The empirical method to index development is typically used when the measurement has a practical purpose, for example to predict which patients are most likely to be discharged after rehabilitation is complete. After testing a large number of questions, statistical procedures based on correlation methods are used to select those that best predict the eventual outcome. These “item analysis” statistics are described in the next section. Empirical selection of items has a practical appeal. It does, however, suffer the weakness that the user cannot necessarily interpret why those who answer a certain question in a certain manner tend to have better outcomes: the questions were not selected in relation to any particular theory of rehabilitation. Many illustrations exist; the questions in the Health Opinion Survey were selected because they distinguished between mentally ill patients and unafflicted people, and although they succeed in doing this, debates over what exactly they measure have continued for 35 years. A more recent example is Leavitt’s Back Pain Classification Scale, which was developed empirically to distinguish between pain of an organic origin and pain related to emotional disorders. It succeeds well, but Leavitt himself commented: “Why this particular set of questions works as discriminators and others do not is unclear from research to date.” Accordingly, the back pain scale may have clinical value, but it does not advance our understanding of the phenomenon of pain as a response to emotional disorders and of how psychological factors may modify the pain response.