جفت جویی پس از طلاق: آزمون دو فرضیه با استفاده از مدل نهایی

Abstract We analyze data from 927 remarried men and women to examine the association between spouses' educational attainment, social class, and age in their first and current union. Applying marginal homogeneity models, we test two competing hypotheses: current unions of remarried people are more homogamous than their first unions (the learning-hypothesis) and remarried people's current unions are less homogamous than their first unions (the marriage market hypothesis). With respect to education, the evidence supports the learning-hypothesis for remarried men, but not for remarried women. With respect to social class, the evidence supports neither the learning-hypothesis nor the marriage market hypothesis. Finally, with respect to age, we find, for both men and women, support for the marriage market hypothesis. We conclude that the remarriage market may have become more beneficial to remarrying men to find a more educationally homogamous partner than their first partner. Moreover, greater age heterogeneity of available spouses in the remarriage market appears to be an important determinant of weaker age homogamy in remarriage.

1. Introduction People tend to choose a partner with a similar social background. Sociologists have extensively studied assortative mating with respect to such social background characteristics as education, class, and religion (Hendrickx et al., 1995; Kalmijn, 1994; Mare, 1991). These studies have focused predominantly on assortative mating in people's first marital or cohabiting unions. However, due to the increase in the number of divorces—in The Netherlands, about 30 percent of existing marriages will dissolve into divorce—remarriages and re-cohabitation have become increasingly important. For example, in The Netherlands it is estimated that about 66 percent of those who were married and who experienced a divorce will have entered a new union within 6 years (van Huis and Visser, 2001). Furthermore, on an average 58 percent of those who cohabited and subsequently dissolved their union will choose to live with a new partner within 4 years (Keij and Harmsen, 2001). Thus, only few people prefer to stay alone after dissolution of their union. When a person experiences divorce and starts a new union, the same or a differing degree of homogamy with respect to social background characteristics, as compared to the first union, will characterize this new union. Common wisdom has it that people who marry for the second time ought to have learned something from their past experiences and mistakes. Yet it is also commonly believed that divorced people will repeat the same mistakes. Either they do not really learn from their mistakes or they are unable to put this learning into practice. For instance, people may have learned what kind of mate they really want but then they may be unable to find such a person, because their opportunities to do so may be restricted. The important question, then, is whether people actually `do things differently' the second time around and, if so, what they do differently (Benson-von der Ohe, 1987). In this paper, we examine changes in assortative mating from first to second unions by testing two competing hypotheses. The first hypothesis pertains to changed preferences for certain spousal characteristics among those willing to re-partner following divorce. It states that individuals who have experienced divorce will prefer a new partner who is more similar to them with respect to social background than their first partner was. More specifically, divorcées have learned from the adjustment problems of a non-homogamous first union that their second union should be more homogamous (Dean and Gurak, 1978). In short, former spouses are assumed to have `learned from their mistakes' (Whyte, 1990) and make a better choice when re-partnering. The learning-hypothesis is interesting for several reasons. In the first place, it reflects the popular and optimistic idea that people learn from their experiences, especially in relationships. However, whether this is actually the case has only rarely been tested empirically, and with contradictory results. In addition, there are several reasons to