Predicting Student Success: Rethinking GPA

In Academic Impressions' recent edition of Higher Ed Impact: Monthly DIagnostic, "Success Leaves Clues: Predictive Modeling in Higher Education," we interviewed a number of experts to provide a conceptual overview of how institutions can take steps toward a more rigorous mining of their current and historical student data to identify predictors not only of which students will be at-risk -- but predictors of success. Learning the shared characteristics of persistent and academically successful students can drive more informed investments in your recruitment and enrollment efforts.

This week, we wanted to isolate one of the "big" factors -- GPA -- and take a brief look at how you can take a more sophisticated look at high school GPA and first term GPA as predictors. For a few tips, we interviewed Bernadette Jungblut, West Virginia University’s director of assessment and retention, and Jim Scannell, president of Scannell & Kurz Inc. Here are some of the thoughts they shared with us.

Rethinking High School GPA

“Don’t just rely on overall high school GPA,” Jungblut warns, noting that often one’s first impulse is to isolate high school GPA as a more effective predictor than standardized test scores and other frequently-used indicators. But Jungblut suggests that you could learn a lot more by looking at GPA for high-school courses in specific subjects and then weight that by the number of units the student took in those subjects. For example, key data elements for your analysis could include:

  • High school math units
  • High school math GPA
  • High school science units
  • High school science GPA
  • High school social studies units
  • High school social studies GPA
  • High school English units
  • High school English GPA
  • High school foreign language units
  • High school foreign language GPA

This is not an exhaustive list; you will want to look at many other factors, such as advanced placement units and honors units. But this partial list does illustrate the type of data you will want to review in order to isolate specific factors that are more predictive of students' academic performance and persistence than other factors.

For example, Jungblut notes: “We have found that high school English GPA and English units are an incredible predictor of student success across all majors, much more so than the GPA in high school science courses. This is because in college you need to be able to write well, read well, and communicate well, whatever your discipline.”

In another example, you may find that GPA in specific mathematics courses and the number of mathematics courses taken is a good predictor of success for your engineering students.

First-Term GPA: Pinpointing the Cliff

"When you're working to understand the drivers behind who retains and who doesn't, typically first-term GPA is the gorilla in the room, the big driver of persistence to the sophomore year," Jim Scannell notes. He suggests isolating that factor with a high degree of specificity. "Where is the dropoff, the cliff?" he asks. Is there a specific grade point average at which the percentage of students who persist drops sharply? For example, if you find that at 2.0 and 2.5 there are relatively similar percentages of students who persist, but at 1.9 that percentage drops sharply, then 1.9 is your cliff.

Once you know this, use univariate analysis to begin defining two models:

  • Students with a 2.0 and above in the first term -- what are the common attributes of these students?
  • Students with a 1.9 or below in the first term -- what are the common attributes of these students?

You are looking for shared characteristics that may help explain the drivers of academic success for your unique student population, as well as the drivers of lack of academic success. Do most of your 2.0 and above students share a similar level of academic preparedness? Do most of them live on or near campus? Are your 1.9 and below students out-of-state students? Are they working students? Are there differences in the number of credit hours each cohort attempted, or in the number of credit hours each group earned?

Once you can describe shared characteristics of students within each of these two cohorts, this can guide more sophisticated analysis and predicitve modeling, and can help you make more informed decisions either in recruiting or in allocating resources to particular forms of student support.

Read our recent free issue of Higher Ed Impact: Monthly Diagnostic, "Success Leaves Clues: Predictive Modeling in Higher Education," for a conceptual overview of mining your data for clues as to the drivers behind students' academic success, retention, and affinity as future alumni.