Descriptive Statistics and Preliminary Analyses

Descriptive statistics for the independent variables in both studies are presented in . As would be expected in samples of newlyweds, husbands and wives tended to rate their problems as not very severe, on average. Further, spouses were rated by observers as engaging in a low proportion of negative behaviors during their problem-solving discussions, on average. Nevertheless, consistent with the idea that the newlywed period can also be a time of transition and adjustment (e.g.,

Bramlett & Mosher, 2002;

Cherlin, 1992), standard deviations in all these independent variables reveal substantial between-person variability. The behavioral codes demonstrated some positive skew, however; thus, as others have done (e.g.,

Johnson et al., 2005), we subjected those codes to logarithmic transformations and used the improved distributions in all subsequent analyses.

| **Table 2**Descriptive Statistics for Independent Variables Measured at Time 1 |

Correlations among the independent variables are presented in . Several results are worth highlighting. First, although correlations between observed negative behavior and problem severity failed to reach significance for husbands, rates of observed negative behavior and self-reported problem severity were positively associated for wives in both studies, suggesting that wives who tended to report more severe problems also tended to exhibit more negative behavior. Second, rates of each type of negative behavior tended to be positively correlated among both husbands and wives, supporting the tendency for prior research to collapse across those codes. Nevertheless, consistent with the current prediction that different negative behaviors may demonstrate different effects on changes in marital satisfaction, at least half the variance in every negative code was unshared with the variance in the other negative codes, and correlations between some negative codes failed to reach significance. Finally, the cross-spouse correlations reveal that spouses tended to share some agreement regarding the severity of their problems, and tended to engage in similar levels of negative behavior during conversations about those problems.

| **Table 3**Correlations among Independent Variables. |

Describing Trajectories of Marital Satisfaction

Of the 72 marriages in Study 1, 9 (13%) were dissolved at Time 8. Of the 135 marriages in Study 2, 5 (4%) were dissolved at Time 3. Mean marital satisfaction scores, standard deviations, and the number of spouses reporting at each wave of data collection in each study are reported in . As can be seen there, husbands and wives in both studies appeared to be relatively satisfied at the outset of the study. Nevertheless, consistent with other longitudinal studies of newlyweds (e.g.,

McNulty et al., 2008), husbands and wives in both studies appeared to experience declines in satisfaction over time.

| **Table 4**Mean Marital Satisfaction Scores across Waves of Measurement for Husbands and Wives. |

Growth curve modeling (e.g.,

Bryk & Raudenbush, 1987) was used to estimate such within-person change in satisfaction over time in each study. Specifically, the following equation was estimated in the first level of a 3-level model:

where,

*Y*_{ij} is the marital satisfaction of individual j at Time i;

*π*_{0ij} is the marital satisfaction of individuals j at Time 0 (i.e., the initial satisfaction for individual j);

*π*_{1ij} is the rate of linear change in marital satisfaction of individual j; and

*e*_{ij} is the residual variance in repeated measurements for spouse j, assumed to be independent and normally distributed across spouses. This model can be understood as a within-subjects regression of an individual’s marital satisfaction onto time of assessment, where the autocorrelation due to repeated assessments was controlled in the second level of the analysis, and the shared variance between husbands’ and wives’ data was controlled in a third level of the analysis. Notably, because trajectories could be computed for all spouses who participated in two or more assessments, the majority of these couples were included in the analyses. Specifically, growth curve analyses were based on all 144 individuals in Study 1 and 234 (87%) of the 270 individuals in Study 2.

Mean estimates, standard deviations, and effect sizes of the growth curve parameters estimated by

Equation 1 are presented in .

^{5} As the table reveals, on average, individuals tended to report relatively high levels of satisfaction initially, but then tended to experience significant declines in satisfaction over time. Nevertheless, according to the standard deviations of these parameter estimates, there was substantial between-subjects variability in all parameters of these trajectories, suggesting that some spouses began the relationship with higher or lower levels of satisfaction than others, and that some spouses experienced more or less change in their satisfaction than others. The primary aim of the current studies was to examine whether variability in changes in satisfaction could be explained by the interaction between negative behavior and initial problem severity. Notably, entering a dummy code for sex into both level 2 equations to account for between-person variance in the intercepts and slopes estimated by

Equation 1 indicated that husbands and wives did not differ in either of the two parameters estimated in either of the two studies.

| **Table 5**Trajectories of Marital Satisfaction. |

How are behaviors associated with initial levels of and changes in satisfaction over time?

