Although it is often the last thing to be considered by the software
designer, the appearance and intelligibility of the screen display is the first
thing to be assessed by the user. "What is it telling me? What does it mean?
What does it expect me to do?" are all questions that the user is likely to ask
on first encountering the system. One may brush off these questions and say
that after the user has worked with the system long enough, everything will
make sense. But this is not always true. Experienced users have even more
telling questions: "Why is that phrase used when it's not performing that
function at all? Why is the screen organized this way? Why do I keeping
making the same error over and over again?" In general, users can adapt to a
less than optimal human/computer interface, but it is always at the expense of
training time and cognitive effort. When these are in short supply, interface
design is critical.
Screen display and the formatting of a menu frame is an art. The phrasing of stems, leaves and other information in the menu frame is a wordsmithing activity that without proper attention can lead to many problems. Fortunately, a growing body of literature in cognitive psychology and more directly in ergonomic research is helping to define the factors of menu design and to set guidelines for "do's and don't's" in designing menus. Many of these guidelines and the research supporting them will be reviewed in this chapter.
6.1 Formatting the Menu Frame
The purpose of a well-designed menu frame is to convey information to the user in the most efficient manner possible. It must be remembered that users read and search for information with expectations. They do not necessarily read the menu frame word-by-word from the beginning. Rather they scan it and are highly influenced by its organization and layout. It follows that organization and layout should facilitate the search process rather than work against it.
Much has been written about screen display in computer systems (e.g., Galitz, 1981; Cropper & Evans, 1968; Engel & Granda, 1975; Foley, Wallace, & Chan, 1984; Mitchell, 1983; Peterson, 1979; Shurtleff, 1980; Tullis, 1981, 1983). Many of these references serve as excellent guides for designing menus.
In terms of menu display, the designer must decide what information to display, in what form, and where. As seen in Chapter 2, menus include information about the context, the stem and leaves of the choice options, and response and feedback information.
6.1.1 Amount of Information per Screen. The first decision that must be made has to do with how much information should be displayed. Most guidelines in the literature suggest that only information essential to the user's needs should be displayed. Irrelevant information clutters up the screen and makes it more difficult and time consuming for the user to locate the option that he desires. The problem is knowing what the user needs and determining what is irrelevant vs. what is necessary.
The designer must contend with a set of tradeoffs:
Amount of Information vs. Scan/Reading Time. The more information, the more difficult and time consuming it is to search for target information and to read the information. Research in human information processing indicates the obvious: it takes longer to search for a word as the amount of surrounding information increases. When time is a critical factor in performance, brevity of the menu is essential in reducing human processing time. Furthermore, readability is affected by the density of text on the screen. It is suggested that no more than 30% of the available character spaces on the screen should be used (Danchak, 1976; Poulton & Brown, 1968; Ramsey & Atwood, 1979).
Level of Experience vs. Need for Explanation. The less familiar the user is with the system, the more detailed the explanations need to be of menu options. Menus need to be information rich for novices. The implication is that performance will be slow, but this is generally not a major concern for novices and casual users. On the other hand, menus may be information lean for experienced users. The terms serve only as cues to jog the memory of the user and to indicate the allowable options at the current state. Furthermore, the speed of scanning and selection is a function of the training of the user. In terms of the theory of cognitive control, the experienced user has acquired sufficient knowledge about the system so that information flow from the computer to the user is no longer necessary.
Amount of Information per Screen vs. Number of Screens. The less information presented per screenful, the more screens that are necessary. One of the major concepts in menu systems is to subdivide information and user control into a network of screens. Typically, when all of the information and options do not fit on one screen, the designer divides the menu into two or more screens. Ultimately, this leads to the depth vs. breadth tradeoff which will be discussed in a later chapter. At this point it is sufficient to say that when screens must be subdivided, it adds to the cognitive complexity of the system. Performance of both novice users and experienced users is lowered due to the added steps necessary to access additional screens. This problem is alleviated in part by larger screens or multiple screens displaying longer menus, but then one must contend again with the first trade-off having to do with scanning and reading time.
6.1.2 Focusing Attention on the Menu. Principles of good screen layout may be derived from theories of perception in Gestalt psychology and more recently cognitive psychology. One of the basic processes in perception is the segmentation of the field into figure and ground. The figure contains those objects that are the focus of attention; whereas, the background is amorphous and undifferentiated. Studies in visual perception, indicate that the figure-ground distinction is the first and most fundamental aspect to be perceived in a display. However, the figure-ground distinction may vary with the stimulus. At one extreme, figure-ground may be a potent aspect of the display that forces the viewer to perceive one area as figure and the rest as ground. On the other hand, the distinction may be weak and driven primarily by the current need state of the viewer. For some displays, we are quite capable of totally reversing figure and ground or of refocusing attention to generate new segmentations of the field.
When a user is accessing a menu, the menu becomes the figure and the rest of the screen, which may be displaying previous output becomes the background. Although the focus of attention is on the foreground, the background can either serve to enhance this focus or to distract from it. Figure 6.1 illustrates three levels of figure-ground distraction for a pull down menu.
When a gray background is used as in the top panel in Figure 6.1, figure-ground segmentation is maximized; when other information clutters the background as in the middle panel, segmentation is reduced; and when the information in the background is similar to the figure as in the bottom panel, segmentation is minimal. Systems supporting the cluttered desktop metaphor can suffer from figure-ground ambiguity. Fortunately, the figure-ground distinction can be enhanced by the proper use of graphic devices such as borders, highlighting, overlapping, graying out, colors, etc.
6.1.3 Perceptual Grouping. Perceptual grouping is an important process that aids in interpreting what something is and its relationship to other objects in the field. Characteristics of the objects themselves suggest an organization of the field of view. We perceive objects that are similar in appearance or close in proximity as belonging together. Such perceptual grouping affects subsequent processing of the information in terms of the meaningfulness and expectations regarding functions. These concepts may be effectively used in menu design; or if not attended to, perceptual effects can be quite detrimental.
Studies have found that spatial proximity of options in a menu affect the grouping of items in the long term memory of the user. Items may be "chunked" together. Chunking is the cognitive process of storing/retrieving a set of items as a unit. For example, single letters may be chunked as a word, and words may be chunked as a sentence. Card (1982) used a recall test in order to assess the chunking process in menu selection. After a series of search trials using an 18 item menu with boxes containing 2 to 4 items, subjects were asked to write down all of the items that they could remember. When items are recalled from memory, items that are in the same chunk tend to be recalled together with a short time interval between recall. Items in different chunks tend to be separated by a longer time interval. Card found that items next to each other or in the same box in the list tended to be clustered together in recall.
Similarity of appearance also leads to perceptual grouping. Items may be similar in phrasing or in graphic shape. One menu uses the items "SAVE" and "SAVE AS ..." The first option saves the current memory buffer by writing over a file that had been previously opened thus destroying the old contents. The second option saves the current memory buffer by opening a new file and writing to it while doing nothing on the old file. The similar phrasing of the items helps the user to group them together as having a similar function with respect to writing the memory buffer to a file. It may be detrimental, however, by obscuring the effect on the old file. The user often has to learn such subtle distinctions the hard way.
Spacing of alternatives can lead to perceptual grouping and has a strong effect on visual search. In their study of menu selection using large menus of 64 items, Parkinson, Sisson, and Snowberry (1985) compared menus having either no spacing between category groups or having one additional space between groups. Menus with the additional space were searched in 25% less time. It is interesting to note that this difference was not substantially reduced with practice. Perceptual grouping induced by spacing seems to have several beneficial effects. First, it appears that grouping helps users to recognize the type of organization and structure of the list (e.g., organization by alphabetical or natural categories). Second, perceptual grouping allows the user to mark and locate the beginning and ending of groups when visually scanning for target items.
Menu systems that display icons can also make good use of perceptual grouping. Scanning and perceptual grouping is often enhanced by graphic images. Figure 6.3 shows the effect of adding icons. In the top panel the names of 12 files are displayed. The bottom panel adds icons indicating the type of file. It is expected that search time for selecting files with the icons will be faster than without the icons even though files in both cases are clustered by type. Icons serve to add perceptual features that aid in grouping and search processes.
6.1.4 Menu Context. Screen design involves setting a context in which the items are perceived and interpreted. Two types of context must be considered. A "sensory" context is set by the background and explicit context information displayed. The recognition and meaning of items in the list are affected by this surrounding information. A "nonsensory" context is set by previous events remembered by the user and by his current task and goal states. In a bibliographic search, for example, sensory context may be established by displaying that the user has located a set of 10 bibliographic citations by a title search and that menu options refer to this set. The nonsensory context of the user may be that he is searching for a particular book and he has not been able to locate it by an author search.
The interpretation of menu items and their consequences are highly influenced by the context. Terms such as "list", "print", "kill", "find", "change", "load" are highly context dependent. "List" may refer to a command to display a listing of a program or it may refer to a list of words. The problem is that words are often ambiguous outside of a clear context. For terms such as these, the user must disambiguate the meaning on the basis of their frequency of use and the recency with which a particular meaning was evoked (Anderson, 1988). The point is that menu systems must provide sufficient context for the users to easily disambiguate the meaning of the menu items.
Context that relies on nonsensory mechanisms is often unpredictable. Consequently, guidelines suggest that every screen explicitly display the user's last selection which led to the current screen or state of the system in order to aid in a step-wise refinement of the interpretation of items. For example, a search through a hierarchical database of employees in an organization may retain the context by adding to the title after each menu selection in the following manner:
6.1.5 Ordering of Menu Items. What is the best order of the items in a menu? There are three major purposes behind the ordering of items. The first is to facilitate search for an item in a list. When items are organized along some characteristic or dimension, the user's search strategy can use that information to locate items faster. Alphabetic ordering helps the user know approximately where to look in a list for a word. The second is to convey additional information about the structure and relationships among the items. This information may aid in helping make choices by knowing the similarities and dissimilarities among items. The third is merely to agree with the user's inferred ordering so as not to disrupt cognitive processing by creating dissonance between the user's knowledge base and the system's knowledge base. Items may be organized according to a number of characteristics.
Random Order. As a baseline, we will consider the case of no perceived organization. For all intents and purposes, the list appears to be random. There may be some underlying software rationale for the order depending on system constraints, but the reason for this order is not apparent to the user and has nothing to do directly with the human/computer interface. For example, the order of desk accessories on the Apple Macintosh is determined by tool priorities. The user can generate no algorithm to facilitate search other than a rote memorization of item position. However, it should be mentioned that frequent users of systems quickly acquire a knowledge of the spatial location of items in a randomly ordered (but constant) list. Card (1982) and others have shown that differences in performance due to the organization of the items are reduced or eliminated entirely with practice. Subjects apparently learn the organization, albeit random, and know where to look for items in the list. Consequently, the detrimental effects of random order, are observed when the user has not yet gained extended familiarity with the items or when a different random order occurs every time the menu is viewed.
Alphabetic Ordering. An alphabetic ordering of alternatives may be used when there is a particular keyword, command name, or title that can be meaningfully alphabetized and scanned.
