Engineering Psychology PSY 378S

1
Engineering PsychologyPSY 378S

• University of Toronto
• Spring 2004
• L7 Attention

2
Outline

• Attention
• Three kinds of attention
• Selective attention
• Supervisory control sampling
• Eye movements/UFOV
• Visual search
• Divided vs. focused attention
• Global (parallel) and local processing
• Emergent features
• Spatial proximity
• Response conflict
• Object-based proximity
• Proximity compatibility principle Display
  design recommendations

3
Attention as Spotlight

• We attend where beam falls
• Use this metaphor to distinguish among 3 types of
  attention

4
Attention as Spotlight

• 1) Selective
• Information sometimes outside of beam we should
  attend to it
• need to attend to something else
• Need to change direction of spotlight
• e.g., so engrossed in task, forget about other
  things

Hey, over here!
5
Attention as Spotlight

• 2) Focused
• Other irrelevant information in beam, need to
  narrow beam
• Need to concentrate on just one source
• Other sources of information in the environment
  serve as distraction
• e.g., someone talking, youre trying to study

Must..focus!
6
Attention as Spotlight

• 3) Divided
• Must get information from various sources, need
  to widen the beam
• Need to monitor many things at once
• e.g., harvesting wheat and hearing the results of
  the Vandal football game on the radio--
  Touchdown
• cant go driving all over the field

Need..to..widen!
7
Selective Attention

• Supervisory Control Sampling
• Operator scans display of a complex system in
  order to control it (power plant, aircraft)
• Location known, information unknown
• Target Search Sampling
• Operator trying to locate target (radar, search
  and rescue)
• Information known, location unknown
• Directing Attention (in text)
• Cueing location (Hey over here!)
• Attention in Depth (in text)
• Potential for reduced clutter

8
Supervisory Control Sampling
Which channel should be sampled more?
9
Supervisory Control Sampling

• Engineering psychologist divides world into
  channels along which events can periodically
  occur
• Signal detection problem Hit (detecting an
  event) Miss (do not observe an event)
• Sensitivity, Bias
• Optimal sampling

10
Experimental Results inSupervisory Control

• Mental model guides sampling
• When and where do events occur
• Improves with expertise
• Design frequently sampled displays should be
  placed centrally
• Sequentially sampled pairs of displays should be
  located together
• Adjustment to event rate
• people sample channels with higher event rates
  more frequently
• e.g., speedometer vs. gas meter vs. engine
  temperature
• sluggish beta

11
Experimental Results inSupervisory Control

• Sampling affected by arrangement
• Operators more likely to make horizontal than
  diagonal scans
• Reluctant to diagonally scan to high event rate
  channel
• Event rate not enoughposition matters
• Memory imperfect sampling imperfect
• Oversampling of channels with low event rates
  (sluggish beta)
• People forget to sample especially if must
  request/select info (channel info not directly
  observable)
• Importance of sampling reminders

12
Experimental Results inSupervisory Control

• Preview helps
• When observers given preview of scheduled events,
  sampling becomes more optimal
• Compromised by number of channels, complexity of
  information
• Processing strategiescognitive tunneling
• Downside of mental model
• When one channel fails, stop processing other
  channels
• Fixate (lock on to) channel being controlled

13
Visual/Target SearchEye Movements

• Based on expectancies (top-down, ?)
• Novice vs. expert
• Scan path will change with task
• Display factors and salience (bottom-up, d?)
• Large, bright, colourful, dynamic, abrupt onset ?
  attract attention
• Physical location (left to right, centre first
  edge effect)
• Singletons (unique stimuli)

14
Visual Fixations

• Foveal area of fixation (approx 2 deg)
• Useful Field of View (UFOV)
• Effective area within which information can be
  extracted
• Min(distance between successive fixations) in
  visual search task
• Varies between 1 and 4 deg
• Dwell time
• Length of fixation

15
UFOV

• Affects
• Driving Behaviour (smaller UFOV?visual search,
  driving)
• Affected by
• Density of information
• Discriminability of target from background
• Aging ? UFOV
• UFOV doesnt blanket area
• Targets may not be identified in UFOV
• Airborne search and rescueonly 53 of terrain
  covered (Stager Angus, 1978)
• Fixated but not detected (implies decision
  criterion)
• Training as solution

16
Dwell Time

• Typical fixation 250-500 ms
• Longer dwells (fixations) on
• Less legible displays
• Less familiar words
• More difficult text, display instruments (greater
  information pickup)
• Unusual objects, out of context
• Survey vs. examination dwell

17
Visual Scanning Implications

• Scanning results tell us about
• Internal expectancies
• Novice vs expert scanning patternstell us about
  mental model/search strategy
• Diagnosticsfrequently watched instruments are
  those most important to operator
• Place in prominent location or close together

18
Factors Affecting Search Rate

• Number of items
• Typically serial search
• Target distinguished by single level of
  dimension/simple rule
• Typically a parallel search
• Discriminability
• Hard to distinguish ? serial (inefficient)
• Easy to distinguish ? parallel (efficient,
  preattentive, pop-out)

Serial
RT
Parallel
Set Size
19
Factors Affecting Search Rate (contd)

• Absence/presence of target features
• Absent (O in Qs) ? serial
• Present (Q in Os) ? parallel
• Spacing
• Little effect scanning and clutter trade off
• One or more targets
• Generally less time if single type of target
• Unless all targets discriminable along single
  dimension (e.g., diagonal stroke in K, X)

20
Factors Affecting Search Rate (contd)

• Training
• If one can train to automaticity, leads to
  parallel processing
• Only where targets are consistent, must maintain
  mapping

21
Visual Search Applications

• Symbol coding
• Unstructured search Icons on desktop,
  commercial and information signs while driving
• Vehicle dispatcher
• Scanning digital map to locate the vehicle that
  is not in service and has large capacity
• Military commander
• Find infantry platoons on digital terrain model

22
Break
23
Divided Attention vs. Focused Attention

• Often good to divide attention (multi task)
  e.g., in high demand air traffic controlit is
  sometimes impossible to narrow when needed
• Divided attention mandatory rather than optional
• Display designs or principles that facilitate
  divided attention impair focused attentiona
  trade off

Need..to..widen!
Must..focus!
24
Parallel Processing and Divided Attention

• Preattentive phase
• Automatic
• Organizes visual world into objects and groups
• Global or holistic processing
• Short term sensory store
• Subsequent selective attention
• Further elaboration
• Perception
• Local processing

25
Selective
Pre-attentive
26
Display on right allows for preattentive grouping
of elements
27
Organized Displays

• Follow Gestalt principles
• And information theory
• Probability in context redundancy
• Knowledge of location of one display item allows
  better prediction for another
• Information theory based measures of display
  organization (Palmiter Elkerton, 1987 Tullis,
  1988)

28
Global and Local Processing

• Global processing
• All items of organized display processed together
  in parallel
• Local processing
• Processing single object in display

29
What is small letter?
30
What is small letter?
31
What is large letter?
32
What is large letter?
33
Global and Local Processing

• Response Conflict (Navon, 1977)
• Global?Local
• Automatically process large, affects small
• Asymmetric interference from small does not
  affect large (Local does not affect Global)

34
Emergent Features

• Related to global processing concept
• Emergent feature is global property of set of
  stimuli (or displays) not evident as each is seen
  in isolation
• Speeds processing (because preattentive)
• Only works if
• organization is compatible with task
• Gestalt principles (based on information theory
  expectation/redundancy constructs) upheld

35
Spatial Proximity and Divided Attention

• How can we improve divided attention?
• Expect spatial proximity will support parallel
  processing ? divided attention.
• There is evidence that it does
• Head-Up Displays (HUDs) as example
• Idea is to superimpose aircraft instrument
  information on view of real world

36
Do HUDs Help?

• Both information channels (instruments and view
  out the windshield) can be attended in parallel,
  or at least reduce transition time
• HUDs improves control of position during landing
  (Wickens Long, 1995)
• However, spatial proximity does not guarantee
  that parallel processing will occur
• Airplane poised to move out less likely to be
  seen with HUD (Wickens Long, 1995)

37
Expectation

• HUD helps divided attention when pilot expects
  stimulus
• And hinders when the stimulus is unexpected
  (airplane on runway)

38
(No Transcript)
39
(No Transcript)
40
Spatial Proximity Conflicts

• Penalty involved in putting information close
  together
• If two attentional channels are close together,
  they will both be processed, even if only one is
  desired
• A failure to focus attention

41
Response ConflictEriksen Eriksen (1974)

• Control
• H
• Display Clutter
• K H K

• Redundancy Gain
• H H H
• Response Conflict
• F H F

H right button F Left button
42
Response ConflictEriksen Eriksen (1974)

• Display clutter
• Confusion, failure to focus between relevant and
  irrelevant
• Response conflict and redundancy gain
• Proximity allows two channels to be processed in
  parallel, even if undesired
• Information competes at perceptual level when
  there are different implications for action
• If they disagree, RT ? -- leads to response
  conflict
• If they agree, RT ? -- leads to redundancy gain

43
Object-Based Proximity

• Similarity can be based on more than spatial
  proximity
• What if different information channels are
  different dimensions of the same stimulus?
• Colour, shape, location, size, etc.
• Stroop task name ink colour
• XXXX XXXX
• GREEN RED
• Stroop effect RT and error ? when colour name
  and ink colour conflict (relative to neutral
  control)
• Redundancy gain also observed

44
Object-Based Model

• Multiple dimensions belonging to single object
  likely to be processed in parallel
• Likely to help if parallel processing required
• But hinder if one dimension is irrelevant
• GREEN RED
• Attention not just space-based (spotlight)
• Object-based model says concurrent processing
  occurs when elements lie within single object,
  independent of its spatial dimensions
• Judgments made about two parts of same object
  made faster than when parts are from different
  objects, even when distance constant (Behrmann et
  al., 1998)

45
Object Displays

• So we can take advantage of this object based
  proximity
• Integrating information sources into one display
  if parallel processing required
• Emergent features property of whole, not seen
  when parts in isolation
• Connect spokes to form polygon

Safety parameter display (Westinghouse)
46
Focused Attention Tasks?

• What if the task requires focussed attention,
  will the object display be good?
• Not really
• Use appropriate object type for task(s)
• E.g., bar graph

47
Putting it TogetherProximity Compatibility
Principle

• If task requires HIGH processing proximity, then
  should have HIGH display proximity
• If task requires LOW processing proximity, then
  should have LOW display proximity
• Processing proximity extent to which
  information sources are used in the task
• Display proximity how close display elements
  are (distance or object-based)

48
Using Multiple Sources of Information

• Integrating multiple information sources by
  proximity or common representation (objectness,
  colour, etc.) results in
• Increased probability of parallel processing
• Increased likelihood of emergent features with
  improved performance if they are mapped to the
  task
• Possible creation of clutter or response conflict
  if the task requires focussed attention on one
  dimension

49
Using Multiple Sources of Information

• Separating multiple information sources by moving
  them apart or putting them in separate objects
  results in
• Increased probability of local processing
• Reduction of clutter or response conflict if the
  task requires focussed attention on one dimension
• May still allow emergent features (bars in bar
  graph)

50
Summary

• Attention
• Three kinds of attention
• Selective attention
• Supervisory control sampling
• Eye movements/UFOV
• Visual search
• Divided vs. focused attention
• Global (parallel) and local processing
• Emergent features
• Spatial proximity
• Response conflict
• Object-based proximity
• Proximity compatibility principle Display
  design recommendations