Predicting preschool problems

Abstract

Being able to predict which children will develop social-emotional problems is important for targeting interventions and efficiently allocating resources to preschools. We used the Strengths and Difficulties Questionnaire (SDQ) and preschool teachers’ assessments of 908 children aged 2–7 in 16 preschools in one Danish municipality, data from administrative registers, and a range of prediction models to examine how well social-emotional problems in preschool be can predicted. Although machine learning models typically make better predictions than linear or logistic regression, no model predicted either child or preschool-level social-emotional problems well. However, using the best-performing machine learning model, we obtained a predicted rank of preschools close to the observed rank (Spearman’s r = 0.69), which improved upon predictions based on income and earlier SDQ measures. Our results indicate that although using register data to target interventions to individual children is difficult, prediction models can improve the identification of preschools in need of extra resources.