The development and validation of a user's model for interactive text editing

Abstract

Interactive computer editors are an integral part of today's growing computer systems. In spite of this, their design is often determined by tradition or designer intuition, rather than empirical evidence. These studies explore a method of deriving naive and expert user models by use of a clustering algorithm. The models, which consisted of a core set of commands, were developed by applying a hierarchical cluster analysis to the paper and pencil responses of naive and expert subjects. It was expected that in an actual interactive editing environment, performance would be better when the operator used the appropriate model editor.

A validation study tested the models in an actual interactive editing environment. Analysis of variance results showed that novice and expert subjects did not have fewer errors, use fewer commands, or take less time to edit when using an editor based on the appropriate user model. In general, user (novice and expert), practice (first or second editing session) and some higher order interactions were the only significant effects. It was hypothesized that the high heterogeneity of variance might be obscuring some editor differences. A cluster analysis performed on the validation study data showed low agreement between subjects and with the original models. Differences in command frequency counts between the paper-pencil and interactive editing environments suggest that deeper aspects of interactive editing must be included in the user models. Not only command set, but mode change and current line location must be considered. It is suggested that interactive editors will more closely conform to user models as they more closely simulate the paper-pencil editing environment.