Summary

Fitts' Law by design predicts a movement time. Typically, these times are less than one second for a graphical interface. Attempting to measure these sub-second movement times using stopwatches would probably lead to a great deal of error, so this experiment instead measures the errors committed by the subjects.

Using a metronome, a steady tempo is established so that the subjects can hear each beat. A metronome is a device musicians use to set the tempo of a piece of music -- metronomes click at a certain steady rate, say 160 beats per minute. The individual subjects are instructed to tap the point of a pen between two targets on a sheet of paper at the tempo of the metronome. After 30 seconds, the subjects stop tapping and count the number of marks left by the pen outside the targets. The number of errors each subject made should be predicted by the Index of Difficulty in Fitts' Law.

It is important to stress to the subjects that the goal is to make a tap at each click of the metronome without regard to errors (taps outside the targets). Accuracy is secondary to keeping up with the metronome.

Method

1. Assemble printouts of the (described below), some extra ball-point pens, a metronome, and a calculator.
2. Pass out the practice sheets to the subjects. Instruct the subjects to tap between the targets with the point of the pen in tempo with the metronome. State that it is more important to keep up with the metronome than to make errors.
3. Set the metronome to 160 beats per minute, or a value you find yields a good mixture of errors over the different target width/amplitude pairs. Start the metronome. (Note: during our experiment, the subjects had trouble hearing the metronome during the trial. Consider finding a way to amplify the sound of the metronome).
4. Start the practice trial. Time the subjects tapping between the targets for 30 seconds. This should yield approximately the same number of trials for each subject. After the trial, have the subjects count their errors (marks left by the pen outside the target).
5. Repeat the previous step for the second set of targets on the practice sheet.
6. Hand out one of the four experiment sheets to each subject. It is best to ensure that an equal number of sheets of each kind are distributed.
7. Perform trials as described above for each of the four target width/amplitude combinations. After each trial, count the errors (this gives time for the subjects to rest).

Analysis

This experiment should demonstrate that the Index of Difficulty predicts the difficulty of the task, and thus predicts the number of errors. The Index of Difficulty is Log base 2 ((2 * Amplitude)/Width).

The files have the same four target width/amplitude combinations, ordered differently from top to bottom. Each target is either 1/4" or 2/4" inch wide, and the amplitude is either 15/4" or 30/4". If we let 1 unit equal 1/4", then the targets are either 1 or 2 units wide, and the amplitudes are either 15 or 30 units. This simplifies the calculation of the Index of Difficulty. Below is a table of the target width/amplitude pairs and their corresponding indices of difficulty.

Combination	Amplitude	(2A/W)	Index of Difficulty
A	2 units	15 units	(2 * 15)/2 = 15	3.91
B	1 units	15 units	(2 * 15)/1 = 30	4.91
C	2 units	30 units	(2 * 30)/2 = 30	4.91
D	1 units	30 units	(2 * 30)/1 = 60	5.91

The data collected from the subject's trials should be predicted by the Index of Difficulty. Combination A should have the least number of errors on average, Combinations B and C should have nearly the same number of errors on average, and Combination D should have the most errors on average.

Results from our experiment

Combination	Index of Difficulty	Average # of Errors
A	3.91	6.9
B	4.91	22.4
C	4.91	16.0
D	5.91	27.3

The results were not as close between combinations B and C as we would have liked, but the general trend of lower Index of Difficulty leading to lower number of errors is present.

Files

Patrick, make the files available in different formats

Extension

In his 1954 paper, Fitts briefly states "The [subjects] were uniformly more accurate in terminating flexor than extensor movements." The above exercise can be extended to demonstrate this.

In arm movements, a flexor movement is controlled by the bicep, and an extensor movement is controlled by the tricep. For a right-handed person, moving to the 'Start' button on a Windows95 interface corresponds to a flexor movement, and moving to the upper right corner of a maximized window corresponds to an extensor movement. In general, a flexor movement of the arm moves the hand closer to chest, and an extensor moves the hand away from the chest.

Demonstrating the accuracy differences between flexor and extensor movements can be incorporated into this experiment by counting the errors for the right and left hand targets separately. For a left-handed person, the right-hand target is the flexor target, and the left-hand target is the extensor target. Below is a table of data from our Sept 12, 1996 experiment.

Combination	Flexor Errors	Extensor Errors	Total # of Errors
A	3.8	3.1	6.9
B	11.0	11.4	22.4
C	7.6	8.4	16.0
D	12.2	15.1	27.3

With the exception of combination A, the trend of less errors in flexor movements is present.