1 Introduction

The recent decades of family change—including the increases in divorce and separation rates, single parenthood, cohabitation, and step family formation—led to an explosion in popular and academic interest in the consequences of family dynamics for children’s well-being and life chances (cf. Amato 2000, 2010; Amato and James 2010; Ribar 2004; Sweeney 2010; McLanahan et al. 2013). Most notably, previous studies have found that children who do not live with both biological parents fare somewhat worse than those who do in terms of psychological well-being, health, schooling, and later labor market attainment, and differ with respect to their own family lives in adulthood. Scholars have interpreted these findings through a relatively small group of factors that include parental and children’s stress associated with family transitions, family conflict, changes in economic resources, and parenting styles. Beyond these established findings, however, several questions remain imperfectly answered.

This Special Issue on “Family Dynamics and Children’s Well-Being and Life Chances in Europe” consists of this introductory article and four empirical studies that address some of these open questions. In general, they give more nuance to the overall association between growing up with both biological parents and child outcomes. More precisely, do these associations differ according to the type of family structure studied? Are these differences in child outcomes due to causal effects of family structures and transitions, or do they reflect preexisting disadvantages between families? And finally, are all children equally affected by family structures and transitions?

In this introduction, we first introduce the theme of family dynamics and children’s outcomes by giving an overview of the findings of parental separation and child outcomes (Sect. 2). Parental separation has been the family transition that has attracted most attention among social scientists, and many of our examples later in the article consider this research too. In addition to summarizing the evidence on the relationship between parental separation and psychological well-being, education, social relationships, and own family lives, we discuss how parental separations have been conceptualized, an issue we return to in the subsequent sections.

Parental separation is, however, just one of the family transitions children can experience during their childhoods. The first open question that in our view requires more attention regards the effects of these other family transitions and forms, namely the number of transitions, stepfamilies, and joint residential custody after parental separation (Sect. 3). Two of the articles in this Special Issue contribute to this stream of research. Mariani et al. (2017) present the first European analysis of the effects of family trajectories on children born to lone mothers. Radl et al. (2017) investigate, in addition to parental separation effects, whether co-residing with siblings or grandparents is related to child outcomes and whether the latter condition the former effects.

The second open question concerns the causal status of the estimated effects (Sect. 4): Do family structures and their changes really affect child outcomes, or do the associations reflect some unmeasured underlying factors? This question has attracted deserved attention (e.g., Amato 2000; Ribar 2004; McLanahan et al. 2013), and we review some commonly used methods, using the effects of parental separation as our example. We pay attention to what effects the methods can estimate, in addition to assessing which unobserved variables the different methods adjust for. This discussion highlights the importance of thinking about methodological choices and interpretations of the results in light of the underlying theoretical model of parental separation. The article in this Special Issue by Bernardi and Boertien (2017) provides also an empirical contribution to this field.

Finally, the last question refers to the heterogeneity in the effects of family dynamics: Are the consequences of parental separation and other family transitions similar for all children? Existing evidence suggests that the answer is no (Amato 2000; Demo and Fine 2010), but the conclusions about who suffers and who does not remain imperfect, as discussed in Sect. 5. Three of the articles of this Special Issue analyze these questions, one from a cross-national perspective (Radl et al. 2017), one by comparing parental separation effects by socioeconomic background (Bernardi and Boertien 2017), and one by immigrant background (Erman and Härkönen 2017).

In the final section of this introduction (Sect. 6), we discuss some ways forward for future research on family dynamics and children’s outcomes. Two articles in this Special Issue fulfill part of this research agenda by providing evidence on how several less often studied family forms relate to child outcomes in the European context (Mariani et al. 2017; Radl et al. 2017). The two other articles in this Special Issue (Bernardi and Boertien 2017; Erman and Härkönen 2017) contribute to the research on heterogeneous consequences of parental separation by clarifying some open questions regarding variation in these consequences by socioeconomic and immigrant background.

2 Parental Separation and Children’s Outcomes

In the 2000s, the share of children who experienced their parents’ separation before age 15 ranged from 10 to 12% in countries such as Bulgaria, Georgia, Italy, and Spain to 35–42% in France, Estonia, Lithuania, and Russia (Andersson et al., forthcoming). In the late 1980s/early 1990s, the corresponding figures ranged from 7 to 30% (Italy and Sweden, respectively, Andersson and Philipov 2002).

Parental separation changes children’s lives in many ways. Many scholars conceptualize separations as processes, which often begin way before and last well beyond the actual separation (e.g., Amato 2000; Demo and Fine 2010; Härkönen 2014), even if these starting and ending points can be hard to define. The pre-separation process often involves increasing estrangement and conflict between the parents. These can themselves have negative effects on children’s well-being, and parental separation might therefore already start leaving its traces even before the parents have formally broken up. Not all separations follow such a trajectory. Some families may have had long-lasting conflicts, and other separations might have ended relatively well-functioning partnerships with at least moderate levels of satisfaction (Amato and Hohmann-Marriott 2007). The parental separation can in such cases come as an unexpected event for children.

