Selection of Action

1
Selection of Action

• chapter 9

2
Response Times

• Two classes of RTs
• Simple RT (aka Reaction Times)
• Choose whether to respond or not
• Go No-go tasks
• Example drag racing
• Choice RT (aka Response Times)
• Must choose which response to make
• Exampl avoiding an accident
• Steer left?
• Slam on brakes?

3
Variables that influence All RTs

• Stimulus Modality
• Auditory is quicker than visual
• Stimulus Intensity
• RTs decrease as intensity increases
• Accumulation model
• Assumes that a decision is made once enough
  evidence has been accumulated.
• Increasing the intensity increases the rate that
  information is transmitted, decreasing the time
  it takes to accumulate a satisfactory amount of
  evidence.

4
Simple Accumulation Model
5
Simple Accumulation w/noise
6
Temporal Uncertainty

• Terminology
• Warning Signal
• Informs that the Imperative stimulus will appear
  shortly.
• ReadySet
• Imperative Stimulus
• The stimulus that is responded to
• GO!
• Warning Interval
• The interval between the Warning Signal and the
  Imperative Stimulus

7
Temporal Uncertainty

• Temporal Uncertainty
• The predictability of the warning interval
• The more predictable the warning interval, the
  easier it is to focus attention on a specific
  time window.
• If the imperative stimulus (go-signal) occurs
  during the expected time window, it is responded
  too faster.
• However, if the warning interval is highly
  variable, people wont be ready for the
  imperative stimulus and responses will be slowed.

8
Temporal Uncertainty

• Attention increases rate of accumulation.
• That is, attention increases our sensitivity to
  a stimulus.
• ? The more attention you pay to an interval in
  time, the more quickly you will accumulate
  evidence about the imperative stimulus, and the
  quicker you can come to a decision and make a
  response.

9
Warning Interval Length

• People arent good at judging time, and the
  longer the time, the greater the variability in
  the judgment.
• Example an exactly 5 second interval will show
  less variability in judgments than a 30 second
  interval.
• ? the greater the warning interval, the greater
  the temporal uncertainty, and therefore the
  greater the RT.

10
Warning Interval Length

• Exceptions
• Too short to prepare
• If the warning interval is too short, people
  might not have enough time to prepare.

11
Temporal Uncertainty Applications

• Warning intervals that are too short might not
  give enough time to get ready.
• e.g. Yellow traffic lights that are too short in
  duration
• Warning intervals that are too long can lead to
  complacency
• e.g. 30 second draw-bridge warning.

12
Expectancy

• The variability of warning intervals, on average,
  slow responses.
• Even if the warning interval is random, people
  can still pick up on the pattern of randomness
  and use it to their advantage.
• That is, the pattern of variability also affects
  RTs.

13
Expectancy

• Example
• If you run track, and you know that the
  regulations say that the gun must go off between
  5 and 10 seconds after set then you have some
  idea about the when the gun will go off.

14
Expectancy

• Lets say that the probability that the gun will
  go off at any given time is equally distributed

15

• What happens if at given time, the gun hasnt
  gone off yet?
• The probability that the gun will go off at the
  next moment increase!

probability that Go! will occur at time t
given that it already hasnt gone off by time t
-1.
16
Expectancy

• Therefore, as time goes on and the window for the
  Go! event narrows down (you become more
  certain), you response times will actually speed
  up!
• This function of certainty over time for when an
  event will occur (given that it hasnt already)
  is known as a Hazard function.

17

• Probability Density Function
• Probability that an event will happen at time t.
• For the starters gun, the probability was flat.
• That is, each time had an equal opportunity to be
  randomly chosen..

18

• Cumulative Distribution Function
• The cumulative probability that an event will
  have happened by time t.
• For the starters the gun must go off after 10
  seconds, so that P() 1 at that point.
• Calculated by integrating PDF

19

• Hazard Function
• The instantaneous probability that an event will
  happen at time t given that it already hasnt
  happened.
• For the starters the gun must go off after 10
  seconds, so that P() 1 at that point.

