Relationship satisfaction trajectories among low-income ethnic minority couples before and after a relationship education intervention

Theoretical framework

Bronfenbrenner’s (2005) bioecological theory of human development describes the continuous and bidirectional influence of the systems within which people live, learn, and grow. The final iteration of the bioecological model included four core concepts—person, process, context, and time. We include each component of the bioecological model as a framework to account for process (learning new relationship skills), person (individuals distinguished by sex and how exposure to new skills influences social resources for relationship satisfaction), context (individuals with low income navigating the mesosystem between the work environments and the couple/family relationship), and time (how new skills influence relationship satisfaction over time). The Prevention and Relationship Enhancement Program (PREP; Stanley et al., 2010) is a behavioral approach to psychoeducation that emphasizes communication skill building and adapted for couples with low income. Previous research provides theoretical and empirical support for the connection between relationship behaviors, such as communication and conflict management strategies, and satisfaction (Bloch et al., 2014; Johnson et al., 2005; Lavner & Bradbury, 2012) as well as acquisition of new relationship skills to improve couple satisfaction (Halford & Bodenmann, 2013; Williamson et al., 2016).Therefore, the current study is grounded in Bronfenbrenner’s (2005) model of human development influenced by process, person, context, and time as well as studies of behavioral couple intervention for the connection between relationship behavior, skill development, and satisfaction over time.

Efficacy of RE

In an effort to address the relational challenges experienced by couples and families with low income, the federal government designed funding opportunities for community-based RE initiatives to promote healthy relationships and teach skills for effective communication and conflict management. RE is typically a manualized psychoeducation intervention provided to couples interested in learning about communication and other relationship skills (Hawkins & Ooms, 2012) that often are utilized in a healthy and satisfying relationship. In fact, negative patterns in communication and relationship behavior predict a couple’s level of relationship satisfaction (Lavner & Bradbury, 2012). Subsequently, relationship satisfaction and distress are predictive of relationship dissolution or divorce (Solomon & Jackson, 2014). RE interventions demonstrate potential to ameliorate couple communication behavior and satisfaction in the relationship (Carlson, Barden, Daire, & Greene, 2014; Halford & Bodenmann, 2013; Williamson et al., 2016) and may help to mitigate unhealthy relationship trajectories in terms of dissatisfaction. Whether relational gains persist in a manner that keeps couples together long-term as intended by the initial funding initiative is less clear (Halford et al., 2001; Johnson & Bradbury, 2015; Markman et al., 2013; Rogge et al., 2013). The building stronger families (BSFs) and supporting healthy marriages (SHMs) randomized-controlled trials of RE yielded small to modest effects (Lundquist et al., 2014; Wood et al., 2014).

Overall, RE interventions for couples with low income yield small but positive relationship effects (d = .25) (Hawkins & Erickson, 2015) that occur dyadically (Braithwaite & Fincham, 2011; Halford & Wilson, 2009). McGill et al. (2016) studied pre-post RE outcomes, and results show dyadic effects between men and women; where women experienced the largest gains when at baseline both members of the couple indicated instability in the relationship and when the male partner reported low relationship quality. Additionally, RE interventions demonstrate efficacy among couples with low income to reduce individual distress (Carlson, Daire, & Bai, 2014) and improve parental alliance (Carlson, Barden, Daire, & Swartz, 2014)—both contributors to relationship quality. While originally posited as a prevention-focused intervention for nondistressed couples, Bradford et al. (2015) note the trend for more relationally distressed couples to attend federal RE programs. In fact, couples presenting with higher levels of distress seem to benefit most from RE intervention (Carlson et al., 2017; Quirk et al., 2014). Evidence exists in support of RE as an effective intervention to improve couple satisfaction, distress, and relationship quality for couples with low income. Few studies examine the longitudinal and dyadic pattern of post-RE outcomes, and researchers highlight the need for multiple assessment points after RE intervention to better capture change and couple-level moderation of change (Markman & Rhoades, 2012; Wadsworth & Markman, 2012).

