It seems natural that highly practiced actions should require the least
time and effort to perform. Rapid action is due in part to the learning of
automatic behavior on the part of the user, but it is also due to the fact that
people arrange objects in their environment for fast and easy access. When an
electrician or a carpenter begins a task, he or she makes sure that the needed
tools and parts are within easy reach. Less frequently used items are left in
the truck or back at the shop. In the same way frequently chosen menu items
should be easily and rapidly accessible to the user. Less frequently used
functions may be buried under layers of menus. There are many situations in
which items are selected so frequently or in rapid repetition that designers
are challenged to minimize selection time and effort. The question is "How?"
Previous chapters have dealt with the overall efficiency of menu structure and access. These chapters and the research they have reported have assumed a more or less uniform use of menu options. In reality, access of menu options is extremely uneven. Some options are accessed with high frequency at the rate of hundreds of times per session. Style and editing functions in a word processor; shapes, lines, and patterns in a graphics package; and rows, columns, and functions in a spread sheet program are selected with great frequency. Other items, however, are accessed only rarely though they may nevertheless be vital to the operation. For example, options to open or close a file may only be used once per session but are absolutely necessary. The differences in frequency of use can be very drastic. Figure 10.1 shows the frequency of use of 49 options in the pull-down menu bar of MacWrite(TM) accessed by one user.
Laverson (1985) introduced the notion of a "menu utilization profile" to characterize the uneven access of items in a hierarchical menu system. He hypothesized that the efficiency of different rapid access methods would depend on the utilization profile. His ideas need to be extended, however, since the differential access of items varies as function of two important factors. The frequency of use may be either task specific in that the particular task performed determines the frequency of use or user specific in that the type of user determines the frequency of use. As an example of task specific utilization profiles, the frequency of use of menu items in a word processor varies substantially between the task of composing a paper and editing a paper. As an example of user specific utilization profiles, the frequency of complex, technical items may higher for experienced users and simple, direct items for inexperienced users of a graphic program.
Two basic methods of rapid access have been explored. One is to arrange items in the menu frame and/or hierarchy to minimize the distance from the initial menu state to the location of high frequency targets. The second is to provide various forms of menu "bypass," "jump-ahead," or direct commands that reduce the distance by defining shortcuts through the menu space. In this chapter, studies will reviewed that bear on the efficacy and importance of such approaches.
10.1 Location in the Menu
Although it may seem obvious that highly used options should be the most readily accessible items, many systems bury these items in the middle or at the bottom of menu lists and hierarchies. Such designs may be driven in part by the fallacy that menu selection makes everything uniformly rapid and easy. The idea is that since menus are inherently easy to use, all items must be easy to access. While this may be the perception of novice users and some designers, experienced users all too quickly detect the bottlenecks and frustratingly slow access to frequently required items.
10.1.1 Pull-down Menus. This is particularly true of pull-down menus. All items appear to be easily and rapidly available. Consequently, the order of items in pull-down menus is rarely designed around frequency of use.
Empirical results, however, indicate that access time depends on the location of items in the pull-down menu. In one experiment, 24 items were presented in 4 different menu configurations varying in the number and length of the pull-down menus (Norman & Chin, 1988). The conditions were as follows: 3x8, 4x6, 6x4, and 8x3, where the first number in each product is the number of pull-down lists and the second number is the length of each list. In addition, prior knowledge of target location was varied by using three different types of menus (see Figure 10.2). For array menus, items were simply 24 letter-number combinations. The letter indicated which pull-down list to access and the number indicated the target item in that list. Array menus provided the user with complete prior knowledge as to the location of target items before menus were pulled down. For alphabetized menus, items were 24 common words each beginning with a different letter of the alphabet. Pull-down lists indicated a letter range and items in each list were in alphabetical order. Although the menus provided a degree of prior knowledge as to location, it was somewhat more difficult and incomplete in comparison to the array menus. Finally, for randomized menus, the items were the same 24 words, but on each trial the words were cast in a different random order across all of the lists. The random menus provided no prior information to the user as to location. In effect this condition simulated the novice user on all trials having no experience.
Subjects were randomly assigned to one type of menu item and were tested on each of the four menu configurations in a counterbalanced order. Sixteen subjects participated in each of the three menu conditions. For each configuration, subjects were given a set of practice trials and then tested on two replications of each of the 24 target items in a randomized block order. On each trial the target item was displayed in a small window just to the left of center on a Macintosh(TM) screen. Subjects used a mouse to pull down and select the target item from the menu bar at the top of the screen. Times were recorded from the onset of the target to the selection response. If the subject selected an item that was not the target, it was recorded, a tone was sounded, and the target item was recycled in the pool of targets to be presented.
Response times were fairly fast for the array menus, averaging about 2.6 sec and varied significantly with target location (see Figure 10.3). Access times were fastest for the first positions in each pull-down and increased linearly with position down the list. Response times, however, did not vary with the position of the pull-down but only with position of items in the list. Linear regression equations of access time as a function of list position indicated that the time increment per item in the pull-down menu was 104 ms for the 3x8 condition, 147 for the 4x6 condition, 251 for the 6x4 condition, and 228 for the 8x3 condition. Consequently, although the lists were shorter in the 6x4 and 8x3 conditions, the pull-down time per item position was significantly longer than in the 3x8 and 4x6 conditions. Access times to the pull-down lists (as indicated by the intercept of the regression lines) were fairly similar across conditions, and there was no significant difference in overall response time among the four conditions. It would appear that for highly predictable menus, the trade-off between list position and pull-down time cancels out when items are accessed with equal frequency. However, if the designer knows that some items will be more frequently accessed than others, they should be placed in a broad menu at the top positions of the pull-downs to minimize access time.
Response times for the alphabetized menus were somewhat longer averaging about 3.2 sec (see Figure 10.4). For these menus, the first and last positions in each list tended to show the fastest access times. The limits of the alphabetic ranges probably served as markers that aided location for these items. This effect occurred both within a pull-down list and across lists for the 3x8 and 4x6 conditions. The access times for the first and last pull-downs are skewed such that time increased with list position for the first pull-down (A-H), but decreased with list position for the last pulldown (Q-Z) with the exception of the first position (Q). Overall access time decreased significantly as the breadth increased and the lists were shorter. The reason for this also lies in the fact that broader menubars display more alphabetic markers that aid in locating items. Consequently, for alphabetic search, designers should increase the number of pulldown lists.
