Is Romantic Desire Predictable? Machine Learning Applied to Initial Romantic Attraction.
Level 4 - case-series / case-control
Cross-sectional predictive modeling study using speed-dating cohorts (graded by design analogy).
PubMed 28853645 · doi:10.1177/0956797617714580
What was done
Across two speed-dating studies, unattached participants completed more than 100 pre-date self-report measures on traits and mate preferences. Participants then engaged in 4-minute speed dates with every opposite-sex participant at their event. Researchers trained random forest machine learning models to test how well pre-date measures predicted general tendency to desire others (actor variance), general tendency to be desired (partner variance), and unique romantic desire for specific partners (relationship variance).
What was found
Random forest models predicted 4% to 18% of actor variance and 7% to 27% of partner variance. Crucially, the models were unable to predict relationship variance (unique dyadic attraction) using any combination of self-reported traits and preferences measured prior to the dates.
Why it matters
These findings challenge commercial matchmaking algorithms and relationship theories that claim partner compatibility and mutual attraction can be accurately forecasted from pre-interaction questionnaire profiles.
Limits
The abstract does not state the sample size (n), participant demographics, or performance metrics beyond variance ranges. Findings are limited to brief 4-minute opposite-sex speed-dating interactions and do not measure attraction that unfolds over extended timeframes.
Cited by
- supports The traits that people report filtering for in the abstract are generally uncorrelated with what actually appeals to them when meeting a partner face to face.
- supports Predicting whether two unacquainted individuals will like each other based on measured actual similarity across multiple assessed traits performs no better than chance (a coin flip).