Longitudinal patterns in relational satisfaction

Husbands and wives of average income report significant declines in relationship quality beginning in the first year of marriage; yet it is unclear whether relationship quality was high or increasing before marriage and then declined early in the marriage. Longitudinal examinations of relationship quality yield no agreed upon pattern or trajectory for relationship quality. In fact, researchers tracked the pattern for relationship quality over time and results indicated a U-shaped, curvilinear, or cubic pattern (Kurdek, 1999; Orbuch et al., 1996; Vaillant & Vaillant, 1993). Although patterns are inconsistent, relationships typically experience periods of increased distress and risk of dissolution following the birth of a child and during early parenthood (Doss et al., 2009; Keizer & Schenk, 2012; Mitnick et al., 2009), early in the relationship, and later in a marriage after the couple’s first child turns 14 years of age (Gottman & Levenson, 2000). In contrast, relationship satisfaction findings are consistent for couples with low income, unemployment, and fewer years of education (Amato, 2010; Charles et al., 2006; Conger et al., 2010). These couples are most at risk for divorce, relationship distress, and poor relational outcomes.

More recent literature (Williamson & Lavner, 2019) supports that low-income newlywed couples have relatively stable, satisfying relationships in the first 5 years of marriage. It was only those who started with a low relationship satisfaction that ended up with a big decline and higher rate of marriage dissolution during the early marriage years. Similarly, Proulx et al. (2017) found that individuals who start off with lower satisfaction are more likely to experience continued poor marital quality and marital dissolution. Those who start off high in marital quality are likely to remain stable or experience minimal decline. Lavner and Bradbury (2010) found contrary to honeymoon-is-over effect because as many as four in five spouses in their sample reported relatively high levels of satisfaction and small declines in satisfaction over the first 4 years of marriage. In another study, the majority of couples reported high positive marital relations and low negative marital relations over the year after the birth of the second child, a time generally considered as “at-risk” for couples. However, these couples had little difficulty managing the transition to the second child (Volling et al., 2015).

RE intervention and satisfaction trajectories

Halford and Wilson (2009) examined the longitudinal effect of RE intervention on measures of relationship satisfaction among 66 primarily Caucasian and well-educated Australian couples. Results suggested that across 4 years’ postintervention, a mean decline per year in satisfaction occurred for men (small effect) and for women (moderate effect). Similarly, analysis of the SHM yearlong intervention with low-income couples suggested that gains obtained by couples most at risk may reduce over time up to the 30-month follow-up point (Gubits et al., 2014). A general decline in satisfaction seems to occur for couples post-RE intervention. However, the homogenous sample demographics for Halford and Wilson’s (2009) study and variance in intervention curriculum, format, and dosage for SHM leave room for further inquiry for the longitudinal effects from RE.

In contrast, Halford and Bodenmann (2013) reviewed 17 studies that included follow-up of at least 12-month postintervention for couples early in their relationship, couples transitioning to parenthood, and couples in established relationships. Overall, 14 of the 17 studies determined that RE facilitated maintenance of relationship satisfaction for couples presenting with higher levels of risk for future relational problems or dissolution. Another study found at 3-year follow-up that participation in RE reduced rates of divorce from 59% of couples divorced in the control group to 51% in the treatment group (Wood et al., 2012). Participation in RE seemed to sustain relational gains for couples at varying lengths of time in their current relationship. Hawkins et al. (2008) in their meta-analysis determined that gains in relationship satisfaction following RE persisted up to 3–6 months after completion (d = .31). Short-term relationship gains seemed to be maintained for couples attending RE; yet initial studies of RE focused predominantly on homogenous groups lacking racial and economic diversity. Therefore, less is known about the longitudinal outcomes of RE for racially diverse couples with low income.

Amato (2014) evaluated how social and economic disadvantage moderated the effects of a 15-month RE program, BSF. The interaction of disadvantage and RE-treatment condition did not significantly predict whether couples stayed together, yet predicted perceived quality of the relationship. Participation in RE seemed to mitigate harmful effects to relationship quality for couples with lower income and less social support.

The aims of the present study were four-fold: (a) to examine couples’ trajectories of relationship satisfaction before and after an intervention of RE up to 120 days (i.e., preintervention enrollment; post-15-hr PREP intervention; 30, 60, and 90 days after completion of PREP); (b) to examine the influence of men and women on each other’s relationship satisfaction growth over time; (c) to investigate the effects of economic and social indicators of disadvantage (i.e., employment status, income, education, length of relationship) on the trajectories of couples’ relationship growth; and (d) to investigate the effect of baseline levels of relational distress to relationship growth after intervention.

