Matt McGue and David T. Lykken (1992)
Department of Psychology, University of Minnesota
Abstract – Although it has long been recognized that there is increased risk of divorce among the children of divorced parents, the extent to which genetic and environmental factors contribute to this familial resemblance has been a matter of speculation only. In order to resolve the separate influence of genetic and environmental factors on risk of divorce, divorce status of 1,516 same-sex twin pairs (722 monozygotic. MZ, and 794 dizygotic, DZ), their parents, and their spouses’ parents was determined. Concordance for divorce was significantly higher in MZ than DZ twins; this was true overall, in both the male and female samples, for both younger and older twin pairs, and both when the twins’ parents had been divorced and when they had not been divorced. The robustness and magnitude of the MZ-DZ difference in divorce concordance indicates a strong influence of genetic factors in the etiology of divorce. Moreover, family background of both spouses contributed independently to couples’ divorce risk, suggesting that, in many cases, divorce may be largely the result of characteristics the two spouses bring to the union rather than to interaction effects. These results also suggest that the adjustment difficulties seen with some children of divorced parents may be dice to an interaction between genetic and environmental factors rather than environmental influences alone, as is assumed in many theories of divorce’s effects.
If current rates persist, more than a third of all U.S. children will experience the breakup of their parents’ marriages (Bumpass, 1984). The impact of this disturbing demographic trend on offspring functioning is not fully known, but is likely to be complex and diverse. While it is clear that parental divorce is associated with significant psychological disturbance in some children, others appear to be largely unaffected (Emery, 1988).
Furthermore, recent longitudinal surveys in both the United States and Great Britain indicate that the adjustment difficulties seen with the children of divorced parents predate the time of parental separation (Block, Block, & Gjerde, 1986; Cherlin et al., 1991), suggesting that parental conflict rather than separation per se may be the critical variable influencing offspring adjustment. An alternative explanation, which receives some empirical support in the present report, is that parental divorce and offspring adjustment difficulties share, in part, a genetic diathesis. That is, genetic risk factors that contribute to liability for divorce in the parents may also contribute to psychological maladjustment in their children.
Although much is known about divorce’s correlates, relatively little is known about its causes. Significant secular changes in divorce rates substantiate the importance of concomitant cultural changes in, for example, judicial practices (Weitzman, 1981) and female employment (Cherlin, 1981). Nonetheless, it remains difficult to forecast accurately within a given cohort which marriages will or will not end in dissolution. While a previous divorce (Martin & Bumpass, 1989), an early age at marriage (Norton & Glick, 19791, premarital pregnancy (Billy, Landale, & McLaughlin, 1986), and unstable employment (Kelly, 1982) are all related to divorce risk, the vast majority of marriages ending in divorce carry none of these known risk factors. Perhaps the most robust predictor of divorce risk is family background.
In their classic study, Pope and Mueller (1979) summarized results from five large epidemiological surveys and reported that approximately 10% more of the marriages of respondents with divorced parents ended in divorce than the marriages of respondents without divorced parents. Although this association has been consistently replicated (e.g., Glenn & Kramer, 1987), the magnitude of the familial effect has been characterized as “modest” (Kulka & Weingarten, 1979). These previous family studies of divorce have considered the family background of only one of the principals. As we show here, the risk of marital dissolution is a function of the aggregate risk brought to the marriage by both spouses; when family background of both spouses is considered, the importance of familial factors no longer appears modest.
Parent-offspring resemblance for divorce may be a consequence of shared genes or shared experiences; in attempting to account for the intergenerational transmission of divorce, however, researchers have typically focused on environmental etiologies. Pope and Mueller (1979) hypothesized that parent-offspring resemblance for divorce was due to social modeling. Greenberg and Nay (1982) suggested that experiencing the divorce of one’s parents attenuates the social stigma associated with divorce, and thus renders it a more acceptable alternative than if a parental divorce had not been experienced. Although several prominent researchers have speculated about the possible role of genetic factors in divorce (Emery, 1988; Glenn & Kramer, 1987), until now there has been no systematic behavioral genetic study of divorce. In the present study, divorce status was obtained on a sample of 722 monozygotic (MZ) and 794 same-sex dizygotic (DZ) twins, their parents, and their spouses’ parents. Results strongly implicate a genetic influence on divorce risk.
