In Part One of this series, I explored the concept of Target Grad Rates, a fundamental component of high
quality data-driven advising. In this installment, we are going to dive into the other two key pillars: affordability and admissibility.
Before diving in, I do want to give thanks to Matt Niksch, Dr. Aide Acosta, and the team at Noble Schools out of Chicago (IL). Much of my knowledge on this topic comes from this network, and I am forever grateful
for their innovation and openness to sharing out their best practices.
So let’s start by giving a simple definition of admissibility and affordability in the lens of high quality, data-driven advising:
Admissibility: The likelihood a student is admitted to a particular postsecondary option.
Affordability: The amount a student (and their family) can financially afford annually for postsecondary education.
Why Admissibility and Affordability?
As we discussed in Part One, the Target Grad Rate metric helps us define what a quality option is for a student. However, without admissibility, we might set completely unrealistic targets for a given student. Let’s say the student has a Target Grad Rate
of 50%, meaning matriculating to any four-year university with a six-year graduation rate above 50% should be their goal. If that student decides to apply to Harvard, Yale, Stanford, and no other options, they likely will not be admitted to a single
college. So without taking admissibility into account, it is impossible to deliver high-quality advising.
What about affordability? Well, in the same vein, if you are advising students towards options that they and their family cannot afford, we are putting them at significant risk of stopping out. The foremost reason students drop out of college is an
inability to afford college costs. If we can control for this risk on the front end during college
applications and selection, we are giving our students a better chance of persisting through and completing postsecondary education.
On Admissibility
If you have ever heard about the idea of “college match,” then you also have some exposure to admissibility as a concept. There are a lot of ways to take a systematic approach to determining admissibility, but the goal is the same: you want to analyze
your historical application data to learn how to advise students in the future. I will explore a number of ways of determining this, including the pros on cons for each approach.
Visually with Scatterplots
Scatterplots are one of the most widely utilized tools by high school students because many college and career readiness platforms plot colleges on some kind of X by Y chart (usually a standardized test score on one axis and GPA on the other). The idea
is to use past data to inform future decisions. However, this comes at a cost. Unfortunately, this approach tends to lead to undermatching, as demonstrated in this study by Tomkins et al. This conclusion is quite logical; willfully leaning into the prospect of rejection is daunting, so students will look for options where they are firmly in the green. This, by definition, is undermatching, what students are doing
is finding likely schools instead of searching out a mix of match, reach, and likely schools.
Pros to Scatterplots
Automatically populated by most college and career readiness platforms/solutions
“Easy” to identify where a student falls
Visually represents general odds of acceptance
Cons to Scatterplots
Can lead to undermatching
Difficult to visualize if the data is too noisy
Can single out single applicants, like an athlete who was admitted below the standards of the average profile
Difficult to create match, reach, and likely break points
Categorization
Another technique that is commonly used for college match is a concept I will call categorization. The goal here is to place universities into groups of similar admissions characteristics and match a student to a particular group of universities. Many
schools employing this technique traditionally leverage Barron’s Ratings, which are no longer updated, so schools have taken to manually updating their groupings annually based on trends they see in college applications. Based on a student’s national
test score and GPA combination, they are assigned to one of these groups. All universities in that group are considered a match, more selective groups are a reach, and less selective groups are a likely school.
Pros to Categorization
Simple concept to explain
Introduces a method for labeling schools as match, reach, and likely schools
Visuals can be created that help a student or counselor quickly find their group of schools
Easy to connect academic growth goals to an increase in postsecondary institutional quality
Cons to Categorization
With the loss of Barron’s Categories, grouping institutions can be tedious.
Students in the top selectivity group can be deemed to have no reach schools, which we know is not true, as an Ivy League school will be a reach (or far reach) for almost every applicant.
There is no degree of “likely,” “match,” or “reach.” A particular school might be on the threshold of two categories for a particular student but this method would fail to identify that.
Unless combined with rigorous statistical analysis on the outcomes of students you serve, the groupings can be rather arbitrary.
