One-and two-way analysis-of-variance procedures are shown logically to be appropriate for testing hypotheses in successive treatment reversal designs for one-subject and Nsubject experiments, respectively. The applicability of these designs is demonstrated through analyses of typical data.The preponderance of studies conducted within the paradigm of operant behavior employ the intrasubject replication design (often called, simply, the reversal design), in which various treatments are successively applied to and removed from the same subject (Sidman, 1960). For example, the A-B-A-B design (in which A -baseline of no reinforcement for a certain response, B = the contingent availability of a reinforcing stimulus following that response) is widely used to demonstrate that if a certain reinforcer is made available contingent upon a response (treatment B), the effect is to increase the frequency of that response above operant level (condition A). Then conditions are reversed and treatment A is reinstated, during which time the response rate is expected to revert to its operant level. Finally, treatment B is re-applied and the response rate is expected once again to increase to above operant level.Control changes in behavior as a result of changes in the treatment conditions provide the most convincing demonstration of functional relationships.The importance of this general procedure for the experimental analysis of behavior can hardly be overemphasized. This design, or variants of it, has been the vehicle for many principles of behavior developed in the last several decades and for most of the successes of operant behavior modification procedures in practical settings. (Indeed, the general form of this argument-if A, then B; if not A, then not B-is one of the most fundamental arguments in scientific methodology.)Nevertheless, there are some disadvantages of the reversal design (see Bandura, 1969, pp. 242-244), one of the most serious of which is the interpretive problem of how large does the behavior change from treatment to treatment have to be to be considered a significant change. As Bandura points out, interpretation is not difficult provided that large successive behavior changes occur rapidly and consistently for many subjects. The interpretive problem arises in those cases in which the behavioral changes are not dramatic or in which some individuals remain unaffected by repeated exposure to one of the treatments. Such findings are especially prevalent in situations, such as classrooms, in which laboratory controls for creating favorable experimental conditions are difficult to achieve. The problem as stated reduces to a statistical one: 193 1972, 5, 193-198 NUMBER 2 (SUMMER 1972)

No abstract

No abstract

Three experiments were performed to test the hypothesis that the skewness of the serial‐position curve is determined, at least in part, by memory span. Expt. I showed that the degree of skewness of the serial‐position curve is positively related to the length of the subject's immediate memory span. Memory span for sequences of colour‐forms was determined for 47 subjects (university students). Eleven subjects with the highest span (HS) and 11 subjects with the lowest span (LS) were compared in the serial learning of a 9‐item list of colour forms. The HS group produced significantly more skewed serial position curves both for overt errors and for failures to respond than did the LS group. Expt. II investigated the effects of prolonged practice on serial learning. Three subjects each learned a different 9‐item list of colour‐forms every day for 4 weeks (20 days). Practice increased the ease of serial learning to the point that the entire list could be comprehended within the subjects' memory span. There was a corresponding increase in the skewness of the serial‐position curve as a function of amount of practice. Expt. III tested the hypothesis that a list of items for which subjects have a relatively long memory span would produce a more skewed serial‐position curve than would a list composed of items for which subjects have a relatively short memory span. Forty subjects each learned a 12‐item list of single letters (in a random order) and a 12‐item list of 3‐letter nonsense syllables. As predicted, the list composed of single letters produced a significantly more skewed serial‐position curve than did the list of nonsense syllables.

No abstract