AI analysed over 11,000 couples' relationships. This is what it found

A first-of-its-kind artificial intelligence (AI) study of romantic relationships based on data from thousands of couples has identified the top predictors that make partners feel positively about their relationship – and the findings show romantic happiness is about a lot more than simply who you're with.

Researchers conducted a machine-learning analysis of data collected from over 11,000 couples, and found that relationship-specific characteristics (personal evaluations of the relationship itself) were significantly more powerful predictors of relationship quality overall than variables based on individual characteristics.

In other words, the type of relationship you build with a partner may be more important to your happiness than either of your individual characteristics - in the study, they looked at traits like how satisfied a person was with life, how anxious they were, or whether their parents' marriage worked out.

"Relationships-specific variables were about two to three times as predictive as individual differences, which I think would fit many people's intuitions," says lead researcher and psychologist Samantha Joel from Western University in Canada.

"But the surprising part is that once you have all the relationship-specific data in hand, the individual differences fade into the background."

Relationship science has existed for decades and prompted huge amounts of psychological theory about what makes (and doesn't make) for happy couples.

Yet the researchers say a key challenge for their still-maturing field is bringing cumulative data together on a larger scale, to bolster the findings made in smaller, standalone studies (which can be expensive and time-consuming to conduct, in terms of recruiting and interviewing participants).

One analytical solution could be AI, which has the ability to sift through huge amounts of data collected from individual laboratories. In a new study, Joel's team employed that very approach, using a machine learning system called Random Forests, which can test the predictive power of a large number of variables fed to it.

The technique, having analysed mostly self-reported measures collected from 11,196 couples across 43 separate datasets, determined what kind of reported variables seemed to matter the most in terms of predicting relationship quality.

"Results revealed that variables capturing one's own perceptions of the relationship (eg. conflict, affection) predicted up to 45 percent of the variance in relationship quality at the beginning of each study," the authors write in their paper, noting that the predictive effect did diminish over the course of the studies.

In terms of these relationship-specific variables that most reliably predicted relationship quality, the most reliable were: perceived partner commitment (eg. "My partner wants our relationship to last forever"); appreciation (eg. "I feel very lucky to have my partner in my life"); sexual satisfaction (eg. "How satisfied are you with the quality of your sex life?"); perceived partner satisfaction (eg. "Our relationship makes my partner very happy"); and conflict (eg. "How often do you have fights with your partner?").

By contrast, predictors related to individual characteristics - observations respondents reported about themselves, ranging from their personality traits through to their age and gender - at most explained only 21 percent of variance in relationship quality.

The individual characteristics that most strongly predicted the quality of a relationship were: 'satisfaction with life'; 'negative affect' (eg. feeling distressed or irritable); 'depression'; 'avoidant attachment' (eg. "I prefer not to be too close to romantic partners"); and 'anxious attachment' (eg. I worry a lot about my relationships with others).

"Experiencing negative affect, depression, or insecure attachment are surely relationship risk factors," the researchers write in their paper.

"But if people nevertheless manage to establish a relationship characterised by appreciation, sexual satisfaction, and a lack of conflict - and they perceive their partner to be committed and responsive - those individual risk factors may matter little."

While there's a huge range of potential statistical insights on offer here – and numerous interpretive limitations too, it must be said, imposed by these kinds of analytical methods – at its heart, the findings boil down to a pretty simple truth, Joel says.

"Really, it suggests that the person we choose is not nearly as important as the relationship we build," Joel explained to Inverse.

"The dynamic that you build with someone — the shared norms, the in-jokes, the shared experiences — is so much more than the separate individuals who make up that relationship."

The findings are reported in PNAS.

#Humans | https://sciencespies.com/humans/ai-analysed-over-11000-couples-relationships-this-is-what-it-found/