The Truth About Confidence Intervals in Score Predictors
Why every honest Step score predictor returns a range, not a single number. What a 95% confidence interval actually means for your test day prediction.
Open any Step 2 CK score predictor and you'll see two very different presentations. Some return a single number ('Your predicted score: 247'). Others return a range ('Your predicted score: 240-254, 95% CI'). The range is honest. The single number is marketing.
What a 95% confidence interval actually means
A 95% CI of 240-254 means: if you took the exam many times under identical conditions, 95% of your scores would fall inside that window. It does NOT mean you have a 95% chance of scoring exactly the midpoint. The window encodes both random variance from a single practice form and structural uncertainty in the conversion model.
Why a narrow CI is suspicious
If a predictor returns a CI tighter than ±4 points from a single NBME, it's almost certainly under-reporting uncertainty. The published NBME-to-Step correlation (~0.85) implies a 95% CI of at least ±10 points from one form. Narrower than that requires multiple forms or a much higher correlation than NBME itself reports.
More practice exams shrinks the CI. Two NBMEs → ±7 points. Three → ±5.5 points. After 4-5 forms, the floor is ±5 because real test-day variance dominates.
Source: signal averaging of r ≈ 0.85 per-form correlation. Applies to NBMEs; UWSA forms are wider due to structural over-prediction bias.
How to interpret YOUR CI
- Lower bound: realistic worst case if test day goes poorly
- Midpoint: best single-number summary of your current level
- Upper bound: realistic best case if test day goes well
- Width: how much uncertainty remains in the prediction
When the CI matters most: borderline pass scores
If your midpoint is 218 with a CI of 208-228, you're firmly above the 209 pass threshold midpoint but the lower bound straddles it. That's a real risk signal worth taking seriously. A predictor that reported only 218 would obscure that risk.
Our calculator always returns the 95% CI alongside the midpoint, and shrinks the window as you add more practice exams.
Ready to predict your Step score?
Free, no signup. Multi-source aggregation, 95% confidence interval, and a personalized study plan.
Run the calculatorRelated articles
NBME 31 Curve: Easier or Harder Than NBME 30?
A direct comparison of the NBME 31 score curve vs NBME 30. Question style differences, predictive accuracy, and when to take which form.
Average Step 2 CK Score: What's a 'Good' Score in 2026?
The 2026 average Step 2 CK score, what counts as a competitive score for each specialty, and how to benchmark your practice scores against residency targets.
How Accurate Are NBME Score Predictions? An Honest Breakdown
What the published correlation between NBME self-assessments and real Step 2 CK actually says, where score predictors add value, and where they break down. A no-marketing look at predictive accuracy.