Method

Participants

The current sample included 728 heterosexual couples, for a total sample size of 1,456. Thus, 728 women and 728 men participated in the Within Our Reach Plus (WOR Plus) program in Year 2 (290 couples), Year 3 (239 couples), or Year 4 (199 couples). Project staff utilized active (e.g., face-to-face interaction) and passive (e.g., flyers, project website) recruitment strategies to inform potential participants about the program in partnership with local health and Women, Infants, and Children departments, workforce development offices, churches, libraries, and other varied social service agencies. Participants enrolled in the program through attendance on the first night of an RE workshop where they completed a group intake that included informed consent and pen-and-paper completion of study assessment instruments.

The majority of this study’s participants were couples of primarily low income and ethnic minority backgrounds. According to the Population Reference Bureau (PRB), poverty thresholds are updated annually by the U.S. Census Bureau and defined by a specified minimum dollar amount necessary for families to meet their basic needs. In 2018, the poverty threshold for a family of four was US$25,465 (PRB, 2018). Male study participants had a mean monthly income of US$1,738 per month and female participants US$925 per month (see Tables 1 and 2 for additional demographic information). There were 195 male participants (26.8%) unemployed and 528 (72.5%) employed at the baseline with 0.7% (n = 5) missing data. There were 391 female participants (53.7%) unemployed and 333 (45.7%) employed at the baseline with 0.5% (n = 4) missing data. Of the male participants, 74.9% (n = 545) reported a racial or ethnic minority background and 17.3% (n = 126) were nonminority represented by White, non-Hispanic population. Approximately 7.8% (n = 57) participants were missing data on either race or ethnicity status. Of the female participants, 76.2% (n = 555) reported a racial or ethnic minority background and 18.4% (n = 134) were nonminority. Approximately 5.4% of females (n = 39) were missing data on either race or ethnicity status. The percentage of the sample that is married is 63% among men and 64% among women.

OPEN IN VIEWER

Table 1.

Participant demographics.

	N	M	SD	Range
Male
Age	725	36.67	9.60	18–67
Years of education	704	13.53	2.56	2–23
Length of relationship (months)	723	100.89	81.05	2–465
Average monthly income	673	1,738.46	1,922.29	0–30,000
Female
Age	727	34.07	8.34	18–61
Years of education	713	14.08	2.55	2–24
Length of relationship (months)	727	100.85	81.35	1–465
Average monthly income	662	924.92	1,277.31	0–15,000

OPEN IN VIEWER

Table 2.

Participant racial and ethnic distribution.

	N	%
Male
American Indian/Alaska native	17	2.3
Asian	6	0.8
Black/African American	192	26.4
Native Hawaiian/Other Pacific Islander	4	0.5
White	256	35.2
Other	206	28.3
Missing	47	6.5
Female
American Indian/Alaska native	8	1.1
Asian	18	2.5
Black/African American	182	25.0
Native Hawaiian/Other Pacific Islander	0	0
White	278	38.2
Other	209	28.7
Missing	33	4.5

Note. “Other” category was not further classified.

Researchers collected five waves of data: preintervention assessment (Wave 1), after-intervention assessment (Wave 2), 30-day after-intervention assessment (Wave 3), 60-day after-intervention assessment (Wave 4), and 90-day after-intervention assessment (Wave 5) for a total of 120 days of observation. Researchers collected Wave 2 of the data at the final night of the RE workshop. For Waves 3–5, project case managers contacted participants via phone to complete assessment instruments telephonically. Wave 1 data include 723 males and 725 females; Wave 2 includes 599 male (83% completeness) and 602 female (83% completeness); Wave 3 includes 357 male (49% completeness) and 391 female (54% completeness); Wave 4 includes 312 male (43% completeness) and 317 female (44% completeness); and Wave 5 includes 272 male (38% completeness) and 283 female (39% completeness). The retention rate is about 38% through Wave 1 to Wave 5. Women’s baseline satisfaction was significantly associated with their missingness at Wave 5 (p < .0005), but this was not happening to men. Specifically, the less satisfied female individuals were more likely to drop from the study at Wave 5. Comparative to other community-based RE programming, we found a similar retention rate (i.e., 67% completed a 12-week follow-up after a higher effort couples RE intervention; Busby et al., 2015). We also tested whether the missingness at Wave 5, which had the largest missingness, was related to any pretest characteristics including participants’ age, length of relationship, years of education, and income, using independent samples t-tests with an α level set at .0125 to control for inflated type I errors. ¹ Length of relationship (male: p = .0002; female: p = .002), income (female: p = .004), and years of education (male: p = .0002) significantly predicted missingness and thus were included into the model as predictors along with employment status which was the expected target covariate. In this way, we accounted for the variables that predicted missingness and the variables did not bias the estimates.