Before estimating the interactive effects of behavior, we examined the main effects of initial problem severity and each type of behavior on the intercepts and slopes estimated by

Equation 1. For these analyses, problem severity or one of several measures of behavior (blames, commands, rejections, a summed index of these 3 direct negative behaviors, or the index of indirect negative behaviors) was separately entered into both level 2 equations, along with a dummy code for sex. As described above, we conducted these analyses on each study separately and, in an effort to maximize power, also conducted analyses that collapsed across both studies. In the analyses that collapsed across the studies, we entered a dummy code representing whether the data were collected in Study 1 or Study 2 to control for variance in the average of each parameter estimated at level 3. Additionally, we conducted subsequent analyses to test whether any of the main effects varied by sex by entering the appropriate Sex X Problem or Sex X Behavior interactions at level 2, and whether any of the main effects varied by study by entering the study dummy code to account for variance in the average of all effects estimated at level 3.

Associations between these variables and initial satisfaction, i.e., the intercepts of the trajectories estimated by

Equation 1, are reported in the top section of . As can be seen there, initial problem severity was negatively associated with initial satisfaction in both studies, indicating, not surprisingly, that happier couples tended to report less severe problems. Regarding associations between behavior and initial satisfaction, a consistent pattern of negative associations emerged between each behavior and initial satisfaction, indicating that spouses who exhibited more negative behaviors were less satisfied with their relationships initially. Although some of these associations did not reach significance in analyses of each separate study, each behavior was significantly negatively associated with initial satisfaction in the analysis that yielded the most power by collapsing across study. Importantly, tests of the Study X Problems and all Study X Behavior interactions revealed that none of these main effects that emerged on the intercepts varied significantly across study. Likewise, tests of all Sex X Problems and Sex X Behavior interactions revealed that none of these main effects varied across sex.

| **Table 6**Main Effects of Negative Behavior on the Trajectory of Marital Satisfaction. |

Associations between these variables and changes in satisfaction, i.e., the slopes of the trajectories estimated by

Equation 1, are reported in the bottom section of . As can be seen there, initial problems were negatively associated with changes in satisfaction in Study 1 and in the analysis that collapsed across both studies, suggesting, somewhat surprisingly, that spouses who reported the least severe problems initially tended to decline in satisfaction at a steeper rate. Additionally, consistent with some prior research (e.g.,

Filsinger & Thoma, 1988;

Gottman et al., 1998), all but one main-effect of behavior on changes in satisfaction were non-significant, even when the power to detect effects was maximized in the analyses that collapsed across study, indicating that, on average, negative behaviors were unrelated to changes in relationship satisfaction over time. The one exception was that commands were marginally negatively associated with changes in satisfaction in Study 2. Nevertheless, the central prediction driving the current research suggests such average effects may mask important effects that emerge among spouses experiencing more versus less severe problems. Importantly, tests of the Study X Problem, all Study X Behavior, the Sex X Problem, and all Sex X Behavior interactions revealed that none of these effects varied significantly across sex or study.

Does initial problem severity moderate the association between initial negative behavior and initial satisfaction?

To address whether each type of negative behavior interacted with initial problems to account for each parameter of the growth curves estimated by

Equation 1, each measure of behavior (blames, commands, rejections, a summed index of these 3 direct negative behaviors, and the index of indirect negative behaviors) was centered and simultaneously entered into both level 2 equations (i.e., slopes and intercepts), along with centered reports of problem severity, one appropriate interaction term at a time, and a dummy code representing sex (see footnote 3). In the analyses that collapsed across the two studies, a dummy code representing whether the data were collected in Study 1 or Study 2 was entered to control for variance in the average of the two parameters estimated at level 3. We also tested whether any of these interactive effects varied by sex by entering the appropriate Sex X Problem or Sex X Behavior interactions at level 2, and whether any of these interactive effects varied by study by entering the study dummy code to account for variance in the average of all of the effects estimated at level 3.

