Research on Marital Satisfaction and Stability in the 2010s: Challenging Conventional Wisdom

Abstract

Although getting married is no longer a requirement for social acceptance, most people do marry in their lifetimes, and couples across the socioeconomic spectrum wish their marriages to be satisfying and long-lasting. This review evaluates the past decade of research on the determinants of satisfaction and stability in marriage, concluding that the scholarship of the past ten years has undermined three assumptions that were formerly accepted as conventional wisdom. First, research exploiting methods like latent class growth analyses reveal that, for most couples, marital satisfaction does not decline over time but in fact remains relatively stable for long periods. Second, contrary to predictions of behavioral models of marriage, negative communication between spouses can be difficult to change, does not necessarily lead to more satisfying relationships when it is changed, and does not always predict distress in the first place. Third, dyadic processes that are reliably adaptive for middle-class and more affluent couples may operate differently in lower-income couples, suggesting that influential models of marriage may not generalize to couples living in diverse environments. Thus, the accumulated research of the last ten years indicates that the tasks of understanding and promoting marital satisfaction and stability are more complex than we appreciated at the start of the decade, raising important questions that beg to be answered in the years ahead.

Despite what you may have heard, the institution of marriage is alive and well. It is true that marriage is no longer a requirement for social acceptance and advancement (Cherlin, 2004). It is also true that more of the population of the United States is currently unmarried than ever before (United States Census Bureau, 2017), because people have been cohabiting more frequently (Rosenfeld & Roesler, 2019), entering marriage later (Kreider & Ellis, 2011), and yet divorcing at similar rates (Cohen, 2016). Notwithstanding these trends, however, the vast majority of people across levels of socioeconomic status still aspire to marry (Kuo & Raley, 2016; McDonald, Pini, Bailey, & Price, 2011), and well over 90% of people do marry at some point in their lives (Kiersz, 2017; Yau, 2015).

The persistence of marriage makes sense considering that the intimate relationship enshrined by marriage continues to be vital to well-being. When marriages are stable and fulfilling, spouses are healthier (Robles, Slatcher, Trombello, & McGinn, 2014), happier (Be, Whisman, & Uebelacker, 2013), and live longer (Whisman, Gilmour, & Salinger, 2018); when the relationship falters, other pillars of well-being are also at risk (Holt-Lunstad, Smith, Baker, Harris, & Stephenson, 2015). The evidence is stronger than ever that the critical driver of these effects is the quality and longevity of the connection between the partners (Musick & Bumpass, 2012). Spouses themselves understand this: across the socioeconomic spectrum, people do not merely want to get married, they want to marry someone they love, and to remain in love together for the rest of their lives (Geiger & Livingston, 2019; Kefalas, Furstenberg, Carr, & Napolitano, 2011; Trail & Karney, 2012; Van Hooff, 2013).

Yet, as the goals and benefits of a happy marriage endure, so do the challenges, reflected in the fact that nearly 50% of first marriages will end in divorce (Kreider & Ellis, 2011). How do people sustain close intimate connections within marriage over time? Why is this so difficult for so many well-intentioned couples? When we first reviewed the accumulated literature addressing these questions twenty-five years ago (Karney & Bradbury, 1995), research on how marriages remain stable or change over time was relatively young, and methods for addressing change were still being developed. The subsequent decades have seen an explosion of research and the blossoming of new methods for conducting that research. In this review, we focus on efforts within the last ten years to account for stability and change in marriage, in hopes of setting the stage for the decade ahead. Further, because the bulk of research on marriage continues to focus on different-sex couples (despite the legalization of same-sex marriage halfway through the decade), this review will likewise focus primarily on research on different-sex couples. Yet even within those restrictions, we cannot do justice to the breadth and depth of contemporary research on the determinants of marital quality and stability (for recent reviews that complement this one, see Finkel, Simpson, & Eastwick, 2017; McNulty, 2016; Proulx, Ermer, & Kanter, 2017; Robles et al., 2014). Instead, our current goal is more narrow: to identify specific areas where research over the last ten years has overturned, or at least called into question, long-standing and widely-held ideas about marital quality and stability. In particular, this review will focus on how the research of the last decade has undermined previous assumptions about how marriages change, how that change comes about, and to what extent our conclusions about these issues generalize across populations.

Trajectories of Marital Satisfaction: Most Couples Don’t Decline Much

No one marries intending to get divorced. The frequency of divorce therefore suggests that the initial connection and optimism that characterizes spouses on their wedding day is hard for many couples to maintain. Indeed, one of the most consistent results in longitudinal research on marriage is that, on average, marital satisfaction declines significantly over time (e.g., Kurdek, 1999; Umberson, Williams, Powers, Chen, & Campbell, 2005). This dispiriting pattern, which has been described as a “typical honeymoon then years of blandness” (Aron, Norman, Aron, & Lewandowski, 2002, p. 182), is as close to a truism as exists in marital research: marital satisfaction declines on average in nearly every longitudinal study of marriage ever conducted, and in newer and older marriages alike (VanLaningham, Johnson, & Amato, 2001). Given that couples uniformly wish to preserve their initial happiness, the ubiquity of these declines is the mystery that most longitudinal studies of marriage are designed to explain.