Measures and variables

On the first night of an RE workshop, participants completed several instruments and an informed consent approved by the university institutional review board. Project staff led the group intake and provided instructions for each instrument. Instruments relevant to the current analysis include a demographic Intake Form and the Relationship Assessment Scale (RAS; Hendrick, 1988; Hendrick et al., 1998).

Intake form

The Project TOGETHER researchers developed a 60-item intake form to collect participant demographic information (age, relationship status, educational background, employment status, income, data related to needs and barriers for the case management component of the project, contact information, and contact preferences).

Relationship Assessment Scale

The RAS is a 7-item instrument that measures general relationship satisfaction and relationship distress. The 7 items are presented on a Likert-type scale ranging from 1 to 5. Some questions include “How much do you love your partner?” and “How well does your partner meet your needs?” This instrument has shown a correlation of .80 with the Dyadic Adjustment Scale, a well-used instrument to measure relationship satisfaction (Hendrick, 1988; Spanier, 1976). According to Hendrick et al. (1998), a higher average score of 4.0 indicates greater relationship satisfaction, while an average score closer to 3.5 indicates relationship distress for men, and an average score ranging from 3.0 to 3.5 for women indicates relationship distress and overall dissatisfaction (Hendrick et al., 1998). Thus, both men and women in the current investigation reported fair amounts of relationship satisfaction at intake (i.e., 270 couples [37.1%] reported preintervention RAS scores below the established cutoff). The reliability of men’s RAS score across five waves was .90, .87, .87, .88, and .88. The reliability of women’s RAS score across five waves was .92, .91, .89, .91, and .91. Table 3 presents RAS descriptive statistics.

OPEN IN VIEWER

Table 3.

RAS descriptive statistics.

	N	M	SD	Range
Male
RAS average—Wave 1	723	3.90	.78	1.14–5
RAS average—Wave 2	599	4.09	.65	1–5
RAS average—Wave 3	357	4.21	.67	1–5
RAS average—Wave 4	312	4.20	.68	1.29–5
RAS average—Wave 5	272	4.24	.66	1.29–5
Female
RAS average—Wave 1	725	3.66	.88	1-5
RAS average—Wave 2	602	3.93	.79	1-5
RAS average—Wave 3	391	4.086	.73	1.57-5
RAS average—Wave 4	317	4.09	.80	1.29-5
RAS average—Wave 5	283	4.09	.76	1-5

Note. RAS = relationship assessment scale.

Data analyses

The researchers computed descriptive statistics first with SPSS v24, and then adopted latent growth curve (LGC) modeling to examine couples’ relationship satisfaction growth in Mplus v7. LGC modeling within the framework of structural equation modeling is considered one of the most powerful approaches to analyze growth trajectories because it permits the examination of both within- and between-person differences in change (Little, 2013; Wei et al., 2015). The first step in building an LGC model is to examine the within-person growth trajectory. In the current study, a linear model initially included two growth parameters: (a) an intercept parameter that represented a female partner’s or male partner’s RAS score at pre-assessment and (b) a slope parameter that represented a female partner or male partner’s rate of growth from pre-assessment to 120 days after enrollment in the RE. Figure 1 shows a visual representation of this LGC model. The next step is to interpret between-individual differences in change through examination of the means and variances of the parameters (e.g., intercept and slope). While the means provide information about the average intercept and slope values, the variances imply individual differences in intercept and slope values (Byrne, 2012). The researchers addressed missing data with a full information maximum likelihood estimation in Mplus, in addition to including key predictors of missingness as described above. Hu and Bentler’s (1999) combinational criteria for acceptable model were adopted: (a) comparative fit index (CFI) or Tucker–Lewis index (TLI) > .95 and standardized root mean square residual (SRMR) < .09, or (b) root mean square error of approximation (RMSEA) < .05 and SRMR < .06.

