Abstract
Obesity not only leads to immense medical costs associated with treating obesity-related illness but is also associated with lower employment prospects and earnings. This study shows that sunshine-induced vitamin D may have a preventive effect on obesity for children. We investigate the relation between sun intensity from pregnancy until infancy on obesity at age six, using population data of more than 600,000 children. Our findings show that the effects of sun intensity on subsequent obesity are concentrated in the first six months of life: 100 hours of additional sunshine over this period reduce overweight by 1.1 percent and severe obesity by 6.2 percent. We offer two main explanations for this pattern. First, infants’ vitamin D levels are particularly sensitive to sunshine in the first six months of life, when lactation is highest. Second, the first six months of life are a sensitive period for later obesity, as this is the period when infants rapidly gain weight, and adipose tissue develops.
I. Introduction
Childhood obesity in the United States has risen from 5 percent to 10.4 percent between 1974 and 2000 among two-to-five-year-olds and further increased afterwards (Cawley 2015). Other Western countries show similar trends (for example, Zellner et al. 2007 for Germany; Black, Hughes, and Jones 2018 for Australia). Obesity causes large costs, estimated at $14.1 billion per year in the United States through higher medical services utilization of the affected children alone (summarized in Cawley 2010). Childhood obesity also strongly correlates with later obesity; about a third of obese preschool children and about half of obese school-age children become obese adults (Serdula et al. 1993). Adulthood obesity not only leads to immense medical costs associated with treating obesity-related illness, amounting to $147 billion dollars per year in the United States (Finkelstein et al. 2009), but is also associated with lower employment prospects and earnings, in particular, for women (for example, Averett 2014). The long-term societal and individual costs of childhood obesity therefore greatly exceed its direct medical costs and make it vitally important to better understand its causes and ways to prevent it.
In this work, we seek to understand the effects of vitamin D on childhood obesity. While an existing literature has explored this relationship, it has been challenging to disentangle the impact of vitamin D exposure from that of other socioeconomic factors. We shed light on this issue by exploiting plausibly exogenous variation in vitamin D levels through sunlight exposure—the primary source of vitamin D for children and adults (for example, Hossein-Nezhad and Holick 2013). There is clear evidence of correlations between individuals’ exposure to sunlight and vitamin D levels on the one hand (for example, Gorman et al. 2017) and between vitamin D levels and obesity in adults and children on the other hand (Pereira-Santos et al. 2015; Turer, Lin, and Flores 2013; Wortsman et al. 2000).1 The direction of causality, however, is unclear (Vanlint 2013). Do higher levels of vitamin D levels reduce the risk of obesity? Or do obese individuals spend less time in the sun, leading to lower vitamin D levels in their bodies? Our analysis is the first large-scale study that establishes a causal link from children’s early sunlight exposure that induces vitamin D production to subsequent childhood obesity. Although we do not directly measure the impact of sunshine on individual vitamin D production, we consider vitamin D the most plausible channel through which sunshine affects obesity.
Two recent studies adopt a similar design as us and establish a causal link between increased sunshine exposure during pregnancy and birth weight for Black (but not white) mothers who are particularly likely to be vitamin D deficient (Trudeau, Conway, and Menclova 2016) and between increased sunshine exposure during pregnancy and the risk of asthma later in life (Wernerfelt, Slusky, and Zeckhauser 2017). Our work complements these papers by considering an additional outcome—adiposity at age six—and by pinpointing the exact period—pregnancy versus the first three, six, 12. or later months of life—when sunlight exposure is most relevant for childhood obesity.
Our empirical analysis is based on data from compulsory school entry examinations and health screenings undertaken by pediatricians and covering all children prior to school entry at age six in 44 German districts over nearly two decades. We combine this data source with information on daily sun hours at the district level. To obtain the causal effects of sun intensity on childhood obesity during pregnancy and during different stages of a child’s life, our research design holds constant important confounding factors, such as district of birth, which may be correlated with children’s health through, for example, differences in lifestyles or diets (for example, Sofi et al. 2014; Tucker and Gilliland 2007); year of birth, which may be correlated with children’s health through, for example, business cycle effects (for example, Dehejia and Lleras-Muney 2004); and month of birth, which the existing literature has shown to correlate with demographic characteristics (for example, Currie and Schwandt 2013) that are themselves correlated with obesity.
Our analysis shows that 100 hours more sunshine in the first six months after birth—equivalent to the increased sun intensity from a two-and-a-half-week-long holiday in winter to a destination where sunshine is similar to that in summer—reduces overweight at age six by 1.1 percent and severe obesity at age six by 6 percent. These estimates imply 1,119 fewer overweight and 420 fewer severely obese children per birth cohort (677,947 children were born in 2010 in Germany) per 100 additional sunshine hours in the first six months after birth. In contrast, increased sunshine hours in the second half of the child’s first year and beyond have no detectable impact on overweight or severe obesity at age six. Increased sun intensity in the last trimester of pregnancy also tends to reduce the risk of subsequent adiposity, but this effect is smaller in magnitude than that of sun intensity in the first six months of life and statistically significant in only some specifications.
There are three main explanations for why the effects of sun intensity–induced vitamin D production own subsequent obesity is concentrated in the first six months of life. First, the first six months of life coincide with the period when lactation is highest. In this period, the typical infant acquires vitamin D primarily through the ingestion of breast milk, which is likely to contain lower levels of vitamin D when sun intensity is low. As the child gets older, baby food, which is typically enriched with vitamin D (for example, formula or baby porridge), becomes a more important part of the child’s diet, making their vitamin D levels less sensitive to sun intensity. Second, vitamin D may directly reduce infants’ weight and BMI in the first six months of life, as it affects lipogenesis, a metabolic process through which certain molecules are converted to body fat (Wood 2008; Abbas 2017). Since BMI in the first six months of life is strongly correlated with BMI later in life (for example, Taveras et al. 2009; Baird et al. 2005; Roy et al. 2016), increased sun intensity in the first six months of life may directly lead to higher risk of obesity at age six. Third, the functions of the adipose tissue primarily develop in the first six months after birth. Vitamin D deficiency hinders this development, increasing the risk of obesity later in life (Ding et al. 2012). The second and third explanations suggest that the first six months of life are a sensitive period for subsequent obesity.
