Patterned Predictions of Chance Events by Children and Adults

A few years ago psychological textbooks designed for elementary students frequently described a phenomenon called "random behavior" as a common precursor of goal-achieving actions. The distinguishing marks of responses thus depicted were their pragmatic futility for the protagonist and nuisance value for the psychological observer. More recently, numerous sequential two-choice "guessing" experiments (with an occasional daring extension by powers of two) have presented problems allowing only incomplete solutions, thereby conspiring to produce a good portion of "random behavior" on the part of unsuccessful guessers. The principal emphasis of investigators employing "guessing" experiments has been with acquisition learning possibilities, since it was soon discovered that probability matching of guessed predictions to stimulusevent frequencies took place with a sufficient number of trials. Much of this work, together with theoretical explanations, is described by Hake (13) and Bush and Mosteller (4) . What takes place when the probability-matching finding is trivial for a probability-prediction situation, as with a 50-50 random stimulus sequence, has been of interest-suggestive exceptions are an experiment by Hyman and Jenkin ( 14) and some unpublished analyses by Goodnow and associates ( 1 1 ) --chiefly as a control condition for acquisition groups which show palpable improvement. The purpose of the present study was to describe the characteristic response patterns produced by random orders on a basis other than probability matching. Generality, it was reasoned, might be gained by the very lack of improvement possibilities with 50-50 sequences. Would not response patterns obtained in such a context appear also in the initial stages of many kinds of attempted problem solution, as the textbooks had intimated, where complete problem solution was eventually possible. The more than half-serious conclusion was thus reached that a widely applicable problem-solving paradigm occurs when there can be no problem solution. The recent book by Bruner, Goodnow, and Austin ( 2 ) has delineated skillfully the contrary approach of giving full information for solvable problems. To be sure, in probabilistic situations this is only a matter of degree. And, we must confess, as the present study progressed, a few partially solvable problems