Numeric Ordering. Numeric ordering can be used for the selection of alternatives that are associated with numbers such as type size, baud rate, and number of copies.
Chronologic Ordering. Items that refer to dates and times may be ordered from oldest to newest. For example, articles in a news service, entries on an electronic bulletin board, and the list of months may be chronologically ordered.
Frequency of Use. Items may be ordered by frequency of use with often used options listed first and infrequent options listed last. Frequency of use may be determined a priori by a logical analysis of the typical tasks or a posteriori from observed frequency of use.
Sequential Processing Order. Items may be listed according to their inferred order in a process or according to a cognitive ordering of items. The items "OPEN FILE", "ADD RECORD","CHANGE RECORD","DELETE RECORD","CLOSE FILE","QUIT" imply a logical ordering of steps. The task analysis not only dictates the functions that are required but also their usual order of operation. Cognitive order refers to the order that users would list items. For example, in a news network the items "WORLD", "NATIONAL", "STATE", "LOCAL" are ordered in terms of an underlying dimension of scope.
Semantic Similarity Ordering. Items may be ordered in terms of some semantic dimension such as impact, reversibility, potency, etc. Items that are most similar appear close to each other in the list. For example, a menu of style of type may be ordered in terms of emphasis: Plain, Underlined, Italicized, Bold. Parkinson et al. (1985) give an example of semantic ordering in which words sharing the most semantic features are presented adjacent to each other. For the category of topography the order of the following names would be: Mojave, Sahara, Everest, Matterhorn, Atlantic, Pacific, Huron, Erie. The first four are land areas consisting of two deserts and two mountains. The last four are bodies of water consisting of two oceans and two lakes.
Categorical Grouping. Items may be categorized and then ordered within groups according to the characteristics listed above. Categorization operates on the similarities and dissimilarities of the items. Categorization allows for hierarchical search and decision processes on the part of the user as discussed in Chapter 3. Designers must also consider that categories and items within categories must themselves be ordered.
The order of items in a menu should be consistent with the expectations and cognitive order held by the user. When the menu order is incongruent with the cognitive order, it makes it difficult for the user to locate items. Alphabetic order clearly reduces search time in many listed, but imagine the confusion in searching an alphabetized list of months, days, or numerals:
The effects of menu organization on search time and on learning have been studied by a number of researchers. The results have been mixed due to additional factors such as the number of items in the list, type of search task, and experience of the subjects.
One clear finding, however, is that alphabetic and categorical ordering are superior to random ordering. What is not clear is the relative advantage of alphabetic vs. categorical ordering. The Card (1982) found that for menus of 18 items in a visual search task, there was an advantage of alphabetic ordering over a condition with categorical ordering and a condition with random ordering. On the other hand, for a 16 item list, Liebelt, McDonald, Stone, & Karat (1982) found a slight advantage for categorical organization over alphabetical and random orderings. However, the difference was not statistically reliable.
Often items are displayed according to several ordering schemes at once. Items may be grouped and then alphabetized or ranked in some way within groups. One study illustrates the importance of forming major groupings of items. McDonald, Stone, and Liebelt (1983) formed categorical groups with either similarity, alphabetical, or random ordering within groups. Menus consisted of 64 items arranged in 4 columns of 16 items. In three conditions, the four lists were categorized by food, animals, minerals, and cities. In Condition CC, the four lists were further ordered by obtaining similarity ratings and ordering the terms using multidimensional scaling and hierarchical clustering techniques. In Condition CA, the four lists were alphabetized, and in Condition CR the items were random. Two other conditions mixed the categories. Condition A displayed a complete alphabetic ordering of the 64 items and Condition R a random order of all items. A second factor in the experiment was of interest. Half of the subjects were shown the explicit word to search for and the other half were shown a single-line dictionary definition. The task was to locate the item on the display screen and press the key corresponding to the identification letters displayed immediately to its left. The identification letters were randomly assigned to items on each trial to prevent memory association that would circumvent the search process. Subjects were obtained from a temporary employment agency for secretarial help. Response time was measured over a series of 5 blocks of 64 trials. Figure 6.3 shows the results.
Overall subjects that were given definitions took longer to respond than subjects given the explicit word. Organization of the list had a significant effect on response time. For the explicit word, random ordering was much worse than any other conditions. Subjects were able to perform quickly in all of the categorical conditions and the alphabetized list. It would seem that once the subject identified the category and scanned the column of 16 items, order within the column was not an important factor. On the other hand, for the definitions, the categorical ordering (CC) was slightly superior to the alphabetized or randomized categories. The alphabetic and random lists led to the longest response times with the alphabetic condition only slightly superior to the random order. This result is important since users may or may not be searching for an exact word match. When they are searching for an item to satisfy a definition (or a command to perform a function), categorical ordering is important.
One would expect that the ordering of the list would become increasingly important as the length of the list gets longer. Perlman (1984) found this to be true. He varied the length of the list to have 5, 10, 15, or 20 items. The items were in one case the numbers 1 to 20 and in another case words beginning with the letters "a" to "t". The lists were either sorted or random. Four groups of 8 subjects searched either sorted words, sorted numbers, random words, or random numbers. Each subject searched all different list lengths. The target number or word was shown on the screen and the subject then searched for the item and then pressed the key corresponding to the item. Figure 6.4 shows the results. Response time increased with the length of the list. Response times for sorted items were shorter than for random items and for numbers than words. What is interesting about these results is the interaction between the list length and the type of list searched. When lists are sorted, the increase due to list length is less that when lists are random. This suggests that when menus include a lot of items, it is extremely important that the designer consider organization of the list in order to reduce search time.
It is not clear, however, whether alphabetic or semantic organization results in superior performance. Parkinson et al. (1985) found little difference in search time between alphabetic and semantic organization within categories of words when the categories were sufficiently separated by spacing. This is an important finding for designers since alphabetic ordering is easy to achieve while semantic ordering requires extensive pretesting. On the other hand it should be pointed out that if the designer wishes to convey a mental model of the system, semantic ordering may help to achieve this objective. For example, in command specification Parkinson et al. point out that given the task of executing a set of functions using a menu of commands, naive users would probably learn the system more efficiently if the commands for similar functions were grouped together in the display.
6.1.6 Orientation of the List. Generally, lists of options have been displayed in a vertical list (columns) rather than a horizontal one (rows). The question is whether the vertical or tabular listing is indeed superior to a single horizontal line listing. Single line listing saves space on the screen. In cases where the user input is an associated letter or number, one might expect the tabular arrangement to be superior. In the case of touch screen or pointing device, it is not clear. In cases where the single line listing is used, items are generally very brief and often single character options as in the following menu from an electronic bulletin board:
Option (A/T/D/E/T)?
In a study of menu preference by Norman (1987), computer science majors indicated a preference for single line listings of commands as opposed to a tabular listing. On the other hand, non-computer science students preferred a tabular listing.
In a study on the organization of items in large menus of 64 items, Parkinson, Sisson, and Snowberry (1985) compared category groupings arranged by column versus row. Menus with categories arranged in columns were searched in approximately 25% less time (1 sec faster) than menus with categories arranged in rows. From these results, visual search for targets appears to be substantially facilitated by column presentation. However, it must be noted that this finding is currently limited to large menus and to the particular configuration chosen by Parkinson et al. In their experiment all menus consisted of 64 words listed in four columns of 16. Eight words were chosen from each of eight natural categories and arranged in either columns (two categories per column) or rows (2 adjacent rows per category). This arrangement implies that given a target in the menu organized by column, subjects first located the correct category in a 2 x 4 array of categories and then scanned down to find the target within the category to locate the two digit response code for the item. In the case of menus organized by row, subjects first located the category in a vertical array of 8 categories, and then scanned 2 rows of 4 items to find the target. It is likely that having categories split on two rows slowed down the search due to the time to jump from line to line as well as from column to column when searching for a word within a category.
This interpretation is further suggested by the results from two alphabetized menus also used in the same study. In one menu, words were alphabetized from top-to-bottom by column and in the other from left-to-right by row. The difference in search time between these two menus was of approximately the same size as for categorical menus. Furthermore, Parkinson et al. note that the row versus column effects may have been confounded by the fact that more space separated words in adjacent columns than in adjacent rows. All this strongly suggests that it was not simply row versus column organization that led to the effect, but that the visual search distance and the number of jumps to reposition each search within categories accounted for the differences.
Although it is not yet clear that row arrangement is itself detrimental, it is clear that large differences in performance can be due to column versus row arrangement of menus. At present the results suggest that designers should attempt to arrange items so as to minimize visual search distance and the need to jump from one location to another when searching within the same category.
The issue of row versus column presentation is important in systems with pull down menus. Such systems generally display a menu bar which is a horizontal menu listing of vertical pull-down menus. In essence they alternate horizontal and vertical arrays. This idea can be extended as an effective way of handling successive depth of a menu while maintaining context. An example of this is shown in Figure 6.5 for a hierarchical tree of languages having a maximum depth of 5. Menu nodes shown in reverse video indicate the selected path.
At present little is known about the use of alternating menus. However, one would expect that with the use of a pointing device such as a mouse, the selection process is much like tracing a route in a maze. As such it will be highly spatial in character and affected by psychomotor processes.
6.1.7 Fixed vs. Variable Format. The literature suggests that the organization of information elements on the screen should be standardized in order to make effective use of spatial context. The user acquires an expectation of where to look to see particular information. When information could appear at different locations, the user must search for it. Teitelbaum and Granda (1983) compared fixed vs. variable formatting of the information. When the location of information was inconsistent, the time that it took to answer questions about information on the screen was longer than when the information was in a constant location. Moreover, when information was in a constant location, response times improved with practice indicating that users were acquiring expectations as to where information was located. Response times did not improve when information was presented in random locations.
The importance of consistency in screen and menu design cannot be over-emphasized. It is imperative that users develop expectations about the location of information to avoid having to repeatedly search for information. Furthermore, inconsistency probably fosters a level of frustration and confusion in the user that may generalize to the other menu selection processes.
6.2 Writing the Menu
Communication is an art. In human/computer interaction, the art is communicating through an interactive media about the control of that media. Fortunately, art is guided by principles. Menus must convey information to the user about (a) the nature of the choice (What is being accomplished in making this choice? What is the purpose of this choice?) and (b) the nature of the alternatives (What will happen if this alternative is selected?).
On the one hand, one would like to convey as much information as possible, explaining all the options in great detail. On the other hand, brevity is important due to screen limitations and the amount of information that the user wants or desires to read. Consequently, there is a trade-off between thoroughness and brevity. And the trade-off is to be moderated by the experience of the user. Many systems are developing the concept of dynamic help in which menus are highly descriptive for novice users, but as the user gains experience, the menus become brief pneumonic reminders of the alternatives.
6.2.1 Titling the Frame. Titles of menus should be more than place holders. They should be descriptive of the list of options and the context. The title, "MAIN MENU OPTIONS," conveys little information except that the user is probably at the top of the menu tree.
