6. Signal Detection Theory and Threshold Determination.pptx
threshold, Weber's law, decision-making process, neural noise, hits and false alarm, internal noise and response, index of sensitivity, detectability, lax criteria, strict criteria, the ROC curve, methods of threshold determination
Learning objectives
• Tounderstand the background of ‘Signal Detection Theory’,
noise and noise plus signal.
• To understand the type of observations and information
acquisition criteria.
• Understand the stimulus detectability, criteria of decision
process, receiver operating characteristics (ROC).
• State methods of threshold determination
3.
Threshold and Weber’slaw?
• Threshold: The minimum amount of energy
required to detect a stimulus
• Weber’s Law: The just-noticeable difference is a
constant proportion of the magnitude of an initial
stimulus.
4.
Select a correctanswer.
• According to the Weber’s law, which statement is incorrect?
a. it is harder to tell the difference between two stimuli at higher levels of
intensity than at lower levels of intensity.
b. People don't perceive stimuli in terms of their absolute intensity but rather
their intensity relative to other stimuli.
c. The just-noticeable difference is a constant proportion of the magnitude of
an initial stimulus.
d. Two stimuli differing by the same percentage to be perceived as different.
5.
Background
• Human arenot perfect observers. In human, threshold
varies on repetition of measurement.
• Stimulus results in a neural activities. Signal must be
perceived as different than the background neural noise
• If this neural activity is sufficiently strong. The stimulus is
seen.
• In summary, there is no single, fixed value below which a
person never detects the stimulus and above which the
person always detects it.
6.
Signal Detection Theory
•The signal detection theory involves detection of the
stimulus (threshold) in the presence of some uncertainty as
a decision-making process.
This process is dependent on
• nature of the stimulus (intensity)
• sensitivity of a person to the stimulus
• cognitive factors
• Decision criteria
• Attention
• Internal neural noise
7.
Signal Detection Theory
•The level of neural noise fluctuates constantly.
• Neural Noise: Neurons are constantly sending information
to the brain, even when no stimuli are present.
• The brain must decide whether the neural activity reflects
• noise alone, or also a signal
• This makes the observer’s task to differentiate:
• A. The signal and noise combination
• B. The noise alone
8.
Summary
• The noiseis random and fluctuating
• The signal is constant
• The noise is always present, and the signal is superimposed
• The larger the signal, the easier it is for the observer to
detect
9.
Relationship between Hitsand False
Alarm
• In a threshold
detection
experiment,
there are two
possible
responses,
"Yes" or "No."
and two
different
possibilities for
the stimulus,
“present” or
“absent”.
10.
Observer Responses
• FalseAlarm
• An observer reports stimulus when stimulus is absent
• Correct Reject
• An observer does not report stimulus when stimulus is
absent
• Hit
• An observer reports stimulus when stimulus is present
• Miss
• An observer does not report stimulus when stimulus is
present
11.
Information acquisition criterion
Specificity= Correct rejection
(CR) / CR + False alarm
Sensitivity = Hit / Hit + Miss
•Sensitivity (true positive rate) is the probability of a positive test result, conditioned on the individual truly being
positive.
•Specificity (true negative rate) is the probability of a negative test result, conditioned on the individual truly
being negative.
Internal response andinternal noise
• External noise: environmental factors. e.g., smugs, room
light, temperature, noise, etc.
• Internal noise: background neural responses.
• Internal response:
• determines the one’s impression whether a stimulus is
present.
• Is the state of the mind is reflected by neural activity
14.
Relationship between Hitsand False Alarm
• There is always a trade-off between the number of Hits and
False Alarms
• When a person is very willing to say that the signal was present,
that individual will show more Hits, but will also have more False
Alarms.
• Index of sensitivity (detectability, d’) : determines the
sensitivity of an individual for any given pattern of Hits and
False Alarms.
15.
Detectability
• Definition: Thedifference
between the means of N
and N + S
• Detectability increases as
the distributions of N and N
+ S become further apart
• With a very large ