Customer Story

Details Make A Difference: How EDUCAUSE Used Data Science Methods to Create Peer Groups with Purpose

Company Overview

EDUCAUSE is a not-for-profit association and the largest community of technology, academic, industry, and campus leaders advancing higher education through the use of information technology. The association provides guidance to its members—more than 2300 academic institutions—on what their peers are doing in the IT world, and how they’re using information technology to advance higher education.

Problem: Lack of data needed to provide valuable information

EDUCAUSE equips its community of members with the knowledge, resources, and community building opportunities needed to help inform decision-making around IT projects. One of the ways it does this is by providing a web portal where its members can identify peer groups to benchmark against each year. Users can create multiple types of peer groups based on institutional criteria—size, degrees offered, public/private, location, etc.—depending on the project they are investigating.

In an effort to incentivize its members to fully optimize peer grouping, as well as to participate in the research the association conducts, EDUCAUSE sought to provide a recommender system to members. The association wanted to provide guidance on how to best select peer groups—based on not only institutional criteria but using data from IT-level projects as well. This would allow members to drill down further and benchmark against schools that would otherwise not come up as a peer but could still qualify as a valuable comparison.

Solution: Drilling down to core information takes the guesswork out of the equation

Analytics8 developed an advanced analytics solution that allowed EDUCAUSE to segment its member institutions based on various datapoints and make it available to its member base. We did this by:

  • Analyzing an existing dataset from EDUCAUSE to compare institutions based on public information such as degree offerings, size, public/private, and sports affiliations and then adding additional features using Python to enhance the dataset.
  • Performing feature selection to determine which of the 80 features were the most informative or influential to creating potential peer groups.
  • Creating a machine learning model to select the 15 features that contained the most information and would best lead to the identification of peer groups.
  • Creating an analytics solution using Qlik Sense that integrated the machine learning model and showed the results of each cluster and what comprises each group. The solution includes peer group scores, comparison analysis, and ability to drill down into individual institutions.

Results: Data-driven information makes peer-grouping valuable

The analytics solution will provide the association the ability to easily recommend peer groups to members based on data science and further promote their own engagement with the peer-grouping tool. This new process will allow the association to better serve its members and continuously learn the different ways the academic institutions can influence each other.

The robust proof-of-concept has set up the association to retrain the machine learning models as it continues to collect more data from members—something EDUCAUSE is actively doing. With this solution, EDUCAUSE will be able to:

  • Provide guidance on what makes a good peer, which institutions each member should be comparing against, and why based on a scoring system.
  • Provide data on institutions based on an IT perspective—project level, budget level, and resource level—and identify which are similar or which ones are different but can still provide value because of aspirational goals that have been identified.
  • Fill in the gaps where there is a lack of research participation within peer groups and recommend alternative peers based on data science.

The machine learning data model offers value to those members that will still only want to use the basic peer groups because they will be able to see the evaluative score of how similar EDUCAUSE thinks their peer group is to their institution. And for those members that want the custom-level peer grouping, they will find value in the recommendations made available to them.