Before addressing whether observed negative behaviors interacted with initial problem severity to account for the slopes of the trajectory, i.e., changes in satisfaction, we first present the interactive effects of behavior on the intercepts estimated by

Equation 1, i.e., initial satisfaction. These results are reported in the top half of . As that section of the table reveals, a number of negative interactions emerged between direct negative behaviors and initial problems. Specifically, the summed index of negative behavior interacted with initial problem severity to account for variance in spouses’ initial satisfaction in Study 2 and rejections interacted with initial problems to account for spouses’ initial satisfaction in the analysis that collapsed across the two studies. Additionally, three interactions emerged between direct negative behavior and initial problems among wives only. Specifically, a significant Sex X Commands X Problem interaction emerged in Study 1 (

*B* = −144.34,

*SE* = 63.27,

*t*(131) = −2.28,

*p* < .05,

*r* = .20), such that the significant Commands X Problems interaction reported in the top of emerged among wives, whereas the corresponding effect for husbands did not reach significance,

*B* = −7.25,

*SE* = 48.24,

*t*(131) = −0.15,

*p* > .50,

*r* = .01. Likewise, a significant Sex X Rejection X Problem interaction emerged in Study 2 (

*B* = −201.40,

*SE* = 101.09,

*t*(253) = −1.99,

*p* < .05,

*r* = .12), such that the Rejection X Problems interaction reported in the top of emerged among wives, whereas the corresponding effect for husbands did not reach significance,

*B* = −89.02,

*SE* = 56.21,

*t*(253) = −1.58,

*p* = .11,

*r* = .10. Finally, a significant Sex X Commands X Problem interaction emerged in the analysis that collapsed across both studies (

*B* = −136.99

*SE* = 53.31,

*t*(395) = −2.57,

*p* < .05,

*r* = .13), such that the Commands X Problems interaction reported in the top of emerged among wives, whereas the corresponding effect for husbands did not reach significance,

*B* = 9.83,

*SE* = 40.48,

*t*(395) = 0.24,

*p* > .50,

*r* = .01. None of the Study X Behavior X Problem interactions reached significance, indicating that none of the effects varied significantly across study.

| **Table 7**Interactive Effects of Negative Behavior and Initial Problem Severity on the Trajectory of Marital Satisfaction. |

The significant interactions were deconstructed to determine whether and when the simple effects of behavior on initial satisfaction reached significance within the sample. Typically, the simple effects of a moderated variable are determined at specific values of the moderating variable, e.g., one standard deviation above and below the mean as recommended by

Cohen and Cohen (1983). Indeed, researchers and practitioners often want to know the effect of one variable at a specific, meaningful value of another variable. As Cohen and Cohen pointed out, however, because one standard deviation is sometimes arbitrary and sample-specific, those values may not always provide the most complete theoretical description of an interactive relationship, as they may leave meaningful significant simple effects that emerge just beyond these arbitrary limits undetected. For this reason, we followed instructions provided by

Preacher, Curran, and Bauer (2006) to use the Johnson-Neyman method (

Johnson & Neyman, 1936) to identify the exact levels of problem severity at which each direct negative behavior demonstrated significant associations with initial satisfaction – i.e., the regions of significance of the simple effects of behavior. This procedure yielded a similar pattern of results across all significant interactions. Specifically, each negative behavior was significantly positively associated with initial satisfaction among spouses facing more minor problems (lower region of significance was −1.84

*SD* for wives’ commands in Study 1, −1.55

*SDs* for the summed index of negative behavior in Study 2, +0.09

*SD*s for wives’ rejections in Study 2, −1.70

*SD*s for wives’ commands in the analysis that collapsed across both studies, and −2.01

*SD*s for rejections in the analysis that collapsed across both studies), but significantly negatively associated with initial satisfaction among spouses facing more severe problems (upper region of significance was +0.29

*SD* for wives’ commands in Study 1, +0.50

*SDs* for the summed index of negative behavior in Study 2, +1.38

*SD*s for wives’ rejections in Study 2, +0.14

*SD*s for wives’ commands in the analysis that collapsed across both studies, and +1.84

*SD*s for rejections in the analysis that collapsed across both studies). In other words, the spouses who were least happy initially were those who engaged in the most negative behavior and experienced the most severe problems.