Empirical Advances

Over the last decade, an accumulating body of research indicates that this truism may not be as true as previously thought. The insight guiding this new work is that emphasizing the average trend in marital satisfaction has masked theoretically interesting variability across people. Even the more sophisticated approaches to addressing multiwave longitudinal data on marriage (e.g., growth curve modeling; T. E. Duncan, Duncan, & Strycker, 2013) estimate average parameters of change (i.e., intercepts and slopes of marital satisfaction over time), and then use other predictors to account for variability around those average parameters. In contrast, alternative approaches to analyzing longitudinal data treat the variability itself as something worth describing in detail. Whereas procedures like growth curve modeling are variable-centered, the newer approaches are person-centered, seeking to identify distinct groups or clusters of people that are more similar to each other than to members of other groups (Jobe-Shields, Andrews, Parra, & Williams, 2015). Techniques like semi-parametric mixture modeling (Nagin, 1999) and latent class growth analysis (Jung & Wickrama, 2008) are among several allowing researchers to move beyond characterizing average trajectories toward describing different types of trajectories more or less prevalent within a sample or population.

When these approaches were applied to longitudinal data on marriage for the first time, the existing literature supported the strong prediction that, even if distinct classes of change in marital satisfaction could be identified, most people would fall into groups characterized by varying degrees of decline. That’s not what happened. Analyses of 20 years of data from couples in the Marital Instability over the Life Course study did indeed reveal five distinct classes of change in marital satisfaction, for example, but 67.6% of the sample were grouped into the two classes characterized by insignificant change in marital satisfaction over the entire duration of the study (Anderson, Van Ryzin, & Doherty, 2010). The same year, a study examining four years of data from initially newlywed couples also found five distinct classes of marital satisfaction trajectories (Lavner & Bradbury, 2010). In this study, over 80% of the sample were grouped into classes that experienced insignificant or minimal changes in satisfaction over time. In both studies, the average trajectory was the familiar high beginning followed by gradual decline, but this average turned out to result from the small minority of couples that start with low levels of satisfaction and decline rapidly from there.

Once the techniques to identify them became available, other datasets were also found to contain large groups of couples who experienced negligible changes in marital satisfaction over time. In the Early Years of Marriage project, over 50% of wives and over 90% of husbands experienced insignificant or minimal declines in their marital happiness across the first 16 years of marriage (Birditt, Hope, Brown, & Orbuch, 2012). In one study of German parents followed over four years, high marital adjustment with slopes that did not differ significantly from zero characterized 90% of couples (Foran, Hahlweg, Kliem, & O’Leary, 2013). Additional newlywed studies found that between 60% and 85% of spouses experienced insignificant or minimal declines in their marital satisfaction across intervals ranging from 2.5 to 4 years (Lavner, Bradbury, & Karney, 2012; Lorber, Erlanger, Heyman, & O’Leary, 2014).

By the time that Proulx, Ermer, and Canter (2017) reviewed this literature, 14 studies had applied these procedures to longitudinal data on marital satisfaction. Given their diverse samples, study durations, and operationalizations of marital satisfaction, it is not surprising that different studies arrived at different ways of classifying marital satisfaction trajectories. Nonetheless, all of the studies undermined the idea that time spent being married leads to declines in marital satisfaction. Across studies, couples experiencing significant declines tended to be couples undergoing stressful life transitions (e.g., new parents) and couples reporting lower satisfaction initially. Yet the dominant pattern across studies was stability, with the most stability experienced by couples reporting the highest initial satisfaction. In other words, the average trend in marital satisfaction has supported a misleading account of how most spouses experience their marriages. Just as no family has ever had 2.5 children, only a minority of couples actually experience high initial marital satisfaction that declines steadily and significantly over the course of their marriage. Instead, most couples who start their marriage happy stay happy for long periods of time.

Priorities for the Next Decade

This is a drastically different account of our dependent variable than the one that has guided longitudinal research to date, and fully embracing it has several important implications for marital research going forward. First, the fact that it has taken this long to recognize the overall stability of marital satisfaction implies that the field has not devoted sufficient attention to describing how marriages change. The welcome progress of the last ten years continues a decades-long trend toward gathering more detailed data from married couples, moving beyond cross-sectional snapshots to exploit the potential of multiwave longitudinal designs and experience sampling methods (e.g., Walsh, Neff, & Gleason, 2017). Psychometrically sound and theoretically coherent instruments for assessing marital satisfaction are in wide use (e.g., Funk & Rogge, 2007). Nevertheless, the field continues to lack an empirical foundation to guide many of the most basic decisions about how to assess and describe trajectories of marital satisfaction. How often and over what length of time should marital satisfaction be measured to capture meaningful change? What interval between assessments represents the optimal balance between seeking higher resolution and avoiding respondent burnout? Without data addressing these questions, marital researchers have been left to design their studies guided by convenience alone. Research that examined these issues directly would provide a stronger foundation for the longitudinal studies of the future.

