213: User Interface Design and Development
Lecture #11 - March 31st, 2009

1) Cognitive Modeling
2) Keystroke-Level Model (KLM)
3) Other GOMS Variants

Cognitive approach to HCI attempts to predict user performance based on a model of cognition
Start with a model of how humans act
Use that model to predict how humans would complete tasks using a particular UI
Provided theoretical foundation of HCI in the 1970s and 1980s

QuickTime™ and a
decompressor are needed to see this picture.
Source: Stuart Card, Lecture on Human Information Interaction, 2007

Abstract
Quantitative
Approximate
Estimated from experiments
Based on a theory of cognition
Adapted from Rob Miller

Don’t need to implement / prototype
Don’t need to test with real users
Theory has explanatory power
Provides scientific foundation for design, like other engineering fields
Adapted from Rob Miller

KLM (Keystroke-Level Model) - Description of user tasks based on low-level actions (keystrokes, etc.)
GOMS (Goals, Operators, Methods, Selectors) -
Higher-level then KLM, with structure and hierarchy
Fitts’ Law - Predicts how long it will take a user to select a target; used for evaluating device input
Model Human Processor (MHP) - Model of human cognition underlying each of these theories

QuickTime™ and a
decompressor are needed to see this picture.

K Press a key or button
P Point to a target on the display
H Home hands on input device
D Draw a line segment
M Mentally prepare for an action
R (system response time)

Adapted from Lorin Hochstein

Walk through a task, listing the operators needed to complete it
Use heuristics to insert M operators (for example, place M’s in front of all P’s that select a command)
– These can be different for different UI styles
Based on estimates for each operator, calculate the amount of time required to complete the task
Adapted from Rob Miller

Keystroke determined by typing speed
0.28s for average typist (40 wpm), 0.08s for fast typist (155 wpm), 1.20s for worst typist
Pointing determined by Fitts’ Law (or general approximation)
T = a + b log (d/s +1) OR
T = 1.1s
Drawing determined by Steering Law
T = a + b (d/s)
Adapted from Rob Miller

Homing estimated by measurement
T = 0.36s (between keyboard and mouse)
Mental prep estimated by measurement
T = 1.35s
(estimated by taking the total task time, subtracting physical operator time, and dividing by the number of “chunks” of activity)
Adapted from Rob Miller

Basic idea: Put an M before each step requiring access of a “chunk” from long-term memory
Insert M’s before each K and P
K -> MK; P -> MP
Delete M’s in the middle of typing a word or string
MKMKMK -> MKKK
Delete M’s in the middle of composite actions (for example, point and click)
MPMK -> MPK
Adapted from Rob Miller

Using Shift-Click
M
P [start of word]
K [click]
M
P [end of word]
K [shift]
K [click]
H [to keyboard]
M
K [Del]
Total: 3M + 2P + 4K
= 7.37 sec
Using Delete
M
P [start of word]
K [click]
H
M
K [Del] x n [length of word]
Total: 2M + P + H + (n+1) K
= 4.44 + 0.28n sec
Adapted from Rob Miller

KLM can help evaluate UI designs, interaction methods and trade-offs, using parametric analysis
T
Del n times
Shift-click n
If common tasks take too long or consist of too many statements, can provide shortcuts
Adapted from Rob Miller

Source: Card, Moran and Newell, “
The Keystroke Level Model for User Performance Time with Interactive Systems
”

Generate a KLM model for deleting a file from your desktop
Estimate the time it would take using the provided operator times
Compare the predicted time with the actual time

Only applies to expert users doing routine (welllearned) tasks
Only predicts efficiency - not error rate, memorizability, learnability, etc.
Impractical for all but the simplest tasks
Ignores
– Parallel processing
– Goal interleaving
– Mental workload (working memory limits, fatigue)
– Planning and problem-solving (how to select a method?)
Adapted from Rob Miller

GOMS provides a higher-level language for task analysis and UI modeling
Generates a set of quantitative and qualitative measures based on description of the task and user interface
Provides a hierarchy of goals and methods to achieve them
Different GOMS variants use different terms, operate at various levels of abstraction, and make different simplifying assumptions

