An Empirical Investigation of Bayesian Inference with Two People.
Abstract
Twenty subjects from the University of Southern California performed a Bayesian inference task in pairs. Like earlier research in inference, the individuals were asked to infer posterior odds about a pair of hypotheses from a collection of data. Unlike the earlier studies, the individuals were then required to aggregate their posterior odds with those of another individual who had seen a second set of independent data samples to form an opinion about the same pair of hypotheses. Conservatism and radicalism findings of earlier studies were reconfirmed. Individual subjects responses collected before aggregation showed conservatism in the high d' condition and radicalism in the low d' condition. The aggregated final odds from the pairs of subjects seem to reflect some confusion. Some of the subjects apparently used a simple and incorrect averaging strategy. Others did not use this strategy but in general, pairs of subjects were unable to provide anything but conservative final odds when they aggregated their two opinions. The importance of using real stimuli, the way the responses were elicited, and the instructions that were given to the subjects are discussed. Also, a 'mean of means' or arithmetic log likelihood response mode is discussed as an alternative elicitation mode that may be useful in information aggregation when more than one person is involved. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1978
- Accession Number
- ADA073368
Entities
People
- Ward Edwards
- William F. Gabrielli Jr.