Unsolved- and insoluble-problem behavior.

Abstract

One of the features of hypothesis-testing theories has been the concept of sampling from a finite set of hypotheses (Hs). This has led to interest in two related questions: (a) Why do 5s fail to solve a problem ? Previous data have suggested that 5s fail to solve because the solution H is not in their subset (the missing H assumption), (b) What happens when all the Hs S is testing are disconfirmed (i.e., when the subset goes to zero) ? One possibility is that 5s will resample the subset again (the recycling assumption). These assumptions were tested in an experiment comparing performance on seven-dimensional insoluble problems and eight-dimensional unsolved problems. The results were consistent with both assumptions.Recently, workers have been examining human discrimination learning in terms of the testing of hypotheses (Hs) (Bower & Trabasso, 1964;Levine, 1966). In a theoretical paper (Levine, 1971), the problem of why 5s fail to solve a problem has been considered. Based on the notion that 5s sample subsets of several Hs at a time, it is clear that an 5 might sample a subset not containing the solution H. This suggests that 5s fail to solve a problem because the solution H is not in their subset ("missing H" assumption). Levine, Yoder, Kleinberg, and Rosenberg (1968), considering presolution performance, noted that the same description applies to insoluble problems (i.e., problems in which no dimension is relevant). The present study was designed to test the untested assertion made by Levine et al., that behavior during unsolved problems should be similar to that during comparable insoluble problems. A subsidiary purpose was to examine the effects of 5's subset going to zero. Levine et al., using short problems, found that 5s tended to become poor information processors ("giving up"). It is possible that given more time (i.e., feedback trials), 5s may sample and test a new subset ("recycling" assumption).Method.-Forty-eight undergraduates served as 5s, participating as part of the introductory psychology course requirements. Of these, 3 were dropped for apparent cheating (see below) and were replaced.A Kodak Carousel projector was used to present the stimuli. Slides were constructed of eightdimensional (8-D) stimuli using the following dimensions: letter (A or T), color (black or white), size of letter (large or small), line under letter (solid or dashed), border surrounding letter (circle or square), number of borders (one or two), type of borders (solid or dashed), and number of dots above border

Examination of hypothesis-sampling theory.

“…A variety of studies verified this , assumption Erickson & Zajkowski, 1967;Erickson, Zajkowski, & Ehmann, 1966;Guy, Van Fleet, & Bourne, 1966;Holstein & Premack, 196S;Trabasso & Bower, 1966). The second component of this statementthat a correct response should be at chance level-is challenged by the findings of Levine and his associates (Classman & Levine, 1972;Levine, 1971;Levine, Miller, & Steinmeyer, 1967;Levine et al, 1968). They have consistently found that the probability of a correct response, prior to solution, is between .3 and .4, when chance is .5.…”

“…There is actually some evidence suggesting that subjects are not aware of all the potential hypotheses either at the beginning, or during the course, of the experiment. Levine discovered that some subjects actually omit an entire dimension from consideration and that this creates a functionally unsolvable problem for them (Glassman & Levine, 1972;Levine, 1971;Levine, Yoder, Kleinberg, & Rosenberg, 1968). This phenomenon is covered more thoroughly later under a discussion of dimensional sampling.…”

“…Presolution depression. Prior to solution, Levine repeatedly showed the probability of correct stimulus choice to be between .3 and .4, whereas chance responding would be a .5 under his designs (Glassman & Levine, 1972;Levine, 1971;Levine et al, 1967;Levine et al, 1968). Levine (1971) suggested that this finding is due to the absence of one of the four dimensions from consideration by the subject.…”

“…He found essentially identical trial-by-trial, correct-choice probabilities for both groups, thus supporting the insoluble-unsolved assumption, as well as the missing dimension assumption. Glassman and Levine (1972) replicated this design, this time using only subjects who had both unsolved and insoluble problems. While this matching procedure eliminated possible group differences due to differential subject selection, it still caused the poorer subjects to be represented more heavily, overall, because they were more likely to have unsolved problems occur.…”