Machine learning meets partner matching: Predicting the future relationship quality based on personality traits.
Level 3 - non-randomized controlled study
Prospective longitudinal follow-up cohort study applying predictive modeling
PubMed 30897110 · doi:10.1371/journal.pone.0213569
What was done
Questionnaires measuring personality traits were administered to 192 partners to predict relationship outcomes 4 years later, including relationship satisfaction, conflicts, and separation intents. Linear regression models using machine learning techniques were evaluated using 10x10-fold cross-validation to assess the predictive contribution of individual personality traits, partner traits, trait similarity, and general versus relationship-specific traits.
What was found
Machine learning models explained 37% of the variance in relationship quality after 4 years. An individual's perceived relationship quality depended primarily on their own personality traits (actor effects). Partner traits and partner similarity provided no incremental predictive value beyond actor effects, and relationship-related personality traits were stronger predictors than general personality traits.
Why it matters
This indicates that future relationship satisfaction is largely driven by an individual's own relationship-specific dispositions rather than dyadic personality matching or partner characteristics, challenging standard assumptions in partner-matching algorithms.
Limits
The sample size was modest (192 partners), and validation was restricted to internal cross-validation without testing on an independent external cohort. Specific metrics distinguishing separation events from satisfaction scores were not reported in the abstract.