On the other hand, let us make the highly reasonable assumption that the trait is normally distributed such that the majority of the examinee population is between plus and minus 1 standard deviation from the mean. In this situation it may be reasonable to try to maximize the information in the middle of the trait range where most people are, and to deemphasize measurement precision in the extremes. This would be accomplished by identifying highly discriminating items that had difficulty values in the middle of the trait range. Items with large positive and negative difficulties would be eliminated from the final version of the test. In this situation, the test created by IRT methods would be highly similar to the test created by traditional item analysis.