Prev Sci DOI 10.1007/s11121-014-0480-4
The Honeymoon Effect: Does It Exist and Can It Be Predicted? Michael F. Lorber & Ann C. Eckardt Erlanger & Richard E. Heyman & K. Daniel O’Leary
# Society for Prevention Research 2014
Abstract The population-level decrease over time in newlyweds’ marital satisfaction is well established. Yet decreasing marital satisfaction does not occur for all spouses to the same extent, if at all. In the present article, we test for the presence and predictability of a “honeymoon effect”—initially high, but rapidly decreasing, marital satisfaction in newlywed couples. Community couples (N=395) were studied from 1 month prior through 2.5 years after marriage. A supermajority of couples showed initially high marital satisfaction that either slowly decreased (women: 86 %) or remained steady (men: 78 %). Smaller groups of men and women showed initially high (men) and moderately high (women), rapidly decreasing marital satisfaction or steady, low levels of marital satisfaction. Membership in these latter less optimal, classes was most consistently predicted by spouses’ own intimate partner violence (IPV) and depression, as well as by their partners’ marital satisfaction, IPV, and depression. The findings suggest that men at risk for the honeymoon effect (~14 %) can be identified for selective prevention based on such predictors. Women at risk for decreasing marital satisfaction (~10 %) can also be identified based on risk factors, but may also exhibit somewhat attenuated marital satisfaction at engagement. Keywords Marital satisfaction . Newlywed . Trajectory . Longitudinal . Prevention
Decreases in newlyweds’ marital satisfaction are so well known they have achieved axiomatic status in the scientific M. F. Lorber (*) : A. C. E. Erlanger : R. E. Heyman New York University, New York, NY, USA e-mail: [email protected] K. D. O’Leary Stony Brook University, Stony Brook, NY, USA
literature and public consciousness (e.g., Naish 2008). Marital satisfaction starts high (including the transient “honeymoon phase”) and decreases over time (Bradbury and Karney 2004; Kurdek 1999). This population-level decrease in marital satisfaction has been known for decades (Vaillant and Vaillant 1993) and is borne out in meta-analyses (Mitnick et al. 2009). Several empirically supported interventions to prevent declines in marital satisfaction and divorce have been developed (e.g., Halford et al. 2004; Laurenceau et al. 2004). These interventions are usually offered to marrying couples irrespective of risk (Halford 2004), with the exception of some that are targeted to the transition to parenthood (e.g., Halford et al. 2010) or parents of preschoolers (Cowan et al. 2011). However, recent meta-analytic findings suggest that the transition to parenthood does not reliably impact marital satisfaction trajectories (Mitnick et al. 2009). Given the expense of preventive interventions, as well as possible iatrogenic effects for low-risk newlyweds (Halford et al. 2001), scarce resources would ideally be targeted to those newlyweds most in need, namely those at greatest risk of rapidly decreasing relationship adjustment. What has become increasingly clear is that not all newlyweds are either highly satisfied with their marriages or at equal risk of rapidly decreasing marital satisfaction. This fact is illustrated in the findings of four reports published in the last few years (Anderson et al. 2010; Birditt et al. 2012; Lavner and Bradbury 2010; Lavner et al. 2012). These investigations each used statistical methods to identify underlying groups of individuals or couples with similar trajectories of marital satisfaction. Each study has identified one group with enduring high levels of marital satisfaction, and another with initially low, decreasing marital satisfaction. Where these studies’ results have differed is in the number (from one to three) and nature of intermediate classes identified. In Lavner and Bradbury and Birditt et al.—each of whom found more than
Prev Sci
one intermediate trajectory class—classes with lower initial marital satisfaction tended to show greater decreases in marital satisfaction. Lavner et al. found only one intermediate class, and again significant decreases were observed. What the results of these three studies jointly suggest is that marital satisfaction does decrease early in marriage, but that it is largely a function of initial marital satisfaction. Higher initial satisfaction is associated with smaller decreases over time. To the preventionist, the findings of Birditt et al. (2012), Lavner and Bradbury (2010), and Lavner et al. (2012) suggest a stepped approach to intervention. Couples with low initial satisfaction may need clinical intervention/indicated prevention to repair their ailing marriages. Couples with moderate initial satisfaction may need lighter touch selective prevention to avert modest decreases in marital satisfaction. Couples with high initial satisfaction may not need any