In 1989, a questionnaire on marriage and divorce was mailed to twin members born in 10 of the 20 birth cohorts between 1936 and 1955 of the Minnesota Twin Registry. (Details of recruitment procedures and method of zygosity determination for the Minnesota Twin Registry are given in Lykken, Bouchard, McGue, & Tellegen, 1990.) The questionnaire assessed the respondent’s marital history as well as the marital histories of the respondent’s parents, current and former spouses, and current and former spouses’ parents. We received back completed surveys from 3,316 individual twins aged 34 to 53 years. Approximately 7% of the sample had never married, a rate that did not vary significantly across sex-zygosity classes (Table 1). The present report is based on responses from the 1,516 same-sex twin pairs (275 male MZ pairs, 447 female MZ pairs, 288 male DZ pairs, and 506 female DZ pairs) in which both members of the pair were ever married and returned completed surveys. Respondents reported divorce status of the parents of their current.spouses, if never divorced, and their ex-spouses, if ever divorced. Divorce information on parents-in-law was available for only 2,226 of the twins.
Table 1 provides a demographic characterization of the sample. The sample is largely Caucasian (>97%), reflecting the Minnesota state population for the birth years considered. Overall, 20.3% of the individual twins who had ever married reported that they had divorced at least once; this rate did not vary significantly by sex or zygosity, nor did it differ significantly from an expected divorce rate of 18.4% derived from 1980 U.S. census data for Minnesotans with an age distribution identical to that observed in the twin sample (US. Bureau of the Census, 1983). The four zygosity-sex classes are demographically homogeneous and reported similar rates of divorce among their parents (overall rate of 12.2%) and their spouses’ parents (overall rate of 14.3%). The lower divorce rates in the parental versus the twin generation reflect both secular changes in divorce rates and the fact that divorce is less common among unions that produce children (e.g., in our sample, the divorce rate was two and one half times greater among childless unions than among marriages that did produce children).
FAMILIAL RESEMBLANCE FOR DIVORCE
Table 2 reports pair-wise indices of familial resemblance for divorce, by sex of target individual, by age of target individual, and overall. Indices include (a) the estimated conditional probabilities of target divorce, (b) the odds ratio (i.e., the odds of target divorce when index is divorced divided by the odds of target divorce when index is not divorced; Fienberg, 1978), and (c) the tetrachoric correlation (i.e., the correlation estimated under the assumption that both target and index divorce represent dichotomizations of a normal distribution; Hamdan, 1970). Like Pope and Mueller, we found a significant relation between parental divorce and offspring divorce. Indeed, the increase from 19% to 29% in rate of divorce among children of divorced parents falls well within the range reported by Pope and Mueller in four large samples of Caucasian Americans.
As expected, the effect of spouse’s parents’ divorce is similar to the effect of target parents’ divorce. Further, the effect of parental divorce did not differ significantly from the effect of DZ co-twin divorce, but both these effects were substantially and significantly less than the effect of MZ co-twin divorce. The greater similarity of divorce status for MZ versus DZ twins held for both males and females, for both twins aged 40 years and younger and twins older than 40 years, and both when parents were divorced (MZ odds ratio of 5.31 ± 0.37 significantly greater than DZ odds ratio of 1.76 ± 0.34) and when parents were not divorced (MZ odds ratio of 5.37 ± 0.17 significantly greater than DZ odds ratio of 1.75 ± 0.15). The data strongly suggest absence of a sex effect; in no case did sex of target individual moderate familial resemblance for divorce.
In order to consider simultaneously the influence of all family background data, we used a logistic function of birth cohort, zygosity, co-twin’s divorce status, parents’ divorce status, spouse’s parents’ divorce status, and all two-way interaction terms to predict divorce risk. The logistic model that only included significant effects yielded a nonsignificant Goodness-of-fit test statistic (x² = 59.9 on 54 df, p > .25), indicating that the model adequately represented the observed familial risk data. Except for zygosity, all main effects were significant at p < .05. The interaction term for zygosity by co-twin’s divorce status was also significant, confirming the greater MZ than DZ twin similarity noted above. None of the other possible interaction terms was significant, indicating that spouse’s family background of divorce and respondent’s family background of divorce contributed independently to the prediction of marital dissolution. Figure 1 illustrates this lack of interaction between the two family backgrounds. A couple’s risk of divorce was 8% higher if the respondent’s parents were divorced, regardless of the spouse’s background, and 12% higher if the spouse’s parents were divorced, regardless of the status of the respondent’s parents. The contributions of the two sets of parents, although statistically different, did not differ appreciably from one another, suggesting that a couple’s risk for divorce increased by approximately 10% for each set of divorced parents.
