that respondents born from March and April have the lowest education level, with less than 10% of high school graduates, significantly lower than those born in November/December with 15% of high school graduates.

Table 2 High school degree and cognitive test scores for rural adults in 1963 to 1975 birth cohorts by birth month
Full size table

Table 2 also shows the one-way ANOVA results that compare the mean word and math test scores of respondents born in different month categories. The F-test statistics indicate significant differences in both the average math and word test scores across the birth months. Similar to monthly distribution of the high school education, respondents born in March and April have significantly lower average test scores than those born November/December and September/October scores.

3.2 Multiple regression estimates

Having revealed the relevance of birth months for adult education and cognitive ability, we need to examine whether such a relationship may be spurious and confounded by various individual, family and regional factors that can affect both cognitive test outcomes as well as patterns of birth months. Several studies in the US have documented different patterns of conception and births throughout months of the year among people of different social-economic status (Dorélien, 2019). We conducted multiple regression analysis to account for the confounding effects of different control variables. Individual level control variables are the respondent’s gender, birth order and the year of birth. The provincial indicators are included to adjust for regional differences. Finally, the mother’s age at birth, parent’s educational level and political affiliation are added to address the family influence on respondent’s education and cognitive development.

Table 3 Results of logistic regression on association of birth months with educational attainment
Full size table

Table 3 shows the results of logistic regression models that examine the association of birth months with respondent’s probability of receiving high school education. The odds ratios in the unadjusted model (Model 1) tells us that the odds of having a high school education for November/December-born respondents are 1.63 times the odds for those born in March and April. To a lesser extent, respondents born in September and October are also significantly more likely to obtain high school education than those born in March and April. Adding more control variables in later models increased the goodness of fit because correspondents’ gender, birth order and birth year as well as parents’ education all have strong effect on the respondent’s probability of obtaining high school education. However, even after adjusting for the relevant control variables, the odds ratios and significance levels of the birth months have barely changed, indicating robustness of the birth month effect on educational attainment (Fig. 1A).

Fig. 1
figure 1

The coefficients of birth months predicted by the final models for educational attainment, word test and math test, respectively

Full size image

Having confirmed the significant association of birth month with educational attainment, we now address the question whether the month of birth has any significant association with adult cognition. At the baseline survey in 2010, correspondents of the 1963 to 1975 birth cohorts are already over 35 years old, long after their formal schooling. We adopt multiple linear regression to examine the effect of birth month on math and word tests after taking into account the respondents’ current education and logged family income besides the same individual and parental factors included in educational attainment analysis. The results of linear regression models on the association of birth months with word test are shown in Table 4. The unadjusted model (model 1) indicates a clear disadvantage in word test for respondents born in March/April compared to those born in November and December and to a lesser extent in September/October and January/February. Respondents born in November and December score 1.7 points (17% of one standard deviation) more than those born in March and April. As personal demographic attributes and parental factors are added in Model 2, the effect size of birth months is attenuated due to significant gender, birth order and birth year differences in test scores and the positive effects of higher parental education.

Table 4 Linear regression estimates of birth month effects on word test scores
Full size table

In the final model, we add the correspondent’s current family income and highest educational level achieved, which greatly improves the fit of model with a R-squared of 0.55. Both mother’s and father’s education and birth order are no longer significant predictors, indicating that parental education affects the correspondent’s adult cognitive achievement mainly through their contribution to educational attainment. However, even in the final model with all covariates added, the birth months still exert a lingering effect showing the persistent disadvantage of those born in March/April compared to those born in November and December (Fig. 1B).

The multiple linear regression results for math test scores are shown in Table 5. Similar to the models for word test, the unadjusted model (model 1) is followed by two additional models with individual, family and region control variables. Like the model results for word test score, the unadjusted model indicates an apparent deficit in math skills for people born in March/April, especially compared to those born in the fall and early winter months of September to November (coeff. of 0.99 and 0.98, about 17% of one standard deviation). With the addition of correspondent’s current educational level and family income, the final model achieves very excellent fit with a R-squared of 0.81. The results demonstrate the strong association of correspondent’s education level with math skills, while parental education works mainly through its contribution to child education. Although the coefficients of birth months are greatly reduced, respondents born in March/April still have significant disadvantages than those born in September/October (coeff. = 0.25, p = 0.006), indicating a lingering effect of birth month well into adult years (Fig. 1C).

