Thanks for the comments and for checking out this website!
The estimator is only as accurate as the submitted anonymous data that is used to generate the regression models. Like any regression model, the estimates become less accurate the further away the independent variable moves from the data mean. In other words, the predictions are less accurate at the extremes than at more average values.
It should be noted, however, that the NBME exams use the same general mathematical algorithm as used on this website and are therefore subject to the same inaccuracies at extreme values. The dataset that the NBME exams are based on are not necessarily larger, either. I've included a link to the academic publication that the NBME exam predictions are based on to give you a better idea of the NBME dataset.
Overall, I think the NBME exams are a great resource and I believe people should place a strong emphasis on the results. Unfortunately, I also believe there is a high degree of variability inherent to the USMLE that makes high accuracy on these predictive models a virtually impossible feat.
Please remember the intention of this website is not to compete with the NBME exams, but to supplement them with score estimates from other question sources. At all times, we feel that people should be careful in their use of any of these score estimates (including the NBME).
Sawhill et al., "Using the NBME Self-Assessments to Project Performance on USMLE Step 1 and Step 2: Impact of Test Administration Conditions", Academic Medicine, 2004:
I think this predictor seems VERY optimistic. When you plug in the number 320 from the NBME test (which is CLEARLY FAILING), The estimator says you should get around 185 in STEP 1.
According to the NBME a grade of 320 gives you a STEP 1 score around 169
NBME exams are supposed to be the best predictor today.