The Hamilton County Department of Education (HCDE) oversees nine K-12 school districts in and around Chattanooga, Tenn. Evaluating and improving school performance became a critical task for the districts, owing largely to No Child Left Behind. HCDE officials knew the students in their districts were scoring below state target levels, but it was difficult to understand why—and without that understanding, it was nearly impossible change the situation. HCDE also had a high dropout rate among high school students, and officials wanted to reduce that number.

Administrators chose IBM’s SPSS Modeler and SPSS Statistics software to take a deeper look at student performance by combining data sources and exploring variables beyond what the state reports provided.

Now, HCDE evaluates student performance and keeps students on track earlier in their academic careers by analyzing students’ test scores and combining that information with information on student attendance, behavior, parent information, class schedules, and other data. “Then, we can go through and begin to look at making predictions and identifying students who might be at risk,” says Kirk Kelly, director of testing and accountability for HCDE.

For example, even before the school year begins, teachers now have a remarkable amount of data on the makeup of their classrooms and on which students might require additional instruction and focus. A teacher will know in September if the data predict that a student will not perform well on the early college assessment ACT Explore test, which takes place in November. That student then can be given the extra attention needed to bridge the gap and, ideally, exceed expectations.

This has led to an improvement in test scores, with Hamilton County students performing well above the national average for the ACT Explore test in English, math, reading, science, and overall composite categories for the last three years.

HCDE also noticed a trend having to do with dropouts: Kelly learned that 63 percent of all dropouts were over the average student age. In fact, that was the biggest indicator contributing to the high school dropout rate. Kelly looked at students from kindergarten through high school to discover just how and when they become overage students.

“A student might be retained and then retained again, held back a year for issues such as athletics or maturity,” Kelly explains. “Then they run into problems.” What 21-year-old, he points out, wants to remain in school with a bunch of 18-year-old kids?

Understanding the high correlation between overage students and dropout rates allows HCDE to be proactive. Officials can identify a student coming into ninth grade who is already 16 or 17 and help the student before he or she gets into trouble, Kelly says. Even earlier in the process, educators can make sure that students—particularly those who have late birthdays—don’t get held back more than is absolutely necessary.

“By making schools aware, we’ve gotten numbers down to a very small percentage of students who are overage being retained. We’re also taking steps to provide help,” Kelly says. HCDE has been doing this for about seven years now; the group of students containing fewer overage children has begun to move into high school, and dropout rates have improved significantly. In fact, HCDE saw dropout rates go from 30 percent to 22 percent over the past year.

HCDE also uses the IBM solution for teacher incentives. “We go through and estimate the scores a student would make based on past history. We predict where a student will score, and track teachers who beat those predictions. Then we rank those results, and if a teacher is in the top 20, they receive an incentive,” explains Kelly.

Kelly’s department started out as a group of three people several years ago, but now it has eight people using the analytics system and looking at anything that might have an impact on student achievement.

The department purchased the base analytics package for about $4,000 in 1998. As they improved results, administrators recognized a greater need—and the department increasingly received a larger budget.

Kelly suggests that education leaders who want to begin doing predictive analytics bring together a group of stakeholders to decide what kinds of information they want to be able to capture, and to begin making sure they have good data. There might be eight or so data points the stakeholder group agrees on, such as age, ethnicity, income, and other variables. Once those have been decided upon, the group should make sure that every student record contains all of this information, and that it is accurate.

“Look at outliers,” he suggests. “Do you have a two-month-old high schooler, or a 99-year-old first grader? Then you might have a transposed birthday. Flag those. Then go through and flag missing variables. The people who are pulling the data and the people who are entering the data will have to interact.”

Good, clean data strengthens your success rate as you look at analytics to head off problems before they occur, Kelly says.

—J.N.

About the Author:

Laura Ascione

Laura Ascione is the Managing Editor, Content Services at eSchool Media. She is a graduate of the University of Maryland's prestigious Philip Merrill College of Journalism. When she isn't wrangling her two children, Laura enjoys running, photography, home improvement, and rooting for the Terps. Find Laura on Twitter: @eSN_Laura http://twitter.com/eSN_Laura