BEHAVIOUR & INFORMATION TECHNOLOGY, 1986, VOL. 5, NO. 1, 89-95.
(c) Copyright 1986 Taylor and Francis Ltd.
http://www.tandf.co.uk/journals
Short Paper
The role of graphics in item selection from menus
PAUL MUTER
Psychology Department, University of Toronto, Toronto, Ontario, M5S 3G3, Canada, muter@psych.utoronto.ca
and CANDACE MAYSON
Communications Research Centre, Department of communications, Ottawa, Canada
Abstract. The present experiment addressed the question of whether the addition of graphics to the alternatives on computer choice pages facilitates user performance. Twenty-one subjects made choices from pages that resembled videotex choice pages. One third of the time the alternatives were displayed in the usual way (Text-Only condition); one third of the time the alternatives were arranged in a nonlinear fashion and each alternative was accompanied by an illustration (Graphics condition); and one third of the time the alternatives were arranged in a nonlinear fashion but there were no illustrations (Control condition). Graphics had no effect on response time, but a reliable effect on accuracy: the error rate in the Graphics condition was half that in the Text-Only condition. Apparently, videotex information providers and other software designers would be well advised to consider adding simple graphics to alternatives on choice pages.
1. Introduction
In many computer applications, users proceed by choosing alternatives from menus presented on choice pages. Choice pages are particularly important in videotex, which is a technology permitting inexpensive, public, remote access to information by means of modified television sets, computers, and telecommunications links. Videotex systems and other computer systems have the capability to display graphics as well as alphanumerics, and the question arises: does the addition of graphics to the alternatives on computer choice pages improve the speed or accuracy of user performance?
There is evidence that pictures can sometimes be semantically processed faster than words. For example, deciding whether two items belong in the same category (e.g., animals) has been found to be faster for pictures than for words (Pellegrino et al. 1977). Categorizing a single picture is sometimes faster than categorizing a single word (Hogaboam and Pellegrino 1978, Pellegrino et al. 1977, Potter and Faulconer 1975). An indirect source of evidence supporting the same conclusion came from a study by Smith and Magee (1980). They found that categorization of a word is disrupted by the simultaneous presence of an incongruent picture, but that the categorization of a picture is not disrupted by the presence of an incongruent word. All of this evidence suggests that user performance on choice pages might be enhanced if an illustration were provided for each alternative.
On the other hand, Mills (1981) speculated that a graphic for each alternative on choice pages may actually impair the speed or accuracy of performance. The graphics may be redundant, and may distract and mislead the user.
For the present experiment, 24 special videotex pages were created. Each page had three forms: Text-Only, Graphics (with a simple graphic added for each alternative), and Control (to control for spatial arrangement). Across the experiment, each page appeared equally often in each form. Accuracy and response time were the dependent variables.
2. Method
2.1. Subjects
Twenty-one civil servants who had responded to requests for volunteers for videotex experiments served as subjects.
2.2. Apparatus and materials
Three forms each of 24 colour videotex choice pages were created employing word items in the experimental Telidon data-base. (Telidon is a form of videotex developed by the Department of Communications of Canada. See Bown et al. 1980.) These pages were created on a Norpak Information Preparation System. The 24 Text-Only pages were much like ordinary Telidon choice pages: they provided a simple, double-spaced linear arrangement of verbal items. The 24 Graphics pages contained the identical verbal alternatives, now distributed around the page in a nonlinear fashion, supplemented by appropriate simple graphics. One of the 24 Graphics pages is shown in figure 1. The 24 Control pages were identical to the Graphics pages except that the graphics were not included. The Control condition was included to ascertain the effect of the nonlinear arrangement, which may contribute to any differences between the Graphics condition and the Text-Only condition.
The graphics were created by C. M., and were based largely on her intuitions. She has no special training or experience in the graphic arts, and had no prior experience with the Norpak Information Preparation System. Some of the graphics were simplified realistic versions of objects; others were based on a more symbolic relationship with the represented concept.
One-third of the pages contained three or four alternatives; one-third contained five alternatives; and one-third contained six or seven alternatives. To enhance the word-graphic pairing, each pair was colour coded: for example, the word 'Food' and its accompanying graphic were both white, while the word 'Furniture' and its accompanying graphic were both yellow. In all, eight colours and several shades of gray were used.