Our paper contributes to the literature on the causes of childhood obesity. Existing papers have focused on high calorie intake (for example, Currie et al. 2010) and lack of physical activity (for example, Kimm et al. 2005) as the two main causes of childhood obesity. Our study adds an additional, so far unexplored, potential cause of childhood obesity, insufficient exposure to sunlight and the resulting vitamin D deficiency in pregnancy and early life. Even though the effect of sunshine in the first months of the child’s life on later obesity is considerably weaker than that of dietary intake or physical exercise later in life, our findings suggest that vitamin D supplements for infants, due to their low production costs, may be a cost-effective way to reduce childhood obesity.
A small number of studies have directly investigated the effects of vitamin D supplements early in life on children’s health outcomes later in life in randomized control trials.2 Hazell et al. (2017) report some suggestive evidence that vitamin D supplements in infancy reduces fat mass and body fat at age three. This study, however, is based on a small sample (132 children at baseline) and suffers from a large rate of attrition (only 66 percent of the 132 children participated in a follow-up at age three).
The randomized control study by Trilok-Kumar et al. (2015), which establishes a causal link between vitamin D supplements in infancy and height and weight at age five, is based on a considerably larger sample (2079 infants at baseline), although follow-up attrition is high in this study too, with only 912 children (43.8 percent) participating in a follow-up at age five. Our study provides several additional insights. First, while Trilok-Kumar et al. (2015) concentrate exclusively on infants with low birthweight and consequently oversamples children from disadvantaged families, we study the entire population of children. Second, while Trilok-Kumar et al. (2015) focus on BMI as an outcome and do not investigate the effects of vitamin D supplements along the BMI distribution, we provide an analysis on adiposity and thus focus on the effects of sunshine exposure on the upper tails of the BMI distribution. Third, while their investigation is for India, a developing country where both malnutrition and obesity are of concern, our study takes place in Germany, a highly developed country where malnutrition is very rare, but obesity is not. In consequence, the hypotheses in Trilok-Kumar et al. (2015) and our work are different. While Trilok-Kumar et al. (2015) conjecture that vitamin D helps children with low birthweight gaining weight, we instead hypothesize that sunshine-induced vitamin D reduces the risk that healthy children become overweight. Finally, in contrast to the study by Trilok-Kumor, our research design allows us to pinpoint the exact period when vitamin D matters for later obesity.
We further contribute to the small literature that establishes a link between sun intensity and health outcomes. Existing studies typically find that more sunlight is related to lower obesity (see Gorman et al. 2017 for a review). In contrast to our work, these studies typically focus on the contemporaneous link between sunlight and obesity, rather than on the effects of sunlight during pregnancy and early life on obesity later in life. Moreover, these studies exploit variation in sunlight either across districts (using variation in sun intensity due to differences in latitude and altitude as, for example, Woolcott et al. 2014 or Voss et al. 2014) or across seasons (for example, Dietz and Gortmaker 1984 or Visscher and Seidell 2004), and are thus likely to overstate the causal effect of sunlight on obesity that operates through raising vitamin D levels, as people are often more active and eat a healthier diet in more compared to less sunny areas or seasons, both of which are strong predictors of obesity (Sofi et al. 2014; Tucker and Gilliland 2007). Our research design eliminates these confounding factors by leveraging variation in sun intensity within districts and within seasons, across birth cohorts.
Overall, our findings highlight that increased sun intensity in the first six months of life may have a preventive impact on obesity, most likely through raising vitamin D levels. These findings have important public health implications, as vitamin D intake can be increased easily and at low cost through dietary supplements.
In Section II, we review the mechanisms through which increased sun intensity at various stages of a child’s life may affect current and subsequent obesity by raising vitamin D levels, as discussed in the medical and biological literature. In Section III, we describe the data sources used in the empirical analysis. In Section IV, we present the empirical strategy to identify the causal effects of sun intensity at different stages of a child’s life on childhood obesity at age six. We report findings in Section V and conclude with a discussion of our findings in Section VI.
II. Sun Intensity, Vitamin D, and Obesity
Several studies have documented a clear and robust correlation between vitamin D and obesity (Pereira-Santos et al. 2015; Turer, Lin, and Flores 2013; Wortsman et al. 2000). The direction of causality is, however, unclear. On the one hand, obese individuals may simply be less active and spend less time outside and may thus be less exposed to sunshine than nonobese individuals, leading to lower vitamin levels in their body (Vimaleswaran et al. 2013; Tucker and Gilliland 2007). At the same time, obese individuals may consume a diet that includes fewer vitamin D-rich products, such as fish, avocados, or mushrooms (Hyppönen and Power 2007). On the other hand, vitamin D levels may have a causal effect on obesity, as adequate vitamin D intake may enhance weight loss and decrease body fat. The argument here is that vitamin D is produced on the skin and in the adipose tissue under the skin. This process starts only when the skin is exposed to sunlight. During this process of vitamin D creation, extra calcium flows into fat cells, thereby affecting lipogenesis, a metabolic process through which certain molecules are converted to body fat (Martini and Wood 2006).
While this argument provides an explanation for why current vitamin D levels may causally affect current obesity risk, vitamin D deficiency in the first months of a child’s life may have long-lasting effects on obesity once the child is older, as the first few months of life are considered a sensitive period for the development of subsequent obesity, for two reasons. First, the infant’s weight more than doubles in the first six months and about triples in the first year of life (Butte et al. 2000). At the same time, rapid weight gain in the first three and six months of life is strongly associated with higher BMI later in life (for example, Taveras et al. 2009; Baird et al. 2005; Roy et al. 2016). Second, a large proportion of the adipose tissue develops during this period (Poissonnet, LaVelle, and Burdi 1988). The adipose tissue is a structure, like an organ, that has important functions in metabolism, particularly in storing and absorbing fat. Biological research suggests that vitamin D is essential for the tissue to sufficiently develop all its functions (Ding et al. 2012). If the adipose tissue has not developed its metabolism functions sufficiently well, it may store additional fat conditional on similar calorie intake, increasing the risk of obesity later in life.
A further explanation for why the effects of sun intensity on subsequent obesity are concentrated in the first six months of life is related to the idea that the infant’s vitamin D levels are particularly sensitive to sunlight in exactly this period. The first six months after birth are the period in which lactation is highest. Around the world, mothers are strongly encouraged to exclusively breastfeed their child for six months and continue breastfeeding up to the age of two years or beyond (for example, World Health Organization or the National Breastfeeding Commission for Germany). In Germany, 76.7 percent of children born between 1986 and 2005 were breastfed, with an average nursing duration of 4.59 months of sole breastfeeding and 6.9 months of breastfeeding with complementary feeding (Lange, Schenk, and Bergmann 2007). A newborn infant who is solely breastfed can acquire vitamin D through direct dietary supplementation, direct sun intensity, or ingestion of breast milk. In the United States, only between 10 percent and 16 percent of breastfed infants receive direct vitamin D supplements, according to a survey with more than 2,000 participants conducted between 2006 and 2008 (Taylor, Geyer, and Feldman 2010). Pediatric advice, such as that of the American Academy of Pediatrics, strongly advises against direct sun exposure of infants during the first six months of life (American Academy of Pediatrics, Committee on Environmental Health 1999). This leaves breast milk as the main plausible source of vitamin D. As maternal vitamin D supplementation during lactation leads to higher vitamin D levels in their children (Hollis et al. 2015), it is likely that increased sun intensity also leads to higher maternal vitamin D levels that are then transferred to the child through breastfeeding. As the child gets older, baby food—which is typically enriched with vitamin D (for example, formula or baby porridge)—becomes a more important part of the child’s diet, making their vitamin D levels less sensitive to sun intensity.
Similar to a breastfed infant, the vitamin D levels of a fetus are tied to those of its mother. Moreover, while adipose tissue primarily develops in the first six months of the child’s life, this process starts during pregnancy (Poissonnet, LaVelle, and Burdi 1988). Increased sun intensity during pregnancy, in particular during the last trimester, may therefore also lower the subsequent risk of adiposity. As vitamin D take-up is higher among pregnant women than among mothers who have just given birth (Aronsson et al. 2013), sun intensity during pregnancy may have a smaller impact on subsequent childhood obesity than sun intensity during the lactation period.
III. Data and Sample
Our study makes use of a unique feature of the German school system that requires all children to undergo compulsory school entry examinations prior to school entry at age six. Examinations are designed to assess children’s school readiness and identify any developmental delays or health problems, including child obesity. The 45-minute test is typically administered in a nearby elementary school in the child’s municipality in the spring before August school entry. Government pediatricians conduct the test, which includes an interview with the child, a battery of motor skills and physical development tests, and the precise measurement of height and weight. In contrast to most data sets used by economists, which rely on parents’ self-reports of their children’s weight and height, an important advantage of our obesity measure is that it represents standardized assessments by health professionals. This is a major advantage because subjective statements by parents about their children’s height and weight are prone to bias (for example, Cullinan and Cawley 2017; Weden et al. 2013).
Our school entry examinations data cover 16 birth cohorts (1986–2001) from 50 districts in three federal states in the North (Schleswig-Holstein), North-West (Lower Saxony), and East (Brandenburg) of Germany (Online Appendix Figure B1). In addition to precise information on height and weight, we observe some background variables, such as the child’s month and year of birth, district of residence, gender, and ethnicity. Additional socioeconomic background variables, such as parental education and unemployment, are included only for the more recent 1998–2001 birth cohorts.
We supplement the school entrance examination data with information on the sunshine hours in each district in each month of each year. Our measure for sunshine hours comes from 21 of the 74 weather stations across Germany that collect information on hours of sunshine on a daily basis. To create our baseline measure, we first sum up daily hours of sunshine in each month, in each year, and for each weather station. We then identify all weather stations within a 50-kilometer (31-mile) radius from the center of the district (there are at most four such weather stations per district), leaving us with 44 districts that have at least one weather station within a radius of 50 kilometers. After restricting our sample to children who are older than 67 months and younger than 81 months at the date of examination, and excluding children with implausible height and weight values, our main sample includes 666,258 children.3
Finally, we average sunshine hours per month and year over those stations, inversely weighted by distance. Sixteen of the 44 districts in our data are assigned the exact same set of weather stations as other districts, leaving us with 28 unique “weather station districts.” For each child in our sample, we finally compute total sunshine hours, divided by 100, in seven three-month intervals, starting from the beginning of pregnancy to the end of the first year. Our results are robust to alternative cutoff rules (for example, 50 miles rather than 50 kilometers) and alternative weighting schemes of weather stations (for example, uniform weighting across weather stations), as shown in our discussion of robustness checks below.
A drawback of our data is that we only observe the district of residence at the time of the school entry examination, as opposed to the district during pregnancy and the first months of life. Evidence from the Sample of Integrated Labor Market Biographies (SIAB) and the German Socio-Economic Panel (GSOEP) suggests that mobility across districts is relatively small, with at most 15 percent of first-time mothers residing in a different district when their child is six years old than during pregnancy (Online Appendix A). Importantly, it is highly unlikely that families base their mobility decisions on expected sunshine hours in the destination district. After all, future sunshine hours in a specific district, conditional on district-by-year of birth and month of birth fixed effects—the key control variables in our regressions—are impossible to predict. We therefore may assign wrong sunshine hours in pregnancy and infancy for a small fraction of children, possibly leading to unsystematic measurement error in sunshine hours, biasing our estimates toward zero. Our estimates may therefore be best understood as a lower bound for the adiposity-reducing effects of additional sunshine hours in infancy.