Second, if stability is indeed the modal experience in marriage, then by the time a couple gets married, many of the factors that ultimately determine the trajectory of the relationship may already be in place. Where couples start, in other words, appears to reveal a lot about where they will end up (Lavner et al., 2012). And yet, because research on marriage begins with couples who are already married, we know very little about relationship satisfaction across the transition into marriage. Collecting data from couples prior to marriage, and even prior to their engagement, holds great promise for illuminating the paths into marriage that result in more or less satisfied newlyweds.

Third, if couples can be reliably distinguished by the trajectory of their marital satisfaction, then it is time to consider the entire trajectory more precisely. The availability of latent class approaches to analyzing trajectories does not preclude the refinement of variable-centered approaches like growth curve modelling; the two approaches are complementary. To date, growth curve analyses of marital trajectories have almost exclusively estimated levels of marital satisfaction (intercepts) and rates of linear change (slopes), often limited by the availability of relatively few longitudinal assessments. Yet longitudinal changes can be described with many more parameters (see Butler, 2011 for a discussion). For example, with sufficient longutudinal assessments, researchers can estimate not only linear change but curvilinear change, i.e., changes that accelerate or decelerate over time. Even among couples with flat linear slopes, fluctuations around the trend line may reveal hidden turbulence. A starting point for refining theories of marriage would be to develop distinct hypotheses about each of these parameters (see Eastwick, Finkel, & Simpson, 2019, for a perspective that moves relationship research in this direction). Time-invariant variables, such as spouses’ personalities and personal histories, may be more strongly associated with levels of satisfaction, whereas time-varying variables, such as couples’ communication and experiences of acute stress, may more strongly account for rates of change and fluctuations in satisfaction over time. In this way, greater precision about our dependent variable highlights directions for developing more precise theory about the effects of our independent variables. To the extent that the results of latent class analyses depend on the variables are entered into the analysis, a more refined assessment of the individual trajectories might lead to different latent classes of trajectories as well.

Fourth, the observation that marital satisfaction remains stable for most couples raises pressing questions about the decision to divorce. For many decades, classical marital research has assumed that deteriorating marital satisfaction is the most proximal reason why marriages dissolve (Gottman & Levenson, 1992; Lewis & Spanier, 1982). Research adopting the latent class approach confirms that couples in which marital satisfaction declines most steeply are indeed much more likely to divorce. Yet these studies also report that couples in stably happy marriages divorce at non-negligble rates (e.g., 14% over ten years, Lavner & Bradbury, 2010), similar to earlier work on national datasets that observed substantial numbers of couples divorcing without preceding signs of marital distress (Amato & Hohmann-Marriott, 2007). One possible explanation for such results is that long intervals between assessments in longitudinal studies fail to measure declines in satisfaction that are in fact occurring shortly before couples make the decision to dissolve. A second possibility is that the explicit measures used in most research on marital satisfaction are not sensitive to changes in spouses’ experience. As McNulty and his colleagues have demonstrated in the last decade (McNulty & Olson, 2015; McNulty, Olson, Meltzer, & Shaffer, 2013), automatic attitudes toward the marriage assessed with the Implicit Associations Test can predict later marital outcomes even when standard self-report measures cannot. Still a third possibility is that some couples may be motivated to end their marriages even when they are relatively satisfied. For example, as social exchange perspectives have long argued (e.g., Levinger, 1976), couples who hold each other in high esteem may still dissolve if they: a) experience few barriers to leaving the relationship, or b) perceive superior alternatives outside of the relationship. Evaluating the evidence for these diverse possibilities will require higher-resolution assessments of the dynamics of this highly significant life transition.

Finally, the relatively high degree of stability that characterizes most marriages has implications for interventions that are devised to prevent relationship distress. Psychoeducational programs for couples have proliferated over the past decade, under the assumption that the average relationship will gradually decline in quality, and with the goal of maintaining the high degree of satisfaction that couples report early in marriage (for a review, see Bradbury & Bodenmann, in press). These initiatives, while well-intentioned, may be misguided to the extent that they are focusing on unselected couples from the general population. Many of these couples will not experience a decline in relationship satisfaction even in the absence of intervention, rendering primary prevention strategies inert and diverting resources away from the subpopulation of couples who will experience relationship distress and perhaps dissolution. Thus secondary prevention—in this case, disseminating interventions specifically to couples who display early indications of risk for distress, whether by virtue of their personal, interpersonal, or environmental vulnerabilities—holds greater potential for promoting healthy marriage (e.g., Barton et al., 2018; also see Heyman et al., 2019).

Marital Interaction: Poor Communication May Not Cause Distress

In 1974, Raush and colleagues argued that self-report methods were inadequate for understanding marital communication and, as a consequence, “theories about interaction between intimates are neither sufficiently firm nor sufficiently specific at present to provide us with other than rough lines of guidance” (Raush, Barry, Hertel, & Swain, 1974, p. 4). Raush et al.’s admonition to observe couples directly has since produced rich insights into couples’ overt patterns of behavioral interdependence (for a critical review, see Heyman, 2001). Five decades later, clarifying the interpersonal behaviors that support healthy relationships remains a crucial task for several reasons. First, prominent models emphasize couple interaction as a fundamental source of the many advantages associated with marriage (e.g., health and well-being for adults and children). Second, research-based efforts to treat and prevent marital distress are enhanced by understanding basic processes of communication. Third, the uptake of social policies intended to benefit families likely pivots on how well they disrupt or enable relationship-promoting patterns of behavior. Yet despite persisting as a vibrant topic of inquiry for family scholars, fundamental questions remain about the nature of ‘interaction between intimates.’ Indeed, empirical findings reported in just this past decade offer a fresh perspective on three enduring assumptions about couple communication.