VARIANCE COMPONENTS ESTIMATION
MZ twins are genetically identical, but DZ twins share, on average, only 50% of their genes identical by descent. Therefore, the data summarized in Table 2, showing that the odds of divorce increase nearly sixfold if one has an MZ co-twin who is divorced, but less than twofold if one has a parent or DZ co-twin who is divorced, suggest that divorce risk is strongly associated with genetic factors. This is not to suggest the existence of a single “divorce gene.” The British geneticist Falconer (1965), in his work on diabetes and congenital malformations, proposed a polygenic-threshold model to account for the transmission of multigenically influenced qualitative characters. Under a polygenic-threshold model, an unobserved normally. distributed liability is assumed to underlie expression of the observable categorical response. Like other quantitative phenotypes, liability may be the result of multiple genetic and multiple environmental factors. The qualitative character (here, divorce) is expressed whenever an individual’s liability score exceeds some critical value (i.e., threshold) along the liability continuum.
Using now standard procedures (Rice & Reich, 1985), the proportion of liability variance associated with genetic contribution by one of the spouses was calculated from the twin data  under the following assumptions: (a) divorce is a threshold character; (b) three factors contributed independently and additively to liability variance, the proportion of variance associated with additive genetic factors (h²),with shared environmental factors (c²; i.e., environmental factors shared by both members of a twin pair), and with nonshared environmental factors (e²; i.e., environmental factors not shared by both members of a twin pair); (c) shared environmental factors contributed equally to MZ and DZ twin similarity; and (d) there is no assortative mating for divorce risk (an assumption that is supported by the data summarized in Fig. 1).
1. Only twin data were used in the variance component analysis because twin and parental data could not be compared directly. Because the parents produced offspring, as a group they were at a relatively low risk for
Given these assumptions, the expected tetrachoric correlations are h² + c² for MZ twins and 1/2h² + c² for DZ twins. The likelihood of the observed twin concordances was written as a function of the unknown variance components, which were then estimated simultaneously using the method of maximum likelihood. The proportion of variance in divorce liability associated with the genetic contribution by one of the spouses was estimated as .523 ± .089 for males, 526 ± .067 for females, and 325 ± .054 for the combined male and female sample. In all three cases, the proportion of variance associated with shared environmental factors was estimated at a boundary value of zero, leaving the balance of liability variance to be accounted for by nonshared environmental factors, by far the most important of which must be the contribution to the divorce risk brought to the marriage by the other spouse. Our finding no evidence for shared environmental effects, although seemingly remarkable, is, nonetheless, commonplace in the behavioral genetics literature (Plomin & Daniels, 1987). Nonetheless, it will be important to determine whether absence of shared environmental effects can be replicated using alternative methodologies (e.g., adoption studies).
Although the assumptions that underlie the threshold model calculations have been widely verified for other behavioral characteristics, the validity of these assumptions for divorce has not been assessed directly. Consequently, we view the variance components estimates as preliminary until replicated using alternative strategies. Nonetheless, these analyses suggest that divorce risk is strongly associated with genetic factors, even though the full contribution of family background factors to divorce liability has not been estimated.  If both spouses contributed equally and independently to divorce liability, 50% would represent the theoretical upper bound for the variance contribution of any one partner.
2. It was possible to estimate formally the genetic contribution of only one of the spouses because, for the unions studied, only one of the spouses (i.e., the twin) provided a genetically informative comparison.
The present study reports that divorce is strongly familial and implicates genetic factors as playing a major role in the familial transmission of divorce. Regardless of sex and parental divorce status, the risk of divorce among co-twins of divorced MZ twins was substantially greater than the risk of divorce among co-twins of divorced DZ twins. The consistency of this observation, along with the size of the sample, lends confidence to the conclusion that genetic factors represent an important influence on divorce risk. Although previous researchers have characterized the familial basis of divorce as modest, the present study suggests that this conclusion, based on studies in which family background was incompletely assessed, is unwarranted.
