The Impact of Control-Display Gain on User Performance in Pointing Tasks
description
Transcript of The Impact of Control-Display Gain on User Performance in Pointing Tasks
The Impact of Control-Display Gain on User Perfor-mance in Pointing Tasks
Department of Information management EngineeringKorea University
Gery Casiez, Daniel Vogel, Ravin Balakrishnan
Human-Computer Interaction, 2008, Volume 23, pp.215~250
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
226
Contents
1. Introduction
2. Related work
3. Point Acceleration performance model
4. Experiment 1 : Desktop size display
5. Experiment 2 : Very large, high resolution dis-
play
6. Conclusion
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
326
Background Pointing at a target is a fundamental and frequent task in GUIs A marginal improvement in pointing performance can have a large
effect on a user’s productivity. But, Pointer acceleration has not been thoroughly studied.
Pointer Acceleration (PA) manipulates the CD gain between the input device and the display
pointer as a function of the device velocity※ device velocity : high → CD gain high ( > 1), slow → CD gain low ( < 1)
Control-Display(CD) gain The ratio of the amount of movement of an input device and the con-
trolled objects (i.e. typically a cursor)
Constant Gain (CG) is the simpler method for manipulating CD gain via a constant multi-
plier regardless of device movement characteristics.
1. Introduction (1/2)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
426
Previous Research Compared PA with constant CD gain : harms performance or little dif -
ference Effect of CD gain is more extensive, but no definitive results Performance follows a U-shaped curve No effect at all
Object1) Propose a model for TA with PA, which adapts the Fitts’ law ID to ac-
commodate the effective motor-space changes created by the PA function
2) Compares the effects of constant CD gain and PA in two experiments3) Propose a model identifying boundary constant CD gain levels to ac-
count for quantization effects
1. Introduction (2/2)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
526
2.1. Fitt’s Law Used to model direct pointing where the hand taps physical objects
• Also robust for indirect pointing where control device and display pointer are decoupled • Decoupling creates 2 Space : display space, motor space
Total movement time D : target’s distance, W : target’s width, a & b : empirically determined constant, logarithmic term : pointing task’s index of difficulty (ID)
2.2. Constant CD gain CD gain : the ratio of pointer velocity to device velocity
V : velocity
Consideration• Quantization problem : resolution of device and display• Clutch : device movement area constraint or comfortable range of arm movement • No term for CD gain in Fitt’s law
2. Related Work (1/4)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
626
2. Related Work (2/4)
Source : Jellinek, H. D., & Card, S. K. (1990). Figure 3. movement time vs. gain
2.3. Prior Studies
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
726
2.4. Dynamic Gain : PA PA increases CD gain as the velocity of the control device increases This behavior is motivated by the hybrid optimized initial impulse mo-
tor control model for human pointing motions
PA is one of many techniques that influence the motor-space through which the device travels during target acquisition:
• High gain reduces the motor distance during ballistic movement• low gain increases the motor size of the target during corrective
action
2. Related Work (3/4)
Figure 2. Decomposition of a pointing movement into the ballistic and corrective phases
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
826
2.4. Dynamic Gain (cont’) PA function f produces a CD gain G from the device motor space ve-
locity v • the function may map motor-space velocity directly to display space veloc-
ity, but this is equivalent G = f(v) ; f (PA function), v (device motor space velocity)
2. Related Work (4/4)
Figure 3. Plotting the control device velocity against Control Dis-play gain shows the characteristic curve of pointer acceleration
functions.
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
926
PA causes dynamic modification to the target’s D and W in motor-space. This modification can be modeled to predict the extent to which the Fitts’
law ID changes in motor space for each target. Motor space index of difficulty (IDmot) formula for PA.