Empirical Advances

Perhaps the most basic assumption underlying research on marital interaction is that specific behaviors, and specific patterns of interaction, reliably predict relationship distress and dissolution for most couples (e.g., Jacobson & Margolin, 1979). Early success at differentiating satisfied from distressed couples cross-sectionally held out promise that a distinct behavioral profile might characterize at-risk couples, as evidenced by distressed couples’ higher rates of negative behaviors, heightened reactivity to and reciprocation of negative behaviors, and sustained cycles of negativity (e.g., Margolin & Wampold, 1981; see Woodin, 2011 for a review), particularly during problem-solving conversations. In longitudinal studies, however, these same behavioral patterns have not emerged consistently as predictors of declines in relationship satisfaction (for a review, see Bradbury & Lavner, 2012). One possible explanation for the inconsistent findings is that the behavioral differences identified in cross-sectional studies may be a consequence rather than a cause of relationship distress. Another possibility is that the behaviors displayed by clinically distressed couples provide a misleading starting point for understanding how well-functioning relationships are maintained and how they change.

Counter-intuitive longitudinal findings that might have been dismissed as statistical artifacts in prior decades have emerged ever more frequently in the past decade, casting doubt on the assumption that positive exchanges are uniformly good for relationships, and that negative exchanges are uniformly costly. For example, observed positive communication early in marriage does not routinely predict levels of marital adjustment or declines over time, while observed negative communication early in marriage sometimes predicts slower declines in satisfaction (Markman, Rhoades, Stanley, Ragan, & Whitton, 2010). Studies also show that negative communication strategies can be beneficial to relationships over the longer term (despite being perceived by partners as unsuccessful in the short term; Overall, Fletcher, Simpson, & Sibley, 2009), with negative thoughts and behaviors proving to benefit couples with more frequent and more severe problems (McNulty, 2010; McNulty & Russell, 2010). When repeated assessments of couple behavior and relationship satisfaction are conducted, predictive effects in either direction tend to be weak, and higher levels of satisfaction predict improvements in observed communication more consistently over 27 months than communication predicts improved satisfaction (Lavner, Karney, & Bradbury, 2016). While more extreme forms of negative behavior, such as verbal hostility and physical aggression, tend to show expected effects on relationship satisfaction and dissolution (e.g., Hammett, Karney, & Bradbury, 2018; Shortt, Capaldi, Kim, & Laurent, 2013), behaviors coded during more ordinary conversations appear to have a more complex and uncertain association with relationship outcomes over time. In sum, cross-sectional characterizations of distressed versus satisfied couples has proven to be easier to achieve than identification of behavioral antecedents of relationship distress, and longstanding assumptions about the predictive effects of positive and negative exchanges on relationship satisfaction have proven to be difficult to corroborate.

A second guiding assumption, following closely upon the first, is that couples can learn to improve their communication behaviors. At the foundation of virtually all preventive and tertiary interventions for couples, this premise has been subjected to numerous randomized controlled tests. Relying heavily on data from well-educated, middle-class samples, an earlier meta-analysis of interventions aimed at preventing relationship distress and dissolution indicated that group-based educational programs did produce intended changes in observed couple communication (Blanchard, Hawkins, Baldwin, & Fawcett, 2009). At the same time, a large-scale multisite government effort to extend communication-based interventions to couples living with lower incomes, known as the Strengthening Healthy Marriages initiative, yielded negligible improvements overall in observed communication, relative to couples receiving no intervention, despite intensive training (Lundquist et al., 2014). Further analyses of these data showed that interventions improved effectiveness and reduced negativity (but failed to increase positivity) only among couples who were at risk by virtue of their younger age at marriage, lower incomes, and fewer years of formal education (Williamson, Altman, Hsueh, & Bradbury, 2016). Thus improvements in communication are certainly possible, though questions remain about the durability and display of learned skills outside the context of formal observational assessments.

When distressed couples participate in empirically-supported forms of therapy, and thus are presumably motivated to grapple with and repair interpersonal deficits, a different understanding of couple communication emerges: from pre-treatment to two years following completion of treatment, observed rates of negativity and withdrawal decline by nearly half, but rates of unilateral positivity decline as well, to levels below those observed prior to the start of treatment (Baucom, Sevier, Eldridge, Doss, & Christensen, 2011). Whereas conventional behavioral models held that well-functioning marriages were characterized by less negativity and more positivity, some scholars now argue that positivity during problem-solving discussions can sometimes reflect avoidance and a reluctance to engage difficult issues. As Baucom et al. (2011) elaborate, “If one partner is upset with the other it is not necessarily helpful to hide these feelings and exclusively express positivity …. Although researchers often make assumptions about the functions of these behaviors, these assumptions are based solely on the topography of the behavior rather than the functional role they actually play in the relationship” (p. 575). In short, with sufficient resources and motivation it does appear that couples can improve their communication, yet the nature of those improvements may well depend upon the quality of the relationship, the specific problem at hand, and both partners’ ability to balance their individual needs with the needs of the larger partnership (see also Baucom, Baucom, & Christensen, 2015). More critically, by essentially requiring couples to discuss relationship problems, standard observational paradigms likely fail to sample how couples naturally engage or avoid potentially divisive issues, and therefore may not provide all of the information needed to understand how communcation changes.

