Step 2 Score Predictor

The short version

• A 20-point swing is normal. The median gap between a student's best and worst practice test is 24 points. 95% of students swing 10+ points.
• Most of the swing is the measurement, not you. Practice forms differ in harshness by about 11 points, and a 200-question test is noisy on its own.
• The real exam beat the worst practice test 99% of the time and landed within about a point of the best, at the median.
• A bad test near exam day predicts almost nothing. A 10-point late drop moved the expected real score by about 1 point.

How much swing is normal

Each student's gap between their best and worst practice test, n = 2,176 students with 3+ score-scale tests.

The middle half of students swing between 17 and 31 points. 83% swing 15 or more, 65% swing 20 or more. If your scores bounce around, you are not the exception; the student with four nearly identical scores is.

Half your swing is the test, not you

The forms are not equally harsh. Holding timing fixed (tests taken 7–30 days out), here is how far below the real exam each form typically lands:

Take UWSA 2 and UWSA 3 in the same week and the 11-point "drop" is built into the forms, with zero change in your knowledge. Adjusting every score for its form cuts the typical student's score variability from 8.0 to 6.9, and most of what remains is ordinary test noise: a 200-question sample of a giant subject wobbles a few points on its own (the USMLE itself reports a standard error of about 7 points on the real exam). Form difficulty plus sampling noise is most of your swing. The conversion pages quantify each form separately.

Does inconsistency hurt your score?

No. Holding the practice-test average fixed, the most erratic third of students finished slightly higher than the steadiest third:

The reason is mundane: a big swing usually contains an early low score you have since outgrown, so an erratic record is often just an improvement curve wearing a scary costume. Erratic profiles are not harder to predict either; prediction error is flat across swing sizes in this data.

Which test is the real you?

Among the 2,061 students whose tests spanned 10+ points, the real exam landed a median of +1 points from their best test, +13 from their average, and +26 from their worst. It beat the worst test 99% of the time and beat the best 55% of the time.

So the worst test is not a prophecy; it is the score most contaminated by a harsh form, a bad day, or an early date. The best test is usually the most recent one on a fairly-curved form, which is exactly why the real exam tracks it. Treat the best score as roughly where you stand, not as a lucky outlier, and treat the average plus the usual offset as the honest expectation.

The bad test a week before your exam

The classic panic: scores were fine, then the last NBME before test day comes in low. Three facts from 1,485 students whose final practice test fell within 3 weeks of their exam:

• It is rare. Only 4% saw their final test land 5+ points below their earlier average. Most people peak at the end.
• It did not matter. The 26 students whose final test dropped 8+ points still finished a median of +9 points above their earlier average, the same as students whose final test was steady (+10). They beat the scary last test itself by +18 points. Small group, so treat it as illustrative.
• Across everyone, the final test's deviation from your average barely predicts. A 10-point late drop moved the expected real score by about 1 point, not 10.

One bad form is noise wearing a deadline. If a reschedule decision has to be made, make it on the level of your whole average, never on the wiggle of the last data point.

Can you grind your way out of a flat trajectory?

First, trajectory is real signal, not another illusion. To measure it without circularity, take each student with 4+ dated tests, fit a slope through the first half of their tests only, then watch how they finished relative to what that early level alone would predict:

A climbing line keeps climbing past test day; a flat line tends to stay put. So the question matters: when the line is flat, can you force it upward? The instinctive answers are more questions, more weeks, or a new question bank. Each one has a number, and none of the numbers cooperate:

• More questions: no. Inside every trajectory group, the most-questions third finished no better than the least-questions third; the best any group managed was -1.5 points, noise for samples this size. Volume that has already failed to move the line does not start working because there is more of it.
• More weeks: no, and the sign is ugly. Flat-trajectory students with the longest dedicated periods finished about 7 points worse against expectations than flat-trajectory students with the shortest (n = 59 per group). Read that carefully: it is mostly selection, since the students who extend are the ones struggling hardest, not proof that extra time hurts. But it does mean the data contains no trace of extra weeks rescuing a flat line.
• Switching banks: no. The same percent correct points to the same real score on either major bank, so the switch buys a new interface, not new points. The head-to-head data.

One move did show a measurable payoff for students who were not already climbing: a second pass of weak material, worth about +0.9 points across flat-to-moderate trajectories (83 repeaters), on top of the larger effect already documented for anyone under 75% on their first pass. For students already climbing steeply it added nothing, which is the tell for what a flat line actually is.

Here is the way to think about it. Your practice scores are the output gauge of your studying: when the gauge climbs, your study loop is converting hours into points; when it sits flat while the hours keep going in, the loop is broken somewhere between doing questions and retaining what they teach. Pouring more hours into a broken loop shows up in this data as exactly nothing. What fixes a flat line is changing what happens after each question block: an honest accounting of why each miss was missed, a second pass aimed at the weakest systems rather than fresh questions everywhere, and treating each practice exam as a diagnosis to act on rather than a verdict to mourn. The students who did the repeat-and-repair version of volume gained; the students who did the more-of-the-same version did not.

A flat line is a method problem, not an hours problem: change how you review before you change your test date. More on what does and does not move scores: the full evidence review.

Common questions

Is a 15-point swing between practice tests normal?

Yes. Among 2,176 students with three or more practice tests, the median gap between best and worst was 24 points, and 83% had a gap of 15 or more. A swing is the normal experience, not a warning sign.

My NBME dropped right before my exam. Should I reschedule?

One bad test is weak evidence. Only 4% of students saw their final practice test land 5+ points below their earlier average, and students with big late drops went on to score about the same relative to their earlier tests as everyone else. Across all students, a 10-point final-test drop predicted about 1 point on the real exam. Decide based on the level of your average, not on one swing.

Should I trust my best or my worst practice test?

Neither extreme is a prophecy, but the real exam runs much closer to the best: among high-swing students it beat the worst practice test 99% of the time and landed within about a point of the best at the median. The honest summary is your average plus the usual offset, which is what the full predictor computes.

My scores are flat. Should I extend my dedicated period?

The data argues against it. Students with a flat early trajectory who studied the longest finished about 7 points worse against expectations than flat-trajectory students with short dedicated periods, partly because struggling students are the ones who extend. What showed a measurable benefit for stalled students was a second pass of weak material, worth about +0.9 points. Change the method, not the calendar.

Get a real prediction, not a rule of thumb

The full predictor combines all of your practice tests with their timing, shows calibrated probability ranges, and was the most accurate of every predictor tested against real scores. See the head-to-head comparison.

Predict my Step 2 score →

More charts & guides