Finally for the randomized menus, response times were much longer, averaging about 4.7 (see Figure 10.5). Subjects seemed to search this list in a linear order viewing the first pulldown from top to bottom, then the second pull-down, and so on until the target was found. A multiple regression analysis of pull-down position and list position for each condition indicated that the time increment for for processing each pull-down was 935 ms for the 3x8 condition, 479 ms for the 4x6 condition, 582 ms for the 6x4 condition, and 346 ms for the 8x3 condition. These differences are due to the time required to visually search each pull-down list. The longer the list, the longer it took to search. The increment for list position within a pull-down was 207 ms for the 3x8 condition, 249 ms for the 4x6 condition, 307 ms for the 6x4 condition, and 154 ms for the 8x3 condition. These increments are be due to both search time and cursor positioning time and reveal the magnitude of added time to locate items buried in a pull-down. Finally, overall response time was somewhat faster for the 3x8 and 4x6 conditions. Apparently search within a few long pull-downs is faster than within a number of short pull-downs.
While only minor differences on the order of .5 sec were found in overall performance due to menu structure, major differences on the order of 2 sec were found due to position in the menu. Consequently, the recommendation is to locate frequently needed items in positions that show the fastest access times. The efficacy of menu location, however, also depends on the user's prior knowledge as to the location of the item. Novice users spend most of their time searching for the location of an item (as simulated by the random menu). Frequently needed items should, consequently, be placed in the pull-down locations searched first (in this case the top left-most pull-down). Experienced users spend most of their time in cursor movement (as simulated by the array menu). Consequently, it doesn't matter in which pull-down a frequently needed item is located, but it is important that it is in a top position.
10.1.2 Minimizing Distance/Maximizing Size. When items are selected using a mouse, selection time is a function of the distance of the cursor to the target and the size of the target according to Fitts' law (see Chapter 8). Consequently, rapid access may be achieved by making items closer to the initial cursor position and larger in physical size. Pull-down menus have the disadvantage in that the larger the target the greater the distance to all but the first item in the list.
An innovative solution to this problem is the pie menu (Hopkins, Callahan, & Weiser, 1987). In the pie menu items are placed at an equal radial distance around the circumference of a circle (see Figure 10.6). The starting or home position of the cursor is at the middle of the circle rather than at the top of a pull-down list. Consequently, all items are equidistant from the home position. The selectable area is a pie-shaped wedge beginning at the center of the circle. Consequently, the distance to the target is very short. Items may be increased in size by making the circle larger. Thus, pie menus seem to have a number of advantages over linear pull-down menus.
In a recent study, performance using pie menus was compared with linear pull-down menus (Callahan, Hopkins, Weiser, & Shneiderman, 1988). Three types of items were used to see if performance on pie menus versus pull-down menus depended on the organization of the items. It was thought that pie menus would be particularly good for items that could be organized around a circle, such as compass locations (e.g., North, NE, East, SE, South, SW, West, and NW). Pull-down menus may be more appropriate for linear organizations of items (e.g., First, Second, Third, Fourth, Fifth, Sixth, Seventh, Eighth). Finally, unclassified groupings of items were used (e.g., Center, Bold, Italic, Font, Move, Copy, Find, Undo) with no prediction as to which would be more appropriate. Five menus were generated for each type of organization and were displayed twice in both the pie and pull-down formats. A completely within-subject design was used in which subjects selected items in all of the menu types and formats in a randomized order of presentation. Subjects had little or no prior experience using the mouse. On each trial, a target item was displayed at the top of the screen. Item selection involved three stages. The user invoked the menu by pressing the mouse button (invocation), held the mouse button down and moved to an item which was then highlighted (browsing), and released the mouse button to confirm the selection (confirmation). Selection times were measured from the point when the mouse button was pressed to when it was released.
Overall response times were significantly faster for pie menus (2.26 sec) than for pull-down menus (2.64 sec ), a performance advantage of about 15% . Although response times for the unclassified items were longer than for the compass and linear types, there was no interaction between menu format and organization of items. Pie menus were uniformly superior across all types of items. Furthermore, fewer selection errors were made using the pie menu; however, the difference was not statistically significant.
A closer analysis of results indicates that the advantage of pie over pull down menus is due to the items in the lower half of the list. Response times for Items 1 through 4 in the list were approximately the same for pie and pull-down menus. However, for Items 5 through 8, response times for the pie menu were considerably faster than for the pull-down menu (see Figure 10.7). Response times increased linearly for pull-down menus from the first to the last item. For pie menus, response times are fastest for the top and bottom points on the circle. Response times appear to increase clockwise around the circle from the top position but decrease sharply at the bottom position where they again increase clockwise around the circle until the top is reached.
Although the items are all equidistant from the home position in pie menus, response times are not uniform. Visual search time probably accounts for this effect. Items in the top and bottom positions of the compass are mostly quickly located. Search then proceeds in a linear manner around the compass.
Pie menus offer a promising alternative to linear pull-down menus. However, it is not yet clear when and how they should be used. Clearly they are limited in the number of items that can be displayed around the circle and require substantially more space on the screen. It is clear that cursor movement time is reduced, but it is not clear what happens with visual search around a circle. Further research is needed to partial out visual search time from movement time in such studies. In addition, it is not clear how experience affects selection time with pie menus. In the study by Callahan et al., subjects were only exposed to the same menu twice. Consequently, subjects had little prior knowledge as to the location of target items. The real advantage of pie menus may be for highly predictable or well-learned menus were visual search time is at a minimum.
10.1.3 Position in the Hierarchy. Hierarchical menus pose a greater problem for rapid access. Each level down the hierarchy requires time and effort. If frequently needed items are buried several levels down, the user must enter a series of menu selections each time the item is required. Many systems place all target items at the same level in the hierarchy for the sake of consistency. Unfortunately, this places all items at the same level of effort.
