Predicting Divorce and Separation from Social and Demographic in US Adults: A Linear Regression Model

A linear regression model was developed to identify social and demographic factors associated with divorce and separation among US adults from the 2024 General Social Survey (GSS). Eleven predictors  (age, sex, education, number of children, income, religiosity, happiness, job satisfaction, hours worked, race, and political affiliation) came together to create three models which were then compared, including: A simple logistic regression modeling age, a multiple logistic regression, and an Elastic Net regularized regression (ENR). 10-fold cross validation and testing using withheld test data was used to assess the performance of the model. The ENR model displayed the best performance (ROC=0.722, test data AUC=0.671), resulting in moderate prediction ability between “Together” and “Divorced/Separated” (DivSep) respondents. Age was the strongest predictor, second being sex, followed by number of children, and religious attendance, in relationship dissolution. Happiness and job satisfaction were predictive of success. The sensitivity of the model was high, approximately 0.99, while the specificity was low, likely due to data imbalance (82% Together, 18% DivSep). The results suggest the use of demographic data for prediction models may hold valuable input to relationship stability. Future studies should incorporate longitudinal data, a larger variety of psychological predictors, and more balanced data.  Machine learning may have greater implications in helping decipher social patterns that are relevant to counseling, relationship research, and industry-dating-algorithms.