Table 5 Linear regression estimates of birth month effects on math test scores
Full size table

4 Sensitivity checks

To check the robustness of our findings, we re-analyzed the data using Stata’s survey data analysis that take into account survey design factors and sampling weights. With the respondents born during 1963 to 1975 as the subpopulation, we ran a series of survey regression models with the same covariates to estimate the association of birth months with educational attainment, and cognitive test scores. Results of the survey regression models on educational attainment are similar to those shown in Table 3, indicating significantly lower probability of obtaining high school degree for those born in March/April than respondents born in November/December. For word and math tests, the unadjusted model shows significant deficits of the March/April-born respondents compared to November/December. However, in the final models with parental education and respondent’s education included, magnitude of differences in the coefficients between the peak winter months and spring months are attenuated and no longer significant at the 5% level.

To examine regional differences of birth month effects, we divided the respondents into south and north regions based on their home provinces and analyzed them separately. The findings are similar for educational attainment in both southern and northern regions. Respondents born in November and December and, to a lesser extent, September and October are more likely to gain high school education compared to people born in March/April. However, in the South, people born in July/August are also less likely to achieve high school degree. For word and math tests, the unadjusted models indicate a significant disadvantage of people born in March/April compared to those born in November/December and September/October. In the final model with all covariates included, the sample in the south shows only marginally significant effect for birth months. This may partly be due to the smaller sample size of the respondents in the southern region.

As we mentioned, our study examined the respondents born from 1963 to 1975 during the last period of Mao’s reign when rural plight and economic deprivation was the norm. Based on the fetal programming hypothesis, we expect that the overall undernutrition and seasonal changes in nutritional intake by pregnant mothers will contribute to the seasonal patterns in cognitive differences. To check on the soundness of this hypothesis, we examined the educational attainment and cognitive performance of later and earlier birth cohorts. First, we examined the respondents born from 1976 to 1986 when China began to abolish the Mao’s planned economy and implement the 9-year compulsory education. With the living standards of rural population on the rise and malnutrition less of a problem, we expect the seasonal pattern of cognitive performance should not be as apparent. The multiple regression models on high school attainment indicate that respondents born in the late fall and winter months have higher probability of a high school education, but the months with lowest probability is not consistent for unadjusted (March/April) and adjusted model (January/February). For both word and math test performance, although the respondents born in December to November enjoys significant advantages, the slack months include both May/June and March/April births with different set of control variables. Then, we conducted a series of multiple regression models to examine the respondents born from 1950 to 1959. A period of relative economic stability ended with radical collectivization leading to the great famine in 1960 and 1961 and later rural stagnation. The famine and the subsequent “great cultural revolution” started in 1966 disrupted the life and normal education of these cohorts of rural children. Our analysis for the 1950’s cohorts did not reveal any significant seasonal patterns for either high school education or cognitive performance.

We do not choose the urban sample for our study because of the relatively higher living standards in urban areas where prenatal nutritional deficit is not as much a problem as for rural women. Besides, with high school education more accessible in urban areas, high school enrollment is much less selective. Thus, we do not expect any strong birth month effects on educational attainment and cognitive performance. With this caveat in mind, our examination of the urban sample from 1964 to 1975 shows that urban respondents born in November/December have a higher probability in obtaining high school education than those born in January/February, with a marginal significance of p = 0.057. However, birth effects have no significant effect on word and math test scores.

5 Summary of findings

Our analysis reveals significant association of birth months with educational attainment and cognitive performance for rural residents born during the lean years of the 1960s and early 1970s in China. People born in the winter months of November and December and late fall of September and October are more likely to earn high school degree than those born in late spring months of March and April. Similar seasonal patterns are revealed for cognitive performance even after school education level is taken into account, which extend the previous study with long-term effect (Bai et al., 2018). The word recognition test shows a significant advantage for people born in November and December over those born in March and April. For math test, people born in September/October also excel over those born in March and April. These associations are robust even with the adjustment of relevant individual and family covariates.

Considering that the survey was conducted in 2010 when the youngest birth cohort are already in their mid-30s, the significant association of their birth months with the performance in the cognitive tests is truly remarkable. Although the effects of birth months are moderate, they are meaningful in practice. For example, the mean-adjusted predicted probability for high school education is 11.5% for the November/December-born respondents, over 4% points more than the 7.4% for respondents born in March/April. Considering the low accessibility of high school education, 4% is a big difference. For the word test, the predicted score in the unadjusted model for November/December-born respondents is 17.96, a difference of 1.72 point compared to the score of 16.24 for March/April. With an overall standard deviation of 9.8, the difference in the test score is about 17% of a standard deviation. For the math test, the unadjusted model gives a predicted test score of 9.79 for September/October vs. 8.80 for March/April. The difference of 1.0 is over 17% of a standard deviation (SD = 5.80) and is significant.

