7 Dec 2006
Visual Feedback: Why Modes Kill
Let me set the scene. It’s the comedy film “Airplane“. The flight crew is violently ill and Striker, a shell-shocked, former pilot, is forced to land a jet full of passengers in dire need of medical attention. The air is heavy with fog, rain pounds on the cockpit windows. Over the static-filled radio comes the voice of ground control desperately talking Striker through the landing.
Striker: No, never.
Ground Control: Thinking the radio is off. #@&*#! This is a waste of time… there’s no way he can land that plane. Striker starts to tremble.
How did Ground Control make this mistake? The answer is simple. Mode error.
Don Norman defines mode errors as occurring when a user misclassifies a situation resulting in actions which are appropriate for the conception of the situation but inappropriate for the true situation. In Airplane, the action could not have been more inappropriate for the situation.
Mode errors are a ubiquitous bane: they cause us to lose our work and to kill hundreds of people. Despite the millions of dollars spent on the design of airplane cockpits, between 1988 and 1996 five fatal airline crashes were the direct result of mode errors. Many more crashes were probably indirectly caused by mode errors. What’s the lesson? If a system contains modes, people will make mode errors; if we design systems that are not humane—responsive to human needs and considerate of human frailties, we can be guaranteed that people will make mistakes with sometimes cataclysmic consequences.
Luckily, there are methods for combating mode errors, generally by employing sensory feedback to indicate the current system mode to the user. Let’s return to the radio in Airplane. It could have used visual feedback to indicate mode by employing a light that glows during transmission. Or, the radio could have used kinesthetic feedback by allowing the radio to transmit only while a button was pushed and held. With either of these two methods, ground control might have avoided the costly (albeit humorous) blunder.
Both methods are clearly better than nothing, but which one is more effective at reducing mode error? The answer is kinesthetic feedback. Why? Because kinesthetic feedback is much harder to ignore than visual feedback and, perhaps more importantly, kinesthetic feedback is actively maintained. For critical modes, visual feedback cannot be relied upon to keep the user from making errors—what the user’s eyes are looking at and what their attention is on are two different things. This is a lesson that many designers have yet to truly learn, at the cost of hundreds of lives.
It’s not always possible to use kinesthetic feedback, however, so when I’m designing interfaces and facing modes, I use the following three questions to guide my solutions.
1. Is the mode determinable before the user takes action?
2. Is it hard for the user to ignore feedback about the mode?
3. Does the user actively maintain the mode?
Is the mode determinable?
System designers are in the habit of hiding system state from the user in the name of “clean design”. The Start Menu, for instance, has a dizzying array of preferences which change the start menu’s behavior. Did you know that drag-and-drop support in the Start Menu is a preference? It’s a mode that you can’t discover until you actually try dragging things around, and even then you wouldn’t know why it had mysteriously stopped working. But, non-determinable modes aren’t confined to computers, the living room is a great example.
The state of the lights in a room constitute a mode: on and off. If your living room is like mine, there are multiple light switches that independently control the light. This way the up/down position of any particular switch is not an indication of whether the lights are on or off.
As is inevitable when I’m expecting guests, a light burns out. By the time I get around to replacing it, the state of the light is indeterminate and I don’t know which way to flick the switch to guarantee that the light is off. Changing the light bulb is a game of Russian Roulette with a 50% chance of losing. Not so good. This uncertainty is the result of a mode that cannot be determined before action is taken.
Analogous situations frequently occur when using computers. There’s a nice example from Microsoft Excel here.
Traditional phones also have no mode indication: If you get distracted half-way through dialing you’ll generally hang-up and start from the beginning because there is no way to figure out where you were. Why is dialing a mode? Because the first 6 key presses mean “queue this number” and the 7th key press means “queue the number and dial”. Interestingly enough, the behavior of resetting the system to a known state (even if it’s not necessary) is quite common. Those readers who are familiar with the profoundly-modal text editor vi know that trained users hit escape before issuing any command, just to make sure that the editor is in a sane state before continuing. Escape just becomes part of the command gesture—a habit. Trained users escape their commands even though there is visual feedback for the current mode. Why? Because it’s less of a mental burden than diagnosing the current state of the system and the feedback is easy to ignore. This brings us to question number two:
Is it hard for the user to ignore feedback about the mode?
I usually discover that I’m in the Caps Lock mode by noticing that MY LETTERS ARE ALL CAPS. Then, if Humanized’s upcoming product Enso isn’t installed, I have to go back and retype the whole thing.
