In the first part of this book I discussed general considerations and procedures for analyzing test scores. These analyses treated the test score as a single observation, or variable, and as a given. However, there are two reasons why we need to understand the nature of these scores before we use them to make decisions about people: (1) to ensure that the test score distribution is appropriate for the intended uses of the test, and (2) to ensure that the test scores are as reliable as needed for the intended uses of the test.

When we design a test for a particular purpose, we generally have specific expectations regarding the way the scores will be distributed. For example, for a norm-referenced (NR) test, we generally want the scores to be widely spread, with a large standard deviation, while with a criterion-referenced (CR) test, we may expect a narrower distribution, with a relatively small standard deviation. We also try to design our tests to yield scores that are appropriate for the ability level of the test takers. That is, we do not want our tests to be either too difficult or too easy for the groups we are going to test. Thus, in designing a test, we would like to be able to control the distribution of the scores we obtain so that these will be useful for the intended purpose of the test. We can control the distributional characteristics of the test scores, to a large extent, by controlling the statistical characteristics of the individual tasks or items that we include in the test.