20
Hazard Function

• Increasing Hazard function
• With increasing Hazard functions, you should be
  pay more attention towards the end of the warning
  interval because that has the least uncertainty.
• Keep in mind, that the end might not be the most
  likely time for the imperative stimulus to occur,
  but the end will have the least uncertainty.
• Flat Hazard Functions
• Uncertainty does not increase over time,
  therefore...
• Attention should not increase over time.

21
Famous Hazard Functions

• Serial Self-Terminating Search
• Each item is examined one at a time until the
  target is found. Items are never reinspected.
• Sampling without replacement.
• Flat PDF distribution
• Identical to starters gun example.
• After N-1 items, if you havent found the target
  yet (and you know it is present), you can be
  certain that the next item must be the target.
• That is, P() of finding the target increases as
  more and more items are examined (if the target
  is there).

22
Famous Hazard Functions

• Memoryless search
• Each item is examined one at a time until the
  target is found. Items are chosen at random, and
  there is no limitation in reinspecting items
  (i.e. no memory).
• Sampling with replacement.
• Exponential PDF distribution
• Since you are sampling with replacement, the
  probability that you will randomly stumble across
  a target does not increase as time goes on.

23
Visual search has memory, Peterson et al.,
Psychological Science, 2001in reply to Horowitz
and Wolfes Visual search has no memory,
Nature, 1998
24
Applications of Hazard Functions

• Drag Racing
• Use an exponentially distributed Warning Interval
  to prevent jump starts.
• Engineering Reality there is a real probability
  that the warning interval could last forever (or
  at least a really long time!) Engines could
  overheat, fans could get restless
• Solution Use catch trials in which the light
  does not turn green.

25
Applications of Hazard Functions

• Stop Light
• Goal is to prevent people from running red
  lights.
• Make Warning Interval (how long the yellow light
  stays on) a flat and narrow distribution.
• Maximizes certainty about when the light will
  turn from yellow to red.
• Unexpected Events
• Truly rare events have high uncertainty, and
  therefore are responded to more slowly.
• Example I had a squirrel fall out of a tree once
  and into the path of my car.

26
Things that influence Choice-RT only

• Hick-Hyman Law
• RTs are a function of the amount of stimulus
  information needed to make a decision.
• As the number of alternative choices increases,
  so does the amount of information, and therefore
  RTs ?.
• As number of alternative ?, RTs increase at a
  negatively accelerating rate.

27
Things that influence Choice-RT only

• Hick-Hyman Law
• As number of alternative ?, RTs ? at a negatively
  accelerating rate.

28
Things that influence Choice-RT only

• But, as amount of information increases linearly,
  RTs also increase linearly.
• Information log2(alternatives)

29
Things that influence Choice-RT only

• Things that influence the amount of information
  also influence RTs
• Probability
• Low(rare) lots of information, slow responses
• High(common) less information, faster responses

30
Things that influence Choice-RT only

• Speed-Accuracy Trade-off
• People can adjust their criterion for how much
  evidence they need before responding.
• If they set their criterion too liberal, they
  will need less information and respond more
  quickly, but many of those responses will be
  errors.
• That is, by adjusting their criterion, people can
  trade-off accuracy for speed.

31
Random Walk

• Evidence is accumulated over time, and a decision
  is made when when enough information is
  accumulated.

32
Random Walk

• Lowering your criterion will lead to faster
  responses, but increases likelihood of errors.

Lower decision criterion (barrier) leads to
faster responses.
33
Random Walk

• Lowering your criterion will lead to faster
  responses, but increases likelihood of errors.

The lower the decision criterion (barrier), the
more likely an error will occur.
34
Speed-Accuracy Operating Characteristic

• Fast Accurate
• Sloppy Slow

35
Speed-Accuracy Operating Characteristic

• Trade-off
• Speed and accuracy can very along a single curve.
• The person can choose to be fast and sloppy or
  slow and accurate (or somewhere in between)
• Between curves
• A person might be better at one task than
  another.
• That is, a person might be good (fast and
  accurate) in one task and do more poorly in
  another task (slow and sloppy).