OPEN IN VIEWER

Figure 1.

Latent growth model. Note. Solid lines represent significant correlations/predictions; dashed lines represent nonsignificant correlations/predictions. RAS = relationship assessment scale.

Following the structure and approach of Acock (2013), the researchers simultaneously fit separate growth curves for both members of each couple. This is a common approach when analyzing data from romantic relationships or other couples, as it accounts for the shared variance within each couple pairing while also estimating separate growth curves for each member. To account for the nested relationship, corresponding errors in each growth curve were logically correlated. Nonrandom errors were correlated, and the intercepts and slopes (e.g., i1–i2; s1–s2) were allowed to covary. We explicitly allowed the intercept and slope to be correlated (e.g., i1–s1; i1–s2) based on Acock’s approach. Finally, the researchers added a direct effect from the male’s intercept to the female’s slope and from the female’s intercept to the male’s slope. Figure 1 shows detailed demonstration.

Results

First, the researchers estimated a four-factor LGC model for the linear growth of relationship satisfaction (see Figure 1). The first latent factor defined the intercept of the growth curve for males and the researchers set all time loadings to 1 (Byrne, 2012). The second latent factor defined the slope of the growth curve for males and the researchers set factor loadings to represent the linear change over the five waves of measurement—0, 1, 2, 3, and 4 in the metric of months. The third latent factor defined the intercept of the growth curve for females, and the researchers set all time loadings to 1. The fourth latent factor defined the slope of the growth curve for females and the researchers set factor loadings to represent the linear change over the five waves of measurement—0, 1, 2, 3, and 4 in the metric of months.

R1: What is the nature of couples’ relationship satisfaction trajectories before and after an RE intervention up to 120 days?

In the original model, we examined the nature of couples’ relationship satisfaction trajectories before and after an RE intervention up to 120 days. We first checked the no change (strict stability) model which produced a poor model fit, χ²(45) = 377.42, RMSEA = .10, SRMR = .10. The dyadic linear model for all the couples demonstrated good model fit, χ²(36) = 154.82, p < .0001, CFI = .97, TLI = .96, RMSEA = .07, confidence interval (CI) 90% [.06, .08], SRMR = .07. Table 4 summarizes all growth model estimates. We also examined measurement invariance within dyads. We constrained the estimated slope and intercept to be equal for each couple over time, which produced a poor model fit, χ² (38) = 183.40, RMSEA = .07, SRMR = .12. Thus, the latent slope and intercept aren’t invariant within each couple.

OPEN IN VIEWER

Table 4.

Fit indices for all estimated models.

Models	χ2	df	CFI/TLI	RMSEA (CI 90%)	SRMR
Original model
Unconditional model	154.82**	36	.97/.96	.07 [.06, .08]	.07
Conditional model	232.04**	96	.97/.95	.04 [.04, .05]	.05
Nondistressed couples model
Unconditional model	87.10**	36	.97/.97	.056 [.04, .07]	.09
Distressed couples model
Unconditional model	60.67*	30	.96/.91	.06 [.04, .08]	.07

Note. CFI = comparative fit index; TLI = Tucker–Lewis index; SRMR = standardized root mean square residual; RMSEA = root mean square error of approximation; CI = confidence interval.

*p < .05; **p < .001.

Unstandardized estimates indicated the value of initial levels of relationship satisfaction, given the significant positive mean of the intercept for males (M = 3.95, p < .001) and females (M = 3.75, p < .001). Men’s baseline satisfaction was higher than women’s baseline satisfaction, but both were above 3.5, the cutoff for interpretation of scores as relationally distressed (e.g., Hendrick et al., 1998). In this original model, both men and women expressed a fair degree of satisfaction with their couple relationship at baseline.