The literature predominantly uses the Body Mass Index (BMI, weight in kilograms divided by height in meters squared) to calculate measures of childhood obesity. However, because they show fast gender-specific developmental changes, fixed cutoff rules for obesity for children similar to those for adults are not adequate. Therefore, childhood obesity is defined in gender- and age-specific growth curves in historic reference populations (for example, Barlow 2007). Usually, a BMI above the 85th percentile defines children as overweight, a BMI above 95th percentile as obese, and a BMI above 99th percentile as severely obese. Following the literature, we construct three main outcome variables: overweight, obesity, and severe obesity, defined, respectively, as >85th, >95th, and >99th percentile in the gender- and age-specific (in months) BMI distributions in our data. The 85th percentile corresponds to a BMI cutoff of 17.07 for boys and 17.17 for girls, the 95th percentile to a BMI of 18.81 for boys and 18.90 for girls, and the 99th percentile to a BMI of 21.55 for boys and 21.70 for girls. These BMI cutoffs are in line with international samples that focus on children of similar ages (Centers for Disease Control and Prevention 2000). We assess the robustness of our results by using internationally defined cutoff points for overweight and obesity, as suggested by Cole et al. (2000), based on data from more than 90,000 children five to six years old who were born in the 1980s and 1990s in Brazil, Great Britain, Hong Kong, the Netherlands, Singapore, or the United States.4 We use the metric BMI measure, as well as height and weight, as additional outcome variables to investigate whether sunshine hours affect the entire BMI distribution or only its upper tail and to examine whether any effects on BMI are driven by reduced height or increased weight.
Table 1 provides a first descriptive overview of our sample. In our sample, boys are on average 120.5 cm tall and weigh 22.83 kg, compared to 119.6 cm and 22.45 kg for girls. The average BMI in our data is 15.65 for boys and 15.61 for girls. Based on the internationally defined cutoff points as in Cole at al. (2000), 11.70 percent of boys and 14.99 percent of girls in our sample are overweight, and 3.23 percent of boys and 4.00 percent of girls are obese. For comparison, in the United States, the country with the highest adiposity levels in the sample of countries examined by Cole et al. (2000), 18.1 percent of boys and 16.5 percent of girls are overweight, and 3.3 percent of boys and 4.0 percent of girls are obese. The relatively high rates of adiposity in our sample are in line with studies showing that the prevalence of overweight and obesity for school children is relatively high in Germany in comparison to other countries and close to that in the United States (for example, Wang and Lobstein 2006).
Descriptive Overview: Height, Weight, and BMI
The findings in Panel A of Table 2 highlight that childhood obesity is associated with socioeconomic background characteristics, in line with existing evidence (for example, Wang and Zhang 2006). Minority children (14.6 percent of our sample) are 2.7 percentage points or 19 percent more likely to be overweight and 1.5 percentage points or 32 percent more likely to be obese than majority children.5 Children of mothers with higher education (15.5 percent in our sample) are 22 percent less likely to be overweight and 34 percent less likely to be obese than children of mothers without higher education.6 Children of fathers who are not employed (13.5 percent in our sample) are likewise more likely to be overweight (by 12 percent) or obese (by 28 percent).
Obesity Risk by Socioeconomic Status and District Characteristics
The findings in Panel B of Table 2 further illustrate that childhood obesity correlates with district characteristics. The risk that a child is overweight or obese is 17 and 27 percent higher in districts with an unemployment rate above the mean. The associations between district GDP and measures of childhood obesity are slightly smaller.7
Table 3 and Figure 1 provide descriptive statistics of sunshine hours in our sample. On average, children in our sample experienced 1,223 hours of sunshine during pregnancy and 827 hours during their first six months of life. For comparison, average sunshine hours over one year are 2,771 in Athens, 2,524 in Barcelona, 1,625 in Berlin, 1,481 in London, and 1,203 in Glasgow. The standard deviations of sunshine hours during pregnancy and the first six months of life are 231 and 285, respectively. Children born in November receive the most (1,481 hours) and children born in May the least (958 hours) hours of sunshine during pregnancy. In contrast, children born in April enjoy the most (1,186 hours) and children born in October the least (443 hours) sunshine hours during the first six months of life. In our sample, sunshine hours during pregnancy were on average highest for the 1990 birth cohort (1,349 hours) and lowest for the 1988 birth cohort (1,129 hours), while sunshine hours during the first six months of life were highest for the 1989 birth cohort (969 hours) and lowest for the 1998 birth cohort (691 hours). Children born in the districts of Brandenburg a. d. Havel and Potsdam-Mittelmark were on average exposed to the most sunshine hours, and children born in the district of Leer were exposed to the least sunshine hours during pregnancy and the first six months of life (1,326 versus 1,083 hours for pregnancy and 885 hours versus 735 hours for the first six months of life). These figures illustrate that there is substantial variation in sunshine hours across districts, birth months, and cohorts.
Sunshine during Pregnancy and First Months of Life: Descriptive Statistics
Sunshine during Pregnancy and First Months of Life: Descriptive Statistics
Data sources: School entry examinations, Schleswig-Holstein, Lower Saxony, Brandenburg. Sunshine hours from 21 weather stations.
Notes: The figures display sunshine hours during the first six months after birth and during pregnancy by month of birth (Panel A), year of birth (Panel B), and district (Panel C) in our sample.
It should be noted that sunshine hours are generally higher in the federal state of Brandenburg (Panel C of Figure 1), a district in former East Germany, than in the two other West German states of Schleswig-Holstein and Lower Saxony. At the same time, the unemployment rate and GDP—both of which, as shown in Panel C Table 1, are associated with a higher obesity risk—are higher in Brandenburg than in the two other states. This correlation between sunshine hours and district characteristics underscores the need for a convincing identification strategy to isolate the causal effects of sunshine hours during pregnancy and in the first months of life on subsequent childhood obesity. We describe our identification strategy next.
