Exploring father–adolescent closeness: A random forest approach

Objective

This study demonstrates how machine learning, specifically random forest, can advance family science, particularly in studying father–child relationships.

Background

Fatherhood research faces challenges with fathers' recruitment and retention, complex living arrangements, and lower response rates compared to mothers. Machine learning, a tool of artificial intelligence, effectively examines large and complex data sets, handles missing data, and identifies relationships between predictors and outcomes. Thus, machine learning can help mitigate the methodological challenges of studying fathers and father–child relationships.

Method

We used random forest to predict adolescent-reported father–adolescent closeness in the Future of Families and Child Wellbeing Study (n = 2,927), using 131 predictors measured during the first decade of childhood.

Results

Fathers' residential status with the child was the strongest predictor of father–adolescent closeness. Using random forest results to inform variable selection, we demonstrated how random forest can enhance the development and performance metrics of regression models.

Conclusion

This study highlights the utility of random forest for studying complex questions, such as how family contexts predict adolescents' perceptions of their father–child relationships.

Implications

Random forest is a feasible and useful approach for applied family scientists to incorporate artificial intelligence into their research, moving the field in new and meaningful directions.