Please enable JavaScript.
Coggle requires JavaScript to display documents.
Behavioural Neuroscience - Psychophysics & Signal Detection Theory -…
Behavioural Neuroscience - Psychophysics & Signal Detection Theory
Psychophysics
Psychophysics
Scientific study of relation between external events & sensation
Relationship between physical stimuli & their subjective correlates/percepts
Fundamental methodology in experimental psych & behavioural science
Objectively Measuring Perception
Introspection
Inner feelings & perception in replicable manner
But not lend to quantification
Psychophysics
More quantitative method of analysis
Perceptual Performance
Sensitivity
Size
Bias
Sensitivity
How much stimulus needed to be able to detect it
Size
Size of effect of stimulus parameter on perception
Bias
How to account for possible bias in decision making
Signal Detection Theory
Signal Detection Theory
Provides method to distinguish between perception (signal & noise) & criterion (bias) in behavioural reports
Every investigator has same evidence but may have diff criteria
Can be widely applied to more cog qs & widely used in cog neuroscience
Two main components of information acquisition & criterion
Proportion of hits & false alarms depend on decision threshold
Signal Detection Theory History
Evolved from development of communications & radar equipment in first half of century
Adapted to psych in 1950s/60s as part of sensation & perception to attempt to understand some of features of human behaviour
When detecting very faint stimuli
General Process
Scenario where investigator searching for something
Either signal present or absent
Either investigators does or does not see signal
Four possible outcomes
Possible Outcomes
Hit
Miss
False alarm
Correct rejection
Hit
Signal present & investigator says yes
Miss
Signal present & investigator says no
False alarm
Signal absent & investigator says yes
Correct Rejection
Signal absent & investigator says no
Perception vs Bias
Try to measure what people perceive
But what report is affected by criterion they apply
Biases & criteria vary depending on diff factors
Factors
Individual
Context
Consequences of making wrong decision
Measures of Criterion
Signal
Noise
Signal
Distribution when what measuring for present
Hits vs false alarms
Noise
Distribution when what measuring for not present
Correct rejects vs misses
Distinguish Sensitivity
Separation of signal & noise curves measure observer's sensitivity independent of criterion
Distance between peaks
Changing criterion changes hit rate & false rates even if sensitivity unchanged
Need way to measure relative separation between signal & noise that doesn't depend on criterion
D-Prime
Measure of relative distance between response to signal absent (noise) & signal present (signal)
Calculating
Computed from hit rate (H) & alarm rate (F)
Treated as probabilities
Convert probabilities to z-scores to compute d-prime & bias
d (perceptual sensitivity) & c (bias) in units of standard devitions so independent on actual measurement units
H
Proportion of stimulus present trials
To which subject responded yes
F
Proportion of stimulus absent trials
To which subject responded yes
Calculations
d’ = z(H) – z(F)
c = -[z(H)+z(F)]/2
In Cognitive Neuroscience
Memory
Attention
Pain
Combine audition & vision
Memory (Lockhart & Murdoch, 1970
Attention (Hawkins et al., 1990)
Pain (Rollerman, 1977)
Combine Audition & Vision (Lovelace et al., 2003)
Study: Relation to Brain (Ross & Heeger, 2003)
A: Detection of contrast incrament
M: Yes/no response to slight incrament in background pattern
R: Decision made on each trial by comparing noisy internal response w. fixed criterion
Responses to false alarms are same as hits & diff to misses even in early visual areas
Implication
Internal noise affected stimulus response & resulting percepton
Predicted by SDT
Measuring Threshold
Thresholds
Visual
Auditory
Pain
Visual
Absolute threshold
Only 5-14 photons needed
Example: Visual
Brightnedd
Auditory
Absolute threshold
20 - 20,000 Hz frequency range
At 2,000 Hz humans can detect sounds from air vibrations less than width of atom
Example: Auditory
Amplitude (dB)
Frequency (Hz)
Pain
Differs depending on gender
Male higher than female (Chesterton et al., 2003)
Measuring Threshold
Detection task
Forced choice
Contrast sensitivity
Detection Task
Psychometric function on % seen as threshold level increases
Stimulus present as yes/no q
Threshold is 75% as that the point 3/4th between 0 and 100
Responses shaped by internal noise
Creates sigmoid shape like cumulative Normal Distribution
Issue
Yes/no decisions affected by bias
Measure what people perceive but what report influenced by criteria they apply
Diff intensity of what can see before class it as being seen
May say can see it to be perceived better
Especially if incentive
Two Alternative Forced Choice
Psychometric function on % correct as threshold level increases
