Day 18: What I learned about the Economics of Higher Education
Over the last 18 posts, this series has followed my new course The Economics of Higher Education. Today, I’ll sum up what I learned and what I think colleges should do in this increasingly challenging environment.
Let’s jump right in with…
What is the point of college?
People usually go to college for three reasons:
College helps you build traits that are good
College signals that you have traits that are good
College is fun
Within “traits that are good” social scientists usually talk about two main categories of skills: cognitive and non-cognitive. Cognitive skills are things like learning quickly, learning deeply, remembering information, reasoning, building an argument, applying information from one context to another, problem solving, etc. Non-cognitive skills are things like grit, conscientiousness, ability to work in a team, conflict management, planning, adaptability, etc.1
These cognitive and non-cognitive skills can be both built or signaled through coursework or extra-curriculars. Hard classes will make you learn knowledge and practice grit, and having done well in hard classes shows that you are smart and hard-working. Playing on the basketball team is an opportunity to practice discipline, and having played on the basketball team proves that you have teamwork skills.
We also need to talk about peers. Peers in the classroom help us build both cognitive and non-cognitive skills. Peers can also help signal that we have cognitive and non-cognitive skills. When I went to grad school, I was offered a great scholarship at Michigan. One professor told me he thought it was because of Pat Kline, a current PhD student and also a graduate of Reed College. Pat had surprised the professors at Michigan with his talents and they were hoping that another Reedie (me) might do well also.2 I highlight this because people often think that concurrent peers are the most important ones. That’s true for building cognitive and non-cognitive skills. But for signaling skills, the peers that come before and after you are probably just as important.
Going back to our list up top, within those first two items…
Build traits that are good
Signal traits that are good
…there are really 8 buckets within each item, corresponding to three margins…
cognitive or non-cognitive skills
built or signaled through coursework or extracurriculars
through professors and coaches or with peers
… so there are 16 total buckets there of things that students may care about.
Lastly, item 3: Fun. What’s fun in college? It’s the classes right, tell me it’s the classes. Fine, maybe only some students think that classes are the most fun part of college (like the nerds who read this Substack!). Many students probably think that extra-curriculars are the best part. And plenty of students enjoy their peers for reasons having nothing to do with building or signaling cognitive or non-cognitive skills. They may also enjoy the place (I’m looking at you, Colorado mountains) and many have strong preferences about the dorms, food, and other amenities.
Ok, why am I spelling all this out. Because economics is the study of choices, and economists assume that agents make choices by optimizing within their constraints. In order to talk about what choices we think people will make, we have to understand what they’re trying to optimize. My point in walking through all this is to highlight that college students have many things in their college value function, at least 17 buckets of things by my count.
If you think of it from a product design perspective, colleges then have a really hard problem. Different students are going to have different relative values for each of those 17 buckets. Different activities within each bucket will appeal to different students. All of this matters for current students, the largest source of revenue for most colleges. And because alumni continue to care about the signaling value of college and are also a major source of revenue through gifts and their potential legacy admit children, colleges must keep their perspectives in mind too.
How do you design an optimal product when so many different people buy your product for so many different reasons?
And what happens when what students want changes over time?
On Day 2 we learned that at the turn of the 20th century the Second Industrial Revolution made building cognitive skills (specifically those that connected recent scientific innovation to profitable industrial uses) more valuable in the labor market. We also learned that before 1940 geographic mobility was low, so there wasn’t much competition between schools for students and the labor market was regionally segmented. After 1940 geographic mobility increased and so too did competition in many markets. That made signaling cognitive skills more valuable in the labor market, so students started to care more about the reputation or prestige of their college. Geographic mobility also made it possible for high academic ability students to concentrate in certain schools, so students started paying attention to the composition of their peer group in addition to quality of their professors.
That’s where the articles we read for Day 2 end. I’ll push it further from there. The US began transitioning from a manufacturing based economy to a service based economy in the 1950s, following World War 2. The chart below shows the fraction of US consumptions dollars spent on services (the solid blue line) and goods (the dashed red line). You can see that in 1960, Americans spent about 55% of their consumption dollars on goods and 45% on services. That ratio had switched by 1980 and today Americans spend nearly 70% of their consumption dollars on services.
