Predictive Models in Unpredictable Times

Let’s be real. We don’t know what we don’t know. Predicting anything in the future is difficult. It is especially complex when factors existing outside of your data can have a substantial influence on future behaviors.

Still, this doesn’t mean that your historical data aren’t informative. On the contrary. The COVID-19 pandemic is exactly the time you should lean on historical data and statistical predictions. Even if predictions are less certain than in earlier years, they are better than guessing.

Given the global scope of the COVID-19 pandemic, we hypothesize that, for the most part, students who were planning to go to college in 2020 will still go to college in 2020. Most will enroll at the institutions to which they have already applied and shown demonstrated interest. We will be testing this hypothesis, in real time, on the lead indicator of deposits as the 2020 admission cycle progresses, and we’ll modify our assumptions if the behavior of deposits changes.

At Capture, we are leveraging the most advanced technology in the industry, combined with decades of enrollment management experience, to assist you during these uncertain times:

  1. Students’ deposit behavior might look different over the next couple months but will remain an important leading indicator. Our machine learning models are built to leverage this behavioral data.
  2. Historically, visiting campus has been a powerful predictor of enrollment, but the utility of that measure for predicting 2020 enrollments changed mid-cycle after schools canceled their visit programs. Capture’s data science team has modified the coding of campus visit in our models to ensure that we’re only including visits that occurred before March, both historically and in the current year.

Our advice is to keep up the great work and stick to the fundamentals. Use predictive models to allocate resources. Focus yield efforts on students who are most likely to enroll. Pay attention to deposits.

We’re here to help.

Innovate USA

For more on this topic, be sure to join Capture Data Scientist John Foster Wednesday, June 10, at 1:10 p.m., for his Innovate USA session, “No Guessing: Using Predictive Models in Unpredictable Times.” Predictive models are a foundation for the duties of all enrollment professionals — from financial aid distribution to implementing yield strategies. The pandemic adds another variable for us to work with and reaffirms why we must continue to revise and use data models. 

Predictive Models

In John’s session, you will journey through Capture’s predictive models — APPLY, AID and ENROLL — to see how hundreds of variables and multiple algorithms can be paired with historical and behavioral data to predict who will apply and who will enroll with unrivaled accuracy, giving you the information you need to make your class … even in uncertain times.

Go here for more information about this session and everything else Innovate USA will offer.