breaking fitts law
DESCRIPTION
Study on Fitts Law by Human Computer Interaction Masters students in Georgia TechTRANSCRIPT
Breaking Fitts’ LawAbhishek, Sahithya, Keenan, Xiao
Our Question.
Is it faster to click on targets at the edge of
the screen?
Bounding line simulates edge of screen
Bounding line simulates edge of screen
Bounding line simulates edge of screen
Theoretical Underpinnings:Targets at the edge of the screen effectively have infinite width
We used the Least-of method of determining target in two-dimensions, which MacKenzie and Buxton (1992) found to be comparable to the W’ Model (actual target depth along the approach vector).
MacKenzie, I. S., & Buxton, W. (1992). Extending Fitts' law to two-dimensional tasks. Proceedings of the ACM Conference on Human Factors in Computing Systems - CHI '92, pp. 219-226. New York: ACM.
W
W = ∞
Are movement times lower while selecting targets at the edge of the screen than predicted by Fitts’ law?
Objectified Question
Does the magnitude of effect vary based on target size?
Additional Questions
Bounded mouse movements will be faster than Fitts’ Law would predict.
Hypothesis 1
Bounded mouse movements will be faster than identical unbounded movements.
Hypothesis 2
Simulate the edge of the screen with a ‘bounding box.’
Participants perform an identical set of pointing tasks with a bounding box and without one.
Design
Independent Variables:
Presence of Bounding BoxSize of Target
Dependent Variable:
Observed Movement Time
Addressing Potential Confounds
Screen Resolution Consistent at 1680x1050
Subject Distance from Screen Same chair height and distance from monitor
Type of Mouse Use of identical Dell optical mouse
Fatigue Breaks after 25 trials
Order Effects Randomized trials to eliminate order effects
Device LCD with identical calibration and constrast
Starting Position Always in the center of the screen
Potential Confounds What We Controlled
Methodology1680x1050 Resolution22” Display2 Foot distance from DisplayTargets are 1º and 1.2º of Visual AngleDell optical mouseRandomized order of trials10 second break after 25 trials to reduce fatigueBright green targets on black backgroundPink bounding boxTrial time = Time from start until successful click0.5s fixation time as cursor is auto-centered.Cursor always starts at center of screen8 varying target distancesTwo distinct target sizesSame set of targets4 participants
Data
t=-5.7272p<0.05
t=0.1196p=0.9
t=-7.8984p<0.05
Condition
Aver
age
(Obs
erve
d M
T)Average Observed MT vs. Condition
significant difference between bounded MT and unbounded MT. almost 100 ms difference.
bounding versus no bounding is not significant for large targets,
but, for small targets, the effect is significant, and is close to 100ms
Corre
latio
n
No Bounding Box Bounding Box
0.9
0.7
0.5
0.3
0.1
Correlation between Observed MT and Predicted MT
so, does Fitts law still work? We were trying to break it. It works very well when there is no bounding box (around .93), and it still works fairly well when there is a bounding box (around .83)
Data
Observed MT vs. Predicted MT (Large targets with Bounding Box)
This is a line representing what Fitts law predicts, and box plots for all of the observed MTs at each index of difficulty.
pretty good fit for large targets with bounding box
Data
Observed MT vs. Predicted MT (Large Targets with No Bounding Box)
also a good fit for large targets with no bounding box
Data
Observed MT vs. Predicted MT (Small targets with Bounding Box)
interesting: these boxes tend to be a bit lower than the Fitts law trend line
Data
Observed MT vs. Predicted MT (Small Targets with No Bounding Box)
and here, Fitts law works pretty well again- the bounding box is gone, so it’s just the normal task
Differences of Observed Time and Predicted Time
So, there is no significant difference between bounding box and no bounding box across all targets, although we were a bit faster with the bounding box
for small targets, there is a highly significant difference between predictions and observed times for small targets with a bounding box, but not with no bounding box. With no bounding box, we think that there may have been outliers that made this average so high, even though there was no significant difference.
Finall, for large targets there were no significant differences between predictions and observations.
• There is a significant difference in movement time between bounded and unbounded movements.
• This effect is only significant for small targets.
Findings
• Instruct participants on how to approach the target, in order to control for the effects of strategic differences
• careful aiming versus quick movements
• We did not remove outliers, and our averages may have been skewed by such points
What would we do differently?
★ Perform test on tablet with physical bounding boxes
★ Add additional target sizes between small (20 pixels) and large (100 pixels) to find out when our effect becomes significant.
★ Test for External Validity: Compare differences in tab switching time between browsers
Next Steps
Chrome on Windows
Chrome on Mac OS
External Validity
Questions?