The first principle is that you must not fool yourself—and you are the easiest person to fool.
Richard Feynman
In April 2025, two economists at the Federal Reserve Bank of New York, Jaison R. Abel and Richard Deitz, published two reports on the return on a college degree.1 2 The first, “Is College Still Worth It?”, concluded the median college graduate earns a 12.5 percent annual return on investment, exceeding both stock market and bond returns. The second, “When College Might Not Be Worth It,” published the same day as part two of a two-part series, found that the graduate at the 25th percentile earns a return of just 2.6 percent, that six-year completers see returns fall to 7 percent, and that graduates in fine arts, liberal arts, leisure, and education earn returns well below the median.
The first paper carried the headline number. The second carried the caveats.
That asymmetry matters because the Federal Reserve occupies a unique position in the higher education debate. Georgetown is a research center. The Institute for Higher Education Policy (IHEP) is a nonpartisan research and advocacy nonprofit. The Federal Reserve is the central bank of the United States. In my experience, when the NY Fed publishes a number, most readers treat it as settled. The 12.5 percent figure entered the national conversation with an authority the underlying methodology does not support.
The researchers are competent. The data sources are credible. The problem, as with Georgetown and IHEP, is what the analysis leaves out. The omissions are familiar by now. They start with understated costs, no taxes, and no discount rate. They extend further into how the Fed handles completion, how the Fed packages its findings, and what the Fed already publishes about graduate outcomes in another corner of the same building.
What the Fed Measured
The NY Fed calculates an Internal Rate of Return (IRR) by comparing the lifetime earnings of college graduates to those of high school graduates. The inputs:
The median college graduate earns $80,000 per year. The median high school graduate earns $47,000 per year. The Fed reports the annual premium at just over $32,000, or about 68 percent higher for the college graduate.1
Total costs are estimated at $180,000. The Fed arrives at this figure by combining direct costs ($30,000 over four years) with opportunity costs ($150,000 in forgone wages as a high school graduate during enrollment).1
The IRR, which equates the cost stream to the benefit stream over a working life of more than 40 years, comes out to 12.5 percent.
That number is real for the assumptions that produced it. The assumptions are the problem.
What the Fed Left Out
The Cost Understatement
The Fed calculates direct college costs at approximately $30,000 for four years. The arithmetic: average published tuition of roughly $21,000 per year, minus nearly $15,000 in average grants, other aid, and tax benefits. The Fed puts the resulting average net price, which also covers fees, books, and supplies, at about $30,000 over four years, or roughly $7,500 per year.1
Room and board are excluded from the 12.5 percent. The Fed’s reasoning: students need to eat and live somewhere, whether or not they attend college, so housing and food costs are not incremental to the decision to attend college.1 The Fed’s second post does test part of the gap. Adding half the average on-campus room and board charge drops the return to about 11 percent.2
The Fed’s cost figure falls short on two grounds.
First, in my experience, campus housing and meal plans usually cost more than living at home or sharing an apartment while working. The College Board reports average room-and-board charges of $13,310 at public and $15,250 at private nonprofit four-year institutions in 2024-25.3 That extra cost is a direct consequence of the enrollment decision. Excluding it understates the true cost of attendance. My model goes further. It counts the full cost of attendance, housing and food included, and does not charge the high school graduate for rent or groceries. That treats the whole bill a family pays for four years as the price of the degree, which is a stricter test than the Fed’s.
Second, the $30,000 figure uses average grant aid to reduce average published tuition. The distribution of aid is heavily skewed. In the College Board’s 2019-20 data, many low-income students received enough grant aid to cover tuition and fees.3 Students from middle-income families receive less grant aid than the lowest-income students, so the average can overstate the aid they will see.3 The College Scorecard reports average net prices by income bracket, so a family can check its own figure instead of relying on the Fed’s average.4
The Fed’s own sensitivity test reveals the scale of the problem. When they nearly quadruple out-of-pocket costs from $30,000 to $112,000, the return falls from 12.5 percent to 9.2 percent.1 They present this as evidence the return is robust. A nearly fourfold increase in direct costs produces only a 3.3 percentage point decline in the output. That is less reassuring than it sounds. Direct costs are a small slice of the Fed’s $180,000 total, which is mostly forgone wages. So the test moves the price families argue about and leaves untouched the assumptions that do the real work: pre-tax earnings, a four-year finish, and the median outcome.