A better approach may be to move frequently used items up the hierarchy and infrequent items further down. Figure 10.8 shows a system in which primary functions are accessible at the top level of the menu, secondary functions at the next level, and tertiary functions are buried at the bottom. An enhanced telephone provides a good example of such a system. Options to answer the phone, dial emergency numbers, and disconnect are primary functions that are available at the touch of one button. Options to enter telephone numbers into memory, set time and date, and set mode to pulse or touch tone are secondary or tertiary functions that might require several levels of menu selection.
Response times for functions at successive levels can be estimated using the log model discussed in Chapter 7. Figure 10.9 (gray bars) shows an idealized graph of response times for the menu in Figure 10.8. It can be seen that primary functions are accessed faster than secondary functions and secondary functions are accessed faster than tertiary functions. The combination of primary functions and first order categories at the first level increases user response time due to the larger number of items. This problem may be ameliorated, however, by the way in which the menu is laid out. For example, one may place primary items in a separate window or on a separate keypad. If they are cognitively distinct, the number of primary items npf and the number of first order categories nfoc will not summate to a breadth of npf + nfoc.
Ultimately, one must consider the average access time over all menu items. If all items are accessed equally often, the expected access time is just over 7 seconds. However, when the menu utilization profile is factored in, a clear advantage in having primary functions at the first level should emerge. Figure 10.9 (black bars) shows the case where primary functions are accessed with .7 probability, secondary functions with .2 probability, and tertiary functions with .1 probability. The expected access time for this utilization profile is only 5 sec. Thus, expected access time (5 sec) is more than halved over a menu in which all items are at the third level (10.8 sec).
An additional advantage of varying position in the menu hierarchy is that one can intentionally hide specialized functions and make them more difficult to access. The idea is to place items that change important system parameters or could have major negative consequences at the bottom of the hierarchy. Setting the date and time, configuring the printer, specifying terminal emulation parameters, and reformatting the disk are examples of items that might be hidden at lower levels of a menu hierarchy.
A number of researchers have proposed the idea of adaptive menus that seek to match the menu utilization profile over frequency of use (e.g., Berke & Vidal, 1987). For pull-down menus, the system could reposition items in lists to reduce the distance that the cursor must traverse to select high frequency items. In hierarchical menus, high frequency items could percolate up the tree to reduce the length of the path. Although adaptive approaches to rapid access appear promising at first glance, they may carry a fatal flaw. While the system "learns" and adapts the menu to usage patterns, the user is kept guessing as to location of an item. High speed automaticity on the part of the user may be frustrated by a seemingly inconsistent system. A second problem, is that the menu utilization profile may not be just user specific. It may vary drastically depending on task. As soon as the system has adapted to the user browsing through messages in an automatic message handler, the user may shift tasks and start composing messages. To accommodate such abrupt changes in the menu utilization profile the system must have a more sophisticated model of user behavior. Finally, there is yet no empirical research supporting the idea self-adapting menus and their superiority over user controlled rapid access methods.
10.2 Accelerating through the System
Rapid access may be accomplished by location, but often it is judicious to arrange the menu in a semantic, logical, or even alphabetic order without regard to frequency of use and then provide a jump ahead or direct access facility to a subset of high frequency items. This approach is particularly valuable when the menu utilization profile is task or user specific. When the designer cannot accommodate all profiles by menu location, users can accommodate themselves by learning jump ahead commands. Direct access menus allow the user to select frequently used items by a single command and by-pass intermediate layers of menu selections. Direct access is memory mediated in the sense that the distance from the initial state to the target is short circuited via a memory path rather than a menu path. For example, access to an item in a videotext service that is three or four levels down may be achieved by typing its command name (e.g., Stocks, WPost, or Games). Furthermore, direct access may allow the user to completely bypass the menu hierarchy. The user may enter a direct command at any point in the menu to access any other function.
In the case of cursor-based and/or mouse-based menu selection systems, the user enters a command mode typically by the selection of a special function key. A command line prompt is displayed and the user types the desired command. Consequently, this form of direct access switches the user out of menu selection and into a simple command language as a mode of interaction. The empirical issues regarding command entry have been covered in other texts and will not be repeated here. However, the interaction between commands and menu selection is of great interest and is the subject of the following sections.
10.2.1 Alternate Command Keys. The most abbreviated menu bypass commands are single key (e.g., program function keys) or dual-key commands (e.g., Alt-keys or Control keys). Access to menu functions may be facilitated by providing command key selection in predominantly mouse-based menu systems. Often the reason for providing keyboard selection is to eliminate the time that it takes for the hand to travel from the keyboard to the analog input device (homing time). When a user is entering text on the keyboard and must repeatedly make menu selections to change style, the use of command keys can drastically reduce task time.
There are two basic problems with alternative commands keys: learning and relearning. The user must learn the alternate command keys to make the menu selection. Learning is typically facilitated by (a) listing key equivalents in the menu display and (b) by using keys that are associated with the menu alternative such as the first letter of the menu alternative. While incidental learning of the key equivalents is probably minor, the menu itself provides an online help system for the user to find key equivalents when needed. Alternative command keys pose a design problem when a number of them are needed and when several menu names begin with the same letter.
However, some simple rules can be used to generate keybindings. Walker and Olson (1988) describe three rules that result in easy learning and resistance to forgetting:
(1) Different function keys should be used for the highest level of command structure. For example, the Esc key might be used for all system related commands, the Alt key for all deletions commands, and the Control key for all other types of the actions.
(2) Provide a one-to-one mapping between a key press and an appropriate action or object. For example, if there is only one save command (e.g., Save File), it might be accessed with Esc S. Additional keys might be needed to specific different directions or actions and different objects. For example, Alt FW might mean delete forward a word and Alt BC mean delete back a character.
(3) The order of keybindings should follow the English pattern of Verb-Adjective/Adverb-Object. For example, Alt BW might mean delete back a word.
Walker and Olson found that by using these rules they could create a set of keybindings that were easier to learn and more resistant to forgetting than those provided by a standard word processor. Furthermore, they note that the rules lead to keybindings that are more extensible than rules proposed by others (Green & Payne, 1984). This is extremely important as the complexity and functionality of software increases.
As users transfer among a number of different applications, they often encounter the problem that key equivalents do not transfer. Control-U in one application may perform an undo function and in another Control-U set the text input to superscript. As noted in Chapter 7, transfer of learning may be impaired when lexical changes (renaming items) are made from one application to another.