The present study allows us to estimate the accuracy with which divorce could be forecast if complete background information were available on both spouses. The risk for a hypothetical mating between two (unrelated) MZ twins both of whose co-twins and both of whose parents had been divorced can be estimated by adding the separate logistic regression weights associated with having divorced parents and having a divorced MZ co-twin and then doubling the sum.  This calculation yields a predicted divorce rate of 77.5%. The risk for a hypothetical mating between two MZ twins with completely negative family histories of divorce (i.e., both MZ co-twins and both sets of parents not divorced) can be calculated similarly and yields a predicted divorce rate of 5.3%, only 1/15th as large. Although in practice we may only rarely have the necessary background information, in principle divorce is highly predictable from family background factors.
3. It is appropriate to add logistic weights to obtain an overall prediction of risk because there was no evidence from the logistic regression analysis that the family background of the two spouses interacted in predicting their risk of divorce.
The present results add yet another item to the growing list of significant real-world behaviors on which there are substantial genetic influences (Bouchard, Lykken, McGue, Tellegen, & Segal, 1990). We do not mean to minimize the obvious importance of environmental influences on divorce. It would be difficult to account for the substantial secular changes in divorce rates (Glick & Lin, 1986) or, for that matter, why Minnesotans divorce less frequently than people residing in other U.S. states (U.S. Bureau of the Census, 1985) without considering cultural factors. Our data suggest that cultural factors influence the threshold for divorce while, within a given culture, variations in underlying aggregate risk are strongly influenced by genetic factors. These findings suggest that social learning theories that emphasize parental role models to explain the intergenerational transmission of divorce (e.g., Nye & Berardo, 1973) are overly simplistic in not considering genetic mechanisms by which offspring come to resemble their parents; and add to the growing literature demonstrating that genetic factors influence our experiences (Plomin & Bergeman, 1991).
Our data implicate the importance of genetic factors in the prediction of divorce risk; the data do not, however, specify the mechanism that generates this association. Although it is clear that no single factor could account for the etiology of something as heterogeneous as divorce, we hypothesize that the genetic influence on divorce risk is mediated largely by inherited personality characteristics.
A couple is more likely to maintain a long-term relationship if both members hold strong traditional moral convictions than if they do not; they are more likely to remain married if their union is a source of pleasure than if it is a source of distress. There is much evidence relating personal values and individual capacity for happiness to risk of divorce. Thus, divorce is more likely among individuals who drop out of school (e.g., college dropouts) as compared with individuals who complete their educational commitments (e.g., high school graduates who do not pursue college degrees: Norton & Moorman, 1987); among individuals with an unstable pattern of employment as compared with those who are stably employed (Kelly, 1982); among individuals who marry before the age of 20 years as compared with individuals who wait until they attain the societally prescribed age of majority (Norton & Moorman, 1987); among individuals who have experienced an out-of-wedlock pregnancy as compared with those who have not (Billy et al., 1986); among individuals who are rated highly neurotic as compared with those who derive much happiness from their lives (Kelly & Conley, 1987); and, finally, among those who are seen as impulsive as compared with those who are behaviorally controlled (Kelly & Conley, 1987). We predict that twins’ concordance and discordance for divorce will be accounted for largely by similarity and difference in key personality factors. We are presently collecting the data necessary to test this prediction.
Finally, our findings have broad implications for understanding the effects of parental divorce on offspring functioning. Undoubtedly, some children with divorced parents suffer significant emotional, social, and economic hardships not suffered by other children. Our results suggest further that children whose parents divorce differ genetically from children whose parents do not divorce. Some children of divorced parents experience significant academic difficulties and externalizing behavioral disorders, while others appear to be largely unaffected (Barber & Eccles, 1992), heterogeneity which we suspect may be due to genotype-environment interaction. That is, childhood adjustment difficulties may occur only with those offspring who both inherit.a strong genetic diathesis (as reflected by dense family history) and are exposed to significant environmental provocation (e.g., parental conflict, limited economic resources). Identifying the separate influence of genetic and environmental factors and their interactions on the psychological functioning of children of divorce will require, however, application of behavioral genetic designs-designs that have not heretofore been even considered (Grych & Fincham, 1990).