RECOGNITION OF BIAS IN STRINGS OF BINARY DIGITS

Summary.-The behaviour of humans in several situations may be largely dependent on their ability ro recognize random sequences. This paper reports an experiment on the recognition of bias, in which various strings of binary digits, each varying in degree of bias, were presented to 44 university students. The method used was the method of paired comparisons and the results were subjected to a multiple-comparison test. It was concluded that, in general Ss recognized the differences berween the strings. Much has been written about response bias, and a number of papers (e.g., Weiss, 1964; Tune, 1964; Baddeley, 1966) have been published in which the production of random response has been experimentally investigated. The general findings are that human Ss are not able to produce a random sequence of digits, but that certain factors (increasing the time interval between individual responses, for example) seem to improve this ability. Baddeley (1966) noted that his Ss were, however, able to select random digit sequences from a mixture comprising sequences derived from random number tables and also sequences generated by other Ss under random-response instructions. So far as this author is aware, this is the only report on randomness recognition in the literature. The purpose of the present paper is to report a pilot experiment, in which the ability of Ss to recognize different degrees of bias is investigated. The problem of recognition of bias is central to several topics in psychology. In particular, it could be argued that the results obtained in probability learning experiments (e.g., Grant, Hake, & Hornseth, 1951; Estes & Straughan, 1954) could be explained by the inability or unwillingness of Ss to recognize the random nature of the sequences presented. Goodnow ( 1955 ), in fact, takes such a line when she argues that Ss are trying to solve a problem. When Galanter and Smith (1958) used patterned sequences in a situation similar to a probabilitylearning situation, they showed that, as patterns became more complex, Ss tended to require more trials before they "saw" the structilre of the sequence. Perhaps chis would also help to explain Edwards' (1961) finding that over a thousand uials probability matching did not occur. Certainly, the appropriate behavioiu in a probability-learning task (the consistent choice of the more frequently occurring element) is demonstrated by very few Ss. Other situations in which the recognition of bias is especially important are those in which formal models prescribe a different type of behaviour under a random reinforcing system than under a biased reinforcing system. Decision theory and game theory make such prescriptions; and it should be possible to devise a computer program for dealing with si~ch situations. As a first step, it