In the Spring of 2013 Wisconsin tested a a data-driven early warning system that can identify which middle-school students are at risk-for dropping out of high school. After 5800 students were identified for teachers and counselors to work with, the principals of these schools were surveyed on whether they were already aware that these students were having trouble. With regard to most of the these students, the answer was, “yes”. The principals knew about them before the data told them.
But principals admitted that some of the students were not on their radar screen.
“All the missed students were females,” said Jared Knowles of the Wisconsin Department of Public Instruction, who presented the results of his model at the annual National Center for Education Statistics conference in Washington DC, STATS-DC 2013, on July 17.
Knowles had guessed that his data model might indeed find girls that high school administrators and counselors were overlooking. He suspects that boys who will eventually drop out of high school tend to have more overt behavioral issues. With girls, “it might be more subtle,” he said.
Knowles developed the early-warning-system model using regression. After trying out many variables, he found that he could determine with 60 percent accuracy who would eventually drop out of high school by looking a sixth grader’s attendance record, disciplinary record, state assessment scores and whether the student switched schools. Knowles said that the assessment test scores were particularly powerful in combination with the attendance record in predicting drop outs. And the more assessments he had, the more accurate the model became. Knowles plans to release his source code for any other state or school district to use and customize. “You can add GPA or whatever other data you have,” he said.
Many other U.S. school districts and states have or are currently developing data-driven models to predict drop outs. Administrators from Yonkers, N.Y. also presented their early warning system at the STATS-DC 2013 conference, but I didn’t get a chance to hear their presentation.
What to do about at-risk students once you identify them is still a mystery. The data doesn’t have answers for that…yet.