The standardized estimates were applied to illustrate the slope factors due to the advantage of standardized estimates in interpreting and comparing results. The significant positive mean of the slope factor for men (M = 0.24, p < .001) and women (M = 0.19, p = .001) indicated an overall positive linear growth of relationship satisfaction. So, both men and women continuously kept growing in relationship satisfaction across 120 days and up to 90 days after the intervention (see Figure 2 for the growth curve of couples). The men’s slope value was larger than the women’s. Thus, men had a steeper slope, indicating that they may experience a larger growth in relationship satisfaction over time.

OPEN IN VIEWER

Figure 2.

Growth curve for all couples (standardized results).

In addition, there was significant variance in the intercept factor of both men (s ² = .42, p < .001) and women (s ² = .59, p < .001), which reflected the variation of initial levels of satisfaction among individuals. Standardized estimates also indicated significant variance in the slope factor for males (s ² = .09, p = .04), showing substantial variability in men’s growth patterns. The pattern of variability in growth did not apply to women. In a future study, we would apply a group-based trajectory model to examine whether any patterns of variability exist in men’s growth trajectory.

R2: What is the dyadic influence of men and women for each other’s relationship satisfaction growth over time?

Second, we examined the dyadic influence of men and women on each other’s relationship satisfaction growth over time. Standardized estimates indicated partner effects for both men and women. Specifically, women’s growth rate (s2) negatively regressed on men’s initial status (i1; r = −.20, p = .03) and men’s growth rate (s1) negatively regressed on women’s initial status (i2; r = −.30, p < .001). In each case, a negative relationship existed between an individual’s baseline satisfaction and their partner’s growth rate of satisfaction. The less satisfied the male or female partner before the intervention, the larger the rate of growth observed for their female or male partner after the intervention. Conversely, the more satisfied a person was before the intervention, the smaller their partner’s rate of growth after the intervention.

We also found an actor effect for men, not women, where the male’s initial satisfaction status (i1) negatively correlated with their own growth rate (s1; r = −.18, p = .001). The male actor effect suggested a negative relationship between baseline satisfaction and growth of satisfaction over time among men. The less satisfied the male was before the intervention, the faster their rate of growth after the intervention and over time.

Female’s initial level of satisfaction at baseline (i2) positively correlated with male’s initial status (i1; r = .76, p < .001). Thus, before attending RE, the more satisfied the female partner, the more satisfied the male partner and vice versa. Similarly, a positive correlation existed between male (s1) and female (s2) growth rates (r = .54, p < .001). From the intake through 120-day post-RE enrollment, the faster the growth of relationship satisfaction for one person, the faster the growth of their partner. Likewise, the slower the growth rate of satisfaction a person experienced, the slower the growth rate of their partner.

We calculated the within-person Cohen’s d for the growth of relationship satisfaction to evaluate the effect sizes—the meaningfulness of the changes across Wave 1 through Wave 5 for males and females. The effect sizes for males and females’ growth are .47 and .52, respectively. The medium to large effect sizes demonstrated the effectiveness of the RE intervention for couples’ relationship satisfaction growth over time.

R3: Does baseline employment status, income, years of education, and length of relationship have any impact on the growth of couples’ relationship satisfaction?

Third, we examined the influence of baseline employment status, income, years of education, and length of relationship on the growth of couples’ relationship satisfaction. To do so, we regressed the unconditional LGC model on four covariates: employment status, length of relationship, income, and years of education. The conditional model demonstrated good model fit but was not superior than the unconditional LGC model with an increased value on χ², χ²(96) = 232.04, p < .0001, CFI = .97, TLI = .95, RMSEA = .04, CI 90% [.04, .05], SRMR = .05. The conditional model yielded two significant covariates: female unemployment and male length of relationship. Female unemployment influenced their own baseline relationship satisfaction (r = .17, p = .02), which indicated females were more satisfied in the couple relationship if they were not going out to work. Male length of relationship slightly influenced their own baseline relationship satisfaction (r = .006, p = .04), indicating that males in couple relationships for a longer period of time were more satisfied.

R4: Does baseline relational distress influence the growth of relationship satisfaction after RE intervention?

