Double Trouble: The Burden of Child-rearing and Working on Maternal Mortality

In times of demographic change and aging populations, policy-makers in many countries are concerned about the development of the work force. In order to buffer the systematic aging and potential shrinking of the work force, policies that encourage an increase in female labor supply are often on the agenda. At the same time, family policies aiming at an increase in fertility rates are discussed (Jaumotte 2004). However, for many mothers (and arguably fathers), simultaneously raising children and pursuing a career is challenging because both pursuits are time-consuming. Moreover, balancing family life and careers may affect parental health through stressors occurring from the double burden of working and caring for children. We investigate this (potential) double burden and its effect on maternal health later in life. When designing public policies aimed at increased fertility rates and/or increased female labor supply, any negative health consequences of family and work have to be taken into account. In the United States and many other developed countries, work-family conflict and its potential negative effects are a public concern (see, e.g., Williams and Boushey 2010).

Our analysis is based on administrative birth and death registries from Sweden. To study the effect of childbearing and labor force activity on maternal health, an ideal set-up would provide exogenous variation in both labor market activity and childbearing. In our study, we use twin births as an unplanned fertility shock. Initially, we compare health outcomes of mothers who had twins at first birth (hereafter referred to as twin mothers) with those of mothers of singletons at first birth (nontwin mothers). The Swedish data allow us to follow mothers over a long period, and to our knowledge, we are the first to study the effects of twins on mothers’ old-age mortality. We find that twin mothers have a 3.8 percentage point (13%) higher all-cause mortality risk compared with mothers of singletons. This finding provides some initial evidence on the relationship among family, work, and health. However, although twinning is an exogenous shock,¹ it can affect old-age health via multiple pathways. The twin mothers in our data have, on average, 0.68 more children five years after the first birth, but child spacing also changes: the spacing between the first and the last child is 3.8 years for twin mothers and 7.4 years for nontwin mothers. Twinning may also affect health risks during pregnancy and delivery as well as marital stability. Moreover, if women adjust their labor market choices depending on childbearing (Lundborg et al. 2017), this might reduce any negative health impacts from the fertility shock due to twinning. Therefore, we examine the reduced-form effect and do not use twin births as an instrumental variable for the number of children.

To understand potential mechanisms and to study a double-burden effect of family and work, we pursue two strategies.² First, we stratify the sample along educational attainment and pension income, both of which are strongly related to labor force attachment. Level of education is an ex ante predictor of higher labor market activity, and pension income is a proxy for ex post realized labor market activity. Thus, if the double burden of working and caring for children is important, we expect to see larger health effects of a fertility shock (twin births) for women with high education and/or high pension income because these women more often combine working and child-rearing.³

The second strategy to understand the mechanism behind the increased mortality of twin mothers is to analyze specific stress-related diseases. The underlying hypothesis is that the double burden of working and caring for children at the same time creates stress, which may particularly affect stress-related mortality. Here, one challenge is the measurement of lifetime stress. We analyze two specific groups of medical diagnoses related to stress during life: cardiovascular diseases (heart attacks and strokes) and smoking-related diseases (lung cancer and chronic obstructive pulmonary disease (COPD) (Brotman et al. 2007; Kouvonen et al. 2005). The literature shows that strong predictors of cardiovascular diseases are known to be elevated by chronic stress, which results from caregiving, for example.⁴ Stress from work-family conflicts is also related to smoking behavior (Hurtado et al. 2016; Nelson et al. 2012), which in turn is strongly correlated with lung cancer and COPD. If these stress-related causes of death are affected, we take this as evidence of increased stress due to the double burden.

