Transcript Document

SIMS 213:
User Interface Design & Development
Marti Hearst
Tues, April 6, 2004
Today
Evaluation based on Cognitive Modeling
Comparing Evaluation Methods
Another Kind of Evaluation
Evaluation based on Cognitive Modeling

Fitts’ Law
 Used to predict a user’s time to select a target

Keystroke-Level Model
 low-level description of what users would have
to do to perform a task.

GOMS
 structured, multi-level description of what users
would have to do to perform a task
GOMS at a glance
Proposed by Card, Moran & Newell in 1983
– Apply psychology to CS
• employ user model (MHP) to predict performance of tasks
in UI
– task completion time, short-term memory requirements
– Applicable to
• user interface design and evaluation
• training and documentation
– Example of
• automating usability assessment
Slide adapted from Melody Ivory
Model Human Processor (MHP)
Card, Moran & Newell (1983)
– most influential model of user interaction
• used in GOMS analysis
– 3 interacting subsystems
• cognitive, perceptual & motor
• each with processor & memory
– described by parameters
» e.g., capacity, cycle time
• serial & parallel processing
Slide adapted from Melody Ivory
Adapted from slide by Dan Glaser
Original GOMS (CMN-GOMS)
Card, Moran & Newell (1983)
Engineering model of user interaction
• Goals - user’s intentions (tasks)
– e.g., delete a file, edit text, assist a customer
• Operators - actions to complete task
– cognitive, perceptual & motor (MHP)
– low-level (e.g., move the mouse to menu)
Slide adapted from Melody Ivory
CMN-GOMS
Engineering model of user interaction (continued)
• Methods - sequences of actions (operators)
– based on error-free expert
– may be multiple methods for accomplishing same goal
» e.g., shortcut key or menu selection
• Selections - rules for choosing appropriate method
– method predicted based on context
– hierarchy of goals & sub-goals
Slide adapted from Melody Ivory
Keystroke-Level Model
Simpler than CMN-GOMS
Model was developed to predict time to accomplish a
task on a computer
Predicts expert error-free task-completion time with the
following inputs:
–
–
–
–
–
a task or series of subtasks
method used
command language of the system
motor-skill parameters of the user
response-time parameters of the system
Prediction is the sum of the subtask times and overhead
KLM-GOMS
(What Raskin refers to as GOMS)
Keystroke level model
1. Predict
Action 1
x sec.
Action 2
y sec.
Action 3
+
2. Evaluate
z sec.
t sec.
Time using
interface 1
Slide adapted from Newstetter & Martin, Georgia Tech
Time using
interface 2
Symbols and values
Remarks
Time (s)
K
B
P
H
D
Press Key
Mouse Button Press
Point with Mouse
Home hand to and from keyboard
Drawing - domain dependent
0.2
.10/.20
1.1
0.4
-
M
R
Mentally prepare
Response from system - measure
1.35
-
Operator
Raskin
excludes
Assumption: expert user
Slide adapted from Newstetter & Martin, Georgia Tech
Raskin’s rules
Rule 0: Initial insertion of
candidate M’s
i.e. not when
P points to
arguments
M before K
M before P iff P selects command
Rule 1: Deletion of anticipated M’s
e.g. when
you point and
click
If an operator following an M is fully
anticipated, delete that M.
Slide adapted from Newstetter & Martin, Georgia Tech
K
B
P
H
D
0.2
.10/.20
1.1
0.4
-
M
R
1.35
-
Raskin’s rules
Rule 2: Deletion of M’s within
cognitive units
e.g. 4564.23
If a string of MK’s belongs to a
cognitive unit, delete all M’s but the
first.
Rule 3: Deletion of M’s before
consecutive terminators
e.g. )’
If a K is a redundant delimiter, delete
the M before it.
Slide adapted from Newstetter & Martin, Georgia Tech
K
B
P
H
D
0.2
.10/.20
1.1
0.4
-
M
R
1.35
-
Raskin’s rules
Rule 4: Deletion of M’s that are
terminators of commands
If K is a delimiter that follows a
constant string, delete the M in front
of it.
Rule 5: Deletion of overlapped M’s
Do not count any M that overlaps an
R.
Slide adapted from Newstetter & Martin, Georgia Tech
K
B
P
H
D
0.2
.10/.20
1.1
0.4
-
M
R
1.35
-
Example 1
Temperature Converter
Choose which conversion is desired,
then type the temperature and press
Enter.
Convert F to C.
K
B
P
H
D
0.2
.10/.20
1.1
0.4
-
M
R
1.35
-
Convert C to F.
HPBHKKKKK
HMPMBHMKMKMKMKMK
HMPBHMKKKKMK
.4+1.35+1.1+.20+.4+1.35+4(.2)+1.35+.2
Apply Rule 0
Apply Rules 1 and 2
Convert to numbers
=7.15
Slide adapted from Newstetter & Martin, Georgia Tech
Example 1
Temperature Converter
Choose which conversion is desired,
then type the temperature and press
Enter.
Convert F to C.
K
B
P
H
D
0.2
.10/.20
1.1
0.4
-
M
R
1.35
-
Convert C to F.
HPBHKKKKK
HMPMBHMKMKMKMKMK
HMPBHMKKKKMK
.4+1.35+1.1+.20+.4+1.35+4(.2)+1.35+.2
Apply Rule 0
Apply Rules 1 and 2
Convert to numbers
=7.15
Slide adapted from Newstetter & Martin, Georgia Tech
Example 2
GUI temperature interface
Assume a button for compressing scale
Ends up being much slower
– 16.8 seconds/avg prediction