Taxes
The Fed calculates returns using pre-tax earnings. The federal tax system is progressive. A college graduate earning $80,000 pays a substantially higher effective federal tax rate than a high school graduate earning $47,000.
In the model behind my book, We Need To Talk About Higher Education, the median college graduate pays about $32,300 more in lifetime federal taxes than the high school graduate, measured in present value (the book’s model chapters give the full NPV decomposition). At full cost of attendance, that moves the graduate from about $50,000 behind before taxes to $82,371 behind after them. At the Scorecard median net price, taxes flip the result, from about $14,800 ahead to about $17,500 behind. That tax is money the Fed’s pre-tax return counts but no graduate ever deposits.
An after-tax analysis narrows the wage premium by 20 to 25 percent. The 12.5 percent return is a pre-tax number applied to a decision families experience in after-tax dollars.
The Discount Rate Problem
The Fed uses the IRR rather than the Net Present Value (NPV) approach used in this book. The IRR is the discount rate at which the NPV of the investment equals zero. At first glance, 12.5 percent seems unambiguously strong. Earning 12.5 percent on any investment sounds like a good deal.
In my view, the IRR flatters the result here. A 12.5 percent rate only turns into 12.5 percent growth in wealth if every dollar of the wage premium earned along the way can also earn 12.5 percent. In practice, few families can do that. They spend it, save it at prevailing interest rates, or invest it in a stock market the Fed itself puts at about 8 percent over the long run.1
A modified IRR using a realistic reinvestment rate would produce a lower figure.
NPV is the better tool for evaluating college. The Fed’s IRR does net out the high school path, since it counts forgone wages as a cost and the wage premium as the benefit. But it reports a rate, not dollars. NPV compares the two paths in the same currency: present-value dollars. It simply asks what each stream of cash flows is worth today.
At a 7.8 percent discount rate and full cost of attendance, the median college graduate produces a lifetime after-tax NPV of $402,005 against the high school graduate’s $484,376. That is a net loss against the high school path. At the Scorecard median net price, which includes living costs, the graduate is still about $17,500 behind. The Fed’s own $7,500 a year leaves living costs out. The Fed uses Census Bureau and BLS survey earnings. This model uses College Scorecard earnings for college graduates. The rest of the difference is in what the model includes and how it handles time.
The Completion Assumption
The Fed’s first post acknowledges the completion problem in a single sentence: estimates “apply to college graduates; those who start college but do not complete a degree incur at least some of the costs but enjoy far fewer benefits.”1 The headline return leaves it out. In a later reply to readers on the same page, New York Fed Research estimated that factoring in dropout risk would likely cut the return by roughly 1.5 percentage points.1
That caveat describes roughly 39 percent of students who start at two- or four-year colleges, the share who have not finished any credential six years later.5 As Chapter 28 documents, they pay at least some of the costs and get far fewer of the benefits. The 12.5 percent return describes only students who finish, and only the median one who finishes in four years. The 39 percent who have not finished after six years are left out of the headline number. They absorb their costs and disappear from the calculation.
A rigorous return figure would weight the graduate outcome by the probability of achieving it. By my own calculation, which replicates the Fed’s framework and counts everyone who starts, including the ones who leave and the ones who take five or six years to finish, the pooled return falls to about 10 to 10.5 percent, a little below the Fed’s rough estimate.6 That sounds like a small change, and it hides the real one. A rate of return flatters small investments, so in the same calculation a student who leaves after two years can post a respectable rate and still end up about $31,000 behind at a 7.8 percent discount rate. And the 12.5 percent headline sits at or above the 65th to 70th percentile of everyone who enrolls, so about two in three students who start do worse. That is before correcting for any of the other omissions.
The Two-Paper Problem
The most revealing feature of the Fed’s analysis is not what either paper says. It is the split between them, and what the split allows.
Two papers. Same institution. Same day. One number, and its caveats in a separate post.
| Paper 1 · “Is College Still Worth It?” | Paper 2 · “When College Might Not Be Worth It” |
|---|---|
| 12.5% headline annual return for the median college graduate. | 2.6% return for the college graduate at the 25th percentile. |
| Median college graduate earns $80,000. Median high school graduate earns $47,000. Pre-tax annual premium of just over $32,000. | For students who take six years to complete, the return falls to 7%. |
| Total cost estimated at $180,000 ($30K direct costs + $150K opportunity cost). | Graduates in fine arts, liberal arts, leisure, and education earn returns well below the median. |
| Returns exceed historical stock market (8%) and bond (4%) returns. | Engineering, math and computers, business and economics, and health sciences earn returns above the 12.5% median. |
Source · Federal Reserve Bank of New York, April 2025. Both papers, side by side1 2
“Is College Still Worth It?” delivers the 12.5 percent headline. It travels. It gets cited. It reassures.