ID in motor space equals ID in display space so difficulty of the point-ing task has not changed
Assume unchanged constants for a and b, Fitts’ law predicts the ex-act same MT regardless of the change in CD gain
Ideal PA function• high CD gain : GD (ballistic phase, affects distance to target)• low CD gain : GW (corrective phase, affects target width)• GD , GW : k, j times greater than baseline CD gain level, j < k• D and W of the target in motor space are not reduced equally• ID in motor space is now smaller than the ID in display space by
a factor of j/k• Users are able to take advantage of a reduction of ID in motor-
space and improve performance
3. PA Performance Model (1/2)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1026
CD gain continuously changes with the velocity of the pointing device Mean CD gain used to cover the distance (CDD) and near the target
(CDW)
ID in motor space
CDW/CDD be r and D >> W, then
1/r << D/W, then
3. PA Performance Model (2/2)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1126
4.1. Apparatus Monitor : 20 inch LCD, 1600 × 1200 resolution, 100 DPI Mouse : : 1600 DPI
this provided a maximum CD gain of 16 with no quantization problems
4.2. Task and Stimuli Task : a reciprocal one-dimensional pointing task to select two fixed-sized
targets back and forth in succession Stimuli : when participants selected a target, target would swap color and
if missed target, sound was heard
4. Experiment 1 : Desktop size display (1/7)
Figure 4. Experimental Display
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1226
4.3. Participants 8 volunteers (all male), mean age : 24.5 (SD = 6.3) Divided two groups of 4 : Windows XP/Vista users, those that did not
4.4. Design A within-subjects design Independent variables
Technique : CG (Constant Gain), PA (Pointer Acceleration) Level : 6 CD gain levels for CG, 6 scale factors for PA
• CD gain Levels for the CG technique : 1, 2, 4, 6, 8, 12• Scale Level for the PA technique : 0.1, 0.25, 0.5, 0.75, 1.0, 1.25
Distance between targets : DL = 360 mm, DM = 180 mm, DS = 90 mm Target width : WL = 8 mm, WM = 4 mm, WS = 2 mm 8 D-W combinations gave 5 task IDs : 3.6, 4.5, 5.5, 6.5, 7.5
4. Experiment 1 : Desktop size display (2/7)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1326
4.5. Results and Discussion Error Rate
increases with small widths A pairwise comparison
• W = 2 mm : 6.8%, W = 4 mm : 4.5%, W=8mm : 3.9%. • The overall mean error rate was 5.0%.
No other factors or interactions showed significant effects for error rate
Movement time
4. Experiment 1 : Desktop size display (3/7)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1426
4.5. Results and Discussion (cont’) Mouse Operating Range and Limb Use
operating range decreases proportionally with increasing CD gain for CG Levels
• CG Level 1, 205 mm ⇒ CG Level 12, 18 mm (Figure 8a) • PA Level 0.1, 177mm ⇒ PA Level 1.25, 26mm (Figure 8b)
Limb usage profiles : the percentage of time limbs or combinations of limbs moved in each frame during a trial (Figure 9).
• limbs are rarely used in isolation • As the effective CD gain for each Level increases, there is progres-
sion from using all limbs together to using the hand and fingers in combination to using the hand or fingers individually
4. Experiment 1 : Desktop size display (4/7)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1526
4.5. Results and Discussion (cont’) Overshooting
more pronounced overshooting in high difficulty selections (more distant or smaller targets)
Overshooting also increased with Level with high levels of CD gain caus-ing more overshooting on difficult targets (Figure 10).
4. Experiment 1 : Desktop size display (5/7)
(a) Constant gain (b) pointer acceleration
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1626
4.5. Results and Discussion (cont’) Peak velocity in motor and display space
Peak motor-space velocity (PMV) : the maximum velocity of the mouse Peak display-space velocity (PDV) : the peak velocity of the on-screen
pointer• These are related by the function mapping motor movement to CD
gain— constant for the CG technique and dynamic for the PA tech-nique
PA technique• PMV and PDV increasing with increased D • This confirms that the intensity of a ballistic movement is dependent
on the distance to be covered User Preference
5-point Likert scale• On average, participants preferred CG Level 4 and PA Level 1 for the
two techniques.
