The Inverse Association Between Number of Siblings and Divorce: New Evidence From China and Europe

Abstract

In the U.S., evidence has emerged suggesting that divorce is lower among those with many versus few siblings, a pattern that may indicate that children develop important social skills via their childhood interactions with siblings. However, this pattern has yet to be tested in other countries with varying fertility and divorce rates. We extend the empirical basis of the sibship size/divorce literature by exploring the association in China and Europe, each with unique demographic characteristics. Each additional sibling is associated with an 11 percent decline in the probability of divorce in China and a two percent decline in Europe, net a wide range of covariates. We also test whether these patterns vary across cohorts and alternative coding schemes. The results have implications for our understanding of how growing up with siblings influences later life outcomes and the contextual features that form that relationship.

Keywords

quantitative divorce/separation family demography fertility global/international

The notion that early childhood conditions shape long-term outcomes is widely accepted in the social sciences (Guo, 1998; Heckman & Masterov, 2007), a pattern Alexander et al. (2014) refer to as “the long shadow” of childhood. Because fertility is declining in most countries, one aspect of the childhood environment that is receiving increasing attention is sibship size (Blake et al., 1991; Bobbitt-Zeher & Downey 2013; Downey & Condron 2004; Merry et al., 2020; Polit & Falbo, 1987). Ever since Simmel (1902) articulated the many differences between a dyad and a triad, demographers have considered how small increases in the number of children can result in significant changes in family dynamics.

A common concern is that growing up without any siblings is problematic for the development of social relationships. In the general public, only children have been described as self-centered, selfish, and lacking in social skills (Blake, 1981). While this position is disputed by some research (Polit and Falbo 1987), there is growing evidence that exposure to siblings during childhood is associated with a lower probability of divorce in adulthood (Bobbitt-Zeher et al., 2014; Merry et al., 2020). One explanation for this pattern is that siblings provide interaction “practice partners” during childhood, honing social skills that facilitate the development and maintenance of long-term relationships. If lack of exposure to siblings operates in this way, it is concerning given declining fertility in most countries and the increasing number of children who will grow up with few or no siblings. It raises the question: Is fertility decline producing a more divorce prone populace?

While it seems plausible that exposure to siblings may promote the skills that facilitate long-term relationships, if this is accurate, we should observe the same inverse relationship between number of siblings and divorce in contexts other than the U.S. We expand the empirical basis of this literature by testing the relationship between number of siblings and divorce in two different areas of the world: China and Europe. China represents a context where divorce is rare but increasing and fertility was long restricted by the One Child Policy. Europe is a mostly modern context with heterogeneous fertility and divorce rates. The unique characteristics of each make them attractive candidates for testing the generalizability of the sibship size/divorce relationship.

Number of Siblings During Childhood and Divorce in Adulthood

Two studies have reported that the number of siblings one is raised with during childhood is inversely related to the likelihood of divorce, years later, as an adult (Bobbitt-Zeher et al., 2014; Merry et al., 2020). This research was motivated by the confluence of two patterns. On one hand, number of siblings have been consistently associated with poorer educational outcomes (Blake, 1981, 1989; Downey, 1995, 2001), suggesting advantages for those growing up with few brothers or sisters. On the other hand, children lacking any siblings are rated by teachers as exhibiting the poorest social/behavioral skills (Downey & Condron, 2004), suggesting potential benefits to sibling interactions at home. Combined, the literature suggested that siblings were associated with reduced school performance but increased social skills. Scholarship on the relationship between number of siblings during childhood and the likelihood of divorce as an adult prompted a discussion between these two previously disparate literatures—one on sibship size and one on divorce.

The results revealed an inverse association. Bobbitt-Zeher et al. (2014) analyzed more than 57,000 adults from the General Social Surveys, spanning 1972 to 2012. They found that each additional sibling was associated with a three percent decline in the likelihood of divorce among respondents that had ever been married. The researchers were sensitive to the possibility of confounds and so included a wide range of control variables in their model: years of education, mother’s education, lived with both biological parents at age 16, parents divorced, family structure, race, gender, age, age at marriage, number of children, cohabitation before marriage, income, homeowner, region, rural/suburban/urban, gender role attitudes, religious affiliation, and attendance. The inclusion of the covariates, especially gender role attitudes, religious affiliation, and religious attendance, reduced the magnitude of the sibship size effect by 74%, but the sibling coefficient remained statistically significant (Bobbitt-Zeher et al., 2014; Table 1). The authors concluded that finding an inverse association between number of siblings and divorce was important because it was “one of the few pieces of empirical evidence that siblings may provide value for important later life outcomes” (Bobbitt-Zeher et al., 2014, p. 16).

Table 1.

China General Social Survey (CGSS) Descriptives.^a

Variables	N	Mean	Std. Dev.	Min	Max	Description
Dependent variables
Currently divorced	11,634	0.03	0.16	0	1
Independent variables
Siblings (linear)	11,600	3.11	1.93	0	10	Coded with highest # at 10+
Any siblings	11,600	0.93	0.26	0	1
Siblings (category)
0 Siblings	11,600	0.07	0.26	0	1
1 Siblings	11,600	0.14	0.35	0	1
2 Siblings	11,600	0.20	0.40	0	1
3 Siblings	11,600	0.19	0.39	0	1
4 Siblings	11,600	0.16	0.37	0	1
5+ Siblings	11,600	0.23	0.42	0	1
Control variables
Male	11,634	0.46	0.50	0	1
Birth cohort
Before 1949	11,634	0.19	0.40	0	1
1950–1959	11,634	0.22	0.42	0	1
1960–1969	11,634	0.26	0.44	0	1
1970–1979	11,634	0.22	0.41	0	1
1980–1989	11,634	0.09	0.29	0	1
1990–1999	11,634	0.01	0.09	0	1
Age	11,634	48.13	13.89	18	98
First marriage age	11,374	23.81	4.06	11	75
Current urban hukou	11,634	0.53	0.50	0	1
Urban hukou at 14	11,620	0.34	0.47	0	1
Educational attainment
Primary education and below	11,632	0.35	0.48	0	1
Secondary education	11,632	0.53	0.50	0	1
Tertiary education	11,632	0.12	0.33	0	1
Mother secondary education and above	10,875	0.13	0.34	0	1	Ref: Primary education and below
Annual household income rank	10,910	5.64	2.86	1	10
Any kids	11,634	0.96	0.20	0	1
Religious belief	11,634	0.10	0.30	0	1	Ref: No religious belief
Regions						Ref: East region
Eastern	11,634	0.39	0.49	0	1
Central	11,634	0.23	0.42	0	1
Western	11,634	0.26	0.44	0	1
Northeastern	11,634	0.12	0.32	0	1
Survey year
2006	11,634	0.24	0.43	0	1
2008	11,634	0.45	0.50	0	1
2017	11,634	0.31	0.46	0	1

^aSamples included are those ever-married respondents in the CGSS datasets.

The possibility that more siblings lower the probability of divorce was then bolstered by a second study. Merry et al. (2020) found an even stronger inverse association in the Longitudinal Study of Adolescent Health (Add Health)—this time each additional sibling was associated with a 10 percent decline in the probability of divorce. The researchers analyzed 14,800 respondents who were in 7^th to 12^th grade in 1994–95 and were followed until 2008–2009, when they were between 24 and 34 years old. This is a younger sample than the GSS and so it is notable that the point estimate is larger (10% decline in the probability of divorce for each additional sibling vs. 3% in the earlier study). Again, the researchers included a wide range of covariates in their models predicting divorce: gender, age, race, age married, any children, education level, mother’s education, lived with both biological parents as a child, homeowner, household income, and religiosity. Unlike the previous study, the difference between the sibship size coefficient in models with statistical controls versus those without was negligible (Merry et al. 2020, Table 1, Sibs linear).

Why Are Siblings Associated with Less Divorce?

There are several possible explanations for the inverse association observed between number of siblings and likelihood of divorce. First, the association may be spurious, the result of differences between the kinds of families that have many versus few children that go unobserved in most models. For example, although scholars have statistically controlled for a wide range of potential confounds, there may exist important additional factors that are correlated both with having many siblings and staying married. As one example, children in large families are often raised with more traditional values (Steelman et al., 2002), which could then reduce the likelihood of divorce in adulthood. Consistent with this view, Vogl and Freese (2020) found that individuals with more siblings reported more conservative values with respect to abortion and same-sex marriage. And while indicators of religiosity and gender role attitudes capture part of this traditionalism, they may not completely account for it. Attempting to identify causal relationships with observational data of sibship size has been a consistent challenge for family scholars. Instrumental variable approaches can provide more confident estimates (Angrist et al., 2010; Blaabæk et al., 2020; Conley & Glauber, 2006), but the data requirements for identifying useful instrumental variables make it challenging for studying divorce.¹

A second set of explanations posits that the association is causal—that there is something about growing up with additional siblings that lowers the probability of divorce in adulthood. The idea here is that growing up with siblings promotes the development of meaningful social skills and that, later in adulthood, these skills facilitate the likelihood of staying married. This notion is plausible, given the robust literature describing the importance of sibling relationships during childhood, their consequences for socioemotional development, and their long-term impacts (Deater-Deckard et al., 2002; Dunn & Munn 1985; Jenkins & Dunn, 2009; Polit & Falbo, 1987). Sibling relationships are often intimate and filled with both positive and negative emotions (Kramer,2010). Previous scholars have described siblings as “practice partners” for interaction (Merry et al., 2020), allowing children to take the role of the other and develop conflict resolution skills. Brody (2004, p. 124) explains that siblings “can provide a unique opportunity for children to develop the ability to understand other people’s emotions and viewpoints, to learn to manage anger and resolve conflict, and to provide nurturance themselves.” Along these same lines, Kramer (2010) emphasizes that sibling interactions promote the ability to manage and regulate one’s emotions during frustrating situations, a skill with value for building relationships outside of the nuclear family. Drawing the link to divorce, Merry et al. (2020) point out that these kinds of relationship dynamics experienced during childhood with siblings overlap with what individuals experience later in life when married. The causal mechanism linking siblings to divorce, therefore, is that relationship skills learned in early childhood (via sibling interactions) become tools for the maintenance of long-term relationships (e.g., marriage) in adulthood.²

An additional causal explanation emphasizes how siblings matter during adulthood rather than childhood. The idea here is that additional siblings reduce the likelihood of divorce because of the network ties an individual and their spouse develop with siblings (and their siblings’ spouses) during adulthood. Merry et al. (2020) write:

Imagine a married couple where both individuals come from large families with siblings who are an active part of the couple’s social network. Under those circumstances, relationships with siblings-in-law may become a meaningful deterrent to divorce because those sibling relationships are ‘part of the package’ of marriage. Ending a marriage is more challenging when significant other relationships (in this case, sibling ones) are also at stake. In contrast, a marital relationship that has little to do with extended family relationships can more readily dissolve.