As a result of the separation, children cease to live full-time with both parents, which requires adjustment to the new situation and can start, intensify, or end exposure to parental conflict (Amato 2010; Cherlin 1999; Pryor and Rodgers 2001). Even if joint residential custody of the child post-separation (i.e., children’s alternate living with each parent) is becoming increasingly common, up to one-third and above in Sweden (Bergström et al. 2015), the child often receives less involved parenting from the nonresident parent (usually the father), whereas the resident parent’s (usually the mother’s) parenting styles can be affected by increasing time demands (Amato 2000, 2010; McLanahan and Sandefur 1994; Seltzer 2000). Besides changes in family relationships, a breakup of a household can lead to a drop in economic resources (e.g., Uunk 2004). Depending on the country, separated parents may need to adjust their labor supply to meet their new time and economic demands (Kalmijn et al. 2007; Uunk 2004). Many children also need to move after their parents’ separation, which requires adjustment to a new home environment and possibly a new neighborhood and school. A separation can be followed by further changes in the family structure, such as parental re-partnering, entry of step-siblings, and sometimes, another family dissolution.

Several studies have documented that on average, the lives of children whose parents separated differ from children who lived with both of their parents throughout childhood (Amato 2000, 2010; McLanahan and Sandefur 1994; McLanahan et al. 2013; Härkönen 2014). In the next paragraphs, we provide an overview of the associations of parental separation with some of the most commonly studied child outcomes: psychological well-being and behavioral problems, education, social relationships, and own family lives. In the subsequent sections, we will refine this basic picture by concentrating on other family forms, causality, and heterogeneity in effects.

2.1 Psychological Well-Being and Behavioral Problems

Children of divorce have lower psychological well-being and more behavioral problems than children who grew up in intact families (Amato 2001; Amato and James 2010; Gähler and Palmtag 2015; Kiernan and Mensah 2009; Mandemakers and Kalmijn 2014). In general, parental separation is more strongly related to externalizing than internalizing problems (Amato 2001), and these associations can persist, and even become stronger, into adulthood (Chase-Lansdale et al. 1995; Cherlin et al. 1991; Lansford 2009).

Growing up in a conflict-ridden but stable family can have more negative effects on children’s psychological well-being than parental separation (e.g., Amato et al. 1995; Dronkers 1999; Hanson 1999; Demo and Fine 2010). Kiernan and Mensah (2009) found a role for both maternal depression and economic resources when explaining the lower emotional well-being of children from separated families, whereas Turunen (2013) found that parental involvement explained part of the lower emotional well-being of children with separated parents, but economic resources did not.

2.2 Education

Children of divorce have lower school grades and test scores (Dronkers 1992; Mandemakers and Kalmijn 2014; Grätz 2015), have lower school engagement (Havermans et al. 2014), differ in the kind of track entered in high school (Dronkers 1992; Jonsson and Gähler 1997; Grätz 2015), and have lower final educational attainment (Bernardi and Radl 2014; Bernardi and Boertien 2016a; Gähler and Palmtag 2015).

Lower school grades and cognitive performance explain part, but not all of the effect of parental separation on completed education (Dronkers 1992). A recent study found that British children of divorce were less likely to continue to full-time upper secondary education even though the parental separation did not affect their school grades (Bernardi and Boertien 2016a). Parental separation can therefore affect the children’s educational decisions irrespective of their school performance.

Changes in parental resources are an important explanation for the lower educational performance of the children of divorce (Bernardi and Boertien 2016a; Jonsson and Gähler 1997; McLanahan and Sandefur 1994; Thomson et al. 1994). Studies that have looked into the role of parenting have found differing results, some reporting that parenting partly mediates the effect of separation on educational attainment, while others found parenting to not influence the relationship between parental divorce and school outcomes (Dronkers 1992).

2.3 Social Relationships

Despite the increase in shared residential custody (Bjarnason and Arnarsson 2011), parental separation generally reduces the child’s contact frequency and relationship quality with the nonresident parent (usually the father), with grandparents and, sometimes, the mother (e.g., Kalmijn 2012; Kalmijn and Dronkers 2015; Lansford 2009). These effects can last into adulthood (Albertini and Garriga 2011; Kalmijn 2012). Joint residential custody, good inter-parental relations, and good early child-father relations can improve post-separation contact with the father (Kalmijn 2015; Kalmijn and Dronkers 2015). On the other hand, parental separation can improve the relationships between siblings due to mutual support (Geser 2001), but does not seem to trigger more support from friends and other kin (Kalmijn and Dronkers 2015).

Good parent–child relationships are desirable by themselves and can also improve other child outcomes (Bastaits et al. 2012; Swiss and Le Bourdais 2009). For example, having a close relationship with the nonresident parent who engages in authoritative parenting has been found to foster children’s well-being and academic success (Amato and Gilbreth 1999). At the same time, contact frequency alone is less important and in some cases, the nonresident parent’s involvement may have negative effects if it increases instability and stress for the child (Laumann-Billings and Emery 2000), for example due to continued parental conflict (Kalil et al. 2011).

