Double impact: what sibling data can tell us about the long-term negative effects of parental divorce

Most prior research on the adverse consequences of parental divorce has analyzed only one child per family. As a result, it is not known whether the same divorce affects siblings differently. We address this issue by analyzing paired sibling data from the 1994 General Social Survey (GSS) and 1994 Survey of American Families (SAF). Both seemingly unrelated regressions and random effects models are used to study the effect of family background on offspring's educational attainment and marital stability.

Double Impact: What Sibling Data Can Tell Us about the Long-Term Negative Effects of Parental Divorce* Nicholas H. Wolfinger, Lori Kowaleski-Jones, Ken R. Department of Family and Consumer Studies, University of Utah abstract: Most prior research on the adverse consequences of parental divorce has analyzed only one child per family. As a result, it is not known whether the same divorce affects siblings differently. We address this issue by analyzing paired sibling data from the 1994 General Social Survey (GSS) and 1994 Survey of American Families (SAF). Both seemingly unrelated regres­sions and random effects models are used to study the effect of family background on offspring's educational attainment and marital stability. Parental divorce adversely affects the educational attainment and the probability of divorce of both children within a sibship; in other words, sib­lings tend to experience the same divorce the same way. However, family structure of origin only accounts for a trivial portion of the shared variance in offspring's educational attainment and marital stability, so parental divorce is only one of many factors determining how offspring fare. These findings were unchanged when controlling for a number of differences both between and within sibships. Also, the negative effects of parental divorce largely do not vary according to re­spondent characteristics. m ain Smith INTRODUCTION Parental divorce can have lasting nega­tive consequences for adult offspring. This finding has been replicated with many nationally representative data sets, including the National Survey of Family Growth (McLanahan and Bumpass, 1988), the Study of Marital Instability Over the Life Course (Amato and Booth, 1991, 1997), the General Social Survey (Glenn and Kramer, 1985, 1987; Wolfin­ger, 1998, 1999, 2003a), the National Survey of Families and Households (Am­ato and Keith, 1991; Wolfinger, 2000, 2003b), and others. These studies share a noteworthy design characteristic. Whether by choice or data limitations, almost all * Correspondence to Nicholas H. Wolfinger, Depart­ment of Family and Consumer Studies, 225 South 1400 East, AEB 228, University of Utah, Salt Lake City, UT 84112-0080 (e-mail: [e-mail]). have analyzed one child per family. This leads to a simple but important question: Do the adverse consequences of growing up in a divorced family vary within sib­ships? This paper explores the negative effects of parental divorce on offspring educational attainment and marital stabil­ity using nationally representative sibling data from the 1994 General Social Survey and the Survey of American Families. Recent scholarly debates on the conse­quences of parental divorce have brought renewed attention to the question of sib­ling differences. The much-publicized re­search of Judith Harris (1995, 1998) con­tends that the perceived consequences of divorce can be actually be attributed to a combination of genetics and children's peer groups. If peers determine how chil­dren fare, there should be considerable variation in how siblings react to parental divorce. The genetic argument is best represented by the work of McGue and 58 Vol. 50, No. 1-2 Negative Effects o f Parental Divorce 59 Lykken, (1992), and Jockin, McGue, and Lykken (1996). Analyses of twins show that the intergenerational transmission of divorce, the increased propensity for the children of divorce to dissolve their own marriages, has a large but not preponder­ant genetic component. This finding sug­gests that the effects of parental divorce should be similar within sibships. Many studies have examined sibling resemblance. Generally their efforts have been unsuccessful-siblings tend to be very different, more so than their shared genetic stock might predict (Dunn and Plomin, 1990). In lieu of genetic explana­tions, differential socialization can explain some of the variation within sibships. In particular, siblings who are emotionally closer to their mothers and better inte­grated in their families tend to be better adjusted psychologically following a di­vorce (Dunn and Plomin, 1990). Although many researchers have ex­amined the negative effects of parental di­vorce on offspring, few have considered how the effects of family structure vary within sibships. Moreover, these studies offer contradictory findings. Mekos, He- therington, and Reiss (1996) compare bi­ological families with stepfamilies com­prising sibships with both homogenous and heterogeneous parentage. Siblings within stepfamilies that shared biological parents are similar to one another to the same extent as respondents from intact families, while step-siblings are far more likely to act differently and be treated dif­ferently by parents. This result implies that divorce affects siblings similarly, al­though such a conclusion should be qual­ified on two grounds. First, it was not the stated purpose of the study. Perhaps re­marriage somehow restores similarity among siblings that had been compro­mised by divorce. Second, Mekos et al. (1996) analyze a nonrepresentative sam­ple of same-sex sibships. Kurdek (1989) also demonstrated a positive association in sibling behavior subsequent to divorce, but analyzed no control group of intact families. Thus it cannot be known whether his results reflect shared influ­ences present prior to the divorce. This limitation also applies to a study by Mon­ahan et al. (1993; see also Buchanan, Maccoby, and Dombusch, 1996). In con­trast to Kurdek (1989) and Mekos et al. (1996) they find little similarity between siblings in divorced families. Studies of twins (Jockin, McGue, and Lykken, 1996; McGue and Lykken, 1992) report considerable similarity in how the chil­dren of divorce fare, a result attributed to shared genetic stock. Given that these studies analyze only twins it is unclear whether their results apply to ordinary sibships. The most thorough test of sib­ling differences was conducted by Sandefur and Wells (1999). Analyzing a nationally representative sample from the National Longitudinal Survey of Youth, they find indirect evidence for similar ef­fects of parental divorce on offspring educational attainment: constraining the effects of family structure on siblings' ed­ucational attainment to be equal did not compromise model fit. One reason the previous sibling stud­ies on the consequences of divorce have reported contradictory findings might be sampling. Only Sandefur and Wells (1999) analyzed a nationally representa­tive sample, while two earlier studies did not include control groups of respondents from intact families. We avoid both of these limitations by analyzing paired sib­ling data from the 1994 General Social Survey and Study of American Families, the only existing nationally representative