The Common Data Set asks every college for the same GPA distribution on the same scale. Colleges answer in ways that cannot be compared, and the most selective ones often decline to answer at all.
Read the 2024-25 Common Data Set filings straight down the column and you would conclude that the University of North Carolina at Chapel Hill enrolls a stronger class than Stanford, and that Stanford narrowly edges Harvard. UNC reports that 97 percent of its entering first-year students arrived with a 4.0 high school GPA. Stanford reports 73.3 percent. Harvard reports 72.4 percent.
Two lines lower, the same three filings report an average high school GPA. UNC says 4.49. Harvard says 4.21. Stanford says 3.94. Two of those averages are impossible on an ordinary unweighted 4.0 scale, which tells you something differs in how these colleges are reporting GPA. It does not tell you exactly what.
Three schools, three sets of numbers that cannot safely be interpreted without knowing how each institution built them. That is the finding of this analysis, and it is a more serious problem for families than any single college’s figure.
We matched 188 selective institutions in the repository, collected 345 of their Common Data Set filings, pulled the GPA items from each, and checked every number against the filing it came from.
The Form Asks for One Scale and Gets Several
Item C11 asks for the percentage of enrolled first-year students in each of nine GPA bands, and its instruction is explicit: report the ranges “using 4.0 scale.” Item C12 asks a different question, the average high school GPA of students who submitted one. Nothing in the form requires the two items to rest on the same calculation.
Union College demonstrates the gap. Its 2024-25 filing reports a C12 average of 91, plainly a 100-point figure, while filing C11 in the standard 4.0 bands with 14 percent of its class at 4.0. Stony Brook does the same thing, reporting a C12 average of 92.91 alongside a C11 distribution in 4.0 bands. Both filings are internally coherent only if C11 and C12 are answering on different scales.
This matters for what can and cannot be concluded from the data. A C12 average above 4.0 is a warning that an institution’s GPA reporting is not on a plain unweighted scale somewhere in its filing. It is not proof that its C11 distribution is weighted. A C12 average below 4.0 is not proof of the opposite, since a weighted system can produce a class average of 3.8. Throughout this piece we describe what C12 says and stop there.
Most Available Ivy-Plus Filings Do Not Report a GPA Distribution
Before the comparability problem, there is an availability problem. Of the 188 institutions in our set, 88 produced a usable 2024-25 C11 distribution for this analysis. Forty-two filed the item blank or entered it as not available, and another 23 filings contain no C11 we could locate, which is a limit of our reading rather than a statement about the institution. The remaining 35 fall out for reasons described in the method note: 18 have no 2024-25 filing linked in the repository, 7 report only a test-submitter or weighted column, 4 report bands that do not sum to about 100, 3 report no C12 average, and 2 link to another institution’s file.
The blanks concentrate at the top. Among the twelve Ivy-Plus institutions, five report a GPA distribution: Harvard, Penn, Princeton, Stanford and Chicago. Seven report nothing. MIT is the most explicit, entering “N/Av” in all nine bands and again in the average-GPA box.
| Ivy-Plus institution | Reports C11 for 2024-25 |
| Harvard University | Yes |
| Princeton University | Yes |
| Stanford University | Yes |
| University of Pennsylvania | Yes |
| University of Chicago | Yes |
| Brown University | No |
| Columbia University | No |
| Cornell University | No |
| Dartmouth College | No |
| Duke University | No |
| Massachusetts Institute of Technology | No, filed as N/Av |
| Yale University | No |
Common Data Set item C11, 2024-25 filings, collected from the College Transitions Common Data Set Repository.
The pattern extends past the Ivy-Plus. Williams, Amherst, Bowdoin, Carleton, Claremont McKenna, Grinnell and Hamilton all file C11 blank in both 2019-20 and 2024-25. Caltech files it blank. So do Georgetown and Boston College.
