Identifying Key Factors in Children's School Readiness: Evidence from Machine Learning

• In Event: Language development and executive functioning

Abstract

Introduction. Children’s advanced language ability and spatial reasoning have been considered as crucial factors for their school readiness(Dobinson & Dockrell, 2021). The early language and spatial reasoning abilities are particularly imperative for children from low SES families, as they are at a higher risk of delayed development and poor academic function in later life. In previous studies, lack of diverse parental vocabulary, sensitive parenting, and literacy activities have been identified as factors contributing to the underdevelopment of low SES children (Pancsofar et al., 2010)
Yet there are two major gaps in the existing literature: First, only a few studies have attempted to account for the multiple dimensions that influence children’s language and spatial reasoning. This is because the conventional statistical models have limitations in that as the number of variables increases, the probability of error increases due to the decrease in degrees of freedom (Atumo, 2022). Second, limited attention has been paid to how maternal, paternal, and parental (maternal and paternal combined) language interactions contribute differently to children’s language and reasoning adjustment. Parke (2002) found that while fathers tend to engage in less verbal and more play-centered activities, mothers were likely to engage in more verbal and directed activities. Therefore, in order to fully understand children’s school readiness, it is important to consider their mothers’ and fathers’ distinctive language involvements. To address these issues, we identify key factors that predict children’s language and spatial reasoning abilities by using machine learning methods to jointly consider multiple variables that affect children’s language and spatial reasoning development, and by considering the language contexts and practices of the mother, father, and both parents.

Data and Measures. Participants of this study include children who participated in the Family Life Project (n=1,292). Information on the environment, family, and child's characteristics was collected at 24 months. Parent-child language interactions were coded from observations of the mother-child and/or father-child book reading contexts. Children’s language and spatial reasoning abilities were measured by the Preschool Language Scale and the Wechsler Intelligence Scale at 35 months, and then divided into two groups (above and below average indicated in the standard scales).

Methods. The model comparison test shows that among five different models, the Random Forest model had the highest classification score. Next, we calculated the average accuracy and feature importance by creating a new model with 1,000 independent data sampling using the Random Forest model. We identified the top 10 independent variables among the average feature importance of 1,000 iterations (​​Kirasich et al., 2018).

Results. Our preliminary results show that both mother, father, and children’s own complex language usages are important factors for the later language and spatial reasoning development. Interestingly, in the mother’s context, maternal parenting and language styles are important, but for the fathers, children’s own characteristics in addition to father’s language styles are important factors. This may suggest that different interaction styles between mothers and fathers contribute to children’s language and spatial reasoning development. This study has practical implications for parenting, education, and clinical work for children’s development.

Authors

• Soo Gyeong Ju, University of Pennsylvania

• Baeksan Yu, Gwangju National University of Education / Norwegian Institute of Public Health

• Sungmin Yang, Seoul National University
  Presenting Author

• Douglas A Frye, University of Pennsylvania