The two causal explanations are not mutually exclusive, but they emphasize different life stages: social skill building during childhood versus network ties during adulthood. Neither has received sufficient empirical support but the datasets employed by scholars have had only weak indicators of interpersonal skills or network ties. For example, Merry et al. (2020) gauged social skills in the Add Health data with respondents’ ratings of their “level of satisfaction with the way we handle problems and disagreements” with their partner. They found a positive relationship between siblings and this indicator of social skills, but it failed to mediate the sibship size/divorce relationship, perhaps because this self-report is a weak indicator of social skills. The network ties explanation produces similar data challenges, requiring sibling data during childhood along with sibling network data in adulthood.

While it is important to sort out these potential explanations, determining the generalizability of the sibship size/divorce association may be an even more pressing task. Replication is a key feature of scientific advancement (Freese & Peterson, 2017), and the sibship size/divorce association currently lacks a broad empirical base. It has only been produced in two datasets at this point and so building more confidence in this initial pattern is especially important. Since both previous studies employed U.S. data, an obvious question is whether the pattern generalizes to other contexts with varying fertility and divorce patterns.

Another reason extending this literature to other countries is important is because some sibship size research posits that sibling effects vary depending on the degree to which the nuclear family is buttressed by extended family or the broader community (Gibbs et al., 2016). For example, Gibbs et al. (2016) found that the inverse association between sibship size and years of education was about one-third as strong among Mormons versus Protestants, a pattern they explained as a product of more communal norms among Mormons—a greater tendency for both extended family (i.e., aunts, uncles, grandparents) and non-kin to invest in children’s well-being. In contexts where extended family plays a larger role, sibship size is thought to play a smaller one (Blaabæk et al., 2020; Gibbs et al., 2016).³ For our purposes, China is characterized by communitarian values and greater reliance on extended family, providing an attractive environment for assessing whether a weaker sibship size/divorce relationship is produced in this context.

Siblings and Divorce in China

There are several characteristics of Chinese society that likely influence the sibship size/divorce relationship. One key feature is historically low divorce rates (Ye & Lin, 1998). During the 1960s and 1970s these low rates were due, in part, to the practice of “mediation,” which clearly impeded divorce (Platte, 1988). A key moment was the abandonment of the mediation requirement in 1981, after which the view that divorce was an individual choice gained stature. More legal changes eventually allowed for unilateral divorce in the cases of domestic violence and extramarital relationships (Sun & Zhao, 2016). And while divorce rates in China remain relatively low, many of the social forces scholars believe are responsible for increasing divorce in western countries (e.g., women’s increasing educational attainment, later age at marriage, lower fertility) have also been shaping family life in China for several decades now; the result is rising divorce (Wang & Schofer, 2018). One idea is that China’s increasing participation in global institutions has legitimized particular values, such as individualism, human rights, and gender equality, all of which allow for non-traditional family forms.⁴ These features point to a changing normative climate in China, and the possibility that newer cohorts will exhibit different patterns from older cohorts.

In addition, China’s 1978 One Child Policy shaped fertility for decades. The one-child policy was nearly universal at first and families endured financial penalties if they had a second child. The policy was eventually relaxed in the mid-1980s and rural parents were allowed to have a second child if their first was a daughter.⁵ By 2015, the government established a two-child limit and in 2021 all limits and penalties were removed. The precise number of births averted is debated (Goodkind, 2017; Wang et al., 2018), but the policy clearly had a significant influence on Chinese fertility for several decades.

At present, we know little about the relationship between number of siblings and the likelihood of divorce in China. Some family scholars have found that having more children (especially sons) is associated with a lower probability of divorce in the Chinese Family Panel Study (Xu et al., 2015, 2016). But we know of no scholarship that has asked our research question—whether the number of siblings one grows up with is associated with the likelihood of divorce as an adult.

There is some evidence, however, that children growing up without siblings struggle to develop the kinds of personality characteristics that would promote good relationships. Cameron et al. (2013) compared 421 children born just prior (or just after) China’s One-Child Policy in 1979, using the exogenous policy shock as leverage for identifying the sibship size “effect.” The scholars found that lacking siblings resulted in significantly less trusting, less trustworthy, and less conscientious individuals (Cameron et al., 2013), all characteristics that could strain a marriage and increase the likelihood of divorce. The authors noted that children without siblings were known as “little emperors” lacking social skills. This study suggests that growing up as an only child (at least in China) produces personality outcomes that may not be conducive to long-term relationships. Based on this evidence, our first hypothesis notes that we expect that the inverse relationship between number of siblings and divorce observed in American samples to generalize to China, although weaker due to China’s more communal values.

Hypothesis 1a

In China, number of siblings will be inversely related to the probability of divorce.

Hypothesis 1b

In China, the inverse association between number of siblings and the likelihood of divorce will be weaker than in the U.S.

Siblings and Divorce in Europe

Europe represents another attractive context in which to extend this line of inquiry for several reasons. First, several European countries have experienced below-replacement level fertility, and so it is well-suited for testing the effects of small sibships. For example, the fertility rate in Spain and Italy has plummeted to 1.2 and 1.3, respectively, well below replacement level. And while both China and Europe’s fertility has been shaped by globalization, urbanization, delayed age at first marriage, and changing gender roles, Europe has not experienced the additional level of state influence that China faced with the One Child Policy. As a result, the differences between parents with few versus many children are likely greater in Europe. Because the selectivity issues play an important role in this research, we are interested to see if the sibship size/divorce relationship is similar across contexts where the degree of selectivity arguably varies.

Similar to the U.S., divorce in Europe has increased since the 1960s, levelling off since the 1980s (EUROSTAT, 2018). Several fundamental predictors of divorce (age, age at first marriage, presence of children, gender, income, and education) point in the same direction across the European continent, but the effect sizes vary within regions or specific nations (Härkönen, 2015). For instance, in a meta-analysis of studies predicting divorce risk across Europe, Wagner and Weiß (2006) find that even a well-established predictor of divorce—the presence of children—ranges in effect size from a 25% decline in the risk of divorce in Germany to a 70% decline in the Netherlands. As one possible explanation for such heterogeneity, Rijken and Liefbroer (2012) note that welfare states can shape divorce attitudes. Indeed, attitudes toward divorce tend to be associated with one’s assessment of the consequences of divorce (especially when children are involved). For example, Rijken and Liefbroer (2012) show that lower poverty rates among single parents and higher enrollment rates in early childhood care and education are both associated with less disapproval of divorce. Differences in fertility rates and views on traditional family values (divorce, marriage, cohabitation) also play a role in understanding divorce rates. Stevenson and Wolfers (2007) suggest that the U.S. might resemble European nations like Italy where comparatively few people believe that “marriage is an outdated institution” (p. 39). On the other hand, Sweden provides a powerful contrast where marriage rates are low and non-marital fertility is high—suggesting that childbirth and marriage are not as closely linked.

European scholarship exploring the relationship between sibship size and divorce is limited. The closest connection to our research question is a study which finds that individuals with a divorced sibling are more likely to experience divorce themselves, an effect similar in magnitude to having a divorced parent (de Vuijst et al., 2017). The authors suggested that the association might be due to how siblings, especially older ones, can serve as “behavioral examples on major transitions in the life course” (de Vuist et al., 2017, p. 2). This study highlights how the presence of a sibling (or siblings) may depend on characteristics of the sibling, a point we return to in the discussion.

Overall, Europe’s relatively high divorce rate is useful for our purposes because, combined with the substantial sample size in our European data, it produces a large number of divorced individuals—more than those found in previous U.S.-based studies with smaller sample sizes. This feature presents us with greater statistical power to identify both the bivariate sibship size/divorce association and its relationship net of covariates. The similarities between Europe and the U.S. in terms of fertility and divorce patterns lead us to expect the sibship size/divorce pattern to replicate in both direction and magnitude.

Hypothesis 2a

The association between number of siblings and the likelihood of divorce in Europe is inverse.

Hypothesis 2b

The association between number of siblings and the likelihood of divorce in Europe is similar in magnitude to that observed in the U.S.

Data and Method

We use data from the China General Social Survey (CGSS) and The Survey of Health, Aging, and Retirement in Europe (SHARE). We present sampling and measurement information for CGSS and SHARE individually and then discuss efforts to align our analytic and modeling approaches across the two data sources.

Chinese General Social Survey

Analytic Sample

The China General Social Survey (CGSS) is a cross-sectional survey project repeated annually/biannually to comprehensively assess the social life of people in China. Using a multi-stage stratified random sampling strategy, the CGSS recruited a nationally representative sample of residents who are 18 years old and above in People’s Republic of China (PRC) in each administration. We pooled data from the 2006, 2008, and 2017 years, which included questions about respondents’ sibship size since childhood. The sample size is 10,151 for 2006, 6000 for 2008, and 12,582 for 2017. Only a random sample of respondents (N = 3208 in 2006; N = 4127 in 2017) answered questions about sibship size since childhood. After restricting our sample to those who are currently or ever married, the sample size for the main analysis of pooled data is N = 11,634.

