Applying the Multivariate Lens
Model to Fault Diagnosis
Pratik D. Jha and Ann M. Bisantz
This study investigates the application of a multivariate lens model to judgments of fault diagnosis in a dynamic, process control system. Although there are documented examples of the univariate lens model across numerous domains, applications of the multivariate model are more limited and have not been extended to judgments in more complex, human–machine systems. Such extensions may prove valuable, because many judgments in such systems, such as those concerning fault diagnosis and recovery, are multivariate or categorical in nature. To investigate the utility of the multivariate lens model for this application, a sensitivity analysis was conducted on simulated fault diagnosis data at three levels of performance, within the context of a dynamic process control simulation. Additionally, experimental results were collected and modeled from 16 participants under two interface and two fault severity conditions using the same simulation. The sensitivity analysis showed that parameters of the multivariate model were in fact sensitive to changes in fault diagnosis performance. However, multivariate parameters showed less sensitivity to performance changes in the experimental results because of the nature of the faults that were made and the canonical correlation procedures used to compute the parameters. Further investigation of the canonical correlation outputs was useful in identifying participants' judgment strategies.
In many decision-making situations, important aspects of the environment, such as the values of informative variables or the relationships between variable values, change over time. Feedback about results of actions is typically available, and it may be necessary to make decisions quickly (Cannon-Bowers, Salas, & Pruitt, 1996; Zsambok 1997). In these situations, decisions are often composed of series of smaller decisions and actions, decision choices are impacted by feedback about the results of just-prior actions, and decision makers must identify the state of the situation to act (Brehmer, 1990; Hogarth, 1981). The remainder of this section discusses models of such dynamic decision making, particularly as applied to fault diagnosis, and relate judgments regarding situation state to the lens model formalism. Models of dynamic or naturalistic decision making in the real world emphasize the cyclical nature