intervention, as they show little risk for declining satisfaction. Yet the findings of the long-term longitudinal study of Anderson et al. (2010), in contrast to the three trajectory studies reviewed above, challenge the notion that lower initial marital satisfaction can be used to identify couples in need of intervention to prevent decreases in marital satisfaction and its sequelae. These authors identified a group of couples (10.6 %) who typified what can be thought of as a “honeymoon effect.” Their initial satisfaction was actually somewhat higher than the 21.5 % who showed enduring high levels of marital satisfaction. However, this state of elevated newlywed marital satisfaction was fleeting, with a rapid initial decline to low levels of satisfaction that persisted beyond 30 years in this 50year investigation. Interestingly, marital satisfaction in this group eventually rebounded somewhat between 35 and 50 years, suggesting that there is hope for couples whose marriages survive the initial and protracted decline characteristic of the honeymoon effect. The findings of Anderson et al. (2010) are particularly provocative for prevention in suggesting that some newlywed couples who are at risk for future marital misery are just as happily married as those who will experience years of marital contentment. If such a honeymoon effect proves replicable, selective prevention to avert it would be desirable. Moreover, because those at risk of the honeymoon effect would not be identifiable on the basis of their initial marital satisfaction, prevention would benefit from identifying risk factors that portend decreasing marital satisfaction among such highly satisfied newlyweds. These predictors could then be targeted in selective preventive interventions for happily married newlyweds who are at elevated risk for a rapid decline in marital satisfaction.
The Present Investigation Given the mixed findings on the nature of newlywed trajectory groups, we sought first to reevaluate the latent trajectory
classes underlying newlyweds’ marital satisfaction. We were chiefly concerned with the presence or absence of a class that exhibits the honeymoon effect (i.e., initially high, declining marital satisfaction) because of its aforementioned relevance to prevention. In a second step, we studied risk factors for declining marital satisfaction. Our approach to predictor selection was a combination of the predictor’s empirical track record in relation to marital satisfaction and its availability in our archival data set. We hypothesized that intimate partner violence (IPV), depression and alcohol problems in either spouse, the partner’s low marital satisfaction, lower socioeconomic status (SES), and young age would increase risk for membership in decreasing trajectory classes (e.g., in a honeymoon effect class, should it emerge). Each of these risk factors has shown concurrent and/or longitudinal prediction of marital dissatisfaction. Greater IPV is linked to lower initial levels of, and greater decreases in, marital satisfaction (e.g., Lavner and Bradbury 2010; Lawrence and Bradbury 2007; Testa and Leonard 2001). The rate of IPV in newlyweds is quite high and IPV is remarkably stable over time in early marriage (Lorber and O’Leary 2012). Given the high level of marital satisfaction among newlyweds (Bradbury and Karney 2004), it stands to reason that many newlywed couples are highly satisfied with their marriages, despite experiencing some degree of IPV. Yet the aversiveness of higher levels of IPV is likely to erode relationship satisfaction over time. Noxious couple environments are associated with mood and substance use disorders in epidemiological samples (e.g., Goering et al. 1996; Markowitz et al. 1989; Whisman 2007). In the present sample, we have found that individuals with premarital dysphoria have lower marital satisfaction later in marriage (Beach and O’Leary 1993). Longitudinal studies support a decline in marital satisfaction in couples with discrepant heavy alcohol consumption (e.g., Homish and Leonard 2007), and a lifetime diagnosis of alcohol use disorder in wives directly and negatively relates to decreased marital satisfaction in both partners (Cranford et al. 2010). One’s own depressive symptoms may contribute to low newlywed marital satisfaction by distorting one’s view of the relationship or by peeling back the veneer of overinflated newlywed satisfaction in a manner characteristic of “depressive realism” (Pietromonaco et al. 1992). The interpersonally challenging behaviors (e.g., lack of affection, increased IPV) associated with one’s own or a partner’s depressive symptoms and/or problem drinking may also erode relationship satisfaction (Coyne et al. 2002; Leonard and Senchak 1996). Lower SES predicted decreasing marital satisfaction in Anderson et al. (2010) and Birditt et al. (2012). Per Anderson et al., couples of lower SES may face more stresses (e.g., financial strain, neighborhood dangers) that challenge the relationship.