Measures

Siblings

The key independent variable of this study is the number of siblings. The CGSS surveys asked respondents to count their number of siblings since childhood, including those who are deceased. Based on this question, we constructed three sibling measures, including 1) a linear sibling variable ranging from 0 to a maximum of 10 siblings; 2) a dummy variable capturing the status of being the only child, where 1 = any siblings and 0 = no siblings at all; and 3) a categorical variable indicating sibship size: 1 = no siblings, 2 = one sibling, 3 = two siblings, 4 = three siblings, 5 = four siblings, and 6 = five or more siblings. We create three separate measures in order to assess whether we find a linear association with divorce, like past studies, whether the key distinction is the presence of any siblings (or not), and whether there are non-linear patterns.

Divorce

Respondents were asked about their current marital status at the survey time across the three survey years. We then created a dichotomous measure of whether currently divorced, where 1 = yes and 0 = no, as an indicator of respondents’ divorcing experience in adulthood. Those who reported separated were also coded as currently divorced. We note that this measure underestimates the percentage of respondents who ever divorced because some have divorced and remarried and some widowed respondents may also have experienced a divorce.

Control Variables

We also consider several covariates in our analyses to account for possible confounding of both independent and dependent variables. Drawing closely on previous studies (Bobbitt-Zeher et al., 2014; Merry et al., 2020), we include gender, birth cohort, any kids, first marriage age, religious belief, region, survey year, and several socioeconomic background variables (Xu et al., 2015, 2016). Gender is coded as 1 = male and 0 = female. For the CGSS sample, birth cohort is coded into six groups for every 10 years, where 1 = before 1949, 2 = 1950–1959, and 6 = 1990–1999. First marriage age is measured in years. Any kids is a dummy variable indicating whether respondents have kids or not. We also include several controls of respondents’ family origins and current socioeconomic status. Previous studies documented that hukou status—the unique household registration system in China to classify citizens into urban and rural residents—is related to socioeconomic benefits for urban residents (Wu, 2019). We thus control for respondents’ hukou status at 14 years old and current hukou status in the survey year, where 1 = urban and 0 = rural. We also include respondents’ educational attainment (1 = primary and below, 2 = secondary, 3 = tertiary) and mother’s educational attainment (0 = primary and below, 1 = secondary and above). Annual household income is a rank measure from 1 to 10 to capture one’s income percentile rank among all respondents of the same survey year. Religious belief is a dichotomous variable, where 1 = having religious belief and 0 = no religious belief. We also include a control for respondents’ current geographic regions based on the common approach in previous research (1 = Eastern, 2 = Central, 3 = Western, 4 = Northeastern) (e.g., Qian & Li, 2020). At last, to control for potential differences across the three survey years, we include a categorical indicator for survey years: 1 = 2006, 2 = 2008, and 3 = 2017.

The Survey of Health, Aging, and Retirement in Europe

Analytic Sample

The Survey of Health, Aging, and Retirement in Europe (SHARE) is a cross-national panel database that includes survey responses from more than 140,000 individuals across 28 European countries and Israel. The target population consists of all individuals 50 years or older⁶ (at time of survey) who reside in the respective SHARE country. Data collection was performed via Computer-Assisted Personal Interviewing (CAPI). The survey is ongoing and first started in 2004 (Börsch-Supan & Jürges, 2005). The present study utilizes data from the first seven waves of data collection (2004–2017). In addition to a regular questionnaire administered during each wave, Wave 3 (2008–2009) and Wave 7 (2017) also include specific modules on life-history data. This specific data collection component is referred to as SHARELIFE. The SHARELIFE questionnaire contains information critical to our analysis (household size during childhood and details on relationship formation and dissolution). As such, our analytic sample is necessarily limited to the approximately 92,000 individuals that successfully completed the SHARELIFE interview in either Wave 3 or Wave 7. Our final analytic sample is further restricted to include only those respondents who have ever been married (N = 86,014).

Measures

Siblings

The key independent variable of this study is number of siblings (including full, step, foster, and adopted siblings). The SHARE study includes a direct question on whether the respondent ever had any siblings (only child vs. not), but the measurement of precise sibship is more complicated. We proceed with two different approaches to estimate respondents’ number of siblings. The first approach uses response data from prompts about how many brothers and sisters the respondent has that are still alive. In using this measure, we draw from respondents’ earliest participation in SHARE. If respondents reported zero siblings alive and also reported that they were only children, we coded them as having zero siblings growing up. But if they reported zero siblings alive but did not report having no siblings growing up, we reasoned that they were raised with at least one sibling and this sibling had since passed away. For this group, we coded them as having one sibling, which may underestimate their total number of siblings growing up. We opt for this adjustment as our primary theoretical motivation is tied to the familial dynamic of growing up with siblings, not necessarily how many siblings are still alive later in the life course. The resulting variable for ‘number of siblings alive’ is top-coded at 10 siblings (0–10+) and has a mean value of 2.23 siblings (see Table 2).

Table 2.

Survey of Health, Aging, and Retirement in Europe (SHARE) Descriptives.^a

Variables	N	Mean	Std. Dev.	Min	Max	Description
Dependent variables
Currently divorced	86,014	0.09		0	1
Ever divorced	86,014	0.19		0	1
Independent variables
Siblings (linear)	82,846	3.16	2.09	0	10	Coded with highest # at 10+
Siblings alive (linear)	85,232	2.23	1.89	0	10	Coded with highest # at 10+
Any siblings	82,084	0.90		0	1
Siblings (category)						Siblings in house at age 10
0 Siblings	82,846	.10		0	1
1 Siblings	82,846	.05		0	1
2 Siblings	82,846	.26		0	1
3 Siblings	82,846	.23		0	1
4 Siblings	82,846	.14		0	1
5+ Siblings	82,846	.21		0	1
Siblings alive (category)						At first survey wave
0 Siblings	85,232	.10		0	1
1 Siblings	85,232	.35		0	1
2 Siblings	85,232	.22		0	1
3 Siblings	85,232	.13		0	1
4 Siblings	85,232	.08		0	1
5+ Siblings	85,232	.12		0	1
Control variables
Male	86,014	0.43		0	1
Birth cohort
Before 1920	85,989	<0.01		0	1
1920–1929	85,989	0.06		0	1
1930–1939	85,989	0.18		0	1
1940–1949	85,989	0.31		0	1
1950–1959	85,989	0.33		0	1
1960–1969	85,989	0.11		0	1
1970–1995	85,989	<0.01		0	1
Age	85,989	68.59	10.20	27	105	Age at most recent survey
First marriage age	83,240	24.55	5.51	14	85
Biological Parents	86,014	0.87		0	1	Lived with bio parents, age 10
Urban	77,181	0.40		0	1
Educational attainment	80,792	2.84	1.46	0	6	ISCED levels
Mother educational attainment	50,514	1.49	1.27	0	6	ISCED levels
Annual household income	21,243	7.32	2.71	1	10	Decile rank in country
Any kids	85,822	0.92		0	1
Homeowner	47,716	74.81		0	1	R owns home
Religiosity	58,630	2.75	1.06	1	4	4 = religion is very important
Religious attendance	50,271	0.12		0	1	Yes/no during past month
Western Europe	86,014	0.28		0	1
Northern Europe	86,014	0.21		0	1
Southern Europe	86,014	0.33		0	1
Eastern Europe	86,014	0.18		0	1
Survey year (first wave)	86,014	2010.78	4.69	2004	2017

^aSamples included are those ever-married respondents in the SHARE dataset.

Our second approach to measuring number of siblings involves data from the retrospective life-history questionnaires. A series of questions asks respondents to provide the total number of people living in their household at age 10 and then specify which relations (parents, siblings, grandparents, other relatives, other non-relatives) are included in the estimate. From this question, we can determine whether the respondent had siblings, but we cannot know exactly how many because the specific numbers of individuals are not provided for siblings, grandparents, other relatives, and other non-relatives—only whether they were included in the overall estimate.⁷

Our solution is to assign “one person” to any instance where grandparents, other relatives, or other non-relatives were included in the count. We add this number to the precise number of parents present at age 10 (this information is provided) and subtract this total from the overall number of people reported as living in the household. The resulting difference is our best estimate of how many siblings lived with the respondent at the time. The number is once again top-coded at 10 siblings and has a mean value of 3.16 siblings (Table 2).

We proceed with both siblings variables in our analysis. While neither represents a perfect measure of “number of siblings during childhood,” they both provide unique and useful information. The “number of siblings alive” variable is imperfect in the sense that we do not have the most accurate depiction of the true number of siblings that were ever a part of the respondent’s life. Similarly, the retrospective measure of siblings in one’s house at age 10 is an estimate and also necessarily excludes younger siblings born any time after the respondent was 10 years of age. Even with such limitations in place, we find either measure to be a reasonable gauge for the experience of growing up with varying numbers of siblings. Moreover, both measures come with a high degree of face validity as the patterns are roughly in line with historic fertility trends. The two measures also share a strong correlation (.74) suggesting that while they do vary from one another and capture estimates at different points in the life course, they are depicting largely similar family structure dynamics for the respondents included in our sample. Finally, we use each of these linear operationalizations of siblings to create categorical measures of precise sibship size (no siblings, one sibling, two siblings, three siblings, four siblings, and five or more siblings).

Divorce

The main dependent variable of this study is whether respondents experienced divorce during adulthood. To mirror the approach, we take with the CGSS data, our primary analysis of divorce in SHARE is based on respondent’s most recent response about their current marital status. Among those ever married, we code those who replied with “divorce” or “separated” as 1 = divorced. All other current statuses (married or widowed) are coded as 0 = not divorced. The SHARELIFE questionnaire, however, provides additional leverage with a complete relationship history module. This allows us to predict the odds of “ever divorcing” in supplemental analysis (Appendices C and F).