expect that remarriages will differ substantially from first marriages. Some commonly cited reasons for this include: the lessons learned from the failure of the first union, its continuous imprint on the current union, the fact that first and second unions take place at divergent points in an individual's life and, finally, that remarried individuals may be subject to very different expectations about how to behave in marriage, because they belong to different marriage cohorts (Furstenberg and Spanier, 1987). A closer investigation of the idea that people learn from the experience of a failed union and look for a better match will shed light on the peculiarities of people's second unions, compared to their first ones. Finally, knowledge about decreased differences in homogamy between first and current unions is important because homogamy between spouses may have positive consequences for marital success (Janssen, 2002). However, the extent to which people can do things differently with respect to re-partnering is very much dependent on the opportunities in the marriage market. For divorced people, it may be less easy to find a new homogamous mate, because the pool of potential partners for divorced people is more limited and diverse than for those seeking marriage for the first time. Thus, even though divorced people may have learned that they should look for a more similar partner the second time around, the marriage market may not be favorable to them so they can put into practice what they have learned. According to this marriage market hypothesis, it is expected that second marriages of divorced people will be less homogamous than their first marriages, because the marriage market limits their ability to realize their preference for a similar partner. In this paper, we test the learning-hypothesis and the marriage market hypothesis, using data from a national life-history survey among 551 first married and 1795 ever-divorced people in The Netherlands in 1998 (Kalmijn, de Graaf and Uunk, 1998). We seek to broaden the knowledge on remarriage in several ways. In the first place, although remarriages have grown in importance, little attention has been paid to partner choice in remarriage in the recent literature. In this study, we use more recent data to replicate and extend somewhat older studies by Whyte (1990), Nı́ Brolchaı́n (1988), Jacobs and Furstenberg (1986), Mueller and Pope (1980), and Dean and Gurak (1978). In addition, by studying assortative mating in remarriage for the case of The Netherlands, we intend to fill the existing lacuna on knowledge about remarriage for this country (Uunk, 1999). We have data on assortative mating with respect to the level of education, age, and social class for remarried people and for their first and second partner. Unfortunately, for the current spouse of remarried people we lack information on other interesting social characteristics, such as religious affiliation and race. The strength of these data is that they include not only information on remarried women, but also on remarried men. In previous studies, data for men were, in most cases, only samples of husbands of remarried women. Thus, we are able to investigate assortative mating among couples, which also include characteristics of first and current wives of remarried men. The paper also shows a methodological application of so-called marginal models (Agresti, 1996; Bergsma, 1997), which allow estimation of log-linear models when observations are dependent. Log-linear models are frequently used in studies on assortative mating in first marriages. These studies often analyze categorical data, such as schooling, social class, or race. However, when applying these models to the analysis of changes in homogamy from a person's first marriage to his next, the researcher must take into account not only the categorical nature of the variables, but also the fact that observations come from the same individual when performing statistical tests. Simply applying `standard' log-linear models to answer such longitudinal research questions may lead to misleading results. To our knowledge, this is the first study on assortative mating in remarriage, which systematically tries to deal with this statistical problem.