sample of adult sibling pairs.60 Wolfinger et al. Social Biology Our analyses extend the results of Sandefur and Wells (1999) in four re­spects. First, we examine a sample of adults of all ages, whereas Sandefur and Wells (1999) analyzed National Longitu­dinal Survey of Youth respondents aged 14 to 22 in 1979. A sample thus circum­scribed in terms of age and birth cohort may not reflect the experiences of Ameri­cans in general. Second, we consider two offspring outcomes, described below; Sandefur and Wells (1999) considered only one. Third, we explicitly address whether the effects of parental divorce vary within sibships; Sandefur and Wells (1999) consider this question only indi­rectly. Fourth, we expand on Sandefur and Wells (1999) by evaluating the condi­tions under which siblings may respond differently to the same divorce. RESEARCH PLAN We examine the effects of parental di­vorce on siblings' educational attainment and divorce histories. These were se­lected as dependent variables on the basis of both prior studies and availability. Ed­ucation has been used extensively in prior research on the long-term consequences of parental divorce (e.g., Keith and Fin­lay, 1988; Krein and Beller, 1988; McLanahan, 1985; McLanahan and Sandefur, 1994). Although a large portion of the relationship between parental di­vorce and offspring educational attain­ment can be attributed to differential in­come and residential mobility while growing up (McLanahan and Sandefur, 1994), these factors should not vary sub­stantially within sibships. The second de­pendent variable is respondent divorce history. Numerous studies (e.g., Amato, 1996; Amato and Booth, 1991, 1997; Amato and DeBoer, 2001; Glenn and Kramer, 1987; McLanahan and Bumpass, 1988; Wolfinger, 1999, 2000, 2003b) have shown that people from divorced families are disproportionately likely to end their own marriages, probably be­cause parental divorce reduces marital commitment (Amato and DeBoer, 2001). The GSS contains many other poten­tial outcome variables related to family structure of origin, such as income, occu­pational prestige, and emotional well­being. However, all suffer the liability of being confounded with other conse­quences of parental divorce. For instance, parental divorce increases the likelihood of offspring divorce, which in turn may reduce emotional well-being and, for women, income. Occupational prestige is strongly correlated with education. Given the cross-sectional nature of the data it is impossible to disentangle the causal or­dering of these outcomes. In contrast, ed­ucation and divorce history are dependent variables with reasonably clear causal connections to family structure of origin. Both prior research and common sense provide evidence that the effects of fam­ily structure may vary according to re­spondent demographic characteristics. Glenn and Kramer (1987) found that the transmission of divorce between genera­tions was stronger for whites than for African Americans, and stronger for women than men. McLanahan and Sande­fur (1994) showed similar albeit weaker results for the effect of parental divorce on offspring educational attainment. Al­though the findings on the intergenera- tional transmission of divorce were not upheld by more recent research (Wolfin­ger, 1999, 2000, 2003b), their underlying logic merits further scrutiny. Divorce may have more adverse consequences for fe­male offspring because girls often have more difficulty adapting to life in step-Vol. 50, No. 1-2 Negative Effects o f Parental Divorce 61 families (Hetherington, 1993). Glenn and Kramer (1987) speculated that women contribute more to the maintenance of their own marriages than do men, so the disadvantages conferred by parental di­vorce may undermine marital stability more for women. Both Glenn and Kramer (1985) and McLanahan and Sandefur (1994) argue that parental divorce may have fewer adverse effects on African Americans because divorce is more insti­tutionalized in Black communities than in their white counterparts. Furthermore, ethnicity appears to be an ongoing issue in the divorce literature: a recent review by Amato (2000) concluded that the num­ber of studies employing adequate sam­ples of non-white families is too small to form firm generalizations. Based on this review, testing for ethnic differences should be an important concern when studying the negative effects of parental divorce on offspring. Taken together the arguments consid­ered in this paper imply similar ethnic ef­fects within sibships, but different gender effects. All children in a sibship will share the same ethnic influences. On the other hand, girls and boys may react dif­ferently to the same set of family condi­tions. In addition, we test whether the ef­fects of parental divorce vary with the number of additional siblings present. Additional brothers or sisters may pro­vide stability against the tumult of a di­vorcing family. If this is the case, parental divorce may have fewer negative effects on respondents from large families. Our data allow us to ascertain whether the po­tential palliative effects of additional sib­lings apply equally within sibships. Our overall analytic strategy is to ex­ploit the two important advantages that sibling data offer. Seemingly unrelated regression (SUR) allows us to ascertain whether parental divorce affects siblings differently. Random effects models con­trol for the dependency between siblings (i.e., shared unobserved heterogeneity), thereby permitting us to study the marital stability and educational attainment of each sibling as being conditionally inde­pendent of the other. METHODS Data We analyze data from the 1994 Gen­eral Social Survey (GSS) and the Study of American Families (SAF). A national probability sample of English-speaking households within the continental United States, the GSS has been conducted an­nually or biennially since 1972 (Davis and Smith, 1994). Within each household an adult is randomly selected to be the re­spondent. The SAF is a survey of ran­domly selected siblings of 1994 GSS re­spondents (Hauser and Mare, 1994). The two surveys contain many identical items, so by combining them it is possible to an­alyze sibling pairs. This research design presents two po­tential sources of bias. First, only chil­dren obviously cannot be included in an analysis of sibling pairs. If they are par­ticularly sensitive to the effects of parental divorce, parameter estimates would be biased in comparison to results obtained from a conventional sample. However, ex­ploratory analyses using only the GSS data indicate that parental divorce does not affect only children differently. Second, SAF response rates are negatively corre­lated with parental divorce, resulting in selection bias. Our solution is a two-stage sample selection correction model. We estimate a selection equation where the dependent variable is whether a SAF re­spondent is available for the substantive62 Wolfinger et al. Social Biology regressions (in which education and re­spondent divorce are the dependent vari­ables). For purposes of identification the selection equation must contain at least one independent variable that does not appear in the substantive equation. This requirement is satisfied with two vari­ables. The first is the percentage of each sibship that is female, given that women