Title page
Contents
Why this study? 2
Research questions 3
Findings 7
Limitations 14
Implications 15
References 17
Table 1/Table 3. Number of students identified as at risk at any point in 2016/17 using optimal risk score cutoffs, by outcome and local education agency 13
Table 2/Table 4. Most important categories of predictors in predictive analysis, by outcome and local education agency, 2016/17 14
Figure 1. Heat map showing differences in probability of academic problems for students with prior academic problems in adjacent time periods during the 2015/16... 8
Figure 2. Heat map showing differences in probability of academic problems for students with selected types of human services involvement in adjacent time periods... 9
Figure 3. Model predictions are strong for the Pittsburgh Public Schools sample during the 2015/16 and 2016/17 school years, by grade level 11
Figure 4. Model predictions are somewhat weaker for the Propel Schools sample than for the Pittsburgh Public Schools sample during the 2015/16 and 2016/17 school... 11
Figure 5. Optimal risk score cutoffs by outcome and local education agency, 2014/15-2016/17 12
Boxes
Box 1. Key terms 4
Box 2. Data sources, samples, and methods 5
Box 3. Interpreting the heat maps 7
Box 4. Relationship between the number of observations and model performance 12
Box Tables
Box table 1. Data elements provided by the Allegheny County Department of Human Services, 2012/13-2016/17 6
Box table 2. Description and level of observation of student academic outcome data for Pittsburgh Public Schools and Propel Schools, 2014/15-2016/17 6