Control Variables

We attempt to adjust for the effects of obvious confounds in our models predicting divorce. The following control variables are included in our analysis of the SHARE data: gender, birth cohort, age, first marriage age, urban residence, whether the respondent has any kids, religiosity, religious attendance, region, survey year, and several socioeconomic background variables. Gender is coded as 1 = male and 0 = female. For the SHARE sample, birth cohort is coded into seven groups (typically by decade), where 1 = before 1920, 2 = 1920–1929, 3 = 1930–1939 and so on, up to 7 = 1970–1995. First marriage age is measured in years. Urban residence is a dummy variable (1 = lives near or in urban center 0 = not near or in urban center). Any kids is a dummy variable indicating whether respondents have kids or not (1 = yes; 0 = no). Religiosity is a retrospective measure that assesses the strength of religious affiliation in one’s childhood home (1–4 scale; where 4 = religion is very important), while Religious Attendance measures whether the respondent attended religious services during the past month (1 = yes; 0 = no) as of the most recent survey. Survey Year indicates respondent’s first wave of SHARE participation (ranges from 2004 to 2017). Region refers to European region (Western Europe, Northern Europe, Southern Europe, and Eastern Europe) as classified by the United Nations Geoscheme. Controlling for European region of residence is a common approach in the social science literature and even specifically for studies utilizing the SHARE dataset (Scheel et al., 2019; Walkden et al., 2018). However, we also acknowledge that country-specific context may play a role in shaping the association between sibship and divorce outcomes. In supplemental analysis (Appendices D, E, and F), we present condensed models that control for country of residence in lieu of region.

Additional controls for family background and socioeconomic status include whether or not the respondent grew up with both biological parents (1 = yes; 0 = no), educational attainment, mother’s educational attainment, household income, and homeownership. For both the respondent and respondent’s mother, educational attainment is measured using the ISCED-97 categories (0 = no schooling; 6 = advanced degree). Household income is measured via decile rank location (1–10) relative to the income distribution in one’s nation of residence. Homeownership is a dummy variable (1 = yes, homeowner; 0 = no, tenant/renter).

Aligning Models Across Datasets

Where possible, we made every effort to align our measurement strategies and modeling choices across the two datasets. To reiterate our approach to operationalizing our independent variable of siblings, our measurement strategy is three-fold: 1) a dichotomous variable measuring whether the respondent grew up with any siblings. In other words, whether or not the respondent is an only child. 2) a linear siblings variable, top-coded at 10+ siblings (see the measurement section above for a detailed discussion of our coding procedures). Recall that we utilize two distinct linear siblings variables from the SHARE dataset—an estimate of how many siblings the respondent was living with at age 10 as well as the number of siblings still alive as reported in the respondent’s earliest wave of SHARE participation. 3) a categorical measure capturing the unique experience of growing up with a specific number of siblings, from one sibling to five or more. In all SHARE analyses, we once again provide two different approaches—one based on each of the two aforementioned linear measurements.

To predict whether respondents are currently divorced among the CGSS and SHARE samples, we fit logistic regression models. We first test the bivariate relationship between siblings and currently divorced and then add control variables into the full models to examine whether the relationship still holds. Several of these controls are present in both the SHARE and CGSS datasets. The following variables are coded identically across the two sources: age at first marriage, birth cohort by decade, gender, whether the respondent currently resides in an urban area, and whether the respondent has any children. We report respondents’ age at the time of the most recent survey (in years), as part of our descriptive analysis but we do not include age in our models due to multicollinearity with our measure for birth cohort. Our coding strategy differs slightly for the following variables as a result of initial measurement decisions during data collection: respondent’s educational attainment, mother’s educational attainment, income, and religiosity (see previous section of measurement for more details on measurement differences).

Finally, our full models differ in some aspects because variables that were available in one dataset were not in the other. In the SHARE dataset, these variables include the following: current religious attendance, whether the respondent owns their home, and whether the respondent grew up with both biological parents. The final SHARE models also include a series of dummy variables for distinct European regions (Western, Eastern, Northern, and Southern Europe). Meanwhile, CGSS models include unique control variables for urbanicity at age 14 (hukou status) as well as a region variable (1 = Eastern, 2 = Central, 3 = Western, 4 = Northeastern). In supplemental analysis (Appendices A and B) interaction terms between sibship size and current urban residence (hukou in CGSS) and between sibship size and birth cohorts are added to determine whether the primary relationship under investigation varies by theoretically selected subgroups.

In all analyses, we used the chained equations of multiple imputation in Stata 15 to generate 10 datasets to deal with the missing data. All variables used in the analyses are included in the imputation models. To avoid the unnecessary random noise of imputed values of dependent variables, we restrict analyses to only those cases with observed values of the dependent variables after imputation (von Hippel & Paul, 2007).

Results

Number of Siblings and Divorce in China

Among those ever married, three percent of our Chinese sample reported that they were currently divorced (Table 1). Recall that the divorce variable in the Chinese data is limited to those who are currently divorced, and so likely underestimates the number who have ever been divorced. The mean level of siblings is 3.11. That number appears high, given the 1979 One Child Policy, but we note that nearly 90 percent of the sample was born prior to that policy’s implementation. The average age is 48.13, and just seven percent are only children.

Table 3.

Logistic Regression Models Predicting Current Divorce in China General Social Survey (CGSS).^a,b,c

Variables	Sibs (Linear)		Any Sibs		Sibsize
Variables	Empty	Full Model	Empty	Full Model	Empty	Full Model
Siblings (linear)	0.886***	0.890**
Siblings (linear)	(0.028)	(0.032)
Any siblings			0.912	1.017
Any siblings			(0.196)	(0.239)
1 Sibling versus none					1.266	1.327
1 Sibling versus none					(0.309)	(0.336)
2 Siblings versus none					1.119	1.077
2 Siblings versus none					(0.264)	(0.272)
3 Siblings versus none					0.907	0.888
3 Siblings versus none					(0.222)	(0.236)
4 Siblings versus none					0.762	0.765
4 Siblings versus none					(0.197)	(0.216)
5 and more siblings versus none					0.622	0.660
5 and more siblings versus none					(0.156)	(0.183)
Male		1.107		1.125		1.106
Male		(0.132)		(0.135)		(0.132)
Birth cohort (1949 and before omitted)
1950–1959		2.689***		2.582***		2.686***
1950–1959		(0.656)		(0.629)		(0.657)
1960–1969		4.491***		4.444***		4.467***
1960–1969		(1.055)		(1.043)		(1.053)
1970–1979		3.522***		3.804***		3.435***
1970–1979		(0.884)		(0.950)		(0.865)
1980–1989		1.487		1.764		1.502
1980–1989		(0.512)		(0.605)		(0.517)
1990–1999		1.582		1.896		1.521
1990–1999		(1.225)		(1.466)		(1.181)
First marriage age		1.053***		1.054***		1.053***
First marriage age		(0.013)		(0.013)		(0.013)
Current urban hukou		2.148***		2.160***		2.149***
Current urban hukou		(0.397)		(0.399)		(0.398)
Urban hukou at 14		1.967***		2.060***		1.987***
Urban hukou at 14		(0.325)		(0.339)		(0.328)
Educational attainment (Primary education omitted)
Secondary education		1.091		1.100		1.076
Secondary education		(0.175)		(0.176)		(0.174)
Tertiary education		0.811		0.835		0.820
Tertiary education		(0.206)		(0.211)		(0.208)
Mother secondary education and above		0.850		0.917		0.850
Mother secondary education and above		(0.161)		(0.172)		(0.161)
Annul household income		0.843***		0.847***		0.842***
Annul household income		(0.023)		(0.023)		(0.023)
Any kids		0.338***		0.327***		0.330***
Any kids		(0.074)		(0.072)		(0.072)
Religious belief		1.768***		1.726***		1.752***
Religious belief		(0.293)		(0.285)		(0.290)
Region (Eastern omitted)
Central		0.651*		0.638*		0.647*
Central		(0.125)		(0.122)		(0.124)
Western		1.102		1.066		1.091
Western		(0.174)		(0.168)		(0.172)
Northeastern		1.185		1.131		1.168
Northeastern		(0.212)		(0.202)		(0.209)
Survey year (2006 omitted)
2008		1.045		1.031		1.046
2008		(0.165)		(0.163)		(0.166)
2017		2.131***		2.120***		2.136***
2017		(0.348)		(0.346)		(0.349)
Constant	0.040***	0.008***	0.030***	0.006***	0.030***	0.007***
Constant	(0.004)	(0.004)	(0.006)	(0.003)	(0.006)	(0.004)
N	11,634	11,634	11,634	11,634	11,634	11,634

*p < .05. **p < .01. ***p < .001.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

Table 3 presents logistic regression coefficients (reported in odds ratio) predicting the probability of divorce among those ever married. The linear effect of siblings (b = .886, p < .001) suggests that each additional sibling is associated with a roughly 11% decline in the probability of divorce. Notably, this point estimate remains intact in a full model controlling for sex, birth cohort, age at first marriage, current urban hukou, urban hukou at age 14, educational attainment, mother’s educational attainment, annual household income, whether the respondent has any children, their religious beliefs, region of the country, and survey year.

Previous scholars have also tested whether the addition of any siblings (vs. none) is associated with the probability of divorce. Like others, we find that it is not (Table 3, Any Sibs). Finally, we tested the relationship between binary variables representing each sibship size category and the likelihood of divorce and again found no statistically significant relationships.

Number of Siblings and Divorce in Europe

Among those ever married, the average number of siblings among respondents in our European sample was similar to the Chinese sample—3.16 (Table 2). The percent ever divorced was significantly higher, however—19 percent. The percent currently divorced, the only number we can directly compare to the Chinese sample, is nine percent, suggesting that divorce is about three times more likely in Europe than in China. Ninety percent of the sample reported having at least one sibling. As expected for a sample focusing on retirement issues, the average age was high--68.59.

Table 4 presents the logistic regression coefficients from predicting divorce among the European sample. The linear sibship size coefficient (b = .931, p < .001) suggests that each additional siblings is associated with a subsequent decline in the likelihood of divorce of about seven percent in the empty model. This is reduced in the full model with covariates, however, to 2.2%. The fact that the association declines so much in the full model suggests that selectivity issues (differences between the kinds of parents having many vs. few children) are more substantial in the European than Chinese sample.

