psychophysical methods
TRANSCRIPT
PSYCHOPHYSICAL METHODSZ. SHI
Course C - Week 5
1
Let’s do a detection task
Please identify if the following display contain a letter T. If Yes, please raise your hand!
T among Ls
2
+
3
L
LL
LL
L
T
L L
X
XX
XX
X
X
X X
1
+
4
L
LL
LL
L
L
L L
X
XX
XX
X
X
X X
2
+
5
L
LL
LL
L
L
L L
X
XX
XX
X
X
X X
3
+
6
L
LL
TL
L
L
L L
X
XX
XX
X
X
X X
4
+
7
L
LL
LL
T
L
L L
X
XX
XX
X
X
X X
5
+
8
L
LL
LL
L
L
L L
X
XX
XX
X
X
X X
6
Results
Trial No Yes No
1 (Present) 1 15
2 (Absent) 0 16
3 (Absent) 0 16
4 (Present) 14 2
5 (Present) 16 0
6 (Absent) 0 16
9
Conditions Presentation time (sec)
P(‘Yes’)
1 0.2 1/16
2 0.4 14/16
3 0.6 16/16
• Non-linear relation between physics and psychology
• Senses have an operating range
Stimuli and sensation
10
Stimulus intensity – physical property
Sen
sati
on
– p
sych
olo
gy
Undetectable region
Saturated region
Point of subjective equality (PSE)
• Is the stimulus vertical?
11
% V
erti
cal r
esp
on
se
0%
100%
50% Point of Subjective Equality - PSE
Just noticeable difference (JND)
• Difference in stimulation that will be noticed in 50%
12
% V
erti
cal r
esp
on
se
25%
75%
–
2= JND
= Upper threshold
= Lower threshold
= Uncertainty interval
JND and sensitivity
• Which psychometric function, full or dashed line, exhibits a greater sensitivity?
• Dashed - the smaller the JND the greater (steeper) the slope, and greater the sensitivity is
13
Psychometric function
• Absolute thresholds (Absolute limen)
the level of stimulus intensity at which the subject is able to detect the stimulus. Some time it is called point of subjective equality (PSE)
• Discrimination thresholds (Difference limen)the difference between two stimuli intensities that the participant is able to detectJust Noticeable Difference (JND)
14
Psychometric function and threshold percentiles
• PSE – threshold percentile is half-way between the minimum and maximum of the function
• BUT – the min and max vary depending on the task
• SO – the exact threshold percentile depends on the task
Threshold percentiles
16
0%
100%
50%
25%
75%
Yes/No
2AFC
3AFC
50%
75%67%
MEASURING THRESHOLDS
17
Measuring psychometric function
18
Psychophysical methods
Classical methods
The method of constant stimuli
The method of adjustment
The method of limits
Adaptive methods
Staircase methods
Parameter estimation by sequential testing (PEST)
Maximum-likelihood adaptive procedures (QUEST, MLP)
1. The method of constant stimuli
• Full control of presented stimuli
• Several fixed stimuli are presented in a random order, many times
• For each stimulus participants perform the same task, e.g., whether or not they see a stimulus
19
Example
20
Method of Constant Stimuli & Psychometric Function
21
• Multiple Intensity levels, multiple repetition
• Prop. of responses • Represented roughly
shape of psychometric curve
• Estimation of Psychometric curve: JND, PSE
100%
Intensity
50%
Proportion of “yes” responses
---+----
-----+--
-+---+--
--++-+--
--++-+-+
+-++-+-+
++++-+-+
++++-+-+
Method of constant stimuli - evaluation
• Whole psychometric function can be estimated
• Many trials are:
• Costly
• not for special groups ( e.g., kids, ADHD)
• Repeating the same stimulus many time can yield learning effects
22
2. Method of limits
1. On some trials the intensity ascends from low to high intensity• When the stimulus or a difference between stimuli is
noticed the trial stops 2. On other trials the intensity descends from high to low
intensity• When the stimulus or a difference between stimuli is not
noticed any more the trial stops3. Averaging the stopping values across several trials yields
thresholds
23
24
Stim
ulu
s In
ten
sity
(B
righ
tnes
s)
n
n
n
n
y
n
n
n
n
n
y
n
n
n
n
n
n
y
95
96
97
98
99
100
101
102
103
104
Transition point
y
y
y
y
y
n
y
y
y
y
y
y
y
n
y
y
y
y
y
y
n
Run 1 Run 2 Run 3 Run 4 Run 5 Run 6
(99+98)/2= 98.5 100.5 99.5 98.5 97.5 99.5 = 99
Example of the method of limits
25
• Sound
Method of limits and psychometric function
26
• Psychometric function• Not required
• Multiple ascending and descending method of limits measure• PSE
Threshold
Internal criterion
Stimuli
ascending
descending
ascending
descending
Prop
. Of ‘
Yes
’
Method of limits - evaluation
• Estimates only certain points on the psychometric curve
• Prone to habituation errors • Falsely increasing thresholds on ascending trials.• Falsely decreasing thresholds on descending trials.