I think this made non-cognitive social skills more valuable in the labor market. Students naturally wanted to build and signal those skills through college. Indeed, that great post from Jay Akridge and David Hummels I talked about on Day 15 says that the biggest gap between what employers say they want from college graduates and the skills college grads actually have is in non-cognitive skills. It’s then not surprising that colleges started investing more in extracurriculars like athletics, a major avenue to build and signal non-cognitive skills. I don’t have good numbers on how much colleges have spent on athletics over time, but we do know that fancy colleges now value athletics enough to give athletes an bump in admissions criteria. We also know that regulating how colleges engaged in sports became important enough that the NCAA named its first executive director in 1951 and the separation between Divisions I, II, and III was established in 1973.
To me all of this means that what college students want has changed tremendously over the last 100 years. Before 1940, students were mostly focused on building cognitive skills. After 1940, geographic mobility made students interested in signaling cognitive skills as well as seeking out a good peer group. As the service economy took off in the second half of the 20th century, students increasingly wanted to build and signal their non-cognitive skills, with their professors and coaches and also with their peers.
The big question now is how GenAI and maybe later AGI will change the labor market value of various skills and thereby the value college students place on building and signaling those skills. There’s an additional question about how young people’s options for building and signaling those skills is changing: college will likely remain a great way to build and signal some kinds of skills, but may fade in relevance, efficacy, or comparative advantage for others. The biggest omission in my syllabus this year is that we didn’t have a day on AI. Next year, I’ll do two: one on AI and the labor market and one on AI and education.3
So why does college cost so much?
Previous economists of higher education have emphasized that the reason why college costs so much is because colleges employ a lot of college educated workers, and employing college graduates has gotten more expensive over time. And through 2000, yes, I agree that was part of the story. But after 2000, I don’t buy it.
Take a look at this chart from Bengali, Valletta, and Zhao (2025) which we read for Day 4. This chart shows the gap between the earnings of college graduates compared to high school graduates. You can see that the relative wage of a college educated worker increased from 1980 through 2000 and then has been either stagnant or maybe falling. College educated workers stopped getting more expensive to employ 25 years ago.
I’ve also written several posts on this Substack digging into the data on expenditures at the top 100 liberal arts colleges. I just don’t see that schools are spending more on instruction since 2000, nor are professor salaries rising.
What I do see is that liberal arts colleges have increased their real spending on sports, arts, and student services staff, also library and academic support staff, also business and financial services staff, and also managers to manage all of those additional staff. And before you shout “I bet all those administrators are living high on the horse!” just, no. Real salaries have decreased for all but administrative assistants, business and financial services staff, and managers, and the raises for those groups have been less than what has been typical for college educated workers in other industries. The reason why real expenditures at liberal arts colleges have increased is because colleges now employ more people.
College is expensive because students now want to build cognitive skills, signal cognitive skills, build non-cognitive skills, signal non-cognitive skills, in the classroom and on the court, with their professors and coaches and also with their peers. Satisfying all of those wants takes a lot of people. That is why college is so expensive.
Will college always cost so much? What’s going to happen next in the economics of higher education?
A big reason why colleges were able to expand their programming in so many areas is because the number of students going to college has increased dramatically over the past century. Higher demand means higher revenues, and for non-profits that means more room for new expenditures.
But now that’s changing.
On Day 9 we talked about the Demographic Cliff and saw that the number of 18 year-olds born in the United States is set to start falling… now. And because college enrollment rates don’t have much room to increase, we should expect the total number of US born college students to decrease. We may even see college enrollment rates decrease. On Day 4 we talked about the returns to a college degree, or how much more money a person can expect to make if they finish college, and I just showed you the punchline chart that the returns to college have stagnated and maybe even started falling. Congress is also trying to reduce government expenditures on financial aid, and that One Big Beautiful Bill will reduce colleges’ own resources for financial aid also because of the expanded endowment tax. That’s sure to reduce demand also. Given the current political climate, the number of international students who want to enroll in US schools will probably decrease also.
