Abstract
This study examines the link between divorced nonresident fathers’ proximity and children’s long-run outcomes, using high-quality data from Norwegian population registers. We follow (from birth to young adulthood) each of 15,992 children born into married households in Norway in the years 1975–1979 whose parents divorced during his or her childhood. We observe the proximity of the child to his or her father in each year following the divorce and link proximity to educational and economic outcomes for the child in young adulthood, controlling for a wide range of observable characteristics of the parents and the child. Our results show that closer proximity to the father following a divorce has, on average, a modest negative association with offspring’s outcomes in young adulthood. The negative associations are stronger among children of highly educated fathers. Complementary Norwegian survey data show that highly educated fathers report more post-divorce conflict with their ex-wives as well as more contact with their children (measured in terms of the number of nights that the child spends at the father’s house). Consequently, the father’s relocation to a more distant location following the divorce may shelter the child from disruptions in the structure of the child’s life as they split time between households and/or from post-divorce interparental conflict.
Introduction
Rising rates of divorce in recent decades have increased political and public concern about nonresident fathers’ involvement in children’s lives. In the United States, for example, many new federal initiatives have been designed to motivate involvement among divorced fathers (Dion 2005), and these have been accompanied by publicly financed ad campaigns promoting the idea of fathers’ integral role in children’s development (Dominus 2005). In particular, there is an increased focus on fathers’ rights to maintain contact with their children following a divorce. In 2004, the California Supreme Court held that several factors—including the distance of the move, the children’s age, and the parents’ relationship—must be considered before children of divorced couples can be moved out of town (McKee 2004). In doing so, the court clarified and effectively overturned a previous ruling that held that custodial parents had the “presumptive right” to relocate unless it would be detrimental to their children or if the move was in bad faith (McKee 2004).
A key underlying assumption of such legal and policy initiatives is that nonresident fathers not only have a right to maintain regular contact with their children but also that doing so will improve those children’s developmental outcomes. Advocates for this position point to evidence concerning risks to development among children with divorced parents as well as fathers’ inputs as important components of children’s development (Amato 2001; Argys et al. 1998; Lamb et al. 1987; McLanahan and Sandefur 1994). However, evidence supporting the positive role of father-child contact following a divorce, per se, is mixed (Amato and Rezac 1994; King 1994a, b).
This study examines the link between divorced nonresident fathers’ proximity and children’s long-run outcomes in a large-scale, national longitudinal sample based on high-quality registry data from Norway. Unlike studies that have relied on small and nonrepresentative samples, our data allow us to follow every child (from birth to young adulthood) born into married households in Norway whose parents divorce during his or her childhood. We observe multiple cohorts of such children, resulting in a large sample that produces precise estimates. Our data allow us to determine the proximity of the child to his or her father in each year following the divorce, and also to assess whether proximity is associated with children’s educational and economic outcomes in young adulthood, controlling for a wide range of observable characteristics of the parents and the child. No other study to our knowledge has adopted such a population perspective on the question of post-divorce father proximity and child outcomes. Our analysis assesses father-child proximity by examining fathers’ moves away from the child to isolate the potential impacts of fathers’ location that are not confounded by other potential impacts of the child’s residential relocation.
To preview our findings, we find no evidence that proximity between divorced fathers and their children benefits children’s educational achievement and human capital attainment in young adulthood. Indeed, our results show that closer proximity to the father following a divorce has, on average, a modest negative association with these outcomes. The negative associations are substantially larger for children of highly educated fathers.
More specifically, our evidence suggests that a child with a college-educated father will increase his or her education by nearly 0.5 years if the father relocates to a more distant location following the divorce. This represents about a 17.5% increase relative to the standard deviation of years of education in our sample. Highly educated fathers’ greater distance post-divorce is also associated with better outcomes for the child in the labor market. A child with a college-educated father is 5.4 percentage points more likely to be working at age 27 if the father relocates to a more distant location following the divorce. Moreover, a child with a college-educated father has earnings that are 12% higher if the father relocates.
Our study investigates possible mechanisms driving these results using survey data of divorced Norwegian parents. Results from the survey data may help to explain why children benefit when highly educated fathers relocate away from their children following a divorce. First, highly educated fathers have a larger degree of post-divorce conflict with their ex-wives. Second, highly educated fathers have more contact (measured in terms of the number of nights that the child spends at the father’s house) with their children post-divorce, which supports U.S. evidence that nonresident father involvement and shared custody arrangements are more common among families of higher socioeconomic status (Amato and Rezac 1994; Cancian and Meyer 1998; Donnelly and Finkelhor 1993; Manning and Smock 1999). Consequently, the father’s relocation to a more distant location following the divorce may shelter the child from disruptions in the structure of the child’s life as the child splits time between households and/or from post-divorce interparental conflict. Nevertheless, the extent to which our findings from Norway extend to other countries is an open question, a point to which we will return in this article’s conclusion.
