Accuracy
How accurate are AdmitBase's odds?
We test them against published national data, show where they are wrong, and fix what the tests find. This page is the record.
The short version
- Published figures are shown as published, with their source and year: school medians, acceptance rates, the AAMC's rate for your MCAT and GPA, LSAC's national admit rate.
- Modelled chances are models. No source publishes the chance that a particular applicant is admitted to a particular school, so nobody can verify one, including us. What we can test is whether our list-level numbers match national totals, and that is what is below.
- The tests found real errors. On 2026-11-05 they showed that onboarding told medical and law applicants a 100% chance of at least one acceptance. That was wrong, and it has been fixed.
Medicine: tested against the AAMC's grid
The AAMC's Table A-23 reports, for each MCAT and GPA band, the share of applicants accepted to at least one US MD school (2023-24 to 2025-26 combined). For each of 45 cells with 100+ applicants (137,318 applicants), we score that applicant against 159 US MD schools, draw 200 random lists of 19 (the average applicant sends 19.5), and combine the schools. Weighted by applicants, by MCAT band:
| MCAT | AAMC reports | AdmitBase model | Model minus AAMC |
|---|---|---|---|
| 494–497 | 11.2% | 35.2% | +24.0 points |
| 498–501 | 22.3% | 39.8% | +17.5 points |
| 502–505 | 35.5% | 46.1% | +10.6 points |
| 506–509 | 47.2% | 52.4% | +5.2 points |
| 510–513 | 62.6% | 58.2% | −4.4 points |
| 514–517 | 71.3% | 62.8% | −8.5 points |
| Above 517 | 80.1% | 66.4% | −13.7 points |
Source: AAMC FACTS Table A-23; model run 2026-11-05. Overall: typical error 12.3 points, average bias +2.5.
- It runs high for lower scores and low for higher ones. A model this flat cannot see that weaker applicants build different lists or that non-score screens bite harder at the bottom.
- That is why the AAMC's number comes first. Wherever a medical applicant enters an MCAT and GPA inside the grid, we show the AAMC's own rate for that band, labelled Published, ahead of any model.
- Before the fix it was far worse. Multiplying school chances as if independent gave a 510 to 513 MCAT, 3.60 to 3.79 GPA applicant 85.6% where the AAMC reports 60.0%. A shared applicant factor (correlation 0.5) brings it to about 57%.
Law: reconciled with LSAC's national totals
LSAC publishes no admit rate by LSAT or GPA, so there is no grid to test bands against. It does publish, for every year, applicants, applicants admitted to at least one ABA-approved school, and applications. The ABA 509 disclosures give the share of applications that ended in an offer. If decisions were independent, those would predict how many applicants are admitted somewhere. They do not:
| Year | Applicants | Applications each | Offer rate per application | Independence predicts admitted somewhere | LSAC reports | Correlation that fits |
|---|---|---|---|---|---|---|
| 2021 | 70,893 | 6.76 | 28.3% | 89.4% | 67.8% | 0.48 |
| 2022 | 62,507 | 6.86 | 30.1% | 91.4% | 69.7% | 0.49 |
| 2023 | 61,319 | 6.78 | 31.6% | 92.3% | 69.9% | 0.52 |
| 2024 | 64,779 | 6.60 | 32.0% | 92.0% | 69.4% | 0.53 |
| 2025 | 76,480 | 6.84 | 26.5% | 87.8% | 65.1% | 0.49 |
Sources: LSAC Admission Trends (updated December 2025); offer rate from AdmitBase's compilation of ABA 509 disclosures, whose application count is within 2% of LSAC's each year.
- Independence overstated by 23 points in 2025. The correlation that closes the gap is 0.48 to 0.53 in every year, the same 0.5 that was fitted to the AAMC's medical grid from an unrelated source.
- The limit. This is one national point per year, with applicants treated as alike. Without a grid we cannot say whether law runs high or low by score. We show LSAC's national rate as context, labelled Published, and the list figures as Model.
Dentistry: one check, and it runs high
ADEA publishes applicants, applications by school, enrollees and the applicants' mean GPA and DAT, but not the share admitted to at least one school, and our data carries an acceptance rate for too few dental schools to run the law reconciliation. So the correlation between schools is assumed (0.5, borrowed from medicine and law, where it was measured) and the test is a single point: score the average 2025 applicant on a typical list.
| Applicants, applications each | 13,657; 9.9 |
| Mean applicant | GPA 3.53, DAT 19.6 |
| Enrolled somewhere (ADEA) | 51.8% |
| AdmitBase, typical 10-school list | 72% |
| Before the fix | 100% (every school, independent) |
Source: ADEA, U.S. Dental School Applicants and Enrollees, 2025 Entering Class (March 2026), Tables 1, 5 and 9; model run 2026-11-05. Enrolled understates admitted: in law about 86% of admitted applicants enrolled, so the admitted share is roughly 52% to 60%.
