Extending goms to human error and applying it to error-tolerant design

The primary purpose of this work is to develop a methodology for designing human-error tolerant systems using GOMS (Goals, Operators, Methods, Selection Rules) theory, a family of techniques for characterizing human performance, as a foundation. GOMS has been successfully used to model human performance for system design, but has typically been limited to practiced, error-free behavior. In this dissertation I present a general framework for error recovery and discuss extensions to GOMS theory to allow modeling of erroneous behavior. The GOMS extensions are implemented in GLEAN (GOMS Language Evaluation and Analysis), a software tool for automating the execution of GOMS models. I also propose a technique, using GOMS, for analyzing systems to prevent human error. The technique is applied on WebStock, a web application designed to elicit human error, and the results are used to redesign WebStock's user interface. A human subjects experiment is used to evaluate the utility of GOMS error analysis by comparing the original WebStock interface with the interface improved using the technique. Some of the improved tasks were only changed in non-procedural ways (e.g. fonts, contrast, etc.). For other tasks, the procedures were improved (e.g. to reduce working memory load) in addition to the non-procedural changes. Over 200 errors were observed, the bulk of which involved some sort of memory failure. In the tasks where both non-procedural and procedural changes were made, there was an 80% decrease in the non-memory errors, and a 91% decrease in the memory errors. The results show that using GOMS error analysis can substantially reduce errors related to both the procedural and non-procedural aspects of an interface.