Background
There is substantial U.S. evidence that children of divorced parents have poorer academic, behavioral, and health outcomes than comparable children in intact families (Amato 2001; McLanahan and Sandefur 1994; Seltzer 1994). In Norway as well, children of divorce demonstrate poorer outcomes to an extent comparable to U.S. samples despite the existence of a generous Scandinavian welfare state that buffers single mothers from serious economic hardship (Breivik and Olweus 2006; Steele et al. 2009; Størksen et al. 2006). For example, Steele et al. (2009), using Norwegian registry data and paying careful attention to selection effects, concluded that parental divorce during early childhood is associated with significantly lower levels of educational attainment.
There is reason to hypothesize that fathers’ proximity to the child following a divorce could have a positive, a negative, or no effect on children’s development, or that the effect could vary for different types of children (e.g., by gender or age at divorce). We will discuss each possibility in turn.
Hypothesis 1: Fathers’ proximity is beneficial because it promotes involvement/investment in children.
Many studies argue that poor child development outcomes following a divorce result from the loss of fathers’ investments that occurs when a marriage dissolves. In the economics literature, these inputs are characterized in general terms by fathers’ inputs of time and money (Becker and Tomes 1986). In the psychology literature, father involvement is conceptualized as a multifaceted construct with three distinct parts: father’s accessibility to, engagement with, and responsibility for his child (Lamb et al. 1987). Accessibility reflects a father’s contact with and availability to the child, irrespective of the quality of their interactions. Engagement is defined by a father’s interactions with his children—including caregiving, play, and teaching activities—and includes both the quantity and quality of father-child interactions (Lamb et al. 1987). Responsibility is conceptualized generally as a father’s involvement in the management of the child’s daily routines, health, and child care, as well as his role in making major decisions about the child.
Nonresident father-child contact (facilitated by living closer rather than farther apart) builds affection and provides opportunities for fathers to engage in active forms of parenting (Sobolewski and King 2005). Proximity to the child following a divorce should therefore be a positive predictor of fathers’ active support and involvement in children’s lives and could ameliorate some of the problems faced by children of divorce.
Some empirical evidence suggests that nonresident divorced fathers’ contact with their children is correlated with good child outcomes, although much of this evidence comes from small and nonrepresentative samples or cross-sectional data. Amato and Gilbreth (1999), in a widely cited meta-analysis, reported that frequency of contact with nonresident fathers has small but statistically significant associations with child outcomes, including higher levels of academic success and lower levels of behavior problems, such as depression and sadness. In contrast, King’s (1994a, b) analysis of large-scale U.S. data found little evidence that father-child visitation has any positive impacts on child development. Seltzer (2000) found that nonresident fathers who maintain closer contact with their children are more likely to pay child support, perhaps because they can observe more directly how that money is supporting the child’s welfare, which could contribute to better outcomes because child support is a significant positive correlate of children’s cognitive development (Argys et al. 1998). Moreover, nonresident fathers tend to bundle their involvement, such that those who see their children frequently also are more likely to be engaged with them, assume parenting responsibility, and provide in-kind support compared with those who see their children infrequently (Coley and Chase-Lansdale 1999; Kalil et al. 2005; Rangarajan and Gleason 1998; Ryan et al. 2008; Sobolewski and King 2005).
The few Norwegian studies that exist on this topic suggest a somewhat stronger positive role for father proximity following divorce than do the U.S. data. Notably, these studies do not use population-level data as we do here, nor do they follow the children of divorce into adulthood. Most of these studies focus on youth’s psychological well-being. Størksen et al. (2005) argued that father absence (i.e., not living with the biological father) accounts for poorer developmental trajectories during adolescence and lower measures of psychological well-being for the children of divorce, especially for boys. Similarly, Breivik and Olweus (2005) argued that children of divorce have optimal psychological outcomes in joint physical custody arrangements (relative to the alternatives), which are undoubtedly highly correlated with divorced fathers’ proximity.
Hypothesis 2: Fathers’ proximity is detrimental because it sparks interparental conflict or undermines stability in children’s lives.
A large literature documents the conflict that often characterizes the divorce process (Furstenberg and Cherlin 1991). Indeed, much of the negative impact of divorce on children can be attributed to adverse pre-divorce conditions in families and children (Morrison and Cherlin 1995). As such, some research suggests that greater increases in behavior problems are seen among children who remain in high-conflict marriages than among children whose parents separate or divorce (Morrison and Coiro 1999). If conflict between the parents persists after the divorce, closer proximity of the fathers may create more opportunities for disagreement, to the detriment of the children. Amato and Rezac (1994), using a large national U.S. sample, found that nonresident father contact was detrimental in terms of boys’ behavior problems when interparental conflict was high (but not when interparental conflict was low). In contrast, for girls, there was no association between nonresident father contact and behavior problems.
Fathers’ proximity post-divorce could also be detrimental if it positively predicts the child’s splitting time between two households, which could reduce stability in the child’s life (Lamb et al. 1997) or increase the likelihood of interparental conflict. The most common living arrangement among divorced families in Norway lets the child visit the nonresident parent (usually the father) one afternoon weekly in addition to every second weekend (Størksen et al. 2006). Children who live closer to their nonresident fathers may have greater contact with those fathers, such as by spending more nights at fathers’ homes. It is possible that frequently moving between two parents’ houses is stressful for children because it requires adjusting to different sets of parental rules and routines or because it disrupts children’s daily activities.