Does problem severity moderate the association between initial negative behavior and subsequent changes in relationship satisfaction?

But the central question addressed by the current study was whether direct negative behaviors interacted with the severity of initial problems to account for

*changes* in marital satisfaction over time. Accordingly, regression coefficients and effect size estimates of the extent to which all behaviors interacted with initial problems to account for variance in the slopes estimated by

Equation 1 are presented in the bottom half of . As that section of the table reveals, consistent with predictions, a pattern of significant positive interactions emerged across all direct negative behaviors and across bith studies. Notably, in the analyses that collapsed across both studies and thus provided the most power, the summed index and every type of direct negativity significantly interacted with initial problem severity to predict changes in satisfaction over time. Also notably, consistent with predictions, none of the interactions involving indirect negative behavior reached statistical significance. Tests of the Study X Behavior X Problem interactions revealed that only one of the effects that emerged in the analyses that collapsed across both studies varied significantly across study: the interactive effect of the summed index of direct negative behaviors was significantly stronger in Study 1 than in Study 2. Nevertheless, that effect was significant in the separate analyses conducted separately on each study. Tests of the Sex X Behavior X Problem interactions revealed that a few of the interactive effects also varied by sex. Specifically, the Sex X Commands X Problem interaction in Study 1 was significant (

*B* = 21.83,

*SE* = 9.75,

*t*(131) = 2.24,

*p* < .05,

*r* = .19), such that the Commands X Problems interaction reported in only emerged among wives, whereas the corresponding effect for husbands did not reach significance,

*B* = 8.13,

*SE* = 11.37,

*t*(253) = 0.72,

*p* = .48,

*r* = .06. The Sex X Direct X Problems interaction in Study 2 was also significant (

*B* = 46.39,

*SE* = 20.04,

*t*(255) = 2.32,

*p* < .05,

*r* = .14), such that the Direct X Problems interaction reported in only emerged among wives, whereas the corresponding effect for husbands did not reach significance,

*B* = 3.95,

*SE* = 15.37,

*t*(255) = 0.26,

*p* > .50,

*r* = .02. And the Sex X Rejections X Problems interaction in Study 2 was also significant (

*B* = 258.53,

*SE* = 106.99,

*t*(253) = 2.42,

*p* < .05,

*r* = .15, such that the Rejection X Problems interaction reported in only emerged among wives, whereas the corresponding effect for husbands did not reach significance,

*B* = −45.11,

*SE* = 39.29,

*t*(253) = −1.15,

*p* = .25,

*r* = .07. However, when the data were collapsed across the two studies, thus maximizing the power to detect effects, only the Sex X Commands X Problems interaction was significant (

*B* = 22.44,

*SE* = 9.97,

*t*(395) = 2.25,

*p* < .05,

*r* = .11), such that the Commands X Problems interaction reported in only emerged among wives, whereas the corresponding effect for husbands did not reach significance (

*B* = 5.39,

*SE* = 10.95,

*t*(395) = 0.49,

*p* > .50,

*r* = .02).

To view the nature of these interactions, the predicted means of change in satisfaction were plotted for spouses one standard deviation above and below the mean on each negative behavior at all levels of problem severity. Plots depicting the results of the analyses that collapsed across both studies are depicted in the first column of . As can be seen there, consistent with predictions, higher levels of each measure of direct negative behaviors were associated with steeper declines in satisfaction in the context of marriages facing relatively minor problems, whereas those same negative behaviors were associated with more stable satisfaction in the context of marriages facing more severe problems. In contrast, but also consistent with predictions, as can be seen in the last plot of the first column of , consistent with the non-significant interactions involving indirect negative behavior, the association between indirect negative behaviors and changes in satisfaction did not vary even across all 6 standard deviations of the severity of the problems experienced in these samples.