“When College Might Not Be Worth It” delivers the caveats. Graduates at the 25th percentile earn 2.6 percent, so a quarter earn that or less. Six-year completers see returns fall to 7 percent. Fine arts, liberal arts, leisure, and education graduates earn returns well below the median. Engineering, math and computers, business and economics, and health sciences earn returns above the 12.5 percent median.2
These are not minor qualifications. They describe outcomes for millions of graduates. The bottom quartile, at 2.6 percent or less, earns less than the 4 percent the Fed itself cites for bonds. The students choosing fine arts or education are not making irrational decisions. They are choosing fields with high social value and lower financial return. The 12.5 percent median includes them and the engineers, presented as though both face the same investment.
Separating the optimistic and cautionary findings into two papers allows selective citation, even though the first post previews the second’s finding in its opening paragraph. The first paper’s number can be quoted without the second’s caveats. Most families making the decision at 18 will meet the number before the caveats.
The Underemployment Number
The NY Fed maintains a data dashboard that tracks the labor market for recent college graduates. That dashboard, updated regularly, showed about 41 percent of recent college graduates (ages 22 to 27) underemployed in early 2025, working in positions not requiring a college degree.7
That figure has not dipped below 36 percent since tracking began in 1990.
For all college graduates, not only recent ones, the underemployment rate has held near 33 percent on average across the same period.7 At any given time, about one in three college graduates works in a job not requiring the credential they paid for.
Consider what that means for the 12.5 percent return. The 12.5 percent describes the median graduate, and the earnings pool behind it already includes the underemployed. But a median hides the spread. The underemployment data, published by the same institution, shows that at any point in time a substantial share of graduates work in jobs that do not require their degree, where the premium may be smaller or absent.
The 12.5 percent and the 41 percent come from the same building. They appear on different pages. Neither of the Fed’s two posts reconciles them.
Not every underemployed graduate earns high school pay. The Fed’s dashboard notes that many work in non-college jobs that are “fairly skilled and well paid” and move into better roles with experience. But take a graduate who earns $47,000 in a job not requiring a degree, and whose pay never pulls ahead of the high school path. On pay, that graduate looks the same as the median high school graduate in the Fed’s own earnings data. By the Fed’s estimate, that graduate gave up $180,000, most of it in forgone wages, to reach the salary the comparison group reached without that cost. The return on that specific investment is not 12.5 percent. It is negative.
What the Book’s Model Shows
The wage premium is not in dispute. The Fed’s analysis draws on Census Bureau and BLS survey earnings. This book’s model draws on College Scorecard earnings for college graduates. Both show the median college graduate earning more than the median high school graduate. The wage premium is real.
The question is what happens to the return when you account for what families experience: after-tax earnings, realistic costs including room and board, and the time value of money.
This book’s model, fully documented in Chapters 29 and 31, produces a lifetime after-tax NPV of $402,005 for the median college graduate at full cost of attendance and $484,376 for the median high school graduate. The Fed says 12.5 percent return. The book’s model says negative $82,371. At the Scorecard median net price, which includes living costs, the median graduate still ends up about $17,500 behind. The −$82,371 uses graduate earnings, the basis most favorable to college. The guided calculator at collegeroi.org also shows enrollee figures.
The Fed says 12.5% return. At full cost of attendance, the book’s model says negative $82,371. Different earnings data, and four variables make the difference.