Whatever the reason, and there are several plausible ones, the result for an applicant is the same: there is no published GPA distribution to interpret. Counselors should stop describing a 3.9 as “below the middle” at a school that does not publish a middle.
Fifteen of the Twenty Highest 4.0 Shares Come With a C12 Average Above 4.0
Among the 88 institutions that do answer, the reported 4.0 share ranges from 2.8 percent to 97 percent. That spread is not a ranking of student quality.
Fifteen of the 88 report a C12 average above 4.0, and two report on a scale that is not a 4.0 scale at all. Those fifteen report a median 4.0 share of 67.9 percent. The 71 institutions whose C12 average is 4.0 or below report a median of 27.3 percent. The first group’s typical figure is two and a half times the second’s.
Figure 1. Share of enrolled first-year students reported at 4.0, 2024-25, the 20 highest of 88 reporting institutions. Colors report what each filing’s own C12 average says, which is a warning about comparability rather than a determination of how the institution built its C11 distribution. Source: Common Data Set item C11, institutions’ own 2024-25 filings, College Transitions Common Data Set Repository. Analysis by College Transitions.
Whatever is driving it, a table that ranks colleges by “percentage of students with a 4.0” cannot be assumed to rank directly comparable quantities across schools. Fifteen of the top twenty carry a C12 average their own filing shows cannot be an ordinary unweighted mean, and none of the twenty documents how its C11 distribution was built.
Below 3.25, the Class Thins Dramatically
Set the top of the table aside and look at the 71 institutions whose C12 average is 4.0 or below. This is a consistency screen rather than a guarantee. It removes the most obvious warning sign, a C12 average above 4.0, without establishing that the institutions left behind calculate C11 identically.
The median institution in that group enrolls 59.6 percent of its class at 3.75 or above and 7.0 percent below 3.25. At 22 of the 71, students arriving under a 3.25 make up less than 2 percent of the class.
| Institution | At 3.75 or above | Below 3.25 | Admit rate |
| University of Florida | 94.0% | 0.0%* | 24.2% |
| Princeton University | 94.0% | 0.4% | 4.6% |
| University of Michigan | 92.8% | 0.4% | 32.0% |
| Johns Hopkins University | 92.4% | 1.2% | 6.4% |
| University of Pennsylvania | 90.0% | 0.0% | 5.4% |
| Stanford University | 89.8% | 0.6% | 3.6% |
| University of California, Berkeley | 89.2% | 0.7% | 11.0% |
| Vanderbilt University | 89.0% | 1.6% | 5.9% |
| University of Wisconsin-Madison | 86.1% | 1.0% | 45.2% |
Enrolled first-year GPA distribution, 2024-25, for institutions reporting a C12 average of 4.0 or below. * Florida leaves its six lower bands blank rather than entering zero; the three bands it does report sum to 100, so the 0.0 percent shown is our reading of that, not a figure Florida published. Source: Common Data Set items C11 and C1, College Transitions Common Data Set Repository. Analysis by College Transitions.
Wisconsin admits 45.2 percent of applicants and 1 percent of its entering class arrived below a 3.25. Selectivity alone clearly does not explain these distributions, which the next section takes up directly.
After a Conservative Screen, the 4.0 Share Still Rose at Most Institutions
Forty-seven institutions report a C12 average of 4.0 or below in both 2019-20 and 2024-25. Screening on both years removes one obvious source of incompatibility. It cannot prove that an institution’s GPA calculation stayed the same across five years, and two cases below show why that caution is necessary.
Within those 47, the 4.0 share rose at 37, fell at 8, and was unchanged at 2. The median change is +2.6 percentage points. The movement is concentrated rather than uniform. Boston University went from 19.0 percent of its entering class at 4.0 to 41.0 percent and Purdue from 11.8 to 33.3 percent, the two largest increases in the screened group. Spelman rose from 22.7 to 41.0 percent, Princeton from 58.2 to 68.5, and Wisconsin crossed the halfway line from 43.1 to 52.7.