IV. Estimation Approach
Our baseline specification is a linear probability regression that links adiposity indicators of child i in district r born in year y and month m to the sum of hours of sunshine exposure in specific age intervals (indexed by the subscript j) from the onset of pregnancy until the child is 80 months old:
1
Here, 
denotes total sunshine hours, divided by 100, during interval j for children residing in district r (there are 44 districts in our main sample) and born in year y (our sample includes 16 birth cohorts) and month m. In our baseline specification, we distinguish four age intervals: pregnancy, the first six months after birth, months 6–11 after birth, and months 12–80 after birth. In a more detailed specification, we distinguish eight age intervals: three three-month intervals during pregnancy, four three-month intervals during the child’s first year of life, and one interval from the child’s first birthday until they are 80 months old. District-by-year of birth fixed effects are denoted by δy,r, month of birth fixed effects by μm, and ɛi,r,y,m is an error term. We cluster standard errors by “weather station districts”—that is, districts with unique weather station combinations. There are 44 districts, 21 weather stations, and 28 “weather districts” in our main sample.8
The coefficients of interest γ j measure how 100 additional sunshine hours in interval j affect adiposity at age six. By conditioning on district-by-birth year fixed effects, δy,r, we allow for the possibility that children and their parents in districts with more sunshine have a different upbringing, lifestyle, and diet than children in less sunny districts (Sofi et al. 2014; Tucker and Gilliland 2007). Children in districts with more sunshine may also generally differ from children in districts with less sunshine. District-by-year of birth fixed effects further account for the possibility that the state of the business cycle, which may be correlated with sunshine hours, affects child health (Dehejia and Lleras-Muney 2004). Month of birth fixed effects, μm, address the concern that disadvantaged mothers (whose children tend to exhibit a larger risk of obesity) are more likely to give birth in seasons with less sunshine (Currie and Schwandt 2013). This approach essentially exploits variation in sunshine hours that stems from some districts, but not others, receiving an unusual amount of sunshine (relative to the district average) in specific months of the year.
V. Results
A. Baseline Results
We present our baseline findings based on regression Equation 1 in Table 4. The estimates in Column 1 show that increased sunshine hours during the first six months after birth reduce the adiposity risk at age six, regardless of which adiposity indicator is used. In contrast, the effects of sunshine during pregnancy, in months 6–11 after birth, or beyond the child’s first year on childhood adiposity are considerably smaller in magnitude and not statistically significant from zero. To assess the magnitude of the impact of sunshine in the first six months of life, consider an increase of 100 sunshine hours, which is equivalent to the increased sun intensity from a two-and-a-half-week-long holiday in winter to a destination where sunshine is similar to that in summer. This reduces the probability of being overweight by 0.167 percentage points or, as 15 percent of children in our sample fall into this category, 1.1 percent (0.00167/0.15). Moreover, 100 additional sunshine hours reduce obesity and severe obesity by 0.129 and 0.062 percentage points, or 2.6 and 6.2 percent, respectively. In terms of absolute numbers of adipose children, these estimates imply 1,119 fewer overweight, 868 fewer obese, and 420 fewer severely obese children per birth cohort (677,947 children were born in 2010 in Germany) per 100 additional sunshine hours in the first six months after birth.
Effects of Sunshine Hours during Pregnancy and First Years of Life on Adiposity Risk at Age 6
As mothers and infants only spend part of the additional sunshine hours outside in the sun, our estimates should be interpreted as “intention-to-treat” estimates and should be considered as lower bounds for the impact of direct sunshine exposure on subsequent adiposity. Evenly spread over the first six months of life, 100 additional sunshine hours correspond to 33 minutes of additional sunshine per day. To put this number into perspective, mothers would have to spend between three and eight minutes in the sun (in Boston, MA, from April to October at 12 p.m. EST, Terushkin et al. 2010) to synthesize vitamin D levels equivalent to the recommended daily supplement for breastfeeding mothers (400 International Units).
To give an additional interpretation of the effect sizes, we compare them with the effect sizes of interventions targeted to reduce child obesity. One such intervention is the Head Start program. Two papers have found that participation in this program at ages three to five reduces the risk of overweight (by 26 percent among boys aged 12–13, according to Carneiro and Ginja 2014) and obesity (by 2.3 percentage points, 14 percent, independent of gender, according to Frisvold and Lumeng 2011). Although the effects of participation in Head Start in reducing childhood adiposity are considerably larger than the effects of 100 additional sunshine hours that we uncover (2.3 percentage points vs. 0.13 percentage points), such preschool programs are far more expensive than vitamin D supplementation.
In contrast, Prina and Royer (2014) do not find significant effects on children’s obesity in an information intervention that aimed to increase parental knowledge and shift parental attitudes about children’s weight in Mexico. In a similar vein, Bhattacharya, Currie, and Haider (2006) report that a school breakfast program in the United States aimed to improve children’s diets had no significant impact on children’s risk of overweight or obesity. The latter two interventions clearly document the difficulties in implementing cost-effective strategies to reduce childhood obesity. The meta-study by Brown et al. (2019) further illustrates this difficulty. This study includes 16 randomized control trials with the explicit goal of preventing obesity in children and focuses on children’s BMI. The study finds that interventions that simultaneously target diet and physical activity reduced BMI by 0.07 kg/m2. In contrast, interventions that target only diet or only physical activity did not significantly reduce BMI. In comparison, our estimates, discussed in Section V, suggest that 100 additional hours of sunshine during the first six months after birth decrease BMI at age six by 0.02 kg/m2 for the whole population and by 0.11 kg/m2 for those children in the highest BMI percentile.
Since the production costs of vitamin D supplements are low, our findings support the view that vitamin D supplementation in infancy is a cost-effective way to reduce childhood obesity.
B. Robustness Checks
Our findings are robust to adopting alternative estimation methods, the inclusion of additional control variables such as district-by-month of birth fixed effects, alternative measurements of sunshine hours, and the inclusion controls for weather.
1. Alternative specifications and control variables
We display estimates obtained from a probit model in Column 2 of Table 4. Marginal effects from this specification are very similar in magnitude to those obtained from the linear probability model, presented in Column 1. Flexibly controlling for age by including dummy variables for each age month group interacted with gender in Equation 1 likewise hardly affects our estimates (Column 3). In Column 4, we estimate an even tighter specification than our baseline specification and replace the month of birth fixed effects (μm in Equation 1) with district-by-month of birth and year of birth-by-month of birth fixed effects. Our results are largely unchanged. Column 5 adds dummy variables for each age month group interacted with gender (as in Column 3) to the specification estimated in Column 5. Again, this has little impact on our estimates.