• Prone to expectation errors• Anticipation of the stimulus arrival and prematurely report.• Falsely decreasing thresholds on ascending trials.• Falsely increasing thresholds on descending trials.
27
How to prevent errors
• Shorten trial series – avoids habituation errors• Variable starting points in the series – avoids
expectation errors• When comparing different stimuli, i.e., standard and
test counterbalance their position and order
28
ascending
descending
ascending
descending
Stimuli Intensity
3. The method of adjustment
• Subject controls (adjust) the stimulus intensity
• Absolute threshold – Adjust stimulus intensity so that the stimulus is barely perceived
• Discrimination threshold – Adjust one stimulus so that it match the standard stimulus
• Average error of all trials
29
• Threshold is the mean of all trials
Example
30
Trial 1 Trial 2
Method of adjustment and Psy. function
31
• Psychometric function• Not required
• Threshold• Internal criterion of
responses• Continuous
adjustment of stimuli intensity
• Fine tuning at the end of trial• PSE
Threshold
Internal criterion
Stimuli
Prop
. Of ‘
Yes
’ Ti
me
Trial 1.
Trial 2.
Method of adjustments - evaluation
• Estimates only certain points on the psychometric curve
• Faster than the method of limits – a few trials suffice
• Habituation errors still present
• Expectation errors still present
32
Methods of adjustments in clinics
33
Measuring psychometric function
34
Psychophysical methods
Classical methods
The method of constant stimuli
The method of adjustment
The method of limits
Adaptive methods
Staircase methods
Parameter estimation by sequential testing (PEST)
Maximum-likelihood adaptive procedures (QUEST, MLP)
Staircase method
• Addresses the problem of choosing “the right” stimulus values• Psychometric function is only sensitive to the middle range
of stimulus values
• Presenting stimuli beyond this range is not informative
• Up – Down rule: after a transition point, present a stimulus that goes in the opposite direction:• 20 dB – yes, 15 dB – no, 20 dB
• 15 dB – no, 20 dB – yes, 15 dB
35
Zhuanghua Shi 36
From the method of limit to staircase
Stim
ulus
Int
ensi
ty
(Bri
ghtn
ess)
n n n 95
n n n 96
n n n 97 n y n n 98
n y y n y 99 y y y n y 100
y y y y 101 y y y 102 y y y 103 y y y 104
n
n
n
Stim
ulu
s In
ten
sity
(B
righ
tnes
s)
n
y
n
n
y
n
n
y
95
96
97
98
99
100
101
102
103
104
Transition point
y
n
y
y
y
n
y
y
y
y
y
y
n
Run 1 Run 2 Run 3 Run 4 Run 5 Run 6
98.5 100.5 99.5 98.5 97.5 99.5
37
= 99
One-up – one-down staircase
• Procedure:• Start from above (below) threshold• Stop after a given number of transition points• Average final several transition points
38
How to estimate other points on the curve
• One-up – one-down represents the 50th percentile
• Varying the number of up/down steps converges at different percentiles
• One-up – three-down, 79%
• One-up – two-down, 71%,
• One-up – one-down, 50%
• Two-up – one-down 29%
• Three-up – one-down, 21%
39
How to estimate other points on the curve?