2.4 Own Family Lives

Children of divorce tend to start dating and have their sexual initiation earlier (Wolfinger 2005) and many move out of the parental home at a younger age (e.g., Ní Bhrolcháin et al. 2000; Ongaro and Mazzuco 2009), often because of conflict with parents and their potential new partners (Wolfinger 2005). Some studies have also found that children of divorce start cohabiting earlier, are more likely to cohabit than to marry, and have partners of lower socioeconomic status (Erola et al. 2012; Reneflot 2009; but see also Ní Bhrolcháin et al. 2000).

The most consistent family demographic finding is that children whose parents divorced are more likely to divorce themselves as adults (e.g., Diekmann and Engelhardt 1999; Dronkers and Härkönen 2008; Kiernan and Cherlin 1999; Lyngstad and Engelhardt 2009; Wolfinger 2005). Differences in the life course trajectories before forming the union explain part of this association (Diekmann and Engelhardt 1999; Kiernan and Cherlin 1999). Other studies have pointed out that parental separation can lead to poorer interpersonal skills and set an example of a feasible solution to relationship problems (Wolfinger 2005).

3 What About Other Family Forms?

We have so far focused on parental separation and its relation to child outcomes. Parental separation is not the only family transition children can experience. Between <5% (much of Europe) and up to 15% (Czech Republic, Russia, UK, and USA) of children are born to lone mothers (Andersson et al., forthcoming; Mariani et al. 2017, this Special Issue). Furthermore, between 14% (Italy and Georgia) and 60% (Belgium) of European children whose parents separate end up living with a stepparent within 6 years (Andersson et al., forthcoming) and often, with step-siblings (Halpern-Meekin and Tach 2008). Children’s residence arrangements likewise vary, with some residing primarily with one parent (usually the mother), whereas others alternate between parents (joint residential custody). Extending the focus of research beyond parental separation is necessary to form a more comprehensive view of the effects of the changing family landscape on children’s lives (King 2009; Sweeney 2010).Footnote 1

One argument puts forward that family stability rather than family structure matters for children’s well-being (cf. Fomby and Cherlin 2007; Waldfogel et al. 2010). From this perspective, children born to lone mothers who do not experience any family transitions during their childhood (such as the entrance of a stepparent) should do better than children who were born in a two-parent family but experienced a family transition (such as parental separation). Others claim that specific family forms and movements between them do matter beyond general family instability (Magnuson and Berger 2009; Lee and McLanahan 2015). The findings of Mariani et al. (2017, this Special Issue) are among those that speak against the general instability thesis and show that the types of family transitions experienced by children born to lone mothers matter for their well-being.

Stepfamilies have gained the attention of many scholars. Children in stepfamilies tend to have poorer outcomes compared to those from intact families and display patterns of well-being closer to single-parent families (Amato 1994, 2001; Gennetian 2005; Jonsson and Gähler 1997; Thomson et al. 1994). Indeed, children in stepfamilies can even have lower psychological well-being and educational achievement than children living with a single mother (Amato 1994; Biblarz and Raftery 1999; Thomson et al. 1994).

Reasons for the poorer performance of children with stepparents include the added complexity in family relationships that is often introduced by the presence of a stepparent. This can lead to ambiguity in roles and to conflict in the family (Thomson et al. 1994; Sweeney 2010), which is among the reasons why having a stepparent often leads to an earlier move from the parental home, especially among girls (Ní Bhrolcháin et al. 2000; Reneflot 2009). Another explanation points to the presence of step-siblings as stepparents may put less time and effort into their stepchildren than their biological ones (Biblarz and Raftery 1999; Evenhouse and Reilly 2004). However, having a stepparent can also have positive effects as (s)he can provide financial resources or help in monitoring the children (Thomson et al. 1994; King 2006; Sweeney 2010). Erola and Jalovaara (2016) showed how a stepparent’s SES was more predictive on adulthood SES than the nonresident father’s SES, and as predictive as the biological father’s SES in intact families. All in all, the effects of step-parenthood are complex and can differ between children who experienced a parental separation and those who never lived with their biological father (Sweeney 2010).

The increase in joint residential custody after parental separation has raised interest in its consequences for children. Many studies have reported that children in joint residential custody fare better than children who reside with only one of the parents (usually the mother) on outcomes such as health and psychological well-being, and contact and relationships with their parents and grandparents (Bjarnason and Arnarsson 2011; Turunen 2016; Westphal et al. 2015). However, questions of causality remain unresolved and parents who opt for joint custody might have been particularly selected from those with higher socioeconomic status and lower levels of post-separation conflict. Indeed, many studies find that joint custody may have negative consequences for children in case of high parental conflict (e.g., Vanassche et al. 2014; also, Kalil et al. 2011). This suggests that policy changes toward joint custody as a default solution may produce unwanted consequences.

