Using Machine Learning Methodology to Identify Predictors of Non-Response to MTSS-B: a Focus on Discipline Problems and Juvenile Justice Involvement.

Abstract

Multi-tiered systems of support for behavior (MTSS-B) is a widely-used tiered preventive intervention currently used in over 25,000 schools across the USA. This study leveraged data from 16,907 elementary students across 42 schools that participated in a prior randomized controlled trial (RCT) of MTSS-B, where schools were randomly assigned to implement Tier 1 + 2 (intervention) supports or Tier 1 only (comparison). The RCT data, collected between 2008 and 2012, were linked to administrative records of behavior in grades 6-12 collected between 2008 and 2024. We used machine learning methods to identify elementary school predictors of three long-term outcomes (i.e., in-school suspension (ISS), out-of-school suspension (OSS), and arrest) signifying Tier 1 non-response. For ISS, the strongest predictors were family involvement, being female, and internalizing problems, which serve as protective factors. For OSS, they were family involvement, internalizing problems, and academic performance, all of which were protective. For arrest, key predictors included family problems, where lower problems were protective against arrest, and prosocial behavior and family involvement, which were both protective. The top predictors were similar for students in the Tier 1 + 2 and Tier 1 only conditions. These findings can inform early screening for possible non-response to MTSS-B and highlight the need for family engagement practices.