Eliciting and Analyzing Expert Judgment: A Practical Guide

Chapters 12 through 14 emphasized investigation and preliminary analyses for the purpose of gaining familiarity with the information gathered from the elicitation. This chapter and chapter 16 focus on what might be termed final analysis procedures that establish interpretable conclusions. This chapter concentrates on forming models whose results provide inferences. Chapter 16 concentrates on forming aggregations or combinations of the experts' judgments also for inference purposes.
This chapter presents modeling techniques and suggestions for describing the experts' answers in terms of other variables. These models are chosen based upon information already gained from the exploratory analyses done on the data base. Some modeling techniques use standard statistical procedures (e.g., least squares regression) as their foundation. These techniques fall under the heading of general linear models (GLMs). Applying multivariate methods such as factor, discriminant, and cluster analyses appropriately model the multivariate structure of the data base. However, as stated in previous chapters, we do not recommend the use of these techniques for final conclusions because of the assumptions required to use them. Other, more applicable modeling techniques are based on decision analysis methods and can be described as conditional models.
In general, the model is a functional relationship, h, between the answer variables, Y, and the ancillary variables, X, and is expressed as
GLMs define the expected value of the answer variable as a function of the ancillary variables:
GLMs also define the observed values of Y as the function, h, of the observed values...