4 But What About Causality?

There is a long-standing debate that concerns whether associations between family types and child outcomes reflect causal effects, or whether they are confounded by unmeasured variables. For example, parents who separate can have different (unmeasured) personality traits from those who do not. Other examples include parental unemployment, mental health, or a developing substance abuse problem, which may not only lead to separation, but also affect the parent’s children.

Researchers have used increasingly sophisticated methods to control for different unmeasured sources of bias (for reviews, Amato 2000, 2010; Ribar 2004; McLanahan et al. 2013). In this section, we discuss some of these methods. We focus on studies that have estimated the effects of parental separation, which serves to illustrate some of the questions involved.

Like most similar reviews, we discuss which (un)measured confounders can be controlled for by the different methods and provide examples of studies that have used them. We also discuss some of the limitations to causal inference in these methods, particularly in light of the underlying theoretical model of parental separation that is assumed. Above, we discussed how parental separations are often theorized as processes that can follow quite different trajectories for different families (Amato 2000; Demo and Fine 2010; Härkönen 2014). Some separations are characterized by a downward spiral of increasing conflict, which can leave its mark on children already before the parents physically separate. Other separations end relatively well-functioning families and can come as a surprise to the children, whereas in some cases the families had high conflict levels for a long time. In this section, we discuss causal inference in light of these underlying models. In the next section, we discuss how these different types of parental separations can have different effects on children.

In addition, we engage in a related but much smaller discussion of what causal questions the different methods can be used to answer (cf., Manski et al. 1992; Ní Bhrolcháin 2001; Sigle-Rushton et al. 2014). A major issue in this regard concerns the counterfactual scenario assumed by different methods. In most studies, the estimated effects are interpreted as telling about how the parents’ physical separation (the separation event) affected the children compared to the counterfactual case in which the parents did not separate. This is, however, not the only possible effect that can be estimated, nor is this interpretation necessarily the correct one in each case.

First, knowing about the effects of the parental separation event is obviously important, but scholars, parents, counselors, and policy makers could likewise benefit from knowing about the “total” effects of parental separation that include the effects of the preceding separation process as well. Second, instead of asking what the effect of the parental separation (compared to them staying together) is, one can ask what the effect is of the parents separating at a specific point in time (the effect of postponing separation) (cf. Furstenberg and Kiernan 2001). Our discussion below points to these issues and suggests how some methods can be more appropriate for answering certain questions than others. Rather than providing a comprehensive discussion on this relatively uncovered topic, we wish to stimulate closer consideration of these issues in future research.

4.1 Regression Models

Before discussing methods that adjust for unmeasured confounding factors, we briefly discuss estimation of parental separation effects with linear and logistic (or similar) regression models, which are by far the most common methods used. With these methods, one compares the outcomes of children who experienced parental separation to the outcomes of children from intact families, adjusting for observed confounding variables. Because the possibilities for controlling for all factors that may bias the results are limited, the estimates from regression models cannot usually be interpreted as causal effects (e.g., McLanahan et al. 2013; Ribar 2004).

Pre-separation parental conflict is often pointed out as an omitted variable that can threaten causal claims. Controlling for pre-separation conflict generally leads to a substantial reduction in the effect of parental separation (e.g., Hanson 1999; Gähler and Garriga 2013), suggesting that exposure to the parental conflict rather than the parental separation event is largely responsible for the poorer performance of the children of divorce. This example can be used to think about the correspondence between the specified regression model and the underlying theoretical model of parental separation. Controlling for the level of pre-separation parental conflict (or related measures of the family environment) is most appropriate if it is reasonable to assume that families’ conflict levels remain stable; comparing children from separated and intact families at similar levels of earlier conflict can then inform about how the children of divorce would have fared had the parents remained together. However, this is not obvious if the separation followed an increase in parental conflict, because the family environment may have continued to worsen had the parents not separated.

If the above and other conditions for making causal claims are met, which effects do they inform us about? A regression model that controls for pre-separation parental conflict or other related measures is best seen as telling about the effects of the parental separation event. However, an increase in parental conflict is often an inherent part of the parental separation process, and controlling for levels of parental conflict close to the parental separation would not be warranted if one is interested in understanding how exposure to the parental separation process, in addition to the separation event, affects children’s outcomes (cf. Amato 2000). The choice of control variables should thus be done with a consideration to the underlying model of parental separation and the effect one wants to estimate.

4.2 Sibling Fixed Effects

Sibling fixed effects (SFE) models compare siblings from the same family who differ in their experience of parental separation before a certain age or life stage, or in the amount of time spent in a specific family type (cf. McLanahan et al. 2013; Sigle-Rushton et al. 2014). SFE controls for factors and experiences that are shared by the siblings, such as parental SES and many neighborhood and school characteristics. This has made SFE a popular method, not least in Europe. Some SFE studies found no effects of parental separation or other family forms on educational outcomes (Björklund and Sundström 2006). Others have found a weak to moderate negative effect on various outcomes even in an SFE design (e.g., Ermisch et al. 2004; Sandefur and Wells 1999; Sigle-Rushton et al. 2014; Grätz 2015).