We acknowledge that level of education may interact with child-rearing in ways other than through the stress caused by the double burden of family and work. For instance, highly educated mothers receive higher incomes, which in turn may allow them to better cope with the increased stress. Higher education is also associated with jobs that offer more flexible working hours and more control over work content, which may help to alleviate conflicts between children and jobs. In this study, we do not attempt to disentangle these different mechanisms behind the relationships among level of education, labor supply, child-rearing, and maternal health. Unfortunately, we lack data on important factors, such as work content and control over work time, because we do not know which jobs mothers in our sample held in the past. However, most of these alternative mechanisms—such as income and work conditions—suggest that highly educated mothers are better able to compensate for the twin burden. Nevertheless, we see that highly educated mothers are more affected by twinning, suggesting that the double burden of family and work outweighs these other mechanisms; we take this as evidence of a double-burden effect of family and work. In robustness analyses, we also use additional data to examine some of these alternative mechanisms.⁵

Given that we compare twin and nontwin mothers, all women in our sample have at least one child. Thus, we are able to study the effect of fertility only at the intensive margin—that is, comparing having a singleton versus twins at first birth. A comparison at the extensive margin (i.e., having any children versus none) would be at least as interesting as the intensive margin comparison: by definition, women without children do not experience the double burden of children and work. However, it is hard to find exogenous variation on the extensive margin effect. In fact, in a lone exception, Lundborg et al. (2017) used the success of in vitro fertilization (IVF) treatments to obtain exogenous variation at the extensive margin. Interestingly, they found that the effects of fertility on earnings are larger at the extensive margin than at the intensive margin. Given this evidence, it seems reasonable to extrapolate that there also are important double-burden effects at the extensive margin.

Our study is linked to the literature on the interactions among fertility, working life, and maternal health. Kahn et al. (1964) were among the first to note that a conflict between work and nonwork roles are a major source of stress. A large literature has researched the work-family conflict empirically. However, many early studies were based on small or highly selective samples (Greenhaus and Beutell 1985). Studies based on large, representative, or administrative data include Weatherall et al. (1994), Martikainen (1995), and Waldron et al. (1998). More specifically, related to our study, Nelson et al. (2012) found a strong relationship between work-family conflicts and smoking behavior in a sample of English long-term care workers. Similarly, Väänänen et al. (2005) documented a negative association between conflicts due to paid and domestic work and health among men and women. Using retrospective data from the Health and Retirement Study, Sabbath et al. (2015) categorized mothers along their past marriage, fertility, and working histories, concluding that working single mothers experience the highest mortality rates in old age. Van Hedel et al. (2016) and Berkman et al. (2015) found that single working motherhood is associated with higher likelihood of stress-related heart diseases. Although suggestive of a double-burden effect, these studies and other related studies were not able to address self-selection of women into specific work-family profiles based on their health and other unobserved related factors. By using twins at first birth as a fertility shock, we try to overcome this limitation.

A growing literature has examined the relationship between family life and female careers. Several studies found that childbearing has negative effects on female labor force participation, earnings, and wages (Angrist and Evans 1998; Lundborg et al. 2017) and that male and female earnings diverge after the birth of the first child (Angelov et al. 2016; Bertrand et al. 2010; Kleven et al. 2019a, b). Other studies examined how public policies, such as parental leave policies, affect labor force participation and childbearing (Del Boca 2002; Lalive and Zweimüller 2009; Schönberg and Ludsteck 2014) or sickness absence from work (Guertzgen and Hank 2018). We study a related aspect of family life and female careers: the double burden of simultaneously working and caring for children.

Two closely related studies are Cáceres-Delpiano and Simonsen (2012) and Kruk and Reinhold (2014). Using multiple births as an instrumental variable, Cáceres-Delpiano and Simonsen (2012) found that a higher number of children implies worse health for U.S. mothers aged 20–45. Based on data from the Survey of Health, Ageing and Retirement in Europe, Kruk and Reinhold (2014) showed that an increase in the number of children has a negative impact on mental health of older women but no effect on older men. Kruk and Reinhold used twin births and sibling sex composition as instruments for the number of children. We add to this literature by focusing on heterogeneity in the long-term effects of fertility on health. We are the first to study effects of fertility on (cause-specific) old-age mortality. In addition, unlike Cáceres-Delpiano and Simonsen (2012) and Kruk and Reinhold (2014), we rely on administrative rather than survey and census data.

We use the Swedish multigeneration register, which links all individuals to their biological mother (and father), even if they do not live in the same household or have died. The register contains parental information for persons born in 1932 or later, including information on year and month of birth (for further details about the register, see Ekbom 2011). Twins are identified as being born to the same mother in the same year and month as another sibling.

From the registry, we identify 404,286 mothers aged 55–65, alive, and resident in Sweden in 1990. Of those mothers, 2,684 (0.66%) had twins at first birth.⁶ We follow the mothers for 20 years, from 1991 to 2010. From the National Causes of Death Register, we know whether they died and, if so, the cause of death. We focus on two groups of diseases that may be related to stress during life: cardiovascular diseases and smoking-related diseases. For the former, we study heart attacks and strokes; for the latter, lung cancer and COPD. Here, we follow the strategy by Evans and Moore (2012) to classify the diagnoses into specific disease categories.⁷

Among all mothers born between 1925 and 1935, 8.6% were not observed in 1990 because they died (75%) or moved abroad (25%) before 1990. In Table A1 of the online appendix, we investigate whether twin and nontwin mothers differ in the probability to be included in our study sample (column 1) or in the probability of dying before 1990 (column 2); we find no significant differences. Thus, although the sample as a whole may suffer from survival/migration bias, our results are unlikely to be biased because twin and nontwin mothers are affected symmetrically.

We have a set of socioeconomic characteristics from population registers. Table 1 describes our variables for the full sample and stratified by mothers’ educational attainment. Education is defined in three categories: primary, secondary, and tertiary schooling. Primary schooling includes elementary schooling (Folkskola), which comprised seven years for most of the cohorts considered here, and junior secondary schooling (Realskola). The designation of secondary schooling means that mothers had some education beyond the primary level but no tertiary education. This level also includes vocational training at the upper secondary level. Tertiary schooling indicates that mothers experienced some higher education: that is, the attainment of some college or of a university education or a PhD.

On average, the mothers were 60 years old in 1990, had their first child at age 24.5, and have 2.4 children. The majority of the mothers completed primary education (59%), about 30% completed secondary education, and roughly 11% earned a tertiary degree. The age at first birth is, on average, three years higher in the highly educated group compared with the lower-educated group, and the average number of children is about the same.

Overall, about 66% of the women aged 55–65 are still active on the labor market (positive labor income) in 1990. However, the fraction of working women varies considerably by education. Whereas 89% of the women with a tertiary degree receive labor income, only 57% of mothers with primary schooling receive income from work at those ages. About 29% of the women in our sample died between 1991 and 2010, with large variation by education. About one-third of the lower-educated mothers died in the 20-year time window we consider, compared with only 26% (19%) among mothers with medium (high) levels of education. We see similar patterns for the different causes of death.

From the population registers, we also obtain information on pension income at age 72. Pension income is the best available proxy for lifetime labor force participation in our data set.^8,9 Pension income at age 72 follows the expected pattern; the mean pension income of mothers with tertiary schooling is more than 125,000 SEK and approximately 85% higher than the pension income of mothers with primary schooling or less.¹⁰

To investigate differences between twin and nontwin mothers, we specify the following linear regression model:where y_i is the outcome variable, and is an indicator equal to 1 if mother i gave birth to twins at first birth. Depending on the specification, the outcome variables are indicators equal to 1 if the mother died from any cause between 1991 and 2010, from a heart attack or stroke, or from lung cancer or COPD between 1991 and 2010. The control variables, x_i, include dummy variables for seven education levels, dummy variables for mothers’ birth cohorts, and a quadratic polynomial in age at first birth.^11,12

One worry when comparing twin mothers with nontwin mothers is that twinning may not be entirely random. For instance, Bhalotra and Clarke (2016) documented that twin mothers are positively selected with respect to health and health behaviors. However, Farbmacher et al. (2018) showed that this selection issue is primarily attributable to dizygotic (fraternal) twins. Dizygotic twins become more likely with increasing age of the mother (Fauser et al. 2005; Reddy et al. 2005) and the use of IVF treatments (Thurin et al. 2004), creating a correlation between twin births and maternal health. Monozygotic twins, on the other hand, are considered to be truly random (MacGillivray et al. 1988; Tong and Short 1998).

We have several strategies for addressing these issues. First, as mentioned earlier, we control for age at first birth in all our analyses. Second, to investigate a possible selection bias stemming from dizygotic twins, we compare estimates between mothers of all twins and only same-sex twins (while also controlling for age at first birth). Because monozygotic twins necessarily have the same sex, their share must be higher among same-sex twins (Black et al. 2007; Figlio et al. 2014). Third, IVF is less of a concern in our data given that more than 99% of the mothers gave birth to their first child between 1940 and 1970, when IVF treatment was not yet available. This is illustrated by Fig. 1, which shows the twin rates in Sweden across the first child’s year of birth. The share of twins remained fairly constant between 1940 and 1980 but increased strongly thereafter. Although the steady but mild rise in the twin rate after 1980 could be attributed to delayed childbearing, the steep increase in the twin rate since 1990 mainly follows the availability of IVF. The vertical lines in the graph indicate the cohorts included in our analysis by birth year of the first child(ren).

Twinning affects the number of children the mother has over a specific period and after completed fertility. However, whereas many previous studies used the birth of twins as an instrumental variable for fertility, we study the so-called reduced-form effects of twinning. That is, we estimate differences between twin and nontwin mothers, thereby capturing all mechanisms through which twinning affects mortality. Along with the main mechanism—the increase in the number of children—other effects may exist. First, twin pregnancies and delivery, on average, pose a greater health risk to the mother than singleton births, which may translate into higher old-age mortality rates (Buhling et al. 2003; Rauh-Hain et al. 2009). Second, because twins are extremely closely spaced, having twins may influence the spacing of further children, which in turn might have a direct effect on mother’s health. Third, twinning may affect the probability of divorce, which may act as an additional channel of stress. Fourth, mothers may adjust their labor market choices if they have twins (Lundborg et al. 2017), perhaps by decreased labor force participation, which in turn may have health effects. We therefore focus on the overall effect of twinning and interpret any effects as a general effect of child-rearing. As a background, we also show how twinning affects completed fertility and birth spacing.

To analyze the double-burden effect, we study the effect of twinning for subsamples defined by level of education and pension income. Education is an important predictor of labor force participation, working hours, and earnings because highly educated mothers are more prone to pursue careers. Therefore, we expect that given the same unplanned fertility shock, highly educated mothers are more likely to experience a double burden of working life and child-rearing. Thus, a finding of larger health effects of twinning for highly educated mothers would be evidence of a double-burden effect. Note that we can use education for stratifying the analysis because most of the mothers in our sample completed their education before giving birth to their first child; the education level is less influenced by fertility than is labor force participation itself.¹³

We use twinning as a fertility shock and assume that highly educated mothers experience more stress because they are more likely to work. However, level of education is also correlated with other factors that may influence the effect of twinning on maternal health. First, the Grossman (1972) model predicts that highly educated individuals are better at using medical care goods and services and are therefore more able to mitigate negative health effects. Second, highly educated mothers receive higher incomes, which may allow them to better cope with the increased stress due to twinning. Third, higher education is also associated with jobs that offer more flexible working hours and more control over the work content, which may also help alleviate conflicts between child-rearing and work. Fourth, because of educational homogamy, fathers’ involvement may differ between higher- and lower-educated mothers. Fifth, because higher education induces more specialized jobs, highly educated women might move farther from their families and have different networks. However, most of these alternative mechanisms indicate that highly educated mothers should be less affected by twinning. In other words, a finding that highly educated mothers are more affected by twinning would suggest that the double burden of family and work outweighs these other mechanisms. Thus, even if we are not able to completely disentangle the different mechanisms behind the relationships among level of education, labor supply, child-rearing, and maternal health, we can provide some evidence on the double burden of family and work.

