Key Predictors of Economic Well-Being of Single Mother Families: A Random Forest Model

Background and Purpose: Poverty is prevalent among single-mother households. Despite 79% of single mothers working full-time, 30% still live in poverty (Salas-Betsch, 2024). A substantial body of research shows that economic hardship is associated with increased family stress (Madsworth and Berger 2006) and can constrain parenting practices (Beasley et al. 2022), with downstream implications for child development (Engle and Black 2008). However, single-mother households are not a homogeneous group, and considerable variation exists in their economic circumstances. It is therefore critical to examine the predictors of economic well-being within this population to identify which families are most vulnerable, inform targeted policy responses, and guide more tailored interventions. In addition, we will utilize random forest, a machine learning approach that offers improved prediction accuracy and stability compared to traditional regression. This study is a preliminary analysis of predictors of single mothers’ economic well-being. The study aims to 1) evaluate the utility of machine learning for forecasting longitudinal economic well-being and 2) identify key predictors that may inform policy and practice.

Methods: Using data from the Fragile Families and Child Wellbeing Study (FFCWS), we examined economic well-being among single mother families at Year 9 (N=3,511). The outcome variable was a standardized composite index incorporating income-to-poverty ratio, vehicle access, housing stability, and material hardship indicators. Predictors included a total of 174 baseline and Year 3 variables capturing demographic characteristics, employment and income, parenting and father involvement, and relationship quality for both parents. We used random forest as the primary analytic method and compared its performance to a baseline model using Ordinary Least Squares (OLS) regression. Shapley Additive exPlanations (SHAP) were then employed to assess the direction and magnitude of each predictor's effect.

Results: Compared to OLS regression (Test R² = 0.20), random forest modestly improved prediction (Test R² = 0.26), supporting the added value of a machine learning approach. The strongest predictors were mother-father married at baseline (Importance Score (IS) = 0.089), father’s income-to-poverty ratio at baseline (IS = 0.078) and Year 3 (IS = 0.076), and mother’s college or graduate education at baseline (IS = 0.073). Mother-father disagreement at baseline (IS = 0.025) also ranked among the top 15 variables. SHAP scores indicated that prior marriage to the father and higher maternal education were associated with better economic well-being at Year 9, as was higher father income-to-poverty ratio—though most fathers reported low values. Greater mother-reported disagreement was associated with worse economic well-being.

Conclusion and Implications: Findings indicate that single mothers’ economic well-being is primarily shaped by structural resources and family-linked economic capacity. Marital status, paternal income, and maternal education were the strongest predictors, suggesting that these factors capture both access to economic resources and selection into more advantaged circumstances. Fathers’ sustained economic influence underscores the importance of considering family systems beyond household boundaries, while relationship conflict, though secondary, likely affects the consistency of support. Together, these findings call for policies that strengthen mothers’ human capital and engage fathers’ economic contributions, moving beyond individually focused or behavioral interventions alone.