Abstract
To use the Bayesian approach to statistical inference, the auditor's subjective evidence must be quantified in the form of a prior probability distribution (PPD). Prior to this study there had been anecdotal evidence that normative expertise supplied from an external source is effective in the assessment of PPDs. Often, external normative expertise has consisted of (1) graphical feedback concerning the shape of the assessed distribution, and (2) validation questions concerning the implications of an assessor's responses. The current study examines the effectiveness of these techniques in guiding the assessment of PPDs. Use of the Cumulative Distribution Function (CDF) elicitation technique requires an assessor to understand relative probability densities. The anecdotal evidence had suggested that the relative probability density between two points could be communicated to an assessor by drawing a graph of the assessed distribution and by asking questions that describe the relative densities. Student-subjects assessed (1) the distribution of starting salaries for Texas A&M graduates, and (2) the distribution of Grade Point Records (GPRs) in the College of Business Administration at Texas A&M University. While all subjects assessed both distributions, the presence of graphical feedback and the order of the cases was varied across subjects. The effect of graphical feedback was measured as a function of both (1) the accuracy of the subject's final estimate, and (2) the change in accuracy from the initial to the final estimate. The effect of validation questions was examined in each treatment cell by examining whether the change in accuracy from the initial to the final estimate was zero. For both cases, graphical feedback was found to have a positive effect on the accuracy of the final estimate and the change in accuracy from the initial to the final estimate. The validation questions were found to be more effective in the Salary distribution case; validation questions were not as effective in the GPR case.
Viator, Ralph Edward (1986). A computer-based approach to the assessment of prior probability distributions : an empirical investigation. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -22418.