Figure 2. Share of enrolled first-years at 4.0, 2019-20 (open circle) to 2024-25 (filled circle), for a selected 22 of the 47 institutions passing the C12 screen, being the largest movers in each direction. Source: Common Data Set item C11, 2019-20 and 2024-25 filings, College Transitions Common Data Set Repository. Analysis by College Transitions.
The two largest declines are the most instructive rows in the whole dataset. Vanderbilt reports 76.5 percent of its 2019-20 class at 4.0 and 35.9 percent in 2024-25, with its C12 average barely moving, from 3.84 to 3.89. Houston reports 38.8 percent and then 2.8 percent, with its average falling from 3.73 to 3.49. Both passed the screen. A forty-point swing in the top band alongside a nearly unchanged average is difficult to explain as a straightforward shift in the entering class alone.
Their filings do not document how either distribution was constructed in either year, so the cause cannot be established from this data. That is the honest limit of this comparison. The top-band percentage and the average GPA can move in opposite directions, and without documentation of an institution’s calculation method there is no way to say why. Read the direction of travel across 47 schools, not any single school’s five-year line.
Admit Rate Has Only a Weak Relationship With Reported 4.0 Share
Families reach for GPA distributions as a proxy for how hard a college is to get into. Across the 66 institutions passing the C12 screen and disclosing enough of item C1 to compute an admit rate, that use does not hold up.
The correlation between admit rate and reported 4.0 share is -0.32. In this sample, a simple linear relationship with admit rate accounts for about 10 percent of the observed variation in reported 4.0 share.
Figure 3. Admit rate against share of enrolled first-years at 4.0, institutions whose C12 average is 4.0 or below, 2024-25. Source: Common Data Set items C11 and C1, institutions’ own 2024-25 filings, College Transitions Common Data Set Repository. Analysis by College Transitions.
The University of Alabama admits 76.7 percent of applicants and reports 42.0 percent of its entering class at 4.0. Davidson admits 13.4 percent and reports 40.6 percent. Alabama at Birmingham admits 88.2 percent and reports 41.2 percent. Three colleges whose admit rates span 75 percentage points enroll classes with nearly identical reported 4.0 shares.
The distributions underneath those three figures are not identical. Davidson reports 73.4 percent of its class at 3.75 or above against Alabama’s 61.0 and Birmingham’s 58.2. The top band converges while the rest of the distribution does not, which is a reason to read the whole table rather than the headline number. Factors such as honors programs and merit-aid recruitment may help explain why some less selective institutions enroll large numbers of 4.0 students, though nothing in the Common Data Set establishes that.
One obvious objection is worth answering before it is raised. If a college reports GPA for only a fraction of its class, its distribution may not represent the class at all. Restricting the analysis to institutions that collected a GPA from at least 90 percent of their enrolled first-years leaves 47 of the 71, and the central findings hold: the median share at 3.75 or above moves from 59.6 to 64.0 percent, the median below 3.25 from 7.0 to 6.7 percent, and the longitudinal finding holds too: among the 30 institutions in that five-year group clearing the same 90 percent threshold in both years, the 4.0 share rose at 28 rather than the 37 of 47 reported above. The admit-rate relationship stays weak at every coverage threshold we tried, measuring -0.20 among institutions reporting GPA for at least 95 percent of the class. Nothing in the piece depends on the filings with thin GPA coverage.
What These Numbers Cannot Tell You
Item C11 describes enrolled students, not admitted ones. It is the profile of the class that showed up after yield, not the bar an applicant had to clear. Nothing in the Common Data Set distinguishes the two.
The bands are coarse. Between 3.75 and 3.99 is a single bucket, so a 3.76 and a 3.99 are indistinguishable. Colleges report only for students from whom they collected a GPA. Among the filings that state that share, it runs from 100 percent at many institutions down to 16.4 percent at Villanova, and a number of filings do not state it at all. A distribution built on one-sixth of an entering class is answering a materially different question from one covering essentially everyone.