36
Micro trade-offs

• Fast Guess
• When the stimuli are highly salient
• Responding before an adequate amount of
  information has been accumulated.
• Random-walk example / Speed-Accuracy Trade-off.
• Errors faster than correct responses
• Slow Guess
• When saliency is low or stims are difficult to
  process.
• If the correct answer isnt readily apparent,
  people give up and guess.
• Errors are slower than correct responses.

37
Conclusions

• RTs affected by
• modality
• rate of information accrual
• decision bias
• expectancy (attention)
• Expectancy affected by
• warning interval length
• experience with the variability in the warning
  interval
• rarity of the event

38
Departures from Information Theory

• Stimulus Discriminability
• The more similar stimuli are to each other, the
  longer the RT.
• Example
• vs o ? X ?

L
L
L
T
T
39
Departures from Information Theory

• Repetition Effect
• When the same answer occurs twice in a row, the
  2nd response is faster.
• Why? Recently priming of pathways used for
  response.
• Exceptions
• Long Delays the same response may be slower
  Gamblers fallacy its unlikely for too many
  identical events to occur in a row.
• Rapid response with same finger there is a
  refractory period. If the delay is too short,
  the finger might not have recovered in time.

40
Departures from Information Theory

• Response Factors
• Effects of confusability
• using different fingers on the same hand is
  slower than using fingers on different hands.
• Controls with different shapes and feel are less
  likely to be confused

41
How do you change the channel?
42
Departures from Information Theory

• Response Factors
• Effects of Complexity
• The more complex the response, the slower the
  response.
• Latency to initiate typing a word is slower than
  latency for a single button press.

43
Departures from Information Theory

• Practice
• The more highly practiced, the quicker the
  response
• automaticity takes over
• the more difficult the task, the more it benefits
  from practice
• Example steering (small benefit from practice)
• manual shifting (large benefit f/ practice)

44
Departures from Information Theory

• Task Switching (Executive Control)
• When switching from one task to another, it takes
  some time to become ready for the new task.
• That is, before a new task can be worked on, the
  rules for the task must be loaded

45
Task Switching

• How many numerals? Which numeral?
• 111 333 3 333 1 333

46
Stimulus-Response Compatibility

• Location Compatibility
• Colocation Principle
• Controls should be located next to the items that
  they control.

47
Stimulus-Response Compatibility

• Location Compatibility
• Congruency
• The spatial layout of the controls matches the
  layout of what they control.

48
Stimulus-Response Compatibility

• Location Compatibility
• Rules
• Establishing consistent rules.
• Inside front burners
• outside back burners

49
Stimulus-Response Compatibility

• Movement Compatibility
• Its best to have the direction of the control
  movement match the pattern of display movement.
• Rotary dial displays with rotary controls
• Linear sliders should be used to move things

50
Stimulus-Response Compatibility

• Movement Compatibility
• Good Design Mercedes-Benz seat controls
• Controls resemble miniature seat.
• To adjust seat, move control in same direction

51
Poor design Apple Quicktime 3 player scrub
mouse pointer up to rotate thumbwheel. Movement
(up) is incongruent with movement of display
(rotatation)
52
Stimulus-Response Compatibility

• Common Movement rules
• When movement compatibility is not possible, its
  best to stick with these common rules
• Example volume control
• Clockwise -gt increase
• move control up -gt increase
• move control right -gt increase
• move control forward -gt increase

53
Stimulus-Response Compatibility

• Movement Proximity (Warwick Principle)
• The closest edge of a control should move in the
  same direction as the display, as if they were
  mechanically connected.
• That is, spatial location also affects movement
  compatibility.

