|
|
|
-- Kitajima & Polson (1995)Kitajima, M. & Polson, P. (1995). Mechanisms of slips in display-based human-computer interaction: a model-based analysis. Advances in Human Factors/Ergonomics, 20B (HCI International'95), 515-520. Elsevier.
Mechanisms of slips in display-based human-computer interaction: a model-based analysisA puzzling and frequently-ignored fact in the human-computer interaction literature is that skilled users have surprisingly high error rates (10 ~ 15%). Card, Moran and Newell (1983) studied individual skilled users performing two tasks, manuscript editing and electronic circuit design editing. The manuscript editing experiment involved a detailed evaluation of a single expert user doing 70 edits presented in marked up manuscript. Errors were made on 37% of the command sequences describing edits. Over half of the errors were detected and corrected during generation of the editing commands. 21% (15 out of 70) of the commands issued by this very skilled user generated the wrong result and required additional edits to correct these errors. In a second study of a single expert carrying out an electronic circuit design editing task, the user had an error rate of 14% on 106 edits. Hanson, Kraut, and Farber (1987) studied 16 researchers and managers who were intermediate and expert level users of UNIX performing document preparation tasks and e-mail. They logged over 10,000 commands. The overall error rate was 10% with error rates ranging from 3% to 50% on different commands. The experiments briefly reviewed here are representative of results from a wide range of studies in the human-computer interaction literature. Error rates for expert users range from 5 to 20%. In all studies of experts, users eventually produced the correct results. Approximately 50% of the errors are detected during the generation of a command and corrected. Detection and correction of errors is an integral part of expert skill. The literature on errors has concluded that there are two qualitatively different types of errors (Norman, 1981; Reason, 1990). The first is errors of commission, or mistakes. Such errors are committed by users who are carrying out a novel tasks and fail to immediately discover the correct action sequence. The other is slips, where users have the correct intention but fail to successfully execute the correct action sequence. Most part of errors described above is slips. Sellen (1990) reviews classes of models that provide principled, qualitative accounts for slips. She argues that all of these models have a hierarchical representation of action sequences that include representations of top-level task goals and lower-level goals that actually control execution of elementary actions. Reason (1990) argues that control of attention is a critical determinant for generating correct performance from a hierarchical representation of action sequences that include representations of top-level task goals and lower-level goals that actually control execution of elementary actions. Failure to adequately attend to the ongoing process and coordinate the interaction between the various schema causes a wrong low-level schema to become activated, generating related but incorrect actions for the current task. In HCI tasks, the users could be focusing on the task of composing new text or drawing a figure, and so on. This would lead to insufficient attention being allocated to subtasks involved in operating the interface. Card, et al. (1983) proposed that experts accept high error rates in order to increase their productivity, because for them error recovery can be done easily and rapidly. Experts trade speed for accuracy, causing slips. In this paper, two mechanisms of slips, attention failures, and speed-accuracy tradeoffs are simulated by a comprehension-based cognitive model of display-based human-computer interaction proposed by Kitajima and Polson (1992, 1994a, to appear), showing that they could account for the rate of slips made by skilled users interacting with graphical user interfaces (Kitajima and Polson, 1994a, 1994b).
Download
|