Time-Series Forecasting

excellent discussion and explanation of Bayesian techniques is given on pages 230–233. Chapter 12 discusses initial procedures for looking at data. Topics covered include the granularity of the data and its possible conditional structure. The authors recommend quantifying as much information as possible and suggest ways to do this. This quantiŽ ed information should be placed in the database for further analysis. Chapter 13 explores ways of understanding the resulting database structure. It provides methods for exploring the relationships between ancillary variables and answer-response variables and also ways of exploring relationships within each of these variable sets. The results of these analyses are not meant to be used as the Ž nal results within the investigation, but rather are preliminary to further statistical investigation. Correlation, factor analysis, cluster analysis, and general linear models are some of the techniques used to explore relationships between variables. Chapter 14 discusses correlation and how it relates to the application of expert judgment. The relationship of correlation to bias is also covered, with a subsection devoted to a detailed description of a 14-step method for detecting correlation among experts. Chapters 12–14 are concerned with investigating and conducting a preliminary analysis of data for the purpose of becoming familiar with information gathered from the elicitation. Chapters 15 and 16 focus on Ž nal analysis procedures that establish interpretable conclusions. Chapter 15 presents modeling techniques that may yield inferences and also suggests ways for describing experts’ answers and terms of the variables. General linear models are emphasized. Multivariate methods, such as factor, discriminate, and cluster analysis, are used to appropriately model the multivariate structure of the database. But the authors do not recommend using these techniques for the Ž nal conclusions because of the assumptions required to use them. They recommend more applicable modeling techniques based on decision analysis methods, which can be described as conditional models. Chapter 16 discusses combining schemes for aggregating the responses of experts and the environments in which the schemes would be used. It also discusses the relationship of the aggregation problem to the problem of characterizing uncertainties. Particular emphasis is placed on Saaty’s method of weight determinations. Chapter 17 is concerned with characterizing uncertainties. It discusses the four basic sources of uncertainty and reviews the procedures for obtaining uncertainty measures, modeling uncertainties, and comparisons of the methods. Chapter 18, the book’s Ž nal chapter, is concerned with making inferences from the data. Two types of inferences are discussed: general and statistical. The authors emphasize that expert information does not allow one to make inferences beyond the available knowledge base. They state that the inferences made do not necessarily represent a true state of nature, nor are they statistically based inferences. This book belongs on the shelf of every researcher who is using or thinking of using expert information in their work. It provides a comprehensive guide to simple, but appropriate ways to elicit expert data, as well as how to analyze it.