Hypothesis 3: Fathers’ proximity will not matter.
There is also reason to think that divorced fathers’ proximity to the child will not matter for children’s development. Economic theories suggest that divorce occurs upon the discovery that a marriage “match” is of poor quality (Becker et al. 1977). For instance, divorce may result from negative surprises about the father’s ability to invest productively in the family’s welfare or the child’s development (Charles and Stephens 2004; Chiappori and Weiss 2006; Kalil et al. 2010; Rege et al. 2009). If so, divorce may have the effect of removing fathers from the household who are least skilled at parenting, making fathers’ post-divorce proximity irrelevant for child development. In support of this hypothesis, Sobolewski and Amato (2007) showed that when children experienced their parents’ marital conflict and divorce growing up, the children had no higher levels of subjective well-being in young adulthood if they were emotionally close to both parents rather than to one parent only. In such “incongruent patterns” of parent-child closeness, the children were more likely to be close to their mothers only, rather than their fathers only, suggesting that emotional closeness to fathers has little “value added” for young adults’ subjective well-being in the aftermath of parental marital conflict and divorce, given an emotionally close relationship with mothers.
Indeed, many of the observed differences in developmental outcomes between the children who do and do not grow up with their father stem from the factors that selected families into divorce in the first place. At least two studies have suggested that marriage confers little benefit to (and may even detract from) children’s development when fathers lack the “skills” (i.e., education, emotional stability) necessary for promoting positive child outcomes (Jaffee et al. 2003; Ryan 2008). Of course, the divorce process (like the marriage market) is imperfect; fathers skilled at parenting can still end up divorced, and the involvement of these nonresident fathers may benefit children’s development. This suggests that the effects of post-divorce father proximity on child outcomes may also depend on fathers’ characteristics, such as his education. It is also possible that fathers’ proximity matters when mothers have limited human capital or parenting skills, or that strong bonds to nonresident fathers matter for children with weak ties to their mothers (King and Sobolewski 2006; Ryan et al. 2006).
Hypothesis 4: Fathers’ post-divorce proximity will have heterogeneous effects, depending on the child.
Researchers have speculated that nonresident father contact and involvement might be more important for boys than for girls (Coley and Chase-Lansdale 1999), given the importance of a male role model for boys’ identity development. Family size and birth order might also moderate the influence of nonresident fathers’ proximity. In large families, for example, where each child’s share of the mothers’ resources is smaller than in families with fewer children, it may be especially important to garner the attention and inputs of the nonresident father. Similarly, later-born children, because they receive fewer resources than earlier-born children within families (Conley 2004), may be especially likely to benefit from nonresident fathers’ inputs.
The age at which the divorce occurs might moderate the influence of nonresident father contact, although it is unclear what sign this interaction would take. The share of post-divorce years spent living in proximity to the nonresident father may matter more for children whose parents divorced at an early age simply because the benefits or detriments associated with fathers’ post-divorce proximity accumulate over a larger number of childhood years. Alternatively, the limits on father-child interactions imposed by a lack of post-divorce proximity could represent a greater loss to those children who have spent a greater proportion of their childhood living with their fathers in a married household. In this case, the benefits of proximity may be larger when the divorce occurs later in the child’s life.
Data
Our empirical analysis utilizes several registry databases provided by Statistics Norway. The data include a rich longitudinal data set containing records for every Norwegian from 1975 to 2006. The variables captured in this data set include individual demographic information (sex, age, marital status, number of children) and socioeconomic data (years of education, earnings). Importantly, the data set includes personal identifiers for one’s parents, allowing us to link children to their parents and siblings. Moreover, the data set includes geographic identifiers for economic region of residence, allowing us to construct a measure for father’s proximity.1
The Norwegian territory covers about 149,400 square miles, an area about the size of California or Germany, with around 13% and 6% of those regions’ populations (in 2008), respectively. The country is dominated by mountainous or high terrain, as well as a rugged coastline stretching about 1,650 miles, broken by numerous fjords and thousands of islands (making the coastline approximately 10 times longer if the length of the fjords were included). An “economic region” constitutes a regional level between a county (of which there are 19) and a municipality (of which there are 435), approximating regional labor markets. There are 46 economic regions in Norway with an average population (in 1975) of about 86,900 people. The main criteria used for defining the economic regions are labor market, trade and service patterns, and commuting and internal migration patterns.2 In general, there are long driving distances between the populated areas of the economic regions because they are mostly far apart or partitioned by mountains and/or the fjord-gashed shoreline.
As a measure of a father’s proximity, we construct a variable denoting the proportion of years that the child and father live in the same economic region between the divorce and the child’s 18th birthday.3 Most fathers either live away from the child most of the time or, conversely, live mostly in the same region as the child. This leads us to focus the empirical analysis on these extremes. Henceforth, we will refer to a proximate father as a father who lived in the same economic region for more than 90% of the years between the divorce and the child’s 18th birthday. Conversely, a distant father is a father who lived in the same economic region for less than 10% of the years between the divorce and the child’s 18th birthday. The motivation for letting economic region, rather than municipality, form the basis for defining proximity is to make sure that distant fathers live sufficiently far away from the child to affect father-child contact and minimize the possibility for shared custody arrangements.
