In behavior research using cognitive and affective measures, there is often incongruity between the statistical analysis employed and the intended inference. This paper argues that incorporating items as levels of a random facet via generalizability theory allows the statistical examination of the inferential question in the desired universe of inference. (BS)