flinn@seismo.UUCP (E. A. Flinn) (12/15/83)
From a paper by A. Benjamin in a recent American Scientist: "Briefly stated, the findings are that when presented with an array of data or a sequence of events in which they are instructed to discover an underlying order, subjects show strong tendencies to perceive order and causality in random arrays, to perceive a pattern or correlation which seems a priori intuitively correct even when the actual correlation in the data is counterintuitive, to jump to conclusions about the correct hypothesis, to seek and to use only positive or confirmatory evidence, to construe evidence liberally as confirmatory, to fail to generate or to assess alternative hypotheses, and having thus managed to expose themselves only to confirmatory instances, to be fallaciously confident of the validity of their judgments (Jahoda, 1969; Einhorn and Hogarth, 1978). In the analyzing of past events, these tendencies are exacerbated by failure to appreciate the pitfalls of post hoc analyses."
trc@hou5a.UUCP (12/16/83)
Seems to me that the experiment sounds biased - the subjects are *instructed* to find an underlying order, which implies to them that the sequence *is* a generated one. A better experiment would have been to give them a mix of generated and random sequences, (and of a reasonable length - IE not so short that the subject cannot judge one way or the other - which is what is implied by the instructions of *this* experiment) and have the subjects sort them into "generated" and "random" sequences. The subjects should be informed that there is a chance that all the samples would be random or all would be generated. I suspect that errors will still be made in many cases, but that the errors will not be biased towards "jumping-to-conclusions". Tom Craver hou5a!trc
steve@lpi3230.UUCP (Steve Burbeck) (12/19/83)
Despite the skepticism of trc@hou5a.UUCP, the tendency of humans (and other organisms as well) to find "order and causality" in randomness is well known in experimental psychology. It is thought to underly what is commonly thought of a superstition. A simple experiment to see for yourself is to listen to a geiger counter. The clicks will seem to cluster and the usual perception is one of "I can feel that there is a pattern there and I can ALMOST characterize it." What is (or ought to be) disturbing about the first article on this subject, is that trained scientist are subject to the same problem with respect to their treatment of data that is inherently somewhat random. We need to be reminded occasionally of those sorts of weaknesses. lpi3230!steve