4. Results Consider Table 1, which presents the marginal tables AB and CD for remarried men and women with respect to educational endogamy. Comparison of these marginal tables for remarried men shows that some change in partner choice occurs between their first and current unions. This change occurs particularly among remarried men with middle and higher level schooling. Whereas 37.4 percent of remarried men with intermediate schooling had a wife with low educational attainment, this percentage drops to 22.9 percent in current unions. However, the same group of men appears to be slightly more likely to re-partner with a woman of the same educational level (53.5 percent in the first union compared to 60.0 percent in the second union). They are also substantially more likely to be married to women with a higher educational attainment (9.1 percent in the first union versus 17.1 percent in the current union). Turning to the higher educated remarried men, we see there is a rather large increase in the number of highly schooled men who remarry to a woman with the same educational attainment (43.1 percent in the first union versus 54.7 percent in the current union). For remarried men, there thus appears to be a change towards marrying more highly educated women the second time around, at the expense of lower educated women. Turning to remarried women, we see that the number of women with an intermediate level of schooling, and who have a spouse with lower educational attainment, is substantially lower in current unions than in first unions (37.0 percent in first unions versus 25.8 percent in current unions). On the other hand, the number of women with intermediate level educational attainment who have a husband with higher educational attainment is considerably higher the second time around than the first time around (18.5 percent in first unions versus 28.0 percent in current unions). Finally, we also see a slight increase in the percentage of higher educated women who re-partner with higher educated men (59.4 percent versus 64.4 percent). Comparison of the marginal distributions of sub-tables AB and CD shows that current unions of remarried women are characterized by relatively fewer husbands with only lower educational attainment and by relatively more husbands with higher educational attainment in comparison to first unions. Of course, such crude outflow percentages give only very basic information about the degree of assortative mating in first and current unions. A main disadvantage is that one does not adjust for the marginal distributions of husband and wife's characteristics, thereby taking into account the effect of relative group size in a population on homogamy (Kalmijn and Flap, 2001). Therefore, we have estimated several marginal models, which model the association between the characteristics of partners and which take into account the relative group sizes. They also allow more systematic testing for differences in endogamy between first and current unions. Table 2 gives selected parameter estimates and goodness-of-fit statistics for these models. Table 2. Educational endogamy models Model Remarried men Remarried women Selected parameters: First union Current union First union Current union M3. Homogeneous overall endogamy .855 (.078) .855 (.078) 0.776 (.078) 0.776 (.078) M4. Heterogeneous specific endogamy Lower 1.794 (.263) 2.719 (.273) 1.588 (.263) 1.607 (.273) Middle −.852 (.239) −1.204 (243) −.763 (.239) −.340 (.243) Higher 2.594 (.284) 2.779 (.256) 2.345 (.284) 1.750 (.256) M5. Homogeneous specific endogamy Lower 2.195 (.215) 2.195 (.215) 1.60 (.74) 1.60 (.74) Middle −1.004 (.194) −1.004 (.194) −.56 (1.30) −.56 (1.30) Higher 2.656 (.221) 2.656 (.221) 2.04 (.58) 2.04 (.58) M6. Heterogeneous uniform association 1.116 (.090) 1.458 (.085) 1.038 (.090) .958 (.084) M7. Homogeneous uniform association 1.275 (.075) 1.275 (.075) .996 (.075) .996 (.075) Goodness of fit X2 G2 df p BIC X2 G2 df p BIC M1. Marginal homogeneity 49.0 69.7 8 .000 21.46 34.4 46.3 8 .000 −2.77 M2. Homogeneous association 5.24 5.26 4 .264 −18.86 3.71 3.72 4 .447 −20.81 M3. Homogeneous overall endogamy 61.3 82.0 7 .000 39.78 48.9 58.2 7 .000 15.27 M4. Heterogeneous specific endogamy .23 .24 2 .109 −11.82 .25 .25 2 .118 −12.02 M5. Homogeneous specific endogamy 5.26 5.36 5 .385 −24.79 3.76 3.76 5 .416 −26.91 M6. Heterogeneous uniform association 5.64 5.81 6 .465 −30.37 10.5 10.5 6 .105 −26.30 M7. Homogeneous uniform association 9.74 9.62 7 .204 −32.60 11.0 10.9 7 .139 −32.03 Conditional likelihood-ratio tests: Model 5–Model 4 5.12 3 .163 3.51 3 .320 Model 7–Model 6 3.81 1 .051 .40 1 .473 Note. Standard errors between parentheses. Table options We start by comparing tables AB and CD, and by testing the most restrictive hypothesis, namely that they are identically distributed (Model 1). The results presented in Table 2 show that this model yields a bad fit for both remarried men and remarried women. The prior inspection of tables AB and CD already showed that the marginal distributions are different: the marginal distribution of A differs from C and B differs from D. The following models test the weaker hypothesis that the association in tables AB and CD is similar. Model 2 asserts the equality of local odds ratio's between table AB and CD. This model fits the data well for both remarried men and remarried women. Thus, the null hypothesis of no change in association between AB and CD cannot be rejected. Model 3 tests whether we can describe the association between husbands' and wives' in-marriage with only one diagonal parameter and whether this parameter is equivalent for tables AB and CD. The test results for this model presented in Table 2 indicate strong lack of fit, both for remarried men and remarried women. However, the less parsimonious variants of the endogamy models, which estimate one parameter for each educational category, fit the data well. This holds for both the homogeneous and heterogeneous variants of the endogamy models and for the models for remarried men and remarried women. As we would expect based on the learning-hypothesis, the specific endogamy parameters under the heterogeneous model show somewhat greater educational endogamy among remarried men. However, this is not true for remarried women, where the degree of endogamy appears to be weaker in current unions than in first unions, especially among women with intermediate and higher levels of education. This seems to contradict the learning-hypothesis and to support the marriage market hypothesis. However, the conditional likelihood ratio test between Model 4 and Model 5 is not significant, showing that the heterogeneous specific endogamy model does not provide a significantly better fit than the homogeneous variant. Finally, the same kind of result is found when testing Model 6, which asserts that the uniform association in table AB is different from the uniform association parameter in table CD, and Model 7, which asserts that uniform association is equal in both sub-tables: both models show strong fit to the data. However, the conditional likelihood-ratio test does provide evidence in favor of the heterogeneous variant of uniform association—rather than the homogeneous variant—for remarried men, but not for remarried women. The estimates for the uniform association parameters under Model 6 also reflect this: The association changes from 1.126 among first unions to 1.458 in current unions of remarried men. Among remarried women, the change in uniform association appears to be negative, although the homogeneous uniform association model does not have a poor fit compared to the heterogeneous uniform association model. This suggests that the null hypothesis of no change need not be rejected. In conclusion, the evidence supports the learning-hypothesis for remarried men, but not for remarried women. Next, we consider the degree of assortative mating with respect to social class. The marginal tables AB and CD for both remarried men and women are presented in Table 3. Note that in this case, the results should be interpreted with caution, especially for remarried women. Differential selection of women into the labor market may introduce selection bias. In other words, remarried women who were working at their first and current marriage may be a selective group.2 Table 3. Marginal tables of class endogamy in first and current unions B. Class first wife D. Class current wife Salariat Middle Working Total Salariat