How the College Financial Health 'Strategic Compass' Determines Peer Sets

Name: How the College Financial Health 'Strategic Compass' Determines Peer Sets
Uploaded: 2024-12-11T17:47:36+00:00
Duration: 2 min 35 s
Channel: Gary Stocker
Description: In this explanatory video, Gary Stocker walks through the method the College Financial Health “Strategic Compass” uses to generate a peer set for a…

11 Dec 2024 · 3 min

Watch on YouTube

2:35

In this explanatory video, Gary Stocker walks through the method the College Financial Health “Strategic Compass” uses to generate a peer set for a target institution, illustrated with Bethany College. He describes a straightforward machine‑learning approach—K‑nearest neighbors (KNN)—that treats every U.S. college and university (public, private, for‑profit) as a point in a high‑dimensional space defined by dozens of institutional characteristics. The algorithm computes the distance between the target school and all others across metrics such as geographic distance, total and graduate enrollment, selectivity, endowment size, revenue composition, and Carnegie classification. By default, Stocker selects the 35 closest institutions (K = 35) as the KNN peer set. He also incorporates the self‑reported peer set that schools submit to the Department of Education, merging the two to check robustness and ensure the peer set reflects both data‑driven similarity and institutional perception. This peer‑set construction underpins the Strategic Compass’s benchmarking and financial‑health analysis for colleges.

Chapters

First, let me just give you a little sense of how this works. So there's an AI, not really fancy AI, it's a nearest neighbors machine learning algorithm to try to find the peer set. It's called KNN for short, K-nearest neighbors. Basically, intuitively what it is, is you have the whole world, the whole United States population of universities, right? And colleges, and have all the characteristics of private colleges, universities. And private. Yeah, I have private and public. I have for-profit. I have them all together, all the data. And what I do is feed all the characteristics of these schools into a big database. And in this case, we're talking about Bethany. So you have this target school of Bethany. And basically what the algorithm does is on all those characteristics, measure the distance between Bethany and all the other schools in the world. In the country on every metric and then sort of summarize that distance, right? And selects the K, uh, K I pick, right? The number I usually pick, 35. It'll pick the 35 closest peers on all, on cumulatively on all of those metrics. And not just geographic peers, right? Yes. So one of them is geographic, like how far away in miles as a bird would fly is the school from Bethany. How big is the school in terms of enrollment, different types of enrollment, graduate enrollment, professional student enrollment? How selective are they? What's the size of their endowment? How big is the revenue? What kinds of revenues do they have? How big are those? What's their Carnegie classification? So there are at least 2 dozen, probably more than that in this algorithm. And it's measuring essentially how similar other schools are to Daphne, and then I select the ones that are 35 that are the most similar usually. And that peer set here is the KNN peer set. Actually, in this case, it looks like I did— I didn't do 35 in this case. I did— I want to do 35 total. So I did 35 from that. I did like, I think I did 25 or so from the KNN. And then there's a self-reported peer set. And I get this. So the school, every school almost reports a peer set to the Department of Education. And I use that as well. And it's nice to have different peer sets. So I usually do several peer sets to just check how robust results are, but