When a title is superfluous, then it may be omitted. To the extent that the list of options define the context, the title is redundant. This is particularly true for experienced users. Titles merely clutter the screen. It is an empirical question as to whether titles help and when they have served their usefulness and should be dropped. One possibility is to design a system in which the user may hide titles or reveal them as necessary.
The title, however, can be a potent stimulus in defining the context and increasing comprehension of the menu. An experiment by Bransford and Johnson (1972) illustrates this fact. They assessed comprehension and recall of a passage describing in detail the steps that one must go through in order to accomplish a certain function. In one case, subjects were given the title of the passage, "Washing Clothes" either before or after reading it. In a control condition, no title was given. Subjects given the title before reading the passage reported higher comprehension and recalled about twice as many items from the passage than other subjects. Subjects given the title after reading the passage did no better than the control subjects given no title. In the same way, menu titles have the potential of increasing the comprehensibility of menus and facilitating choice.
We may think of a schema as a conceptual framework that provides "slots" to be filled in with specific information. Menus are themselves schemata of the form: "Of the following options, select one: (a) ... (b) ... (c) ..." Users assimilate the particular options into this framework. Menus are also a part of a larger schema in which they are understood. Without establishing the context of the choice by instantiating a schema, comprehension of the menu frame may be severely limited. Consider, for example, the following menu:
Select function:
Open
Close
Activate top element
Activate bottom element
Clean
The purpose of this choice and the function of the alternatives is not at all apparent. However, if you are told that the schema has to do with the operation of an oven, the menu is comprehensible and easy to remember.
6.2.2 Wording the Alternatives. Alternatives may be single words, brief phrases, or longer descriptions of options. It is generally recommended that when brief phrases are to be used that they are verb phrases that are written in parallel construction (Shneiderman, 1987). For example, rather than phrasing the items as "Print," "Execute a program," and "Disk eject," they should be "Print a file," "Execute a program," and "Eject a disk." In general, it is recommended not to change the construction of a phrase within the same menu frame. In addition, Shneiderman lists the following guidelines:
(1) Use familiar and consistent terminology.
(2) Ensure that items are distinct from one another.
(3) Use consistent and concise phrasing.
(4) Bring the keyword to the left.
An important aspect of wording is not only that it conveys information, but that the alternatives are interpreted as distinctly different. Distinctiveness refers to the semantic aspects of an alternative that enhances its difference from other alternatives in the set. The object is to word items so as to emphasize differences rather than the commonality of function. Consider, for example, a menu that contains the items (1) Services for Professionals, (2) Home Services, (3) Business and Financial, (4) Personal Computing. Items 1 and 2 are not very distinct in that they both emphasize services. Furthermore, the distinction of what is meant by "professional" versus "home" is not clear. Items 1 and 3 lack distinctiveness altogether. Item 1 could be a subset of 3 or vice versa. Items 2 and 4 lack distinctiveness since "home" and "personal" share common aspects. The terms "services" and "computing" are not distinct since any services provided on a time sharing system must involve computing. Such a menu poses a dilemma for the novice user who, for example, wants to balance his checkbook. The question is how to design a menu system so as to maximally enhance the distinctiveness of items and minimize their confusability.
In a study by Schwartz and Norman (1986) novice computer users searched either (a) the original menu of a commercial time sharing system in which items were not worded in a holistic manner or (b) a modified menu of the system in which items were holistically worded. They hypothesized that the effect of item distinctiveness would vary with the type of search required. In explicit search problems, subjects searched for the exact wording of targets (e.g., "Search for 'Backgammon'"). In information find problems, subjects searched for menu items that provided information about subject matter topics (e.g., "Find information concerning how to balance your checkbook"). These two types of menu problems correspond to the two general uses of menu selection systems. In general, it was found that subjects using the modified menu (a) took less time per problem; (b) found targets in a more direct path; and (c) gave up on fewer problems than subjects using the original menu. Although information find problems took longer and required a greater number of frames to be traversed than explicit search problems, there was no interaction with distinctiveness. Distinctiveness had a similar effect on search processes for both explicit and information find problems.
Schwartz and Norman (1987) suggest that distinctiveness may be controlled to some extent by the way in which choices are clustered in the menu tree. Items may be shifted from one menu frame to another to enhance the differences among options within a particular frame. When the structure of the menu is fixed, the wording of alternatives may be altered to enhance differences. They suggest the following strategies for improving distinctiveness:
(1) Induce holistic, intact, and integral processing of the alternatives. Holistic alternatives are not likely to be decomposed into aspects that may be shared among the items. Options should appear as different as apples and oranges even though they may share many common aspects (e.g., both fruits, both round, both about the same size, etc).
(2) Maximize aspect redundancy and intercorrelations within alternatives and minimize them between alternatives. For example, the position and color of traffic lights are perfectly correlated and redundant. Consequently, both aspects can be used to discriminate between the alternatives of stopping and going.
(3) Present all alternatives simultaneously rather than one by one or page by page. When alternatives are processed sequentially (as in a slow display), a subjective emphasis on the common aspects of the alternatives may occur. Users look for shared aspects that lead from one alternative to another rather than contrasting the aspects when all are seen simultaneously.
(4) Explicitly define the universe of options so that each alternative may be understood in this context. If the alternatives are Diet-Rite, Diet Coke, 7-Up, and Fruit Juice, the universe of options is unclear. With one generic drink (Fruit Juice) and three specific drinks, one might over generalize to the universe of liquids as the reference, while others might restrict the universe to drinks that refresh. Avoiding this type of ambiguity is important in effective menu design.
(5) Ensure that the variability among the alternatives is high in terms of the rationale for the choice. For example, if the universe of choice is among soft drinks and the alternatives are Diet-Rite, Diet Coke, and Diet Pespi, there is little perceived variability. However, if the universe of choice is limited to dietetic soft drinks, then the perceived variability is high.
6.2.3 Graphic Alternatives
An increasing number of software applications are using graphic symbols called "icons" to represent objects and operations. Graphic symbols have long been popular in cartography. Graphic symbols can be visually more distinct from one another than words, and it is easier to spot a graphic symbol on a map than it is to locate a word. Similarly in human/computer interaction graphic symbols conserve space and yet are more distinct on busy display screens. However, there are a number of questions about the relationship between graphic symbols and the objects and commands that they represent when used in menu selection. Hemenway (1982) distinguishes between icons that depict (a) the objects on which the commands operate, (b) the operations or operators themselves, and (c) the operations on the objects. Figure 6.6 gives examples of these three types.
The first panel shows icons that represent objects operated on by some command. Since only the operand is depicted, it is not always clear what the operation is. Often, as in the case of the camera, it is simply an on/off function. In other cases, users must learn that more complex commands are implied. The middle panel shows icons that depict operations. The arrow has the conventional meaning of movement or direction and the "X" has the meaning of deletion or negation. Both meanings more or less represent the intended command. On the other hand, the operations represented by the paint brush and scissors are implied by what the tools do. Finally, the icons in the last panel combine both objects and operators. The first icon combines the box (image area) and the "X" (delete) to imply delete image area. The second and third icons show objects before and after the operation. The operation is inferred by the difference between the two. The last icon (fill in) shows a snapshot of the operation in progress. Animation completes richness of meaning and provides yet a stronger basis for implying the operation by showing the operation in action.
Hemenway proposes a simple model for the interpretability and effectiveness of icons. When a new icon is encountered its interpretability depends on (a) its comprehension (e.g., the ability to discover what the icon depicts) and (b) its effectiveness as a retrieval cue (e.g., the ability to form a link between what is depicted and the corresponding command). For experienced users, the effectiveness of an icon ability depends on (a) the ability to recognize what the icon depicts and (b) the ability to retrieve the link between the icon and the command from memory. It is predicted that highly familiar, conventional symbols will be easier to learn and recall than obscure icons that lack distinguishing features. Furthermore, icons that directly depict objects and operations are predicted to be more effective than ones that resort to analogy and convoluted links.
Often the organization and inter-relation among commands can be expressed in shared features of the icons. The icon for a tool to generate rectangles and filled rectangles shares the shape but differs in shading. Similarly icons for drawing lines may vary in thickness corresponding to the width of the lines created (see Figure 6.7). Hemenway suggests that when elements are shared across icons the user has less to learn since familiarity on the shared feature transfers to the new icon. Furthermore, repetition of the elements can make it easier for users to link the elements to aspects of the command and to recognize how icons vary in terms of their features.
An empirical study was conducted by Rogers (1987) to test the prediction that direct linking between icons and commands results in better interpretability of icons. Four types of icon sets for file manipulation, text editing, and printing were designed that were either (a) abstract symbols, (b) concrete analogies, (c) concrete objects, or (d) concrete objects operated on with abstract symbols. Abstract symbols and concrete analogies constitute indirect mapping; whereas, concrete objects provide a direct mapping. For comparison a fifth group was composed of verbal names consisting of high imagery words. The icon set with the most direct mapping (concrete objects operated on with abstract symbols) resulted in the fewest accesses to the help facility and least number of errors. However, there was also an interaction between the type of command and the form of representation. File operations such as "to save" appeared to be easier to use and recall when the commands were verbal . Text editing commands such as "insert a space" were more discriminable as icons.
These results suggest that designers should not only use icons that have a direct linking but that they should carefully analyze the nature of the command and its representation. Icons are perhaps best used to represent the operations of graphic tools and objects; whereas, verbal labels are best for formal commands.
A final possibility is to add graphic and verbal labels. It has been speculated that graphic symbols added to menu items could impair the speed and accuracy of performance (Mills, 1981). Studies on categorization provide indirect evidence about how pictures and verbal labels interact. Smith and Magee (1981), for example, found that when subjects were asked to categorize words, the simultaneous presentation of an incongruent picture disrupted processing (e.g., a picture of a car shown with the word "table"). However, when subjects were asked to categorize pictures, the presentation of an incongruent word did not disrupt processing (e.g., the word "car" shown with a picture of a table). Furthermore, it has been shown that the categorization of pictures tends to be faster than of words (Pellegrino, Rosinski, Chiesi, & Siegel, 1977). Such evidence suggests that graphic symbols may facilitate menu selection or at least not degrade performance.
To test the effect of adding graphic symbols, Muter and Mayson (1986) compared three conditions using a videotext data-base: (a) text-only menus in which verbal items were displayed in a double-spaced linear list, (b) text plus graphic menus in which verbal items were displayed with graphic symbols of the items and distributed in a nonlinear fashion on the page, and (c) control menus in which verbal items were distributed in a nonlinear fashion on the page. To enhance the pairing of graphics and labels both were shown in the same color. Subjects were shown menu pages and asked a series of 12 questions (e.g., "Where can you buy a low priced bedroom suite?"; "Where can you locate a handsaw?"). Instructions emphasized both speed and accuracy. It was found that the text plus graphics condition resulted in significantly fewer errors (2.8%) than either the text-only (5.6%) or the control (4.6%) conditions. Overall response times did not differ. Thus, the addition of graphics halved the error rate without significantly increasing the processing time.
On the other hand, a study by Wandmacher and Müller (1987) found no difference in error rate between graphic menus and word menus, but did find a difference in reaction time in both a search-and-select paradigm and a recognition paradigm. On each trial of the search-and-select experiment subjects were given a task description (e.g., "You want to print a document") and a menu of either word commands or icons was then displayed (See Figure 6.8). Subjects made their selections by entering a number corresponding to the item. The order of the menu items was randomized on each trial to so that subjects could not memorize the numbers. On each trial of the recognition experiment only one item was displayed and subjects had to respond by pressing a prelearned number. In both the search-and-select and the recognition paradigm, response time were faster for icons than for words. These authors suggest that the gap between the meaning of an expression and its form is smaller for icons than for word commands. Hutchins, Holland, and Norman (1986) refer in general to this gap as the "articulatory distance." Icons may facilitate menu selection because they provide a more direct access to the meaning of the functions involved. Less time is required to translate an intention into an actual selection.
The use of graphics in menu selection appears to be a promising design feature. However, the effective use of graphic symbols requires thoughtful consideration. Arend, Muthig, and Wandmacher (1987) point out that it is not sufficient to just recommend the use of icons. In the same way that Schwartz and Norman (1986) demonstrated that discriminability has an effect on word menus, distinctiveness may have an effect on graphic menus. Arend et al. suggest that the "global superiority" effect in visual perception may affect the degree to which icons facilitate menu selection. The global superiority effect basically is that global features of figures (e.g., shape, color, size) can be selected and responded to considerably faster than local features (e.g., lines and structures within a figure).
Arend et al. contrasted the effects of (a) icon distinctiveness defined in terms of global superiority and (b) icon representativeness defined in terms of articulatory distance. Distinctive icons were abstract and emphasized one global feature. Representational icons included local features to make the meaning of the icon more apparent. A search-and-select paradigm was used in which subjects were given task descriptions and then had to select the appropriate menu item. Either a word command menu, a distinctive icon menu, or a representational icon menu was presented to different groups of subjects. In addition menus were constructed of either 6 or 12 items. Subjects made selections by pressing the key corresponding to the position of the menu item on either a 3 x 2 or 3 x 4 key matrix depending the menu size condition. Figure 6.9 shows the menu sets used in the 12 item condition.
The results clearly favored the distinctive icon menu over either the word menu or the representational menu. Figure 6.10 shows the mean response time for each menu. Over all conditions, the distinctive icon menu resulted in response times that were nearly twice as fast as for word or representational icon menus. Consequently, the type of icon used has a great effect on performance. Distinct, abstract icons which have high global superiority result in the best performance. Representational icons that reduce the articulatory distance, however, do no better than word menus. Apparently, icon distinctiveness is much more important than icon representativeness, at least in this task.
Interestingly, response time increased substantially with menu size for word menus and representational icon menus; however, the increase for the distinctive icon menu was not significant. Distinctive menus seem to be impervious to menu size. Word menus seem to be searched relatively slowly and sequentially. Similarly, representational menus are searched in a slow and sequential manner. In contrast, distinct, abstract icons seem to be searched in a rapid and parallel fashion. The global features of abstract items appear to facilitate parallel visual processing of the items.
These experiments clearly support the positive benefits of graphic menus. But it is clear that icons must be carefully selected. They should depict concrete objects and actions rather than analogies or abstractions. They should be highly discriminable and emphasize global features. Furthermore, icons should be clearly interpretable. Their relationship to commands should be direct and obvious. Finally, one should consider using both icons and verbal labels to retain the advantages of both.
6.3 Selection Response
Once the user has formed an intention and decided upon the desired menu option, the selection must be activated. The "selection response" is the overt act on the part of the user that impacts on the computer. As important as it is, the selection response often seems to be the least of one's concerns in designing a menu selection system. However, it is the focal point of the user's communication with the computer. Clarity, simplicity, speed, and accuracy are essential. For the novice user, instructions must be given. It can be extremely frustrating to the novice user to key in the desired alternative and wait five minutes with no response because the return key had not been pressed. For the experienced user, continual frustration may occur when the user habitually hits the wrong key or overshoots the item with a mouse. Added up over the life of the system, these human/computer errors may reduce the efficiency of the system by 5-25% depending on the extent of interaction.
Many different types of input devices have been employed in menu systems as noted in chapter 2. In this section we will be interested not so much in the physical device as its functionality, location, and state on the display. Whatever the device, some instructions must be given to the user (especially to the novice user) as to how it functions. Furthermore, the response generally is given a visual representation at some location on the screen. Finally, the display conveys the current state of the response. For example, the display may indicate the following states: no selection, tentative selection, implemented selection.
6.3.1 Response Instructions. Instructions are needed as to how to make the response selection in two cases: (1) the person using the system for the first time and (2) the person returning to the system who has forgotten how to make a response selection. In the first case, instructions need to explicitly state the sequence of actions, such as "Enter number and press RETURN." Response methods may vary among systems leading to habitual errors. If a user is used to terminating input by pressing the RETURN key, and the system is designed to respond following one key entry, the habitual press of the RETURN key may lead to an error in a subsequent menu. Such a system needs to be able to ignore the RETURN key if pressed within a certain time interval.
On the other hand, if the user is not used to entering a terminating character, he may wait a long time for the system to respond before realizing the problem. One solution to this is to use a time out, such that after a period of time, the system responds to whatever has been input.
6.3.2 Position of Response. With touch screens and pointing devices, the response input is conceptually spatial and congruent with the option; that is, the location of the option is the location of the response. This adds a persuasive sense of compatibility. On the other hand, when the input is via a keyboard or key pad, the spatial position of the response is distal from the option. In the case of special key pads, the entry is conceptually at the key pad. For a standard keyboard input, we generally think of the response as being inserted on the screen. The location of this insertion point generally follows the list of responses. Some systems locate the insertion point following the stem. In a survey of novice and experienced computer users, Norman (1987) found that novices preferred the location following the stem and before the options and that experienced users preferred the location after the options.
One question is whether a distal insertion point is needed at all. Given keyboard input, it need not be echoed on the screen; but rather cause a highlighting of an alternative. Recent studies comparing "embedded" versus "explicit" menus suggest that when menu items are integrated or embedded in the task or the text being read, users work faster than when items are explicitly listed at a distal point. For example, Powell (1985) compared user search through a hierarchical data in order to answer twenty questions in a 15 minute period. Figure 6.11 shows an example of one of the frames of the embedded menu. Significantly more questions were answered correctly with the embedded menus than with the explicit menus. Furthermore, subjects preferred using the embedded menus over the explicit menus. Koved and Shneiderman (1986) found similar results for information search in online maintenance manuals.
6.3.3 Response Compatibility. An important factor in the theory of cognitive control is the concept of "stimulus-response compatibility" or as referred to in the human factors literature "control-display compatibility." The idea is that aspects of the stimulus should be congruent with aspects of the response. Congruency may be defined along a number of dimensions. Spatial congruency refers to a match between the relative location of the stimulus and the relative location of the response. For example, if the alternatives are arrayed left to right on the screen, responses are congruent if they are arrayed left to right on the keyboard. Spatial congruency is often violated by systems using the keyboard. For example, function keys will be listed along the bottom of the screen in order from 1 to 10. But they appear as two columns on the left of the keyboard. The result is that the user must go through an extra cognitive process in order to translate screen location to keyboard location.
A second type of congruency has to do with letter and number pairing. Table 6.1 illustrates the types of congruency and incongruity that can result.
Table 6.1
Examples of Congruent and Incongruent Option-Response Pairs
Congruent Incongruent (Worse Case) Mixed
Ordered Options Ordered Responses
______________________________________________________________________________
D - Debug B - Debug D - Edit D - Compile
E - Edit G - Edit E - Buffer E - Edit
F - File C - File F - Graph F - Graph
G - Graph H - Graph G - Halt G - Halt
H - Halt D - Halt H - Debug H - Read
______________________________________________________________________________
Figure 6.12 shows the results. Compatible letter responses led to the fastest responses. Subjects were able to circumvent the need to locate the option and then press the response. Compatible number responses were superior to incompatible numbers and incompatible letters were slowest. This suggests that if compatible letter responses can be generated they are preferred. Since this is usually not the case, compatible number responses are the next best design choice.
6.3.4 Response Verification and Feedback. The menu display is generally used to indicate the current state of response selection. Three states exist:
Waiting for response. In this state the user has not yet selected an option. The system may explicitly prompt the user by providing an insertion point for the response and direct the user to the keyboard or pointing device.
Tentative response selected but not implemented. In this state the user has input a selection but has not yet been verified that it is the selection to be implemented. This state is crucial in systems where selections have consequential effects and where response error is a possibility. It allows the user to correct a response before it is implemented. Response error may occur due to pressing the wrong key or missing the target using a pointing device. The disadvantage is that the user has an extra response to make in order to implement the selection. In many cases this is merely pressing the RETURN or SEND key. In other cases, with simultaneous menus, the implement response may activate a whole set of tentative selections.
Response selected and implemented. In this state the user has made a response to implement the selection. The system may provide feedback to the user that the desired selection has been made. If the effect of the selection is not immediately apparent or delayed in some way, the system should provide feedback to the user confirming the selection and informing them of its progress.
6.4 Summary
Each menu frame in a system conveys information and provides a forum for user control. The graphic format of the frame affects attention, interpretation, context, and search time. Well-designed menus help to focus attention on the relevant dimensions of choice; they help to interpret the function of alternatives and the reason for the choice; they help to establish the context for the choice; and they help to facilitate visual search for items in the menu list.
The wording of the items is crucial. To the degree that the items convey unambiguous and highly discriminable options, users will be able to avoid wrong selections and reduce the time that it takes to traverse menu structures. It is noted that guidelines need to be developed to assist designers in writing and evaluating the wording of alternatives. As noted in the subsequent chapter on prototyping, evaluation of the system should be at the individual word level rather than at the global performance level.
The selection response and instructions on how to respond vary greatly among systems. The lack of consistency among systems can result in problems as users go from one system to another. To the extent that any system differs from the expectations of the user, explicit instructions are required. It is noted that the position of response insertion on the display should be consistent with the user's mental model of the system. Responses should be compatible with the alternatives. Finally, the system should give the user an opportunity to verify the selection and change it if necessary before commitment and should provide acknowledgement that the selected alternative was properly accessed.
The amount of menu detail and response instructions needed by the user will vary with training and experience as discussed in the next chapter. In principle, however, the variables pertaining to format and phrasing will remain important at all levels of experience.
Screen display and the formatting of a menu frame is an art. The phrasing of stems, leaves and other information in the menu frame is a wordsmithing activity that without proper attention can lead to many problems. Fortunately, a growing body of literature in cognitive psychology and more directly in ergonomic research is helping to define the factors of menu design and to set guidelines for "do's and don't's" in designing menus. Many of these guidelines and the research supporting them will be reviewed in this chapter.
6.1 Formatting the Menu Frame
The purpose of a well-designed menu frame is to convey information to the user in the most efficient manner possible. It must be remembered that users read and search for information with expectations. They do not necessarily read the menu frame word-by-word from the beginning. Rather they scan it and are highly influenced by its organization and layout. It follows that organization and layout should facilitate the search process rather than work against it.
