This investigation was completed at the Oregon Research Institute and supported by NIMH grant MH-0892-01 and NSF grant GB-257. Computations were performed at the Western Data Processing Center and the Health Sciences Computing Facility, University of California at Los Angeles.

For readers who may not have encountered the domain formulation of reliability, the following remarks and examples may help to clarify the issues involved. Reliability is seen as indicating the degree to which one may generalize from the observed variable to some domain which the variable is supposed to represent. The definition of the domain is quite arbitrary, being subject entirely to the purposes of the investigation. Thus, an observed variable has as many different “reliabilities” as there are domains to which the variable may be considered to belong. For example, a domain might consist of all possible administrations of a particular psychometric test. Then the reliability of one administation should indicate the degree to which it is representative of the results which would be obtained from the other administrations. With respect to the grade given a student's performance on a nonobjective achievement test, one might ask how representative it is of the grades which would be given by other graders from the same department, or by graders from all universities and colleges, or how representative it is of grades on all other performances by the subject on similar tests.