Our analysis takes into account individual demographic attributes and parental factors as covariates as they can affect education attainments (see Table 1). Our results from the regression analysis shows that the year of birth indicates that the probability of obtaining high school education has dropped from 16% to 1963 birth cohort to the lowest 8.2% for birth cohort 1967, and only cohort 1975 surpassed cohort 1963 reaching 18%. Gender disparity is also revealed by the much higher probability of high school education for males. Birth order makes a big difference in high school education with a clear advantage for the eldest sibling and a clear disadvantage of the middle child. Respondents born to a mother younger than 20 have lower probability of getting a high school education. Parents’ educational level is a strong predictor of the respondent’s high school education. However, parents’ political status as a party member or administrative position is only marginally significant for high school education when parents’ education is adjusted.

6 Possible mechanisms

In this section, we will examine several potential mechanisms for the association of birth season with educational attainment and cognitive functioning. As we have reviewed in the introduction, the “age position” effect induced by school entry policies on student school performance and later educational attainment and cognitive ability have been widely confirmed in different countries (Day et al., 2015; Dhuey et al., 2017; Melkonian & Areepattamannil, 2018). Older children in a grade level tend to have better academic performance and higher educational attainment. The advantage of older kids in a grade cohort has been attributed to their relative psychological maturity, competitiveness and confidence, as well as teacher’s extra attention. The “age position” theory does conform to certain findings in our analysis. We found that rural respondents born in September and October (the school year starts in September) have the highest average scores in math tests, compared to those born in the spring months of March/April/May. However, there exist obvious deviations from predictions of the age position theory. Although our findings generally confirm the better cognitive performances of those born in fall and early winter months, the peak birth months for better educational attainment and word test score is not in September and October, but in November and December. Besides, the lowest performance for math and word tests and educational attainment is associated with the birth months of March and April instead of late spring and summer months. The underperforming respondents born in spring are clearly not the youngest in their school grade. Therefore, other mechanisms must also be at work to produce the birth season effect on cognitive performance.

In the past two decades, various studies on the birth outcomes and later health ramifications have accumulated ample evidence in support of the “fetal programming” hypothesis that points to the intrauterine origins of birth outcomes and later child development and health. Fetal nutrition status and environmental exposure during pregnancy such as ambient temperature, air pollution and disease prevalence are found to have direct relevance to birth outcomes and later development of children. Recent studies in the US have pointed to the strong correspondence between seasonal influenza and birth outcomes: exposure to seasonal influenza in the 3rd trimester often trigger preterm birth leading to lower birthweight of the children. As the seasonal influenza often peaks in the first months of the year, a higher proportion of the children born in late January and early February (who are conceived in May) tend to suffer a shortened gestation with a lower birth weight (Currie & Schwandt, 2013). However, exposure to influenza around conception and in the first trimester can lead to structural defects of embryo leading to fetal loss and infant death (Dorélien, 2019).

The influenza season in China usually peaks in winter months of December and January in the north and in the spring months in the southern areas (Gao et al., 2009). Although we have found no systematic studies of in utero exposure to seasonal influenza on birth outcomes in China, the seasonal patterns of influenza prevalence should not bode well for children born in the late winter and spring months. They should have higher proportions of the preterm births and infants with low birth weights, which in turn can adversely affect cognitive development and education attainment. Unfortunately, the CFPS survey does not provide adult respondents’ birthweights or gestation periods. Only children who are born after 1995 and aged 15 and younger have birth weights. A simple means test indicates that the sample of rural children born in March have the highest proportion of low birthweight ( < = 2.5 kg), but it is not significantly different than for the winter months of December, September and January.

Limited research on the effect of ambient temperature on birth outcomes in China shows low temperature may be a risk factor for preterm births, unfavorable for infants born in the winter months (Liang et al., 2016). Air pollution, especially in urban areas, has become a serious problem with rapid industrial development in the last three decades. Several epistemological studies in China have revealed strong association of maternal exposure to air pollution such as PM2.5, NO2 and other pollutants on child birth outcomes (Wu et al., 2018; Liang et al., 2019). However, air pollution did not pose a real problem in rural China in the 1960 and 1970 s when our respondents were born.

However, as we described earlier, the undernutrition due to food insecurity and insufficient nutrients intake in rural China during Mao’s reign did affect maternal fertility and birth outcomes (Cai & Wang, 2005; Song, 2012) and contributed to long-term adult morbidity and mortality (Shi et al., 2013; Wang et al., 2017). Recent studies have corroborated the adverse effect of the great famine around 1960 on the cognitive functioning of adults who were born or conceived during the famine (Kim et al., 2016; Xu et al., 2018). Other studies have documented the maternal dietary patterns and nutritional inadequacy for pregnant women in rural areas in recent decades (Cheng et al., 2009; Gao et al., 2013) and their effect on negative birth outcomes (e.g., Ronnenberg et al., 2003).