Goals - What the user wants to achieve; can be broken down into subgoals
Operators - An action performed in service of a goal; can be perceptual, cognitive or motor acts
Methods - Sequences of operators and subgoal invocations that accomplish a specified goal
Selection Rules - Represent the user’s knowledge of which method should be applied
Source: John and Kieras, “The GOMS Family of User Interface Analysis Techniques”

Goal: delete text (n chars long)
Select: method 1 if n > 10 method 2 if n < 10
Method 1: Goal: highlight text & delete
Goal: highlight text
Point
Click
Point
Shift
Click
Verify
Click
Method 2: Goal: delete n chars
Adapted from Rob Miller

The difference between goals and operators is simply the level of detail chosen by the analyzer
Goals are usually important end-user intentions
Operators usually represent primitive user actions, that have a fixed execution time regardless of the context (or that is a constant function of some parameter), that can be estimated empirically
As a result, operators usually stop at the command or keystroke level

KLM - Simplest version (Card, Moran, Newell 1983)
CMN-GOMS - Original formulation; includes methods and selection rules (Card, Moran, Newell 1983)
NGOMSL - GOMS using natural language; also can model memory usage and learning times (Kieras
1988)
CPM-GOMS - Models parallel processing by cognitive, perceptual and motor systems (John 1990 )
Source: John and Kieras, “The GOMS Family of User Interface Analysis Techniques”

Provides for methods to achieve explicit goals and subgoals
Selection of methods are predicted by the system based on user’s rational decision-making
Mental operations are modeled either as:
– VERIFY operators
– Subgoal invocations
– Selection rules

Generate task description
– Pick high-level user Goal
– Write Method for accomplishing Goal (may invoke subgoals)
– Write Methods for subgoals (recursive)
– Stop when Operators are reached
Evaluate description of task
Apply results to UI
Iterate
Adapted from Chris Long, Marti Hearst

Structured natural language notation for representing GOMS models
Can predict learning time based on the number of
NGOMSL statements in a method (plus loading any additional static data in memory)
Can be used for developing training materials, help modules, etc.
Models the operation of working memory by including Retain and Recall statements
More specific notation for mental operations
Still does not address parallel processing, goal interleaving, fatigue, problem-solving, etc.

Source: John and Kieras, “The GOMS Family of User Interface Analysis Techniques”

Cognitive-Preceptual-Motor or Critical-Path-Method
Works at an even lower level of detail then KLM
Primitive operators are very simple perceptual, cognitive or motor acts
Explicitly models parallelism by considering three mental processors and storage systems that can work in parallel (based on Model Human
Processor - to be discussed in next lecture)
Execution time predicted using critical path - the longest path through the task based on cognitive limitations and information flow dependencies

PP
CP
MP right
0 start start eye move mouse move
50 50 perceive target
100 move mouse
480 verify target
50
MP left
MP eye move eye to target
30 perceive cursor
100 start
Shift press
50 press
Shift
100
Adapted from Rob Miller

PP
CP
MP right
0 start start eye move mouse move
50 50 perceive target
100 move mouse
480 verify target
50
MP left
MP eye move eye to target
30 perceive cursor
100 start
Shift press
50 press
Shift
100
Adapted from Rob Miller

Phone company considering redesign of a workstation for telephone operators
– Reduced keystrokes needed for common tasks
– Put frequently-used keys closer to user’s fingers
New design was 4% slower than old design
– 1 sec/call = $3 million/year
Keystroke-level model has no explanation
But CPM-GOMS explained why:
– Keystrokes removed were not on the critical path
– Used during slack time, while greeting customer
– A keystroke was moved from the beginning of call
(during slack time) to later (putting it on the critical path)
Adapted from Rob Miller

Generate GOMS models (CNM-GOMS or
NGOMSL) for deleting a file from your desktop using two methods
– Drag-and-drop
– Using terminal window
Compare the two models for deciding which should be used, and in which contexts

Assumes extreme expert behavior
Tasks must be goal-directed
Does not model problem-solving, exploration, etc.
Difficult to model behavior at this level of detail
Still hard to model anything but the simplest tasks
Not as easy as heuristic analysis, guidelines, or cognitive walkthrough
Conclusion: GOMS is an interesting theoretical model, but is hardly ever used in practice

We will talk about the conceptual theories underlying KLM and GOMS
– Fitts’ Law
– Model Human Processor
Interactive Prototype #2 and Experiment
Design due on April 13th!