The following points and ideas came from a class discussion of Fitts' Law on Sept 19, 1996.

If your target button is not near many other interface parts, then there is no concept of errors. You just keep trying until you hit the button to finish the task. The distance of other interface parts nearby can in fact introduce errors -- if two buttons are close together you might hit the wrong one. This is especially true for palettes of small buttons and multiple cascaded windows where the title bars a close together.

Moving targets exist in real interfaces. For example, when choosing a font from a scrollable list, the list may scroll before you reach the font of interest. Moving targets are not addressed by Fitts' Law.

Buttons at the edge of the screen change the movement -- it may be acceptable to slam the pointer into the side of the screen and then make small adjustments to hit the target. This style of movement is not covered by Fitts' Law

How can Fitts' Law be (theoretically) applied to interface design?

• Things done more often should be assigned a larger button.
• As interface designers, we have more control over the amplitude (A) to a button than the width (W) of the button. Thus things done more often should be closer to the average position of the user's cursor. Potentially, Fitts' Law was responsible for new kinds of interface widgets -- context menus, popup dialogs, and tear-off menus. All these widgets reduce the amplitude of movement.

One idea of applying Fitts' Law to an interface is making buttons grow proportional to the distance from the pointer -- this is not a realistic approach (many problems exist with this approach).

We didn't find any useful mention of Fitts' Law in 10 or so textbooks on designing user interfaces. None of the books used Fitts' as a justification for any design decision.

There are tradeoffs when applying Fitts' Law to interface designs. Fitts' Law suggests that interface components should be made larger and positioned closer to the average cursor position. These suggestions may act in opposition to other factors that make a larger difference on interface efficiency, such as organization and use of available screen real estate.

We analyzed the Microsoft Word for Windows 95 interface with respect to Fitts' Law. Using three different button sizes and an arbitrary amplitude, it was determined that the differences between movement time to the smallest button and to a standard toolbar button were around 0.2 seconds. The conclusion of this analysis is that Fitts' does not make a significant effect in standard user interfaces.

The three buttons used were

1. a toolbar button,
2. the close button, and
3. the Number of Copies up/down spin button (on the Print . . . dialog).

The amplitude used represents the length in pixels from the center of the screen to a corner. Our screen was in 1024x768 resolution, so the length of the diagonal is 1280 pixels. 640 pixels (half of 1280) represents the distance from the center of the screen to a corner.

The constants used were a=52 and b=148. This yielded the following IDs and movement times:

	Button Width	Button Height	ID	Movement Time
Up/Down arrow	18 pixels	9 pixels	7.15	1110.50
Close button	17 pixels	16 pixels	6.32	987.65
Toolbar button	25	24	5.74	901.07

We used the 'smaller-of' model when computing the ID.

This demonstrates that the savings in movement time between typical interface parts is minimal due to size. Thus, Fitts' Law may not be a effective tool to increase efficiency in a general interface.

In expert performance, planning of a movement may overlap the actual movement. For example, if a user knows approximately where a dialog button will appear, they may move the cursor to the position before the dialog appears. This prediction decreases the amplitude of the motion and speeds expert performance. Thus Fitts' Law may over-estimate the movement time in some interfaces.

Many Fitts' Law studies have come up with different empirical constants for the same device -- what could cause this?

• change in mouse technology (old vs. recent)
• change in society -- today's youth have grown up with computers, thus they are accustomed to the devices

Do new ergonomic mouse designs increase accuracy? They probably improve the homing time to the mouse, which is included in Fitts' Law as part of the constant a.

How might Fitts' Law be extended to 3D movements? What is the effective width of a 3D target?

Each added dimension increases the degrees of freedom, but also increases the number of dimensions of potential errors.

In 3d, the input device may be a dataglove instead of a mouse, so homing time would be eliminated.

Can 3d lessen the tradeoff between organization/layout optimization and Fitts' Law optimization? Amplitude to a 3d target may be more than a mouse can handle (what if I need to walk across a virtual room to click a virtual light switch?) Additionally, judging distance in a 3d environment is much more difficult.

How do 3d input devices compare to each other? Input devices should be chosen according to task (would you want to fly an airplane with a mouse?)

In discussing different models for 2d amplitude, a difference exists between

1. using distance from starting point to the center of the object, and
2. from the starting point to closest inside point of the object.

What about targets where one dimension is much larger than another dimension (such as title bars)? Does the 'smaller of' model continue to be a good model for predicting the target width? In this case, the W' model (that measures width along the approach angle) may be more appropriate.

One source of experimental error may be that subjects were in the habit of clicking and starting trial before planning the move. This is contrary to Fitts' requirement that movements be aimed and direct. The software used in the experiment prevented the start of the next trial until the mouse had been at rest for a short period of time, but the time may not have been long enough to prevent an effect from this habit.

In the experiment, we should have adjusted the mouse/puck drivers to different gain values (so that they were both at their unique optimal value) instead of choosing the same value for both. We were attempting to make the 'gain' a non-factor, but instead may have created a factor.

We were using a device designed for 3d input (space puck) in a 2d environment. How could this have effected the performance numbers?