Behaviorism admits a variety of theoretical concepts into the science of behavior. These concepts differ in how closely they are linked to the behavioral data language. Transformations within the behavioral data language are securely linked and pose no problem. State, dispositional, and operationally defined concepts, however, provide only partial definitions since they are indeterminate and, in the case of the first two, open as well. In practice, even operationally defined concepts are also open.
Intervening variables tie together a set of interrelated independent variables with a set of interrelated dependent variables. Behaviorists differ as to whether the benefits of intervening variables outweigh their dangers. This debate centers on the tactical issue of whether concepts facilitate or inhibit research and explanation rather than on the philosophical respectability of intervening variables.
A similar debate arises over the admission of hypothetical constructs, theoretical concepts which explicitly refer to unobserved entities. Proponents argue that hypothetical constructs are necessary if the science of behavior is to be reducible, explanatory, and fertile. Opponents see these constructs as diluting the objective and empirical nature of the science. Although none of these arguments is conclusive and many depend on individual intuitions about how research is best performed, most behaviorists do incorporate hypothetical constructs in their theories. Nevertheless, they implicitly impose restrictions on the postulated features of these constructs so that the dangers are reduced.
For the reasons discussed in the preceding chapter, the boundaries of the behavioral data language are not sharply defined. Yet, wherever they are drawn, the study of behavior must exceed them to establish a science. It must transcend the immediate momentary observation to impose or discover coherence in its subject matter. Therefore, concepts must be introduced which do not qualify as terms in the behavioral data language.
This requirement, however, raises serious problems for the behaviorist program. By tying the behavioral data language to observation and intersub