| Variable | NY Fed’s treatment | This book’s treatment | Effect on the result |
|---|---|---|---|
| Wage premium | Median college $80K vs. median HS $47K. Pre-tax. | College Scorecard earnings for graduates, a different source. | Both show a real premium |
| Direct costs | $30,000 total over four years (using average aid against average tuition). | Full cost of attendance (College Scorecard). At the Scorecard median net price, which includes living costs, the median graduate is still about $17,500 behind. | Lower return |
| Room and board | Excluded from the 12.5%. Argument: students must eat and live somewhere anyway. Paper 2 tests half the on-campus charge (about 11%). | Included in the full cost of attendance. Average campus room and board runs about $13K–$15K per year.3 | Lower return |
| Taxes | Pre-tax earnings used throughout. | After-tax earnings. The progressive federal tax system narrows the wage premium by 20–25 percent. | Lower return |
| Discount rate | IRR. Reports a rate, not dollars, and never selects an explicit discount rate. | NPV at 7.8%, the cost of capital families actually face. | Lower return |
| Completion risk | Acknowledged in one sentence in the post and left out of the 12.5%. A later reply to readers estimated about 1.5 points. The 12.5% describes the median four-year finisher. | Not in the −$82,371, which counts graduates only. By the author’s own calculation in the Fed’s framework, counting everyone who enrolls cuts the return to about 10 to 10.5%.6 | Lower return |
| Headline output | +12.5% annual return | −$82,371 lifetime NPV (median college $402,005 vs. median HS $484,376 at full cost of attendance, about −$17,500 at the Scorecard median net price) | Opposite conclusion |
Source · NY Fed, “Is College Still Worth It?” (April 2025)1 · This book’s full model in Chapters 29 and 31
Both analyses use median earnings. Both compare college graduates to high school graduates. Beyond the earnings source, the difference in the −$82,371 is four variables: the direct cost of attendance, room and board, taxes, and a discount rate that reflects the real cost of capital and the time value of money. Remove any one of them and the college case improves. Include all four and it does not. Completion risk, counted separately, lowers the return further.
What This Means
Chapter 6 identified the pattern running through the major studies of the College Wage Premium: researchers trained in labor economics apply their discipline’s tools to a question those tools were not designed to answer. The four studies examined in these appendices confirm that pattern in specific detail.
The Federal Reserve is not acting in bad faith. Neither is Georgetown. Neither is IHEP. The researchers are competent professionals. But their unit of analysis is the population, not the individual.
Their question is whether education increases earnings at scale. The family’s question is whether this degree at this institution at this cost will generate a return. Those are different questions requiring different tools.
Across all four appendices, the pattern I see is consistent.
No taxes. Economists study pre-tax earnings because they measure productivity. A financial analyst would not evaluate an investment without accounting for the tax treatment of the returns.
No discount rate, or an artificially low one. Economists measuring a wage gap at a point in time do not need to discount cash flows. When they do discount, they reach for a low, near risk-free rate. In my view, that fits a method that treats education as a macro-level good rather than a risky individual investment. The CEW’s rankings used 2 percent, then zero percent.8 The Fed used IRR, which avoids selecting a discount rate at all.
Aggregation to the median. Economists report medians because they are studying the population effect. A financial analyst would not tell a client the median stock returned 12.5 percent without disclosing the variance, the probability of loss, and the specific risk profile of the investment under consideration.
Completion risk acknowledged in passing and excluded from the calculation. Economists studying the wage premium study graduates, because graduates are the population exhibiting the premium. The 39 percent who have not finished after six years are outside the frame. A financial analyst evaluating expected return would weight the outcome by the probability of achieving it.
The pattern is not conspiracy. It is not negligence. It is a discipline applying its own tools to a question those tools were not designed to answer, and, in my view, an industry amplifying the results because they confirm its preferred narrative.
When the Fed publishes a 12.5 percent return, the implicit message to families is: college is a good investment. When the same institution publishes a 41 percent underemployment rate for recent graduates, the implicit message is: at any given moment, the investment is not delivering a degree-level job for two in five recent graduates. Both numbers are real. Both are published by the same source. The 12.5 percent is the number I see reach families. The 41 percent sits in a different report, on a different page.
The data to evaluate college as an investment is public. The College Scorecard, IPEDS, the BLS, and the Census Bureau provide everything a student needs. What they do not provide is the methodology to assemble those inputs into an honest answer.
The studies in these appendices applied the methodology of economics. The model in this book applies the methodology of finance. The questions are different. The answers are different. The methodology is documented. The assumptions are stated. The tool is live at collegeroi.org. Substitute your own inputs and rerun if you disagree.
The appropriate response to the Fed’s 12.5 percent is not to reject the data. It is to apply the right discipline to the question the data is supposed to answer.
About this analysis
This critique appears as Appendix D (Federal Reserve Bank of New York) in We Need To Talk About Higher Education by Leon Shivamber.