Comparison of siblings from the same family is a core aspect of the SFE design. This affects the data requirements and the interpretation of the results. To fix ideas, we can use an example of the effects of parental separation on children’s school grades at age 15. For an SFE analysis, one needs data on multiple siblings, some of whom experienced the parental separation before age 15 whereas others did not. This requirement reduces the effective sample size. The sibling who did not experience the parental separation is always the older one, and her grades are used to infer about the counterfactual grades of her younger sibling, had she not experienced the parental separation. SFE controls for everything shared by the siblings, but additional controls are needed to adjust for differences between them. Some of these—such as birth order and birth cohort and/or parental age (Sigle-Rushton et al. 2014)—are available in many datasets, but remaining unobserved differences (as well as measurement error) can cause important bias to the estimates (Ermisch et al. 2004; Frisell et al. 2012).

SFE models are most informative of the effects of parental separation if it is reasonable to assume that the family environment (including levels of parental conflict) would remain stable in the absence of the parental separation (Sigle-Rushton et al. 2014). In such a case, it is most likely that the younger sibling would have experienced a similar family environment as the older sibling, had the parents not separated. The interpretation of SFE results becomes more problematic if the parental separation is the culmination of a deterioration of the family environment (such as increased parental conflict). It is likely that the family environment would have continued to deteriorate had the parents not separated, and the younger sibling would have been taking her grades in a more conflictual family (than her older sibling experienced). Without additional measures, SFE models thus generally rely on the assumption of the stability of the family environment (cf. Sigle-Rushton et al. 2014).

SFE models estimate the effect of the event of the parental separation rather than the separation process. Because SFE models are estimated from a subsample of families that dissolved, the estimates are difficult to generalize without making additional assumptions. Also, because the estimates tell about differences between siblings who experienced parental separation but at different ages, or experienced a different amount of time in a separated family, the estimates are best interpreted as effects of the timing of the separation, as argued in detail by Sigle-Rushton and colleagues (2014).

4.3 Longitudinal Designs

Research with longitudinal data has been more applied  in the USA than in Europe (McLanahan et al. 2013), possibly because of data access issues. Such data can be analyzed using many methods, but unlike with SFE, these methods can only be used to analyze outcomes that are measured more than once. Similar to SFE models, longitudinal studies generally report weaker effects on child outcomes of parental separation and other family transitions than found in cross-sectional analyses.

4.3.1 Lagged Dependent Variables

In lagged dependent variable (LDV) analyses, one controls for the dependent variable at an earlier measurement point (before parental separation) (Johnson 2005; McLanahan et al. 2013). The idea is to adjust for initial differences in outcomes between children from separated and intact families. LDV is mostly used in cohort and other studies with just two or few measurement points. Early examples include studies in Britain, which found that although children of divorce had lower psychological well-being already pre-divorce, parental divorce had negative long-term effects (Cherlin et al. 1991; Chase-Lansdale et al. 1995). Limitations of LDV models include that the estimates are sensitive to omitted variables that affect both the separation and the pre-separation outcome, as well as measurement error in the latter (Johnson 2005).

The pre-separation measurement point can correspond poorly to the stages of the parental separation process, especially in cohort studies in which measurements are often done several years apart. LDV models are therefore most appropriate if the differences in the outcome between children who experienced parental separation and those who did not can be assumed to be stable. If one assumes that the child’s well-being deteriorated prior to the separation, the lagged dependent variable can capture part of the effect of the separation process. However, if the measurements are taken several years apart, it is even more difficult than usual to tell whether the outcome was measured before or during the pre-separation deterioration in well-being and consequently, how the estimated coefficient should be interpreted.

4.3.2 Individual Fixed Effects

Individual fixed effects (IFE) models are based on comparing individuals before and after the parental separation and in effect, use individuals as their own control groups to control for time-constant unobserved factors. In an early British IFE study, Cherlin et al. (1998) concluded that experience of parental separation had weak to moderate negative effects on adulthood psychological well-being, and Amato and Anthony (2014) reported similar effects on educational, psychological, and health outcomes in the USA. Other American studies have used IFE designs to analyze the effects of the number of transitions (e.g., Fomby and Cherlin 2007), of different family transitions (e.g., Lee and McLanahan 2015), or combined SFE and IFE approaches (Gennetian 2005).

IFE methods estimate the effect of parental separation if it is reasonable to assume that the child whose parents separated would have experienced similar (age-specific) outcomes in the absence of separation as observed before the separation (Aughinbaugh et al. 2005). Again, this is most feasible if the child’s level of well-being can be assumed to have remained stable. This is less likely if the child’s well-being began to deteriorate already before the separation, because this deterioration could have continued had the parents not separated. Two US studies attempted to address this issue by tracing behavioral problems and academic achievement before and after the parental separation (Aughinbaugh et al. 2005) and by using a triple-difference approach, which compares trends (and not just levels) in the outcome between children from separated and intact families (Sanz-de-Galdeano and Vuri 2007). Neither study found the event of parental separation to have appreciable effects.

