Social inference as a function of the number of instances and consistency of information presented.

Abstract

The present study investigated two major hypotheses: Instances which confirm or disconfirm a generic assertion (e.g., "Bill hurts Antuvians") combine additively to yield the judged likelihood that the assertion is true; and the less expected a piece of credible information, the greater the impact of that piece of information on a person's likelihood judgment. The subjects were asked to decide how likely they thought it was that a man (described as either kind, cruel, or by no adjective) would perform a certain act (which was either evaluatively positive, neutral, or negative) after being given 0, 1, or 2 instances of confirming and/or discontinuing information. Within each Adjective X Act combination, the likelihood data were well described by an additive model. As expected, information was found to have greater impact when (a) the subject was given an instance of information which was inconsistent with other instances presented in an item; (ft) the subject was given a discontinuing instance of a positively evaluated assertion or a confirming instance of a negatively evaluated assertion; and (c) the subject was given a disconfirming instance of a balanced situation or a confirming instance of an imbalanced situation.

Test of an additive model of social inference.

“…This result would ordinarily be interpreted as inconsistent with additive models. Two points that have not been noted previously deserve emphasis: (a) The values of instances model (Gollob et al, 1973) can account for this result; (b) the values of instances model is not equivalent to the averaging model and can be tested against it with a suitable experiment.…”

“…Certain experimental designs do not allow the distinction between constant-weight averaging models and additive models. When appropriate designs have been used, research in several information integration domains has ruled out additive models in favor of averaging formulations (Anderson, 1971;Birnbaum, 1974; Birnbaum et al, in press).It is important to note that the "additive, values of instances model" of Gollob et al (1973) is not additive in the traditional sense and can account for a result that would ordinarily be interpreted as critical evidence against adding models. Gollob et al (1973) assume that the impact of redundant information will be less.…”

“…However, information integration experiments often find that J(L + LH) > J(L), yet J(H + LH) < J(H), where L is a lowvalued item, H is a high-valued item, LH is a combination, and J represents the judgment. The data of Gollob et al (1973) show that the addition of a disconfirming-confirming combination (HL) to a disconfirming 655 …”

“…Gollob et al (1973) assume that the impact of redundant information will be less. This assumption is termed the expectancy-impact hypothesis: Expected information should have smaller effects.…”

“…However, information integration experiments often find that J(L + LH) > J(L), yet J(H + LH) < J(H), where L is a lowvalued item, H is a high-valued item, LH is a combination, and J represents the judgment. The data of Gollob et al (1973) show that the addition of a disconfirming-confirming combination (HL) to a disconfirming 655 instance (L) raises the rating from 3.58 to 4.S7, whereas the addition of the same HL combination to an H item lowers the rating from 6.8S to 6.67. This result would ordinarily be interpreted as inconsistent with additive models.…”