A third major assumption in observational studies of couple interaction is that improved communication, once achieved, will enhance the quality and stability of the relationship. Once again, experimental tests of interventions, and the distinction between preventive and therapeutic interventions, prove to be informative. In the aforementioned Strengthening Healthy Marriages project, the small effects that the communication-based interventions produced on relationship satisfaction 18 months later were not mediated by observed communication (Williamson et al., 2016). This indicates that learning to communicate more effectively and with less negativity does not necessarily translate into better functioning relationships and, conversely, that relationships can improve without overt improvements in communication (cf. Barton et al., 2017; Cordova et al., 2014). Indeed, a brief intervention focused only on enhancing relationship awareness, with no emphasis on skills training, performed as well as 15-hour intensive skill-based workshops, possibly because heavy emphasis on behavioral skills can inadvertently sensitize couples to shortcomings in their communication (Rogge, Cobb, Lawrence, Johnson, & Bradbury, 2013).

Other work reveals interesting disjunctions and continuities between behavior change and changes in satisfaction. According to the largest and longest couple therapy outcome study undertaken to date, changes in observed positivity and negativity from before to after couples therapy are entirely unrelated to spouses’ reports of relationship quality and stability two years later (Baucom et al., 2011). Thus, while empirically-supported couples therapy can change the communication behaviors long emphasized in interpersonal models of marriage, at least some of these changes may have little bearing on subsequent perceptions of the relationship. Observed pre-to-post treatment increases in problem-solving, and decreases in withdrawal, however, do predict better relationships over this span (Baucom et al., 2011). Thus, at least for couples in an acute state of distress, acknowledging and addressing specific deficits or shortcomings in a relationship can enhance evaluations of the relationship, and apparently more so than reductions in negative behavior.

Priorities for the Next Decade

This past decade has witnessed notable progress toward clarifying how communication between spouses is associated with their evaluations of their relationship. Research on couple interaction appears to be in a transitional stage—yielding too few clear empirical results to justify complete adherence to what has been an unusually generative set of models and paradigms, while still casting about for concepts and tools that promise novel insights into how couples conduct their intimate lives. Marked heterogeneity in how researchers define and code specific behaviors no doubt underlies some of this confusion, and important steps are being taken to rectify this problem (e.g., Overall & McNulty, 2017). However, the critical uncertainties that remain likely reflect theoretical imprecision regarding exactly when and how specific elements of couple communication come to affect judgements of relationship quality and longevity, a problem compounded by reliance on observational paradigms that provide a rich, compelling, but ultimately constrained appreciation for how intimate partners co-create a dynamically complex behavioral repertoire. While we cannot fully anticipate how this transitional stage will be resolved, several of the findings reviewed here point to gaps in knowledge that, if addressed, might advance understanding in the decade ahead.

First, although we have no strong empirical rationale for doubting that behaviors can have rewarding and punishing properties, and that these properties contribute to variability in judgments of marital satisfaction, evidence is accumulating that what makes a specific behavior rewarding or punishing differs across spouses and couples. Behaviors do not ‘speak for themselves,’ in that the effects of specific behaviors on satisfaction likely vary as a function of immediate but unobservable dyadic and relational microcontexts (e.g., power imbalances in the relationship, the severity and controllability of the problem under discussion, the capacity and willingess of partners to change, partners’ commitment to the relationship; see Overall, 2017; Overall & McNulty, 2017). Careful parsing of potential moderators would shed needed light on when specific classes of behaviors do and do not predict changes in satisfaction.

Second, the temporal dimension of couple interaction, including behavioral sequences and patterns that unfold over time, is growing less visible in the field, depriving us of access to critical features of how partners communicate with each other. As a result we are left with decontextualized behaviors displayed by, and inadvertently attributed to, just one partner, thus grossly oversimplying the emergent and dyadic character of couple interaction (S. Duncan, Kanki, Mokros, & Fiske, 1984). Sequential analysis has proven to be difficult to implement—in part because repeated instances of behavioral sequences are necessary for reliable estimates of those sequences—but this problem can be overcome, e.g., by collecting continous ratings of behavior (e.g., Ross et al., 2017), by studying how behaviors unfold over the entire span of conversations (e.g., Kuster et al., 2015; Leuchtmann et al., 2019), and by undertaking intensive analysis of specific behavioral sequences hypothesized to be critical for problem-solving or relationship maintenance. For example, in a methodologically elegant study, Bloch, Haase and Levenson (2014) focused intensely on how spouses mutually downregulate negative emotions and experiences during a problem-solving conversation, with results suggesting that ineffective attempts to contain negative emotions, moreso than the mere presence of negative affect, are most likely to compromise wives’ relationship satisfaction. Thus it may be premature to abandon the traditional observational paradigm entirely, provided that specific and temporal patterns are the focus of inquiry.