Our empirical analysis focuses on the 1975–1979 cohorts in order to ensure availability of outcome measures when the child reaches the age of 27. Our analytic sample is restricted to children whose parents were married when the child was born, which makes up about 90% of the aforementioned sample. The reason for this choice is that our family data do not allow us to distinguish between cohabitants and single parents, nor do we observe the dissolution of cohabitation. When interpreting our results, keep in mind that our findings do not speak to the effect of a distant father after the dissolution of cohabitation. For example, it is possible that ex-cohabiting fathers, compared with divorced fathers, have different relationships with their children or their former partners. Further, we restrict the sample to children whose parents were divorced before the child turned age 13,4 and who reside in the same municipality as their mother until age 18. To ensure that our “treatment” variable is well defined, we also exclude a small number of children whose father died before the child’s 18th birthday. With these restrictions, we have a sample of almost 20,800 children. Next, we exclude about 2,700 children who moved to a different region before age 18 to ensure that our estimate of the effect of father’s geographic proximity is not due to the child’s or mother’s relocation. Finally, as explained earlier, we restrict the sample to children whose father is categorized as either proximate or distant. Applying these restrictions provides us with a sample of 15,992 children.
Our key outcome variable is the child’s years of education at age 27. Figure 1 illustrates the distribution of this variable for our analytical sample. We also construct two dummy variables denoting whether the child obtained at least 12 years of education (high school graduate) and whether the child obtained at least 15 years of education (college graduate). Further, we utilize several outcome variables capturing socioeconomic status at age 27. In addition to annual earnings, which consist of all market income from wages and self-employment, these include dummy variables indicating whether the child is working, unemployed, receiving welfare, or married at age 27. Finally, we examine whether the youth had a child of his or her own prior to age 21.
Our rich data set allows us to construct several variables capturing important child, father, and mother characteristics to include in our regression analyses. Notably, characteristics that are possibly endogenous to the relocation event (such as earnings) are captured prior to the relocation. The set of control variables includes dummy variables for the economic region of residence of the family at the child’s birth; the child’s sex, year of birth, first- or second-generation immigrant status, number of siblings, birth order, and age at parental divorce; the father’s age at the birth of the child (six dummy variables for ages <20, 20–24, 25–29, 30–34, 35–39, and ≥40); father’s age at the birth of his oldest child (six dummy variables for ages <20, 20–24, 25–29, 30–34, 35–39, and ≥40); father’s years of education (four dummy variables for ≤9, 10–12, 13–15, and ≥16); and father’s average earnings in the two years prior to divorce (entered in linear and quadratic form). A set of mother characteristics is also controlled for and is coded in an identical fashion to the father variables. These include her age at birth of the child, her age at birth of her oldest child, the number of children she had prior to the child’s birth with men other than the father of the child, and her education and pre-divorce earnings.
In some regressions, we will also add controls for mother and father post-divorce circumstances. Specifically, we construct a dummy variable indicating whether the father remarried prior to the child’s 16th birthday and a dummy variable indicating whether the father had a new child prior to the child’s 16th birthday. Similar dummy variables were constructed for the mother.
Empirical Strategy
The parameter of interest is the estimated distant father coefficient, η, which captures the incremental increase in education from having a distant father relative to children whose father is proximate. Estimation of Eq. 1 will produce unbiased estimates of η provided that there are no omitted determinants of children’s outcomes that are correlated with father’s relocation.5
The identifying assumption may be difficult to defend for several reasons. For example, more productive fathers may have greater access to more employment opportunities outside their economic region and may be more mobile as a result. If children of more productive fathers tend to achieve more education, our estimate of η will be biased upward. Alternatively, our estimate could be biased downward if fathers predisposed to invest less time in their children are also more likely to move away. In our empirical analysis, we address these potential sources of bias by investigating the robustness of our estimate to inclusion and exclusion of our large set of covariates capturing important parental characteristics. Admittedly, in the absence of experimental data, the threat of selection bias cannot be fully resolved. In particular, a reverse causality story is possible: fathers’ decisions to relocate may be affected by their expectations for the child’s development.
Empirical Results
Summary Statistics
Our sample consists of 15,992 children, of whom 14.5% have a distant father and 85.5% have a proximate father. In Table 1, we separately report the means and standard deviations for the sample of children with distant fathers and the sample of children with proximate fathers. We note that child educational outcomes are slightly higher for children with a distant father.
Table 1 also shows that child and parent’s pre-divorce characteristics are fairly similar across the two groups. This is important because large differences in the observable characteristics of children with proximate and distant fathers may suggest unobservable differences, calling our empirical strategy into question. Across the two samples, there are, however, some differences in post-divorce circumstances. In particular, mothers are less likely to remarry if the father is distant, and distant fathers are more likely to have a new child.
