Nonadditivity in Inference Judgments.
Abstract
It has been known for some time that subjects in Bayesian tasks produce data that look more like averages than like inferences. Shanteau suggested that the proper descriptive rule for the data is a weighted average. Wallsten, however, pointed out that a constant weighted averaging rule such as Shanteau used is formally equivalent to the Bayesian rule in that both are qualitatively additive. In principle, however, averaging can be differentially weighted, in which case it becomes non-additive. This paper reports two experiments that test for additivity in an inference task. Although the group data were, on the whole, additive, there were numerous violations of additivity in the single subject data. These comprised three general categories: (1) violations that are interpretable as resulting from a serial adjustment process in which adjustments are toward the value of the new information presented, (2) violations that are interpretable as resulting from a serial adjustment process in which adjustments are Bayesian-like in their direction, and (3) violations that appeared to be systematic but had no ready interpretation. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1982
- Accession Number
- ADA124232
Entities
People
- Lola L. Lopes