Besides the overall bleak nutritional status for rural residents in China before the 1980s, there also existed apparent seasonal patterns of nutritional intake (Han et al., 2010). Rural residents often scrimped and saved throughout the year just to have a well-provided Chinese New Year that usually falls in later half of January and first half of February, when people can eat their fill with meat, cereal and vegetables. The holiday season is also a favorite occasion for wedding ceremonies and banquets, providing an opportunity for people to gain nutrition, especially animal protein. These macro nutrients are necessary to sustain the nutritional requirement for a healthy pregnancy and essential for the fledgling embryo as it is forming major functioning structures. Epidemiological studies on maternal nutrition and birth outcomes have revealed the adverse effect of nutritional deficiency during periconceptional period and early pregnancy on child birth outcomes (e.g., Ekamper et al., 2015; Chia et al., 2019; Abu-Saad et al., 2020). Therefore, it is plausible that the food bonanza around the Chinese New Year in late January and February should benefit children conceived during this period (with expected births in October, November and December). Given the established associations of preterm birth and low birth weight with adverse health and cognitive outcomes later in life (Hochstedler et al., 2020), it is not surprising that rural respondents born in November and December should have significant educational and cognitive advantages. In the same vein, the well-known nutrition deficit in the months of May and June in rural China when food grain from past year’s harvest was almost depleted and new crops was not yet ripe for harvesting may well lead to the disadvantages of children conceived in May or June and born next March and April.

In summary, results of our analysis tend to support the fetal origin theory that pins down the source of adult health and intelligence on the in-utero environment critical for fetal development. Our findings not only prove to be consistent with the fetal origin hypothesis, but also point out that context matters. Nutritional intake in rural China of the 1960s and 70s demonstrated clear seasonal patterns: the relative food abundance during the Chinese New Year (in late January and early February) and the lean months of spring with food shortage (in May and June). Since adequate maternal nutrition during the periconceptional period is critical for child birth outcomes, we believe that the seasonal patterns of nutritional intake by rural women may well explain the association of birth months with educational attainment and adult cognition of their children.

7 Limitations and further research

Although our study has revealed a significant association of birth months with educational attainment and adult cognitive ability, one limitation of the study is that limited information is available about respondents’ family and parents at the time of birth and during school. We only know the parents’ educational level and, to some extent, political affiliation. Although rural China during the Maoist period is largely an egalitarian society, there did exist certain economic disparities among households that might affect birth outcomes and school achievement. Lacking the information on family income or living conditions in our analysis may cause omitted variable bias of the estimated effects. Besides, since birth outcomes such as birth weight or gestational length are not available, we are unable to examine the seasonal patterns of birth outcomes and explore the pathways that links birth months to educational attainment.

We would also prefer to conduct a family fixed effects analysis in order to draw more robust causal inference on birth-month patterns of educational attainment and cognitive performance. Although we know the number of old and young siblings of the respondents, many respondents do not provide information about their siblings’ educational level. Besides, none of the siblings are interviewed or take the math and word cognitive tests. Therefore, a family fixed-effects analysis comparing the siblings in the same family is not feasible.

Although our study lends support to the fetal origin theory of educational attainment and adult cognitive ability, our study has only scratched the surface of a very important issue. Understanding seasonal patterns of birth outcomes and its association with later cognitive development and physical and mental health have not only academic value but also practical significance for public health. It will help shed light on the roles of various ecological factors for maternal health and fetal development.

Although this topic has been widely studied by scholars of various disciplines in Western countries, very little research has been done to examine the factors affecting seasonal patterns of birth outcomes in China. Due to data limitations, we can only surmise about the possible impact of seasonal pattern of nutritional conditions during pregnancy on adult educational attainment in rural China. Although with economic development people in China is undergoing a nutritional transition, there are still pockets of poverty where malnutrition affects the health of women and children (Wang et al., 2011; Zhang et al., 2016; Li et al., 2017; Yang et al., 2017). As data systems on maternal health, birth registration, disease control and meteorological information become increasingly available, rigorous epidemiological research can be conducted to learn about the relationship of different environmental factors at each stage of pregnancy with the birth outcomes and child development. Equipped with the insights into mechanisms of the intricate human-environment interaction during pregnancy, we will be better able to inform policies and practices aimed at improving the health conditions of mothers and children during the critical development windows.