A concern in applying a parametric regression model under the assumption of selection on observables is lack of overlap in the covariate distribution. As emphasized by Imbens and Wooldridge (2008), this can be assessed by the (scale-invariant) normalized difference measure. For each covariate, the normalized difference is defined as the difference in averages by treatment status, scaled by the square root of the sum of variances. Imbens and Wooldridge suggested that as a rule, parametric regression models tend to be sensitive to the functional form assumption if the normalized difference exceeds one-quarter. The last column of Table 1 displays normalized differences for child and parental covariates, indicating that lack of overlap should be of little concern for our estimation results.
Main Results
As described earlier, our empirical analysis compares educational outcomes of children with a proximate father to children with a distant father. The estimated coefficient captures the incremental increase in education from having a distant father relative to children whose father is proximate. Table 2 presents our main results. We run each model specification for three different educational outcome variables: years of education, high school graduate, and college graduate. Standard errors in Table 2 (and subsequent tables) allow for dependence of residuals across siblings with the same mother.
Controlling only for cohort fixed effects, Model 1 finds that the educational attainment of children with a distant father is significantly higher compared with children with a proximate father. Models 2 and 3 show that this result is robust to the inclusion of controls for the child’s age at divorce, region fixed effects, and child characteristics.
As discussed earlier, an estimate of the impact of a distant father on a child’s education is likely biased by omitted variables. In particular, a positive effect estimate may be due to fathers’ being more likely to relocate if the mother is resourceful and can manage most of the parenting herself. Alternatively, more resourceful fathers may be more likely to relocate due to better job opportunities. Consistent with an omitted variable bias story, we see that our estimate declines somewhat when controlling for mother or father characteristics in Models 4 and 5. However, the estimates in the different models are not significantly different. Moreover, we can see in Models 4 and 5 that controlling for either parent’s characteristics has similar effects on the estimate, whereas Model 6 shows that further adding of the other parent’s characteristics does not greatly affect the estimates. Even though we cannot rule out omitted variable bias, the robustness to inclusion of father and mother characteristics is reassuring in this regard.
Post-divorce circumstances give rise to another source of omitted variable bias. For example, as our summary statistics suggest, a father may relocate because he is starting a new family. If this is the case, our estimate may be capturing the effect of a father’s remarriage and new childbearing, rather than the effect of his relocation. In Models 7 and 8, we investigate such a possibility by including controls for mothers’ and fathers’ post-divorce characteristics. Post-divorce characteristics are dummy variables capturing the individual’s new family situations and are described in the “Data” section. Admittedly, the analyses in Models 7 and 8 suffer from an endogeneity problem because a father’s relocation can also affect the likelihood of remarriage and new childbearing. Nevertheless, the fact that our point estimates barely move when we include fathers’ and mothers’ post-divorce characteristics suggests that the estimate of having a distant father is not seriously biased by the effect of the parents’ new family situations.6
Due to the endogeneity problems in Models 7 and 8, we view Model 6 as our preferred specification. In the first set of rows of regression results, we can see that this model suggests that children with a distant father, on average, have 0.13 years more education than children with proximate fathers. This represents a 5% increase relative to the standard deviation of years of education in our sample. Alternatively, in the second set of rows of regression results, we see that a distant father increases the likelihood of graduating from high school by 2 percentage points. About 66% of our sample graduate from high school. Although statistically significant, one can arguably question the economic significance of these estimates. For example, an increase of 0.13 years of education corresponds to a mere 0.65% increase in annual earnings.7 Nevertheless, the empirical analysis in Table 2 suggests that nonresident fathers’ proximity to their children does not have a beneficial impact on those children’s educational achievement. In fact, the evidence suggests that closer proximity to the father following a divorce has a modest negative association with these outcomes.
The second and third set of rows of regression results report estimates using high school graduate and college graduate as dependent variables. These results are consistent with a linear effect of having a distant father on years of education. Hereafter, we therefore focus on years of education as our outcome variable.
Heterogeneous Effects
The “Background” section discusses several reasons for why the effect of fathers’ relocation could be different for different subgroups of children (e.g., by gender or age at divorce). This raises the possibility that the modest estimates in Table 2 mask meaningful effects within certain subgroups of children. Table 3 investigates this possibility. Model 1 replicates our preferred Model 6 from Table 2. In Models 2–5, we add terms interacting distant with child characteristics. While these interaction effects are somewhat imprecisely estimated, we find no significant differences between the different groups of children.
In the “Background” section, we also hypothesize that children with a high-skilled father may benefit more from paternal input than a child with a low-skilled father. If so, we should expect that children of fathers with higher education benefit more from proximity than other children. In Table 4, we investigate whether the effect of having a distant father depends on fathers’ and mothers’ education and pre-divorce earnings. Model 1 replicates our preferred Model 6 from Table 2. Model 2 adds an interaction term between distant and whether the father attended college. Interestingly, from the interaction term, we see that the positive effect of having a distant father is substantially larger if the father has some college education, while the point estimate for children of fathers with no college education is small and no longer statistically significant.
