Most Important Predictors of Father Involvement Among Politically Conservative Men: An Explainable Artificial Intelligence (XAI) Approach to Random Forest Modeling

Background and Purpose: In the 2024 Presidential Election, a majority (55%) of men voted for Donald Trump, with increased support observed among men across all races and ethnic groups. In parallel, 54% of fathers–men with children under age 18–identified with the Republican Party. These patterns reflect a general shift towards conservative values among men, as well as perceived challenges or unmet needs under the leadership of the Democratic Party. Fathers are included in this group of men who have felt disaffected or unheard. That said, little is known about politically conservative fathers, including what individual, family, and sociopolitical factors predict involvement with their children. Combining a heuristic model of the ecology of father-child relationships and an explainable approach to machine learning, we identified the most important predictors of father involvement among politically conservative fathers.

Methods: Data came from a national online survey of fathers who voted for Donald Trump in the 2024 Presidential Election (N=367). Predictors included 27 variables on men's beliefs about the role of fathers (e.g., essential for child well-being that fathers spend time playing with children), paternal status centrality (e.g., being a father is one of the most fulfilling experiences for a man), fathering challenges (e.g., unemployment, substance use), masculinity (e.g., traditional male role norms), mental health (e.g., depression, life satisfaction), political decision style (e.g., choosing a familiar political candidate), family functioning (e.g., spending time with family), and sociodemographic characteristics (e.g., race/ethnicity, education, employment, geographic region, child age) served as predictors. Father involvement with children (e.g., teaching, praising, providing) in the past year served as the outcome. Random forest was employed, combined with Shapley Additive exPlanations (SHAP) to enhance model transparency and quantify each predictor’s contribution to the model prediction.

Results: Descriptive statistics showed that fathers were on average 45 years old. Although fathers were predominantly White, married, and employed, they were diverse with respect to educational levels, income, and geographic regions. Fathers reported high levels of involvement with their children (M=3.27, SD=0.61, range: 0-4). The final random forest model showed moderate predictive performance, with an R²=.28 in the testing data (MSE=0.22, RMSE=0.47). Men’s beliefs about the role of father (SHAP=.09), life satisfaction (SHAP=.08), family functioning (SHAP=.07), paternal status centrality (SHAP=.06), and traditional male role norms (SHAP=.03) were the top variables predicting higher father involvement.

Conclusion and Implications: Politically conservative fathers are involved with their children at moderately high levels. Furthermore, beliefs about fatherhood and psychosocial well-being were some of the most important predictors of father involvement with children in the past year. Findings highlight the relevance of paternal identity-based processes and healthy family contexts for elucidating father involvement in politically conservative populations. Social work researchers advancing child and family well-being research through AI should employ explainable AI best practices to bring novel yet transparent insights that inform fatherhood programs and family policies in supporting healthy father-child relationships.