Reconciling discrepant universal screening data to improve decision-making: A Bayesian logistic regression approach

There are a substantial number of students with mental health needs who do not receive timely support. Universal screening is a promising practice for facilitating early intervention services. Multi-informant assessment is noted as a best practice for valid decision-making. However, this practice has not yet been applied to universal screening data. Universal screening utilizing a single rater (teacher) likely results in a significant number of students not being identified for support. This study (1) employed a Bayesian statistical model to incorporate students' background information (e.g., demographic variables; disciplinary referrals; social, academic, and emotional risk statuses) to generate estimates of academic risk, (2) used this background information to generate cut scores in a training sample and validate them in a test sample, and (3) identified the unique value of adding teacher and student self-reports with regard to sensitivity and specificity. Results demonstrated the promise of incorporating background information in the accurate identification of students with low, medium, and high risk for mental health needs. Implications for research and practice are discussed.