In Model 3, we add an additional interaction term between distant and a high (pre-divorce) earnings dummy variable for the father, defined as having earnings above the median among fathers in our sample. We see that the point estimate for having a distant father with high education barely changes, and that fathers’ earnings appear to be irrelevant for the effect of distant. Model 4 adds similar interaction terms for mothers’ earnings and education. The point estimate on having a distant father with high education is also robust to these additional covariates. Moreover, mothers’ education and earnings are not statistically significant mediators of the distant effect. Models 5 and 6 assure that the model is robust to the inclusion of fathers’ and mothers’ post-divorce characteristics interacted with distant father, suggesting that the effect estimate of having a distant college-educated father is not biased by the effect of the parents’ new family situations.
The evidence in Table 4 thus suggests positive effects of having a distant father post-divorce if the father is highly educated, with small (and perhaps no) effect if the father has no more than a high school diploma.8 Specifically, the results from Model 2 suggest that a child with a college-educated father will increase his or her education by nearly 0.5 years if the father relocates to a more distant location following the divorce. This represents about a 17.5% increase relative to the standard deviation of years of education in our sample.
Additional Outcomes
Table 5 investigates the effect of having a distant father relative to children whose father is proximate on other socioeconomic outcomes (defined in the “Empirical Strategy” section).9 All models include the control variables of Model 6 in Table 2. In line with the results on education, the results presented in the first row of regression coefficients suggest that having a distant father does not, on average, seem to matter for the child’s socioeconomic status at age 27. Our investigation of heterogeneous effects in Table 4, however, suggests that children of fathers with a college education benefit from fathers’ relocation in terms of higher educational attainment. The last row of regression coefficients includes interaction terms to investigate similar heterogeneity along these outcomes.
The analyses in Table 5 provide additional evidence of the adverse impacts of father’s proximity post-divorce if the father has a college education. In particular, if the father is college-educated, then his distance post-divorce strengthens the labor force attachment, increases earnings, and reduces welfare dependency of the child at age 27. For example, the first column of regression coefficients showing the interaction results suggests that a child with college-educated father is 5.4 percentage points more likely to be working at age 27 if the father relocates to a more distant location following the divorce. This represents more than an 8% increase relative to the work force participation rate in our sample. Similarly, the second column suggests that a child with college-educated father will increase his or her earnings by more than 12% if the father relocates.
Investigation of Mechanisms Using Survey Data
The results in Tables 4 and 5 contradict the hypothesis that children of highly skilled fathers benefit more from fathers’ proximity post-divorce than children of low-skilled fathers. Instead, highly educated fathers’ proximity seems to be negatively associated with children’s educational attainment and future economic success. A possible explanation for this could be that highly educated fathers are more involved in their children’s lives following the divorce, including a greater likelihood of having the child stay with the father for overnight visits and weekends and possibly a greater likelihood of shared custody arrangements (Amato and Rezac 1994; Cancian and Meyer 1998; Donnelly and Finkelhor 1993; Manning and Smock 1999). As discussed earlier, fathers’ involvement of this sort may not be beneficial to the child if it contributes to post-divorce interparental conflict, or if splitting time between two households disrupts needed stability in the child’s life (Lamb et al. 1997). Consequently, it may be beneficial to the child if a highly educated father relocates because it shelters the child from interparental post-divorce conflict and/or creates a more stable home environment.
This hypothesis would be supported by evidence that divorced college-educated fathers have more contact with their nonresident children and a greater degree of conflict with their ex-spouses. Our register data do not allow us to test this because we cannot observe post-divorce interparental conflict and father-child contact. However, we can utilize responses to a mail survey performed by Statistics Norway in 2002 to investigate this issue. The survey asked divorced parents about nonresident father-child contact and interparental conflict. The responses were matched to register data, providing a rich set of variables capturing the respondents’ socioeconomic status. The survey had a response rate of about 60%. In total, the survey interviewed 2, 309 parents, including 749 couples in which both the mother and the father of a child responded (i.e., 1,498 responses).
In the analysis of these survey data, we strive to parallel the sample in our investigation of the register data. Consequently, we utilize data on all fathers who were living with the mother at the birth of their (first) child and divorced before the (first) child turned age 13, where the mother has custody of the child(ren). We also construct control variables capturing child, mother, and father characteristics similar to those described in the “Empirical Strategy” section. We have two main variables of interest: (1) a dummy variable measuring whether the father reports a high level of post-divorce conflict with the child’s mother;10 and (2) a dummy variable indicating that the child spends more than one of five nights at the father’s home.
Table 6 investigates whether a high level of post-divorce conflict is more prevalent among fathers with a college education. Consistent with the mechanism outlined in Hypothesis 2, Model 1 finds that college-educated fathers report a higher level of post-divorce conflict compared with fathers with no college education. Models 2 and 3 show that this correlation is robust to the inclusion of post-divorce characteristics and father and child characteristics. In line with other studies, these data also show a correlation between fathers’ remarrying/cohabiting and post-divorce conflict (Cherlin and Furstenberg 1994). Over the sample for which we have responses from both the mother and father, Models 4 and 5 replicate Models 2 and 3 (respectively), while Model 6 controls for characteristics of the mother. Estimates are modestly larger for this sample but remain robust to the inclusion of mother’s characteristics.11
We also find in these survey data that compared with other children, children of college-educated fathers spend more time with their father post-divorce. This correlation is robust to the inclusion of post-divorce characteristics, in addition to father and child characteristics. Again, the correlation appears modestly stronger when estimated over the sample for which we have responses from both parents and is robust to the inclusion of mother’s characteristics (results not presented in tables but available upon request).
