Comments on Cone's “The relevance of reliability and validity for behavioral assessment”

The thesis of Cone's paper is that generalizability theory, an alternative to classic reliability and validity theory, may provide a means for evaluating the adequacy of behavioral assessment techniques. Specifically, Cone suggests that these devices be evaluated with respect to scorer, item, method, setting, time, and dimension generalizability. He proposes that all the universes of generalizability "that are relevant to traditional assessment apply to behavioral assessment as well." The contention of the present commentary is that, although generalizability studies may have some utility in the evaluation of behavioral assessment devices, generalizability across all or even most suggested universes is not a necessary characteristic of precise and useful behavioral assessment techniques. Assessment tools which produce generalized responding may not always be useful within a behavioral assessment framework. A major goal of behavioral assessment is to specify the functional relationship of behaviors to the environmental variables affecting their occurrences. The purpose of behavioral assessment is to ascertain the differential responsiveness of the subject to a variety of stimulus conditions. These controlling variables are the hallmark of a functional analysis of behavior, the usual precursor to behavioral intervention programs. Thus, only by determining this differential responsiveness will the variables controlling behavior be discovered. For example, although behavioral researchers have computed the item generality or the internal consistency of self-report inventories such as the Sexual Interaction Inventory (LoPiccolo & Steger, 1974) and the Fear Survey Schedule (Dickson, 1975) and have reported fairly high correlations, there is no reason why a behavioral clinician would predict or desire the client to respond similarly to each test item. Similarly, in the assessment of assertive behavior, generalized assertive responding to a variety of role playing situations would not be particularly useful in