So dental runs about 12 to 20 points high for the average applicant. That is far better than 100%, and still too high, which is why the dental list figure is worded as an upper estimate.
What the tests found, and what we changed
- 2026-11-05, medicine. A 512 MCAT, 3.72 GPA applicant was shown 100% for at least one acceptance across a spread of 19 schools; onboarding said 100% across 159 schools with 34 expected acceptances. The AAMC's figure for that band is 60.0%. Fixed: a medicine-specific curve, a measured correlation between schools, a headline for a typical 20-school list, and the AAMC's own rate shown first.
- 2026-11-05, dentistry. Onboarding told the average applicant 100% across every school (9.7 expected acceptances). Fixed: a typical 10-school list (ADEA: 9.9 applications per applicant) and an assumed correlation, giving 72% where about 52% enrolled. Not yet right, and labelled as such.
- 2026-11-05, law. Onboarding told every law applicant 100% across about 198 schools, with 15 to 116 expected acceptances. Fixed: the same correlation and a typical 7-school list (LSAC: 6.6 to 6.9 applications per applicant). For an LSAT 158, GPA 3.5 applicant that is now 86% with 1.8 expected acceptances, against a national average of 1.79 offers per applicant.
What we cannot check
- Any single school's probability, for any program. It is not published, so it is a comparison tool and labelled Model.
- Residency. The models use whole-class figures. A resident is understated at a state school and a non-resident overstated. The AAMC's residency split is shown on each medical school page, with no probability attached.
- Veterinary. The typical list size (5 applications) comes from AAVMC figures quoted in the trade press for 2014 to 2022, not a current primary source, and the correlation is assumed as in dentistry. Lowest confidence of the programs that show a list figure.
- Pharmacy, optometry and MBA. We found no national total we could verify, so these programs show no single “chance of at least one acceptance”, only where you stand school by school. A list you build yourself in the Cycle Report still multiplies school chances and will overstate.
- Medians we estimate. Where a school publishes only a median, we estimate the spread and mark the figure Estimate.
How to read a number on AdmitBase
- Published
- An official source released the figure and we show the source and year: a school's class profile, an ABA 509 disclosure, the AAMC, LSAC. We do not alter it.
- Estimate
- AdmitBase estimated a figure the school does not publish, such as the spread around a median. It is marked wherever it appears.
- Model
- A modelled chance. It orders schools and compares lists. It is not a rate any source reports, and the tests above say where it errs.
For advisors: citing this page
Quote national rates to students, not modelled chances: for medicine the AAMC's rate for their MCAT and GPA band (the calculator shows it first), for law LSAC's national admit rate. Tell them the bands are a measure of fit. Suggested citation:
AdmitBase. “How accurate are AdmitBase's odds?” admitbase.com/accuracy. Underlying sources: AAMC FACTS Table A-23 (2023-24 to 2025-26); LSAC Admission Trends (December 2025); ADEA, U.S. Dental School Applicants and Enrollees, 2025 Entering Class.
If a number looks wrong, email support@admitbase.com. We check it against the source, fix it, and add it to the log above.
Questions
How accurate are AdmitBase's admission odds?+
For medicine we compare the model's chance of at least one acceptance with the AAMC's national grid across 45 MCAT and GPA cells. The typical error is about 12 points, and it is systematic: the model runs high for MCAT scores under 510 (by 24 points at 494 to 497) and low above 513 (by 14 points above 517). For law we reconcile the model with LSAC's national totals, where multiplying school chances predicted 87.8% of 2025 applicants admitted somewhere and LSAC reported 65.1%. Per-school probabilities cannot be verified by anyone, because no source publishes them.
Does AdmitBase publish its errors?+
Yes. This page shows the medical backtest by MCAT band, the law reconciliation for 2021 to 2025, what the tests found, what we changed because of them, and what we cannot check.
What do the Published, Estimate and Model labels mean?+
Published means an official source released the figure and we show its source. Estimate means AdmitBase estimated a figure the school does not publish, and says so. Model means a modelled chance that no source publishes, so it cannot be verified directly; it is tested against national totals instead.
Why don't you show a probability for each school?+
We do show a modelled chance for each school, labelled Model, but no source publishes per-school admission probabilities, so it is a comparison tool, not a rate. For medicine the AAMC's own national rate for the applicant's MCAT and GPA band is shown first because it is published data.
See it applied: medical school chances, law school chances, and the math behind a balanced school list. Found something wrong? support@admitbase.com.