Producing long-term forecasts of individual student outcomes: An application of chain-linked models with short-span data and education policy regime change
Jeff Dominitz, Soobin Kim, Robert H. Meyer, & Andrew Rice
Parents, students, teachers, and counselors routinely monitor student outcomes to assess progress toward educational goals. Although many goals may be short-term (e.g., promotion to the next grade, proficiency on annual tests, or improved grades), each of these stakeholders is also concerned with students’ long-term goals and outcomes. We focus in this paper on using chain-linked predictive analytics models to produce forecasts of high school graduation and college persistence outcomes for students as early as 3rd grade. We estimate these models using four years of panel data on over two million students in a consortium of more than 120 California school districts participating in the CORE Data Collaborative. As these forecasts are updated over time with new information on student attendance, standardized test scores, coursework, and other intermediate outcomes, progress toward long-term goals may be tracked and obstacles identified as they arise.