Prev Sci
Finally, a meta-analysis of five American data sets indicates that young couples experience less marital satisfaction (Glenn et al. 2010), after controlling for socioeconomic confounds. Glenn et al. cite theories pointing to the psychological immaturity and poorer impulse control of youth, as well as “marriage market” dynamics (e.g., marriage to an ill-suited spouse by a young person who has not adequately tested their desirability on the marriage market), to explain the age–marital satisfaction link. Youth is also associated with greater IPV (Schumacher et al. 2001), which is inversely associated with marital satisfaction.
Method Analyses of the data of an archival, longitudinal data set of newlyweds (O’Leary et al. 1989) were performed. Latent trajectory classes in marital satisfaction have not been previously analyzed in this sample. Participants Community couples (N=396) planning their first marriages in Onondaga and Suffolk counties, NY, were recruited via radio or newspaper announcements between 1983 and 1986. Demographics are reported in Table 1. The couples were all White, mostly in their 20s, and most had at least some college education. Further details on recruitment of the sample have been reported elsewhere (O’Leary et al. 1989). Procedure Couples made laboratory visits at engagement (1 month prior to marriage), as well as at 6, 18, and 30 months post-marriage, where they completed a battery of questionnaires.
Physical Intimate Partner Violence At the engagement assessment, each spouse completed the Conflict Tactics Scale (CTS; Straus 1979), a widely used questionnaire measure of the frequency of violent and non-violent conflict tactics. Physical IPV was measured with the eight-item physical aggression subscale. The CTS items increase in behavioral severity— from less (e.g., throwing something at the partner) to more (e.g., using a gun or knife)—rather than sampling the IPV construct with a number of related items. Item responses consisted of frequency ranges, coded from 0 (no occurrences of that specific act in the past year) to 6 (20 or more instances of that specific act in the last year). Item averages from both partners’ reports were subsequently averaged to yield IPV scores for each individual that took into account the person’s reports of perpetration and her/his partners’ reports of victimization (e.g., one’s reports of slapping and the partner’s reports of being slapped). Depressive Symptoms At the engagement assessment, each spouse completed the Beck Depression Inventory (BDI; Beck et al. 1988), a widely used 21-item measure of depressive symptoms that has high convergent validity with other measures of depression and discriminates depression and anxiety symptoms (Beck et al. 1988). The BDI scores analyzed were item sums. Cronbach’s alphas were .77 and .75 for men and women, respectively. Alcohol Problems At the engagement assessment, spouses completed the Michigan Alcoholism Screening Test (MAST; Selzer 1971), a widely used measure of the physical, social, and psychological consequences of problem drinking. The MAST has a weighted scoring system and total scores can from 0 to 53. The MAST differentiates people who have alcohol problems (e.g., convicted of drunk driving or drunk and disorderly conduct) from controls (Selzer 1971). Analytic Strategy
Measures Demographics Descriptive statistics on age, annual income (averaged across partners), and years of education, as well as all other study variables, are reported in Table 1. Marital Satisfaction Each spouse completed the Short Marital Adjustment Test (MAT; Locke and Wallace 1959), a widely used self-report measure of global marital satisfaction. The MAT discriminates between clinic and non-clinic couples (e.g., O’Leary and Arias 1987) and correlates very highly with more modern measures of marital satisfaction (e.g., r=.91 with the Couples Satisfaction Index; Funk and Rogge 2007). The MAT has a weighted scoring system and total scores can range from 2 to 158, with scores below 100 indicative of marital distress.