Logistic Regression
Many advanced college counseling departments use a method called "logistic regression" to predict college admissions outcomes. It’s a statistical tool that analyzes a student's information—such as their grades, test scores, and other relevant factors—to
estimate the likelihood of being admitted to a specific college. Think of it like a sophisticated calculator that identifies patterns in the data and assigns a probability. For example, after analyzing a student's academic profile and comparing it
to past admissions data, logistic regression might predict that the student has a 75% chance of being accepted to a particular university. A logistic regression, in this context, will output a probability that a particular student will be accepted
to an institution they are considering.
To do this on your own, you can leverage solutions like R Studio and exports from your college and career readiness platform to generate a unique logistic regression model for your students. The goal is to figure out the degree to which a student’s inputs
(e.g., SAT/ACT, GPA, extracurriculars) impact their odds of admission. Logistic regression does not force an even assignment of significance in the inputs, so you could find that your students’ GPAs carry far more weight than their SAT scores in the
analysis.
Having done this many times for our partners, here is some advice on how to perform logistic regression on your own:
Use YOUR data. General models will produce results that do not mirror how you would advise a student.
Clean the data, remove outliers as best as you can.
Group institutions like you would for categorization and run a logistic regression on data you have from that group. It can help you build a more flexible model for all schools (and increase the sample sizes from which the model can draw).
More variables is not better. Typically, looking at one-to-two characteristics about a university and two-to-three characteristics about the student produces statistically significant results without overfitting the data. For example, group the universities
and look at their admit rates while considering GPA and SAT score for the student.
If you have a lot of data for a single college, build a model just for that institution.
Test your model. Ask questions like, “Does this mirror how I would advise a student?” You do not want the technology to differ from the guidance you would give without it.
Group your probabilities of admission into student-friendly terms. For example,
<15% is Far Reach, 15-50% is Reach, 50-75% is Match, and 75%+ is Likely. This will allow you to set list-building goals for your students quickly (i.e., 3 reach, 5 match, 3 likely applications).
Pros to Logistic Regression
Provides probabilistic insights (i.e., ”how likely to occur is X outcome?”).
The model will show how much each factor (GPA, SAT scores, etc.) affects the likelihood of acceptance, making it easier for counselors to explain the results to students and provide targeted advice.
Counselors can customize logistic regression models based on data relevant to particular types of schools, student profiles, or application cycles, making the advice more personalized.
Supports data-driven decision making.
Informs an application strategy.
Adaptable to multiple factors.
Cons to Logistic Regression
The accuracy of predictions depends heavily on the quality of data used to build the model. If data is outdated or biased, it can lead to incorrect predictions, misleading students.
College admissions decisions can be complex and subjective. Logistic regression may not effectively capture nuanced factors that impact admissions like personal essays, recommendation letters, or interviews.
Students might rely too heavily on the acceptance probability and discount the importance of other factors, potentially leading to unnecessary stress or overconfidence.
If the model is trained on historical data with inherent biases (e.g., preference for certain demographics or schools), it may unintentionally reflect and perpetuate those biases in its predictions.
Not all counselors may have the technical background to understand or explain logistic regression, potentially requiring training or external support to implement effectively.
Probabilities can be hard to comprehend. For example, 85% ≠ 100% yet students might feel like it is.
The Second Leg of the Stool
Admissibility, graduation rate targeting, and affordability comprise a three-legged stool that savvy district, schools, networks, community organizations, and others employ to help connect students with their next, best step after high school. No matter
the approach, whether one of the three described above or another that an organization employs, considering a student’s likelihood of admission is critical.
In the third post of this series, I’ll consider affordability, both how to calculate it and how to convey it to students and families making their matriculation decision.
Ryan Hoch is the Co-Founder and CEO of Overgrad. Overgrad is an all-in-one postsecondary access and success platform that guides students to and through high quality postsecondary pathways. From advisors to administrators, Overgrad streamlines processes and ensures data-informed practices for improved student outcomes.
Email to a Friend