In summary, the survey evidence suggests that highly educated fathers report a higher degree of post-divorce conflict with their ex-wives and are more involved in their children’s lives, at least as measured by the number of nights the child spends at the father’s house. Consistent with Hypothesis 2, this suggests an explanation for why children may benefit when highly educated fathers relocate: the father’s relocation may shelter the child from post-divorce conflict and/or from persistent disruptions in the structure of the child’s life as he or she splits time between households.
Conclusion
In the United States and elsewhere, several recent legal and policy initiatives have been undertaken to motivate involvement of divorced fathers in their children’s lives (Dion 2005), including laws restricting the relocation of families following divorce (McKee 2004). These initiatives are generally motivated by the notion that involvement of nonresident fathers is beneficial for their children, a hypothesis supported by evidence from many fields highlighting fathers’ inputs as important components of children’s development (Argys et al. 1998; Lamb et al. 1987; McLanahan and Sandefur 1994). Nonetheless, there is a scarcity of compelling evidence supporting the positive role of father-child contact following a divorce (King 1994a, b).
Utilizing high-quality longitudinal population data on children of divorce who have reached young adulthood in Norway, we find no support for the assumption that nonresident fathers’ proximity to their children has a beneficial impact for those children. Instead, our evidence indicates that fathers’ post-divorce proximity is negatively associated with children’s educational achievement and human capital attainment in young adulthood. Importantly, these associations are statistically significant (and are substantial) only for highly educated fathers. These findings should not be interpreted as suggesting that the quality of fathers’ relationships with children following a divorce does not matter to child development, nor that fathers are unimportant in children’s lives. The findings do suggest, however, that living close to a nonresident father following a divorce does not, in and of itself, promote good child outcomes.
A number of plausible hypotheses could explain why living closer to a divorced father fails to promote better child outcomes. For instance, it is possible that fathers’ proximity is not strongly correlated with fathers’ engagement in their children’s lives. If so, this hypothesis runs counter to numerous studies that show a positive correlation between father-child contact and other dimensions of fathers’ parenting involvement (Coley and Chase-Lansdale 1999; Kalil et al. 2005; Ryan et al. 2008; Sobolewski and King 2005).
A second hypothesis is that fathers who divorce are commonly those who have not been successful in the roles of husband and father. It is well established that men’s risk factors for divorce include poor relationship skills, antisocial behavior, and low human capital (Jaffee et al. 2003). The divorce may be the result of the same problems with the mother. To the extent that divorced fathers are characterized by poor parenting skills or serve as poor role models, proximity to such fathers may fail to benefit their children and may even be detrimental to their children’s development (Ryan 2008). Our finding that fathers’ proximity is uncorrelated with children’s outcomes for fathers with less than college education is broadly consistent with this hypothesis, although it fails to explain the negative associations for college-educated fathers.
A third possibility is that any benefit of increased fathers’ engagement, facilitated by close proximity, is offset or even undermined by the challenges in accommodating that engagement (Amato and Rezac 1994). In particular, fathers’ active engagement could come at the cost of increasing post-divorce conflict among the ex-spouses or disrupting stability in children’s day-to-day lives, each of which could adversely affect children’s development. Among these mothers and fathers who did not succeed as a married couple, frequent interaction may make cooperation more difficult (Furstenberg et al. 1987). This hypothesis could explain why fathers’ proximity generally fails to improve children’s outcomes as well as the substantial negative associations for college-educated fathers. Prior analyses have found that highly educated divorced fathers are more likely to be involved in their children’s lives and to request joint custody (Amato and Rezac 1994; Cancian and Meyer 1998; Donnelly and Finkelhor 1993; Manning and Smock 1999). Our analysis of Norwegian survey data finds that college-educated divorced fathers spend more nights with their children and also report higher degrees of post-divorce conflict with their ex-spouses. Together, these results suggest that fathers’ relocation to a more distant location following the divorce may benefit children by sheltering them from interparental conflict and/or from persistent disruptions in the structure of the child’s life as they split time between households. Although these findings pertain to the effects of fathers’ relocation following a divorce, they are also important given recent legal and policy initiatives to limit custodial mothers’ relocation choices following a divorce because these policies are motivated by concerns about father-child post-divorce proximity.
As noted earlier, we wish to avoid implying that nonresident fathers’ relationships do not matter for children’s development following a divorce. Our data contain no measures of the quality of relationships or the extent of responsive parenting between nonresident fathers and their children. These are the two dimensions of nonresident fathers’ behavior that have been most often linked to positive outcomes for children (Amato and Gilbreth 1999; King and Sobolewski 2006). The limited number of studies that have examined the role of father-child contact, per se, have found inconsistent associations with children’s developmental outcomes (Amato and Gilbreth 1999; Amato and Rezac 1994; King 1994a, b). Our findings are thus generally consistent with these prior studies, although none examined long-run outcomes as we do here, and many are inconclusive on the issue of causality given a reliance on cross-sectional data (e.g., King and Sobolewski 2006). It is also possible that data limitations and analytic flaws characteristic of many of these prior studies reduced their ability to identify the negative associations between father-child proximity and children’s attainment that our study has demonstrated. At the same time, our study examines outcomes that are more strongly related to cognitive than socioemotional development (given information available in the registry data). It is possible that fathers’ proximity would not have had negative associations with socioemotional outcomes.