The 2024-25 form splits C11 into three columns covering test submitters, non-submitters, and all enrolled students. We used the all-enrolled column throughout, and excluded seven institutions whose filings offer no all-enrolled figure to use. Clemson and West Point filled only the test-submitter column. Florida State replaced the three columns with its own pair headed “Unweighted %” and “Weighted %”, reporting 12.4 percent at 4.0 on the first and 87.6 percent on the second. Ranking any of these against all-enrolled figures would compare different populations.
One caution about sourcing, and it concerns our own collection. Three links in the College Transitions repository resolve to another institution’s filing and were excluded. Two are duplicated links: the cell for Stanford’s 2019-20 filing carries the same file as St. Olaf’s, and the cell for Worcester Polytechnic’s 2024-25 filing carries Whitman’s. A fourth link, for Princeton’s 2024-25 filing, returned an error and the filing was retrieved directly instead. Anyone working from these files should confirm the institution named inside the document before using a number from it.
How to Use This in a Family Meeting
Read the average GPA line before quoting any college’s 4.0 share. If that average is above 4.0, something in that filing is not on a plain unweighted scale. That is a warning not to compare its C11 figure mechanically with another institution’s until you know how each distribution was constructed, which the filings do not say. The check takes ten seconds.
For the institutions that publish nothing, say so plainly rather than substituting a number from a peer. A family is better served by “Williams does not disclose this” than by an inference drawn from Amherst, which also does not disclose it.
And retire the idea that a high 4.0 share, by itself, signals a hard admit. Alabama and Davidson enroll classes with nearly the same reported share of 4.0 students and admit rates that differ by a factor of nearly six. What a rising 4.0 share does tell a family is that at many institutions a perfect GPA is now common among the students who enroll. It does not tell us how admissions officers weighed that GPA against course rigor, testing, essays, recommendations, or activities.
Sources and method. All figures come from institutions’ own Common Data Set filings for 2024-25 and 2019-20, collected from the College Transitions Common Data Set Repository. We matched 188 institutions in the repository across the Ivy-Plus group, top-50 national universities, state flagships and top-25 liberal arts colleges, and collected 345 filings covering 186 of them, and extracted items C11 (GPA distribution of enrolled first-year students), C12 (average high school GPA and share submitting) and C1 (applicants and admits). A filing enters the cross-section only if it reports C11 for all enrolled students or in a single undifferentiated column, its bands sum to between 98 and 102, it reports a C12 average so its comparability can be assessed, and the document names the institution it is filed under. That leaves 88 institutions for 2024-25 and 92 for 2019-20, out of the 188 matched. The full 2024-25 disposition of all 188 is itemised in the accompanying dataset. Findings were re-run restricting the sample by the share of enrolled students for whom each college reports collecting a GPA, at thresholds of 80, 90 and 95 percent; the distribution and longitudinal findings are materially unchanged and the admit-rate correlation weakens rather than strengthens. Institutions reporting only the test-submitter column are excluded rather than ranked against all-enrolled figures. C12 is used to describe what each filing says about its own average, never to classify an institution’s C11 methodology as weighted or unweighted, because the CDS instructs C11 on a 4.0 scale and does not require the two items to share a calculation. Admit rates are computed from C1. Gender rows are summed and checked against any total row in the filing; where the two disagreed, the gender-row sum was used, because it can be checked line by line, and the disagreement was recorded. Several CDS filings print residency breakdown rows in the same shape as a total row, so a recorded disagreement is not always a defect in the filing. Excluded and logged with reasons: three repository links resolving to another institution’s filing, filings whose bands do not sum to approximately 100, and filings reporting no C12 average. Princeton’s 2024-25 filing was obtained directly after its repository link returned an error. No value was inferred, interpolated or filled. Item C11 covers enrolled students, not admitted students. Analysis by College Transitions, September 2026.