Although higher education holds the ex ante potential only for greater labor force attachment, pension income is a proxy for lifetime income and is thus an ex post realization of the former. Pension income in Sweden is independent of the partners’ income. Our hypothesis is that mothers with a higher pension income are more likely to experience a double-burden effect. We therefore stratify the sample at the median pension income, with an income greater than the median implying more labor market attachment. Pension income is a direct result of labor market participation, which could be reduced because of childbirth. Thus, pension income is an endogenous outcome. However, pension income is also the best available proxy for lifetime labor force participation in our data set, and we believe that splitting the data set based on it reveals valuable insights. One caveat of the split by pension income is that a specific level of pension income can result from working many hours at a medium or low wage as well as working few hours at a high wage. A high number of hours worked is usually what creates the conflict with child-rearing and thus the double burden. However, if women with low and high pension income do not differ in the number of hours worked but only in their average income, we should see no differences in the effect of having twins by pension income. Additionally, twin births could have a direct effect on survival and retirement until age 72, which may create selective sorting. Table A1 in the online appendix (columns 4–5) shows no differences between twin and nontwin mothers with respect to whether the pension income at age 72 is missing or the amount of pension income at age 72.

Our results are derived for the cohorts of mothers born between 1925 and 1935 who had their first children between 1940 and 1970. To get an idea how these cohorts differ from younger cohorts in terms of mother’s age at first birth, the number of children, and birth spacing, we plot the main differences by mothers’ birth cohorts in the online appendix (Figs. A1–A4). For the cohorts born between 1925 and 1935, age at first birth declined from around 25 years to about 24 years, remaining there until around age 24 for the birth cohort until the late 1940s and increasing continuously thereafter (Fig. A1, online appendix). The number of children per mother fluctuated between just more than 2.4 and 2.2 children per mother for all cohorts, with the highest average number of children among women born in the early 1930s and the lowest for women born around 1945 (Fig. A2, online appendix). Birth spacing between the first and the second as well as the first and the last child declined continuously for mothers until the cohorts born around 1940. It increased slightly for mothers born until 1950 but continued to decrease thereafter (Figs. A3 and A4, online appendix).

In summary, we find that having twins increases all-cause mortality significantly for women older than age 55. Twin mothers also have a higher probability of dying from lung cancer/COPD and heart attacks/strokes. The effects of twins on maternal health are the strongest for high-educated mothers and mothers with above-median pension income.

Our results fit into a recent line of epidemiological and sociological literature on the adverse effects of the work-family strain, which offers three general theoretical considerations. First, there are selection effects: women who are employed, married, and have children are healthier than their childless, unmarried, and unemployed counterparts. Second, according to the role accumulation theory, combining family and work is beneficial for women’s health. Third, multiple role theory states that combining work and family roles leads to stress, with negative consequences for health (Kahn et al. 1964).

In our study, we use twins at first birth as a shock to fertility, which allows us to partly overcome the selection problems arising in this literature. We find that all-cause mortality and mortality from lung cancer and heart diseases are significantly elevated among mothers who give birth to twins. The higher probability of death due to lung cancer and heart diseases indicates that at least part of the effect is stress-related, given that the medical literature strongly associates these causes of death with stress from work-family conflicts and caregiving. That is, the additional burden on women resulting from having two versus one child at their first birth takes its toll on their health later in life.

We also find strong mortality effects among mothers with higher education and above-median pension income. We argue that this can be taken as evidence of a double-burden effect. Highly educated mothers have a higher likelihood of pursuing a career despite having children, and higher pension income indicates that women worked more. In other words, higher education is an ex ante predictor of higher labor market attachment, and pension income is a proxy for ex post realized labor market activity. Thus, higher all-cause and stress-related mortality rates among mothers with higher education and above-median pension income combine to point in the direction of a double burden from child-rearing and working that negatively affects maternal health in old age.

The specific mechanisms behind these findings deserve further research, especially because women of younger generations are increasingly more likely to stay attached to the labor force while raising children (Goldin and Mitchell 2017). Fathers’ roles in supporting their families, both financially and by taking a more active role in raising children, change as well. Changes in family planning have to be taken into account when it comes to the external validity of our results for younger generations. We shed some light on this issue by depicting the averages of age at first birth, number of children, and birth spacing for younger and older generations in Sweden. In any case, more research is necessary to find policies that can protect mothers from the double burden of family and work. The recent study by Avendano et al. (2015), for example, showed that more generous parental-leave policies reduce maternal depression in old age. This is particularly relevant in times of demographic change, when many countries aim to increase both female labor supply and birth rates.