Every admissions office uses a predictive model.
That’s a bold statement, but it’s true. Sure, plenty of schools don’t contract with a vendor to build them a predictive model, but every admissions office is making some sort of prediction about who is most likely to enroll and how many students will enroll.
Here are some examples of those models. For the first few, the admissions goals become the prediction, regardless of any evidence to the contrary.
The Last-Year-Plus-One Model
“We’re going to enroll one more student than last year.”
The Wish-Fulfillment Model
“The university president said we need 50 more enrollments, a 2-point bump in our median ACT and a 2 percent decrease in our discount rate. So that’s what we think the class will look like.”
The Out-of-Thin-Air Model
“I think we’ll enroll about 500 students this year.”
The High-Water-Mark Model
“Three years ago, our yield was 30 percent. I see no reason why we won’t do that again.”
The Better-Than-the-Last-Guy Model
“The last director of admissions was a bum. Surely I can increase enrollments by 25 students compared to that guy.”
It’s good to have lofty goals. The problem is when those goals become the prediction. The reason that’s a problem is because you don’t have a good way to know if you are likely to miss your goal. You don’t have any early warning signal that you need to do something different. You just end up with disappointment in September when you didn’t make your class.
Now, here are some examples that get a little more scientific but are still somewhat arbitrary.
The Single-Factor Model
“Students closer to campus are more likely to enroll,” or “students who’ve visited campus are more likely to enroll,” or “students from that state aren’t gonna come here.”
The Deposit-Rate-Comparison Model
“We’re up 20 percent in deposits date-to-date, so I think our enrollment will be up 20 percent.”
The Theoretical-Logit Model
“We think these seven things predict whether a student enrolls. Let’s test them in a logistic regression model. If the p-values look good, let’s roll with it.”
Or you could take an empirical approach, meaning you’re going to base your prediction on the things that can actually be shown to predict enrollment.
The Machine-Learning Model
“Take everything we know about students who’ve enrolled in the past versus students who didn’t enroll and let the machine sort through them and find the patterns that empirically predict whether a student enrolls or not.”
I’ll let you guess which approach Capture takes in our predictions.
See for yourself how Capture uses logistic regression, multiple types of data, automatic feature selections, dynamic scoring as well as how we can retrain your model within the enrollment cycle. Take a Free Envision Challenge and let us run your data for free. We’re eager to show you what we can do.
By John Foster, Senior Data Analyst, Capture Higher Ed