Linear versus process tracing approaches to judgment modelling: A new perspective on cue importance

For a bankruptcy prediction problem, the judgment formation process is studied using linear models and process tracing models. The linear models are constructed using traditional linear discriminant analysis techniques. The process tracing models are constructed using computer-generated algorithmically-based decision nets. All the models presented show good predictive accuracy. However, the linear models and process tracing models diverge widely on several measures of cue importance. This divergence, for a fairly straightforward problem, is intriguing since virtually all the evidence in the accounting literature about cue importance is based on linear models research. The importance of different information cues to decision-makers is clearly a critical issue in the design of effective accounting information systems. Thus, this study suggests the need for much more careful attention to the complex question of assessing cue importance.