Middle Working Total Remarried men A. Husband's class first union Salariat 12 (25.5) 30 (63.8) 5 (10.6) 47 (100) C. Current union 30 (46.9) 30 (46.9) 4 (6.3) 64 (100) Middle 6 (12.0) 34 (68.0) 10 (20.0) 50 (100) 11 (23.9) 28 (60.9) 7 (15.2) 46 (100) Working 5 (6.1) 46 (56.1) 31 (37.8) 82 (100) 9 (13.0) 29 (42.0) 31 (44.9) 69 (100) Total 23 (12.8) 110 (61.5) 46 (25.7) 179 (100) 50 (27.9) 87 (48.6) 42 (23.5) 179 (100) B. Class first husband D. Class current husband Salariat Middle Middle Total Salariat Middle Working Total Remarried women A. Wife's class first union Salariat 15 (28.3) 30 (56.6) 8 (15.1) 53 (100) C. Current union 23 (27.4) 48 (57.1) 13 (15.5) 84 (100) Middle 9 (17.0) 34 (64.2) 10 (18.9) 53 (100) 10 (18.9) 30 (56.6) 13 (24.5) 53 (100) Working 8 (7.1) 60 (53.6) 44 (39.3) 112 (100) 8 (9.9) 30 (37.0) 43 (53.1) 81 (100) Total 32 (14.7) 124 (56.9) 62 (28.4) 218 (100) 41 (18.8) 108 (49.5) 69 (31.7) 218 (100) Note. Outflow percentages between parentheses. Table options The comparison of outflow percentages for remarried men shows, first, that men who belong to the salariat are more likely the second time around to have a wife who also belongs to the salariat than in their first union (25.5 percent versus 46.9 percent). On the other hand, men who belong to the salariat are less likely the second time around to have a wife who belongs to the middle classes than in their first union (63.8 percent versus 46.9 percent). In addition, the number of men who belong to the middle classes and who marry a woman who belongs to the salariat is relatively larger in current unions than in first unions, whereas the number of middle-class men who marry a woman from a middle class or working class background drops from first to second marriage. Finally, working class men are more likely to marry a woman from the salariat the second time around. They are also more likely to re-partner with a woman of the same class background, but less likely than in their first union to have a partner who comes from the middle classes. Among remarried women, we find that middle-class women are less likely to have a middle-class husband the second time around. They also appear to be slightly more likely in their current union to have a husband with a working class background compared to the first union. Finally, working class women appear to be less likely the second time around to have a husband who belongs to the middle classes, whereas they are more likely in remarriage to have a husband who belongs to the working class. The marginal distribution of B and D differs mainly in the percentage of remarried women who had a husband from the middle classes: in remarriage, this percentage drops from 56.9 percent to 49.5 percent. Table 4 provides the results of the testing of several marginal models for the association between husbands' and wives' class position in first and current unions. Table 4. Class endogamy models Model Remarried men Remarried women Selected parameters: First union Current union First union Current union M3. Homogeneous overall endogamy .733 (.078) .733 (.078) .521 (.111) .521 (.111) M4. Heterogeneous specific endogamy Salariat 1.342 (.263) 1.285 (.273) 1.135 (.420) .663 (.388) Middle/intermediate classes −.555 (.239) −.380 (.243) −.406 (.398) −.378 (.371) Working 1.381 (.284) 1.902 (.256) 1.207 (.377) 1.605 (.337) M5. Homogeneous specific endogamy Salariate 1.339 (.159) 1.339 (.159) .899 (.303) .899 (.303) Middle/intermediate classes −.478 (.194) −.478 (.194) −.456 (.248) −.456 (.248) Working 1.608 (.221) 1.608 (.221) 1.438 (.261) 1.438 (.261) M6. Heterogeneous uniform association .688 (.090) .833 (.084) .624 (.142) .616 (.129) M7. Homogeneous uniform association .772 (.075) .772 (.075) .619 (.103) .619 (.103) Goodness of fit X2 G2 df p BIC X2 G2 df p BIC M1. Marginal homogeneity 21.5 22.1 8 .006 −19.40 20.1 21.5 8 .010 −21.58 M2. Homogeneous local odds ratio's 1.62 1.64 4 .195 −19.11 2.15 2.14 4 .292 −19.40 M3. Homogeneous overall endogamy 16.74 19.26 7 .019 −17.05 20.38 21.90 7 .005 −15.79 M4. Heterogeneous specific endogamy .24 .24 2 .113 −10.14 .81 .81 2 .333 −9.96 M5. Homogeneous specific endogamy 1.83 1.87 5 .128 −24.07 2.24 2.23 5 .185 −24.69 M6. Heterogeneous uniform association 2.65 2.66 