While the results so far refer to our baseline definitions of overweight, obesity, and severe obesity, in Column 6, we use internationally defined cutoff points to classify children as overweight and obese as in Cole et al. (2000). The pattern remains the same. Whereas sunshine hours in the first six months of life reduce the risk of childhood adiposity, sunshine hours during pregnancy and beyond the first six months of life have no significant impact on childhood adiposity. In terms of magnitude, estimates are similar to our baseline specification. For example, 100 additional sunshine hours in the first six months after birth reduce the risk of overweight at age six by 0.22 percentage points or 1.7 percent and the risk of obesity by 0.08 percentage points or 2.2 percent.9
2. Alternative measurements of sunshine hours
In Columns 2–5 of Table 5, we explore the robustness of our results to alternative measurements of sunshine hours. For comparison, in Column 1, we report our baseline estimates as in Column 1 of Table 4. In Column 2, we increase the distance rule to the next weather station and discard weather stations more than 50 miles (instead of 50 kilometers) away from the center of the district. In Column 3, we only use information from the closest weather station, while in Column 4, we uniformly weight across up to four weather stations in the district to compute sunshine hours in each interval. Our estimates are very similar across alternative specifications.
Effects of Sunshine Hours during Pregnancy and First Years of Life on Adiposity Risk at Age 6: Alternative Measurements of Sunshine Hours and Weather Controls
Sunshine hours in winter months do not generate the same vitamin D levels in the body as sunshine hours in summer months because the sun is less intensive during this period. Moreover, infants’ and mothers’ skin may be largely covered due to low temperature. To account for this, we weight our baseline measure of sunshine hours with a sinus function, where the months June and July receive the highest weight of one. Estimates, reported in Column 5 of Table 5, are virtually identical to those in our baseline specification. This is not surprising, as the specification controls for month of birth fixed effects, and conditional on month of birth fixed effects, the baseline and weighted measures of sunshine hours are highly correlated.
3. Controlling for temperature and rainfall
Weather may have a direct impact on childhood obesity or early fetal development, conditional on sunshine (for example, Barreca, Deschenes, and Guldi 2018). We probe the robustness of our estimates to the inclusion of weather controls in Columns 6 and 7 of Table 5. Adding average daily temperature (Column 6) or cumulative rainfall (Column 7) in each of the four intervals as controls to our baseline specification barely changes our estimates. In Online Appendix Table B1, we report results from a regression of children’s adiposity risk on average daily temperature and cumulative rainfall, instead of sunshine hours, in each of the four intervals. Coefficients are consistently small in magnitude and not statistically significant from zero. Overall, these findings highlight that weather in pregnancy and infancy primarily affects childhood adiposity through sunshine hours rather than temperature or rainfall, as we would expect if vitamin D generation reduces the adiposity risk.
4. Placebo regressions
As a final specification check, we randomly assign “fake” sunshine hours to children in each of the four intervals from pregnancy to age six to check whether regressions of adiposity risk on “fake” sunshine hours produces equally large effects in the first six months of life as our baseline regression. To this end, we randomly draw a month of birth, a year of birth and a district for each child and assign sunshine hours in each interval accordingly. We then regress adiposity risk at age six on “fake” sunshine hours during pregnancy, months 0–5, months 6–12, and months 13–80, controlling for the same variables as in our baseline regression. We do this 300 times. In Figure 2, we plot the cumulative distribution of coefficients that we obtain from these “fake” or “placebo” regressions for the zero to five months interval for which we find the largest effects. For all measures of adiposity, point estimates are closely centered around zero. For overweight, 13 out of the 300 point estimates are more negative than our baseline estimate of −0.165; for obesity and severe obesity, the number is even smaller (7 and 2, respectively). These findings cast some strong doubt that our estimates are simply spurious and driven by random variation in our data.
“Placebo” Regressions of Adiposity Risk at Age Six on “Fake” Sunshine Hours Zero to Six Months after Birth
Data sources: School entry examinations, Schleswig-Holstein, Lower Saxony, Brandenburg. Sunshine hours from 21 weather stations.
Notes: The figure displays the cumulative distribution of point estimates obtained from 300 regressions of adiposity risk at age six on “fake” sunshine hours for the interval zero to six months after birth for which we find the largest effects. For each child in our sample, we randomly assign a month of birth, a year of birth and a district. We then compute “fake” sunshine hours during pregnancy, the first six months after birth, months 6–12 after birth, and months 13–80 after birth accordingly and regress our measures of adiposity risk at age six on “fake” sunshine hours in the four intervals, conditioning for the same set of control variables as in our baseline specification. We repeat this exercise 300 times. The vertical line indicates our baseline point estimate. For overweight, 13 out of 300 “fake” point estimates are more negative than our baseline estimates; for obesity and severe obesity, the numbers are seven and two.
C. Additional Findings
1. Detailed sunshine intervals
In Figure 3, we distinguish eight (as opposed to four) sunshine intervals: seven three-month intervals ranging from the beginning of pregnancy to the end of the child’s first year and (as in Table 4) one interval from the children’s first birthday until they are 80 months old. The figure once again visually illustrates that increased sunshine hours in the first six months of life (that is, in intervals 0–2 and 3–5) reduce the risk of adiposity at age six. In contrast, increased sunshine hours in months six to 11 after birth or beyond the children’s first year do not have a significant impact on childhood adiposity. In terms of magnitude, 100 additional sunshine hours in months three to six after birth reduce the risk of being overweight, obesity, or severe obesity by 0.19, 0.18, and 0.09 percentage points or 1.3, 3.6, or 9 percent, respectively.
Effects of Sunshine Hours during Pregnancy and First Years of Life on Adiposity Risk at Age Six
Data sources: School entry examinations, Schleswig-Holstein, Lower Saxony. Sunshine hours from 21 weather stations.
Notes: The figures show the effects of 100 additional sunshine hours (in percentage points) in three-month intervals from conception to 11 months after birth, and between 12 and 80 months after birth, on children’s risk of overweight, obesity, and severe obesity. Overweight, obesity, and severe obesity are defined, respectively, as ≥85th, ≥95th, and ≥99th percentile in the gender- and age (in months)-specific BMI distributions. Effects are estimated by a linear probability model and regressions control for, as in Equation 1, district-by-birth year and birth month fixed effects. The horizontal line depicts a zero effect. The vertical lines show 95 percent confidence intervals calculated based on standard errors clustered at the “weather station district” level (districts that are assigned different weather station combinations, 28 clusters).
