Skip to main content

Verified by Psychology Today


What Really Brings Couples Together

... and why the resource-allocation model leads to success.

Key points

  • A recent study tested four computer models to predict who pairs with whom in real life based on mate preferences and characteristics.
  • One model nicknamed "invest more if they like me back" was the winner, recreating 46% of real-life couples compared to 2% expected by chance.
  • The study suggests that computer models will be used as matchmakers and matchbreakers (predictors of who shouldn't be together) in the future.

For centuries professional matchmakers all over the world have used their intuition to predict compatibility between two people looking for love and marriage. But with the onset of technology and computer modelling, it feels almost inevitable that gut feeling will be replaced by data-driven predictions (although ChatGPT has a long way to go!).

A first step might be to see if computer models can actually predict who pairs with who in real life, and a recent paper in Personality and Social Psychology Review did just that. In the paper, 191 real couples were recruited and both partners had their mate qualities and preference recorded to create 382 “profiles." Next, the researcher (Dr. Daniel Conroy-Beam, UC Santa Barbara) programmed four different computer models to pair up the profiles into couples based on their mate preferences and characteristics.

These models were based on theories from the scientific literature about how couples form and a central idea of the paper was that the model that was able to “recreate” the most original couples best reflected how couples are created in real life. Before getting to the results, let’s look at the models.

Model 1: The “You’ve worn me down!” model (The Kalick-Hamilton Model)

In this model, people go on dates and make offers of commitment based on their date’s attractiveness. If both offer commitment, a couple is formed. The “offer commitment” threshold changes depending on how many bad dates someone has been on. As the number of dates increased, the attractiveness threshold lowers.

Model 2: The “Men do the rounds, women stick and twist” model (The Gale-Shapley Algorithm)

Here, single men offer commitment to the women they most prefer. Women choose among these offers and pick the best one. If she already has a partner, he’s included in the judgement and jettisoned if a better offer comes along. Disappointed men move on to their next best option.

I think this model best reflects the “red pill” perspective of how couples form, with women having a key role as choosers while men have little impact.

Model 3: The “Invest more if they like me back” (The Resource Allocation Model)

This model says that people spread their efforts around, trying to attract many different mates, and giving better options more attention. Then, people adjust their effort based on how much their actions are reciprocated. Eventually, a single pair forms who invest all their efforts in each other.

Model 4: The “Keep your options open (for a while)” model (The Aspiration Threshold Model)

The final model suggests that people pair with the first willing partner who meets an attractiveness threshold. Then for a while, they remain open to switching to a better alternative if they come along. The attractiveness threshold changes; when people receive lots of offers from unattractive suitors, and no offers from attractive ones, then the threshold decreases.

Some time ago, I posted these four models on the Darwin Does Dating Twitter account and asked people to predict which models would win. And I’ve presented them above in descending order of popularity. Models 1 and 2 received 7% and 21% of the votes respectively while models 3 and 4 both received 36% of the votes.

Which model won?

When simulations were run based on data from real-life couples, Model 3 (Invest if they like me back) was able to recreate 46% of couples. That might not sound like much, until you realize that assembling couples by chance would recreate real ones around 2% of the time.

Were you a fan of Models 2 (Men do the rounds) and 4 (Keep your options open)? Well, these were able to recreate 36% and 29% of couples. While poor old Model 1 (You’ve worn me down!) recreated only 8% of couples. Better than chance… but not by much. “Invest more if they like me back” was a clear winner, and incidentally it was the model most grounded in modern evolutionary theory while also integrating basic behaviorist principles.

Predicting who shouldn’t be together

Couples weren’t recreated perfectly, of course. Part of this is because models are just that — basic representations. Mate choice is constrained and guided by many processes almost impossible to capture within a single model. At the same time, further analysis revealed that the couples which the algorithms did manage to recreate had higher real-life relationship satisfaction than those who didn’t. This implied that some of the couples in the study weren’t best suited to each other in the first place. Perhaps, in the future, we might turn to computers not just as matchmakers, but also as matchbreakers.

This method of testing which models can most faithfully recreate real mating decisions not only allows scientists to examine which of their mate choice theories is best, but it also gives exciting insight into the potential matchmaking technology of the future. It also gives people who question the validity of stated (vs revealed) preferences something to think about.

Facebook image: PeopleImages com - Yuri A/Shutterstock


Conroy-Beam, D. (2021). Couple Simulation: A Novel Approach for Evaluating Models of Human Mate Choice. Personality and Social Psychology Review, 25(3), 191–228.

More from Andrew G. Thomas PhD CPsychol MBACP
More from Psychology Today