In selecting a domain for psychology, behaviorism distinguishes between behavior and physiology. A priori attempts to draw this distinction having failed, an empirical or theoretical definition of behavior emerging from a science of behavior is probably the most useful. The distinction is important for molar behaviorism which argues, mostly on tactical grounds, for an autonomous science of behavior independent of physiology.
For a science of behavior to develop, observation must generate descriptive reports, couched in the categories of language and perception. Some epistemologists conclude that all observation is dictated by theory. Nevertheless, it seems that the degree to which an observation is contaminated by theory varies, depending on the observation. On this continuum between the purely observational and the purely inferential, intersubjective agreement can be used as a convenient measure, with objectivity increasing with increasing intersubjective agreement.
In constructing a behavioral data language, some behaviorists insist that only physical descriptions be used. However, this insistence is not only impractical; it is also based on false intuitions about the ontological superiority of physical properties. Psychology must be free to use properties it finds effective as long as they are intersubjectively identifiable.
Most behaviorists prefer to exclude from the behavioral data language any descriptions expressed in: action language, the intensional mode, purposive terms, or molar categories. In practice, this preference is often violated. The reasons both for the preferences and for the violations are varied, and they further clarify behaviorist notions of empiricalness and objectivity.
Functional definitions of stimulus and response are derived from empirical findings. They often specify properties that are psychological--i.e., derived from psychological experiments and not used by the other sciences. Despite appearances to the contrary, a functional approach is not circular, nor does it contaminate the behavioral data language with theory.