Analyses were carried out in four phases, using Mplus version 7 (Muthén and Muthén 1998–2012). In the first phase, unconditional latent growth curves (LGC) were fit to identify the trajectories of women’s and men’s marital satisfaction for the entire sample; they were also a preliminary model identification step to the second phase analyses. In the second phase, growth mixture models (GMMs; Muthén and Muthén 2000) were estimated to investigate the presence and nature of unmeasured or latent classes within which individual marital satisfaction trajectories were similar but between which individual trajectories differed significantly. The presence/absence of a group of people who exhibited the “honeymoon effect”— initially high, but rapidly declining marital satisfaction—was of chief concern. In the third phase, we explored the correspondence within couples of men’s and women’s membership
Prev Sci Table 1 Descriptive statistics Women
Age Education Incomea MAT engagement MAT 6 months MAT 18 months MAT 30 months Aggression Depression Alcohol problems a
Men
M
SD
Min.
Max.
M
SD
Min.
Max.
23.41 14.49 $15,254 123.96 119.16 116.83 114.71 1.28 4.40 1.92
2.98 1.97 $5,805 17.30 22.83 25.79 26.44 0.59 4.06 3.08
15.00 7.00 $2,100 49.11 28.00 31.00 35.00 1.00 0.00 0.00
36.00 20.00 $40,000 157.00 157.00 157.00 158.00 5.00 24.00 30.00
25.22 14.59 – 121.23 116.29 114.53 111.35 1.14 3.77 3.41
3.66 2.33 – 17.76 21.82 24.28 26.53 0.38 4.17 5.30
16.00 6.00 – 64.08 35.00 29.00 26.00 1.00 0.00 0.00
37.00 21.00 – 159.70 156.00 156.00 154.00 6.67 30.00 40.00
Income is a couple level variable, averaged across partners
in the trajectory classes identified in the GMMs. In the fourth phase, we conducted regression analyses to test our hypotheses concerning the prediction of latent trajectory class by demographic factors, personal and partner adjustment, IPV perpetration and victimization, and partner marital satisfaction, each measured at the engagement assessment.
Results Missing Data The rate of missing data was 10.14 %, primarily reflecting attrition, with the exception of two women and one man who did not provide any MAT data and thus were excluded from analysis. Otherwise, missing data were handled with full information maximum likelihood estimation to avoid estimate biasing associated with listwise deletion of cases with missing values (Schafer and Graham 2002). Unconditional Latent Growth Curve Models Unconditional LGC models were estimated for women’s and men’s marital satisfaction trajectories. LGC models use structural equation modeling to form latent intercept (starting point) and slope (change) variables that can then be studied in relation to other variables (Fig. 1). The values of the loadings of MAT scores at each wave on the slope factor determine the placement of the intercept (i.e., the point in time of the trajectories’ origination; 1 month prior to marriage in the present case) and the shape of trajectories that are being modeled. Slope loadings that are fixed to reflect the actual intervals between waves of assessment would specify a linear change model, with a constant amount and direction of change from month-to-month (e.g., a loss of .5 marital satisfaction
points per month). However, nonlinearity is suggested by the sample means presented in Fig. 1. Thus, we used the technique recommended by Little et al. (2006) to allow the data to determine the shape of the estimated marital satisfaction slopes, rather than imposing a particular model (e.g., linear or quadratic) on the shape of change in marital satisfaction. In this technique, two of the slope loadings were fixed to reflect the times of assessment and two were freely estimated. The engagement slope loading was set to 0 and the 30-month slope loading was set to 31, as it occurred 31 months after the engagement assessment. The slope loadings for the 6- and 18-month assessments were freely estimated to allow for nonlinearity in the slopes. Intercept loadings were all set to 1. The unconditional LGC models each adequately fit women’s [χ2 (3) = 5.55, p = .136, CFI = 1.00, TLI = 0.99, RMSEA=0.05] and men’s MAT data [χ2 (3)=2.23, p=.526, CFI=1.00, TLI=1.00, RMSEA=0.00]. The final slope loadings were 0.00, 19.63, 29.28, and 31.00 for women’s and 0.00, 17.90, 25.79, and 31.00 for men’s engagement, 6-, 8-, and 30month assessments. Figure 1 shows that the estimated marital satisfaction means fairly closely track the sample (i.e., observed) means. Mean intercepts (i.e., MAT scores at engagement) were 123.96 (SE=0.88) for women and 121.24 (SE= 0.89) for men. Both women (M=−0.25, SE=0.04, t=−5.93, p
The Honeymoon Effect: Does It Exist and Can It Be Predicted? Michael F. Lorber & Ann C. Eckardt Erlanger & Richard E. Heyman & K. Daniel O’Leary
