GUI Users Have Trouble Using Graphic Conventions on Novel Tasks

Catherine A. Ashworth

Department of Psychology & Institute of Cognitive Science
University of Colorado
Boulder, CO 80309 USA
Tel: +1-303-492-5394
E-mail: ashworth@clipr.colorado.edu

ABSTRACT

Twenty-five Macintosh users performed poorly when attempting novel tasks in Macintosh-like applications. The tasks tested subjects' understanding of the meaning of ten different GUI graphic conventions (such as the symbol for a Pop-Up Menu). Subjects who had used more applications had greater accuracy rates. The trials testing Ellipses and the Walking Menu symbol revealed that even when subjects knew the convention, they were also guided by the quality of the semantic match between the command label and the task goal. An analysis of likely reasons why subjects did not know Radio Buttons or X-Boxes suggests users can employ a strategy of "re-exploring" an interface object on each use. These findings have implications for current GUI design guidelines and for theories of GUI learning and use.

Keywords:

Graphic User Interface, GUI, display-based computing, display-based skill, exploratory learning, user testing, Macintosh.

INTRODUCTORY SECTION

Although the details of how people use command-line interfaces are fairly well known [1,2], less is known about how people learn and use graphic user interfaces (GUIs). It is known that in great contrast to command-line users, GUI users appear not to have highly detailed precompiled plans, but can be uncertain about the exact consequences of a command (e.g., resulting cursor location) [8]. In addition, users can have trouble recalling details of screen contents [7] and of their action sequences [7,8]. Theories of the skill of using a GUI have recently been developed [2,4,9]. In general, these theories posit that users determine their next goal or next action by combining information from the screen with general information in their head.

However, other than studies of how to craft menus (REF), there has been little analysis of the GUI objects which are intended to guide users' inferences about the available functionality and how to interact with it. Simply put, how do users learn about the graphic conventions, what do they learn about them, how do they use them, and why? These are the questions that this research addressed.

METHOD

Twenty-five Macintosh users representing a wide range of employment, age, and experience were paid to participate in the experiment. Three types of data were collected from subjects: performance data, declarative data, and interview data.

Subjects performed four blocks of ten trials each - one for each of the ten conventions tested. The ten conventions tested were: Radio Buttons, X-Boxes, Ellipses, Dim Command Text, Dark Command Text, Keyboard Menu Shortcuts, Walking Menu Symbol, Pop-Up Menu, Heavy Button Border, and Highlighted Text in a Text Field. On each trial, the computer display contained a task description and a screen shot from a fictitious yet Macintosh-like application. Subjects' first response was captured and scored. Subjects also produced extensive declarative knowledge reports for each convention in which they explained: the goals that can be achieved, how to do so, what they see when they do it, and, when applicable, the nature of other objects that the convention makes appear. Finally, subjects were interviewed concerning their years using a Macintosh, where they use the Macintosh, and the applications that they use.

FINDINGS

In general, subjects performed poorly on the computerized tasks in which they had to use their knowledge of the conventions to achieve a goal. Interestingly, the only reliable predictor of subjects' accuracy was the number of applications that they had used (r= .37, p. = .08, trend).

Ellipses and Walking Menu

Each trial for Ellipses and Walking Menu had two buttons that were semantic matches to the task goal. One of the synonyms was the better semantic match; one had the correct convention graphic. These button characteristics were counterbalanced.

When subjects did not know the convention, they selected the button with the better semantic match. This occurred on the Ellipses trials on which there was no main effect of the convention, but there was a main effect of semantic match (cumulative probability of obtained counts given the expected counts equaled .00005). When subjects knew a convention better than chance but not well, they were more likely to select the button with the correct graphic when its label was the better semantic match. This occurred on the Walking Menu trials on which there was a main effect of convention graphic (probability equaled .00009), no main effect of semantic match, but a marginal interaction of the two factors (c2 (1) = 3.46, .10 > p. > .05). Specifically, when the command with the correct graphic was chosen, the proportion of trials with the best semantic match (versus worst) was well over chance (probability equaled .039).