To the extent that designers can adopt standard key equivalents across applications for high frequency items, there should be a positive transfer of performance. Unfortunately, this is a near impossible design objective primarily because of the variability in user specific and task specific menu utilization profiles. For one user /task the key equivalents are fine, but for another the most frequently used menu items have no key equivalents. The only solution may be to allow users to define their own key equivalents in addition to, or as an alternative to, those provided by the application. At present there is no major research on the issue of user defined key equivalents.
10.2.2 Direct Access vs. Type Ahead. When rapid access is required for all terminal items or at least a number larger than the number of command keys available, access may be provided by entering a command string. At least two types of command strings can be used. One method is to provide a mnemonic name for the menu item for direct access. In essence, this implements a vast, but unlisted set of options at each choice point. The user must remember the full name in order to jump to that item.
An alternative to command entry is the type-ahead method in which the user enters a number of sequential choices at one time. All of the choices are processed at one time so that intervening menus are bypassed. Users may enter as many levels as they can recall. If the user enters the entire path, the system implements that function. If the user enters only a partial path, the system implements selections up to that point and displays the next menu in the series. While the direct access method of jump-ahead is essentially all-or-nothing, the type-ahead method allows partial jump-ahead.
The direct access and type-ahead methods were evaluated to determine which (a) takes the least number of trials to learn, (b) produces the lowest error rates, and (c) receives the greater approval by users (Laverson, Norman, & Shneiderman,1987). Thirty-two subjects searched for targets in a hierarchical data base of college course information clustered by academic division, department, and level (see Figure 10.10). In the direct access condition, subjects learned to find targets using five letter names (e.g., HUCMO for human and community resources, MAPST for mathematical and physical sciences; see Figure 10.11). In the type-ahead condition, subjects learned to find targets using a series of five letters normally selected along the path to access the target. Subjects participated in both conditions in a counterbalanced order.
The results indicated that the direct-access method (a) required significantly fewer trials to learn than the type-ahead method, (b) resulted in significantly lower error rates, and (c) significantly reduced response time. Furthermore, subjective evaluations indicated that subjects preferred the direct access method over the type-ahead method. Subjects indicated that direct access commands were easier to learn, easier to recall, easier to use, and resulted in greater overall satisfaction. Most subjects reported that they used rote memory for learning and recalling jump-ahead commands. However, if subjects attempted to derive the commands in the type-ahead method, they had to recall the items chosen in the correct order. If subjects attempted to derive the commands in the direct access method, they had to figure out the nomenclature of names.
Although the study by Laverson et al. favored the direct access method, it is not clear whether the advantage of direct access over type ahead is due to the jump ahead method or due to differences in the meaningfulness of the two sets of commands. Meaningfulness is a very potent factor in the ability to commit information to memory. A study by Jofre & Pino (1987) illustrates this point. They implemented a menu tree in which the concatenation of selection codes produced a pronounceable strings of letters. They note that in the Spanish language pronounceable strings can be easily constructed by alternating consonants and vowels. Subjects that used the pronounceable codes performed significantly better than subjects that used randomly generated codes. Although Jofre et al. did not compare the type-ahead method with the direct access method, it is clear that meaningfulness, or a strong correlate of meaningfulness (i.e., pronounceability) is an important factor in developing menu by-pass commands. It would appear that whichever method can be implemented with greater meaningfulness of the commands should be implemented. In general, direct access allows the designer to generate commands with the highest meaningfulness.
10.3 Speed-Accuracy Tradeoff in Rapid Menu Selection
When users are forced to speed up on a task there is generally an increase in errors. This is known as the speed-accuracy tradeoff. In menu selection this tradeoff is evidenced by wrong selections. Selection errors are generally of two types: discrimination errors and movement errors. Discrimination errors occur when the user selects an item that is similar in physical or semantic features to the target, for example, selecting "Save" rather than "Save As." Movement errors occur when the user attempts to select the correct item, but misses it and selects another. Movement errors generally result in the selection of an item that is one-off from the target, for example, one above or below in a pull-down list or an item keyed next to the correct key on a keyboard (e.g., "v" or "n" rather than "b"). On pull-down menus, the proportion of errors in timed tasks is on the order of 2-4% with about 80% being one-off errors (Norman & Chin, 1988).
Time pressure may be forced on users by the environment or by the system itself. In many cases, the user must complete a menu search before a deadline. In other cases the system may institute a time out if the user has not responded within a set period of time. When the menus are long, the user may be hurried scanning the list and making a selection. With short menus, scanning time is reduced, but the user has to make more selections and potentially more errors. It should be no surprise that empirical results indicate a significant advantage for broad menus. Wallace (1988) investigated menu search under time pressure in hierarchical menus of 256 items varying the breadth and depth in three structures: 28, 44, and 162. While there was no significant difference in the number of searches that timed out using the three menu structures, the narrow/deep structure ( 28) resulted in 174% more errors than the broad/shallow structure ( 162). Clearly, broad/shallow menus are required when users are under time pressure.
10.4 Summary
When users access menus hundreds of times per hour over extended periods of work, they become highly proficient and demand rapid access to frequently used functions. The way in which a menu is organized can go a long way toward decreasing access time. For pull-down menus, frequently accessed items should be placed in the top positions in pull-down lists. In other mouse-based systems, innovative menu design such as pie menus can reduce access time by minimizing cursor movement and maximizing size of the selectable area. The location of pop-up menus and the use of other selection topologies such as grids may also reduce response time.
Access time may also be reduced by minimizing switching from one mode of input to another. Many mouse-based applications provide alternate command keys to eliminate homing time. In hierarchical menus, access time may be reduced by placing frequently used items at higher levels in the menu tree and burying infrequent items at the bottom. Unfortunately, there is often a conflict between frequency placement and logical placement. Logically, a menu item may belong in a group of items four levels down the hierarchy, but its frequency of use may suggest that it should be placed at the top of the tree.
A solution to this problem is to provide rapid access via a jump ahead method. Jump ahead may be provided by either a type-ahead method in which the user enters a series of menu selections all at once or a direct access method in which the user enters a frame name. Research favors the method that can provide the most meaningful string of characters. In general, design favors the use of meaningful frame names for direct access.