6 .149 −28.46 4.85 4.91 6 .437 −27.40 M7. Homogeneous uniform association 3.02 3.08 7 .117 −33.23 4.85 4.92 7 .322 −32.77 Conditional likelihood-ratio tests: Model 5–Model 4 1.63 3 .347 1.42 3 .299 Model 7–Model 6 .42 1 .483 .01 1 .080 Note. standard errors between parentheses. Table options Again, we start with the test of the marginal homogeneity hypothesis, which holds that both marginal tables AB and CD are identically distributed. This model yields a bad fit. Model 2 asserts homogeneous association in both marginal tables. This model fits well and suggests that we do not have to reject the null hypothesis of no change. Model 3, which estimates a single endogamy parameter for the diagonal for both marginal tables, performs poorly for both the data of remarried men and remarried women. However, the less restricted heterogeneous specific endogamy model (Model 4), as well as the homogeneous specific endogamy model (Model 5) both perform well. The conditional likelihood-ratio test between these models is not significant at the 5 percent level. Although the heterogeneous specific endogamy model does not fit better than its homogeneous counterpart does, inspection of the parameters of Model 4 shows some interesting results, which are in line with both the learning-hypothesis and the marriage market hypothesis. In particular, class endogamy decreases from first to second marriage for remarried men and especially for remarried women who belong to the salariat. This is in accordance with the marriage market hypothesis. In contrast, class endogamy increases for remarried men and women who belong to the working class, which is in line with the learning-hypothesis. Finally, Model 6 and Model 7 assume uniform association in both marginal tables; these models yield a good fit. The uniform association parameters estimated under Model 6 show an increase in the association between first and current unions of remarried men, whereas the association parameters in both sub-tables for remarried women are about equal. The likelihood-ratio test shows that the homogeneous uniform association model does not have a worse fit than the heterogeneous uniform association model. It appears that the data do not allow a clear decision on which model to prefer. However, the BIC measures suggest that there is positive evidence favoring the homogeneous uniform association model. We conclude here that the evidence supports neither the learning-hypothesis nor the marriage market hypothesis; this applies to remarried men and remarried women. Finally, we turn to the analysis of age homogamy among remarried men and women (Table 5). Table 5. Marginal tables of age homogamy B. Age first wife D. Age current wife <20 21–25 26–30 31–35 36–40 >40 Total <20 21–25 26–30 31–35 36–40 >40 Total Remarried men A. Husband's age First union <20 49 (79.0) 12 (19.4) 1 (1.6) 0 (0) 0 (0) 0 (0) 62 (100) C. Current union 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (100) 21–25 85 (36.3) 136 (58.1) 10 (4.3) 3 (1.3) 0 (0) 0 (0) 234 (100) 2 (22.7) 7 (77.8) 0 (0) 0 (0) 0 (0) 0 (0) 9 (100) 26–30 20 (20.6) 52 (53.6) 22 (22.7) 2 (2.1) 1 (1.0) 0 (0) 97 (100) 14 (19.7) 30 (42.3) 17 (23.9) 9 (12.7) 1 (1.4) 0 (0) 71 (100) 31–35 0 (0) 5 (29.4) 6 (35.3) 4 (23.5) 1 (5.9) 1 (5.9) 17 (100) 5 (5.7) 21 (23.9) 34 (38.6) 24 (27.3) 3 (3.4) 1 (1.1) 88 (100) 36–40 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 4 (3.9) 16 (15.5) 30 (29.1) 25 (24.3) 17 (16.5) 11 (10.7) 103 (100) >40 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 1 (0.7) 3 (2.2) 19 (13.7) 27 (19.4) 20 (14.4) 69 (49.6) 139 (100) Total 154 (37.6) 205 (50.0) 39 (9.5) 9 (2.2) 2 (0.5) 1 (0.2) 410 (100) 26 (6.3) 77 (18.8) 100 (24.4) 85 (20.7) 41 (10.0) 81 (19.8) 410 (100) B. Age first husband D. Age current husband <20 21–25 26–30 31–35 36–40 >40 Total <20 21–25 26–30 31–35 36–40 >40 Total Remarried women A. Wife's age First union <20 134 (84.8) 140 (13.6) 5 (1.5) 0 (0) 0 (0) 0 (0) 66 (100) C. Current union 0 (0) 1 (50.0) 1 (50) 0 (0) 0 (0) 0 (0) 2 (100) 21–25 56 (84.8) 9 (13.6) 1 (1.5) 0 (0) 0 (0) 0 (0) 279 (100) 1 (2.5) 11 (27.5) 18 (45.0) 6 (15.0) 3 (7.5) 1 (2.5) 40 (100) 26–30 20 (21.3) 57 (60.6) 15 (16.0) 2 (2.1) 0 (0) 0 (0) 94 (100) 0 (0) 11 (11.3) 40 (41.2) 33 (34.0) 9 (9.3) 4 (4.1) 97 (100) 31–35 3 (17.6) 8 (47.1) 4 (23.5) 1 (5.9) 1 (5.9) 0 (0) 17 (100) 0 (0) 4 (4.8) 26 (31.0) 34 (40.5) 15 (17.9) 5 (6.0) 84 (100) 36–40 0 (0) 3 (75.0) 0 (0) 0 (0) 1 (25.0) 0 (0) 4 (100) 0 (0) 0 (0) 11 (14.7) 34 (45.3) 20 (26.7) 10 (13.3) 75 (100) >40 0 (0) 3 (100.0) 0 (0) 0 (0) 0 (0) 0 (0) 3 (100) 0 (0) 2 (1.2) 5 (3.0) 19 (11.5) 23 (13.9) 116 (70.3) 165 (100) Total 213 (46.0) 223 (47.0) 25 (5.4) 3 (0.6) 2 (0.4) 0 (0) 463 (100) 1 (0.2) 29 (6.3) 101 (21.8) 126 (27.2) 70 (15.1) 136 (29.4) 463 (100) Note. Outflow percentages between parentheses. Table options Because an analysis of age at marriage for remarried people has to cover a broad age range, a relatively large number of empty cells characterize the marginal tables AB and CD.3 The marginal tables reflect the difference in age homogamy, which exist between first and current marriages. None of the remarried men is older than 35 when he enters his first marriage and most of the partners are well below the age of 35 as well. However, when we consider marginal table CD for the remarried men, we see a considerable difference in age homogamy. Most remarriages of divorced men occur from the age of 26 onwards, and it appears that a large number of these divorcées prefer a wife of considerable younger age in remarriage (61 percent of all observations occur below the diagonal of table CD). Among remarried women, also most first marriages occur well below the age of 35, and most of their first partners are of the same age or younger. However, in remarriage 23 percent of the observations are above the diagonal (compared to only 4 percent in first marriages), indicating that a substantial number of remarriages of divorced women occur with an older husband. Notice, however, that 29 percent of the remarried women in table CD have a new husband who is younger than the wife. This substantial shift in age homogamy among remarried men and women is reflected in the test of marginal homogeneity, which shows a very poor fit, both for remarried men and remarried women (Table 6, Model 1). Table 6. Age homogamy models Model Remarried men Remarried women Selected parameters: First union Current union First union Current union M6. Heterogeneous uniform association 1.03 (.129) .512 (.055) .801 (.110) .613 (.052) M7. Homogeneous uniform association .698 (.055) .698 (.055) .661 (.047) .661 (.047) Goodness of fit X2 G2 df p BIC X2 G2 df p BIC M1. Marginal homogeneity 315.16 446.94 35 .000 236.37 402.61 558.52 35 .000 343.7 M6. Heterogeneous uniform association 37.72 38.90 48 .143 −249.88 83.96 80.22 48 .001 −214.39 M7. Homogeneous uniform association 52.65 53.56 49 .335 −241.23 89.56 84.85 49 .000 −215.90 Conditional likelihood-ratio tests: Model 7–Model 6 14.66 1 .000 4.63 1 .031 Note. Standard errors between parentheses. Table options For remarried men, we find that the heterogeneous uniform association (Model 6) model, as well as the homogeneous uniform association model (Model 7) both show a satisfactory fit. However, the conditional likelihood-ratio test between these models yields a significant result, which indicates that the homogeneous uniform association model fits poorly compared to the heterogeneous uniform association model. For remarried women, we find that neither the heterogeneous uniform association model nor the homogeneous uniform association model holds. Substantive conclusions based on these models can therefore not be drawn. However, the results suggest that also the current unions of remarried women are less homogamous with respect to age than in their first unions: the conditional likelihood-ratio test between Model 7 and Model 6 is significant, indicating that the homogeneous uniform association model fits poorly compared to the heterogeneous uniform association model. This is evidence against the learning-hypothesis and in favor of the marriage market hypothesis. In accordance with the marriage market hypothesis, the strength of uniform association decreases from 1.03 for remarried men in their first union to. 512 in their second unions. Among remarried women, a decrease in the strength of association can also be detected (.801 in first unions versus .613 in current unions), although it is not as pronounced as among remarried men.