2. Nonlinearity
So far, we have assumed that sunshine hours in three-month intervals from the start of pregnancy to the child’s first year affect childhood adiposity in a linear way. We investigate the possibility of a non-linear relationship between sunshine hours in the first six months of life and subsequent childhood adiposity in Figure 4, by separating sunshine hours in the first six months of life into ten equally sized groups. In addition to these ten dummy variables, we control for district-by-year of birth and month of birth fixed effects (as in Equation 1) as well as sunshine hours during pregnancy, months 6–11, and months 12–80 after birth (as in Table 4). The figure highlights that the relationship between sunshine hours in infancy and adiposity at age six is indeed roughly linear. For example, moving from the lowest to the highest decile—equivalent to an increase of 800 sunshine hours—reduces the risk of obesity by one percentage point, or 20 percent.
Nonlinear Effects of Sunshine Hours in First Six Months of Life on Adiposity Risk at Age Six
Data sources: School entry examinations, Schleswig-Holstein, Lower Saxony, Brandenburg. Sunshine hours from 21 weather stations.
Notes: The figures plot the effects (in percentage points (pp.)) of sunshine hours in the first six months after birth, categorized in deciles, on children’s overweight, obesity and severe obesity risk. Effects are estimated by a linear probability model and regressions control for, as in Equation 1, district-birth year and birth month fixed effects. Overweight, obesity, and severe obesity are defined, respectively, as ≥85th, ≥95th, and ≥99th percentile in the gender- and age (in months)-specific BMI distributions. Sun deciles are constructed by ranking districts by their sunshine hours in the first six months after birth and dividing them into ten equally sized groups. For example, in the second decile districts receive 491 sunshine hours on average, compared to 777 hours and 1,287 hours in the fifth and tenth decile. The horizontal line depicts a zero effect. The vertical lines show 95 percent confidence intervals, based on standard errors clustered at the “weather station district” level (districts that are assigned different weather station combinations, 28 clusters).
3. BMI, Weight, and Height
Figure 5 shows the impact of sunshine exposure in eight detailed intervals on BMI, which is a combination of a child’s weight and height. Panel A clearly shows that sunshine hours in the first six months of life significantly reduce BMI, but effects beyond six months after birth are close to zero and insignificant. Moreover, sun intensity in the last trimester of pregnancy also significantly lowers overall BMI.
Effects of Sunshine Hours during Pregnancy and First Years of Life on BMI, Height, and Weight at Age Six
Data sources: School entry examinations, Schleswig-Holstein, Lower Saxony, Brandenburg. Sunshine hours from 21 weather stations.
Notes: Panel A shows the effects of 100 additional sunshine hours (in percentage points) in three-month intervals from conception to 11 months after birth and between 12 and 80 months after birth on children’s BMI (Panel A), height (Panel B), and weight (Panel C). Effects are estimated by a linear probability model and regressions control for, as in Equation 1, district-by-year of birth and month of birth fixed effects. The horizontal line depicts a zero effect. The vertical lines show 95 percent confidence intervals, calculated based on standard errors clustered at the “weather station district” level (districts that are assigned different weather station combinations, 28 clusters).
In Panels B and C of the figure, we explore the impact of sunshine hours in pregnancy and infancy separately for height and weight. The graphs show that sunshine hours generally have little impact on children’s height at age six, regardless of the sunshine interval. Even though the effect of increased sunshine hours on height is statistically significant for the interval three to five months after birth, the effect is very small in magnitude: 100 additional sunshine hours in that interval reduce height by 0.8 millimeters, implying an increase in BMI through the height channel of 0.013 percent if evaluated at mean weight and height for boys.
For weight, in contrast, a similar pattern as for our obesity measures emerges, and the negative impact of sunshine hours on weight at age six is concentrated in the first six months of life. The coefficient for the interval three to five months shows that 100 additional sunshine hours reduce weight at age six by 60 grams, implying a reduction in BMI through the weight channel of 0.26 percent if evaluated at mean weight and height for boys—an effect that is 20 times larger in magnitude than that through the height channel. The negative impact of sunshine hours in the first six months of life on average BMI at age six is therefore driven by a reduction in weight and not an increase in height.
4. Effects along the BMI distribution
In a final step, we examine the effects of 100 additional sunshine hours in the first six months after birth along the BMI distribution at age six. To this end, we estimate unconditional quantile regressions based on Equation 1 for each decile as well as the 95th and 99th percentile of the BMI distribution. The results, presented in Figure 6, show that while increased sunshine hours somewhat reduce BMI at all deciles, the effect becomes increasingly more negative at higher percentiles of the BMI distribution beyond the 80th percentile. This finding is important, as higher BMI within the normal BMI range has no negative effects on health or medical costs; only children with an extreme BMI show worse health and generate high costs for the health system. The finding further suggests that increased sunshine exposure in the first six months of life prevents extreme cases of dysfunctional adipose tissue development and extreme forms of insufficient metabolism. More generally, the finding that the effects of increased sunshine hours are concentrated at the right tail of the BMI distribution mirrors those for bone density, where additional vitamin D primarily benefits individuals with very low bone stability (Dawson-Hughes et al. 1997).
Effects of Sunshine Hours during the First Six Months of Life along the BMI Distribution
Data sources: School entry examinations, Schleswig-Holstein, Lower Saxony, Brandenburg. Sunshine hours from 21 weather stations.
Notes: The figure plots the effects of 100 additional sunshine hours on BMI in the first six months after birth along the BMI distribution. Each square shows the effect at a certain decile. Additionally, effects at the 95th and 99th percentile are shown. The vertical lines show 95% confidence intervals, based on standard errors clustered at the “weather station district” level (districts that are assigned different weather station combinations, 28 clusters).
5. Heterogeneous responses by subgroups
Do the effects of sunshine hours in infancy on childhood obesity vary across children? We first distinguish between minority and majority children.10 Minority children benefit more from additional sunshine exposure in infancy than majority children, for example, because mothers of minority children are less likely to take vitamin D supplements (for example, Moffat et al. 2015) or because of their darker skin color, which generally makes them more prone to vitamin D deficiency (for example, Martin, Gowda, and Renzaho 2016; Hintzpeter et al. 2008). Minority children are also more likely to be overweight or obese than majority children (Table 2). The results in Online Appendix Table B2 tentatively suggest that sunshine hours in the first six months after birth reduce the risk of adiposity at age six more for minority than majority children. However, the difference in effects is statistically significant only for the outcome of overweight.