Finally, it is worth emphasizing that the extent to which our findings from Norway extend to other countries is an open question. Our sample consists of children born into married households in the years 1975–1979 whose parents divorced in his or her early childhood. By then, Norway had already cemented its generous welfare state, including cash transfers for families with children in general and single parents in particular, which may have reduced the importance of fathers’ proximity for financial support and care.
It should also be noted that there is undoubtedly heterogeneity in our sample that we are unable to identify given data limitations. For example, if the average small negative effect of father proximity on children’s attainment is largely a result of conflict that persists between the mother and the nonresident father, this would suggest that proximity matters differently for families who are able to effectively resolve the conflicts of the divorce or who are better able to maintain clear and compatible rules and routines across households, perhaps especially in the case of shared-custody arrangements (see, e.g., Amato and Rezac 1994). Indeed, our findings could be interpreted as support for policy initiatives to better equip divorced parents to minimize the conflicts and disruptions that arise when caregiving responsibilities are shared. It might also be the case that emotional ties to fathers, whose formation is facilitated by proximity, are more beneficial when children have weak emotional ties to their mothers (King and Sobolewski 2006; Ryan et al. 2006). At the same time, a conclusion that is strongly supported by our data is that policies that attempt to regulate father-child proximity following a divorce are unfounded if the motivation for such policies is the positive development of the child.
Acknowledgments
The authors contributed equally to the manuscript and are listed in alphabetical order. Financial support from the Norwegian Research Council (194347/S20) is gratefully acknowledged. This work is also part of the research activities at the Centre of Equality, Social Organization, and Performance (ESOP) at the Department of Economics, University of Oslo. ESOP is supported by the Research Council of Norway. The authors would also like to thank Rebekah Coley, Rebecca Ryan, Ingunn Størksen, participants at the ESOP seminar at the University of Oslo, and seminar participants at the University of Stavanger for helpful comments.
Notes
Unfortunately, we do not know the exact geographical distance between the father and child.
The classification of economic regions is analogous to the NUTS 4 level in EU’s regional classification, whereas county and municipality correspond to the NUTS and the NUTS 5 levels, respectively. See Bhuller (2009) for a detailed description of the classification of economic regions that we use. See Kalil et al. (2009: Figures A1 and A2) for a map of the economic regions of Norway.
We focus on proportion of years instead of number of years because the latter is highly correlated with age of divorce.
We restrict the sample to children whose parents were divorced before the child turned 13 to ensure a reasonable amount of time between divorce and adulthood.
Some of the outcomes of interest are limited dependent variables. In these cases, our linear probability model will be the best least-squares approximation of the true conditional expectation function. As noted by Angrist (2001), if there are no covariates or they are mostly discrete, as in our case, linear models are no less appropriate for limited dependent variables than for other types of dependent variables. In any case, we have checked that our results are robust to alternative approximations of the conditional expectation function, estimating logit and probit models. The marginal effects from these models are very similar to the estimates from the linear probability model.
As a further robustness check, we also added controls for educational attainment of mothers’ and/or fathers’ new spouse. Our estimates barely move. In general, post-divorce covariates add little explanatory power to the model, so the robustness of our effect estimates to their inclusion is not surprising.
Hægeland et al. (1999) estimated the average return on earning per completed year of education to be about 5%.
Kalil et al. (2009: Table A1) presented regression results to investigate the robustness of this finding to alternative covariates, as we performed for the “main effect” in Table 2. The estimates are found to be quite robust to alternative specifications.
About 20% of the children with distant father and 18% of children with proximate father were undertaking education at age 27. Excluding these children from the analysis of the impact of distant father on children working and their earnings at age 27 does not materially affect our results. Specifically, the estimated effects of distant father corresponding to the upper rows of Table 5 are −0.006 (0.011) for working and −1,880 (4,170) for earnings. In comparison, the estimated effects of distant father corresponding to the lower rows of Table 5 are −0.013 (0.012) for working and −7,580 (4,600) for earnings, whereas the estimated effects of distant father interacted with father attended college are 0.049 (0.027) for working and 33,000 (10,860) for earnings. Note, however, that excluding these children creates a potential endogeneity problem because distant father (interacted with father attended college) affects children’s educational attainment and thus the probability of undertaking education at age 27.
This measure was originally scaled with values of 1, 2, 3, and 4, where 4 denotes “high conflict,” and 1 denotes “no conflict.”
The finding of higher conflict reported by college-educated fathers is much less pronounced when we analyze the level of conflict as reported by the mother. Using mothers’ report of conflict, the probability of high post-divorce conflict is only modestly (and insignificantly) higher for college-educated fathers. However, these results are estimated on a smaller sample (the analysis requires a subsample where both the mother and father respond as fathers report their own education) and are very imprecise.