Table 4.

Logistic Regression Models Predicting Current Divorce in SHARE.^a,b,c

	Sibs (Linear)		Sibs Alive (Linear)		Any Sibs		Sibsize		Sibsize Alive
Variables	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model
Siblings (linear)	0.931***	0.977***
Siblings (linear)	(0.006)	(0.006)
Siblings alive (linear)			0.971***	0.992
Siblings alive (linear)			(0.006)	(0.007)
Any Siblings					0.834***	0.915*
Any Siblings					(0.033)	(0.036)
1 Sibling versus none							1.048	1.016	0.867***	0.938
1 Sibling versus none							(0.063)	(0.063)	(0.035)	(0.039)
2 Siblings versus none							0.956	0.937	0.900*	0.947
2 Siblings versus none							(0.039)	(0.039)	(0.038)	(0.419)
3 Siblings versus none							0.878**	0.934	0.810***	0.880*
3 Siblings versus none							(0.037)	(0.041)	(0.039)	(0.044)
4 Siblings versus none							0.795***	0.901*	0.840**	0.967
4 Siblings versus none							(0.038)	(0.045)	(0.046)	(0.056)
5+ Siblings versus none							0.670***	0.867**	0.756***	0.901
5+ Siblings versus none							(0.030)	(0.041)	(0.038)	(0.049)
Male		0.829***		0.829***		0.829***		0.828***		0.830***
Male		(0.021)		(0.022)		(0.021)		(0.021)		(0.022)
Birth cohort (1919 and before omitted)
1920–1929		1.202		1.212		1.205		1.200		1.209
1920–1929		(0.333)		(0.336)		(0.333)		(0.332)		(0.335)
1930–1939		1.950*		1.972*		1.955*		1.947*		1.970*
1930–1939		(0.525)		(0.531)		(0.526)		(0.524)		(0.531)
1940–1949		3.409***		3.455***		3.421***		3.405***		3.452***
1940–1949		(0.916)		(0.929)		(0.919)		(0.915)		(0.929)
1950–1959		4.792***		4.859***		4.807***		4.790***		4.858***
1950–1959		(1.290)		(1.310)		(1.294)		(1.290)		(1.310)
1960–1969		4.666***		4.741***		4.691***		4.665***		4.739***
1960–1969		(1.266)		(1.287)		(1.272)		(1.266)		(1.287)
1970–1995		2.861**		2.916**		2.879**		2.859**		2.909**
1970–1995		(0.921)		(0.939)		(0.927)		(0.920)		(0.937)
First marriage age		0.996		0.996		0.996		0.996		0.996
First marriage age		(0.002)		(0.002)		(0.002)		(0.002)		(0.002)
Biological parents		0.807***		0.800***		0.803***		0.808***		0.802***
Biological parents		(0.027)		(0.027)		(0.027)		(0.027)		(0.027)
Urban		1.349***		1.352***		1.351***		1.349***		1.351***
Urban		(0.035)		(0.035)		(0.035)		(0.035)		(0.035)
Educational attainment		1.046***		1.049***		1.049***		1.046***		1.048***
Educational attainment		(0.011)		(0.011)		(0.011)		(0.011)		(0.011)
Mother’s Ed. Attainment		1.054***		1.056***		1.056***		1.054***		1.055***
Mother’s Ed. Attainment		(0.012)		(0.012)		(0.012)		(0.012)		(0.012)
Household income		0.976**		0.977**		0.977**		0.976**		0.977**
Household income		(0.006)		(0.006)		(0.006)		(0.007)		(0.007)
Homeowner		0.496***		0.497***		0.497***		0.497***		0.497***
Homeowner		(0.015)		(0.015)		(0.015)		(0.015)		(0.015)
Any kids		0.797***		0.793***		0.795***		0.797***		0.795***
Any kids		(0.034)		(0.034)		(0.034)		(0.034)		(0.034)
Religiosity		0.894***		0.890***		0.889***		0.893***		0.890***
Religiosity		(0.016)		(0.015)		(0.015)		(0.016)		(0.016)
Religious attendance		0.846***		0.844***		0.844***		0.846***		0.844***
Religious attendance		(0.035)		(0.035)		(0.035)		(0.035)		(0.035)
Eastern Europe		0.660***		0.662***		0.664***		0.661***		0.663***
Eastern Europe		(0.025)		(0.025)		(0.025)		(0.025)		(0.025)
Southern Europe		0.432***		0.433***		0.433***		0.432***		0.433***
Southern Europe		(0.016)		(0.016)		(0.016)		(0.016)		(0.016)
Northern Europe		1.037		1.039		1.043		1.038		1.041
Northern Europe		(0.034)		(0.034)		(0.034)		(0.034)		(0.034)
Year of first Survey		1.018***		1.018***		1.018***		1.018***		1.018**
Year of first Survey		(0.003)		(0.003)		(0.003)		(0.003)		(0.003)
Constant	0.131***	0.000***	0.112***	0.000***	0.123***	0.000***	0.122***	0.000***	0.121***	0.000***
Constant	(0.003)	(0.000)	(0.002)	(0.000)	(0.005)	(0.000)	(0.004)	(0.000)	(0.004)	(0.000)
N	86,014	86,014	86,014	86,014	86,014	86,014	86,014	86,014	86,014	86,014

*p < .05. **p < .01. ***p < .001.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

Similar to the U.S. and Chinese patterns, the strongest support for the siblings/divorce relationship is evident when siblings are operationalized as a linear term, with other operationalizations producing less consistency. Recall that we also developed a number of “siblings alive” variable for the European data. This variable produces a statistically significant association with divorce (b = .971, p < .001) but this relationship becomes insignificant in the full model, undermining the sibling network explanation. We also distinguished between respondents who had any siblings versus none. Here, our empty model suggested that those with any siblings were 17% less likely to divorce than those without siblings (b = .834, p < .001), which declined to eight percent in the full model (b = .915, p < .05). Finally, our categorical siblings variables produced some statistically significant patterns—namely those in larger sibships were less prone to divorce.

Discussion

Our results reveal that divorce is less likely among those with many versus few siblings in both China and Europe. Combined with the U.S. patterns from previous studies, there is now evidence of this inverse association across multiple contexts. And while the point estimates vary, the bigger news is that the inverse association between number of siblings and divorce is observable in societies with widely different demographic and cultural contexts. This result prompts us to seriously consider the possibility that sibling experiences during childhood may have a long-term influence on divorce as an adult.

Inspired by scholars emphasizing the importance of context (Blaabæk et al., 2020; Gibbs et al., 2016), we anticipated that the pattern would replicate in China, but that the communal aspect of Chinese culture would produce a weaker sibship size/divorce relationship than in the U.S. Scholars emphasizing a contextual understanding of siblings have argued that what occurs within nuclear families is conditioned by the extent to which children’s development is shared with a broader community. In largely individualistic societies, children’s outcomes are shaped more by the nuclear family, conditions in which siblings should matter more. In more communal societies, such as China, children’s development may not be so dependent on nuclear family processes. But we found that the consequences of siblings for divorce were as strong (or stronger) in China, contrary to hypothesis 1a.⁸ This pattern adds to the empirical basis for evaluating the ways in which context shapes (or does not shape) the consequences of sibship size. Current evidence suggests that the association between number of siblings and the likelihood of divorce is remarkably durable across context.

This overall conclusion comes with several caveats, however. To maintain comparability across the Chinese and European datasets, we presented results for “currently divorced,” because “ever divorced” was unavailable in the Chinese data. It was available in the European data, however, and so in Appendix C we present these results. They are very similar but suggest that the reduction in the likelihood of divorce associated with each additional sibling is three percent, more in line with the U.S. studies. If “ever divorced” were available in the Chinese data, we assume that it would show an even stronger association with siblings. Another complicating factor is that we found no evidence that the association between siblings and divorce is stronger among urban (typically more modern) than non-urban respondents (see Appendices A and B).

A second caveat is that the inverse association is strongest when siblings are coded as a linear variable. Alternative codings, such as any siblings versus no siblings, or categorical groupings, produce less consistent findings. This is similar to what previous scholars studying U.S. data found and so an important remaining question is: Why does the linear operationalization produce the most consistent pattern? If siblings reduce the probability of divorce by providing social skills practice partners, one would expect a distinction between having any versus no siblings to produce a consistent association with divorce. It may be that, although the number of siblings is generally associated with a lower probability of divorce, siblings vary in meaningful ways in terms of how much they promote social skills. Perhaps if we were able to distinguish kinds of siblings more precisely (older vs. younger, sisters vs. brothers, closely vs. widely-spaced), we would be able to identify the kinds of siblings that reduce the likelihood of divorce. Drawing on previous research that has been able to make these finer distinctions, one might predict, for example, that an older, widely-spaced sister would reduce the probability of divorce more than a younger, closely-spaced brother (Powell & Steelman, 1990, 1993). Along these same lines, we recognize that not all siblings are the same—some likely promote social skills while others do not. The quality of the sibling relationship, therefore, may matter more than the mere presence of siblings (Yucel and Downey, 2015). It is not possible, however, to explore these more nuanced issues with our data.⁹

In addition, the theoretical mechanisms by which siblings are supposed to reduce divorce (via promoting social skills or greater network ties) remain largely unexplored. Previous scholars have found little support for these mechanisms, but they have lacked quality indicators. As a result, there is currently a disjuncture between the relatively robust association observed between number of siblings and divorce and the rather weak ability of scholars to test potential mechanisms for this association. This problem will persist until we overcome an important data limitation: we need the ability to produce strong tests of the sibship size/divorce relationship and the sibship size/social skills and sibship size/network ties relationships with the same data. One particular challenge for the literature is that the kinds of datasets that provide good estimates of the sibship size/divorce relationship do not tend to have strong indicators of the kinds of mechanisms (e.g., social skills) thought to explain the relationship. For example, the GSS data have indicators of whether the respondent thinks of themselves as “sociable,” but it is unlikely that this kind of self-report is a good indicator of the ability to develop and maintain long-term relationships. Accordingly, scholars have had little success explaining the sibship size/divorce relationship with theoretically meaningful mediators.