generally do a disproportionate share of maintaining kinship ties (Rossi and Rossi, 1990). Thus, heavily female sib­ships should be in closer overall contact, yielding higher SAF response rates. The second variable measures whether the GSS respondent has moved since age 16, either within or out of state. All else be­ing equal, sibling propinquity should pro­duce higher response rates. Dummy vari­ables measuring family structure of origin are also included as independent vari­ables in the sclcction equation. Results from the selection equation are then used to generate the inverse of the Mills ratio (\), which is included as a regressor in the substantive equations in order to ac­count for the effects of sample selection. Another data issue concerns the clus­ter design used to collect the initial GSS sample. Data generated by cluster sam­pling but analyzed assuming simple ran­dom sampling produce standard errors that are biased downward, yielding artificially inflated Z-ratios. A common solution is to calculate Huber-White standard errors (e.g., Greene, 1993), which take the clus­tering into account. This correction cannot be implemented readily for the random effects models that we estimate. However, preliminary analyses showed that the clus­tering has no appreciable effect on stan­dard errors calculated from GSS data. Missing data are minimal and there­fore deleted listwise for all variables ex­cept parental education (N = 160). For this item we include an additional dummy that measures whether an observation has missing data for parental education. More sophisticated missing data techniques, such as multiple imputation, do not per­form appreciably better (Paul, McCaffrey, Mason, and Fox, 2002). The sample sizes are 1,100 sibling pairs for analyses of education and 404 sibling pairs for analyses of respondent divorce. The sample size for the latter is smaller for two reasons. First, never-married re­spondents are omitted. Second, the 1994 GSS employed a split ballot design and therefore only half of all respondents were queried about the marital informa­tion needed for our analyses. We repeated the analyses of education using only those cases available for the divorce re­gressions and obtained results similar to those obtained from the full sample. Table 1, showing summary statistics for all variables, reveals various differ­ences between the GSS and SAF subsam­ples that at first glance are suggestive of poor data quality. For instance, 26 per­cent of GSS respondents report the head of their parental family as having less than a high school education, compared to 21 percent of SAF respondents. This probably reflects changes in educational attainment occurring between the times each sibling left the parental household. Residential mobility may account for the different levels of urbanicity reported by siblings (61 percent for GSS respondents, 57 percent for SAF respondents). In con­trast, relatively immutable variables such as Catholicism or total number of siblings show almost no difference between the GSS and SAF subsamples. This provides positive evidence about the quality of the data.Vol. 50, No. 1-2 Negative Effects o f Parental Divorce 63 TABLE 1 Descriptive Statistics for the Combined Sample, GSS, and SAF Subsamples Parental Family Structure Intact 0.76 0.43 0.77 0.42 0.76 0.43 Divorced mother 0.07 0.26 0.07 0.26 0.07 0.26 Remarried mother 0.02 0.15 0.02 0.15 0.02 0.15 Other 0.14 0.35 0.14 0.35 0.15 0.35 Race White 0.88 0.32 0.89 0.31 0.87 0.33 Black 0.08 0.27 0.08 0.27 0.08 0.27 Other 0.04 0.19 0.02 0.15 0.05 0.22 Parental Education Not H.S. graduate 0.24 0.43 0.26 0.44 0.21 0.41 H.S. graduate 0.46 0.50 0.47 0.50 0.46 0.50 Some college 0.03 0.18 0.03 0.18 0.03 0.17 College graduate 0.11 0.32 0.12 0.32 0.11 0.31 Post graduate 0.08 0.26 0.08 0.27 0.07 0.26 Missing data 0.08 0.27 0.04 0.19 0.12 0.32 Other Ever divorced 0.33 0.47 0.36 0.48 0.30 0.46 Years of education 13.67 2.79 13.63 2.82 13.71 2.75 Rural 0.59 0.49 0.61 0.49 0.57 0.49 Catholic 0.30 0.46 0.30 0.46 0.29 0.46 Birth cohort 49.68 15.48 49.61 15.51 49.74 15.45 # of siblings 3.45 2.25 3.45 2.25 3.45 2.25 Age - age 1st wed 26.81 15.00 26.87 14.83 24.74 15.18 Age difference between sibs 5.17 3.96 5.17 3.97 5.17 3.96 Male 0.48 0.50 0.49 0.50 0.46 0.50 Sibs are same sex 0.51 0.50 0.51 0.50 0.51 0.50 Total GSS SAF Mean SD Mean SD Mean SD Dependent Variables The first dependent variable is years of completed education, ranging from zero to twenty. Education is treated as continuous to provide greater variability. The second dependent variable is a dichotomy measur­ing whether or not a respondent has ever been divorced. The coding for these and other variables appears in Appendix A. Independent Variables The GSS and SAF include two items that measure the structure of respondents' families of origin. Respondents were first queried about household composition at the age of 16. If respondents were not living with both biological parents, a sec­ond item ascertained the reason. Eighty- five percent of respondents report three varieties of family structure: intact two- parent families, mother-only families re­sulting from divorce or separation, and mother/step-father families resulting from divorce or separation. Each of the three categories is coded as a dummy variable, with respondents from intact families as the reference category.64 Wolfinger et al. Social Biology An additional dummy is coded for the 15 percent of respondents not falling into the three aforementioned categories. These comprise a variety of other family structures, as well as respondents whose living situations at age 16 were the prod­uct of parental military service, parental incarceration, or parental death. These re­spondents are pooled and included for an­alytic reasons. Too few people from this diverse group hail from any particular family background to produce individu­ally meaningful parameter estimates, but omission of any respondent falling into the "other" category necessarily means omitting both siblings from the analysis. Doing so could bias the sample, given that individual siblings from "other" fam­ily backgrounds are disproportionately likely to have come from divorced fami­lies. Retaining "other" respondents and identifying them with their own dummy variable solves this problem. However, these respondents comprise such hetero­geneous family backgrounds that it is im­possible to draw substantive conclusions about them. Other independent variables fall into three categories. First, we control for a variety of demographic differences be­tween sibships, including parental educa­tion, race, Catholicism at age 16, number of siblings, and rural vs. urban origins. These items allow us to understand whether within-sibship variation on the dependent variables can be explained by sociodemographic differences between families. Two additional variables, re­spondent education and age at marriage, could potentially be included in analyses of marital stability. They are omitted to preserve consistency with the analyses of educational attainment; moreover, prior research shows that other factors are largely responsible for the relationship between parental divorce and offspring divorce (Amato, 1996; Amato and De­Boer, 2001; Wolfinger, 1999, 2000, 2003b). A second set of independent vari­ables includes characteristics that