Much has been written about screen display in computer systems (e.g., Galitz, 1981; Cropper & Evans, 1968; Engel & Granda, 1975; Foley, Wallace, & Chan, 1984; Mitchell, 1983; Peterson, 1979; Shurtleff, 1980; Tullis, 1981, 1983). Many of these references serve as excellent guides for designing menus.
In terms of menu display, the designer must decide what information to display, in what form, and where. As seen in Chapter 2, menus include information about the context, the stem and leaves of the choice options, and response and feedback information.
6.1.1 Amount of Information per Screen. The first decision that must be made has to do with how much information should be displayed. Most guidelines in the literature suggest that only information essential to the user's needs should be displayed. Irrelevant information clutters up the screen and makes it more difficult and time consuming for the user to locate the option that he desires. The problem is knowing what the user needs and determining what is irrelevant vs. what is necessary.
The designer must contend with a set of tradeoffs:
Amount of Information vs. Scan/Reading Time. The more information, the more difficult and time consuming it is to search for target information and to read the information. Research in human information processing indicates the obvious: it takes longer to search for a word as the amount of surrounding information increases. When time is a critical factor in performance, brevity of the menu is essential in reducing human processing time. Furthermore, readability is affected by the density of text on the screen. It is suggested that no more than 30% of the available character spaces on the screen should be used (Danchak, 1976; Poulton & Brown, 1968; Ramsey & Atwood, 1979).
Level of Experience vs. Need for Explanation. The less familiar the user is with the system, the more detailed the explanations need to be of menu options. Menus need to be information rich for novices. The implication is that performance will be slow, but this is generally not a major concern for novices and casual users. On the other hand, menus may be information lean for experienced users. The terms serve only as cues to jog the memory of the user and to indicate the allowable options at the current state. Furthermore, the speed of scanning and selection is a function of the training of the user. In terms of the theory of cognitive control, the experienced user has acquired sufficient knowledge about the system so that information flow from the computer to the user is no longer necessary.
Amount of Information per Screen vs. Number of Screens. The less information presented per screenful, the more screens that are necessary. One of the major concepts in menu systems is to subdivide information and user control into a network of screens. Typically, when all of the information and options do not fit on one screen, the designer divides the menu into two or more screens. Ultimately, this leads to the depth vs. breadth tradeoff which will be discussed in a later chapter. At this point it is sufficient to say that when screens must be subdivided, it adds to the cognitive complexity of the system. Performance of both novice users and experienced users is lowered due to the added steps necessary to access additional screens. This problem is alleviated in part by larger screens or multiple screens displaying longer menus, but then one must contend again with the first trade-off having to do with scanning and reading time.
6.1.2 Focusing Attention on the Menu. Principles of good screen layout may be derived from theories of perception in Gestalt psychology and more recently cognitive psychology. One of the basic processes in perception is the segmentation of the field into figure and ground. The figure contains those objects that are the focus of attention; whereas, the background is amorphous and undifferentiated. Studies in visual perception, indicate that the figure-ground distinction is the first and most fundamental aspect to be perceived in a display. However, the figure-ground distinction may vary with the stimulus. At one extreme, figure-ground may be a potent aspect of the display that forces the viewer to perceive one area as figure and the rest as ground. On the other hand, the distinction may be weak and driven primarily by the current need state of the viewer. For some displays, we are quite capable of totally reversing figure and ground or of refocusing attention to generate new segmentations of the field.
When a user is accessing a menu, the menu becomes the figure and the rest of the screen, which may be displaying previous output becomes the background. Although the focus of attention is on the foreground, the background can either serve to enhance this focus or to distract from it. Figure 6.1 illustrates three levels of figure-ground distraction for a pull down menu.
When a gray background is used as in the top panel in Figure 6.1, figure-ground segmentation is maximized; when other information clutters the background as in the middle panel, segmentation is reduced; and when the information in the background is similar to the figure as in the bottom panel, segmentation is minimal. Systems supporting the cluttered desktop metaphor can suffer from figure-ground ambiguity. Fortunately, the figure-ground distinction can be enhanced by the proper use of graphic devices such as borders, highlighting, overlapping, graying out, colors, etc.
6.1.3 Perceptual Grouping. Perceptual grouping is an important process that aids in interpreting what something is and its relationship to other objects in the field. Characteristics of the objects themselves suggest an organization of the field of view. We perceive objects that are similar in appearance or close in proximity as belonging together. Such perceptual grouping affects subsequent processing of the information in terms of the meaningfulness and expectations regarding functions. These concepts may be effectively used in menu design; or if not attended to, perceptual effects can be quite detrimental.
Studies have found that spatial proximity of options in a menu affect the grouping of items in the long term memory of the user. Items may be "chunked" together. Chunking is the cognitive process of storing/retrieving a set of items as a unit. For example, single letters may be chunked as a word, and words may be chunked as a sentence. Card (1982) used a recall test in order to assess the chunking process in menu selection. After a series of search trials using an 18 item menu with boxes containing 2 to 4 items, subjects were asked to write down all of the items that they could remember. When items are recalled from memory, items that are in the same chunk tend to be recalled together with a short time interval between recall. Items in different chunks tend to be separated by a longer time interval. Card found that items next to each other or in the same box in the list tended to be clustered together in recall.
Similarity of appearance also leads to perceptual grouping. Items may be similar in phrasing or in graphic shape. One menu uses the items "SAVE" and "SAVE AS ..." The first option saves the current memory buffer by writing over a file that had been previously opened thus destroying the old contents. The second option saves the current memory buffer by opening a new file and writing to it while doing nothing on the old file. The similar phrasing of the items helps the user to group them together as having a similar function with respect to writing the memory buffer to a file. It may be detrimental, however, by obscuring the effect on the old file. The user often has to learn such subtle distinctions the hard way.
Spacing of alternatives can lead to perceptual grouping and has a strong effect on visual search. In their study of menu selection using large menus of 64 items, Parkinson, Sisson, and Snowberry (1985) compared menus having either no spacing between category groups or having one additional space between groups. Menus with the additional space were searched in 25% less time. It is interesting to note that this difference was not substantially reduced with practice. Perceptual grouping induced by spacing seems to have several beneficial effects. First, it appears that grouping helps users to recognize the type of organization and structure of the list (e.g., organization by alphabetical or natural categories). Second, perceptual grouping allows the user to mark and locate the beginning and ending of groups when visually scanning for target items.
Menu systems that display icons can also make good use of perceptual grouping. Scanning and perceptual grouping is often enhanced by graphic images. Figure 6.3 shows the effect of adding icons. In the top panel the names of 12 files are displayed. The bottom panel adds icons indicating the type of file. It is expected that search time for selecting files with the icons will be faster than without the icons even though files in both cases are clustered by type. Icons serve to add perceptual features that aid in grouping and search processes.
6.1.4 Menu Context. Screen design involves setting a context in which the items are perceived and interpreted. Two types of context must be considered. A "sensory" context is set by the background and explicit context information displayed. The recognition and meaning of items in the list are affected by this surrounding information. A "nonsensory" context is set by previous events remembered by the user and by his current task and goal states. In a bibliographic search, for example, sensory context may be established by displaying that the user has located a set of 10 bibliographic citations by a title search and that menu options refer to this set. The nonsensory context of the user may be that he is searching for a particular book and he has not been able to locate it by an author search.
The interpretation of menu items and their consequences are highly influenced by the context. Terms such as "list", "print", "kill", "find", "change", "load" are highly context dependent. "List" may refer to a command to display a listing of a program or it may refer to a list of words. The problem is that words are often ambiguous outside of a clear context. For terms such as these, the user must disambiguate the meaning on the basis of their frequency of use and the recency with which a particular meaning was evoked (Anderson, 1988). The point is that menu systems must provide sufficient context for the users to easily disambiguate the meaning of the menu items.
Context that relies on nonsensory mechanisms is often unpredictable. Consequently, guidelines suggest that every screen explicitly display the user's last selection which led to the current screen or state of the system in order to aid in a step-wise refinement of the interpretation of items. For example, a search through a hierarchical database of employees in an organization may retain the context by adding to the title after each menu selection in the following manner:
- West Coast
- West Coast Sales Division
- West Coast Sales Division Manager
- West Coast Sales Division Manager of Government Services
6.1.5 Ordering of Menu Items. What is the best order of the items in a menu? There are three major purposes behind the ordering of items. The first is to facilitate search for an item in a list. When items are organized along some characteristic or dimension, the user's search strategy can use that information to locate items faster. Alphabetic ordering helps the user know approximately where to look in a list for a word. The second is to convey additional information about the structure and relationships among the items. This information may aid in helping make choices by knowing the similarities and dissimilarities among items. The third is merely to agree with the user's inferred ordering so as not to disrupt cognitive processing by creating dissonance between the user's knowledge base and the system's knowledge base. Items may be organized according to a number of characteristics.
Random Order. As a baseline, we will consider the case of no perceived organization. For all intents and purposes, the list appears to be random. There may be some underlying software rationale for the order depending on system constraints, but the reason for this order is not apparent to the user and has nothing to do directly with the human/computer interface. For example, the order of desk accessories on the Apple Macintosh is determined by tool priorities. The user can generate no algorithm to facilitate search other than a rote memorization of item position. However, it should be mentioned that frequent users of systems quickly acquire a knowledge of the spatial location of items in a randomly ordered (but constant) list. Card (1982) and others have shown that differences in performance due to the organization of the items are reduced or eliminated entirely with practice. Subjects apparently learn the organization, albeit random, and know where to look for items in the list. Consequently, the detrimental effects of random order, are observed when the user has not yet gained extended familiarity with the items or when a different random order occurs every time the menu is viewed.
Alphabetic Ordering. An alphabetic ordering of alternatives may be used when there is a particular keyword, command name, or title that can be meaningfully alphabetized and scanned.
Numeric Ordering. Numeric ordering can be used for the selection of alternatives that are associated with numbers such as type size, baud rate, and number of copies.
Chronologic Ordering. Items that refer to dates and times may be ordered from oldest to newest. For example, articles in a news service, entries on an electronic bulletin board, and the list of months may be chronologically ordered.
Frequency of Use. Items may be ordered by frequency of use with often used options listed first and infrequent options listed last. Frequency of use may be determined a priori by a logical analysis of the typical tasks or a posteriori from observed frequency of use.
Sequential Processing Order. Items may be listed according to their inferred order in a process or according to a cognitive ordering of items. The items "OPEN FILE", "ADD RECORD","CHANGE RECORD","DELETE RECORD","CLOSE FILE","QUIT" imply a logical ordering of steps. The task analysis not only dictates the functions that are required but also their usual order of operation. Cognitive order refers to the order that users would list items. For example, in a news network the items "WORLD", "NATIONAL", "STATE", "LOCAL" are ordered in terms of an underlying dimension of scope.