Furthermore, as in SFE models, IFE effects are estimated only from those children who actually experienced the separation. This generally means a reduction in sample size. For the same reason, IFE results generalize primarily to that group.

4.3.3 Placebo Tests and Growth-Curve Models

Longitudinal data can also be used to conduct “placebo tests,” that is, to analyze whether future separation (e.g., t + 1) predicts earlier outcomes (t, or earlier). Bernardi and Boertien (in this Special Issue) found with British data that although children who experienced parental separation before age 16 had a lower probability of transitioning to post-compulsory secondary education, this was not the case for children whose parents separated between ages 17 and 19 (i.e., after the educational transition age). This supports the view that the separation, and not the family environment that preceded it, had an effect on educational decisions.

Finally, longitudinal data have been analyzed with growth-curve models (GCM) to track trajectories in children’s outcomes. Cherlin et al. (1998) reported that the effects of parental separation on psychological problems increased through adolescence and young adulthood. Even though growth-curve models enable analysis of how effects develop, they are not immune to confounding from unmeasured variables that can affect both the initial level of well-being and its development over time (McLanahan et al. 2013). To address this, Kim (2011) combined matching methods with GCM and found that cognitive skills and non-cognitive traits developed negatively already through the separation period and the effects were amplified by the separation event.

4.4 Interpreting Causal Effects

Controlling for measured and unmeasured confounders practically always leads to reduced effect sizes, which means that children who experienced parental separation would have fared differently to children from intact families regardless. Some studies have found no effects, but the prevailing conclusion is that parental separation can have weak to moderate negative effects (Amato 2000, 2010; McLanahan et al. 2013; Ribar 2004).

Increasing adoption of advanced methods to control for unmeasured variables improves our understanding of the consequences of family change. None of the methods are, however, completely immune to confounding by unobserved variables. Relatedly, they also correspond differently to underlying theoretical models of parental separation, which affects their interpretation.

We repeatedly mentioned how the methods are most robust if it is reasonable to assume that the family environment, and the children’s well-being, remained stable before the separation and would have remained stable in its absence. Such a scenario characterizes some separations but provides a poorer description of many others where separation was a culmination of a deteriorating family environment (Amato 2000; Demo and Fine 2010; Härkönen 2014). In some cases, additional (time-varying) control variables (e.g., Ermisch et al. 2004; Lee and McLanahan 2015) or more complex research designs (e.g., Sanz-de-Galdeano and Vuri 2007) can be used to alleviate these problems. When choosing the appropriate variables or designs, one should decide whether one is interested in the effects of the separation event or the exposure to the whole separation process. Both are relevant, and their analysis each carries specific challenges. We also discussed how some estimates might be better interpreted as indicators of the influence of the timing of parental separation (cf. Furstenberg and Kiernan 2001), another relevant yet different question. All in all, scholars should pay attention to which effects their methods estimate and think of this in light of the underlying theoretical model of parental separation or other family dynamics they are interested in (cf. Manski et al. 1992; Ní Bhrolcháin 2001).

5 For Whom, When, and Where are Family Transitions Most Consequential?

Most studies reviewed above analyzed what happens on average. Whereas the finding that children growing up in non-traditional families have different outcomes is very consistent, this result hides a large variation in effects at the individual level. A minority of children suffer from a parental separation, but a somewhat smaller minority shows improvements in well-being and performance, and even if parental separation can be a taxing experience associated with sadness and feelings of loss, a large minority or even a majority of children do “just fine” without robust effects in either direction (Amato 2000, 2010; Amato and Anthony 2014; Amato and James 2010; Demo and Fine 2010). Next, we discuss how this heterogeneity in effects is related to pre-separation parental conflict and children’s and parents’ socio-demographic attributes. After that, we review what is known about variation in the effects over time and cross-nationally.

5.1 For Whom Does It Matter?

Which children are more likely to suffer from parental separation than others? Studies both from the USA (Amato et al. 1995; Hanson 1999; Booth and Amato 2001) and Europe (Dronkers 1999) have found that pre-separation parental conflict moderates the effects of the separation. Parental separation can be beneficial for children from high-conflict families, but is more likely to have negative effects when parental conflict was low and the separation came as a relative surprise.

Other studies have analyzed variation in the effects of parental separation by demographic characteristics. Although some studies have found gender-specific effects, most have not, leading Amato and James (2010) to conclude that the gender differences in effects are modest at most. Similar variation in findings characterizes research on effects of stepfamilies (Sweeney 2010).

Child’s age at parental separation has been another moderator of interest. Breakups occurring while children are adults have no or the smallest effects (Cherlin et al. 1998; Kiernan and Cherlin 1999; Furstenberg and Kiernan 2001; Lyngstad and Engelhardt 2009). Studies on educational outcomes often find the effects to be most pronounced when parents divorced close to important educational decision points (Jonsson and Gähler 1997; Lyngstad and Engelhardt 2009; Sigle-Rushton et al. 2014). Otherwise, findings differ in their conclusions about the childhood stages most sensitive to family disruption, and the specific pattern of heterogeneity is likely to depend on the outcome studied.