may vary within sibships. These include re­spondent gender, whether or not siblings are of the same gender, and the absolute age difference between siblings. In partic­ular, it is important to control for respon­dent gender when divorce is the depen­dent variable because men misrepresent their own marital histories more often than women (Bumpass, Martin, and Sweet, 1991). Additional within-sibship measures would be useful, but are not available in the GSS/SAF. Third, we con­trol for birth cohort when education is the dependent variable, given that both aver­age educational attainment and the inci­dence of parental divorce have changed over time. An additional variable is necessary for models predicting respondent divorce. Divorce can be right-censored; that is, re­spondents may still dissolve their mar­riages after the GSS/SAF interviews. Sur­vival modeling is preferable for this sort of situation, but the GSS/SAF do not pro­vide information on divorce timing and therefore lack adequate data for survival modeling. To contend with right censor­ing we use a procedure employed by other analysts of GSS data (Glenn and Kramer, 1987; Wolfinger, 1999). A new variable was constructed by subtracting age at first marriage from current age. A lowess model (Cleveland, Grosse, and Shyu, 1992) revealed a curvilinear and non-monotonic relationship (not shown) between respondent divorce and (AGE minus AGEWED), so we modeled the difference between them as a quadratic.Vol. 50, No. 1-2 Negative Effects o f Parental Divorce 65 Including this variable in regression equations should largely ameliorate the right censoring bias by modeling the du­ration of exposure to the hazard of di­vorce. The two terms of this quadratic are highly correlated with respondent birth cohort (r = 0.94 for each), so it is not pos­sible to include both variables in the models predicting divorce. One liability of the GSS/SAF is the absence of data on respondents prior to parental divorce, given that people from divorced families often exhibit signs of poor well-being even before their par­ents' marriages end (Cherlin, Chase- Lansdale, and Kieman, 1995; Cherlin, Chase-Lansdale, and McRae, 1998; Cher­lin et al., 1991; Furstenberg and Teitler, 1994; Kieman and Cherlin, 1999). How­ever, these studies also show that parental divorce affects children irrespective of pre-divorce well-being. The upshot is mostly rhetorical: parental divorce should be thought of as a process, representing an accretion of events, rather than a sim­ple demographic transition. Analyses Seemingly Unrelated Regression (SUR) Analysis. We use SUR models to see if the effects of parental divorce vary within sibships. In the analysis of educa­tional attainment siblings are randomly assigned to one of two equations that as­sume the usual ordinary least squares specification: (la) Education( 1) = /301 -I- /3,^(ParentalDivorce) + /32| {Family) + fi^Individual) + e] (16) Educaiion(2) = (3q2 + pl2(ParentalDivorce) ~t" fi22(Family) + fi ^(Individual) + e2 where Equations la and lb describe the relationship between education and fam­ily structure for each of the two siblings. The vectors "Family" and "Individual" respectively represent the familial vari­ables and person-specific independent variables. SUR models provide more effi­cient estimates of parameters and stan­dard errors when the error terms are al­lowed to be correlated across equations la and lb (i.e., r[1,2)]^0). This is as­sumed to be the case because siblings are affected by the same unmeasured vari­ables. The SUR models also allow formal testing of the hypothesis Ho: Pu = P12. A comparable SUR model is used for the sibling analysis of marital stability, in which divorce is treated as a dichotomous dependent variable. Siblings are randomly assigned to one of two probit equations: (2a) Divorce(l)* = /301 + Pu(ParentalDivorce) + fi2](Farnily) + /3 ^(Individual) + e] (2b) Divorce(2)* = /?02 + fil2(ParentalDivorce) + {322(Family) + (3^(Individual) + e2 where Divorcedand Divorce(2)* are unobserved latent dependent variables. We are able to measure only the observ­able dichotomies Divorce(l) and Di- vorce(2), where Divorce(l) = 1 if Di- vorce(l)* > 0 and Divorce(l) = 0 otherwise; a similar specification applies to Divorce(2). The model assumes that E[£j] = E[e2] = 0 and Var[e,] = Var[e2] - 1. We estimate p = Cov[£j,£2], which measures the degree to which the residual for Sibling 1 is associated with that for Sibling 2. This SUR model also permits us to test the hypothesis Ho: Pn = P12- A more straightforward way to esti­mate the SUR models would have been to assign respondents from each subsam­ple (GSS, SAF) to a separate equation. However, this is problematic because any differences across equations could simply be a product of the sample selection bias noted earlier. Our solution, above and66 Wolfinger et al. Social Biology beyond the two-stage selection correction described earlier, is to randomly assign each sibling in a pair to one of the two equations. Random Effects Models. Another ap­proach to studying the effects of parental divorce on the subsequent well-being of adult children is to use random effects models. Random effects models make the simple assumption that regression param­eters will vary between sibling pairs due to unmeasured differences across the sample of sibships. The sibship-specific regression coefficients are treated as a "random" sample of coefficients drawn from some specified distribution; in our application, the distribution of random ef­fects is assumed to be normally distrib­uted. The random effects model assumes the following general form: (3) Yit = a + Xu/3 + Zt+ ek where Y is years of education completed for individual i of sibship t, X is a vector of dummy variables measuring family structure for individual i of sibship t, Z is a family-specific residual shared by both members in a sib pair, and e is a residual term that has a zero mean, is uncorrelated with X and Z, and has a constant vari­ance. The random effects model can be estimated based on the following: (4) y„-p7, = o -p)« + <x« - + to - p)z, where p = (t\ 1((t\ + <x^). If p = 0, then cr2z = 0 and Z( = 0 for all sib pairs and the model reduces to an OLS regression. If p = 1, then a2s - 0 and eit = 0, which means that there is no random error in the model (i.e., R2 = 1). Values of p therefore measure the proportion of unobserved variation in the dependent variables that is shared by siblings within sibships. Comparable random effects logistic regression models are estimated where divorce is the outcome. These models have the following functional form: Z,)=l+exp(^+Z|) </)■„= I and f«„g+Z.>- !