Industry wide, demand for college will fall. Lower demand means lower revenues and maybe lower prices. For non-profits that must mean lower expenditures. And in an industry that is expensive because it employs a lot of people, lower expenditures will have to mean fewer employees.
The next question is which departments, programs, and services will be trimmed and by how much. Most colleges will cut something, because they are realistic and inherently risk-averse. But colleges shouldn’t cut too much, because some other schools will surely go out of business and our college could capture some of their previous customer base. If we cut too many or the wrong programs, students will go elsewhere. If we cut too few programs, we’re vulnerable to a few years of bad yields and then we’ll be the one going out of business.
So what should we cut?
If I were advising a college, I would tell them to go back to fundamentals. What do you think your students think is the point of college? Are they primarily looking to build skills or signal skills? Cognitive or non-cognitive skills? How do they think AI is going to change the value of their cognitive and non-cognitive skills? Do they want to build and signal through coursework or through extracurriculars? With professors and coaches or with their peers? How have their options for building and signaling skills outside of college changed? How much are they just looking for some fun? Which of the 17 buckets is your college great at now, and which can you continue to be great at even with some cuts? And even more important, which buckets do your students care about least? Decide in advance what you will be strategically, purposefully bad at so you have room to be great at the most important things.4
For the last 45 years, it was the best strategy for most colleges to try to do all things for all students. Colleges can’t afford that anymore. They’ll need to specialize. In what, exactly, is up to each.
And that closes out the series! I hope you enjoyed reading along with my class.
I’m going to take a break from this Substack for a while so I can get rolling on my next research project. Data collection wrapped up in Nairobi last week so I get to spend my summer playing with new data. There is nothing better than that!
Thanks for reading and I’ll see you in a few months.
The importance of non-cognitive skills was popularized in economics by Nobel Prize winner Jim Heckman, Colorado College alum of 1965. I got to hang out with him for a whole day once when he was in town to celebrate a dorm being named after him. Two students in this class lived in the Heckman House this year. They say he sends them Chicago-style popcorn every year.
Pat made full professor at Berkeley only 11 years after getting his PhD, his papers have been cited 18,000 times, and he’s currently a co-editor at Econometrica. I’ll leave it to you to discern whether I’ve lived up to his legacy or not.
I would LOVE your help in building ideas for my two days on AI and education and AI and the labor market. Right now I’m just flagging papers as I find them and in the late fall I’ll do a more concerted lit review.
I found this a list by Will Rinehart at AEI here. There have been several interesting NBER papers recently, and I expect we’ll see lots over the next 6 months. On the labor market side, there’s this one from Denmark (ahhh the beautiful administrative data in Denmark…) showing that while workers are using AI, they’re not working fewer hours because of it. There’s this RCT in the US (by another grad school classmate, Go Blue!) that shows that workers are using AI to reduce their time with email, but it hasn’t reduced their time in meetings. On the education side, this paper on learning in coding classes at a university shows that LLM usage has both positive and negative effects for students, while this paper on learning in high school math classes shows negative effects.
Those are the best ones I’ve seen so far. What have been your favorites?
This is my best advice for living a good life, by the way. It’s all about the opportunity costs.
This was so much fun to watch from afar!
Thank you for sharing. Good luck on your summer data analysis, it sounds like a joy.
Please keep me in the loop, I would absolutely love to learn and chat more about the role of AI in the workforce - anecdotally, many things are changing and I agree with you: specialization not only of colleges, but of inherent people skills will be the path ahead. Selling yourself and your specialization will be critical in the workforce.
Take care!
Amanda
I thought they were only trying to optimize money! " we have to understand what they’re trying to optimize " You need money to buy stuff, whatever it is you want to buy! No?
It was interesting to read how you roll. I don't "signal" anything at all because I am an Autistic individual. And we do not signal. However, you might like my posts under ECONO Posts since I have had an original idea that I classify under the subject area "Economics," although I do NOT enjoy dealing with data.