Despite the emphasis on rapid access, selection errors in well practiced menus tend to be rare. Generally they are one-off errors in which the user inadvertently selects an item next to the target item. It is only with extreme time constraints that errors appear with any great frequency.
Previous chapters have dealt with the overall efficiency of menu structure and access. These chapters and the research they have reported have assumed a more or less uniform use of menu options. In reality, access of menu options is extremely uneven. Some options are accessed with high frequency at the rate of hundreds of times per session. Style and editing functions in a word processor; shapes, lines, and patterns in a graphics package; and rows, columns, and functions in a spread sheet program are selected with great frequency. Other items, however, are accessed only rarely though they may nevertheless be vital to the operation. For example, options to open or close a file may only be used once per session but are absolutely necessary. The differences in frequency of use can be very drastic. Figure 10.1 shows the frequency of use of 49 options in the pull-down menu bar of MacWrite(TM) accessed by one user.
Laverson (1985) introduced the notion of a "menu utilization profile" to characterize the uneven access of items in a hierarchical menu system. He hypothesized that the efficiency of different rapid access methods would depend on the utilization profile. His ideas need to be extended, however, since the differential access of items varies as function of two important factors. The frequency of use may be either task specific in that the particular task performed determines the frequency of use or user specific in that the type of user determines the frequency of use. As an example of task specific utilization profiles, the frequency of use of menu items in a word processor varies substantially between the task of composing a paper and editing a paper. As an example of user specific utilization profiles, the frequency of complex, technical items may higher for experienced users and simple, direct items for inexperienced users of a graphic program.
Two basic methods of rapid access have been explored. One is to arrange items in the menu frame and/or hierarchy to minimize the distance from the initial menu state to the location of high frequency targets. The second is to provide various forms of menu "bypass," "jump-ahead," or direct commands that reduce the distance by defining shortcuts through the menu space. In this chapter, studies will reviewed that bear on the efficacy and importance of such approaches.
10.1 Location in the Menu
Although it may seem obvious that highly used options should be the most readily accessible items, many systems bury these items in the middle or at the bottom of menu lists and hierarchies. Such designs may be driven in part by the fallacy that menu selection makes everything uniformly rapid and easy. The idea is that since menus are inherently easy to use, all items must be easy to access. While this may be the perception of novice users and some designers, experienced users all too quickly detect the bottlenecks and frustratingly slow access to frequently required items.
10.1.1 Pull-down Menus. This is particularly true of pull-down menus. All items appear to be easily and rapidly available. Consequently, the order of items in pull-down menus is rarely designed around frequency of use.
Empirical results, however, indicate that access time depends on the location of items in the pull-down menu. In one experiment, 24 items were presented in 4 different menu configurations varying in the number and length of the pull-down menus (Norman & Chin, 1988). The conditions were as follows: 3x8, 4x6, 6x4, and 8x3, where the first number in each product is the number of pull-down lists and the second number is the length of each list. In addition, prior knowledge of target location was varied by using three different types of menus (see Figure 10.2). For array menus, items were simply 24 letter-number combinations. The letter indicated which pull-down list to access and the number indicated the target item in that list. Array menus provided the user with complete prior knowledge as to the location of target items before menus were pulled down. For alphabetized menus, items were 24 common words each beginning with a different letter of the alphabet. Pull-down lists indicated a letter range and items in each list were in alphabetical order. Although the menus provided a degree of prior knowledge as to location, it was somewhat more difficult and incomplete in comparison to the array menus. Finally, for randomized menus, the items were the same 24 words, but on each trial the words were cast in a different random order across all of the lists. The random menus provided no prior information to the user as to location. In effect this condition simulated the novice user on all trials having no experience.
Subjects were randomly assigned to one type of menu item and were tested on each of the four menu configurations in a counterbalanced order. Sixteen subjects participated in each of the three menu conditions. For each configuration, subjects were given a set of practice trials and then tested on two replications of each of the 24 target items in a randomized block order. On each trial the target item was displayed in a small window just to the left of center on a Macintosh(TM) screen. Subjects used a mouse to pull down and select the target item from the menu bar at the top of the screen. Times were recorded from the onset of the target to the selection response. If the subject selected an item that was not the target, it was recorded, a tone was sounded, and the target item was recycled in the pool of targets to be presented.
Response times were fairly fast for the array menus, averaging about 2.6 sec and varied significantly with target location (see Figure 10.3). Access times were fastest for the first positions in each pull-down and increased linearly with position down the list. Response times, however, did not vary with the position of the pull-down but only with position of items in the list. Linear regression equations of access time as a function of list position indicated that the time increment per item in the pull-down menu was 104 ms for the 3x8 condition, 147 for the 4x6 condition, 251 for the 6x4 condition, and 228 for the 8x3 condition. Consequently, although the lists were shorter in the 6x4 and 8x3 conditions, the pull-down time per item position was significantly longer than in the 3x8 and 4x6 conditions. Access times to the pull-down lists (as indicated by the intercept of the regression lines) were fairly similar across conditions, and there was no significant difference in overall response time among the four conditions. It would appear that for highly predictable menus, the trade-off between list position and pull-down time cancels out when items are accessed with equal frequency. However, if the designer knows that some items will be more frequently accessed than others, they should be placed in a broad menu at the top positions of the pull-downs to minimize access time.
Response times for the alphabetized menus were somewhat longer averaging about 3.2 sec (see Figure 10.4). For these menus, the first and last positions in each list tended to show the fastest access times. The limits of the alphabetic ranges probably served as markers that aided location for these items. This effect occurred both within a pull-down list and across lists for the 3x8 and 4x6 conditions. The access times for the first and last pull-downs are skewed such that time increased with list position for the first pull-down (A-H), but decreased with list position for the last pulldown (Q-Z) with the exception of the first position (Q). Overall access time decreased significantly as the breadth increased and the lists were shorter. The reason for this also lies in the fact that broader menubars display more alphabetic markers that aid in locating items. Consequently, for alphabetic search, designers should increase the number of pulldown lists.