-(, </>, = 0 where F(.) is the cumulative probability distribution, Xit is again a vector of dummy variables measuring family struc­ture for individual i of sibship t, and Z( is a family-specific effect shared by both members in sibship t. In this model, Z( ~ N(0,tr2z). The statistic p = ct2z / ( 1 + cr2z) measures the proportion of total variance that is explained by the sibship- level component. These models are esti­mated using generalized estimating equa­tions (Liang and Zeger, 1993). Random effects models offer three contributions. First, when analyzing sib­ling pairs or other clustered data random effects models yield more efficient esti­mates than do ordinary least squares. Sec­ond, unlike the SURs they allow us to generate a single set of parameter esti­mates for the entire sample, rather than results based on two subsamples. This should provide a more accurate appraisal of the effects of parental divorce, particu­larly when respondent divorce history is the outcome variable and the subsample sizes become somewhat small (N = 404 for each of the two subsamples in the SUR models). Finally, random effects models allow us to ascertain the extent of within-sibship variation that can be ac­counted for by differences in parental family structure.Vol. 50, No. 1-2 Negative Effects o f Parental Divorce 67 RESULTS Seemingly Unrelated Regression Models Table 2 presents results from the seemingly unrelated regression (SUR) analyses of respondent education on fam­ily background and other variables. Sib­ling A and Sibling B refer to the random­ized assignment of siblings into two subsets. As predicted, all models show that parental divorce has a negative effect on educational attainment, with step- parenting consistently having stronger ef­fects than single-mother parenting. Also, the estimated effects of family structure do not change substantially from Models 1 to 2, suggesting that sociodemographic differences between respondents cannot account for the relationship between fam­ily structure of origin and educational at­tainment. Model 3 adds controls for within-sibship factors, including age dif­ferences and the gender composition of the sibship. This also had little effect on the coefficients measuring the effects of family structure on offspring education. The SUR models allow us to ascertain whether parental divorce affects siblings differently. For each regression model we test the hypothesis bA = bB for the effects of single-mother parenting and step- parenting on offspring educational attain­ment. The results of these tests are shown in the columns to the right of each regres­sion model. Although the magnitude of some coefficients and their corresponding significance tests vary between Equations A and B none of these differences are sta­tistically significant, which indicates no meaningful within-sibship variation in the effects of parental divorce on off­spring education. In other words, with re­spect to educational attainment siblings react to the same divorce the same way. Taking all models into account, divorced- mother parenting on average reduces off­spring educational attainment by up to three-fourths of a year (Models 2.B and 3.B). The figures for remarried divorced mothers are larger, with average educa­tional attainment declining by up to two years (Models 2.A and 3.A). A similar story emerges when consid­ering how parental divorce affects the marital stability of adult offspring. Table 3 shows the results of the seemingly unre­lated bivariate probit analyses of respon­dent divorce history on family structure of origin and other independent variables. In all of the B equations divorced-mother parenting significantly increases the like­lihood that respondents dissolve their own marriages. Although the correspond­ing coefficients in the A equations are small and statistically insignificant, the differences across equations in the mag­nitude of the coefficients are themselves not significant. Perhaps the coefficients vary across models because of the re­duced size of each sib-specific subsam­ple. In any event, the equality of coeffi­cients across models in Table 3 offers the same conclusion as do the SUR analyses of respondent education: the effects of parental divorce do not vary within the same family. Moreover, controlling for differences both within and between sibships does not affect the relationship between parental divorce and offspring divorce. One surprising result suggested by Table 3 concerns the lack of a relation­ship between step-parenting and respon­dent marital stability. This is contrary to previous research (e.g., Glenn and Kramer, 1987; Wolfinger, 2000), which showed especially strong rates of divorceSeemingly Unrklated Regression Estimates of Respondent Education on Family Structure of Origin and Other Variables TABLE 2 Model i Model 2 Model 3 Sibling A Sibling B b SE b SE ^A " Sibling A Sibling B b SE b SE bA - bB Sibling A Sibling B b SE b SE bA = bB Divorced mother -0.42 0.32 -0.79** 0.30 n.s. -0.18 0.30 -0.70* 0.29 n.s. -0.19 0.30 -0.70* 0.29 n.s. Remarried mother -1.55** 0.51 -1.26* 0.55 n.s. -1.97*** 0.49 -1.39** 0.53 n.s. -1.99*** 0.49 -1.41** 0.53 n.s. Intact family - - - - - - - - - - - - Other -1.50** 0.27 0.66* 0.27 _1 33*** 0.26 -0.28 0.26 _1 32*** 0.26 -0.30 0.27 Black _ _ - - -0.45 0.29 -0.06 0.28 -0.48 0.29 -0.07 0.28 White - - - - - - - - - - - - Other - - - - -0.36 0.36 0.34 0.39 -0.36 0.36 -0.35 0.40 Parental education Not a H.S. grad H.S. graduate - - - - 0.80*** 0.19 ]]7*** 0.19 0.81*** 0.19 1.18*** 0.19 Junior college - - - - 1.55*** 0.42 2 47**+ 0.43 1.59*** 0.42 2.48*** 0.44 College grad - - - - 2.15*** 0.28 2 49*** 0.27 2.15*** 0.28 2.50*** 0.27 Post graduate - - - - 3 32*** 0.31 2 83*** 0.31 3 34*** 0.31 2.85*** 0.31 Data missing - - - - -0.54+ 0.31 -0.22 0.29 -0.53+ 0.31 -0.22 0.29 Birth cohort 0.03*** 0.01 0.04*** 0.01 0.01 0.01 o.or 0.005 0.01 0.01 0.01* 0.005 Rural - - - - -0.51** 0.15 -0.35* 0.15 -0.51** 0.15 -0.35* 0.15 Catholic - - - - 0.54** 0.17 0.35* 0.16 0.54** 0.17 0.34* 0.16 # of siblings - - - - -0.13*** 0.04 -0 19*** 0.03 -0 13*** 0.04 -0.20*** 0.04 Male - - - - - - - - 0.16 0.14 0.17 0.14 Age difference - - - - - - - - 0.01 0.02 0.02 0.02 between sibs Same sex siblings - - - - - - - - 0.26+ 0.15 0.13 0.14 -2.34 1.93 -0.81 1.94 -4.28* 1.80 -1.90 1.82 -4.27* 1.81 -1.99 -1.83 Constant 16.40*** 0.86 15.91*** 0.87 15.72*** 0.85 14.56*** 0.87 15.46*** 0.86 14.38*** 0.88 + p < 0.10 ; * p < 0.05 :** p < 0.01 ; *'+* p < 0.001 Wolfinger et al. Social BiologyTABLE 3 Seemingly Unrelated Bivariate Probit Estimates of Respondent Divorce History on Family Structure of Origin and Other Variables Model i Model 2 Model 3 Sibling A Sibling B b SE b SE Sibling A Sibling B b SE b SE Sibling A Sibling B b SE b SE X? II < XI Divorced mother Remarried 0.07 0.28 0.55* 0.27 n.s. 0.09 0.29 0.66* 0.28 n.s. 0.04 0.29 0.66* 0.28 n.s. mother -0.26 0.47 0.04 0.59 n.s. -0.23 0.48 0.09 0.62 n.s. -0.32 0.47 0.20 0.62 n.s. Intact family - - - - - - - - - - - - Other -0.28 0.24 0.32 0.23 -0.29 0.26 0.42+ 0.25 -0.35 0.26 0.46 0.25+ Black __ - - __ -0.13 0.33 -0.17 0.33 -0.13 0.33 -0.17 0.33 White - - - - - - - - - - - - Other - - - - 0.47 0.39 0.39 0.40 0.53 0.39 0.40 0.40 Parental education Not a H.S. grad - - - - - - - - - - - - H.S. graduate - - - - 0.06 0.17 0.30+ 0.17 0.07 0.17 0.27 0.17 Junior college - - - - 0.34 0.37 0.55 0.42 0.38 0.37 0.54 0.42 College grad - - - - -0.20 0.30 -0.23 0.29 -0.18 0.30 -0.23 0.29 Post graduate - - - - 0.34 0.32 0.49 0.33 0.38 0.32 0.43 0.34 Data missing - - - - 0.12 0.28 0.13 0.27 0.19 0.28 0.11 0.27 Age-age Is1 wed 0.09*** 0.02 0.07*** 0.02 0.09*** 0.02 0.07*** 0.02 0.10*** 0.02 0.07*** 0.02 Age-age Is' wed2 -0.002*** 0.0003 -0.001*** 0.0003 -0.002*** 0.0003 -0.001*** 0.0003 -0.002*** 0.0003 -0.001*** 0.0003 Rural - - - - 0.05 0.15 -0.12 0.15 0.05 0.15 -0.11 0.15 Catholic - - - - 0.03 0.15 -0.02 0.16 0.05 0.15 -0.03 0.16 # of siblings - - - - 0.002 0.03 -0.06+ 0.03 -0.01 0.03 -0.05 0.03 Male - - - - - - - - 0.13 0.14 -0.16 0.14 Age difference between sibs - - - - - - - - 0.02 0.02 -0.03 0.02 Same sex siblings - - - - - - - - 0.01 0.15 0.08 0.14 X -2.49 1.61 -0.88 1.66 -2.48 1.68 -0.96 1.73 -2.73 1.69 -0.63 1.75 Constant -0.25 0.73 -1.06 0.76 -0.43 0.80 -1.00 0.82 -0.56 0.81 -1.02 0.84 ' p < 0.10 ; * p < 0.05 ;** p < 0.01 ;**«<>< 0.001 Vol. 50, No. 1-2 Negative Effects of Parental Divorce70 Wolfinger et al. Social Biology transmission for children reared by step­parents. Although puzzling, this finding is consistent within sibships and therefore accords with the other results presented here. It may be the product of a small sample size, given that only 2 percent of GSS/SAF respondents hail from step­parent families. Alternately, it may reflect the historical weakening of the transmis­sion of divorce between generations (Wolfinger, 1999). In additional models (not shown), we explore whether the negative effects of parental divorce vary by respondent gender, ethnicity, or presence of siblings. These analyses were only conducted for offspring educational attainment; prelimi­nary analyses showed that the sample size was inadequate for the seemingly unrelated bivariate probit models of offspring marital instability. The analyses of educational attainment reveal only one statistically significant interaction, be­tween respondent family structure back­ground and offspring gender, ethnicity, and siblings: respondents of "other" eth­nic background (neither African Ameri­can nor white) and "other" parentage (not from an intact family, a divorced single­parent family, nor a divorced step-family) have significantly higher levels of educa­tional attainment. For reasons described earlier in the paper, it is impossible to in­terpret results associated with "other" parentage. Based on these results, the negative ef­fects of parental divorce do not appear to vary by offspring ethnicity, gender, or presence of siblings, either between or within sibships. However, the random ef­fects results presented below do not di­vide the sample between two equations and therefore offer a better test of demo­graphic differences in siblings' reactions to parental divorce. Table 4 presents random effects esti­mates for the regressions of educational attainment on parental divorce and other independent variables. We first estimate a baseline model (1), in which birth cohort and the sample selection parameter (X) are the only independent variables. This is compared to Model 2, which includes the family structure variables, in order to ascertain how much within-sibship varia­tion in educational attainment can be explained by parental family structure. Model 3 includes a vector of respondent characteristics that do not vary within sibships, while Model 4 contains mea­sures that may vary within sibships. Model 2 shows a significant negative association between family background and educational attainment. Furthermore, the results are consistent in magnitude with those produced by the SUR analy­ses. Models 3 and 4 respectively control for differences between and within sib­ships. These results are also similar to their SUR counterparts, with the control variables having little effect on the rela­tionship between family of origin and off­spring educational attainment. The other noteworthy result concerns the statistically significant rho (p) pa­rameters, which indicate the amount of within-sibship variation in educational at­tainment not explained by the indepen­dent variables. Parental divorce accounts for very little of this variation: p changes slightly, from 0.47 to 0.46, between Model 1, which omits the family struc­ture variables, and Model 2. Although differences between sibships further re­duce p (Model 3), observed within- sibship differences do not (Model 4). These results show that factors aside from family structure play an important role in R a n d o m E ffects M odels Vol. 50, No. 1-2 Negative Effects of Parental Divorce 71 TABLE 4 Random Effects Estimates of Education on Family Structure of Origin and Other Variables Model j Model 2 Model 3 Model 4 b SE b SE b SE b SE Divorced mother _ - -0.65** 0.24 -0.50* 0.23 -0.51* 0.23 Remarried mother -]44*** 0.40 -1.67*** 0.37 -1.70*** 0.37 Intact family - - - - - - - - Other - - -1.08*** 0.20 -0.81*** 0.19 -0.82*** 0.19 Black - - - - -0.25 0.23 -0.26 0.23 White - - - - - - - - Other - - - - 0.01 0.29 0.02 0.29 Parental education Not a H.S. grad - - - - - - - - H.S. graduate - - - - 0.97*** 0.15 0 0.15 Junior college - - - - 1.95*** 0.32 1.98*** 0.32 College grad - - - - 2 28*** 0.21 2 29*** 0.21 Post graduate - - - - 3 03*** 0.24 3.05*** 0.24 Data missing - - - - -0.37+ 0.21 -0.37+ 0.21 Birth cohort 0.04*** 0.004 0.04*** 0.004 0.01 + 0.004 0.0 r 0.004 Rural - - - - -0.41*** 0.12 -0 42*** 0.12 Catholic - - - - 0.44** 0.13 0.44** 0.13 # of siblings - - - - -0.16*** 0.03 -0.17*** 0.03 Male - - - - - - 0.18+ 0.10 Age difference - - - - - - 0.02 0.02 between sibs Same sex siblings - - - - - - 0.19 0.12 X 3.84** 1.28 -1.65 1.57 -3.22* 1.42 -3.27* 1.43 P Constant 0.47*** 13.71*** 0.56 0.46*** 16.19*** 0.70 0.33*** 15.24*** 0.67 0.33*** 15.02*** 0.68 *p < 0.10 ; V < 0.05 ;**()< 0.01 ;***;)< 0.001 explaining sibling variability in educa­tional attainment. Table 5 presents the results of random effects logistic regressions of offspring marital stability on family structure of origin and other independent variables. The results accord with those already presented, so they will be described only briefly. Consistent with the SUR analyses of marital stability (Table 3), divorced- mother parenting increases the chances of offspring divorce whereas step-parenting does not. As for the random effects analy­ses of offspring education, family struc­ture of origin accounts for only a small proportion of the variation in how sib­lings fare. Finally, controlling for observ­able differences both between and within sibships did not affect the relationship between parental divorce and offspring divorce. Additional random effects models (not shown) test whether the consequences of parental divorce vary by respondent gender, ethnicity, or presence of siblings. These models have the advantage, in comparison to the seemingly unrelated regressions, of larger sample sizes and therefore provide better tests of potential demographic variation in the negative ef­fects of parental divorce, both within and between sibships. However, very little72 Wolfinger et al. Social Biology Random Effects Logit Estimates of Respondent Divorce History on Family Structure of Origin and Other Variables TABLE 5 Model j Model 2 Model 3 Model 4 b SE b SE b SE b SE Divorced mother Remarried - - 0.58 0.38 0.70+ 0.39 0.71 + 0.39 mother - - -0.23 0.71 -0.23 0.71 -0.21 0.71 Intact family - - - - - - - - Other - - 0.02 0.32 0.08 0.34 0.09 0.34 Black - - - - -0.27 0.46 -0.28 0.46 White - - - - - - - - Other - - - - 0.75 0.51 0.75 0.51 Parental education Not a H.S. grad - - - - - - - - H.S. graduate - - - - 0.32 0,23 0.32 0.23 Junior college - - - - 0.88+ 0.52 0.89+ 0.52 College grad - - - - -0.42 0.41 -0.42 0.42 Post graduate - - - - 0.79+ 0.46 0.79+ 0.49 Data missing - - - - 0.20 0.37 0.20 0.37 Age - age 1st wed 0.16*** 0.03 0.16*** 0.03 0.16*** 0.03 0.16*** 0.03 Age - age 1sl wed2 -0.002*** 0.0004 -0.003*** 0.0005 -0.002*** 0.0004 -0.002*** 0.00 Rural - - - - -0.06 0.20 -0.06 0.20 Catholic - - - - -0.04 0.22 0.03 0.23 # of siblings - - - - -0.05 0.04 -0.05 0.05 Male - - - - - - 0.00 0.18 Age difference between sibs - - - - - - -0.01 0.02 Same sex siblings - - - - - - 0.07 0.19 X P -4.00* 0.47** 1.87 -3.46 0.46** 2.32 -3.62 0.43* 2.36 -3.50 0.43* 2.37 Constant -0.93 0.85 -1.19 1.05 -1.25 1.11 -1.31 1.13 *p < 0.10 ; • p <0.05 : ** p <0.01 ; *** p <0.001 variation was observed. The only statisti­cally significant interaction concerns edu­cational attainment: male step-children from divorced families have lower levels of educational attainment than do their female counterparts. Since this result ac­cords with neither prior research (e.g., Hetherington, 1993) nor our other analy­ses, we are cautious about attaching too much importance to it. Moreover, Type II error may be responsible given the large number of interactions we test. ■> DISCUSSION - This paper has examined the effects of parental divorce on offspring marital sta­bility and educational attainment using nationally representative sibling data from the General Social Survey (GSS) and the Survey of American Families (SAF). Our primary finding is straight­forward: parental divorce affects siblings similarly. Moreover, the effects of parental divorce are uniformly negative. PeopleVol. 50, No. 1-2 Negative Effects of Parental Divorce 73 from divorced families complete fewer years of school and are more likely to dis­solve their own marriages than are people from intact families. Step-parenting fur­ther reduces offspring educational attain­ment but offsets the negative effects of parental divorce on offspring marital sta­bility. Controlling for a variety of differ­ences both within and between sibships does not attenuate the consequences of parental divorce, and we observed almost no variation according to respondent sex, race, or presence of siblings. Given its gravity as a life course tran­sition, it makes sense parental divorce affects siblings similarly. Divorce often disrupts almost every family routine (Wallerstein and Kelly, 1980), so it is not surprising that both members of a sibship may be affected negatively. Our results are also not surprising given the robust negative effects of divorce chronicled by earlier research. The children of divorce are no more or less likely to graduate from high school on time if they spend many or few years in a single-parent family, are old or young when first expe­riencing parental divorce, or live through multiple divorces (McLanahan and Sandefur, 1994). Similarly, recent studies show that the intergenerational trans­mission of divorce does not vary by sex, race, religion, or gender (Wolfinger, 1999, 2000, 2003b). Given that the con­sequences of parental divorce are so ro­bust in other ways, it makes sense that all children within a family should be af­fected the same way. Our results also ac­cord with what we know about the mech­anisms responsible for the negative effects of parental divorce. Divorce af­fects offspring educational attainment largely through income and residential mobility (McLanahan and Sandefur, 1994), while the intergenerational trans­mission of divorce can probably be attrib­uted to the lessons children learn about marital commitment (Amato and DeBoer, 2001). None of this should vary within sibships, so it is logical that siblings are affected similarly. Although parental divorce can have large negative effects on offspring, it is only one of numerous factors that may do so. This is demonstrated by the fact that parental divorce accounts for only a small portion of the shared variance in siblings' educational attainment and propensity for divorce. Numerous other factors, most of them probably unmeasurable, also make a difference. This is why siblings can be so different even in the face of the large and homogeneous effects of parental di­vorce, a result that is in accordance with Andrew Cherlin's (1999) recent presiden­tial Address to the Population Association of America. Parental divorce sometimes has negative effects on offspring, but these are hardly inevitable given the num­ber of other conditions affecting how children fare. One limitation of this study concerns the data. The random effects models failed to account for considerable within- sibship variance in educational attain­ment and divorce history. Data on family processes and psychological well-being, both before and after the divorce, would permit more insight into whether the ef­fects of family structure ever vary within sibships. Measures of pre-divorce well­being would be particularly useful, given that children's problems may be evident long before the divorce itself. In addition, the GSS/SAF measure of parental divorce, family structure at age 16, is relatively crude compared to those available in other surveys. More precise information might74 Wolfinger et al. Social Biology yield useful information on how siblings respond to divorce. Although we have examined parental divorce as a socio-environmental life event among children, the sequela from parental family structure to offspring outcomes are affected by both social and biological fac­tors. Our data do not allow us to measure the separate contributions of environmen­tal, genetic, and gene-environment interac­tions that might account for why siblings of divorced parents are at (a common) greater risk of divorce and reduced edu­cational attainment compared to siblings from intact families. We cannot rule out the presence of heritable factors such as depression and personality traits that may be shared among siblings (and indeed their parents), thereby predisposing them to similar adverse consequences ostensi­bly resulting from life events such as parental divorce (Jockin, McGue, and Lykken, 1996; McGue and Lykken, 1992). Further research is warranted to identify these heritable, shared biological factors. Even with these limitations, the pres­ent study contributes to the literature by analyzing a large and nationally represen­tative data set to provide findings on the nature of the long-term effects of divorce on adult children. Future research should attempt to discover why parental divorce affects siblings similarly. In the mean­time, our results may be of interest to mental health professionals and others working with children adversely affected by divorce. Race Religion Family structure at age 16 Education of head of respondent parental family Urbanicity Sex Education Respondent divorce history Coded 0 if no, 1 if yes APPENDIX A Coding of Variables Set of two dichotomous indicators, each coded 1 if re­spondent is: Black, non-white/non-Black; white is the reference category Coded 1 if Catholic, 0 if not Catholic Set of three dichotomous indicators, each coded 1 if re­spondent family of origin was headed by a divorced sin­gle mother, a remarried divorced mother, or other; intact family is the reference category Set of five dichotomous indicators, each coded 1 if family head: was not a high school graduate, had a ju­nior college graduate, was a four-year college graduate, has a post-graduate degree, or if data are missing; not a high school graduate is the reference category. Coded 0 if respondent lived in a city 50,000 or more at age 16 or a suburb of a larger city, coded 1 if respon­dent lived in a town of under 50,000 or in a rural area Coded 0 if female, 1 if male Continuous variable measuring years of completed educationVol. 