Criticism Against Einhorn and Horgarth`s Theory of Diagnostic Inference

Abstract

所謂診斷性推論，是指人們會依所發生的事件，來推測造成這些現象之原因的推測方式。艾宏及侯佳士(1982)認為診斷性推論是一個主動的，強調因果關係的，由果推因的歷程，所關心的是對特殊事件而非一般性事件的推論。他們批評一般統計模式無法處理有這些特質的診斷性推論，所以提出一個新的描述性模式說明人們處理這類問題的歷程。事實上，他們對於一般統計模式的批評是不合理的，他們錯將所處理的貝氏問題以點估計問題視之，以至於提出很多錯誤的論點。同時他們提出的新模式雖然得到研究結果的支持，但在研究方法上卻有缺失之處。這篇文章主要就艾宏－－侯佳士文中有關統計及研究方法方面提出批判。

The particular issue with which this paper is concerned is diagnostic inference.That is, given the occurrence of a set of outcomes/results/ symptoms, one has to infer to what extent is a particular action or event responsible for the observed effects. Einhorn and Hogarth (1982) argued that the essential aspects of such inferences are that they are causal rather than correlational, backward rather than forward (one goes from effects to prior causes), concerned with specific rather than the general cases, and constructive rather than nonconstructive (one can synthesize, enlarge, or otherwise develop new hypotheses). They further argued that the most common statistical model (e.g., Peterson & Beach, 1967) involving inferences does not consider these four aspects, and they developed a new model to describe how people assess the likelihood that one of two hypotheses is true on the basis of varying amount of evidence for each. I shall show, however, that their claims against the usual statistical model are unfounded and that they, in fact, misconceive the type of statistical problem with which they are faced.They think they are dealing with point estimation problems, when, in fact, the diagnostic problems with which they are dealing are Bayesian problems.Furthermore, even though they concluded that their model fitted the data reasonably well, some methodological considerations provide questions about their conclusion. The main purpose of this paper is to critique Einhorn and Hogarth`s arguments and model in statistical and methodological terms.