To determine the regions at which each behavior was significantly negatively versus positively associated with changes in satisfaction, we identified the regions of significance of the simple effects of negative behavior according to procedures described by

Preacher et al (2006). Results of these simple slopes analyses are presented in . As the table reveals, each negative behavior demonstrated significant negative associations with changes in satisfaction among spouses facing relatively minor problems, with lower regions of significance ranging from 0.27

*SD*s above the mean to more than 3

*SD*s below the mean on initial problem severity, depending on the sample and behavior. Nevertheless, these same behaviors also demonstrated significant positive associations between negative behavior and changes in satisfaction among spouses facing more severe problems, with upper regions of significance ranging from 0.42

*SD*s above the mean to more than 3

*SD*s above the mean, depending on the sample and behavior. In other words, whether spouses’ direct negative behaviors demonstrate negative or positive effects on changes in their marital satisfaction over time depends on the severity of the problems those spouses face in their relationships.

| **Table 8**Regions of Significance of the Simple Effects of Direct Negative Behavior on Changes in Marital Satisfaction. |

Were the interactive effects of direct negative behaviors significantly different from the interactive effects of indirect negative behaviors?

Although the significance of the interactive effects of direct negative behavior and the non-significant interactive effects of indirect negative behavior are consistent with the prediction that direct and indirect negative behavior would differentially interact to predict satisfaction, an additional set of analyses was conducted to determine whether the interactive effect of each measure of direct negative behavior was *significantly* different from the interactive effect of the indirect negative behaviors. Specifically, the analyses that collapsed across the two studies to account for between-subject variance in changes in satisfaction, and thus offered the most power to detect effects, were repeated except this time the interaction involving indirect negative behavior was entered along with and compared to one of each of the four interactions involving each measure of direct negative behavior. The results of this analysis are reported in the top portion of . As can be seen by the *chi-square* values reported there, the interactive effects of the summed index, commands, and rejections on changes in satisfaction were significantly stronger than the interactive effect involving indirect negative behavior.

| **Table 9**Comparing the Interactive Effects of Behavior and Problem Severity on Changes in Satisfaction |

Were the interactive effects of behavior-focused negative behaviors significantly different from the interactive effects of character-focused negative behaviors?

Likewise, although significant interactive effects emerged for rejections, we conducted additional analyses to determine whether those interactive effects were weaker than the interactive effects of blames and commands, as suggested by previous theory. Specifically, the analyses that collapsed across the two studies to account for between-subject variance in changes in satisfaction, and thus offered the most power to detect effects, were repeated except this time the interaction involving blames was entered along with and compared to the interaction involving rejections in one model and the interaction involving wives' commands was entered along with and compared to the interaction involving wives' rejections in another model. The results of these analyses are reported in the bottom portion of . As can be seen by the *chi-square* values reported there, both comparisons failed to reach significance, suggesting the interactive effects of blames and commands on changes in satisfaction were not different from the interactive effects of rejections on changes in satisfaction.

Do direct negative behaviors interact with problem severity to predict changes in marital satisfaction through changes in the severity of the problems themselves?

Finally, we addressed the mechanism through which we expected exhibiting more direct negative behaviors in relationships characterized by more significant problems to predict more stable satisfaction -- changes in the severity of the problems themselves. We tested this mediational prediction by computing asymmetric confidence intervals for the mediated effect, following the procedures described by

MacKinnon, Fritz, Williams, & Lockwood (2007). Those procedures required two sets of additional analyses. First, we conducted analyses to estimate the effects of the interactions between each measure of negative behavior and initial problem severity on the predicted mediator -- changes in problem severity. Specifically, we repeated the analyses testing for the interactive effects of behavior on changes in satisfaction, except this time we substituted spouses’ reports of problem severity at each wave of measurement for their reports of satisfaction in

Equation 1 to form the following level-1 HLM model:

Second, we estimated the association between changes in problem severity and changes in marital satisfaction over time through the following level 1 model:

where we controlled for the effects of each interaction term on every parameter at level 2 (i.e., intercept, time slope, and problems slope). Given that the significance of the interactive effects of behavior on satisfaction did not vary across the two studies, we collapsed across the two studies for these analyses to be most parsimonious and to obtain the greatest power, but, as we did in the analyses on changes in satisfaction, we controlled for variance due to study and sex and also estimated the appropriate Study X Effect and Sex X Effect interactions to determine whether any effect varied across sex or study.

