TASK 2: SIGNAL DETECTION
SIGNAL & NOISE DISTRIBUTIONS
SIGNAL DETECTION PARADIGM
If there are 2 discrete states of the world (signal &
noise) that cannot easily be discriminated
Signal must be detected by human operator
Two response categories (yes & no) 2 states in the
world + 2 response categories 2 x 2 matrix
Hits – signal is present & subject says “yes”
Misses – signal is present & subject says “no”
False alarms – signal is absent & subject says “yes”
Correct rejections – signal is absent & subject says “no”
Perfect answers usually do not occur there is always data in all 4 cells
Signal Stimulus presented to the subject, usually very faint
Noise All other stimuli in the environment, might be mistaken for the signal
Always present
SIGNAL DETECTION THEORY (SDT)
number of occurences per cell
Values are expressed as probabilities:
total number of occurences∈a column
Assumes 2 stages of information processing
Sensory evidence is aggregated concerning the presence / absence of the signal
A decision is made about whether this evidence indicates a signal or not
External stimuli generate neural activity in the brain
There is more sensory / neural evidence in the brain when a signal is present than
when it is absent
Neural evidence increases with stimulus intensity
Evidence variable X
Enough neural activity, X exceeds critical threshold X C operator decides yes
XC = criterion value chosen by the operator – vertical line
All X values to the right of XC (X > XC) response: YES
Too little neural activity operator decides no
Noise = average activity when no signals are present, fluctuates
External noise = from the environment
, Internal noise = neural responses are noisy, even if stimulus is the same on
each trial
Total area under the curve = 1
P ( H )+ P ( M )=1
P ( FA ) + P (CR ) =1
X varies considerably even without stimuli being present because of random
variations in environment & operator’s own baseline level of neural firing
Smaller difference in intensity between signals & noise bigger error probabilities,
because the amount of variation in X resulting from randomness increases relative to
the amount of energy in the signal
“Perceptual effect” the subject experiences on each trial can be seen on the x-axis
Probability distributions tell us what the chances are that a given perception of the
stimulus is due to (N) or to (S + N)
CRITERION SETTINGS & SIGNAL STRENGTH
CRITERION SETTING
2 main components to decision-making-process: stimulus
strength & criterion
Subject needs to pick a criterion location along
internal response axis
Internal response greater than criterion
response: “yes”
Internal response less than criterion response:
“no”
Can adjust the kind of errors they make by
manipulating their criterion
Liberal / risky criterion setting – more prone to saying “yes”, many hits but also many false
alarms
XC places to the left
Low criterion setting
Conservative criterion setting – more prone to saying “no”, making few false alarms but
also many misses
XC placed to the right
High criterion setitng
Behaviour is determined by placing decision criterions XC
STIMULUS STRENGTH
Affects probability density functions in obvious way
Strong signal will shift S + N curve to the right internal response strength is
stronger
SIGNAL & NOISE DISTRIBUTIONS
SIGNAL DETECTION PARADIGM
If there are 2 discrete states of the world (signal &
noise) that cannot easily be discriminated
Signal must be detected by human operator
Two response categories (yes & no) 2 states in the
world + 2 response categories 2 x 2 matrix
Hits – signal is present & subject says “yes”
Misses – signal is present & subject says “no”
False alarms – signal is absent & subject says “yes”
Correct rejections – signal is absent & subject says “no”
Perfect answers usually do not occur there is always data in all 4 cells
Signal Stimulus presented to the subject, usually very faint
Noise All other stimuli in the environment, might be mistaken for the signal
Always present
SIGNAL DETECTION THEORY (SDT)
number of occurences per cell
Values are expressed as probabilities:
total number of occurences∈a column
Assumes 2 stages of information processing
Sensory evidence is aggregated concerning the presence / absence of the signal
A decision is made about whether this evidence indicates a signal or not
External stimuli generate neural activity in the brain
There is more sensory / neural evidence in the brain when a signal is present than
when it is absent
Neural evidence increases with stimulus intensity
Evidence variable X
Enough neural activity, X exceeds critical threshold X C operator decides yes
XC = criterion value chosen by the operator – vertical line
All X values to the right of XC (X > XC) response: YES
Too little neural activity operator decides no
Noise = average activity when no signals are present, fluctuates
External noise = from the environment
, Internal noise = neural responses are noisy, even if stimulus is the same on
each trial
Total area under the curve = 1
P ( H )+ P ( M )=1
P ( FA ) + P (CR ) =1
X varies considerably even without stimuli being present because of random
variations in environment & operator’s own baseline level of neural firing
Smaller difference in intensity between signals & noise bigger error probabilities,
because the amount of variation in X resulting from randomness increases relative to
the amount of energy in the signal
“Perceptual effect” the subject experiences on each trial can be seen on the x-axis
Probability distributions tell us what the chances are that a given perception of the
stimulus is due to (N) or to (S + N)
CRITERION SETTINGS & SIGNAL STRENGTH
CRITERION SETTING
2 main components to decision-making-process: stimulus
strength & criterion
Subject needs to pick a criterion location along
internal response axis
Internal response greater than criterion
response: “yes”
Internal response less than criterion response:
“no”
Can adjust the kind of errors they make by
manipulating their criterion
Liberal / risky criterion setting – more prone to saying “yes”, many hits but also many false
alarms
XC places to the left
Low criterion setting
Conservative criterion setting – more prone to saying “no”, making few false alarms but
also many misses
XC placed to the right
High criterion setitng
Behaviour is determined by placing decision criterions XC
STIMULUS STRENGTH
Affects probability density functions in obvious way
Strong signal will shift S + N curve to the right internal response strength is
stronger