A total of 288 'questions' were developed, 12 for each page. The questions for the Home and Family page (figure 1) are provided in table 1. The questions were so designed that no word in the question matched any word in the correct alternative, and no word in the question matched the most obvious name of the graphic corresponding to the correct alternative. For example, a kettle was used as the graphic for the item 'Appliances', so the word 'kettle' did not appear in any questions for which that alternative was the correct answer.
The videotex pages were photographed, mounted on 35 mm. colour slides, and presented by means of a carousel slide projector. The slides produced an image approximately 60 cm wide by 40 cm high. The subject sat approximately 2 m from the screen. The experimenter used a stopwatch to record the time taken to answer a sequence of eight questions.
2.3. Procedure
Each subject saw any particular page in one form only (Text-Only, Graphics, or Control) throughout the experiment. Each of the 24 pages was tested in the Text-Only condition for seven of the subjects, in the Graphics condition for seven of the subjects, and in the Control condition for the remaining seven subjects.
Each subject answered all 288 questions, twelve on each of the 24 pages. The experimental session was divided into 12 blocks of 24 questions each. Within a block, all 24 pages (eight in each condition) were tested once each. Within a block, there were eight trials in one condition, then eight in a second condition, then eight in, the third condition. The order of the three conditions within a block was determined randomly
for each block for each subject. The order of the questions (across blocks) was determined randomly for each subject.
Subjects were tested individually. First, they were given instructions, which gave equal emphasis to speed and accuracy. For each of the 12 blocks, the following procedure was then followed. The subject was given a sheet of paper with 24 questions printed on it. The questions were initially hidden by another piece of paper. When the experimenter and the subject were ready, the experimenter simultaneously presented the first slide, started the stopwatch, and said 'Go'. On the cue 'Go', the subject exposed and read the first question, looked for the correct answer on the screen, spoke his answer aloud (e.g., 'four'), pressed a button to get the next slide, exposed the next question, and so on. A dashed line on the question sheet and a blank slide indicated the end of the sequence of eight trials. The experimenter recorded the eight responses, and stopped the stopwatch immediately after the eighth response. The experimenter then informed the subject of his total time on that sequence, and of any errors. For errors, the experimenter redisplayed the slide in question, and indicated the correct answer.
The experimenter then rearranged the slide tray, started the second sequence of eight trials, and so on.
2.4. Design
The experiment took the form of a 3 x 12 randomized block factorial design. The factors were condition (Text-Only, Graphics, or Control) and Block (1-12). Each of the 21 subjects was tested in all 36 (3 x 12) cells.
3. Results
3.1. Accuracy
The mean error rate was clearly affected by the presence of graphics (see figure 2). The mean error rate was 2.8% in the Graphics condition, 5.6% in the Text-Only condition, and 4.6% in the Control condition. The analysis of variance showed an effect of Condition, F(2,40) = 9.69, P < 0.01. A Geisser-Greenhouse conservative F test (Kirk 1968, p. 142) also produced a reliable effect, P < 0.01. Tukey multiple comparison tests revealed a difference between the Graphics condition and the Text-Only condition, P < 0.01, a difference between the Graphics condition and the Control condition, P < 0.05, and no difference between the Control condition and the Text-Only condition. Sign tests (on the 21 subjects collapsed across blocks) yielded the same conclusions: Graphics better than Text-Only (17-1, P < 0.001), Graphics better than Control (17-3, P < 0.01), and no difference between Control and Text-Only (13-7, not significant).
The absence of a significant difference between the Text-Only and the Control conditions suggests that most or all of the difference between the Graphics condition and the Text-Only condition was owing to the graphics and not to the different spatial arrangement used in the Graphics condition.
As subjects were given feedback on every trial, there was ample opportunity for learning. In spite of this, there was no effect of Block on accuracy, F(11,220) = 1.02, nor did the effect of Condition depend on Block, F(22,440) < 1.
3.2. Response time
Mean response times were based on all sequences of responses, including those with one or more errors. (In the present experiment, response times are of interest primarily as an index of difficulty; hence the times of all responses are relevant.)
Graphics had no reliable effect on response time (see table 2). The mean response time in the Graphics condition was slightly lower than in the Text-Only and Control conditions, but the effect of Condition did not approach statistical significance, F(2,40) = 1.33.