Stimulus presented as two alternatives
Function no longer starts at 0 but instead at 50%
Even guessing would get 50%
Only below is detect & peversely choose opposite
So threshold is 87.5% as that the 3/4th ways of 50 & 100
Alternatives
Locations
Intervals
Advantage
Reduces bias
Can't always just claim see it
Own criteria no longer comes into play
Contrast Sensitivity
Contrast sensitivity function as decreasing contrast & increasing spatial frequency
Ability to perceive pattern varies w. spatial frequency (DeValois & DeValois, 1990)
Brain most sensitive to intermediate frequencies
Contrast Sensitivity Function
Visual system can only detect high & low spatial frequencies at relatively high contrast
In human visual system makes approximately inverted U shape
Peak suggests system most sensitive to medium spatial frequencies
2-4 cycles/deg
Advantage
Can be estimated for other animals
Allow direct comparison of perceptual abilities
Discriminating Stimuli
Discrimination Task
Measures difference threshold
Measures two key influences
Key Influences
Point of subjective equality (PSE)
Just noticeable difference (JND)
PSE
Point where equal no. times will say points to left & right
50% times answer tilted right
Direction of stimulus pushes away from what perceieved
Tilt Illusion
Surrounding pattern affects perception of orientation of inner lines
Is centre of region vertical vs is centre of region tilted left/right of vertical
Generates sigmoid response if original question rephrased
Discrimination of Tilt Illusion
Psychometric function as % times answered 'tilted right' as degrees from vertical increases
Measure perceived centre tilt for fixed surround orientation of 15 degrees
PSE where stimulus appears tilted to left/right equally often
What Curve Tells Us
Accurately measure tipping point of curve
This way results measured in consistent, systematic way
Is replicable
Instead of measuring single point can measure whole pattern of responses for full picture
Method of Constants
Observers presented w. fixed (chosen) set of stimuli
Size of illusion = (PSE a - PSE b)/2
PSE a = Measuring for fixed surround orientation
PSE b = Measuring for surround orientation
Issue
Takes long time to run experiments
Usually want/need to measure each PSE several times & on many conditions
Need to know approx where PSE lies
Need to sample far enough away & densely enough to have points that fit curve
Can be difficult to get right & piloting needed
Important to randomise presentation
JND
JND = (75% point - 25% point)/2
Steeper curve means smaller JND & higher sensitivity
Method of Adjustment
Staircase
Adaptive
Staircase (Levitt, 1971)
Response of left/right w. angle presented in each trial
Ends after fixed no trials
Details
One-down, one-up staircase converges to PSE
50% left & 50% right decisions
Different rules result in diff % points for threshold detection
Two+ staircases usually interleaved w. diff starting points
Avoid Ps predicting nect stimulus
Random starting points to avoid up-down bias
Adaptive Methods
Adaptive staircases
Adaptive curve selection
Adaptive Staircases
PEST
QUEST
Parameter Estimation by Sequential Training (Taylor & Creelman, 1967, Findlay, 1978, Pentland, 1980)
Quick Estimate by Sequential Training (Watson & Pelli, 1983)
Adaptive Curve Estimation
APE
Bayesian Adaptive Estimation Ψ
APE (Watt & Andrews, 1981)
Bayesian Adaptive Estimation Ψ (Konstevich & Tyler, 1999)
Critical Evaluation
Evaluation of Adjustment
Advantages
Limitations
Limitation
Problem w. task getting harder toward end
Fatigue/loss of concentration
Presentation order less random
Can lead to adaption/expectation effects
Advantage
Staircases much faster than method of constants
Fewer assumptions about PSE & range of stimuli
Though decisions about starting point & sampling need to be made carefully in simple staircases
Critical Evaluation of d-Prime
Advantages
Disadvantages
Notes
Advantages
Measure of sensitivity that is independent of observers's criterion
Does not assume existence of fixed threshold value
Separates out sensitivity from response bias
Limitations
Need to find stimuli that are close to threshold
Give limited info about response relative to stimulus
Notes
Often related measures such as A' used & reciever operating characteristic calculated
Criterion is called β or c and can also be calculated in various ways
Reveiver Operating Characteristic
Curves capture sensitivity & specificity of subject for particular task
Steep = High sensitivity
Area under curve (A') sometimes used as alt measure to d'
Problems w. Psychophysical Methods
Measuring thresholds using single interval detection task is affected by reporting bias
Two-interval tasks rely on memory component
Changes in perception over time (adaption), fatigue & boredom can be issues
Assumptions about shape of psychometric curve/underlying distributions can be wrong
Most methods need to find good starting parameter range