What Tears Couples Apart: A Machine Learning Analysis of Union Dissolution in Germany

Area of Science:

• Demographic Research
• Social Sciences
• Computational Social Science

Background:

• Union dissolution is a significant demographic event.
• Traditional regression models have limitations in predicting relationship outcomes.
• Machine learning offers advanced analytical capabilities.

Purpose of the Study:

• To apply machine learning, specifically Random Survival Forests (RSF), to predict union dissolution.
• To compare the predictive accuracy of RSF with conventional regression models.
• To identify key predictors of union dissolution using a novel methodology.

Main Methods:

• Utilized Random Survival Forests (RSF), a machine learning technique.
• Analyzed data from 2,038 married or cohabiting couples from the German Socio-Economic Panel Survey.
• Compared RSF predictive performance against traditional regression models.

Main Results:

• RSF demonstrated significantly higher predictive accuracy for union dissolution than conventional models.
• Key predictors identified include life satisfaction (both partners) and woman's housework percentage.
• Other significant predictors include woman's working hours and marital status.

Conclusions:

• Machine learning techniques, particularly RSF, offer substantial benefits for analyzing union dissolution.
• RSF can uncover complex patterns and interactions missed by traditional methods.
• The study highlights the potential of ML for broader demographic research.