# Society for Prevention Research 2014
Abstract The population-level decrease over time in newlyweds’ marital satisfaction is well established. Yet decreasing marital satisfaction does not occur for all spouses to the same extent, if at all. In the present article, we test for the presence and predictability of a “honeymoon effect”—initially high, but rapidly decreasing, marital satisfaction in newlywed couples. Community couples (N=395) were studied from 1 month prior through 2.5 years after marriage. A supermajority of couples showed initially high marital satisfaction that either slowly decreased (women: 86 %) or remained steady (men: 78 %). Smaller groups of men and women showed initially high (men) and moderately high (women), rapidly decreasing marital satisfaction or steady, low levels of marital satisfaction. Membership in these latter less optimal, classes was most consistently predicted by spouses’ own intimate partner violence (IPV) and depression, as well as by their partners’ marital satisfaction, IPV, and depression. The findings suggest that men at risk for the honeymoon effect (~14 %) can be identified for selective prevention based on such predictors. Women at risk for decreasing marital satisfaction (~10 %) can also be identified based on risk factors, but may also exhibit somewhat attenuated marital satisfaction at engagement. Keywords Marital satisfaction . Newlywed . Trajectory . Longitudinal . Prevention
Decreases in newlyweds’ marital satisfaction are so well known they have achieved axiomatic status in the scientific M. F. Lorber (*) : A. C. E. Erlanger : R. E. Heyman New York University, New York, NY, USA e-mail: [email protected] K. D. O’Leary Stony Brook University, Stony Brook, NY, USA
literature and public consciousness (e.g., Naish 2008). Marital satisfaction starts high (including the transient “honeymoon phase”) and decreases over time (Bradbury and Karney 2004; Kurdek 1999). This population-level decrease in marital satisfaction has been known for decades (Vaillant and Vaillant 1993) and is borne out in meta-analyses (Mitnick et al. 2009). Several empirically supported interventions to prevent declines in marital satisfaction and divorce have been developed (e.g., Halford et al. 2004; Laurenceau et al. 2004). These interventions are usually offered to marrying couples irrespective of risk (Halford 2004), with the exception of some that are targeted to the transition to parenthood (e.g., Halford et al. 2010) or parents of preschoolers (Cowan et al. 2011). However, recent meta-analytic findings suggest that the transition to parenthood does not reliably impact marital satisfaction trajectories (Mitnick et al. 2009). Given the expense of preventive interventions, as well as possible iatrogenic effects for low-risk newlyweds (Halford et al. 2001), scarce resources would ideally be targeted to those newlyweds most in need, namely those at greatest risk of rapidly decreasing relationship adjustment. What has become increasingly clear is that not all newlyweds are either highly satisfied with their marriages or at equal risk of rapidly decreasing marital satisfaction. This fact is illustrated in the findings of four reports published in the last few years (Anderson et al. 2010; Birditt et al. 2012; Lavner and Bradbury 2010; Lavner et al. 2012). These investigations each used statistical methods to identify underlying groups of individuals or couples with similar trajectories of marital satisfaction. Each study has identified one group with enduring high levels of marital satisfaction, and another with initially low, decreasing marital satisfaction. Where these studies’ results have differed is in the number (from one to three) and nature of intermediate classes identified. In Lavner and Bradbury and Birditt et al.—each of whom found more than
Prev Sci
one intermediate trajectory class—classes with lower initial marital satisfaction tended to show greater decreases in marital satisfaction. Lavner et al. found only one intermediate class, and again significant decreases were observed. What the results of these three studies jointly suggest is that marital satisfaction does decrease early in marriage, but that it is largely a function of initial marital satisfaction. Higher initial satisfaction is associated with smaller decreases over time. To the preventionist, the findings of Birditt et al. (2012), Lavner and Bradbury (2010), and Lavner et al. (2012) suggest a stepped approach to intervention. Couples with low initial satisfaction may need clinical intervention/indicated prevention to repair their ailing marriages. Couples with moderate initial satisfaction may need lighter touch selective prevention to avert modest decreases in marital satisfaction. Couples with high initial satisfaction may not need any intervention, as they show little risk for declining satisfaction. Yet the findings of the long-term longitudinal study of Anderson et al. (2010), in