Third, although studies of couple communication now routinely exploit longitudinal data, they are only beginning to adopt a developmental focus, and thus they do not yet address how behavioral exchanges affect other domains of couples’ relationships and, in turn, various relationship outcomes. Observations of newlywed couples do indicate that behaviors displayed in a social support task predict displays of negative emotion during a problem-solving task one year later, whereas the opposite effect is far weaker, and that negative emotions during problem-solving mediate the effects of social support behaviors on satisfaction and dissolution ten years later (Sullivan, Pasch, Johnson, & Bradbury, 2010). Again, the behavioral paradigm appears to be capturing important facets of couple communication, but that information may be of limited value unless we connnect those behaviors with other behavioral domains and other emotionally engaging experiences in couples’ lives. Finer-grained analyses of the sort made possible by intensive and extensive daily studies, when they are embedded in longitudinal designs, are likely necessary for understanding dyadic interdependence in a more complete way (for a review, see Repetti, Reynolds, & Sears, 2015). Such an approach would permit greater access to important but fleeting moments in couples’ daily lives, such as how verbal intimacy following sexual intimacy enhances relationships (e.g., Muise, Giang, & Impett, 2014), or how couples enter and exit psychologically meaningful moments, including the specific sorts of exchanges that make it more or less likely for partners to devote time to the relationship or to avoid one another following conflict.

We conclude here by raising the possibility that long-standing efforts to identify and alter specific patterns of behavior, however they might be assessed, are framing the problem incorrectly. In addition to considering the unconditional main effects of couple interaction on couple outcomes (which, as we have seen, may be somewhat elusive), it may be fruitful to conceptualize dyadic processes primarily as the means by which events arising outside a relationship come to affect relationship appraisals. Thus, a couple with ostensibly poorer communication could have a better relationship than an otherwise identical couple with better communication, if fewer outside demands were placed on the quality of their communciation or if they were especially good at rising to these demands (e.g., providing symmpathy to an upset partner, or teaming up in the face of an unexpected event such as a sick child or job loss). Stress, and partners’ perceptions of one another’s stress, may influence whether specific behaviors are registered as aversive, suggesting that continued efforts to join behavioral analyses with an appreciation of the stressors that couples face (e.g., Bodenmann et al., 2015; Kuhn, Bradbury, Nussbeck, & Bodenmann, 2018; Neff & Broady, 2011) may go far in providing the theoretical specificity that Raush and colleages were seeking five decades ago.

Do Conclusions about Marriage Generalize? Income as a Moderator

Twenty-five years ago, the vast majority of research on marriage had been conducted on samples composed primarily of middle-class, college-educated, mostly white couples (Karney & Bradbury, 1995). In the intervening years, the profile of the typical sample in marital research has expanded, but samples of convenience continue to dominate, reflecting a widespread assumption about the dynamics of marriage: that what is true for the narrow slice of couples that have been studied extensively is likely to be true for the more diverse couples that lie outside the sampling frame of most studies.

Over the past decade, this assumption has been questioned by burgeoning lines of research on Hispanic marriages (e.g., Orengo-Aguayo, 2015), African American marriages (e.g., Cutrona, Russell, Burzette, Wesner, & Bryant, 2011; Stanik, McHale, & Crouter, 2013), and most recently same-sex marriages (e.g., Chen & van Ours, 2018), to mention only a few dimensions of diversity that the field has explored. To illustrate the consequences of broader sampling for deepening our understanding of marriage, we devote this section to reviewing one dimension of diversity that was examined at length over the past ten years: income.

The possibility that marital process may vary at different levels of income is worth highlighting because, a decade ago, this possibility was distinctly overlooked. When the federal government launched the Healthy Marriage Initiative (HMI) in the early 2000s, the goal of the program was to promote stronger, more stable relationships among lower-income couples (Office of Family Assistance, 2012). Yet at the time data on the relationships of lower-income couples were scarce. Nevertheless, the development and implementation of skills-based relationship education programs proceeded apace, guided by the assumption that the conclusions of prior research on relationship processes among relatively affluent, mostly white couples would generalize to the lower-income, more diverse communities targeted by the new initiative. Within the past decade, the results of nationwide, multisite, longitudinal evaluations of these programs were released, and they were not encouraging. Despite their great cost and the sincere efforts of well-intentioned administrators and educators across the country, the impact of relationship education on the well-being and stability of lower-income couples proved to be negligible (Lundquist et al., 2014; Wood, Moore, Clarkwest, & Killewald, 2014). Abandoning the assumption that the foundations of satisfying and stable marriage are common across income groups, leading observers called for research directly examining how basic processes contributing to successful intimacy may in fact vary among different segments of society (Johnson, 2012; McNulty, 2016).

