A Machine Learning Algorithm Uncovers The Top Five Predictors Of A Healthy Relationship

Relationship quality, or the perception that a romantic relationship is either good or bad, is a powerful determinant of life outcomes.
​
People in positive relationships tend to be happier, healthier, and more productive. They also tend to raise more well-adjusted children.

But what predicts relationship quality? A cross-disciplinary team of researchers think they have an answer. Analyzing data from over 11,000 romantic couples recruited by 29 different research laboratories, the team found that people who viewed their partner as highly committed to the relationship were most likely to report being in a thriving relationship. Interestingly, commitment mattered more to the quality of a relationship than other factors such as trust, support, passion, sexual frequency, and affection.

“The most reliable (top five) relationship variables were perceived partner commitment (‘My partner wants our relationship to last forever’), appreciation (‘I feel very lucky to have my partner in my life’), sexual satisfaction (‘How satisfied are you with the quality of your sex life?’), perceived partner satisfaction (‘Our relationship makes my partner very happy’), and conflict (‘How often do you have fights with your partner?’),” report the researchers.

These findings are largely consistent with previous research, which has also identified commitment, appreciation, sexual satisfaction, perceived partner satisfaction, and conflict as important determinants of relationship quality.

The researchers used a machine learning approach to rank the factors most important to a relationship’s success or failure. They found that people’s subjective impressions of their relationship did a better job predicting relationship quality than individual difference measures such as age, gender, personality, or religion. To be exact, subjective impressions explained approximately 45% of people’s current relationship satisfaction while individual difference measures explained only about 21%. Among the most important individual difference measures were life satisfaction (‘The conditions of my life are excellent’), negative emotion (‘feeling distressed,’ ‘feeling irritable’), depression (‘feeling hopeless’), attachment anxiety (‘I worry about a lot about my relationships with others’), and attachment avoidance (‘I prefer not to be too close to romantic partners’).

​As impressive as all of this might seem, the authors were equally struck by the model’s limitations. With all that data (over a million data points in total), they were unable to predict current relationship satisfaction with a high degree of accuracy. They were also unable to reliably predict changes in relationship satisfaction over time.

“Relationship-quality change (i.e., increases or decreases in relationship quality over the course of a study) was largely unpredictable from any combination of self-report variables,” state the researchers. “These results are consistent with another recent large collaboration showing that life trajectories are generally difficult to predict, even with complex machine-learning methods.”

Perhaps the most striking conclusion, however, was that a partner’s evaluation of the relationship did nothing to increase the model’s ability to predict relationship satisfaction above and beyond the information already provided by one partner. The authors conclude, “Our results suggest that if Amir and Alex each complete many questionnaires about themselves and their relationship, all of the predictable variances in their relationship quality will be explained solely by their own perceptions of that relationship.”

Article by Mark Travers
Mark Travers, Ph.D., is an American psychologist with degrees from Cornell University and the University of Colorado Boulder.