In a second step, we distinguish between who did and did not participate in all nine medical postnatal checkups. The first three of these medical checkups are conducted directly and in the days after birth. After that, checkups take place three, six, 12, 24, 48, and 60 months after birth, conducted by pediatricians in hospitals or at their surgeries. Even though checkups are compulsory, take-up is not complete, and incomplete take-up is correlated with social disadvantage (Kamtsiuris et al. 2007). There is little evidence the effects of sunshine hours in infancy vary between children with complete and incomplete take-up (Online Appendix Table B2).
VI. Discussion and Conclusion
Based on administrative data from school entry examinations that cover all children in three large geographical areas in Germany and span 16 birth cohorts, combined with information on monthly sunshine hours obtained from 20 weather stations, our analysis shows that increased sun intensity in the first six months of life causally reduces the risk of overweight and obesity at age six, likely through higher vitamin D intake.
In terms of magnitude, 100 additional hours of sunshine in this time period, which is equivalent to the increased sun intensity from a two-and-a-half-week-long holiday in winter to a destination where sunshine is similar to that in summer, reduce the risk of being overweight at age six by 1.1 percent, the risk of obesity by 2.6 percent, and the risk of severe obesity by 6 percent. Sun intensity beyond the first six months of life, in contrast, has no discernible impact on subsequent adiposity. Sun intensity during the last trimester of pregnancy also reduces the risk of adiposity at age six, although this effect tends to be smaller in magnitude than during the first six months of life and is statistically significant in only some specifications.
We offer two main explanations for this pattern. First, the first six months of life (and possibly the third trimester of pregnancy) are a sensitive period for later obesity, as this is the period when infants rapidly gain weight and adipose tissue develops. Second, infants’ vitamin D levels are particularly sensitive to sunshine in the first six months of life, when lactation is highest.
Since we cannot observe the child’s and mother’s direct exposure to sunlight, our estimates should be interpreted as intention-to-treat estimates and should be considered as lower bounds for the impact of direct sunshine exposure on subsequent adiposity. Assuming that mothers of infants spend a third of their time in the sun during sunny days (which is likely to be an overestimate), our estimates imply that only 33 hours of additional direct sunshine exposure in the first six months of life—or 11 minutes per day—are required to induce a 1.1 percent reduction in the risk of overweight and a 6 percent reduction in the risk of severe obesity at age six. Medical research suggests that spending three to eight minutes in the sun in Boston, MA, from April to October at 12 p.m. EST synthesizes 400 International Units (IU) of vitamin D (Terushkin et al. 2010), which is equivalent to the recommended daily supplement for breastfeeding mothers.11 As the parts of Germany included in our study are located at a latitude more north than Boston, and mothers unlikely receive additional sun mostly at 12 p.m. when vitamin D generation is highest, our assumed 11 minutes additional sun per day will lead to roughly an additional 400 IU of vitamin D. Therefore, our estimates may also give a hint of the effects of the daily intake of recommended vitamin D supplements.
Overall, a relatively small increase in direct sun exposure during the lactation period, or, alternatively, only the recommended amounts of vitamin D supplements for breastfeeding mothers, can reduce subsequent adiposity of their children. This finding is important for public health resource allocation. Since take-up of vitamin D supplements after birth is low, our results give further reasons, besides other proven effects of vitamin D, why governments should invest in campaigns to increase utilization of vitamin D supplements.
Footnotes
↵1. According to the meta-study by Pereira-Santos et al. (2015), obese and overweight individuals show a 35 percent and 24 percent higher risk of vitamin D deficiency than average weight individuals.
↵2. For adults, studies are likewise scarce and inconclusive. While two double-anonymous randomized control studies find that vitamin D supplementation enhanced weight loss (Mason et al. 2014) or lowered body fat mass (Salehpour, Hosseinpanah, and Shidfar 2012) for specific subgroups, Soares, Ping-Delfos, and Ghanbari (2011), in a summary of the literature, conclude that “current evidence from RCTs did not consistently support the contention that calcium and vitamin D accelerated weight or fat loss in obesity.”
↵3. We drop children with a height below 80 cm and above 140 cm and with a weight above 60 kg and below 10 kg.
↵4. The BMI cutoff points for overweight at age 5.5, 6, and 6.5 are 17.45, 17.55, and 17.71 for boys and 17.20, 17.34, and 17.53 for girls. The respective BMI cutoffs for obesity are 19.47, 19.78, and 20.23 for boys, and 19.34, 19.65, and 20.08 for girls. Cole et al. (2000) define no cutoff for severe obesity.
↵5. Minority children are children whose parents migrated to Germany (for data from Weser-Ems) or children who do not have German citizenship (for data from Schleswig-Holstein). Information about migration background is not available in the state of Brandenburg.
↵6. Mothers with higher education have either obtained a university degree (Brandenburg and Weser-Ems) or a school-leaving qualification that allows them to enter university (Schleswig-Holstein).
↵7. Information on unemployment rate on district level is available since 1998. Information on GDP on district level is available since 2000.
↵8. The standard errors hardly change if we apply Conley standard errors (Conley 1999) to account for spatially autocorrelated error structures.
↵9. The overall rates for overweight and obesity following the international cutoff by Cole et al. (2000) are 13.28 and 3.6.
↵10. Information on minority status is available in only two federal states, Lower Saxony and Schleswig-Holstein, leading to a substantial reduction in sample size. See footnote 6 for a definition of minority status.
↵11. Terushkin et al. (2010) also present evidence that because of the well-known detrimental side effects of ultraviolet irradiation and the high effectivity of oral Vitamin D supplements, oral supplementation remains the safest way for increasing vitamin D levels. This result is supported by Wicherts et al. (2011) who, based on a randomized clinical trial, conclude that vitamin D supplementation is more effective than advised sunlight exposure in reducing vitamin D deficiency.
- Received September 2020.
- Accepted March 2022.
This open access article is distributed under the terms of the CC-BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at: https://jhr.uwpress.org.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.