Recently, scholars have become increasingly interested in whether effects of parental separation differ by parental socioeconomic status (Augustine 2014; Grätz 2015; Mandemakers and Kalmijn 2014). Although having resources can help families to deal with family transitions, children from resourceful families could also lose more from parental separation (Bernardi and Radl 2014; Bernardi and Boertien 2016a). In line with these contrasting predictions, empirical results are mixed, with some findings pointing to stronger negative effects in families with high (Augustine 2014; Grätz 2015; Mandemakers and Kalmijn 2014) or low socioeconomic status (Bernardi and Boertien 2016a; Bernardi and Radl 2014; Biblarz and Raftery 1999; Martin 2012; McLanahan and Sandefur 1994). Bernardi and Boertien (2017, this Special Issue) address this inconsistency. They show that methodological choices underlie part of this variation in results, but their substantive conclusion is that the negative effect of parental separation on educational choices is stronger for children whose high-socioeconomic status father moves out. The greater financial losses are an important part of the explanation, which also suggests that the results might be different for outcomes that are less responsive to financial resources.

Other studies have compared the effects of parental separation and single parenthood between ethnic, racial, and migrant groups. Many US studies have found that Black children are less affected by growing up in a non-intact family than White children (Fomby and Cherlin 2007; McLanahan and Bumpass 1988; McLanahan and Sandefur 1994; Sun and Li 2007). Some European studies have found variation in family structure effects by ethnic and immigrant background (Kalmijn 2010, forthcoming; Erman and Härkönen, this ‘Special Issue’). In general, the family structure effects are weaker in groups in which parental separation and single motherhood are more common, which has been explained by less stigma, better ways of handling father absence, a broadly disadvantaged position with less to lose, or differential selection by unobserved factors, as argued by Erman and Härkönen in this Special Issue.

Instead of analyzing different predictors of separation separately, Amato and Anthony (2014) used several of these predictors together to, first, predict the children’s propensity to experience parental separation, and second, analyze whether parental divorce effects vary by this propensity. They found that the effects were the strongest for children with the highest risk of experiencing parental divorce, a result seemingly at odds with the above-mentioned findings of weaker effects in groups with higher separation rates.

5.2 Stability Over Time

It is straightforward to expect that the effects of family transitions on child outcomes should have waned over time. As non-traditional family forms have become more common, the social stigma attached to them should decrease (Lansford 2009). Children of divorce are also increasingly likely to retain close contact with both of their parents (e.g., Amato and Gilbreth 1999; Gähler and Palmtag 2015) and families and societies may have in general become better in handling the consequences of family change. Yet, several studies have reported remarkable stability in the negative associations between parental separation and educational attainment, psychological well-being, and own family dissolution risk (Albertini and Garriga 2011; Biblarz and Raftery 1999; Dronkers and Härkönen 2008; Sigle-Rushton et al. 2005; Li and Wu 2008; Gähler and Palmtag 2015). Some studies have found changing effects, but in opposite directions: a waning intergenerational transmission of divorce (Wolfinger 2005; Engelhardt et al. 2002), but a strengthening effect of parental separation on educational attainment (Kreidl et al. 2017).

Why this general stability? One possibility is that although some factors associated with parental separation, such as stigma, have become less common, other proximate consequences—including shock, grief, and anger over the separation of the parents (Pryor and Rodgers 2001)—have remained stable. Another potential explanation refers to changing selection into separation. Parental separation has become increasingly associated with low levels of maternal education (Härkönen and Dronkers 2006). The motives for divorce have also changed over time. Fewer parental separations are today preceded by severe conflict and violence, whereas more are characterized by psychological motives and disagreements upon the division of labor (De Graaf and Kalmijn 2006; Gähler and Palmtag 2015). In general, changing selectivity of parental separation can have offset any weakening trend in its effects. The data requirements to disentangle these explanations are high, but those studies which have appropriate variables support the conclusion of a generally stable effect (Sigle-Rushton et al. 2005; Gähler and Palmtag 2015).

5.3 Cross-National Variation

Associations between family structure and child outcomes are robust in the sense that they are generally found in each country (cf. Amato and James 2010) and are often more similar than one might expect (Härkönen 2015). However, many studies have reported cross-national variation in the strength of associations (e.g., Brolin Låftman 2010; Radl et al. 2017, this Special Issue). A series of studies found that countries with policies aimed at equalizing the living conditions between different types of families had smaller family structure gaps in educational achievement (Pong et al. 2003; Hampden-Thompson 2013; however, see Brolin Låftman 2010). Larger family structure differences have also been reported in economically more developed societies, where the nuclear family plays a more important role (Amato and Boyd 2014).