The possible learning effects alluded to earlier did emerge in the response time data. Subjects became faster with practice: the effect of Block was reliable, F(11,220) = 49.6, P < 0.001. This effect did not depend on Condition, F(22,440) < 1.
4. Discussion and conclusion
The addition of graphics to the alternative on videotex choice pages halved the error rate. This effect was due to the graphics, and not to the different spatial arrangement used in the Graphics condition. The effect was obtained in spite of the fact that the creator of the graphics was not a professional artist.
It could be argued that the absolute magnitude of the effect was sufficiently small to be of no practical significance, but it is important to bear in mind that error rates in normal videotex usage may be substantially higher than in the present experiment. In more realistic situations, other investigators have found error rates ranging from 14 to 35% (Lee et al. 1984). With such error rates, the effect of graphics may be of considerable practical significance. Of course, it would probably be impossible to generate useful graphics for all alternatives of all choice pages, but it should be possible for the vast majority.
A limitation of the present study is that slides were used for presenting choice pages. On videotex, it takes time to display graphics, and any benefit of graphics on choice pages will eventually have been weighted against this cost. Because the amount of extra time required to display graphics will depend on several factors, and because the amount of extra time will probably diminish with advances in technology, we felt that it would be wise first to establish whether indeed graphics are beneficial given instantaneous presentation. Subsequent research can compare the benefits to the costs, given particular assumptions about the time required to display graphics.
It is unclear why graphics had an effect on accuracy and not speed. Perhaps subjects elected, consciously or unconsciously, to hold response time constant across conditions and let accuracy vary. Instructions emphasizing accuracy over speed might produce a facilitative effect of graphics on speed but not on accuracy.
Further research is needed to determine which aspect or aspects of the graphics accounted for the performance difference in the present experiment. One possibility is that the graphics provided more relevant information, and that this extra information not only did not lead the user down a garden path (Mills 1981), but was actually useful. For example, for the category clothing, in the Graphics condition extra information was provided in the form of an exemplar: a coat. Most graphics on choice pages will provide extra information of some sort.
A second possibility is that semantic features may be more certain to be processed for graphics than for words. It is fairly easy to fixate a word and not attend to its meaning; it may be more difficult to fixate a graphic and not attend to its meaning. The extraction of meaning from a graphic may be more automatic, more effortless, and more inevitable than the extraction of meaning from a word.
The present experiment suggests that the facilitative influence of graphics on videotex choice pages outweighs any distracting or misleading influence. Videotex information providers and other computer software designers would be well advised to consider adding a simple graphic to each alternative on choice pages.
Acknowledgments
We thank Robert L. Duchnicky, Lochlan E. Magee, and Nancy L. Williamson for helpful comments. This work was supported in part by Research Grant U0149 from the Natural Sciences and Engineering Research Council of Canada to the first author.
References
BOWN, H. G., O'BRIEN, C. D., SAWCHUCK, W., STOREY, J. R., and TREURNIET, W. C., 1980, Telidon Videotex and user-related issues. In Processing of Visible Language 2, edited by P. A. Kolers, M. E. Wrolstad, and H. E. Bouma (New York: Plenum). ,
HOGABOAM, T. W., and PELLEGRINO, J. W., 1978, Hunting for individual differences in cognitive processes: Verbal ability and semantic processing of pictures and words. Memory and Cognition, 6, 189-193.
KIRK, R. E., 1968. Experimental Design: Procedures for the Behavioral Sciences (Belmont, CA: Wadsworth).
LEE, E., WHALEN, T., McEWEN, S., and LATREMOUILLE, S., 1984, Optimizing the design of menu pages for information retrieval. Ergonomics, 27, 1051-1069.
MILLS, M. 1., 1981, A study of the human response to pictorial representation on Telidon. In Telidon Behavioral Research III, edited by D. Phillips (Ottawa: Department of Communications).
PELLEGRINO, J. W., ROSINSKI, R. R., CHIESI, H. L., and SIEGEL, A., 1977, Picture-word differences in decision latency: An analysis of single and dual memory models. Memory and Cognition, 5, 383-396.
POTTER, M. C., and FAULCONER, B. A., 1975, Time to understand pictures and words. Nature, 253, 437-438.
SMITH, M. C., and MAGEE, L. E., 1980, Tracing the time course of picture-word processing. Journal of Experimental Psychology: General, 4, 373-392.