Skip Navigation

Predictors of Near-Term Academic Risks


Description: Many districts across the country are creating and using early warning systems to identify predictors of school dropout, but there has been less emphasis on predicting intermediate problems related to attendance, behavior, and course performance—known as the ABCs. By predicting these intermediate problems early, educators can target resources for the most at-risk students and intervene before problems get more serious. REL Mid Atlantic is working with Pittsburgh Public Schools (PPS), the Propel Schools charter school network, and the Allegheny Department of Human Services (DHS) to develop an approach for predicting which students are at risk for near- and long-term academic problems. This predictive model will use academic and behavior data collected by schools as well as DHS data on child welfare involvement, out-of-home placement, homelessness, receipt of public benefits, residence in public housing, and criminal justice involvement. The inclusion of DHS data will provide school staff a more holistic view of events occurring in students' lives to identify at-risk students in need of intervention or support.

Research Questions:

  1. Which student-level in- and out-of-school characteristics are independently associated with near- and long-term academic problems?
  2. How well do combinations of student-level in- and out-of-school characteristics and events predict near- and long-term academic problems?

Study Design: First, an exploratory analysis will test how individual predictor variables are related to outcome variables. Key predictors include demographic data, baseline measures of key outcomes, and DHS data. We will examine the relationship between each predictor and the following outcomes: chronic absenteeism, in- and out-of-school suspensions, below-grade level performance on math and reading state tests, core course failure, and insufficient credits for promotion to the next grade.

Next, using the same predictors, we will build predictive models that identify students who are most at risk for academic problems. We will use a machine learning approach that is designed to extract the most relevant information from a data set, which is useful when there is no strong theory to guide the way different predictors interact with one another and when events occur over time. The model will not only use information about whether in- or out-of-schools events occurred, but it will also take into account the timing and sequence of events to enhance the accuracy of predictions.

Projected Release Date: Spring 2020

Partnership or Research Alliance: No associated alliance

Related Products: Making Connections report

Principal Investigators & Affiliation:
Julie Bruch, Mathematica Policy Research