Empirical Advances

Research heeding this call over the last decade has revealed numerous ways that basic marital processes vary across levels of socioeconomic status. What does not vary is the value of marriage itself. Although rates of marriage are substantially lower in lower-income communities than in more affluent ones, marriage remains a cherished goal across income groups (Garrett-Peters & Burton, 2015; Scott, Schelar, Manlove, & Cui, 2009; Trail & Karney, 2012). Instead, income groups can be distinguished by the sorts of challenges couples face in attempting to reach that goal. Lower-income couples, like lower-income individuals, report greater mental health problems, higher stress, and, not surprisingly, more financial strain than their higher-income peers (Maisel & Karney, 2012). When asked directly about challenges in their marriages, lower-income couples are more likely to identify finances and substance abuse, rather than communication and household chores, as their most salient sources of conflict (Jackson et al., 2016; Trail & Karney, 2012). To the extent that the problems of lower-income couples are more severe, their marital interactions are likely to be more difficult as well (Williamson, Hanna, Lavner, Bradbury, & Karney, 2013). Indeed, Cutrona and her colleagues, drawing upon observer ratings of videotaped interactions between spouses, have shown that couples living in more disadvantaged neighborhoods express less warmth with each other than couples living in higher income neighborhoods (Cutrona et al., 2003).

Do their greater challenges mean that lower-income couples are less satisfied with their relationships than more affluent couples? In fact, lower-income couples are not less satisfied with their relationships on average, despite their higher divorce rates and the more severe difficulties they encounter. Although perceptions of greater financial strain have been consistently associated with lower relationship satisfaction (e.g., Conger & Conger, 2008), household income and relationship satisfaction are unrelated (Hardie & Lucas, 2010; Maisel & Karney, 2012). Yet poorer couples do experience their marriages differently than wealthier ones. Longitudinal research tracking marital satisfaction across the first years of marriage reveals that, although levels and slopes of marital satisfaction do not differ between more and less wealthy couples, poorer couples report significantly greater variability in their marital satisfaction across time (Jackson, Krull, Bradbury, & Karney, 2017). In other words, lower-income couples experience their relationships as more turbulent than affluent couples, whose greater resources presumably buffer them from stressful events and crises (cf. Henry, Sheffield Morris, & Harrist, 2015).

To the extent that income has direct associations with constructs central to influential models of successful marriage, like mental health, stress, marital interaction, and trajectories of marital satisfaction, then the associations among these constructs are likely to vary across levels of income as well. Over the past decade, analyses of diverse samples that directly compare patterns of associations among more or less affluent couples confirm this prediction. For example, analyses of survey data from Florida, Texas, California, and New York reveal that stress and mental health account for significantly more variance in the relationship satisfaction of poorer couples than wealthier ones (Maisel & Karney, 2012). When couples with more access to resources confront stress or struggle with mental health problems, they have options for protecting their relationships (e.g., by paying for assistance or counseling) that poorer couples facing the same challenges do not.

Ross et al. (2019) went further, showing that even basic marital processes considered fundamental to successful marriage can have categorically different implications for more or less affluent couples. Their focus was the demand/withdraw pattern of marital interaction, in which one partner, seeking change, makes a request, while the other, favoring the status quo, deflects or avoids. The ensuing cycle has been consistently associated with declines in marital satisfaction in numerous studies (e.g., Eldridge, Sevier, Jones, Atkins, & Christensen, 2007) and across multiple cultures (Christensen, Eldridge, Catta-Preta, Lim, & Santagata, 2006). Yet the negative implications of the demand/withdraw pattern rest on the assumption that, when this pattern arises, the change at issue is possible. What makes the pattern so aversive for both sides is the idea that the partner requesting change can reasonably expect the other partner to give in, and that the partner resisting change could accede to the other partner’s request but chooses not to. But what if the requested change is impossible, as it may be for lower-income couples with less control over their circumstances? A response (withdrawal) that is maladaptive for the middle-class couples that comprise most of the samples in research on marital satisfaction may be be adaptive for lower-income couples with fewer options and resources. Indeed, across two longitudinal studies that included observational assessments of marital interactions, Ross et al. (2019) found that, whereas engaging in the demand/withdraw pattern predicts lower marital satisfaction for wealthier couples, the same pattern predicts more stable marital satisfaction for poorer couples.

Such findings begin to make sense of the failure of programs guided by research on upper-income couples to improve the outcomes of the lower-income couples to which they were targeted: the specific challenges and constraints that confront lower-income couples may moderate or negate the impact of interventions that might be helpful in other circumstances. Analyses of data from the Building Strong Families study, an evaluation of interventions designed to encourage marriage among unmarried lower-income parents, support this perspective. Although the evaluation showed that the programs had no effects on relationship quality or rates of transitioning into marriage (Wood, McConnell, Moore, Clarkwest, & Hsueh, 2012; Wood et al., 2014), Williamson et al. (2017) reasoned that programs that included job training and professional education, in addition to relationship education, might have benefitted couples, as these programs targeted the specific problems that lower-income couples name as obstacles to marriage (e.g., Edin & Reed, 2005). In fact, programs that included these elements did have unique effects, but in the opposite direction: men who received professional education were less likely to transition into marriage. Further analyses revealed that men receiving additional hours of education from these programs spent less time at home with their children and contributed less money to their households, accounting for their lower marriage rates. Such unexpected findings highlight the need to understand couples’ experiences – the demands they are facing, the constraints on their time and resources, and the particular dynamics that are adaptive or maladaptive in their specific context– before attempting to develop policy or implement interventions to improve their lives.