Dronkers and Härkönen (2008) found that the intergenerational transmission of divorce was weaker in countries where parental divorce was more common. This fits the intuition of weaker penalties when certain family behaviors are more common. However, other studies have found the opposite (Pong et al. 2003; Kreidl et al. 2017). An explanation is that in societies in which separation is uncommon, it is more often a solution to ending very troubled relationships and therefore more likely to be beneficial for the children.

6 Discussion and Recommendations for Future Research

We set the stage for future research in four directions. First, understanding the effects of heterogeneous family forms and transitions will be a research priority in the future as well (Amato 2010). Most of the research reviewed in this introduction has focused on the effects of parental separation, but scholars have been increasingly aware of and interested in the complexity of family forms in today’s societies. Some of this research was addressed in this article, and the analyses by Mariani, Özcan, and Goisis, and Radl, Salazar, and Cebolla-Boado in this Special Issue are further contributions to this topic: the former being the first to look at the outcomes of children born in lone mother families within one European country (the UK), and the latter providing a cross-national overview of the effects of various types of family structures. Future research, particularly in Europe, should continue addressing questions such as the effects of experiencing multiple family transitions and of complex family life course trajectories during childhood. Family complexity can also mean that the boundaries between family forms become blurred. An example is the increasing popularity of joint residential custody, which questions earlier divisions into single-parent and two-parent families. Understanding the effects of family forms under family complexity thus also means an update in conceptual thinking.

Second, children react to (changes in) family circumstances in remarkably different ways (e.g., Amato and Anthony 2014), which is hidden under the average effects reported in most studies. Three of the papers in this Special Issue address these questions and identify subgroups for which effects appear to be more limited compared to other groups such as low SES families and children from ethnic minorities. Better understanding the sources of vulnerability and resilience in the face of family change will continue to be a priority for research, and in this task, future research will benefit from combining theoretical and methodological approaches from sociology, demography, psychology, and genetics (cf. Amato 2010; Demo and Fine 2010).

Another related task for future research will be to systematize the research on variation in family structure effects across individuals and families, groups, and societal contexts. As reviewed in this article, the findings often point to confusingly different directions. Many studies, including the ones by Erman and Härkönen and Bernardi and Boertien in this Special Issue, have found that parental separation effects on educational outcomes are weaker in socioeconomic and ethnic groups where it is more frequent, but Amato and Anthony (2014) reported that the effects are more negative for children who had the highest risk of experiencing parental separation. Yet another group of studies have reported that the effects of parental separation are more negative when the parents had lower levels of conflict—and presumably, low likelihood of separating—before the separation (Amato et al. 1995; Dronkers 1999; Hanson 1999; Demo and Fine 2010). Many cross-national studies have concluded that these effects are stronger in societies in which parental separation is more common (Pong et al. 2003; Kreidl et al. 2017). At the same time, most studies continue to find that parental separation effects have remained stable even though more children have been experiencing it. Understanding these seemingly contradictory results will need theoretical development and appropriate data and designs to test them. Bernardi’s and Boertien’s study in this Special Issue provides a good example of such research.

Third, future research will undoubtedly continue employing sophisticated methods to analyze whether family structures and transitions have causal effects on children’s lives. Yet as discussed above, conceptual thought of what effects can be estimated with different methods and what effects are of most theoretical interest has not necessarily kept up with the methodological advances (for exceptions: Manski et al. 1992; Ní Bhrolcháin 2001; Sigle-Rushton et al. 2014). Using parental separation as our example, we distinguished between the effects of separations as events and separations as processes, as well as between the experience of separation and its timing. Researchers should pay more attention to these differences in the conceptualization of effects, which essentially boils down to the consideration of the underlying theoretical model of parental separation. Better recognition of these differences can contribute to theory-building and methodological advancement and help in formulating advice to parents, family counselors, and policy makers.

Last, these issues have implications for understanding social inequality in a time of family change. The “diverging destinies” thesis (McLanahan and Percheski 2008) holds that socioeconomically uneven family change, in which the retreat from stable two-parent families is happening particularly among those with low levels of education, can reduce social mobility. Yet whether this is the case depends not only on differences in family structures by socioeconomic background, but also on the strength of the effects of these family structures on the outcomes in question; if the effects are nil or weak, it does not matter who lives in which kind of family. The inequality-amplifying effects of socioeconomic differences in family structures can furthermore be shaped by heterogeneity in family structure effects (Bernardi and Boertien 2016b). Bernardi’s and Boertien’s (2017, this Special Issue) findings, that the negative effects of parental separation are weaker for children whose parents have low levels of education, imply that the socioeconomic differences in family instability are less important in affecting intergenerational inequality than often thought. Erman’s and Härkönen’s (2017, this Special Issue) results show that parental separation effects are weaker among ancestry groups where parental separation is more common suggest the same for ethnic inequalities. Together, these findings refine arguments stating that divergence in family structures will lead to an increase in inequality. Instead, the results imply that whether this happens or not is contingent on the strength of these effects and on whether they are similar across groups.