Semantic Similarity Ordering. Items may be ordered in terms of some semantic dimension such as impact, reversibility, potency, etc. Items that are most similar appear close to each other in the list. For example, a menu of style of type may be ordered in terms of emphasis: Plain, Underlined, Italicized, Bold. Parkinson et al. (1985) give an example of semantic ordering in which words sharing the most semantic features are presented adjacent to each other. For the category of topography the order of the following names would be: Mojave, Sahara, Everest, Matterhorn, Atlantic, Pacific, Huron, Erie. The first four are land areas consisting of two deserts and two mountains. The last four are bodies of water consisting of two oceans and two lakes.
Categorical Grouping. Items may be categorized and then ordered within groups according to the characteristics listed above. Categorization operates on the similarities and dissimilarities of the items. Categorization allows for hierarchical search and decision processes on the part of the user as discussed in Chapter 3. Designers must also consider that categories and items within categories must themselves be ordered.
The order of items in a menu should be consistent with the expectations and cognitive order held by the user. When the menu order is incongruent with the cognitive order, it makes it difficult for the user to locate items. Alphabetic order clearly reduces search time in many listed, but imagine the confusion in searching an alphabetized list of months, days, or numerals:
| Select the month in which you were born: | Select the day on which you were born: | Select a numeral: |
|---|---|---|
|
1 April 2 August 3 December 4 February 5 January 6 July 7 June 8 March 9 May 10 November 11 October 12 September |
1 Friday 2 Monday 3 Saturday 4 Sunday 5 Thursday 6 Tuesday 7 Wednesday |
1 Eight 2 Five 3 Four 4 Nine 5 One 6 Seven 7 Six 8 Three 9 Two 10 Zero |
The effects of menu organization on search time and on learning have been studied by a number of researchers. The results have been mixed due to additional factors such as the number of items in the list, type of search task, and experience of the subjects.
One clear finding, however, is that alphabetic and categorical ordering are superior to random ordering. What is not clear is the relative advantage of alphabetic vs. categorical ordering. The Card (1982) found that for menus of 18 items in a visual search task, there was an advantage of alphabetic ordering over a condition with categorical ordering and a condition with random ordering. On the other hand, for a 16 item list, Liebelt, McDonald, Stone, & Karat (1982) found a slight advantage for categorical organization over alphabetical and random orderings. However, the difference was not statistically reliable.
Often items are displayed according to several ordering schemes at once. Items may be grouped and then alphabetized or ranked in some way within groups. One study illustrates the importance of forming major groupings of items. McDonald, Stone, and Liebelt (1983) formed categorical groups with either similarity, alphabetical, or random ordering within groups. Menus consisted of 64 items arranged in 4 columns of 16 items. In three conditions, the four lists were categorized by food, animals, minerals, and cities. In Condition CC, the four lists were further ordered by obtaining similarity ratings and ordering the terms using multidimensional scaling and hierarchical clustering techniques. In Condition CA, the four lists were alphabetized, and in Condition CR the items were random. Two other conditions mixed the categories. Condition A displayed a complete alphabetic ordering of the 64 items and Condition R a random order of all items. A second factor in the experiment was of interest. Half of the subjects were shown the explicit word to search for and the other half were shown a single-line dictionary definition. The task was to locate the item on the display screen and press the key corresponding to the identification letters displayed immediately to its left. The identification letters were randomly assigned to items on each trial to prevent memory association that would circumvent the search process. Subjects were obtained from a temporary employment agency for secretarial help. Response time was measured over a series of 5 blocks of 64 trials. Figure 6.3 shows the results.
Overall subjects that were given definitions took longer to respond than subjects given the explicit word. Organization of the list had a significant effect on response time. For the explicit word, random ordering was much worse than any other conditions. Subjects were able to perform quickly in all of the categorical conditions and the alphabetized list. It would seem that once the subject identified the category and scanned the column of 16 items, order within the column was not an important factor. On the other hand, for the definitions, the categorical ordering (CC) was slightly superior to the alphabetized or randomized categories. The alphabetic and random lists led to the longest response times with the alphabetic condition only slightly superior to the random order. This result is important since users may or may not be searching for an exact word match. When they are searching for an item to satisfy a definition (or a command to perform a function), categorical ordering is important.
One would expect that the ordering of the list would become increasingly important as the length of the list gets longer. Perlman (1984) found this to be true. He varied the length of the list to have 5, 10, 15, or 20 items. The items were in one case the numbers 1 to 20 and in another case words beginning with the letters "a" to "t". The lists were either sorted or random. Four groups of 8 subjects searched either sorted words, sorted numbers, random words, or random numbers. Each subject searched all different list lengths. The target number or word was shown on the screen and the subject then searched for the item and then pressed the key corresponding to the item. Figure 6.4 shows the results. Response time increased with the length of the list. Response times for sorted items were shorter than for random items and for numbers than words. What is interesting about these results is the interaction between the list length and the type of list searched. When lists are sorted, the increase due to list length is less that when lists are random. This suggests that when menus include a lot of items, it is extremely important that the designer consider organization of the list in order to reduce search time.
It is not clear, however, whether alphabetic or semantic organization results in superior performance. Parkinson et al. (1985) found little difference in search time between alphabetic and semantic organization within categories of words when the categories were sufficiently separated by spacing. This is an important finding for designers since alphabetic ordering is easy to achieve while semantic ordering requires extensive pretesting. On the other hand it should be pointed out that if the designer wishes to convey a mental model of the system, semantic ordering may help to achieve this objective. For example, in command specification Parkinson et al. point out that given the task of executing a set of functions using a menu of commands, naive users would probably learn the system more efficiently if the commands for similar functions were grouped together in the display.
6.1.6 Orientation of the List. Generally, lists of options have been displayed in a vertical list (columns) rather than a horizontal one (rows). The question is whether the vertical or tabular listing is indeed superior to a single horizontal line listing. Single line listing saves space on the screen. In cases where the user input is an associated letter or number, one might expect the tabular arrangement to be superior. In the case of touch screen or pointing device, it is not clear. In cases where the single line listing is used, items are generally very brief and often single character options as in the following menu from an electronic bulletin board:
Option (A/T/D/E/T)?
In a study of menu preference by Norman (1987), computer science majors indicated a preference for single line listings of commands as opposed to a tabular listing. On the other hand, non-computer science students preferred a tabular listing.
In a study on the organization of items in large menus of 64 items, Parkinson, Sisson, and Snowberry (1985) compared category groupings arranged by column versus row. Menus with categories arranged in columns were searched in approximately 25% less time (1 sec faster) than menus with categories arranged in rows. From these results, visual search for targets appears to be substantially facilitated by column presentation. However, it must be noted that this finding is currently limited to large menus and to the particular configuration chosen by Parkinson et al. In their experiment all menus consisted of 64 words listed in four columns of 16. Eight words were chosen from each of eight natural categories and arranged in either columns (two categories per column) or rows (2 adjacent rows per category). This arrangement implies that given a target in the menu organized by column, subjects first located the correct category in a 2 x 4 array of categories and then scanned down to find the target within the category to locate the two digit response code for the item. In the case of menus organized by row, subjects first located the category in a vertical array of 8 categories, and then scanned 2 rows of 4 items to find the target. It is likely that having categories split on two rows slowed down the search due to the time to jump from line to line as well as from column to column when searching for a word within a category.
This interpretation is further suggested by the results from two alphabetized menus also used in the same study. In one menu, words were alphabetized from top-to-bottom by column and in the other from left-to-right by row. The difference in search time between these two menus was of approximately the same size as for categorical menus. Furthermore, Parkinson et al. note that the row versus column effects may have been confounded by the fact that more space separated words in adjacent columns than in adjacent rows. All this strongly suggests that it was not simply row versus column organization that led to the effect, but that the visual search distance and the number of jumps to reposition each search within categories accounted for the differences.
Although it is not yet clear that row arrangement is itself detrimental, it is clear that large differences in performance can be due to column versus row arrangement of menus. At present the results suggest that designers should attempt to arrange items so as to minimize visual search distance and the need to jump from one location to another when searching within the same category.
The issue of row versus column presentation is important in systems with pull down menus. Such systems generally display a menu bar which is a horizontal menu listing of vertical pull-down menus. In essence they alternate horizontal and vertical arrays. This idea can be extended as an effective way of handling successive depth of a menu while maintaining context. An example of this is shown in Figure 6.5 for a hierarchical tree of languages having a maximum depth of 5. Menu nodes shown in reverse video indicate the selected path.
At present little is known about the use of alternating menus. However, one would expect that with the use of a pointing device such as a mouse, the selection process is much like tracing a route in a maze. As such it will be highly spatial in character and affected by psychomotor processes.
6.1.7 Fixed vs. Variable Format. The literature suggests that the organization of information elements on the screen should be standardized in order to make effective use of spatial context. The user acquires an expectation of where to look to see particular information. When information could appear at different locations, the user must search for it. Teitelbaum and Granda (1983) compared fixed vs. variable formatting of the information. When the location of information was inconsistent, the time that it took to answer questions about information on the screen was longer than when the information was in a constant location. Moreover, when information was in a constant location, response times improved with practice indicating that users were acquiring expectations as to where information was located. Response times did not improve when information was presented in random locations.
The importance of consistency in screen and menu design cannot be over-emphasized. It is imperative that users develop expectations about the location of information to avoid having to repeatedly search for information. Furthermore, inconsistency probably fosters a level of frustration and confusion in the user that may generalize to the other menu selection processes.
6.2 Writing the Menu
Communication is an art. In human/computer interaction, the art is communicating through an interactive media about the control of that media. Fortunately, art is guided by principles. Menus must convey information to the user about (a) the nature of the choice (What is being accomplished in making this choice? What is the purpose of this choice?) and (b) the nature of the alternatives (What will happen if this alternative is selected?).
On the one hand, one would like to convey as much information as possible, explaining all the options in great detail. On the other hand, brevity is important due to screen limitations and the amount of information that the user wants or desires to read. Consequently, there is a trade-off between thoroughness and brevity. And the trade-off is to be moderated by the experience of the user. Many systems are developing the concept of dynamic help in which menus are highly descriptive for novice users, but as the user gains experience, the menus become brief pneumonic reminders of the alternatives.
6.2.1 Titling the Frame. Titles of menus should be more than place holders. They should be descriptive of the list of options and the context. The title, "MAIN MENU OPTIONS," conveys little information except that the user is probably at the top of the menu tree.
When a title is superfluous, then it may be omitted. To the extent that the list of options define the context, the title is redundant. This is particularly true for experienced users. Titles merely clutter the screen. It is an empirical question as to whether titles help and when they have served their usefulness and should be dropped. One possibility is to design a system in which the user may hide titles or reveal them as necessary.
The title, however, can be a potent stimulus in defining the context and increasing comprehension of the menu. An experiment by Bransford and Johnson (1972) illustrates this fact. They assessed comprehension and recall of a passage describing in detail the steps that one must go through in order to accomplish a certain function. In one case, subjects were given the title of the passage, "Washing Clothes" either before or after reading it. In a control condition, no title was given. Subjects given the title before reading the passage reported higher comprehension and recalled about twice as many items from the passage than other subjects. Subjects given the title after reading the passage did no better than the control subjects given no title. In the same way, menu titles have the potential of increasing the comprehensibility of menus and facilitating choice.
We may think of a schema as a conceptual framework that provides "slots" to be filled in with specific information. Menus are themselves schemata of the form: "Of the following options, select one: (a) ... (b) ... (c) ..." Users assimilate the particular options into this framework. Menus are also a part of a larger schema in which they are understood. Without establishing the context of the choice by instantiating a schema, comprehension of the menu frame may be severely limited. Consider, for example, the following menu:
Select function:
Open
Close
Activate top element
Activate bottom element
Clean
The purpose of this choice and the function of the alternatives is not at all apparent. However, if you are told that the schema has to do with the operation of an oven, the menu is comprehensible and easy to remember.
6.2.2 Wording the Alternatives. Alternatives may be single words, brief phrases, or longer descriptions of options. It is generally recommended that when brief phrases are to be used that they are verb phrases that are written in parallel construction (Shneiderman, 1987). For example, rather than phrasing the items as "Print," "Execute a program," and "Disk eject," they should be "Print a file," "Execute a program," and "Eject a disk." In general, it is recommended not to change the construction of a phrase within the same menu frame. In addition, Shneiderman lists the following guidelines:
(1) Use familiar and consistent terminology.
(2) Ensure that items are distinct from one another.
(3) Use consistent and concise phrasing.
(4) Bring the keyword to the left.
An important aspect of wording is not only that it conveys information, but that the alternatives are interpreted as distinctly different. Distinctiveness refers to the semantic aspects of an alternative that enhances its difference from other alternatives in the set. The object is to word items so as to emphasize differences rather than the commonality of function. Consider, for example, a menu that contains the items (1) Services for Professionals, (2) Home Services, (3) Business and Financial, (4) Personal Computing. Items 1 and 2 are not very distinct in that they both emphasize services. Furthermore, the distinction of what is meant by "professional" versus "home" is not clear. Items 1 and 3 lack distinctiveness altogether. Item 1 could be a subset of 3 or vice versa. Items 2 and 4 lack distinctiveness since "home" and "personal" share common aspects. The terms "services" and "computing" are not distinct since any services provided on a time sharing system must involve computing. Such a menu poses a dilemma for the novice user who, for example, wants to balance his checkbook. The question is how to design a menu system so as to maximally enhance the distinctiveness of items and minimize their confusability.
In a study by Schwartz and Norman (1986) novice computer users searched either (a) the original menu of a commercial time sharing system in which items were not worded in a holistic manner or (b) a modified menu of the system in which items were holistically worded. They hypothesized that the effect of item distinctiveness would vary with the type of search required. In explicit search problems, subjects searched for the exact wording of targets (e.g., "Search for 'Backgammon'"). In information find problems, subjects searched for menu items that provided information about subject matter topics (e.g., "Find information concerning how to balance your checkbook"). These two types of menu problems correspond to the two general uses of menu selection systems. In general, it was found that subjects using the modified menu (a) took less time per problem; (b) found targets in a more direct path; and (c) gave up on fewer problems than subjects using the original menu. Although information find problems took longer and required a greater number of frames to be traversed than explicit search problems, there was no interaction with distinctiveness. Distinctiveness had a similar effect on search processes for both explicit and information find problems.
Schwartz and Norman (1987) suggest that distinctiveness may be controlled to some extent by the way in which choices are clustered in the menu tree. Items may be shifted from one menu frame to another to enhance the differences among options within a particular frame. When the structure of the menu is fixed, the wording of alternatives may be altered to enhance differences. They suggest the following strategies for improving distinctiveness:
(1) Induce holistic, intact, and integral processing of the alternatives. Holistic alternatives are not likely to be decomposed into aspects that may be shared among the items. Options should appear as different as apples and oranges even though they may share many common aspects (e.g., both fruits, both round, both about the same size, etc).
(2) Maximize aspect redundancy and intercorrelations within alternatives and minimize them between alternatives. For example, the position and color of traffic lights are perfectly correlated and redundant. Consequently, both aspects can be used to discriminate between the alternatives of stopping and going.
(3) Present all alternatives simultaneously rather than one by one or page by page. When alternatives are processed sequentially (as in a slow display), a subjective emphasis on the common aspects of the alternatives may occur. Users look for shared aspects that lead from one alternative to another rather than contrasting the aspects when all are seen simultaneously.
(4) Explicitly define the universe of options so that each alternative may be understood in this context. If the alternatives are Diet-Rite, Diet Coke, 7-Up, and Fruit Juice, the universe of options is unclear. With one generic drink (Fruit Juice) and three specific drinks, one might over generalize to the universe of liquids as the reference, while others might restrict the universe to drinks that refresh. Avoiding this type of ambiguity is important in effective menu design.
(5) Ensure that the variability among the alternatives is high in terms of the rationale for the choice. For example, if the universe of choice is among soft drinks and the alternatives are Diet-Rite, Diet Coke, and Diet Pespi, there is little perceived variability. However, if the universe of choice is limited to dietetic soft drinks, then the perceived variability is high.
6.2.3 Graphic Alternatives
An increasing number of software applications are using graphic symbols called "icons" to represent objects and operations. Graphic symbols have long been popular in cartography. Graphic symbols can be visually more distinct from one another than words, and it is easier to spot a graphic symbol on a map than it is to locate a word. Similarly in human/computer interaction graphic symbols conserve space and yet are more distinct on busy display screens. However, there are a number of questions about the relationship between graphic symbols and the objects and commands that they represent when used in menu selection. Hemenway (1982) distinguishes between icons that depict (a) the objects on which the commands operate, (b) the operations or operators themselves, and (c) the operations on the objects. Figure 6.6 gives examples of these three types.
The first panel shows icons that represent objects operated on by some command. Since only the operand is depicted, it is not always clear what the operation is. Often, as in the case of the camera, it is simply an on/off function. In other cases, users must learn that more complex commands are implied. The middle panel shows icons that depict operations. The arrow has the conventional meaning of movement or direction and the "X" has the meaning of deletion or negation. Both meanings more or less represent the intended command. On the other hand, the operations represented by the paint brush and scissors are implied by what the tools do. Finally, the icons in the last panel combine both objects and operators. The first icon combines the box (image area) and the "X" (delete) to imply delete image area. The second and third icons show objects before and after the operation. The operation is inferred by the difference between the two. The last icon (fill in) shows a snapshot of the operation in progress. Animation completes richness of meaning and provides yet a stronger basis for implying the operation by showing the operation in action.
Hemenway proposes a simple model for the interpretability and effectiveness of icons. When a new icon is encountered its interpretability depends on (a) its comprehension (e.g., the ability to discover what the icon depicts) and (b) its effectiveness as a retrieval cue (e.g., the ability to form a link between what is depicted and the corresponding command). For experienced users, the effectiveness of an icon ability depends on (a) the ability to recognize what the icon depicts and (b) the ability to retrieve the link between the icon and the command from memory. It is predicted that highly familiar, conventional symbols will be easier to learn and recall than obscure icons that lack distinguishing features. Furthermore, icons that directly depict objects and operations are predicted to be more effective than ones that resort to analogy and convoluted links.
Often the organization and inter-relation among commands can be expressed in shared features of the icons. The icon for a tool to generate rectangles and filled rectangles shares the shape but differs in shading. Similarly icons for drawing lines may vary in thickness corresponding to the width of the lines created (see Figure 6.7). Hemenway suggests that when elements are shared across icons the user has less to learn since familiarity on the shared feature transfers to the new icon. Furthermore, repetition of the elements can make it easier for users to link the elements to aspects of the command and to recognize how icons vary in terms of their features.
An empirical study was conducted by Rogers (1987) to test the prediction that direct linking between icons and commands results in better interpretability of icons. Four types of icon sets for file manipulation, text editing, and printing were designed that were either (a) abstract symbols, (b) concrete analogies, (c) concrete objects, or (d) concrete objects operated on with abstract symbols. Abstract symbols and concrete analogies constitute indirect mapping; whereas, concrete objects provide a direct mapping. For comparison a fifth group was composed of verbal names consisting of high imagery words. The icon set with the most direct mapping (concrete objects operated on with abstract symbols) resulted in the fewest accesses to the help facility and least number of errors. However, there was also an interaction between the type of command and the form of representation. File operations such as "to save" appeared to be easier to use and recall when the commands were verbal . Text editing commands such as "insert a space" were more discriminable as icons.
These results suggest that designers should not only use icons that have a direct linking but that they should carefully analyze the nature of the command and its representation. Icons are perhaps best used to represent the operations of graphic tools and objects; whereas, verbal labels are best for formal commands.
A final possibility is to add graphic and verbal labels. It has been speculated that graphic symbols added to menu items could impair the speed and accuracy of performance (Mills, 1981). Studies on categorization provide indirect evidence about how pictures and verbal labels interact. Smith and Magee (1981), for example, found that when subjects were asked to categorize words, the simultaneous presentation of an incongruent picture disrupted processing (e.g., a picture of a car shown with the word "table"). However, when subjects were asked to categorize pictures, the presentation of an incongruent word did not disrupt processing (e.g., the word "car" shown with a picture of a table). Furthermore, it has been shown that the categorization of pictures tends to be faster than of words (Pellegrino, Rosinski, Chiesi, & Siegel, 1977). Such evidence suggests that graphic symbols may facilitate menu selection or at least not degrade performance.
To test the effect of adding graphic symbols, Muter and Mayson (1986) compared three conditions using a videotext data-base: (a) text-only menus in which verbal items were displayed in a double-spaced linear list, (b) text plus graphic menus in which verbal items were displayed with graphic symbols of the items and distributed in a nonlinear fashion on the page, and (c) control menus in which verbal items were distributed in a nonlinear fashion on the page. To enhance the pairing of graphics and labels both were shown in the same color. Subjects were shown menu pages and asked a series of 12 questions (e.g., "Where can you buy a low priced bedroom suite?"; "Where can you locate a handsaw?"). Instructions emphasized both speed and accuracy. It was found that the text plus graphics condition resulted in significantly fewer errors (2.8%) than either the text-only (5.6%) or the control (4.6%) conditions. Overall response times did not differ. Thus, the addition of graphics halved the error rate without significantly increasing the processing time.
On the other hand, a study by Wandmacher and Müller (1987) found no difference in error rate between graphic menus and word menus, but did find a difference in reaction time in both a search-and-select paradigm and a recognition paradigm. On each trial of the search-and-select experiment subjects were given a task description (e.g., "You want to print a document") and a menu of either word commands or icons was then displayed (See Figure 6.8). Subjects made their selections by entering a number corresponding to the item. The order of the menu items was randomized on each trial to so that subjects could not memorize the numbers. On each trial of the recognition experiment only one item was displayed and subjects had to respond by pressing a prelearned number. In both the search-and-select and the recognition paradigm, response time were faster for icons than for words. These authors suggest that the gap between the meaning of an expression and its form is smaller for icons than for word commands. Hutchins, Holland, and Norman (1986) refer in general to this gap as the "articulatory distance." Icons may facilitate menu selection because they provide a more direct access to the meaning of the functions involved. Less time is required to translate an intention into an actual selection.