Caps Lock is a prime example of a mode that gives barely-worthwhile feedback: a small light on the keyboard glows when Caps Lock is engaged. It’s the analog of the hypothetical visual feedback for the radio in Airplane. Of course, the keyboard is exactly where no touch-typist ever looks and, to make matters worse, the Caps Lock light can be unhelpfully unlabeled:
Can you figure out whether Caps Lock is on?
The Caps Lock feedback is so easy to ignore that it just doesn’t work. As often happens in computing, we get band-aid fixes instead of true fixes. For instance, Microsoft added a nicely non-modal message to the Windows login screen that reminds the user of Caps Lock’s state. Even that doesn’t work to prevent mode errors, as I always type my password before noticing the warning.
Some nice visual feedback for Caps Lock that unfortunately doesn’t work reliably to prevent mode errors.
More generally, visual warnings, even when placed in prominent view, are not guaranteed to work: visual feedback is ignorable and overused. But, you don’t have to take my word for it.
Renowned HCI researcher Bill Buxton investigated the mode error reduction ability of visual versus kinesthetic feedback back in 1992. Buxton experimented on 12 novice and 12 expert users whom he asked to perform simple editing tasks in a pared-down vi editor. He found that “feedback using the visual channel is generally avoidable: one can easily choose not to monitor visual information. Kinesthetic and audio feedback, however, are more demanding and inescapable by their very nature.” He continues, “Information delivered through the visual channel is simply not as salient as information delivered kinesthetically… even though the visual cues in the first experiment involved changing the entire screen area pink. This has important implications for systems which rely on more subtle visual cues such as changing the shape of the cursor or the color of the menu bar.”
Go ahead and look at his data. It’s persuasive. It shows that people have the amazing ability to ignore visual cues, even when they are looking directly at them. This illustrates the important distinction between what your eyes are looking at and what your attention is on. The two are not the same, and it’s your attention that matters. And if your attention is on the mode of the system, you won’t make mode errors–but then your attention won’t be on the work you are trying to get done. Visual indication is one method for trying to bring your attention to the system state, but as Buxton showed it has a decent chance of failure. Using sound-based feedback is more likely to succeed, because while you can avoid looking at a visual indication, you cannot avoid listening to an audio indication (as every player who has been forced to listen to Baby Mario’s cry in Yoshi’s Island 2 knows). But even that is inferior to kinesthetic feedback. Which brings us to the last question:
3. Does the user actively maintain the mode?
When the user actively maintains the mode it is known as a quasimode. When the system maintains the mode it is a proper mode. Caps Lock is a mode; Shift is a quasimode. Drag-and-drop employs a quasimode; Click-move-click employs a standard mode. As Buxton showed, quasimodes not only drastically decrease the number of mode errors users make but they also decrease the cognitive burden that is associated with keeping the system state in mind: “Even though many of the expert subjects commented that they were used to keeping track of the mode ‘in their head’, feedback of both kinds significantly reduced their mode errors.” That is, even though the experts thought that they could keep both their work and the system mode as their locus of attention, they could not. This reaffirms the notion that people are fundamentally bad at multitasking, when the multiple tasks require active thought. Interestingly, “experts made more mode errors on average than novices.” In other words, experts were worse than novices at actually keeping the system state in mind.
How does one measure cognitive burden? By distracting a user with an ancillary task and, once that task is finished, measure how long it takes for them to resume the main task. This gives a measure of the effort required to assess the state of the system and task. “Any difference in resume time among conditions must reflect a difference in cognitive operations since there [is] no [difference] in the physical actions required to return to the editing task after servicing the distractor.” Buxton found that not only does the quasimodal method lead to “significantly faster resume times” but also that the result was independent of skill: everyone’s resume time benefited from kinesthetic feedback and not at all from visual feedback. This corroborates the claim that quasimodes reduce cognitive burden, and that they do it by keeping your locus of attention – what you’re thinking about – on your work and not the modal state of the system. Using kinesthetic feedback helps treat your train of thought as sacred.
Actively maintained modes—quasimodes—are a versatile tool. While feedback of any kind is helpful in reducing mode errors, kinesthetic feedback dominates over the congested visual feedback channel. Not only is kinesthetic feedback more potent and less cognitively effortful, but it’s also more enjoyable: one has to look no further than exquisite key-edges of real keyboards versus the doubt-inducing flatness of touch-screen keyboards for verification. So, the next time you battle with design, try the power of kinesthetic feedback by adding quasimodes to your arsenal. Enso, uses a quasimode at its very core. The quasimode keeps your mind on your work instead of on the system state.
by Aza Raskin