contrast to the three trajectory studies reviewed above, challenge the notion that lower initial marital satisfaction can be used to identify couples in need of intervention to prevent decreases in marital satisfaction and its sequelae. These authors identified a group of couples (10.6 %) who typified what can be thought of as a “honeymoon effect.” Their initial satisfaction was actually somewhat higher than the 21.5 % who showed enduring high levels of marital satisfaction. However, this state of elevated newlywed marital satisfaction was fleeting, with a rapid initial decline to low levels of satisfaction that persisted beyond 30 years in this 50year investigation. Interestingly, marital satisfaction in this group eventually rebounded somewhat between 35 and 50 years, suggesting that there is hope for couples whose marriages survive the initial and protracted decline characteristic of the honeymoon effect. The findings of Anderson et al. (2010) are particularly provocative for prevention in suggesting that some newlywed couples who are at risk for future marital misery are just as happily married as those who will experience years of marital contentment. If such a honeymoon effect proves replicable, selective prevention to avert it would be desirable. Moreover, because those at risk of the honeymoon effect would not be identifiable on the basis of their initial marital satisfaction, prevention would benefit from identifying risk factors that portend decreasing marital satisfaction among such highly satisfied newlyweds. These predictors could then be targeted in selective preventive interventions for happily married newlyweds who are at elevated risk for a rapid decline in marital satisfaction.
The Present Investigation Given the mixed findings on the nature of newlywed trajectory groups, we sought first to reevaluate the latent trajectory
classes underlying newlyweds’ marital satisfaction. We were chiefly concerned with the presence or absence of a class that exhibits the honeymoon effect (i.e., initially high, declining marital satisfaction) because of its aforementioned relevance to prevention. In a second step, we studied risk factors for declining marital satisfaction. Our approach to predictor selection was a combination of the predictor’s empirical track record in relation to marital satisfaction and its availability in our archival data set. We hypothesized that intimate partner violence (IPV), depression and alcohol problems in either spouse, the partner’s low marital satisfaction, lower socioeconomic status (SES), and young age would increase risk for membership in decreasing trajectory classes (e.g., in a honeymoon effect class, should it emerge). Each of these risk factors has shown concurrent and/or longitudinal prediction of marital dissatisfaction. Greater IPV is linked to lower initial levels of, and greater decreases in, marital satisfaction (e.g., Lavner and Bradbury 2010; Lawrence and Bradbury 2007; Testa and Leonard 2001). The rate of IPV in newlyweds is quite high and IPV is remarkably stable over time in early marriage (Lorber and O’Leary 2012). Given the high level of marital satisfaction among newlyweds (Bradbury and Karney 2004), it stands to reason that many newlywed couples are highly satisfied with their marriages, despite experiencing some degree of IPV. Yet the aversiveness of higher levels of IPV is likely to erode relationship satisfaction over time. Noxious couple environments are associated with mood and substance use disorders in epidemiological samples (e.g., Goering et al. 1996; Markowitz et al. 1989; Whisman 2007). In the present sample, we have found that individuals with premarital dysphoria have lower marital satisfaction later in marriage (Beach and O’Leary 1993). Longitudinal studies support a decline in marital satisfaction in couples with discrepant heavy alcohol consumption (e.g., Homish and Leonard 2007), and a lifetime diagnosis of alcohol use disorder in wives directly and negatively relates to decreased marital satisfaction in both partners (Cranford et al. 2010). One’s own depressive symptoms may contribute to low newlywed marital satisfaction by distorting one’s view of the relationship or by peeling back the veneer of overinflated newlywed satisfaction in a manner characteristic of “depressive realism” (Pietromonaco et al. 1992). The interpersonally challenging behaviors (e.g., lack of affection, increased IPV) associated with one’s own or a partner’s depressive symptoms and/or problem drinking may also erode relationship satisfaction (Coyne et al. 2002; Leonard and Senchak 1996). Lower SES predicted decreasing marital satisfaction in Anderson et al. (2010) and Birditt et al. (2012). Per Anderson et al., couples of lower SES may face more stresses (e.g., financial strain, neighborhood dangers) that challenge the relationship.