Radio Buttons and X-Boxes

The trials for both Radio Buttons and X-Boxes required subjects to de-select an option. Knowing how to de-select an option is probably strongly related to understanding what the graphic means about the nature of the options. For instance, a subject who knows that Radio Buttons signal that the options are mutually exclusive and that one option is required is likely to (correctly) de-select an option by clicking on another in the set. Since we expected that users interact with these two option selection graphics based on the textual labels, the tasks used nonsense labels.

Subjects performed poorly on trials for both Radio Buttons and X-Boxes although they did better on the X-Boxes. They erred primarily in treating the Radio Buttons as if they were X-Boxes, but were also unclear about how to use those.

DISCUSSION

Our findings have implications for both theories of GUI use and for guidelines for the design of GUIs.

First, the finding that not all interface conventions are well known is interesting since the Macintosh is often so easy to use and seems to provide so much information about itself. Designers should be aware that not all conventions are well known and should perhaps minimize their use of the harder conventions. In addition, design priniciples derived from the relative ease of the various conventions could be used to guide creation of new conventions.

Second, as designers have long suspected, and display-based theories of GUI use [2,4,9[ have recently formalized, users possess a robust semantic matching strategy that is used even when they understand much of what the convention graphic is telling them. The confirmation of semantic matching extends our understanding of why good menus can be so much easier to use than typed commands.

Third, it is very likely that users do not know the meaning of both Radio Buttons and X-Boxes because the textual labels that accompany them (e.g., "Small," "Medium," and "Large" for Radio Buttons) completely convey the nature of the options and therefore how to interact with the graphics. In the event that the user does not fully understand the implications of the textual labels, the graphic provides instructional feedback to both correct and incorrect actions. For example, when the user tries to select an additional Radio Button, the highlight moves to the most recently selected button, and the user sees that only one Radio Button can be selected at a time.

Interface objects whose feedback is so instructional could easily support a strategy of "re-exploration" in which knowledge about the interface object is recreated during each use. "Re-exploration" would be an extremely display-based process that could support use but foil traditional learning. Designers would want to control on which portions of applications users employed a "re-exploration" strategy. Also, theories of exploration-based learning of applications [9] may well be relevant to more than a user's first work sessions. Theories of computing as a display-based skill will want to incorporate this user strategy and determine the conditions that support it.

REFERENCES

Card, S. K., Moran, T. P. & Newell, A. The Psychology of Human-Computer interaction. Hillsdale, N.J.: LEA 1983.
Howes, A. & Young, R. M. A learning and performance model of display-based task-action mapping. Submitted for review 1994.
Kieras, D. E. Towards a Practical GOMS Model Methodology for User Interface Design. In M. Helandar (Ed.) Handbook of Human-Computer Interaction. North-Holland 1983, pp. 135-158.
Kitajima, M. & Polson, P. G. (1994). A model-based analysis of errors in display-based HCI. In Conference Companion of CHI'94. New York, NY: ACM, 301-302.
Mayes, J. T., Draper, S. W., McGregor, A. M. & Oatley, K. Information flow in a user interface: The effect of experience and context of recall of the MacWrite screen. In D. N. Jones & R Einder (Eds.) People and Computers IV. Cambridge, UK: Cambridge University Press 1988, pp. 191-220.
Payne, S. J. Display-based action at the user interface. International Journal of Man-Machine Studies, 35, 1991, pp. 275-289.
Paap, K. R., & Renate J. R.-H. Design of Menus. In M. Helandar (Ed.) Handbook of Human-Computer Interaction. North-Holland 1983, pp. 205-233.
Reiman, J. F. Learning strategies and exploratory behavior of interactive computer users. Unpublished dissertation, University of Colorado, Department of Computer Science, 1994.