Using KLM and Information Theory to
Design More Efficient Interfaces (Raskin)
Armed with knowledge of the minimum
information the user has to specify:
– Assume inputting 4 digits on average
– One more keystroke for C vs. F
– Another keystroke for Enter
Can we design a more efficient interface?
Using KLM to
Make More Efficient Interfaces
First Alternative:
To convert temperatures,
Type in the numeric temperature,
Followed by C for Celcius or
F for Fahrenheit. The converted
Temperature will be displayed.
MKKKKMK = 3.7 sec
Using KLM to
Make More Efficient Interfaces
Second Alternative:
– Translates to both simultaneously
C
F
MKKKK = 2.15 sec
GOMS in Practice
Mouse-driven text editor (KLM)
CAD system (KLM)
Television control system (NGOMSL)
Minimalist documentation (NGOMSL)
Telephone assistance operator workstation
(CMP-GOMS)
– saved about $2 million a year
Slide adapted from Melody Ivory
Drawbacks
Assumes an expert user
Assumes an error-free usage
Overall, very idealized
Fitts’ Law
Models movement time for selection tasks
The movement time for a well-rehearsed
selection task
• increases as the distance to the target
increases
• decreases as the size of the target
increases
Slide adapted from Newstetter & Martin, Georgia Tech
Fitts’ Law
Time (in msec) = a + b log2(D/S+1)
where
a, b = constants (empirically derived)
D = distance
S = size
ID is Index of Difficulty = log2(D/S+1)
Slide adapted from Newstetter & Martin, Georgia Tech
Fitts’ Law
Time = a + b log2(D/S+1)
Target 1
Target 2
Same ID → Same Difficulty
Slide adapted from Pourang Irani
Fitts’ Law
Time = a + b log2(D/S+1)
Target 1
Target 2
Smaller ID → Easier
Slide adapted from Pourang Irani
Fitts’ Law
Time = a + b log2(D/S+1)
Target 1
Target 2
Larger ID → Harder
Slide adapted from Pourang Irani
Determining Constants for Fitts’ Law
To determine a and b design a set of tasks with varying values
for D and S (conditions)
For each task condition
– multiple trials conducted and the time to execute each is recorded and
stored electronically for statistical analysis
Accuracy is also recorded
– either through the x-y coordinates of selection or
– through the error rate — the percentage of trials selected with the cursor
outside the target
Slide adapted from Pourang Irani
A Quiz Designed to Give You Fitts
http://www.asktog.com/columns/022DesignedToGiveFitts.
html
Microsoft Toolbars offer the user the
option of displaying a label below each
tool. Name at least one reason why
labeled tools can be accessed faster.
(Assume, for this, that the user knows the
tool and does not need the label just
simply to identify the tool.)
Slide adapted from Pourang Irani
A Quiz Designed to Give You Fitts
1. The label becomes part of the target. The target
is therefore bigger. Bigger targets, all else
being equal, can always be acccessed faster.
Fitt's Law.
2. When labels are not used, the tool icons crowd
together.
Slide adapted from Pourang Irani
A Quiz Designed to Give You Fitts
You have a palette of tools in a graphics
application that consists of a matrix of
16x16-pixel icons laid out as a 2x8 array
that lies along the left-hand edge of the
screen. Without moving the array from the
left-hand side of the screen or changing
the size of the icons, what steps can you
take to decrease the time necessary to
access the average tool?
Slide adapted from Pourang Irani
A Quiz Designed to Give You Fitts
1. Change the array to 1X16, so all the
tools lie along the edge of the screen.
2. Ensure that the user can click on the
very first row of pixels along the edge of
the screen to select a tool. There
should be no buffer zone.
Slide adapted from Pourang Irani
Comparing Evaluation Methods
Jeffries et al., 1991
Comparing Evaluation Methods
“User Interface Evaluation in the Real World: A Comparison of
Four Techniques” (Jeffries et al., CHI 1991)
Compared:
– Heuristic Evaluation (HE)
• 4 evaluators, 2 weeks time
– Software Guidelines (SG)
• 3 software engineers, familiar with Unix
– Cognitive Walkthrough (CW)
• 3 software engineers, familiar with Unix
– Usability Testing (UT)
• Usability professional, 6 participants
The Interface:
– HP-VUE, a GUI for Unix (beta version)
Comparing Evaluation Methods
Jeffries et al., CHI ‘91
Comparing Evaluation Methods
Jeffries et al., CHI ‘91
On a 9 point scale
Higher is more critical
Comparing Evaluation Methods
Jeffries et al., CHI ‘91
Comparing Evaluation Methods
Jeffries et al., CHI ‘91
Comparing Evaluation Methods
Jeffries et al., CHI ‘91
Conclusions:
– HE is best from a cost/benefit analysis, but requires
access to several experienced designers
– Usability testing second best – found recurring,
general, and critical errors but is expensive to conduct
– Guideline-based evaluators missed a lot but did not
realize this
• They were software engineers, not usability specialists
– Cognitive walkthrough process was tedious