54

• Good Bad Best

?
Obeys common rules Common rules Rules and
Warwick Warwick conflict with match each
other.
55
Stimulus-Response Compatibility

• Compatibility Ambiguity
• Sometimes the general rules for movement
  compatibility do not match operators mental
  model.
• Example Flight control
• According to the general rules, pushing forward
  should increase altitude. Instead, pushing it
  forward decreases altitude.
• Correct mental model.
• Pushing stick forward make the nose pitch
  forward, causing the plane to descend.

56
(No Transcript)
57
Affordance

• Affordance
• The basic shape of the object suggests how it
  should be used.
• Knowledge in the world.
• A handle affords (suggests) grabbing
• A putton affords pushing.
• A table (flat surface) suggests that you can put
  stuff on it.

58
Bad design

• Handles suggest pulling.
• Sowhy do these doors require the handles to be
  pushed in order to exit the corridor?

59
Stages in RT

• Subtraction Method (Donders)
• Each time you add another cognitive function to a
  task, RTs should increase by the amount of time
  it takes to perform that function.
• Example
• Try to find a white car in a parking lot full of
  white cars. As the number of distractors (cars)
  increases, so does search time
• Search time is the aggregate of all of the
  individual examinations.

60
Stages in RT

• Now try to find a red car in a parking lot full
  of white cars. Task is much easier. Donders
  assumption does not hold

61
Additive Factors Technique

• Assumes that brain ( mind) consist of different
  modules, or stages, that are tuned to perform a
  specific task (e.g. perception, language
  comprehension).
• Additive factors technique allows us to see if 2
  independent variables affect the same stage.
• If their effects on RTs interact, then we can
  assume that the independent variables affect the
  same stage of processing.
• If there is no interaction, then the IVs must
  affect different stages of processing.

62
Additive Factors Technique

• Question How does weather affect search for a
  tank in different terrains?

Marginal Means
Cell Means
63
Additive Factors Technique

• By looking at the marginal means, we can see that
  each factor has a main effect.
• Terrain affects detection (6.5 vs. 15 sec)
• Weather affects detection (7.5 vs. 14 sec)

64
Additive Factors Technique

• If the two variables independently affect
  processing, then the main effects (marginal
  means) predict these responses.
• Notice that since the lines are parallel, there
  is no interaction.

65
Additive Factors Technique

• However, the graph of our observed data, the
  lines converge.
• Therefore, the two variables interact.
• That is, the effect of weather when detecting a
  tank in a forest is more than you would predict
  when look at the terrain and weather variables
  alone.

66

• No interaction
• Should weather and type of firing response speed
  to shoot an enemy tank?
• Weather sunny vs. foggy
• Response trigger vs. voice activated.

67
Additive Factors

• Problems
• Requires
• serial assumption of stages of processing
• variables that slow or speed up processing in one
  stage do not affect later stages (as a
  consequence of earlier stages).

68
Psychological Refractory Period (PRP)

• Sometimes we need to perform two tasks in close
  succession. Often, the first task delays our
  completion of the 2nd task.
• Bottleneck both tasks need to use the same
  subsytem, but the subsystem can only handle one
  task at a time.
• That is, the 2nd task must wait for the 1st to
  complete.
• That means that a task (2nd task) that normally
  takes X amount of time will take even longer when
  it follows a task (T1) that uses the same
  subsystem.
• Note that PRP is not the same as task-switching.

69
Psychological Refractory Period (PRP)

• Example
• When you go to the doctors office, you (Task 2)
  often have to wait for the doctor (bottleneck) to
  be done with a patient (Task 1) before you can be
  seen (the delay in processing).

70
(No Transcript)
71
PRP
72
Decision Complexity Advantage

• According to Hick-Hyman, RTs are dependent on
  amount of information that needs to be processed.
• Therefore, can 3 2-bit tasks be completed as
  quickly as 1 6-bit task?
• No.
• Time to perform a task is dependent on
• time to load a task set
• perceptual and motor processes
• Example Morse-code slower than typing

73
Three 2-bit tasks
One 6-bit task
74

• http//www.baddesigns.com/