Development and validation of cognitive models for human error reduction

A three-dimensional conceptual model for human error reduction is proposed and partially validated. The three dimensions are instruction type, task type, and individual differences. Three experiments were conducted to test the model. The independent variables included subjects' knowledge level, task type, and instruction type. The dependent variables included subjects' attitude toward task, the number of finished tasks, errors in all tasks, errors per task, errors per subject, warning related errors in all tasks, warning related errors per task, and warning related errors per subject. Results of the three experiments partially supported the conceptual model and indicated the following statistically significant differences: (1) For low inference tasks, high knowledge subjects using procedural instructions made 57% fewer errors in all tasks and 82% fewer errors per task than high knowledge subjects using conceptual instructions. (2) For low inference tasks, low knowledge subjects using procedural instructions made 37% fewer errors in all tasks than low knowledge subjects using conceptual instructions. (3) For high inference tasks, high knowledge subjects using conceptual instructions made 244% fewer errors per subject than high knowledge subjects using procedural instructions. (4) For high inference tasks, low knowledge subjects using warning instructions made 68% fewer warning related errors in all tasks than low knowledge subjects using procedural instructions. (5) For high inference tasks, high knowledge subjects using explicit instructions made 58% fewer errors in all tasks than high knowledge subjects using implicit instructions. (6) For low inference tasks, high knowledge subjects using explicit instructions made 71% fewer errors per task than high knowledge subjects using implicit instructions. This research demonstrates how different instructions should be selected for different people performing different tasks in order to reduce human error rates.