Finally for the randomized menus, response times were much longer, averaging about 4.7 (see Figure 10.5). Subjects seemed to search this list in a linear order viewing the first pulldown from top to bottom, then the second pull-down, and so on until the target was found. A multiple regression analysis of pull-down position and list position for each condition indicated that the time increment for for processing each pull-down was 935 ms for the 3x8 condition, 479 ms for the 4x6 condition, 582 ms for the 6x4 condition, and 346 ms for the 8x3 condition. These differences are due to the time required to visually search each pull-down list. The longer the list, the longer it took to search. The increment for list position within a pull-down was 207 ms for the 3x8 condition, 249 ms for the 4x6 condition, 307 ms for the 6x4 condition, and 154 ms for the 8x3 condition. These increments are be due to both search time and cursor positioning time and reveal the magnitude of added time to locate items buried in a pull-down. Finally, overall response time was somewhat faster for the 3x8 and 4x6 conditions. Apparently search within a few long pull-downs is faster than within a number of short pull-downs.
While only minor differences on the order of .5 sec were found in overall performance due to menu structure, major differences on the order of 2 sec were found due to position in the menu. Consequently, the recommendation is to locate frequently needed items in positions that show the fastest access times. The efficacy of menu location, however, also depends on the user's prior knowledge as to the location of the item. Novice users spend most of their time searching for the location of an item (as simulated by the random menu). Frequently needed items should, consequently, be placed in the pull-down locations searched first (in this case the top left-most pull-down). Experienced users spend most of their time in cursor movement (as simulated by the array menu). Consequently, it doesn't matter in which pull-down a frequently needed item is located, but it is important that it is in a top position.
10.1.2 Minimizing Distance/Maximizing Size. When items are selected using a mouse, selection time is a function of the distance of the cursor to the target and the size of the target according to Fitts' law (see Chapter 8). Consequently, rapid access may be achieved by making items closer to the initial cursor position and larger in physical size. Pull-down menus have the disadvantage in that the larger the target the greater the distance to all but the first item in the list.
An innovative solution to this problem is the pie menu (Hopkins, Callahan, & Weiser, 1987). In the pie menu items are placed at an equal radial distance around the circumference of a circle (see Figure 10.6). The starting or home position of the cursor is at the middle of the circle rather than at the top of a pull-down list. Consequently, all items are equidistant from the home position. The selectable area is a pie-shaped wedge beginning at the center of the circle. Consequently, the distance to the target is very short. Items may be increased in size by making the circle larger. Thus, pie menus seem to have a number of advantages over linear pull-down menus.
In a recent study, performance using pie menus was compared with linear pull-down menus (Callahan, Hopkins, Weiser, & Shneiderman, 1988). Three types of items were used to see if performance on pie menus versus pull-down menus depended on the organization of the items. It was thought that pie menus would be particularly good for items that could be organized around a circle, such as compass locations (e.g., North, NE, East, SE, South, SW, West, and NW). Pull-down menus may be more appropriate for linear organizations of items (e.g., First, Second, Third, Fourth, Fifth, Sixth, Seventh, Eighth). Finally, unclassified groupings of items were used (e.g., Center, Bold, Italic, Font, Move, Copy, Find, Undo) with no prediction as to which would be more appropriate. Five menus were generated for each type of organization and were displayed twice in both the pie and pull-down formats. A completely within-subject design was used in which subjects selected items in all of the menu types and formats in a randomized order of presentation. Subjects had little or no prior experience using the mouse. On each trial, a target item was displayed at the top of the screen. Item selection involved three stages. The user invoked the menu by pressing the mouse button (invocation), held the mouse button down and moved to an item which was then highlighted (browsing), and released the mouse button to confirm the selection (confirmation). Selection times were measured from the point when the mouse button was pressed to when it was released.
Overall response times were significantly faster for pie menus (2.26 sec) than for pull-down menus (2.64 sec ), a performance advantage of about 15% . Although response times for the unclassified items were longer than for the compass and linear types, there was no interaction between menu format and organization of items. Pie menus were uniformly superior across all types of items. Furthermore, fewer selection errors were made using the pie menu; however, the difference was not statistically significant.
A closer analysis of results indicates that the advantage of pie over pull down menus is due to the items in the lower half of the list. Response times for Items 1 through 4 in the list were approximately the same for pie and pull-down menus. However, for Items 5 through 8, response times for the pie menu were considerably faster than for the pull-down menu (see Figure 10.7). Response times increased linearly for pull-down menus from the first to the last item. For pie menus, response times are fastest for the top and bottom points on the circle. Response times appear to increase clockwise around the circle from the top position but decrease sharply at the bottom position where they again increase clockwise around the circle until the top is reached.
Although the items are all equidistant from the home position in pie menus, response times are not uniform. Visual search time probably accounts for this effect. Items in the top and bottom positions of the compass are mostly quickly located. Search then proceeds in a linear manner around the compass.
Pie menus offer a promising alternative to linear pull-down menus. However, it is not yet clear when and how they should be used. Clearly they are limited in the number of items that can be displayed around the circle and require substantially more space on the screen. It is clear that cursor movement time is reduced, but it is not clear what happens with visual search around a circle. Further research is needed to partial out visual search time from movement time in such studies. In addition, it is not clear how experience affects selection time with pie menus. In the study by Callahan et al., subjects were only exposed to the same menu twice. Consequently, subjects had little prior knowledge as to the location of target items. The real advantage of pie menus may be for highly predictable or well-learned menus were visual search time is at a minimum.
10.1.3 Position in the Hierarchy. Hierarchical menus pose a greater problem for rapid access. Each level down the hierarchy requires time and effort. If frequently needed items are buried several levels down, the user must enter a series of menu selections each time the item is required. Many systems place all target items at the same level in the hierarchy for the sake of consistency. Unfortunately, this places all items at the same level of effort.
A better approach may be to move frequently used items up the hierarchy and infrequent items further down. Figure 10.8 shows a system in which primary functions are accessible at the top level of the menu, secondary functions at the next level, and tertiary functions are buried at the bottom. An enhanced telephone provides a good example of such a system. Options to answer the phone, dial emergency numbers, and disconnect are primary functions that are available at the touch of one button. Options to enter telephone numbers into memory, set time and date, and set mode to pulse or touch tone are secondary or tertiary functions that might require several levels of menu selection.