The results of the first set of additional analyses are reported in the first two columns of . As can be seen there, as was the case regarding changes in satisfaction, significant interactions emerged between initial problem severity and all indices of direct negative behavior. In contrast, the interaction between indirect negative behavior and initial problem severity did not reach significance (*B* = 0.05, *SE* = 0.56, *t*(398) = 0.08, *p* > .5, *r* = .00). Notably, none of these effects varied across sex or study.

| **Table 10**Changes in Problem Severity Accounting for Interactions between Negative Behavior and Changes in Satisfaction. |

Plots of all interactions are depicted in the second column of . As can be seen there, the patterns of these interactive effects of behavior on changes in problem severity were similar to the interactive effects of behavior on changes in satisfaction, except reversed. Specifically, more direct negative behaviors were associated with more substantial growth in problem severity over time among spouses facing relatively minor problems, whereas those same behaviors were associated with declines in problem severity among spouses facing more serious problems. As can be seen in the last plot in the second column of , in contrast, the association between indirect negative behavior and changes in problem severity did not vary across all 6 *SD*s of initial problem severity in these samples. Deconstructing these interactions statistically by identifying the regions of significance demonstrated that each negative behavior was indeed significantly positively associated with changes in problem severity among spouses facing relatively minor problems (lower region of significance was +0.71 *SD* for the summed index of direct negative behavior, +0.77 *SD* for blames, −1.23 *SD*s for commands, and < −3 *SD*s for rejections), but also significantly negatively associated with changes in problem severity among spouses facing more severe problems (upper region of significance was +2.71 *SD*s for the summed index of direct negative behavior, +2.57 for blames, +1.33 *SD*s for commands, and + 2.07 *SD*s for rejections).

The results of the second set of additional analyses that tested the effects of changes in problem severity on changes in satisfaction are presented in the second two columns of . As can be seen there, changes in problem severity were significantly negatively associated with changes in satisfaction even after controlling each interaction between negative behavior and initial problem severity, indicating that couples who experienced worsening problems over time became less satisfied over time. Notably, none of these effects varied across sex or study.

Finally, for each type of behavior, we multiplied these two components of the indirect effect together to obtain an estimate of the mediated effect and computed the corresponding confidence intervals to determine whether those effects were different from zero. The results of those analyses are reported in the last two columns of . As can be seen there, none of the confidence intervals contained zero, indicating that each mediated effect was different from zero. However, whereas the mediated effects involving direct negative behavior, blames, and commands were significant using a 95% confidence interval, the mediated effect of the Rejection X Problems effect was only significant using an 80% confidence interval, indicating it was only marginally significant using a one-tailed test. Notably, once changes in problem severity were controlled, interactions between initial problem severity and the summed index of negative behavior, blames, and rejections no longer significantly predicted changes in satisfaction (for direct negative behavior, *B* = 2.91, *SE* = 3.52, *t*(398) = 0.83, *p* > .40, *r* = .04; for blames, *B* = 1.99, *SE* = 10.14, *t*(398) = 0.20, *p* > .50, *r* = .01; for rejections, *B* = 2.10, *SE* = 7.20, *t*(398) = 0.29, *p* > .50, *r* = .01), suggesting that changes in problems fully mediated the interactive effects of behavior and initial problem severity on changes in satisfaction. Nevertheless, even after changes in problem severity were controlled, the interaction between wives’ commands and initial problem severity continued to significantly account for variance in changes in wives’ marital satisfaction over time, *B* = 19.37, *SE* = 5.91, *t*(395) = 3.28, *p* < .01, *r* = .16. Notably, because none of the components of these mediated effects varied across sex or study, none of the mediated effects varied across sex and study, including the mediated effect of the Commands X Problems interaction for which the direct effect on changes in satisfaction was only significant among wives.