50, No. 1-2 Negative Effects o f Parental Divorce 75 Age-age 1st wed Catholicism Siblings are same sex Absolute age difference between siblings Number of siblings Birth cohort Continuous variable Coded 1 if Catholic, 0 if not Catholic Coded 0 if no, 1 if yes Continuous variable Continuous variable Continuous variable ACKNOWLEDGEMENTS We thank Robert Schoen and the Utah Population Group for helpful comments, and Al Hernandez for research assistance. An ear­lier version of this paper was presented at the 2001 annual meeting of the Population Asso­ciation of America, Washington. Wolfinger's participation in this research was supported by the Bireley Foundation. REFERENCES Amato, P. R. 1996. Explaining the intergenerational transmission of divorce. Journal of Marriage and the Family 58:628-640. ---------. 2000. The consequences of divorce for adults and children. Journal of Marriage and the Family 62:1269-1288. Amato, P. R., and A. Booth. 1991. Consequences of parental divorce and marital unhappiness for adult well-being. Social Forces 69:905-914. ---------. 1997. A generation at risk: Growing up in an era of family upheaval. Cambridge, Massa­chusetts: Harvard University Press. Amato, P. R., and B. Keith. 1991. Separation from a parent during childhood and adult socioeco­nomic attainment. Social Forces 70:187-206. Amato, P. R., and D. DeBoer. 2001. The transmis­sion of divorce across generations: Relationship skills or commitment to marriage? Journal of Marriage and the Family 63:1038-1051. Buchanan, C. M., E. E. Maccoby, and S. M. Dornbusch. 1996. Adolescents after divorce. Cambridge, Massachusetts: Harvard University Press. Bumpass, L. L., T. Castro Martin, and J. A. Sweet. 1991. The impact of family background and early marital factors on marital disruption. Journal of Family Issues 12:22-42. Chase-Lansdale, P. L., A. J. Cherlin, and K. Kiernam. 1995. The long-term effects of parental divorce on the mental health of young adults: A developmental perspective. Child De­velopment 66:1614-1634. Cherlin, A. J. 1999. Going to extremes: Family structure, children's well-being, and social sci­ence. Demography 36:421-428. Cherlin, A. J., F. F. Furstenberg, Jr., P. L. Chase-Lansdale, K. E. Kiernan, P. K. Robins, D. R. Morrison, and J. O. Teitler. 1991. Longitudinal studies of effects of divorce on children in Great Britain and the United States. Science 252:1386-1389. Cherlin, A. J., K. E. Kiernan, and P. L. Chase- Lansdale. 1995. Parental divorce in childhood and demographic outcomes in young adulthood. Demography 32:299-318. Cherlin, A. J., P. L. Chase-Lansdale, and C. McRae. 1998. Effects of parental divorce on mental health throughout the life course. Ameri­can Sociological Review 63:239-249. Cleveland, W. S., E. Grosse, and W. M. Shyu. 1992. Local regression models. In J. M. Cham­bers and T. J. Hastie, eds., Statistical Models in S, pp. 309-376. Pacific Grove, CA: Wadsworth & Brooks. Davis, J. A., and T. W. Smith. 1994. The general social surveys: Cumulative codebook, 1972- 1994. Chicago: National Opinion Research Center. Dunn, J., and R. Plomin. 1990. Why siblings are so different. New York: Basic Books. Furstenberg, F. F., Jr., and J. O. Teitler. 1994. Reconsidering the effects of marital disruption: What happens to children of divorce in early adulthood? Journal of Family Issues 15:173-190. Glenn, N. D., and K. B. Kramer. 1985. The psy­chological well-being of adult children of di­vorce. Journal of Marriage and the Family 47:905-912. ---------. 1987. The marriages and divorces of the children of divorce. Journal of Marriage and the Family 49:811-825. Greene, W. H. 1993. Econometric analysis (2nd ed.). Englewood Cliffs, N.J.: Prentice Hall. Harris, J. R. 1995. Where is the child's environ­ment? A group socialization theory of develop­ment. Psychological Review 102:458-489.76 Wolfinger et al. Social Biology ---------. 1998. The nurture assumption: Why chil­dren turn out the way they do; parents matter less than you think and peers matter more. New York: The Free Press. Hauser, R. M., and R. D. Mare. 1994. Study of American families. Madison, WI: Data and Program Library Service; <http://dpls.dacc.wisc ,edu/Saf/index.html>. Hetherington. E. M. 1993. An overview of the Virginia longitudinal study of divorce and re­marriage with a focus on early adolescence. Journal of Family Psychology 7:39-56. Jockin, V., M. McGue, and D. T. Lykken. 1996. Personality and divorce: A genetic analysis. Journal of Personality and Social Psychology 71:288-299. Keith, V. M., and B. Finlay. 1988. The impact of parental divorce on children's educational at­tainment, marital timing, and likelihood of di­vorce. Journal of Marriage and the Familv 50:797-809. ' Kiernan, K. E., and A. J. Cherlin. 1999. Parental divorce and partnership dissolution in adult­hood: Evidence from a British cohort study. Populations Studies 32:39-48. Krein, S. F, and A. H. Beller. 1988. Educational attainment of children from single-parent fami­lies: Differences by exposure, gender, and race. Demography 25:221-234. Kurdek, L. A. 1989. Siblings' reactions to parental divorce. Journal of Divorce 12:203-219. Liang, K.-Y. and L. Z. Zeger. 1993. Regression analysis for correlated data. Annual Review of Public Health 14:43-68. McGuk, M., and D. T. Lykken. 1992. Gencdc in­fluence on risk of divorce. Psychological Science 6:368-373. McLanahan, S. S. 1985. Family structure and the reproduction of poverty. American Journal of Sociology 90:873-901. McLanahan, S. S., and G. Sandhfur. 1994. Growing up with a single parent: What hurts, what helps. Cambridge, MA: Harvard Univer­sity Press. McLanahan, S. S., and L. L. Bumpass. 1988. In- tcrgcnerational consequences of family disrup­tion. American Journal of Sociology 94:130-52. Mekos, D., E. M. Hetherington, and D. Reiss. 1996. Sibling differences in problem behavior and parental treatment in nondivorced and re­married families. Child Development 67:2148­2165. Monahan, S. C., C. M. Buchanan, E. E. Mac- CODY, and S. M. Dornbusch. 1993. Sibling dif­ferences in divorced families. Child Develop­ment 64:152-168. Paul, C., D. McCaffrey, W. M. Mason, and S. A. Fox. 2002. What should we do about missing data? A case study using logistic regression with data missing on a single covariate. Unpublished paper, RAND. Rossi, A., and P. Rossi. 1990. Of human bonding: Parent-child relations across the life course. New York: Aldinc dc Gruyter. Sandefur, G. D., and T. Wells. 1999. Does family structure really influence educational attainment? Social Science Research 28:331-357. Wallerstein, J. S., and J. B. Kelly. 1980. Surviv­ing the breakup: How children and parents cope with divorce. New York: Basic Books. Wolfinger, N. H. 1998. The effects of parental di­vorce on adult tobacco and alcohol consump­tion. Journal of Health and Social Behavior 39:254-270. ---------. 1999. Trends in the intergenerational trans­mission of divorce. Demography 36:415-420. ---------. 2000. Beyond the intergenerational trans­mission of divorce: Do people replicate the pat­terns of marital instability they grew up with? Journal of Family Issues 21:1061-1086. ---------. 2003a. Parental divorce and offspring mar­riage: Early or late? Social Forces 82:337-353. ---------. 2003b. Family structure homogamy: The effects of parental divorce on partner selection and marital stability. Social Science Research 32:80-97.