Prev Sci
Finally, a meta-analysis of five American data sets indicates that young couples experience less marital satisfaction (Glenn et al. 2010), after controlling for socioeconomic confounds. Glenn et al. cite theories pointing to the psychological immaturity and poorer impulse control of youth, as well as “marriage market” dynamics (e.g., marriage to an ill-suited spouse by a young person who has not adequately tested their desirability on the marriage market), to explain the age–marital satisfaction link. Youth is also associated with greater IPV (Schumacher et al. 2001), which is inversely associated with marital satisfaction.
Method Analyses of the data of an archival, longitudinal data set of newlyweds (O’Leary et al. 1989) were performed. Latent trajectory classes in marital satisfaction have not been previously analyzed in this sample. Participants Community couples (N=396) planning their first marriages in Onondaga and Suffolk counties, NY, were recruited via radio or newspaper announcements between 1983 and 1986. Demographics are reported in Table 1. The couples were all White, mostly in their 20s, and most had at least some college education. Further details on recruitment of the sample have been reported elsewhere (O’Leary et al. 1989). Procedure Couples made laboratory visits at engagement (1 month prior to marriage), as well as at 6, 18, and 30 months post-marriage, where they completed a battery of questionnaires.
Physical Intimate Partner Violence At the engagement assessment, each spouse completed the Conflict Tactics Scale (CTS; Straus 1979), a widely used questionnaire measure of the frequency of violent and non-violent conflict tactics. Physical IPV was measured with the eight-item physical aggression subscale. The CTS items increase in behavioral severity— from less (e.g., throwing something at the partner) to more (e.g., using a gun or knife)—rather than sampling the IPV construct with a number of related items. Item responses consisted of frequency ranges, coded from 0 (no occurrences of that specific act in the past year) to 6 (20 or more instances of that specific act in the last year). Item averages from both partners’ reports were subsequently averaged to yield IPV scores for each individual that took into account the person’s reports of perpetration and her/his partners’ reports of victimization (e.g., one’s reports of slapping and the partner’s reports of being slapped). Depressive Symptoms At the engagement assessment, each spouse completed the Beck Depression Inventory (BDI; Beck et al. 1988), a widely used 21-item measure of depressive symptoms that has high convergent validity with other measures of depression and discriminates depression and anxiety symptoms (Beck et al. 1988). The BDI scores analyzed were item sums. Cronbach’s alphas were .77 and .75 for men and women, respectively. Alcohol Problems At the engagement assessment, spouses completed the Michigan Alcoholism Screening Test (MAST; Selzer 1971), a widely used measure of the physical, social, and psychological consequences of problem drinking. The MAST has a weighted scoring system and total scores can from 0 to 53. The MAST differentiates people who have alcohol problems (e.g., convicted of drunk driving or drunk and disorderly conduct) from controls (Selzer 1971). Analytic Strategy
Measures Demographics Descriptive statistics on age, annual income (averaged across partners), and years of education, as well as all other study variables, are reported in Table 1. Marital Satisfaction Each spouse completed the Short Marital Adjustment Test (MAT; Locke and Wallace 1959), a widely used self-report measure of global marital satisfaction. The MAT discriminates between clinic and non-clinic couples (e.g., O’Leary and Arias 1987) and correlates very highly with more modern measures of marital satisfaction (e.g., r=.91 with the Couples Satisfaction Index; Funk and Rogge 2007). The MAT has a weighted scoring system and total scores can range from 2 to 158, with scores below 100 indicative of marital distress.
