Dennis Wixon and Michael Good
Digital Equipment Corporation
Nashua, New Hampshire, USA
Reprinted with permission from Proceedings of the Human Factors Society 31st Annual Meeting (October 19-22, 1987, New York), Volume 1, pp. 571-575. Copyright © 1987 by the Human Factors and Ergonomics Society. All rights reserved.
The weaknesses inherent in categorizing interfaces are discussed. Questions are raised about using categorical approaches in science and design. Alternative approaches are suggested with an emphasis on dimensional scope and contextual sensitivity. It is argued that interfaces should be seen in terms of their transparency and support for breakdown.1
Many questions can motivate us as we study interaction between humans and computers. We can be motivated to understand the nature of human performance for its own sake. We can be motivated to make a given interface the best that it can be given the time and resource constraints. This paper is motivated by the question: Are we doing all we can to deepen our understanding of how people and computers interact?
We begin by examining the ways we have conceptualized systems in the past, and consider their strengths and weaknesses. Our aim is to provoke debate and in so doing move to a deeper understanding of our approaches to user interfaces.
In 1983 Ben Shneiderman introduced the term direct manipulation to characterize a new class of systems which seemed to offer new possibilities for human-computer interaction. These systems were distinguished from command systems and menu systems. We will take Shneiderman’s descriptions of interface styles (1987) as a starting point for our discussion.
Shneiderman is cautious in distinguishing menus from other interface styles. While recognizing that the distinctions can be blurred, he emphasizes criteria such as “how much the system offers on the display at the moment the selection is made, the form and content of item selection, and what task domain knowledge is necessary for users to succeed” (p. 86). He asserts that menu systems are especially effective for users who are relatively new to a particular system, but cautions that using a menu system does not guarantee an appealing and/or effective system.
Shneiderman distinguishes command systems from menu systems “by the fact that the users of command languages must recall notation and initiate action” (p. 138). In contrast, users of menu systems recognize and choose from a set of alternatives. Command languages are also differentiated from programming languages by their ephemeral nature, although the role of command histories and macros are recognized. Command systems are considered especially appropriate for experienced, frequent system users, and as a method for overcoming certain hardware limitations such as limited screen space and slow display rates.
Direct-manipulation systems are characterized by “visibility of objects and actions of interest, rapid reversible incremental actions, and replacement of complex command language syntax by direct manipulation of the object of interest” (p.180). Examples include display editors (editors which do not require special commands to view results of operations), spreadsheets, spatial data management, and video games. Shneiderman associates direct manipulation systems with enthusiastic responses from users and positive feelings, but again cautions that interface style is not a panacea.
Interface Style Eclecticism
Problems with These Categories
Shneiderman’s descriptions of interface styles seem to reflect some sense in which the systems are different. At a gross level there would probably be little debate if human factors professionals were asked to classify real systems. However, as a way of defining systems, there are problems with these classifications when we examine them more closely.
For example, menus and command systems are not systematically differentiated. Since menus are differentiated from commands in terms of the response of the user, active versus reactive, one might find the same system changing its classification. If the menu system allowed type-ahead (users can type responses to menus which have not yet appeared and their appearance is bypassed), and users became familiar with the system and used this feature extensively, than the usage of the system has “changed” from a menu system to a command system. This is characteristic of how people in our field studies routinely use systems with relatively deep menus. Similarly, as Shneiderman points out, command systems often integrate a menu function, where the user can select from a list of choices.
The criteria for direct manipulation systems also have problems. For example, many objects of interest in direct manipulation systems are not immediately visible. Menus must be pulled down, and files scrolled. Some of the needed information is obscured by other information. In addition, not all direct manipulation systems have rapid operations – commercially available systems in this category have had both slow and fast response times.
The description of direct manipulation systems also appears to encompass very different types of systems. Display editors and spreadsheets on character-cell terminals are grouped under direct manipulation along with more graphical systems like computer games, painting programs, and CAD/CAM programs. Yet there is some level at which the graphical, bit-mapped programs seem to be in a different class than the more textual, character-cell programs.
