In this article we are concerned with the often surprising degree of behavioral control exerted by what are, in many cases, unscheduled and unintended relationships between important experimental events. A distinction is proposed between traditional contingencies (i.e., if event X-then event Y) and a second class of relationships which are termed autocontingencies. The logical relationships which generate autocontingencies are derived from systematic constraints in the distribution of event Y itself, which allow prediction of the subsequent occurrence or absence of event Y.Supportive evidence for behavioral control by autocontingencies is presented from a survey of operant and Pavlovian conditioning procedures which involve both appetitive and aversive stimuli. Autocontingency effects are examined within a variety of situations which include such well-established phenomena as "scalloping" under fixed-interval food schedules, as well as the "reverse scalloping" or negatively accelerated responding which occurs when food-reinforced responding is punished under a fixed-interval schedule. An analysis of several recently published experiments suggests the possibility that subjects may occasionally have become "aware" of an autocpntingency before the experimenter who programmed it did. These experiments, which typically yield puzzling behavioral effects, have often forced the underlying autocontingency into the investigator's attention. Such instances support the notion that autocontingencies involve subtle relationships, despite the fact that their effects are often far from subtle.An additional source of support for and elaboration of the concept of autocontingencies comes from our own conditioned suppression data. We exposed rats to different experimental arrangements of signal and shock. Subjects for whom no signal preceded each shock delivery learned to find "safety" in more subtle features of the experimental situation. For example, these subjects showed reliably increased responding in the brief periods following each shock delivery once they learned that shocks were separated by a minimum 3-min interval. Similarly, a constraint of three shock deliveries per session yielded a "subtle safety signal" and resulted in enhanced responding following offset of the third shock. These data also suggest the function of "need" (i.e., the simultaneous presence of other predictors) in determining the degree of control exerted by autocontingencies. Subjects in our experiment for whom a traditional tone-shock contingency reliably produced warning did not show behavioral control by autocontingencies which were simultaneously available.The use of the term subtle to describe the relationships underlying autocontingencies suggests that a certain degree of neurological sophistication may be necessary to integrate or process the information provided by autocontingencies. Pre-169

No abstract