4. Experiment 1 : Desktop size display (6/7)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1726
4.5. Results and Discussion (cont’) Fitts’ Law Analysis and Relationship to the Model
Fitts’ law models are based on regression analysis of the 8 D-W combina-tions
4. Experiment 1 : Desktop size display (7/7)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1826
5.1. Apparatus, Task and Stimuli Same Apparatus, Task and Stimuli as in Experiment 1
except display : 4.7 m × 1.7 m, 25 DPI
5.2. Participants 8 volunteers (6 male, 2 female), mean age : 23.5 (SD = 1.6) None had participated in Experiment 1
5.3. Design A within-subjects design Independent variables
Technique : CG (Constant Gain), PA (Pointer Acceleration) Level : 6 CD gain levels for CG, 6 scale factors for PA
• CD gain Levels for the CG technique : 2, 5, 8, 12, 16, 20• Scale Level for the PA technique : 0.25, 0.5, 1.0, 1.25, 1.5, 2.0
Distance between targets : DL = 4,500mm, DM = 2,250 mm, DS = 1,125 mm Target width : WL = 36 mm, WM = 18 mm, WS = 9 mm 8 D-W combinations gave 5 task IDs : 5, 6, 7, 8, 9
5. Experiment 2 : Very large, high-resolution (1/6)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
1926
5.4. Results and Discussion Error Rate : 4% with no significant difference across independent variables Movement time
To fairly compare, removed the two first CG and PA Levels no significant difference between the two techniques Unlike Experiment 1, where PA had a 6% advantage over CG for the
smallest targets : suspect that WS was too large to replicate the accu-racy problem
5. Experiment 2 : Very large, high-resolution (2/6)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2026
5.4. Results and Discussion (cont’) Clutching Time
Because of the high amount of clutching that involved both arm and forearm in some conditions, we were not able to perform an analysis of limb usage
5. Experiment 2 : Very large, high-resolution (3/6)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2126
5.4. Results and Discussion (cont’) Mouse Operating Range
operating range decreases with increased Level
5. Experiment 2 : Very large, high-resolution (4/6)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2226
5.4. Results and Discussion (cont’) Overshooting
is significantly more pronounced with PA (M = 1.44%) than with CG (0.78%)
Overshooting increases with Level and Distance Unlike Experiment 1, the significant effect of Width is not caused by in-
creased overshooting on small targets
User Preference 5-point Likert scale
• On average, participants preferred CG Level 16 and PA Level 1.5 for the two techniques.
5. Experiment 2 : Very large, high-resolution (5/6)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2326
5.4. Results and Discussion (cont’) Fitts’ Law Analysis and Relationship to the Model
Fitts’ law models are based on regression analysis of the 8 D-W combina-tions
In spite of strong regression fitness, participants could not fully exploit the ID reduction in motor space
• Clutching eroded the theoretical performance advantage of PA
5. Experiment 2 : Very large, high-resolution (6/6)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2426
6.1. Gain Level On both displays, and in both CG and PA techniques
Low levels of CD gain : negative effect on performance. High levels of CD gain : increased overshooting indicating an issue with muscle control accuracy because of the re-
duced distances in motor-space
Pilot Experiment on High level CD gain CD gain level : 8, 16, 20, 30, 40, 50 Target distances : 4,500, 2,250, 1,125 mm Target widths : 36, 18, 9 mm
MT appears to remain constant for the CD gain levels above 16 CD gain has little effect on pointing performance until human limits
of speed and accuracy are approached
6. Conclusion (1/3)
Figure 18. Usable CD gain range
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2526
Define a usable range of CD gain settings minimum usable CD gain (CDmin)
ORmax : maximum operating range Dmax :: largest expected target distance
maximum usable CD gain (CDmax) Wmin : Minimum expected target width CDlmax : limb precision CDqmax : device quantization CDqmax : device quantization Mouseres , Screenres : ratio of mouse and screen resolution Handres : minimum resolution of the hand and fin-
gers
※ example 400 DPI mouse, 20″ display (100 DPI), maximum TD (360 mm), minimum TW
(2 mm), maximum operating range (250 mm), hand res (0.2 mm) CDmin = 1.4 ; 360 / 250 CDmax = min (CDqmax= 4, Cdlma=10) = 4 ; min (400 / 100, 2 / 0.2) = 4
6. Conclusion (2/3)
Korea University
User Interface Lab
Cop
yrig
ht 2
008
by U
ser I
nter
face
Lab
2626
6.2. Pointer Acceleration Versus Constant Gain On the standard desktop display, PA was 3.3% faster overall, and up to
5.6% faster with small targets We also found that PA follows Fitts’ law with good regression fitness Finally we encourage researchers to use pointer acceleration rather than
constant gain as a base technique for comparing new pointing technique performance
6. Conclusion (3/3)
?Q & A