What we know about how long relationships last
A study covering more than ten thousand couples compared the predictive value of everything research thought it knew. Its conclusions are disappointing, and instructive.
The problem posed
The question looks simple: what makes a relationship satisfying for those living it, and makes it last. Decades of work have produced dozens of partial answers, each established on a different sample, with different measures.
Comparing these answers with one another requires having many studies in a homogeneous form, which did not exist. That is precisely the exercise carried out by Samantha Joel and a very large number of colleagues, in work published in 2020.
The method
The team gathered data from several dozen longitudinal studies, covering more than ten thousand couples, and applied to the whole set statistical learning methods able to measure the share of variation explained by each category of variable.
The interest of this approach is that it does not test one hypothesis but compares them all, on the same data. It thus avoids the bias that consists in favouring the variable one works on.
The result
Individual characteristics known before the relationship (personality, age, income, stated preferences) explain a very small share of subsequent satisfaction. What we know about a person taken in isolation says almost nothing about what their relationship will be.
What explains more belongs to the relationship itself: the satisfaction reported by the partner, the sense of commitment, the perception that the other is available. In other words, the best predictors are properties of the bond, not of the people.
Even these variables leave a considerable share unexplained. The predictive capacity obtained, all factors combined, remains modest.
What this invalidates
This result weakens the idea of matching based on individual profiles. If what we know about people before the encounter explains little, then a selection operated on that information cannot produce more.
It also weakens, in the opposite direction, retrospective diagnoses explaining a separation by the character of one of the two. The factor invoked may be real and explain only a very marginal share of what happened.
What this leaves open
A significant part of what decides a relationship is not captured by existing instruments: material circumstances, life events, what is built in daily exchanges and that questionnaires do not reach.
The honest conclusion is therefore modest. Research describes mechanisms correctly, it predicts trajectories badly, and nothing today indicates it is close to managing that.
Key points
What we know about a person before the relationship explains very little of the satisfaction they will draw from it.
The best predictors are properties of the bond itself, and their predictive power remains modest.
This result weakens both the promises of algorithmic matching and retrospective explanations by the character of one of the two.
This article describes documented mechanisms and disagreements. It recommends no position, no party and no vote, and commits to no reading other than that of the work it cites.
And you, where do you stand?
The Civix test situates your convictions on six dimensions and matches you with one of the twelve profiles, without ever telling you what to think.
Start my test