With weak support for the mechanisms explaining the association, the possibility that we are observing a spurious relationship cannot be dismissed. We found that the sibling coefficient declines considerably when covariates are included in the European model, suggesting important differences between the kinds of families having many versus few children. Although we included a broad range of control variables, it is possible that an additional unobserved variable accounts for the association. As one example, we included several indicators of socioeconomic status in our models (e.g., educational attainment, mother’s educational level, household income, and homeowner status) but these may not capture variation in life opportunities better gauged by wealth, highlighting the difficulty in equalizing respondents with many versus few siblings via covariate adjustment.

There is also meaningful heterogeneity in some of our results, which can be explored primarily in the large European data. In supplemental analyses restricting the European sample to each country or region, one at a time, we found that the inverse association between siblings and divorce is not observed in every occasion, but the majority of national or regional contexts do produce results consistent with our hypothesis (Appendix D). In appendices E and F, we present condensed results based on our full modeling strategy—the only difference is that we control for country of residence instead of region. While only the linear siblings effect remains statistically significant when predicting current divorce (Appendix E), we continue to find robust effects across sibling operationalizations in models predicting ever being divorced (Appendix F).

Finally, our study hints at a potentially pro-social benefit of siblings—greater skill at maintaining marriage—yet if we expand our focus across later life stages, the presence of siblings may have more complex consequences. Adult children negotiate several family issues with siblings, such as aiding aging parents and dividing resources once those parents are deceased. These periods have the potential to bond siblings together toward a common cause or produce significant disputes, often fraught with emotional tension (Conway, 2016). Understanding how siblings matter across the entire life course would provide a more comprehensive understanding of the consequences of growing up with many versus few sisters and brothers.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Doug B. Downey

Joseph Merry

Notes

Appendix A

Table A1.

Interaction Models Predicting Current Divorce in CGSS.^a,b,c,d

Variables	Sibs (Linear)
Variables	Sibs × Urban	Sibs × Cohort
Siblings (linear)	0.861*	0.889
Siblings (linear)	(0.055)	(0.091)
Current urban hukou	1.857*	2.146***
Current urban hukou	(0.542)	(0.398)
Birth cohort (1949 and before omitted)
1950–1959	2.697***	2.939*
1950–1959	(0.658)	(1.310)
1960–1969	4.520***	3.898**
1960–1969	(1.063)	(1.641)
1970–1979	3.537***	3.739**
1970–1979	(0.888)	(1.608)
1980–1989	1.489	1.670
1980–1989	(0.512)	(0.859)
1990–1999	1.547	3.982
1990–1999	(1.199)	(4.104)
Male	1.108	1.103
Male	(0.132)	(0.132)
First marriage age	1.053***	1.053***
First marriage age	(0.013)	(0.013)
Urban hukou at 14	1.976***	1.969***
Urban hukou at 14	(0.326)	(0.326)
Educational attainment (Primary education omitted)
Secondary education	1.084	1.092
Secondary education	(0.175)	(0.176)
Tertiary education	0.814	0.801
Tertiary education	(0.206)	(0.204)
Mother secondary education and above	0.859	0.842
Mother secondary education and above	(0.163)	(0.161)
Annual household income	0.844***	0.844***
Annual household income	(0.023)	(0.023)
Any kids	0.337***	0.341***
Any kids	(0.073)	(0.075)
Religious beliefs	1.768***	1.782***
Religious beliefs	(0.293)	(0.296)
Region (Eastern omitted)
Central	0.650*	0.651*
Central	(0.125)	(0.125)
Western	1.101	1.109
Western	(0.174)	(0.175)
Northeastern	1.181	1.181
Northeastern	(0.211)	(0.211)
Survey year (2006 omitted)
2008	1.043	1.042
2008	(0.165)	(0.165)
2017	2.124***	2.118***
2017	(0.347)	(0.346)
Siblings × current urban hukou	1.049
Siblings × current urban hukou	(0.079)
Siblings × Birth cohort (1949 and before omitted)
1950–1959 × Number of siblings		0.973
1950–1959 × Number of siblings		(0.120)
1960–1969 × Number of siblings		1.047
1960–1969 × Number of siblings		(0.122)
1970–1979 × Number of siblings		0.973
1970–1979 × Number of siblings		(0.126)
1980–1989 × Number of siblings		0.896
1980–1989 × Number of siblings		(0.224)
1990–1999 × Number of siblings		0.318
1990–1999 × Number of siblings		(0.372)
Constant	0.009***	0.008***
Constant	(0.005)	(0.005)
N	11,634	11,634

*p < .05. **p < .01. ***p < .001.

× Cohort is the interaction model between siblings and birth cohort.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

^d× Urban is the interaction model between siblings and current urban hukou.

Appendix B

Table B1.

Interaction Models Predicting Current Divorce in SHARE.^a,b,c,d

Variables	Sibs (Linear)		Sibs Alive (Linear)
Variables	Sibs x Urban	Sibs x Cohort	Sibs x Urban	Sibs x Cohort
Siblings (linear)	0.974**	0.941	0.995	0.984
Siblings (linear)	(0.008)	(0.121)	(0.009)	(0.203)
Urban	1.320***	1.349***	1.370***	1.352***
Urban	(0.059)	(0.035)	(0.057)	(0.035)
Birth cohort (1919 and before omitted)
1920–1929	1.202	1.072	1.212	1.208
1920–1929	(0.333)	(0.563)	(0.336)	(0.498)
1930–1939	1.951*	1.645	1.973*	1.916
1930–1939	(0.525)	(0.849)	(0.532)	(0.770)
1940–1949	3.410***	2.963*	3.458***	3.384**
1940–1949	(0.916)	(1.523)	(0.930)	(1.354)
1950–1959	4.792***	4.222**	4.862***	4.796***
1950–1959	(1.290)	(2.176)	(1.311)	(1.920)
1960–1969	4.666***	4.280**	4.745***	4.825***
1960–1969	(1.266)	(2.208)	(1.289)	(1.940)
1970–1995	2.859**	2.658	2.919**	4.166**
1970–1995	(0.921)	(1.608)	(0.941)	(2.024)
Male	0.829***	0.829***	0.829***	0.829***
Male	(0.021)	(0.021)	(0.022)	(0.022)
First marriage age	0.996	0.996	0.996	0.996
First marriage age	(0.002)	(0.002)	(0.002)	(0.002)
Biological parents	0.807***	0.806***	0.800***	0.798***
Biological parents	(0.027)	(0.027)	(0.027)	(0.027)
Educational attainment	1.046***	1.046***	1.049***	1.049***
Educational attainment	(0.011)	(0.011)	(0.011)	(0.011)
Mother’s Ed. attainment	1.054***	1.053***	1.056***	1.055***
Mother’s Ed. attainment	(0.012)	(0.012)	(0.012)	(0.012)
Household income	0.976**	0.976**	0.977**	0.977**
Household income	(0.007)	(0.007)	(0.006)	(0.006)
Homeowner	0.496***	0.496***	0.497***	0.497***
Homeowner	(0.015)	(0.015)	(0.015)	(0.015)
Any kids	0.797***	0.797***	0.793***	0.793***
Any kids	(0.034)	(0.034)	(0.034)	(0.034)
Religiosity	0.894***	0.894***	0.890***	0.890***
Religiosity	(0.016)	(0.016)	(0.015)	(0.015)
Religious attendance	0.846***	0.846***	0.844***	0.844***
Religious attendance	(0.035)	(0.035)	(0.035)	(0.035)
Eastern Europe	0.660***	0.660***	0.662***	0.662***
Eastern Europe	(0.025)	(0.025)	(0.025)	(0.025)
Southern Europe	0.432***	0.432***	0.433***	0.432***
Southern Europe	(0.016)	(0.016)	(0.016)	(0.016)
Northern Europe	1.037	1.036	1.039	1.038
Northern Europe	(0.034)	(0.034)	(0.034)	(0.034)
Year of first Survey	1.018***	1.018***	1.018**	1.018***
Year of first Survey	(0.003)	(0.003)	(0.003)	(0.003)
Siblings x Urban	1.007		0.994
Siblings x Urban	(0.013)		(0.014)
Siblings x Birth cohort
1920–1929 × siblings		1.034		1.003
1920–1929 × siblings		(0.137)		(0.212)
1930–1939 × siblings		1.053		1.017
1930–1939 × siblings		(0.137)		(0.211)
1940–1949 × siblings		1.042		1.012
1940–1949 × siblings		(0.135)		(0.209)
1950–1959 × siblings		1.038		1.009
1950–1959 × siblings		(0.134)		(0.208)
1960–1969 × siblings		1.022		0.994
1960–1969 × siblings		(0.132)		(0.206)
1970–1995 × siblings		1.016		0.825
1970–1995 × siblings		(0.170)		(0.200)
Constant	0.000***	0.000***	0.000***	0.000***
Constant	(0.000)	(0.000)	(0.000)	(0.000)
N	86,014	86,014	86,014	86,014

*p < .05. **p < .01. ***p < .001.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

^d× Urban is the interaction model between siblings and current urban residence; × Cohort is the interaction model between siblings and birth cohort.

Appendix C

Table C1.

