A Causal Model of Human Error for Safety Critical User Interface Design

This paper describes a method of assessing the implications for human error on user interface design of safety-critical software. In previous work we have proposed taxonomy of influencing factors that contribute to error. In this paper, components of the taxonomy are combined into a mathematical and causal model for error, represented as a Bayesian Belief Net (BBN). The BBN quantifies error influences arising from user knowledge, ability and the task environment, combined with factors describing the complexity of user action and user interface quality. The BBN model predicts probabilities of different types of error, slips and mistakes, for each component action of a task involving user-system interaction. We propose an Impact Analysis Method that involves running test scenarios against this causal model of error in order to determine those user actions that are prone to different types of error. Applying the proposed method will enable the designer to determine the combinations of influencing factors and their interactions that are most likely to influence human error. Finally we show how such scenario-based causal analysis can be useful as a means of focusing on specifically relevant guidelines for safe user interface (UI) design. In the paper the proposed method is demonstrated through a case study of an operator performing a task using the control system for a laser spectrophotometer.