Fourth, we divided the total sample into two subsamples based on participants’ RAS baseline average score. Given prior research which suggests that initially distressed couples experience the largest gains from RE (e.g., Carlson et al., 2017; Quirk et al., 2014), it was important to examine potential differences in model fit between distressed and nondistressed couples. The researchers designated a couple as “distressed” if the couple met either condition for (a) male’s baseline RAS less than or equal to 3.5 or (b) female’s baseline RAS less than or equal to 3.0. On the contrary, the researchers designated a couple as “nondistressed” if the couple met condition for (a) male’s baseline RAS larger than 3.5 and (b) female’s baseline RAS larger than 3.0. As a result, the researchers categorized 270 couples into the distressed group and 453 couples into the nondistressed group with 0.7% missing data (n = 5). No significant differences in demographics existed between the groups (i.e., distressed, nondistressed) for men or women in terms of their age, income, years of education, or length of relationship using independent samples t-tests with an α level set at .0125 to control for inflated type I errors.

Discussion

Results confirm the linear nature of couple’s trajectories in relationship satisfaction. Overall, couples experienced improvements in relationship satisfaction and demonstrated growth from their baseline levels pre-RE intervention across 120 days. These findings add to a growing body of literature for the effectiveness of RE interventions to positively influence relationship satisfaction among couples with low income (Hawkins & Erickson, 2015; Hawkins & Fackrell, 2010). The demographic composition of the study participants (i.e., majority of couples identified a racial or ethnic minority background and low-income status) and dyadic nature of the analysis provide new information for the effectiveness of RE over time with a diverse sample of couples. When we examined couple trajectories separately based on baseline level of distress (i.e., distressed and nondistressed couples), results were less conclusive for the durability of gains from RE over time. Therefore, we provide recommendations for future research and practice to better understand the complexities that seem to influence relationship satisfaction trajectories post-RE completion.

First, a negative relationship existed between a person’s initial relationship satisfaction status and their growth in satisfaction over time. Couples demonstrated a larger rate of relationship satisfaction growth when they reported lower initial levels, and a couples’ initial level of relationship satisfaction could be considered a predictor of their potential growth in satisfaction after RE intervention. Yet, the lower rate of growth among couples with higher initial satisfaction could relate to a type of “ceiling effect.” If couples start with a high degree of satisfaction, it may be more difficult to detect growth or other positive relational outcomes which may exist. Ceiling effects occur in a variety of applied contexts in examination of growth trajectories (e.g., Bradford et al., 2017) and may have influenced the differences in growth observed in the current study.

Second, we found significant individual variance in the two latent factors of the unconditional LGC model. Individuals varied significantly in their initial status and growth patterns of relationship satisfaction. Although in general couples seemed to benefit from RE, not all couples benefited to the same degree. Consistent with prior studies of employment and financial stress (e.g., Gudmunson et al., 2007; Sarmiento & Cardemil, 2009; Williamson et al., 2013), in the current analysis, female unemployment predicted higher levels of baseline relationship satisfaction in the conditional model—an unexpected finding.

We did not anticipate greater baseline relationship satisfaction among unemployed women. However, differential effects from unemployment by sex exist for mental health and overall well-being, where women often fare better during unemployment than their male counterparts (Strandh et al., 2013; van der Meer, 2014). Similarly, women hold more positive beliefs about the effects of acquired job loss for themselves than men (Michniewicz et al., 2014). Study eligibility criteria included that couples denote an active parental status for a child under the age of 18 and current co-parenting responsibility with their partner. So, female unemployment status may have been selected by the couple, rather than acquired through job loss, to accommodate family roles and responsibilities for caregiving. Another potential justification for the unexpected unemployment finding was that women who worked and had to be involved in parental tasks might be weaker in health, which could negatively impact their relationship quality as well. Results also indicated that length of relationship had a slight impact on men’s baseline relationship satisfaction. Men felt more satisfied in the couple relationship when the relationship was longer. Prior studies report that men, in general, tend to experience slightly higher levels of relationship satisfaction compared to their female counterparts. Further, relationship satisfaction and length of relationship may have a bidirectional relationship where one influences the other (i.e., men staying in their relationship once they feel satisfied and vice versa). Men also have longer life expectancies and experience health benefits as a result of being in long-term, satisfying, and committed relationships. These significant and tangible benefits for men may also contribute to continued relationship satisfaction and thus longer lasting relationships.

Surprisingly, baseline income and years of education did not significantly impact the couples’ relationship satisfaction, possibly because their daily life needs were at least minimally met. For example, some couples accessed and utilized community and state social service resources such as identified Project TOGETHER recruitment sites. These couples may have been receiving assistance that helped fulfill their basic needs (i.e., for food, shelter, childcare, health care, career training and development, and assistance with bill payments), thus reducing the negative relational impact and distress often associated with these factors.

Prior to RE intervention, a positive correlation existed between male and female baseline relationship satisfaction—at the start of RE, the more satisfied the male partner, the more satisfied the female partner and vice versa. A positive relationship also existed for male and female growth rates in relationship satisfaction from intake through 120-day post-RE enrollment. Thus, although some research suggests that partners may differ in their perceptions of current relational health (McGill et al., 2016), couples reported a moderate degree of agreement for current relationship distress as well as improvement over time. Both men and women seemed to benefit from RE participation and included bidirectional influence between one another. RE practitioners may therefore continue to offer services, on the whole and to both distressed and nondistressed couples, without concerns of negative or harmful effects.

Moreover, the dyadic analysis illuminated significant partner effects for males and females from baseline satisfaction scores to a partner’s growth over time. In both cases, a person’s initial level of satisfaction negatively predicted their partner’s rate of growth. Here, the less satisfied a person was before the intervention, the larger their partner’s rate of growth from the intervention. Conversely, the more satisfied a person was before the intervention, the smaller their partner’s rate of growth. The partner effect between baseline and post-RE growth in satisfaction is congruent with prior research with pre-post analysis. McGill and colleagues (2016) found the greatest post-RE relational gains occurred for females whose partner reported a low baseline level of relationship quality. However, we did not find sex-based differences in partner effects among couples. Similarly, researchers demonstrated related partner effects where a partner’s relationship self-regulation strategies predicted marital satisfaction (Hardy et al., 2015) and a partner’s social integration predicted dedication to the relationship (Owen et al., 2012). Thus, dyadic associations between members of a couple seem important for relationship satisfaction and contribute to benefits derived from RE for couples.

Finally, we were unable to confirm the linear nature of couple’s satisfaction trajectories when we separated the original sample by their baseline levels of distress (i.e., distressed and nondistressed couples). Thus, our model may not capture moderators that influence baseline distress or the process of change over time following RE intervention. Additional moderators in the intervention process literature such as hope (Hawkins et al., 2017), dyadic coping (Bodenmann et al., 2009), or relationship commitment (Rauer et al., 2014) may differentially influence the trajectory of satisfaction for the distressed and nondistressed couples. Researchers should continue to explore RE moderators and factors that may influence benefits derived from participation in RE, and future RE programming may include tailored recommendations and interventions for couples based on presenting characteristics found influential to sustain positive changes. Practical applications to consider are RE intervention formats that include additional services to support and sustain changes.

The results from nondistressed couples were congruent with the results from Bradford et al. (2017) which found little to no significant improvement from pre- to post-RE in relationship quality, commitment, or interaction quality regardless of program duration. In their study and ours, we acknowledge the potential for ceiling effects to limit detection of relationship gains. Regarding the distressed couples’ results, as greater numbers of distressed couples attend RE (Bradford et al., 2015), more research is needed to understand satisfaction growth in distressed couples and program modifications intended to maintain relationship gains.

Conclusion

Study findings provide an innovative perspective for the dyadic and longitudinal nature of relationship satisfaction growth post-RE with a diverse sample of couples. Overall, couples’ growth followed a linear and positive trajectory. A correlation existed between male and female partner’s baseline levels of relationship satisfaction and growth rate over time. Couples’ initial relationship satisfaction could be considered as a negative predictor of their satisfaction growth over time after an RE intervention. The significant interindividual differences on initial status and growth patterns suggest the importance of moderating variables that could explain interindividual differences; however, the addition of four covariates did not successfully explain the variances in this study. A clear dyadic influence of men and women on each other’s relationship satisfaction growth exists, and male partners seem to have experienced a larger growth in relationship satisfaction over time through RE. Neither linear nor quadratic trajectory was found in distressed or nondistressed couples. More research is needed to replicate and better understand the benefits from RE for distressed and nondistressed couples. Additionally, more research is needed to explore potential moderators and program services that may better address and sustain changes for couples based on presenting distress and couple characteristics.