Response times for functions at successive levels can be estimated using the log model discussed in Chapter 7. Figure 10.9 (gray bars) shows an idealized graph of response times for the menu in Figure 10.8. It can be seen that primary functions are accessed faster than secondary functions and secondary functions are accessed faster than tertiary functions. The combination of primary functions and first order categories at the first level increases user response time due to the larger number of items. This problem may be ameliorated, however, by the way in which the menu is laid out. For example, one may place primary items in a separate window or on a separate keypad. If they are cognitively distinct, the number of primary items npf and the number of first order categories nfoc will not summate to a breadth of npf + nfoc.
Ultimately, one must consider the average access time over all menu items. If all items are accessed equally often, the expected access time is just over 7 seconds. However, when the menu utilization profile is factored in, a clear advantage in having primary functions at the first level should emerge. Figure 10.9 (black bars) shows the case where primary functions are accessed with .7 probability, secondary functions with .2 probability, and tertiary functions with .1 probability. The expected access time for this utilization profile is only 5 sec. Thus, expected access time (5 sec) is more than halved over a menu in which all items are at the third level (10.8 sec).
An additional advantage of varying position in the menu hierarchy is that one can intentionally hide specialized functions and make them more difficult to access. The idea is to place items that change important system parameters or could have major negative consequences at the bottom of the hierarchy. Setting the date and time, configuring the printer, specifying terminal emulation parameters, and reformatting the disk are examples of items that might be hidden at lower levels of a menu hierarchy.
A number of researchers have proposed the idea of adaptive menus that seek to match the menu utilization profile over frequency of use (e.g., Berke & Vidal, 1987). For pull-down menus, the system could reposition items in lists to reduce the distance that the cursor must traverse to select high frequency items. In hierarchical menus, high frequency items could percolate up the tree to reduce the length of the path. Although adaptive approaches to rapid access appear promising at first glance, they may carry a fatal flaw. While the system "learns" and adapts the menu to usage patterns, the user is kept guessing as to location of an item. High speed automaticity on the part of the user may be frustrated by a seemingly inconsistent system. A second problem, is that the menu utilization profile may not be just user specific. It may vary drastically depending on task. As soon as the system has adapted to the user browsing through messages in an automatic message handler, the user may shift tasks and start composing messages. To accommodate such abrupt changes in the menu utilization profile the system must have a more sophisticated model of user behavior. Finally, there is yet no empirical research supporting the idea self-adapting menus and their superiority over user controlled rapid access methods.
10.2 Accelerating through the System
Rapid access may be accomplished by location, but often it is judicious to arrange the menu in a semantic, logical, or even alphabetic order without regard to frequency of use and then provide a jump ahead or direct access facility to a subset of high frequency items. This approach is particularly valuable when the menu utilization profile is task or user specific. When the designer cannot accommodate all profiles by menu location, users can accommodate themselves by learning jump ahead commands. Direct access menus allow the user to select frequently used items by a single command and by-pass intermediate layers of menu selections. Direct access is memory mediated in the sense that the distance from the initial state to the target is short circuited via a memory path rather than a menu path. For example, access to an item in a videotext service that is three or four levels down may be achieved by typing its command name (e.g., Stocks, WPost, or Games). Furthermore, direct access may allow the user to completely bypass the menu hierarchy. The user may enter a direct command at any point in the menu to access any other function.
In the case of cursor-based and/or mouse-based menu selection systems, the user enters a command mode typically by the selection of a special function key. A command line prompt is displayed and the user types the desired command. Consequently, this form of direct access switches the user out of menu selection and into a simple command language as a mode of interaction. The empirical issues regarding command entry have been covered in other texts and will not be repeated here. However, the interaction between commands and menu selection is of great interest and is the subject of the following sections.
10.2.1 Alternate Command Keys. The most abbreviated menu bypass commands are single key (e.g., program function keys) or dual-key commands (e.g., Alt-keys or Control keys). Access to menu functions may be facilitated by providing command key selection in predominantly mouse-based menu systems. Often the reason for providing keyboard selection is to eliminate the time that it takes for the hand to travel from the keyboard to the analog input device (homing time). When a user is entering text on the keyboard and must repeatedly make menu selections to change style, the use of command keys can drastically reduce task time.
There are two basic problems with alternative commands keys: learning and relearning. The user must learn the alternate command keys to make the menu selection. Learning is typically facilitated by (a) listing key equivalents in the menu display and (b) by using keys that are associated with the menu alternative such as the first letter of the menu alternative. While incidental learning of the key equivalents is probably minor, the menu itself provides an online help system for the user to find key equivalents when needed. Alternative command keys pose a design problem when a number of them are needed and when several menu names begin with the same letter.
However, some simple rules can be used to generate keybindings. Walker and Olson (1988) describe three rules that result in easy learning and resistance to forgetting:
(1) Different function keys should be used for the highest level of command structure. For example, the Esc key might be used for all system related commands, the Alt key for all deletions commands, and the Control key for all other types of the actions.
(2) Provide a one-to-one mapping between a key press and an appropriate action or object. For example, if there is only one save command (e.g., Save File), it might be accessed with Esc S. Additional keys might be needed to specific different directions or actions and different objects. For example, Alt FW might mean delete forward a word and Alt BC mean delete back a character.
(3) The order of keybindings should follow the English pattern of Verb-Adjective/Adverb-Object. For example, Alt BW might mean delete back a word.
Walker and Olson found that by using these rules they could create a set of keybindings that were easier to learn and more resistant to forgetting than those provided by a standard word processor. Furthermore, they note that the rules lead to keybindings that are more extensible than rules proposed by others (Green & Payne, 1984). This is extremely important as the complexity and functionality of software increases.
As users transfer among a number of different applications, they often encounter the problem that key equivalents do not transfer. Control-U in one application may perform an undo function and in another Control-U set the text input to superscript. As noted in Chapter 7, transfer of learning may be impaired when lexical changes (renaming items) are made from one application to another.
To the extent that designers can adopt standard key equivalents across applications for high frequency items, there should be a positive transfer of performance. Unfortunately, this is a near impossible design objective primarily because of the variability in user specific and task specific menu utilization profiles. For one user /task the key equivalents are fine, but for another the most frequently used menu items have no key equivalents. The only solution may be to allow users to define their own key equivalents in addition to, or as an alternative to, those provided by the application. At present there is no major research on the issue of user defined key equivalents.
