Effects of information sequence and irrelevant distractor information when using a computer-based decision aid

Decision Sciences, Winter 2001 by Reneau, J Hal, Blanthorne, Cindy

RELATED LITERATURE

The dominant normative model of information-based belief revision is Bayes' theorem. Early research in this area (e.g., Peterson & Beach, 1967) suggested Bayes' theorem provides a relatively good descriptive model of belief revision. In contrast, the results from over 30 years of experimentation document fundamental differences between human judgments and Bayes' model predictions (e.g., Tversky & Kahneman, 1981; Hogarth, 1982). Three specific points are briefly summarized from this literature:

1. The sequence in which information is presented affects judgments in a predictable manner.

2. The presence of irrelevant distractor information reduces judgment accuracy.

3. Neither information sequence nor irrelevant distractor information affect judgment confidence.

Sequence Effects

Historically, experimenters have observed both primacy and recency sequence effects that are normatively irrelevant and, thus, violate the Bayesian model (e.g., Nisbett & Ross, 1980, Ch. 8). However, a theoretical foundation has been established that attempts to explain diverse findings in the literature regarding effects of information sequence. Hogarth and Einhorn (1992) developed and tested a descriptive model of belief revision, the Belief-Adjustment model. The model assumes an anchoring and adjustment process that depends on six task characteristics: (1) type of task (evaluation or estimation), (2) complexity of the task, (3) amount of information presented, (4) type of information (consistent or mixed), (5) presentation order of information, and (6) whether the response mode requires that beliefs be obtained after presenting each piece of information ("Step-by-Step") or all of the information ("End-of-Sequence"). Although the response mode required by a task often parallels the type of processing utilized by decision makers, this experiment requires an End-of-Sequence response mode, yet it is expected that auditors will utilize Step-by-Step processing due to information complexity and the length of the series of information exhibited to auditors.

The Step-by-Step form of the model assumes that one's belief is revised using a sequential anchoring and adjustment process. This process implies multiple adjustments, whereas End-of-Sequence processing assumes a single adjustment based on the aggregate impact of the evidential set. An important, unambiguous prediction of the Step-by-Step model is the presence of recency sequence effects for mixed information sets. That is, information received late in a sequence is given more weight in the overall judgment. This effect has been consistently observed in a variety of unaided judgment experiments using students (Einhorn & Hogarth, 1985; Hogarth & Einhorn, 1992), experienced auditors (see Asare & Messier, 1991, for a review; Messier & Tubbs, 1994), and experienced tax professionals (Pei, Reckers, & Wyndelts, 1990, 1992). The ability of these experiments to examine the descriptive validity of belief revision as posited by the Belief Adjustment model can be questioned due to inherent difficulties in measuring or manipulating the subjective strength of evidence and one's sensitivity to evidence. Yet, the number of studies and consistency of results supporting the prediction of recency effects due to Step-by-Step processing lead us to expect recency effects. It is not the authors' intention to test the descriptive validity of the Belief-Adjustment model. We believe the model has undergone enough testing to support its descriptive validity. Instead, we apply the model in an attempt to further our understanding of the effects) of decision aid structure on judgment. Accordingly, we predict: