Behavioral Aspects of Text Editors

1
Behavioral Aspects of Text Editors

• David W. Embley, George Nagy
• University of Nebraska, Lincoln

2
Assumptions for readers

• Familiar with basic vocabulary of computer
  science
• Sufficient exposure to various text and program
  editors
• Innocent of any formal training in psychology

QED, CMS, TECO, Wylbur, WIDJET and UNIX
3
Interactive Text Editors

• Frequently, Primary means of interaction with
  computer
• Manuscript creation
• Programming
• File System Maintenance
• Email
• Important to make their use easy
• Editors
• General Purpose Interactive editors
• QED, CMS, TECO, Wylbur, WIDJET and UNIX
• Language dependent editors
• BASIC, APL, LISP

4
Editor Design and Evaluation

• Everyone has an opinion, but no consensus
• Some established means
• Introspection
• Field Studies/Observations
• Formal Analysis
• Controlled Experiments
• Psychological Models

?
5

• Introspection

Our own intuition and experience, is what we
depend on when we assume that we know as much
about the topic as the next person and are too
lazy to look further
6
Applicable areas of Psychology

• Cognitive Psychology
• Study of higher mental processes
• LLUMPRT
• Well studied area but limited application to
  study of text editors

7
Overview

• Section 1 Temporal Models
• Section 2 Impact of Editor structure and command
  languages
• How do different editors differ ?
• Section 3 Stimulus and response studies of input
  devices Mouse wins

8
Performance Time Considerations
9
Objective

• To minimize the time incurred by a user
  performing a number of editing tasks over a
  period of time
• Depends on numerous factors
• Expertise of the user
• Learning methods and procedures
• User alertness and motivation
• Out of Paper
• Some are within our control and can be improved

10
Predictive Models
The Keystroke Model CARD
Total time sum of time required to perform
individual unit tasks
To acquire a mental representation of the task
Perform and execute it
11
?
12
Example

• Replacing one word of arbitrary length with a
  five letter word

13
Model Verification

• 12 subjects, 4 different editing tasks, 3
  different editors
• Tasks
• Simple word substitution
• Moving a sentence

14
Results

• Observed times and predicted times match mostly
• Exploring More or less detailed models CARD

15
The Embley Model

• A simpler model for line-oriented editors
• Objective
• Comparing program editor performance as a
  function of time required by a user to perform
  editing tasks

16
The Model
Acquisition time and mental time combined m
number of command response pairs Tc delay per
command mental prep. Time computer response
time n number of keystrokes Tk time per
keystroke
17
GOMS Model

• Attempts to explain How an expert user
  accomplishes routine editing tasks, not just time
  constituents

18
(No Transcript)
19
Can adjust to desired level of detail

• Example

Substitute
OR
Specify substitute command specify argument
number 1 - specify argument number 2 enter
command
Which one is more accurate ?
?
20
Experimental Studies

• Several variations were explored
• 10 different GOMS models
• 16 second operator duration, 8, 4, 2, 0.5 ( type
  an s, home hands on keyboard)
• 5 participated, only 1 studied

DATA
Derivation Data
Cross-validationData
Prediction rules for operator sequences Estimates
for operator duration
For calculation of unit task time using
derivation data results
21
(No Transcript)
22
Predicting the task accomplishment method

• Objective To determine whether a set of simple
  selection rules could account for the methods
  user select.
• The Experiment
• 3 subjects
• Teletypewriter and CRT

23
(No Transcript)
24
Findings

• Each subject appeared to have a dominant method
  the first rule
• S2 applied different dominant method for
  different devices speed difference
• Selection of methods depends on feature of task
  e.g.
• Locating a line number of lines between current
  line and target line

Users are able to quickly select near-optimal
methods by having assimilated heuristic rules
based on a few pertinent task features
25
Contribution of Errors

• Error ignored in previous experiments
• Even for experts 5-30 time in errors and error
  corrections
• For accurate Prediction, errors must also be
  considered

26
Robertss Experiments

• 4 experts ? 4 separate tasks
• Human observer noted time consumed by significant
  errors (gt 30 seconds )
• Findings Much of the subject-to-subject
  variation is due to error rates
• For error free data, variation can be attributed
  to editors than to the subjects

27
Applying the Keystroke model

• Errors were ignored
• Optimal Method prediction ? Predictions 25-30
  too low
• Actual key sequence records ? only 87 accounted
  for.
• Remaining time ? Unknown mental activities

28
Advantage of Keystroke Model

• Assumes that user is so practiced that
• Method selection time would be nil
• Choices optimal
• Entry Flawless
• Provides an upper bound for the editing time
• Comparison between predicted time and observed
  time ? relatively large difference indicates that
  editor is difficult to use optimally

29
(No Transcript)
30
Effects of Computer Response time
31
Effects of Computer Response time

• System Delay and Unpredictability ? Affects user
  Productivity and Satisfaction
• Editing Any perceptible delay may prove
  irritating
• R.B. Miller ? Immediate response is not a
  universal requirement in interactive computing
• Lists various class of user actions and allowable
  delays ? best guesses

32
(No Transcript)
33
Millers experiment

• Effects of varying CRT display rates and output
  delays on user performance
• Delay Increasing the display rate from 1200 to
  2400 baud produced no significant performance or
  attitude changes
• Variance Increasing the variability of the
  output display rate produced a significant
  deterioration in both performance and attitude

34
Grossbergs Experiment

• Problem Solving Context System response time has
  little effect on performanceWhy?
• Users simply adjust their tactics to make best
  use of their time on system
• Response Times in problem solving activities
  varied 1,4,16 and 64
• As mean delay increased , users became more
  cautious and deliberate
• However, no definite effect on time required to
  reach solution

?
35
Transferring to the Editing environment

• Editing ? a routine cognitive skill
• Additional mental preparation time not useful, in
  fact would interfere with the task completion
  time ( because of irritation )
• Experiments are always motivated to complete
  their tasks, but not in the real world

36
Editor Design Considerations
37
How can we make editor easy to use ?

• Depends on the
• Command language of the editor
• Underlying structure ( editor states or modes)
• Tradeoff
• Our inability to learn, remember, and effectively
  use large complex command sets vsdesire to
  achieve editing objectives within minimum time
• Limits range of design options

38
Many approaches

• Popular wisdom
• Principle of Predictable Behavior
• User Engineering Principles
• Observation
• Dzidas Questionnaire study User perceived
  quality as a multi-dimensional concept
• Identified 7 major categories
• Learning Process
• 1. Self teaching through trial/error with machine
  feedback most effective
• 2. Anxiety decreased learning

?
39
Controlled Experiments

• Command language structure and learning ability
• Whether user options are good for everyones
  performance ?
• Experiment Two versions of editors
• Inflexible full commands, no abbrv., extra
  spaces, or defaults allowed
• Flexible lot of freedom

?
40
Results

• Flexibility pros and cons
• More prone to syntax errors
• Completed tasks faster
• Role of English-similar commands
• It is more helpful ?
  
  

?
41

• Tested with two versions of same editor (NOS)
• Typical Notational Syntax
• Legitimate English phrases
• 24 paid subjects
• English version
• Completed more tasks
• Error rate was lower
• Editing efficiency was better
• ? Surface syntax of an editor is surprisingly
  important from human engineering point of view

42
(No Transcript)
43
Input and Output Devices

• Psychological and Human factors underlying design
  and use of keyboards, screen displays and
  pointing devices

44
Key Entry

• The most common means of encoding letters and
  numbers

Keyboard Devices
Oldest typewriters
Teletypewriters
Electric typewriters
?
Detailed research exists in keyboard design
45
Detailed research exists in keyboard design

• Keyboard layout (e.g. QWERTY)
• Numeric keys
• Standard Key size
• Slope of keyboard
• Key depression force required
• Key displacement
• Type of Kinesthetic feedback from key actuation

46
Typing speed

• Some factors
• Finger ballistics
• Reaction time
• Motor learning
• Short term memory
• Human information processing capacity
• Average single finger tapping rate 6 keys/s

Little finger ? Index Finger 20 increase
47
Some interesting stats

• Good typists 0.2 secs per keystroke ( 50
  words/min)
• Less Frequent users 0.7 secs
• Experienced Typists 0.08 secs (12 taps/sec)
• Typing with alternate hands 25 faster than with
  one hand
• ? Control the necessary echo output rate for a
  display

48
Effects of Training

• Poor typing habits are difficult to shed
• Self-taught typists do not reach even half the
  speed expected from entry level typists
• Worth considering the benefits of specialized
  training

49
The Shaffer and Hardwick Experiment

• Limitation of human information processing
  capabilities
• Material
• Difficult but coherent text
• Randomly selected words
• Short words of Random character sequences
• Explanation
• Acquisition of a hierarchy of habits ( ability
  to type a whole word as a single unit)
• Able to read farther ahead, as opposed to random
  characters

?
50
0.159 secs per keystroke
main() unsigned paddr,pdata LOOP
printf("Input port address (hex) ")
scanf("x",paddr) pdata inp(paddr)
printf("Port(xh) xh\n",paddr,pdata) goto
LOOP return 0
More than double the time
0.162 secs per keystroke
51
Error rates

• Error rates vary much from operator to operator
  than does speed
• Effect of visual feedback
• Masking the keyboard
• Masking the printed text

• reduces the speed and accuracy
• Error 0.9 ? 2.6
• Speed 30
• reduces only the accuracy
• Error 40 increase !

52
Other studies

• Signaling errors immediately is helpful ( 25
  faster)
• Automatic error correction many editors
  incorporate it

53
Conclusions

• Lot of scope for studies of human factors aspects
  of the use of editors for searching , inspecting
  and maintaining file systems using interactive
  text-editors
• Many research areas exist(ed)
• Underlying model of information structure
• Techniques for selecting small segments of text
• Form of editor commands
• 2-D editors vs 1-D editors
• Error feedback and benefits of automatic error
  correction
• Split-screen and multiple-screen editing
  operations
• Screen size and material exposed to user
• Color Displays
• Audio Input, audio feedback
• Direct use of eye movement for pointing and menu
  selection