Logistic Regression Models Predicting Ever Divorced in SHARE.^a,b,c

Variables	Sibs (Linear)		Sibs Alive (Linear)		Any Sibs		Sibsize		Sibsize Alive
Variables	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model
Siblings (linear)	0.911***	0.967***
Siblings (linear)	(0.004)	(0.005)
Siblings alive (linear)			0.949***	0.977***
Siblings alive (linear)			(0.005)	(0.005)
Any Siblings					0.81***	0.916**
Any Siblings					(0.023)	(0.027)
1 Sibling versus none							1.133**	1.085	0.865***	0.954
1 Sibling versus none							(0.050)	(0.051)	(0.026)	(0.030)
2 Siblings versus none							0.957	0.949	0.881***	0.948
2 Siblings versus none							(0.029)	(0.031)	(0.028)	(0.032)
3 Siblings versus none							0.851***	0.924*	0.782***	0.867***
3 Siblings versus none							(0.027)	(0.032)	(0.028)	(0.033)
4 Siblings versus none							0.765***	0.896**	0.737***	0.878**
4 Siblings versus none							(0.027)	(0.034)	(0.031)	(0.039)
5+ Siblings versus none							0.602***	0.827***	0.663***	0.837***
5+ Siblings versus none							(0.020)	(0.029)	(0.025)	(0.035)
Male		1.004		1.006		1.004		1.003		1.006
Male		(0.020)		(0.020)		(0.020)		(0.020)		(0.020)
Birth cohort (1919 and before omitted)
1920–1929		1.456		1.482		1.462		1.452		1.483
1920–1929		(0.306)		(0.312)		(0.307)		(0.305)		(0.312)
1930–1939		2.259***		2.322***		2.268***		2.250***		2.329***
1930–1939		(0.466)		(0.479)		(0.467)		(0.464)		(0.480)
1940–1949		3.692***		3.824***		3.708***		3.677***		3.842***
1940–1949		(0.761)		(0.789)		(0.764)		(0.758)		(0.793)
1950–1959		5.222***		5.429***		5.238***		5.205***		5.460***
1950–1959		(1.080)		(1.123)		(1.082)		(1.076)		(1.130)
1960–1969		5.210***		5.438***		5.242***		5.193***		5.467***
1960–1969		(1.085)		(1.134)		(1.092)		(1.082)		(1.141)
1970–1995		6.527***		6.840***		6.584***		6.507***		6.871***
1970–1995		(1.538)		(1.613)		(1.551)		(1.533)		(1.621)
First marriage age		0.990***		0.990***		0.990***		0.990***		0.990***
First marriage age		(0.002)		(0.002)		(0.002)		(0.002)		(0.002)
Biological parents		0.753***		0.745***		0.745***		0.754***		0.746***
Biological parents		(0.019)		(0.019)		(0.019)		(0.020)		(0.019)
Urban		1.304***		1.307***		1.308***		1.305***		1.306***
Urban		(0.026)		(0.026)		(0.026)		(0.026)		(0.026)
Educational attainment		1.040***		1.042***		1.045***		1.041***		1.042***
Educational attainment		(0.008)		(0.008)		(0.008)		(0.008)		(0.008)
Mother’s Ed. attainment		1.071***		1.072***		1.075***		1.072***		1.072***
Mother’s Ed. attainment		(0.010)		(0.010)		(0.010)		(0.010)		(0.010)
Household income		0.996		0.996		0.997		0.996		0.996
Household income		(0.005)		(0.005)		(0.005)		(0.005)		(0.005)
Homeowner		0.589***		0.589***		0.590***		0.590***		0.589***
Homeowner		(0.014)		(0.014)		(0.014)		(0.014)		(0.014)
Any Kids		0.671***		0.668***		0.667***		0.670***		0.668***
Any Kids		(0.022)		(0.022)		(0.022)		(0.022)		(0.022)
Religiosity		0.868***		0.864***		0.861***		0.867***		0.865***
Religiosity		(0.014)		(0.014)		(0.013)		(0.014)		(0.014)
Religious attendance		0.797***		0.795***		0.794***		0.797***		0.796***
Religious attendance		(0.030)		(0.030)		(0.030)		(0.030)		(0.030)
Eastern Europe		0.565***		0.565***		0.570***		0.565***		0.564***
Eastern Europe		(0.016)		(0.016)		(0.016)		(0.016)		(0.016)
Southern Europe		0.331***		0.331***		0.333***		0.331***		0.331***
Southern Europe		(0.009)		(0.009)		(0.009)		(0.009)		(0.009)
Northern Europe		1.212***		1.214***		1.221***		1.211***		1.215***
Northern Europe		(0.030)		(0.030)		(0.030)		(0.030)		(0.030)
Year of first survey		1.007**		1.006**		1.007**		1.007**		1.006**
Year of first survey		(0.003)		(0.003)		(0.003)		(0.003)		(0.003)
Constant	0.311***	0.000**	0.263***	0.000**	0.283***	0.000**	0.278***	0.000**	0.280***	0.000**
Constant	(0.005)	(0.000)	(0.004)	(0.000)	(0.008)	(0.000)	(0.007)	(0.000)	(0.007)	(0.000)
N	86,014	86,014	86,014	86,014	86,014	86,014	86,014	86,014	86,014	86,014

*p < .05. **p < .01. ***p < .001.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

Appendix D

Table D1.

Logistic Regression (Coefficients Only) of Current Divorce on Select Sibling Metrics—Separate Models for each Region and Country in Europe, Controlling for All Covariates in Table 4 (SHARE data).^a,b

Location	Sample Size	Siblings (Linear)	Siblings Alive (Linear)	Any Siblings
Western Europe	24,080	0.994	0.995	0.940
Northern Europe	18,350	1.000	1.020	1.074
Southern Europe	27,959	0.945***	0.972*	0.784**
Eastern Europe	15,625	0.947**	0.953*	0.813*
Austria	3379	0.970	0.965	0.977
Germany	4611	0.995	1.002	0.817
Sweden	3774	1.076*	1.048	1.493*
Netherlands	2071	0.998	1.014	0.655
Spain	5248	1.043	0.999	0.867
Italy	5205	0.934*	0.948	0.856
France	4249	0.995	0.973	1.171
Denmark	3817	1.016	1.035	1.072
Greece	4004	0.901*	0.971	0.698*
Switzerland	2743	0.966	0.951	1.042
Belgium	5818	1.026	1.034	1.000
Israel	2076	1.008	0.985	1.285
Czechia	4912	0.972	0.967	0.870
Poland	5174	0.918*	0.938	0.600*
Ireland	618	0.937	1.028	0.771
Luxembourg	1209	1.005	1.007	1.350
Hungary	1493	0.981	0.998	1.032
Portugal	1240	0.982	1.016	0.826
Slovenia	3542	0.939	0.991	0.900
Estonia	4702	1.035	1.072	1.035
Croatia	2327	0.936	1.031	0.937
Lithuania	1960	0.953	0.944	0.968
Bulgaria	1928	0.863*	0.869	0.790
Cyprus	1213	0.808*	0.754**	0.154**
Finland	1839	0.997	1.009	1.308
Latvia	1640	0.951	1.006	0.994
Malta	1176	0.873*	0.917	0.658
Romania	2069	0.998	1.032	1.038
Slovakia	1977	1.007	0.986	0.836
Overall: locations with a negative association		24 out of 33 (72%)	19 out of 33 (57%)	21 out of 33 (64%)

*< .05. **p < .01. ***p < .001.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

Appendix E

Table E1.

Logistic Regression Models (Sibling Effects Only) Predicting Current Divorce in SHARE (Controlling for All Covariates in Table 4 and Country of Residence in lieu of Region).^a,b,c

Variables	Sibs (Linear)		Sibs Alive (Linear)		Any Sibs		Sibsize		Sibsize Alive
Variables	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model
Siblings (linear)	0.931***	0.985*
Siblings (linear)	(0.006)	(0.006)
Siblings alive (linear)			0.971***	0.998
Siblings alive (linear)			(0.006)	(0.007)
Any Siblings					0.834***	0.944
Any Siblings					(0.033)	(0.034)
1 Sibling versus none							1.048	1.011	0.867***	0.949
1 Sibling versus none							(0.063)	(0.063)	(0.035)	(0.039)
2 Siblings versus none							0.956	0.952	0.900*	0.976
2 Siblings versus none							(0.039)	(0.040)	(0.038)	(0.044)
3 Siblings versus none							0.878**	0.962	0.810***	0.922
3 Siblings versus none							(0.037)	(0.043)	(0.039)	(0.047)
4 Siblings versus none							0.795***	0.935	0.840**	1.017
4 Siblings versus none							(0.038)	(0.047)	(0.046)	(0.059)
5+ Siblings versus none							0.670***	0.908*	0.756***	0.939
5+ Siblings versus none							(0.030)	(0.044)	(0.038)	(0.052)

*p < .05. **p < .01. ***p < .001.

N = 86,014.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

Appendix F

Table F1.

Logistic Regression Models (Sibling Effects Only) Predicting Ever Divorced in SHARE (Controlling for all Covariates in Table 4 and Country of Residence in lieu of Region).^a,b,c

Variables	Sibs (Linear)		Sibs Alive (Linear)		Any Sibs		Sibsize		Sibsize Alive
Variables	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model	Empty	Full Model
Siblings (linear)	0.911***	0.978***
Siblings (linear)	(0.004)	(0.005)
Siblings alive (linear)			0.949***	0.985**
Siblings alive (linear)			(0.005)	(0.005)
Any Siblings					0.810***	0.937*
Any Siblings					(0.023)	(0.029)
1 Sibling versus none							1.133**	1.065	0.865***	0.961
1 Sibling versus none							(0.050)	(0.050)	(0.026)	(0.031)
2 Siblings versus none							0.957	0.949	0.881***	0.959
2 Siblings versus none							(0.029)	(0.031)	(0.028)	(0.033)
3 Siblings versus none							0.851***	0.941	0.782***	0.897**
3 Siblings versus none							(0.027)	(0.033)	(0.028)	(0.035)
4 Siblings versus none							0.765***	0.929*	0.737***	0.916*
4 Siblings versus none							(0.027)	(0.035)	(0.031)	(0.042)
5+ Siblings versus none							0.602***	0.877***	0.663***	0.880**
5+ Siblings versus none							(0.020)	(0.032)	(0.025)	(0.037)

*p < .05. **p < .01. ***p < .001.

N = 86,014.

^aModels are estimated from imputed datasets.

^bCoefficients reported as odds-ratios.

^cStandard errors are in parenthesis.

References

Alexander

K. L.

Entwisle

D. R.

Olson

L. S.

(2014). The long shadow: Family background, disadvantaged urban youth, and the transition to adulthood. Russell Sage Foundation.

