Title page
Contents
Why this study? 2
Research questions 4
Findings 6
Implications 11
Limitations 13
References 14
Appendix A. Methods 16
Appendix B. Supporting analysis 24
Table 1. Definition of outcomes for risk score, outcomes for PPS flags, and prior performance PPS flags 5
Figure 1. Percent of predictions that correctly predict either there will or will not be a problem, 2016/17 6
Figure 2. Among student-quarters with an academic problem, percent correctly predicted to have an academic problem, 2016/17 7
Figure 3. Among student-quarters without an academic problem, percent predicted to not have an academic problem, 2016/17 8
Figure 4. Among student-quarters predicted to have an academic problem, percent that actually had an academic problem 2016/17 9
Figure 5. Percent of student-quarters with a wrong prediction, by race, 2016/17 11
Figure 6. Guide for districts in deciding risk score cutoffs 13
Boxes
Box 1. Key terms 4
Table A1. Data elements provided by Pittsburgh Public Schools, 2012/13-2016/17 17
Table A2. Data elements provided by Allegheny County Department of Human Services, 2012/13-2016/17 17
Table A3. Sample size in Pittsburgh Public Schools, 2016/17 (number of student-quarters) 19
Table A4. Demographic characteristics and school service eligibility of sample, 2016/17 (percent of student-quarters) 19
Table A5. Frequency and duration of student involvement with Allegheny County Department of Human Services, 2016/17 (percent of student-quarters) 20
Table A6. Frequency of outcomes in 2016/17 (percent of student-quarters) 21
Table A7. Confusion matrix 23
Table B1. Characteristics of student-quarters predicted to have at least one academic problem, all outcomes, 2016/17 24
Table B2. R-squared values from regression analysis, 2016/17 (percent of student-quarters) 30
Table B3. Chronic absenteeism 31
Table B4. Course failure 33
Table B5. Low grade point average 35
Table B6. Suspensions 37
Figure B1. Percent of student-quarters predicted to have an academic problem, 2016/17 26
Figure B2. Among student-quarters with an academic problem, percent correctly predicted to have an academic problem, 2016/17 27
Figure B3. Percent of student-quarters over-identified, by race and outcome, 2016/17 28
Figure B4. Percent of student-quarters under-identified, by race and outcome, 2016/17 29