FAST LoginSKORE Login

Limits of Modeling - Monday Musings

By Jennifer Wick on Apr 23, 2012

Predictive modeling is a powerful tool, but it isn’t perfect.

Tags: admissions, enrollment strategies, data


When I was part of an admissions office, every day in April at 10:00 AM, you could have mistaken the tension and excitement in the office for an episode of Discovery Channel’s Deadliest Catch, brought on by the arrival of the mail bin. Enrollment deposits that arrived via phone were harkened by the ringing of a bell, and we each took guesses at what the final tally for the day would be once the web payments were counted. If we had a particularly good day, the news traveled like wildfire across campus. Wherever I went, on campus or off, if I ran into someone I knew, I would invariably be asked how the numbers looked. Anticipation was in the air, even with sophisticated data tracking and aid strategies developed through predictive modeling in place.Predictive modeling has some limits, but is a useful tool.

Why? At the heart of it, predictive modeling is a powerful tool, but it isn’t perfect.

  • Predictive models are based on prior-year data, so major differences in the pool can dampen their accuracy.
  • An econometric model’s ability to predict enrollment behavior is also sensitive to outside factors such as unanticipated changes in competing institutions’ policies and sudden changes in state grant funding.
  • While the variables in an econometric model certainly are key decision factors, even the best models only explain a portion of the variance in enrollment behavior. Variables not in the model (and not even captured by an institution) certainly play a role as well. (These are teenagers, after all!)


While econometric modeling can introduce more predictability about enrollment results, it can’t completely eliminate the anticipation surrounding what enrollment results will be, nor is it a replacement for strong, sound leadership. What it can do is provide powerful information for strategic decision making, when used in the right context.

Image © iStockphoto.


Jennifer WickAbout the author: Enrollment Management Consultant Jennifer Wick joined the Scannell & Kurz team in May 2011. She manages the Financial Aid Strategy Tool (FAST) and provides consulting on a wide range of enrollment management topics from admissions to retention.

Jennifer comes to S&K from Clarkson University where she served for nine years as Director of New Student Financial Assistance. Her experience encompasses both financial aid and admissions responsibilities, specializing in balancing financial aid strategies with enrollment targets and retention concerns. She has conducted enrollment data analysis, managed regional territories, and created successful financial aid strategies for targeted demographics. In addition, Jennifer has a keen interest in retention, which includes predictive model development to identify at-risk first-year students.

Jennifer earned a B.S. and M.S. in Physics, both from Clarkson University. You can connect with her on LinkedIn.


Comments (2)

  1. James Baldwin

    James Baldwin
    University of Pittsburgh at Bradford
    May 01, 2012 at 11:48 AM

    You made excellent points concerning predictive modeling. Another issue is the size of your population (n) by which your models have been developed. Variance is a much larger factor at smaller institutions.

  2. Jennifer Wick

    Jennifer Wick
    Scannell & Kurz
    May 03, 2012 at 08:30 AM

    James, that's a great point. In our work, we use several years of data to counteract this issue. Although sometimes, particularly with transfer students, there just aren't enough records for a stable in model. When this happens, we can still perform meaningful table analyses.






Allowed tags: <strong><em><br>Add a new comment: