Predicting wellbeing in empty-nesters: an ensemble machine learning approach with data-driven feature selection

Area of Science:

• Psychology
• Computational Social Science

Background:

• Psychological research can benefit from machine learning (ML) predictive modeling.
• Worked examples translating ML logic to psychological questions are scarce.

Purpose of the Study:

• Demonstrate a predictive ML workflow using the empty nest as a case study.
• Investigate differences in wellbeing between parents experiencing the empty nest and those with children at home.
• Identify robust predictors of happiness and life satisfaction and examine group-specific predictor effects.

Main Methods:

• Utilized European Values Study data (N=43,670 parents aged ≥45).
• Employed propensity score matching, consensus feature selection (five methods, nested cross-validation), and three-phase group comparison.
• Applied a machine learning workflow with ensemble feature selection.

Main Results:

• No significant differences in happiness or life satisfaction were found between empty-nest parents and those with children at home, even after demographic matching.
• Health status and perceived freedom/control were the strongest predictors of happiness.
• Life satisfaction prediction involved a broader set of factors including personal resources, social connections, institutional confidence, and social capital.
• Predictors largely overlapped between groups, with limited robust group-specific interaction effects.

Conclusions:

• Wellbeing in midlife and later adulthood is influenced by factors beyond empty-nest status, irrespective of family configuration.
• Predictive modeling offers a valuable complement to traditional hypothesis-driven research in psychology.
• Findings align partly with Self-Determination Theory, emphasizing autonomy and relatedness.