• Weighted up/down methods
•where
• For 75th percentile, p = .75, up step size should be 3 times of down step size
40
Staircase method - evaluation
• Hysteresis effect – the starting point makes a difference• Higher thresholds for ascending staircase then for
descending
• Expectation effect• Many trials necessary
41
Controlling for hysteresis and expectation
• Interleaved staircases
42
Parameter Estimation by Sequential Testing (PEST)
• Variant of weighted step-size method
• Large steps at the beginning
• Changing the step size as the run proceeds, in a particular way
43
Parameter Estimation by Sequential Testing (PEST)
1. With each response reversal, the step size is halved
2. When the minimum is reached, the step size is constant
3. When there is no reversal, the first two steps keep the same size
4. From the 3rd onward steps are doubled
5. If a reversal follows a doubling of step size, the 3rd step stays the same
44
Adaptive staircase methods (QUEST and MLP)
• Uses prior knowledge about psychometric function:
• What is the function? cumulative normal, logistic, …
• Successive trials are evaluated as evidence that they belong to one or another psychometric function
• By computing maximum-likelihood that the trial belonged to a function
45
Adaptive staircase methods (QUEST and MLP)
• For response on each trial one can compute its probability on the basis of different functions
• The function that produces the best fits for most trials is the winner
46
50%
Seen
Unseen
Adaptive staircases - evaluation
• Random stimulus presentation – little risk of expectation or habituation errors
• Efficient – uses all responses so only few trials suffice
• Accurate – up to the extent to which the assumptions are satisfied
• Risky – if the assumptions are invalid it will not produce reliable results
• Complex to learn and understand
47
QUEST
• A Bayesian adaptive psychometric method
• Watson & Pelli, Perception & Psychophysics, 1983
• available in Psychtoolbox (psychtoolbox.org).
• Method• User all prior knowledge to guide the placement of the
trials by maximum likelihood estimation of Bayesian prob.
• Psychometric function and shape must be assumed
Zhuanghua Shi 48
Quest provided by Psychtoolbox
• q=QuestCreate()• Create parameters of a weibull psychometric function
based on previous knowledge and guess• q=QuestCreate(tGuess,tGuessSd,pThreshold,beta,delta,ga
mma)• Beta: steepness of the PF• Delta: lapse rate, e.g. 0.01• Gamma: guess rate
• QuestQuantile(q);• set the stimulus level
• QuestUpdate(..);• after response, update probability density function (PDF)
Zhuanghua Shi 49
MLP: Adaptive maximum likelihood
• David Green (1990, 1993)
• Implemented by Grassi & Soranzo, 2009
• MLP: Matlab toolbox for rapid and reliable auditory threshold
• Similar to QUEST, but• Use logistic function as general psychometric function
• Downloadable:
• http://www.psy.unipd.it/~grassi/mlp.html
50
Bayesian methods - evaluation
• Pros
• Efficiency and accurate
• Using all information (responses)
• Cons
• Complex procedure
• Psychometric function must be determined
• Variability of the threshold depends on particular p-target.
Adapted from Grassi, 2009 51
52
Psychophysical methods
Classical methods
The method of constant stimuli
The method of adjustment
The medhod of limits
Adaptive methods
Staircase methods
Parameter estimation by sequential testing (PEST)
Maximum-likelihood adaptive procedures (QUEST, MLP)
Summary
ascendingdescendin
g ascendingdescendin
g
Tim
e
Trial 1.Trial 2.
Summary
• Stimulus-to-sensation mapping is non-linear
• Sigmoidal, logarithmic, exponential
• Describing the mapping function:
• Absolute and discrimination thresholds
• Classical methods – time consuming, but little assumptions
• Adaptive methods – more efficient, but only if assumptions hold
53