Get the book → Read the argument in full → Run your own numbers →
Notes
- Abel, Jaison R., and Deitz, Richard. “Is College Still Worth It?” Federal Reserve Bank of New York Liberty Street Economics, April 16, 2025. Reports a 12.5% Internal Rate of Return for the median college graduate in 2024, calculated using Census Bureau and BLS Current Population Survey data. Total costs are estimated at $180,000 ($30,000 in direct costs plus $150,000 in opportunity costs). Median college graduate earnings: $80,000. Median high school graduate earnings: $47,000. Room and board are excluded from direct costs. Returns calculated pre-tax. In replies to readers posted on the same page by New York Fed Research (April 25, 2025), the Fed adds that raising out-of-pocket costs from $30,000 to $112,000, which it calls quadrupling, reduces the return to 9.2%, and that factoring in dropout risk would likely reduce the return by roughly 1.5 percentage points. The second post (2) models the $112,000 case as no aid plus a room-and-board wedge.
- Abel, Jaison R., and Deitz, Richard. “When College Might Not Be Worth It.” Federal Reserve Bank of New York Liberty Street Economics, April 16, 2025. Published the same day as (1), as part two of a two-part series. Finds a 2.6% return for the 25th-percentile graduate. Six-year completers see returns fall to approximately 7%. Returns vary substantially by field: engineering, math/computers, business/economics, and health sciences produce the highest returns. Fine arts, liberal arts, leisure/hospitality, and education produce the lowest. Adding half the average on-campus room and board charge reduces the return to about 11% (about 1.3 percentage points). One extra year to degree reduces the return by about a quarter, and two extra years by more than 40 percent. Chart data: https://libertystreeteconomics.newyorkfed.org/wp-content/uploads/sites/2/2025/04/LSE_2025_CollegeNotWorthIt_data.xlsx.
- Ma, Jennifer, Matea Pender, and Meghan Oster. Trends in College Pricing and Student Aid 2024. College Board, 2024. Table CP-1: average 2024-25 housing and food (room and board) charges of $13,310 at public four-year and $15,250 at private nonprofit four-year institutions. Figures CP-11 and CP-12 (2019-20 data) show the lowest-income students getting the largest average grants in every sector and selectivity group. Among dependent students with family incomes below $40,000, 79% at very selective public four-year institutions and about 55% at other public four-year institutions received enough grant aid to cover tuition and fees, though their average net total budget after grants was still $14,630 at very selective public four-year institutions.
- U.S. Department of Education College Scorecard. https://collegescorecard.ed.gov/ and https://collegescorecard.ed.gov/data. Net price by income bracket: U.S. Department of Education, “Technical Documentation: College Scorecard Institution-Level Data,” September 2025, https://collegescorecard.ed.gov/files/InstitutionDataDocumentation.pdf.
- National Student Clearinghouse Research Center. “Yearly Progress and Completion.” December 4, 2025. Of students who first entered two- or four-year institutions in fall 2019, 61.1% had completed a credential within six years, so about 39% had not. That 39% includes the 9.0% who were still enrolled after six years. The eight-year completion rate for the fall 2017 cohort was 64.8%.
- Author’s enrollee-weighted calculation (August 19, 2026), replicating the Fed’s framework and counting students who leave after two years and graduates who take five or six years: pooled return about 10 to 10.5 percent, a two-year leaver about $31,000 behind at a 7.8 percent discount rate, and the 12.5 percent headline at or above the 65th to 70th percentile of enrollees.
- Federal Reserve Bank of New York. “The Labor Market for Recent College Graduates.” Updated regularly. Data: https://www.newyorkfed.org/medialibrary/Research/Interactives/Data/college-labor-market/College-labor-data.xlsx.
- Georgetown University Center on Education and the Workforce. “Ranking 4,600 Colleges by ROI (2025).” February 26, 2025. FAQ: “In earlier reports, we used a discount rate of 2 percent on future cash flows” and “For the current version of this data tool, we use a discount rate of zero percent.”
Related critiques
This is one of four close readings of the major college ROI studies. See the others, then read the audit that weighs all four together and the plain-language case beneath them.
- Georgetown’s College Payoff, examined.
- Georgetown’s ROI ranking, examined.
- IHEP’s Rising Above the Threshold, examined.
- All four studies, audited together: Auditing the Numbers That Say College Pays.
- The plain-language version: College Is a Bet.
- The same problem priced against the alternative: Compared to What?
- The full argument: Is College Worth It?