Gendered differences in the factors influencing adolescent academic outcomes : An analysis of the Fragile Families and Child Wellbeing Study (FFCWS) using decision trees, a supervised machine learning method

Sammanfattning: School outcomes play a crucial role in the life of adolescents since they affect them throughout their life course. It is well-known that girls outperform boys academically during adolescence across most OECD countries. Previous studies utilize traditional linear methods to uncover gendered differences in the association between theoretically motivated factors and adolescent academic outcomes. The purpose of this study is to investigate if there are gendered differences in the higher-order non-linear interaction of individual, family, and school-level factors that affect adolescent academic outcomes using decision trees, a supervised machine learning method. Additionally, this approach enables one to see if there are gendered differences in the relative importance of individual, family, and school-level factors associated with adolescent academic outcomes. It will also help in validating previous studies that have shown gendered differences in the factors that influence adolescent academic outcomes.   Data comes from Fragile Families and Child Wellbeing Study (FFWCS), a longitudinal birth-cohort study conducted in the United States that follows parents and their children who were born in 1998-2000. It captures demographic characteristics, parenting behaviour, child behaviour, economic status, and many other variables linked to adolescent academic outcomes.   The results reveal that household income is the most important predictor of academic outcomes measured in terms of grade point average for adolescent girls, whereas fluid intelligence is the most important predictor for adolescent boys. Furthermore, girls from lower and middle-income families with stereotypically masculine behavioural traits are penalized academically. Parental academic expectation appears as a more important factor for boys as compared to girls. Overall, a combination of family resource capital and cognitive factors emerge as the most important factors across most trees for both boys and girls. Policy recommendations are discussed along with limitations and future studies.   Keywords: adolescent academic outcomes, gender, fragile families, supervised machine learning, decision trees