The use of graphics in menu selection appears to be a promising design feature. However, the effective use of graphic symbols requires thoughtful consideration. Arend, Muthig, and Wandmacher (1987) point out that it is not sufficient to just recommend the use of icons. In the same way that Schwartz and Norman (1986) demonstrated that discriminability has an effect on word menus, distinctiveness may have an effect on graphic menus. Arend et al. suggest that the "global superiority" effect in visual perception may affect the degree to which icons facilitate menu selection. The global superiority effect basically is that global features of figures (e.g., shape, color, size) can be selected and responded to considerably faster than local features (e.g., lines and structures within a figure).
Arend et al. contrasted the effects of (a) icon distinctiveness defined in terms of global superiority and (b) icon representativeness defined in terms of articulatory distance. Distinctive icons were abstract and emphasized one global feature. Representational icons included local features to make the meaning of the icon more apparent. A search-and-select paradigm was used in which subjects were given task descriptions and then had to select the appropriate menu item. Either a word command menu, a distinctive icon menu, or a representational icon menu was presented to different groups of subjects. In addition menus were constructed of either 6 or 12 items. Subjects made selections by pressing the key corresponding to the position of the menu item on either a 3 x 2 or 3 x 4 key matrix depending the menu size condition. Figure 6.9 shows the menu sets used in the 12 item condition.
The results clearly favored the distinctive icon menu over either the word menu or the representational menu. Figure 6.10 shows the mean response time for each menu. Over all conditions, the distinctive icon menu resulted in response times that were nearly twice as fast as for word or representational icon menus. Consequently, the type of icon used has a great effect on performance. Distinct, abstract icons which have high global superiority result in the best performance. Representational icons that reduce the articulatory distance, however, do no better than word menus. Apparently, icon distinctiveness is much more important than icon representativeness, at least in this task.
Interestingly, response time increased substantially with menu size for word menus and representational icon menus; however, the increase for the distinctive icon menu was not significant. Distinctive menus seem to be impervious to menu size. Word menus seem to be searched relatively slowly and sequentially. Similarly, representational menus are searched in a slow and sequential manner. In contrast, distinct, abstract icons seem to be searched in a rapid and parallel fashion. The global features of abstract items appear to facilitate parallel visual processing of the items.
These experiments clearly support the positive benefits of graphic menus. But it is clear that icons must be carefully selected. They should depict concrete objects and actions rather than analogies or abstractions. They should be highly discriminable and emphasize global features. Furthermore, icons should be clearly interpretable. Their relationship to commands should be direct and obvious. Finally, one should consider using both icons and verbal labels to retain the advantages of both.
6.3 Selection Response
Once the user has formed an intention and decided upon the desired menu option, the selection must be activated. The "selection response" is the overt act on the part of the user that impacts on the computer. As important as it is, the selection response often seems to be the least of one's concerns in designing a menu selection system. However, it is the focal point of the user's communication with the computer. Clarity, simplicity, speed, and accuracy are essential. For the novice user, instructions must be given. It can be extremely frustrating to the novice user to key in the desired alternative and wait five minutes with no response because the return key had not been pressed. For the experienced user, continual frustration may occur when the user habitually hits the wrong key or overshoots the item with a mouse. Added up over the life of the system, these human/computer errors may reduce the efficiency of the system by 5-25% depending on the extent of interaction.
Many different types of input devices have been employed in menu systems as noted in chapter 2. In this section we will be interested not so much in the physical device as its functionality, location, and state on the display. Whatever the device, some instructions must be given to the user (especially to the novice user) as to how it functions. Furthermore, the response generally is given a visual representation at some location on the screen. Finally, the display conveys the current state of the response. For example, the display may indicate the following states: no selection, tentative selection, implemented selection.
6.3.1 Response Instructions. Instructions are needed as to how to make the response selection in two cases: (1) the person using the system for the first time and (2) the person returning to the system who has forgotten how to make a response selection. In the first case, instructions need to explicitly state the sequence of actions, such as "Enter number and press RETURN." Response methods may vary among systems leading to habitual errors. If a user is used to terminating input by pressing the RETURN key, and the system is designed to respond following one key entry, the habitual press of the RETURN key may lead to an error in a subsequent menu. Such a system needs to be able to ignore the RETURN key if pressed within a certain time interval.
On the other hand, if the user is not used to entering a terminating character, he may wait a long time for the system to respond before realizing the problem. One solution to this is to use a time out, such that after a period of time, the system responds to whatever has been input.
6.3.2 Position of Response. With touch screens and pointing devices, the response input is conceptually spatial and congruent with the option; that is, the location of the option is the location of the response. This adds a persuasive sense of compatibility. On the other hand, when the input is via a keyboard or key pad, the spatial position of the response is distal from the option. In the case of special key pads, the entry is conceptually at the key pad. For a standard keyboard input, we generally think of the response as being inserted on the screen. The location of this insertion point generally follows the list of responses. Some systems locate the insertion point following the stem. In a survey of novice and experienced computer users, Norman (1987) found that novices preferred the location following the stem and before the options and that experienced users preferred the location after the options.
One question is whether a distal insertion point is needed at all. Given keyboard input, it need not be echoed on the screen; but rather cause a highlighting of an alternative. Recent studies comparing "embedded" versus "explicit" menus suggest that when menu items are integrated or embedded in the task or the text being read, users work faster than when items are explicitly listed at a distal point. For example, Powell (1985) compared user search through a hierarchical data in order to answer twenty questions in a 15 minute period. Figure 6.11 shows an example of one of the frames of the embedded menu. Significantly more questions were answered correctly with the embedded menus than with the explicit menus. Furthermore, subjects preferred using the embedded menus over the explicit menus. Koved and Shneiderman (1986) found similar results for information search in online maintenance manuals.
6.3.3 Response Compatibility. An important factor in the theory of cognitive control is the concept of "stimulus-response compatibility" or as referred to in the human factors literature "control-display compatibility." The idea is that aspects of the stimulus should be congruent with aspects of the response. Congruency may be defined along a number of dimensions. Spatial congruency refers to a match between the relative location of the stimulus and the relative location of the response. For example, if the alternatives are arrayed left to right on the screen, responses are congruent if they are arrayed left to right on the keyboard. Spatial congruency is often violated by systems using the keyboard. For example, function keys will be listed along the bottom of the screen in order from 1 to 10. But they appear as two columns on the left of the keyboard. The result is that the user must go through an extra cognitive process in order to translate screen location to keyboard location.
A second type of congruency has to do with letter and number pairing. Table 6.1 illustrates the types of congruency and incongruity that can result.
Table 6.1
Examples of Congruent and Incongruent Option-Response Pairs
Congruent Incongruent (Worse Case) Mixed
Ordered Options Ordered Responses
______________________________________________________________________________
B - Buffer F - Buffer B - Compile B - Assign
C - Compile A - Compile C - Assemble C - BufferD - Debug B - Debug D - Edit D - Compile
E - Edit G - Edit E - Buffer E - Edit
F - File C - File F - Graph F - Graph
G - Graph H - Graph G - Halt G - Halt
H - Halt D - Halt H - Debug H - Read
______________________________________________________________________________
Option-Response compatibility has two directions. One is a forward
compatibility from the option to the response (e.g., if the option is
"Assemble", the response must be "A") and the other is a backward compatibility
from the response to the option (e.g., if the response is "A", the option must
be "Assemble"). Forward compatibility is broken down when there is
uncertainty as to what the response is given the option. Although the menu
requires that there be a unique response for each option, that unique coding
may not be apparent to the user. Backward compatibility is broken down when
there is uncertainty as to what the option is given the response. This occurs
when several options begin with the same letter.
The effect of incompatibility was observed in an experiment by Perlman (1984).
Time to press the correct selection response was compared in four menus which
varied compatibility with letter responses and number responses: (a) a
compatible letter menu in which letter responses were the same as the first
letter of each menu item (as in the first column of Table 6.1); (b) an
incompatible letter menu in which the letters "a" through "z" were randomly
assigned as the responses; (c) a compatible number menu in which the numbers 1
through 8 were assigned as the selection responses; and (d) an incompatible
number menu in which numbers were randomly assigned to menu items. Items were
always listed in the same alphabetic order.Figure 6.12 shows the results. Compatible letter responses led to the fastest responses. Subjects were able to circumvent the need to locate the option and then press the response. Compatible number responses were superior to incompatible numbers and incompatible letters were slowest. This suggests that if compatible letter responses can be generated they are preferred. Since this is usually not the case, compatible number responses are the next best design choice.
6.3.4 Response Verification and Feedback. The menu display is generally used to indicate the current state of response selection. Three states exist:
Waiting for response. In this state the user has not yet selected an option. The system may explicitly prompt the user by providing an insertion point for the response and direct the user to the keyboard or pointing device.
Tentative response selected but not implemented. In this state the user has input a selection but has not yet been verified that it is the selection to be implemented. This state is crucial in systems where selections have consequential effects and where response error is a possibility. It allows the user to correct a response before it is implemented. Response error may occur due to pressing the wrong key or missing the target using a pointing device. The disadvantage is that the user has an extra response to make in order to implement the selection. In many cases this is merely pressing the RETURN or SEND key. In other cases, with simultaneous menus, the implement response may activate a whole set of tentative selections.
Response selected and implemented. In this state the user has made a response to implement the selection. The system may provide feedback to the user that the desired selection has been made. If the effect of the selection is not immediately apparent or delayed in some way, the system should provide feedback to the user confirming the selection and informing them of its progress.
6.4 Summary
Each menu frame in a system conveys information and provides a forum for user control. The graphic format of the frame affects attention, interpretation, context, and search time. Well-designed menus help to focus attention on the relevant dimensions of choice; they help to interpret the function of alternatives and the reason for the choice; they help to establish the context for the choice; and they help to facilitate visual search for items in the menu list.
The wording of the items is crucial. To the degree that the items convey unambiguous and highly discriminable options, users will be able to avoid wrong selections and reduce the time that it takes to traverse menu structures. It is noted that guidelines need to be developed to assist designers in writing and evaluating the wording of alternatives. As noted in the subsequent chapter on prototyping, evaluation of the system should be at the individual word level rather than at the global performance level.
The selection response and instructions on how to respond vary greatly among systems. The lack of consistency among systems can result in problems as users go from one system to another. To the extent that any system differs from the expectations of the user, explicit instructions are required. It is noted that the position of response insertion on the display should be consistent with the user's mental model of the system. Responses should be compatible with the alternatives. Finally, the system should give the user an opportunity to verify the selection and change it if necessary before commitment and should provide acknowledgement that the selected alternative was properly accessed.
The amount of menu detail and response instructions needed by the user will vary with training and experience as discussed in the next chapter. In principle, however, the variables pertaining to format and phrasing will remain important at all levels of experience.
No comments:
Post a Comment