What Do You Think Makes a Relationship Work?

A new study tries to predict what makes and keeps couples satisfied.

Posted Sep 12, 2020

When it comes to relationships, the 64-million dollar question most people have is what will make theirs work. You might have your favorite theories about what keeps a relationship strong. Perhaps you find your own satisfaction to be greatest when you and your partner share a laugh, a night of intimacy, or even a weekend afternoon of completing mundane chores around the house. You might feel, however, that it’s something about the match between you and your partner’s personalities—either your similarities or, more surprisingly, your differences. When you dig deeper, you might decide that your satisfaction depends on how intimate you feel with your partner, how committed you are, how well you communicate, and whether you can come to amicable resolutions of the inevitable differences that pop up in daily life.

Relationship researchers, who try to put measuring sticks onto the quality of people’s feelings about their close partners, have their own theories about what contributes to a couple’s satisfaction and to the longevity of their partnership. In the latest attempt to tease apart the many potential contributors to relationship satisfaction, the University of Utah’s Samantha Joel (2020) led a mega-size group of researchers who analyzed 43 datasets, each consisting of couples studied over two or more time points. A key requirement of each study was that it included both members of the couple, allowing for so-called “dyadic” comparisons to be made.

Published as part of the Open Science Initiative, the Joel et al. study allows other researchers to peer directly into the datasets themselves (with no identification of individual participants). The information that can be gained from such an approach includes not only what the datasets looked like, but also the study’s hypotheses, analytic methods, and even the specific computer programs used. On the plus side, such an approach minimizes the possibility of researchers manipulating the data to get the best results — what’s known as “p-hacking." On the other hand, the data itself, though massive in scope, has its own limitations.

As background to the study, Joel and her collaborators note that it’s important to understand what factors lead to good relationships. Previous research, she writes, shows that “Unhappy marriages are associated with many negative stress-related outcomes, including poor physical health, high blood pressure, poor immune system functioning, mortality, and risk of mental health problems." In other words, a good relationship is good for your health. There is also a negative spillover effect of low marital quality into your other relationships, including those with children, and even your productivity at work. These findings probably resonate with your own experience.

Each study in the dataset had as its key outcome factor, or variable, the individual’s perception of the relationship’s quality, or relationship satisfaction, although a number of studies also included a measure of commitment. Because dyadic data were available, the authors could compare the predictive value of qualities of Partner A with qualities of Partner B, consistent with, the authors note, “the raison d’être of intensive dyadic data collection efforts." Across the 43 studies included in the dataset involving nearly 12,000 couples, the median age of participants was 27, their relationships had lasted between 3 months to 12 years (with the majority toward the lower end of the range), and the couples were tracked for from 2 months to 4 years, with most at the shorter end of the scale—63% were together one year or less. Nearly half of the samples included partners who were dating, and less than a quarter were married; the rest represented a combination of the two. The study’s length did not span the entire relationship of the majority of participants. The conclusions about long-term relationships drawn from this study, then, would be limited to younger adults, those who are not married, and those whose relationships have lasted approximately one to two years.

Taking advantage of the methods now available through machine learning (Artificial Intelligence), the authors used a tool with the picturesque name of “Random Forest,” which tests, one at a time, the strength of each available predictor in a sequential set of decision trees (hence the name). At the end of the process, the researchers are able to determine which predictors are the strongest contributors to the overall model. There were a maximum of 42 Random Forest models per dataset. Essentially, the program “learns” which are the strongest predictors in any given dataset in a sequential fashion, eventually honing in on those that work the best. You can see that this study used a very different approach than you might be used to seeing if you’ve read psychology research involving more traditional methods.

The one dataset with the largest number of predictors had 146 participants followed over time, involving 97 predictors based on individual measures and 50 based on relationship measures. This dataset was based on college students in dating relationships followed over one-to-two years, and whose average relationship at the start of the study was 17 months.

You might wonder what happened to people whose relationship ended before the study was over. By definition, they would not be in the study anymore. The findings, then, wouldn’t help you understand why relationships end, only why they persist, but more factors could go into the analysis than any one correlational study could ever hope to investigate.

Looking now at the main findings, if you were thinking that those "raison d’être" dyadic predictors would show up as important to the final relationship satisfaction measures, you would be surprised to learn that they did not. Indeed, at the first measurement, the only significant predictors of relationship satisfaction were those associated solely with the perceptions of the respondent (i.e. the “actor”). No partner-reported variables added to any of the predictive models at the initial point of the study. At follow-up, virtually nothing predicted relationship satisfaction at all. As the authors concluded, “Self-report variables may be ill-equipped to reliably predict future changes in satisfaction, at least as operationalized here." When it came to commitment, similar findings emerged, but these effects were even smaller both at baseline and follow-up.

Again, somewhat surprisingly, adding individual differences in personality to the model had no impact on the prediction of relationship satisfaction change over time. Breaking down the relationship variables that seemed to have the most predictive power (at either baseline, follow-up, or change in between), the top three predictors of satisfaction were the perception that the partner was committed to the relationship, degree of intimacy (self-reported), and sense of being appreciated by the partner. This analysis suggests that you’ll be more satisfied with your relationship if you believe your partner is committed to it, at least when the relationship is in its early stages.

Taking the overall “success rates” into account with respect to individual differences, the authors reported some support for attachment-related variables including anxiety (“I worry a lot about my relationships with others”) and avoidance (“I prefer not to be too close to romantic partners”). Participants higher in life satisfaction and those lower in depression and negative affect also reported higher relationship satisfaction. Again, though, keep in mind that these effects were small in the grand scheme of things, given how poorly relationship satisfaction and commitment could be predicted at all.

Is it hopeless, then, to predict how your relationship will evolve over the future, especially in its early stages? Although this would seem to be the case from the present findings, the authors suggest another possible set of influences on how couples evolve over time: Perhaps those individual differences that actors report about themselves (e.g. attachment) combine with the partner’s individual differences and the partner’s report about the relationship to influence the actor’s own report about the relationship to influence overall relationship quality.

In simpler terms, according to this so-called “mediational model,” your perception of your relationship would reflect your and your partner’s personalities, along with the satisfaction your partner reports having. In turn, your own relationship satisfaction would statistically become the only factor left that could predict the quality of your relationship.

All of this means that it is your own perception that seems to matter most when it comes to your feelings of satisfaction, at least in a relationship’s early stages. As the authors conclude, “the most proximal predictors are features that characterize a person’s perception of the relationship itself."

To sum up, although this study can’t tell you what will allow your relationship to last over the long haul, it provides insight into the importance of, as the saying goes, “the eyes of the beholder.” Thinking positively about your relationship might very well be the key to getting your relationship off to a start that will keep it fulfilling over time.

References

Joel S, Eastwick PW, Allison CJ, et al. Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proc Natl Acad Sci U S A. 2020;117(32):19061-19071. doi:10.1073/pnas.1917036117