Angrist

Lavy

Schlosser

(2010). Multiple experiments for the causal link between the quantity and quality of children. Journal of Labor Economics, 28(4), 773–824. https://doi.org/10.1086/653830

Blaabæk

Ea H.

Jæger

M. M.

Molitoris

(2020). Family size and educational attainment: cousins, contexts, and compensation. European Journal of Population, 36(3), 575–600. https://doi.org/10.1007/s10680-019-09543-y

Blake

(1981). The only child in america: prejudice versus performance. Population and Development Review, 7(1), 43–54. https://doi.org/10.2307/1972763

Blake

Judith.

1989. Family Size and Achievement. Vol. 3. Berkeley: University of California Press.

Blake

Richardson

Bhattacharya

(1991). Number of siblings and sociability. Journal of Marriage and the Family, 53(2), 271–283. https://doi.org/10.2307/352898

Bobbitt-Zeher

Downey

D. B.

(2013). Number of siblings and friendship nominations among adolescents. Journal of Family Issues, 34(9), 1175–1193. https://doi.org/10.1177/0192513x12470370

Bobbitt-Zeher

Downey

D. B.

Joseph

(2014). Number of siblings during childhood and the likelihood of divorce in adulthood. Journal of Family Issues, 37(15).

Brody

Gene.

(2004). Siblings’ Direct and Indirect Contributions to Child Development. Current Directions in Psychological Science, 13(3), 124–26.

10.

Börsch-Supan

Hendrik

Kirsten

H. A.

Grant

Agar

Dimitrios

Enrica

, … 2005. “The Survey of Health, Aging, and Retirement in Europe-Methodology.”

11.

Cameron

Erkal

Gangadharan

Meng

(2013). Little emperors: Behavioral impacts of China’s one-child policy. Science (New York, N.Y.), 339(6122), 953–957. https://doi.org/10.1126/science.1230221

12.

Conley

Glauber

(2006). Parental educational investment and children’s academic risk: estimates of the impact of sibship size and birth order from exogenous variation in fertility. Journal of Human Resources, 41(4), 722–737. https://doi.org/10.3368/jhr.xli.4.722

13.

Conway

(2016). Where there’s a will....: Law and emotion in sibling inheritance disputes. In The Emotional Dynamics of Law and Legal Discourse (pp. 35–57). London: Bloomsbury Publishing.

14.

de Vuijst

Poortman

A. R.

Das

van Gaalen

(2017). Cross-sibling effects on divorce in the Netherlands. Advances in Life Course Research, 34, 1–9. https://doi.org/10.1016/j.alcr.2017.06.003

15.

Deater-Deckard

Dunn

Lussier

(2002). Sibling relationships and social-emotional adjustment in different family contexts. Social Development, 11(4), 571–590. https://doi.org/10.1111/1467-9507.00216

16.

Development, Organization for Economic Cooperation and, and OECD Family Database . 2021. Fertility Rates.

17.

Downey

Douglas B.

(1995). When Bigger Is Not Better: Family Size, Parental Resources, and Children’s Educational Performance. American Sociological Review, 60(5), 746–61.

18.

Downey

Douglas B.

(2001). Number of Siblings and Intellectual Development: The Resource Dilution Explanation. The American Psychologist, 56(6–7).

19.

Downey

D. B.

Condron

D. J.

(2004). Playing well with others in kindergarten: The benefit of siblings at home. Journal of Marriage and Family, 66(2), 333–350. https://doi.org/10.1111/j.1741-3737.2004.00024.x

20.

Dunn

Munn

(1985). Becoming a family member: Family conflict and the development of social understanding in the second year. Child Development, 56(2, Family Development and the Child), 480–492.

21.

EUROSTAT . 2018. Eurostat.

22.

Freese

Peterson

(2017). Replication in social science. Annual Review of Sociology, 43(1), 147–165. https://doi.org/10.1146/annurev-soc-060116-053450

23.

Gibbs

B. G.

Workman

Downey

D. B.

(2016). The (Conditional) resource dilution model: State- and community-level modifications. Demography, 53(3), 723–748. https://doi.org/10.1007/s13524-016-0471-0

24.

Goodkind

(2017). The astonishing population averted by China’s birth restrictions: Estimates, nightmares, and reprogrammed ambitions. Demography, 54(4), 1375–1400. https://doi.org/10.1007/s13524-017-0595-x

25.

Guo

(1998). The timing of the influences of cumulative poverty on children’s cognitive ability and achievement. Social Forces, 77(1), 257. https://doi.org/10.2307/3006017

26.

Härkönen

(2015). Divorce. Emerging Trends in the Social and Behavioral Sciences, 1–14.

27.

Heckman

J. J.

Masterov

D. V.

(2007). The productivity argument for investing in young children. Review of Agricultural Economics, 29(3), 446–493. https://doi.org/10.1111/j.1467-9353.2007.00359.x

28.

Jenkins

Dunn

(2009). Siblings within families: Levels of analysis and patterns of influence. (pp. 79–93 Vol. 126). Jossey-Bass.

29.

Kramer

(2010). The essential ingredients of successful sibling relationships: An emerging framework for advancing theory and practice. Child Development Perspectives, 4(2), 80–86. https://doi.org/10.1111/j.1750-8606.2010.00122.x

30.

Merry

Bobbitt-Zeher

Downey

D. B.

(2020). Number of siblings in childhood, social outcomes in adulthood. Journal of Family Issues, 41(2), 212–234. https://doi.org/10.1177/0192513x19873356

31.

Platte

(1988). Divorce trends and patterns in China: Past and present. Pacific Affairs, 61(3), 428–445. https://doi.org/10.2307/2760459

32.

Polit

D. F.

Falbo

(1987). Only children and personality development: A quantitative review. Journal of Marriage and the Family, 49(2), 309–325. https://doi.org/10.2307/352302

33.

Powell

Steelman

L. C.

(1990). Beyond sibship size: Sibling density, sex composition, and educational outcomes. Social Forces, 69(1), 181–206. https://doi.org/10.2307/2579613

34.

Powell

Steelman

L. C.

(1993). The educational benefits of being spaced out: Sibship density and educational progress. American Sociological Review, 58(3), 367–381. https://doi.org/10.2307/2095906

35.

Qian

(2020). Separating spheres: Cohort differences in gender attitudes about work and family in China. China Review, 20(2), 19–51.

36.

Rijken

Liefbroer

A. C.

(2012). European views of divorce among parents of young children: Understanding cross-national variation. Demographic Research, 27(2), 25–52. https://doi.org/10.4054/demres.2012.27.2

37.

Scheel

H. L.

Sören

Rune

H. L.

Jeune

Linda Juel

(2019). Cross-National Comparison of Sex Differences in ADL and IADL in Europe: Findings from SHARE. European Journal of Ageing, 17(1), 69–79.

38.

Simmel

(1902). The number of members as determining the sociological form of the group. I. The American Journal of Sociology, 8(1), 1–46. https://doi.org/10.1086/211115

39.

Steelman

L. C.

Powell

Werum

Carter

(2002). Reconsidering the effects of sibling configuration: Recent advances and challenges. Annual Review of Sociology, 28(1), 243–269. https://doi.org/10.1146/annurev.soc.28.111301.093304

40.

Stevenson

Wolfers

(2007). Marriage and divorce: Changes and their driving forces. Journal of Economic Perspectives, 21(2), 27–52. https://doi.org/10.1257/jep.21.2.27

41.

Sun

Zhao

(2016). Divorce, abortion, and the child sex ratio: The impact of divorce reform in China. Journal of Development Economics, 120, 53–69. https://doi.org/10.1016/j.jdeveco.2015.11.006

42.

Vogl

T. S.

Freese

(2020). Differential fertility makes society more conservative on family values. Proceedings of the National Academy of Sciences of the United States of America, 117(14), 7696–7701. https://doi.org/10.1073/pnas.1918006117

43.

von Hippel Paul

(2007). Regression with Missing Ys: An Improved Strategy for Analyzing Multiply Imputed Data. Sociological Methodology, 37(1), 83–117.

44.

Wagner

Weiss

(2006). On the variation of divorce risks in Europe: Findings from a meta-analysis of European longitudinal studies. European Sociological Review, 22(5), 483–500. https://doi.org/10.1093/esr/jcl014

45.

Walkden

G. J.

Anderson

E. L.

Vink

M. P.

Tilling

Howe

L. D.

Ben-Shlomo

2018. “Frailty in Older-Age European Migrants: Cross-Sectional and Longitudinal Analyses of the Survey of Health, Aging and Retirement in Europe (SHARE).” Social Science & Medicine ( 1982 ) 213:1–11.

46.

Wang

Cai

Shen

Ke.

Gietel-Basten

(2018). Is demography just a numerical exercise? Numbers, politics, and legacies of China’s one-child policy. Demography, 55(2), 693–719. https://doi.org/10.1007/s13524-018-0658-7

47.

Wang

C. T.

Schofer

Evan

(2018). Coming Out of the Penumbras: World Culture and Cross-National Variation in Divorce Rates. Social Forces, 97(2), 675–704.

48.

(2019). Inequality and Social Stratification in Postsocialist China. Annual Review of Sociology, 45(1), 363–382. https://doi.org/10.1146/annurev-soc-073018-022516

49.

Qiu

(2016). Is the ‘Seven-Year Itch’ Real?—A study on the changing divorce pattern in Chinese marriages. The Journal of Chinese Sociology, 3(1), 17–23. https://doi.org/10.1186/s40711-016-0038-x

50.

Qiu

(2015). The impact of children on divorce risk. The Journal of Chinese Sociology, 2(1), 1. https://doi.org/10.1186/s40711-015-0003-0

51.

Lin

(1998). Analysis of trend and cause of divorce in contemporary China. Population and Economics(3), 22–28.

52.

Yucel

Downey

D. B.

(2015). When quality trumps quantity: siblings and the development of peer relationships. Child Indicators Research, 8(4), 845–865. https://doi.org/10.1007/s12187-014-9276-0