10.2.2 Direct Access vs. Type Ahead. When rapid access is required for all terminal items or at least a number larger than the number of command keys available, access may be provided by entering a command string. At least two types of command strings can be used. One method is to provide a mnemonic name for the menu item for direct access. In essence, this implements a vast, but unlisted set of options at each choice point. The user must remember the full name in order to jump to that item.
An alternative to command entry is the type-ahead method in which the user enters a number of sequential choices at one time. All of the choices are processed at one time so that intervening menus are bypassed. Users may enter as many levels as they can recall. If the user enters the entire path, the system implements that function. If the user enters only a partial path, the system implements selections up to that point and displays the next menu in the series. While the direct access method of jump-ahead is essentially all-or-nothing, the type-ahead method allows partial jump-ahead.
The direct access and type-ahead methods were evaluated to determine which (a) takes the least number of trials to learn, (b) produces the lowest error rates, and (c) receives the greater approval by users (Laverson, Norman, & Shneiderman,1987). Thirty-two subjects searched for targets in a hierarchical data base of college course information clustered by academic division, department, and level (see Figure 10.10). In the direct access condition, subjects learned to find targets using five letter names (e.g., HUCMO for human and community resources, MAPST for mathematical and physical sciences; see Figure 10.11). In the type-ahead condition, subjects learned to find targets using a series of five letters normally selected along the path to access the target. Subjects participated in both conditions in a counterbalanced order.
The results indicated that the direct-access method (a) required significantly fewer trials to learn than the type-ahead method, (b) resulted in significantly lower error rates, and (c) significantly reduced response time. Furthermore, subjective evaluations indicated that subjects preferred the direct access method over the type-ahead method. Subjects indicated that direct access commands were easier to learn, easier to recall, easier to use, and resulted in greater overall satisfaction. Most subjects reported that they used rote memory for learning and recalling jump-ahead commands. However, if subjects attempted to derive the commands in the type-ahead method, they had to recall the items chosen in the correct order. If subjects attempted to derive the commands in the direct access method, they had to figure out the nomenclature of names.
Although the study by Laverson et al. favored the direct access method, it is not clear whether the advantage of direct access over type ahead is due to the jump ahead method or due to differences in the meaningfulness of the two sets of commands. Meaningfulness is a very potent factor in the ability to commit information to memory. A study by Jofre & Pino (1987) illustrates this point. They implemented a menu tree in which the concatenation of selection codes produced a pronounceable strings of letters. They note that in the Spanish language pronounceable strings can be easily constructed by alternating consonants and vowels. Subjects that used the pronounceable codes performed significantly better than subjects that used randomly generated codes. Although Jofre et al. did not compare the type-ahead method with the direct access method, it is clear that meaningfulness, or a strong correlate of meaningfulness (i.e., pronounceability) is an important factor in developing menu by-pass commands. It would appear that whichever method can be implemented with greater meaningfulness of the commands should be implemented. In general, direct access allows the designer to generate commands with the highest meaningfulness.
10.3 Speed-Accuracy Tradeoff in Rapid Menu Selection
When users are forced to speed up on a task there is generally an increase in errors. This is known as the speed-accuracy tradeoff. In menu selection this tradeoff is evidenced by wrong selections. Selection errors are generally of two types: discrimination errors and movement errors. Discrimination errors occur when the user selects an item that is similar in physical or semantic features to the target, for example, selecting "Save" rather than "Save As." Movement errors occur when the user attempts to select the correct item, but misses it and selects another. Movement errors generally result in the selection of an item that is one-off from the target, for example, one above or below in a pull-down list or an item keyed next to the correct key on a keyboard (e.g., "v" or "n" rather than "b"). On pull-down menus, the proportion of errors in timed tasks is on the order of 2-4% with about 80% being one-off errors (Norman & Chin, 1988).
Time pressure may be forced on users by the environment or by the system itself. In many cases, the user must complete a menu search before a deadline. In other cases the system may institute a time out if the user has not responded within a set period of time. When the menus are long, the user may be hurried scanning the list and making a selection. With short menus, scanning time is reduced, but the user has to make more selections and potentially more errors. It should be no surprise that empirical results indicate a significant advantage for broad menus. Wallace (1988) investigated menu search under time pressure in hierarchical menus of 256 items varying the breadth and depth in three structures: 28, 44, and 162. While there was no significant difference in the number of searches that timed out using the three menu structures, the narrow/deep structure ( 28) resulted in 174% more errors than the broad/shallow structure ( 162). Clearly, broad/shallow menus are required when users are under time pressure.
10.4 Summary
When users access menus hundreds of times per hour over extended periods of work, they become highly proficient and demand rapid access to frequently used functions. The way in which a menu is organized can go a long way toward decreasing access time. For pull-down menus, frequently accessed items should be placed in the top positions in pull-down lists. In other mouse-based systems, innovative menu design such as pie menus can reduce access time by minimizing cursor movement and maximizing size of the selectable area. The location of pop-up menus and the use of other selection topologies such as grids may also reduce response time.
Access time may also be reduced by minimizing switching from one mode of input to another. Many mouse-based applications provide alternate command keys to eliminate homing time. In hierarchical menus, access time may be reduced by placing frequently used items at higher levels in the menu tree and burying infrequent items at the bottom. Unfortunately, there is often a conflict between frequency placement and logical placement. Logically, a menu item may belong in a group of items four levels down the hierarchy, but its frequency of use may suggest that it should be placed at the top of the tree.
A solution to this problem is to provide rapid access via a jump ahead method. Jump ahead may be provided by either a type-ahead method in which the user enters a series of menu selections all at once or a direct access method in which the user enters a frame name. Research favors the method that can provide the most meaningful string of characters. In general, design favors the use of meaningful frame names for direct access.
Despite the emphasis on rapid access, selection errors in well practiced menus tend to be rare. Generally they are one-off errors in which the user inadvertently selects an item next to the target item. It is only with extreme time constraints that errors appear with any great frequency.
No comments:
Post a Comment