Many interactive systems use an eclectic mix of styles. The example of MacWrite2 on page 93 is used to show a pull-down menu and thus is included in the chapter on menus. Yet, it is a display text editor and as a result could be classified as a direct-manipulation system. Additionally, the user must take a positive action to get this list (i.e., click and hold on the style item). But such action seems inconsistent with the view of the menu user as reactive. (That aspect of the definition is probably rooted in the image of users waiting for screens to repaint menus which take up the entire screen). Other parts of the MacWrite system use still other interface techniques, like dialog boxes (an example of form fill-in style).
The more we consider real world examples the more it becomes clear that the terms command, menu, and direct manipulation do not apply to systems as wholes. Systems, whether well designed or poorly designed, are mixtures of styles. Interface styles are used in different portions of the system as the designers feel it is appropriate. For example, some popular spreadsheets combine a direct view of the results with menus and sophisticated macro-programming facilities, which are presented in editable command languages.
Limits of Categories
At a gross level, these descriptions of interface style may be useful despite the problems which arise when studied at a more detailed level. We do not believe that the problems arise from the definition of this particular set of categories, but from the limits of this sort of categorical analysis.
Categorization is a simple and primary form of definition, based on the similarity between objects. This fundamental perception of similarity is what Pepper (1970), in his analysis of explanatory systems, has called a “root metaphor” for cognition. As such it is the basis of many powerful philosophical systems, particularly those derived from ancient Greek philosophy.
Categorization has its limitations. While making major contributions in such sciences as biology (animal and plant classification) and chemistry (classification of elements), it has rarely been successfully applied to human artifacts. It just doesn’t seem to be the case that we can offer adequate definitions of things like chairs. If we look at critical attributes like “has four legs,” then we exclude beanbag chairs. If we look in terms of functions, like “people sit on it,” we could include tables and window sills too. If we can’t adequately define things like chairs how could we expect to define interactive systems?
Should we even make an attempt? Shneiderman points out that a number of factors affect the usability of a given class of system. In our own research we have found command, menu, and iconic systems at all levels of usability and acceptance.
In a fundamental sense Shneiderman is correct. Quibbling about particular definitions is irrelevant. But if interfaces styles cannot be defined in a rigorous way, how can we make generalizations about them? At a more fundamental level the definitions themselves and even the very concept of defining things in this way may be irrelevant to good design and good science. Even if we could offer better definitions of classes of systems — definitions that classified systems in ways consistent with our intuitions — one would be no further in understanding what makes systems usable or enjoyable.
Definitions like these lack dimensional scope and context sensitivity. In terms of dimensional scope, we cannot say how much “direct-manipulation” a system has. Instead, the present analysis establishes direct manipulation as a formal category. A system falls into the direct-manipulation category or it does not. In contrast to such an approach, much of the progress in physical sciences can be traced to setting up dimensions so degrees of things could be measured and functionally related to degrees of other things. Similarly in terms of context sensitivity, the categories offered break down when we look at systems as people actually use them. The context of use of a system determines a large part of its effectiveness. Formal definitions intrinsically and necessarily attempt to ignore context.
Focus and Context
Few would argue with the proposition that good design stems from an intimate acquaintance with the user, the task, and the context of use. But when examined more completely this seemingly simple proposition has profound implications for both design and research.
Beginning with a focus on the user’s work in context implies that the categories of analysis should not be determined before the study begins. Instead one begins with the phenomena of work in context and lets the categories emerge from that study, evolving over time. It is better to consider the study of people using computers in terms of concept formation as compared to hypothesis testing or classification. Sensitivity to the interrelation between work and the interface is what is critical. Such sensitivity produces products like the spreadsheet which began with an understanding of how accountants worked or of the ledgers they worked with and the forecasting they had to do.
Ease of use and ease of learning should not be considered properties of systems just as novice and expert should not be considered properties of people. Ease of learning and use derive from an examination of the details of the operation of the software in relation to what the user is trying to do and what the user brings to the situation. While at some general level this is acknowledged, its implications are not fully appreciated. A set of fixed categories or measures will blind one to other aspects of the situation. An openness is absolutely required to appreciate complexities of what makes software easy or hard to use.
The implication seems to be that there are no generalizations about systems that we can make: each system is unique and must be studied in context of the users and tasks. While each system is unique and in a sense each user interacting with each system at each moment is unique, the designer or the researchers need not approach each system in a completely unfocused way as a tabula rasa. Rather the researcher brings a background based on past experience with systems that forms a set of horizons against which the new system is seen. The use of past experience provides a means by which the current events can stand out.
Transparency and Support for Breakdowns
Well-designed direct-manipulation systems seem to have high values of two elements of usability. transparency and support for breakdowns. These concepts arise in the philosophical work of Martin Heidegger, and have been useful in our field studies of system usability.
In our everyday experience with human tools, we are usually not consciously aware of our use of the tools. In this sense the tools are transparent (or ready-to-hand, in Heidegger’s terminology). Winograd and Flores (1986) present the example of a writer using a word processor. The writer thinks of words which appear on the screen. None of the complex devices used to make the words appear on the screen — including arms and hands, a keyboard, a screen, and the computer — is present to the writer except when something goes wrong, and there is a breaking down. If the screen appears garbled, for instance, a component called the “screen manager” may emerge with a “buggy” property.
Winograd and Flores also mention Heidegger’s example of someone using a hammer. When people hammer, the hammer does not exist to them unless something breaks down; for instance, if the hammer slips and mars the wood.
In observing how people use systems, it appears that systems that we would commonly call “direct manipulation” have the potential, when well-designed, to be more transparent to a larger number of users than well-designed systems which rely more on command- and menu-interaction techniques. Shneiderman briefly mentions transparency as an element of the usability of direct manipulation systems. For instance, in discussing the attributes of these systems, he mentions that “error messages are rarely needed” (p.202). This is one element which results from the transparency of a system. What makes these systems transparent? We do not believe that it is due to certain properties which define direct manipulation systems. If we start off with this categorical approach to interface style, we may blind ourselves to the elements which are really contributing to the user’s experience of usability.
By concentrating on desirable qualities of interactive systems, rather than desirable categories of interactive systems, we may be more receptive to the utility of new interface technologies. Current direct-manipulation systems may work well by current standards, but certainly other techniques have great potential for providing transparent interfaces. Leaving aside the technical and social problems, continuous voice recognition certainly could provide a more transparent interface than direct-manipulation systems.
Engineering transparency into systems critically depends on an intimate understanding of the user’s work, the context of this work, and the background the user brings to bear on the work. To use Shneiderman’s words, you must know what “the objects of interest” are before you endeavor to make them visible. An unproductive and frustrating system is one that shows you what you don’t want to see. Many computer systems, of whatever interface style, are overly complex for the jobs users need done (Ledgard, Singer, and Whiteside, 1981). A direct manipulation interface to an overly-complex system may worsen the usability problems by making the unneeded complexity even more obvious to the user.
A transparent system may be a system in which breakdowns occur less frequently. However, breakdowns are unavoidable, and another important consideration is how well the system supports breakdowns. Does the system provide recovery procedures for the user in event of an error? Does the system take advantage of the breakdown to give the user a greater understanding of the system, which could make the system more usable in the future? Or does the system provide terse, hostile, or non-existent messages in these cases?
Direct-manipulation systems often provide greater support for breakdown than command and menu systems. For example a single step undo is common. We could argue whether this is a property of direct-manipulation systems or not. It seems more productive to concentrate instead on what makes systems supportive of breakdowns, and to view this as a dimension of usability.
The goal of this paper is to change the way designers and researchers think. Rather than think in terms of immutable system categories, designers and researchers are urged to think in terms of dimensions — dimensions for usability. What excites people about high-quality direct manipulation systems may not be the properties of the direct manipulation interface style, but the higher levels of certain aspects of usability that have been found in many of these systems.
At the same time, dimensions are only an initial focus, and a horizon against which to study the use of systems in context. Initial concepts should change as the progress in studying people using systems in the context of their daily work. Categories or dimensions are only a starting point for either research or design.
1The views expressed in this paper are those of the authors and do not necessarily reflect the views of Digital Equipment Corporation.
2MacWrite is a trademark of Apple Computer, Inc.