Priorities for the Next Decade

Although the results of large-scale evaluations of programs aimed at lower-income couples were disappointing, they succeeded in raising crucial questions about how marriages succeed or fail in diverse contexts. By taking up these questions, research on marriage over the past ten years has shed light on ways that macro-level socioeconomic conditions can facilitate, constrain, and even alter marital process previously thought to be fundamentally similar across couples. As federal and local policy-makers continue their efforts to support marriages and families, it is crucial that research on marriage capitalize on the progress of the last decade, partly to continue the process of refining theory and partly to ensure that the next generation of programs has more impact than the previous one. In particular, the studies reviewed here suggest three important recommendations for further developing research on marriage that is sensitive to context.

First, in light of a decade’s worth of research demonstrating how different marriages can be at varying levels of socioeconomic status, a preliminary task for the next decade is for scholars of marriage to be more precise and intentional about sampling. Large-scale studies, usually informed by sociology and demography, generally do an excellent job developing explicit sampling frames and ensuring adequate representation from all levels of income. Smaller-scale studies, usually informed by clinical and social psychology, fare less well, often relying on samples of convenience that overlook harder to reach lower-income and diverse couples. By describing samples in greater detail, and by devoting more effort to sampling couples that vary not only in socioeconomic status but in other relevant dimensions of diversity as well (e.g., race/ethnicity, religion, gender identity), future research on couples will be better able to examine and articulate limits to the generalizability of our findings.

Second, to the extent that socioeconomic contexts affect how couples pursue and maintain intimacy, research on the effects of context and research on dyadic processes within couples can no longer proceed on parallel tracks. On the contrary, research on micro-level cognitions and behaviors must account for the macro-level contexts in which those cognitions and behaviors take place. Prominent scholars are already moving in this direction, combining detailed assessments of partners’ experiences in the relationship with census data on neighborhood conditions (Bryant et al., 2010) and national and local economic indicators (Murray, Lamarche, & Seery, 2018). Detailed self-report interviews that assess couples’ perceptions of the concrete features of their own environments (e.g., Hammen, 2005) may be required to identify the proximal contexts that mediate between distal conditions and dyadic processes within couples’ relationships.

Third, in advance of the more refined models to come, the research of recent years already points out novel avenues for intervention with vulnerable couples. In particular, as evidence accumulates that disadvantaged environments can inhibit couples’ ability to interact effectively, policy-makers and scholars may consider exploring interventions that target those environments directly. Whereas relationship-education and skills-based training has proven ineffective at promoting stable unions among lower-income couples, policies that improve the circumstances of those couples, e.g., through poverty reduction, may be more successful at achieving the same ends (Gassman-Pines & Yoshikawa, 2006; Mamun, 2008). The promise of local efforts to improve the lives of struggling couples suggests that federal policies also be reviewed for their direct and influence on marriage and families, and some have already called for such analyses to be conducted routinely (Bogenschneider, 2013). The research of the past decade suggests that any program that improves the conditions of couples is likely to improve their relationships; such programs may affect a broad range of couples more efficiently than programs that target individual couples directly.

Conclusion

The last ten years of research on marital satisfaction and stability revealed that long-standing assumptions about marriage are either incomplete, misleading, or wrong. For most couples, satisfaction does not decline over time but in fact remains relatively stable for long periods. Contrary to the predictions of behavioral models of marriage, negative communication between spouses can be difficult to change, does not necessarily lead to more satisfying relationships when it is changed, and does not always predict distress in the first place. Dyadic processes that are reliably adaptive for middle-class and more affluent couples may operate differently in lower-income couples, raising serious questions about how well current models of marriage generalize to couples living in diverse environments. By the end of the decade, then, the assembled research of the last ten years has revealed that the tasks of understanding and promoting marital satisfaction and stability are more complicated than we appreciated at the start of the decade, raising important questions that beg to be answered in the years ahead.

The tools that led to these theoretical refinements will be the tools that continue to move this field forward. Novel approaches to data analysis hold the power to reveal new insights lying buried in current and archival data sets. Well-funded, well-powered, large-scale studies (like the Strengthening Healthy Marriage evaluation) have provided invaluable information, even when the predictions guiding those studies could not be confirmed. Progress in understanding complex, multidetermined processes of marital development require support for additional studies of similar scale and breadth. Finally, as the social psychologist Kurt Lewin (1951) observed, our basic theories of how couples succeed or fail are likely to be refined by continued efforts to develop new and more effective interventions, as doing so requires program developers to be explicit about moderators and mechanisms of their desired effects. Advances on these fronts will deepen understanding of marriage and thereby enable many more couples, regardless of their social standing, to reap the many benefits provided by our closest social bonds.