Abstract

Anderson early on concluded from his research on impression formation that data supported averaging but not adding models, in spite of the fact that so-called set size effects, in particular, appeared to favor adding rather than averaging. A close review of methods used to test the models and the outcomes of such studies suggests that Anderson's support of the averaging model is biased, possibly due to greater flexibility of the averaging models. After making a distinction between N(number of input variables)-dependent weights(resulting in averaging) and N-independent weights(resulting in adding), a hypothesis was advanced, which stated that averaging is done when inputs contain conflicting feedbacks regarding the identical target while adding is favored when the input evaluations are consistent and/or number of inputs is made salient. It was further hypothesized that averaging is a mechanism by which people reduce uncertainty and complexity engendered by cognitive inputs.