Analyses were carried out in four phases, using Mplus version 7 (Muthén and Muthén 1998–2012). In the first phase, unconditional latent growth curves (LGC) were fit to identify the trajectories of women’s and men’s marital satisfaction for the entire sample; they were also a preliminary model identification step to the second phase analyses. In the second phase, growth mixture models (GMMs; Muthén and Muthén 2000) were estimated to investigate the presence and nature of unmeasured or latent classes within which individual marital satisfaction trajectories were similar but between which individual trajectories differed significantly. The presence/absence of a group of people who exhibited the “honeymoon effect”— initially high, but rapidly declining marital satisfaction—was of chief concern. In the third phase, we explored the correspondence within couples of men’s and women’s membership
Prev Sci Table 1 Descriptive statistics Women
Age Education Incomea MAT engagement MAT 6 months MAT 18 months MAT 30 months Aggression Depression Alcohol problems a
Men
M
SD
Min.
Max.
M
SD
Min.
Max.
23.41 14.49 $15,254 123.96 119.16 116.83 114.71 1.28 4.40 1.92
2.98 1.97 $5,805 17.30 22.83 25.79 26.44 0.59 4.06 3.08
15.00 7.00 $2,100 49.11 28.00 31.00 35.00 1.00 0.00 0.00
36.00 20.00 $40,000 157.00 157.00 157.00 158.00 5.00 24.00 30.00
25.22 14.59 – 121.23 116.29 114.53 111.35 1.14 3.77 3.41
3.66 2.33 – 17.76 21.82 24.28 26.53 0.38 4.17 5.30
16.00 6.00 – 64.08 35.00 29.00 26.00 1.00 0.00 0.00
37.00 21.00 – 159.70 156.00 156.00 154.00 6.67 30.00 40.00
Income is a couple level variable, averaged across partners
in the trajectory classes identified in the GMMs. In the fourth phase, we conducted regression analyses to test our hypotheses concerning the prediction of latent trajectory class by demographic factors, personal and partner adjustment, IPV perpetration and victimization, and partner marital satisfaction, each measured at the engagement assessment.
Results Missing Data The rate of missing data was 10.14 %, primarily reflecting attrition, with the exception of two women and one man who did not provide any MAT data and thus were excluded from analysis. Otherwise, missing data were handled with full information maximum likelihood estimation to avoid estimate biasing associated with listwise deletion of cases with missing values (Schafer and Graham 2002). Unconditional Latent Growth Curve Models Unconditional LGC models were estimated for women’s and men’s marital satisfaction trajectories. LGC models use structural equation modeling to form latent intercept (starting point) and slope (change) variables that can then be studied in relation to other variables (Fig. 1). The values of the loadings of MAT scores at each wave on the slope factor determine the placement of the intercept (i.e., the point in time of the trajectories’ origination; 1 month prior to marriage in the present case) and the shape of trajectories that are being modeled. Slope loadings that are fixed to reflect the actual intervals between waves of assessment would specify a linear change model, with a constant amount and direction of change from month-to-month (e.g., a loss of .5 marital satisfaction
points per month). However, nonlinearity is suggested by the sample means presented in Fig. 1. Thus, we used the technique recommended by Little et al. (2006) to allow the data to determine the shape of the estimated marital satisfaction slopes, rather than imposing a particular model (e.g., linear or quadratic) on the shape of change in marital satisfaction. In this technique, two of the slope loadings were fixed to reflect the times of assessment and two were freely estimated. The engagement slope loading was set to 0 and the 30-month slope loading was set to 31, as it occurred 31 months after the engagement assessment. The slope loadings for the 6- and 18-month assessments were freely estimated to allow for nonlinearity in the slopes. Intercept loadings were all set to 1. The unconditional LGC models each adequately fit women’s [χ2 (3) = 5.55, p = .136, CFI = 1.00, TLI = 0.99, RMSEA=0.05] and men’s MAT data [χ2 (3)=2.23, p=.526, CFI=1.00, TLI=1.00, RMSEA=0.00]. The final slope loadings were 0.00, 19.63, 29.28, and 31.00 for women’s and 0.00, 17.90, 25.79, and 31.00 for men’s engagement, 6-, 8-, and 30month assessments. Figure 1 shows that the estimated marital satisfaction means fairly closely track the sample (i.e., observed) means. Mean intercepts (i.e., MAT scores at engagement) were 123.96 (SE=0.88) for women and 121.24 (SE= 0.89) for men. Both women (M=−0.25, SE=0.04, t=−5.93, p
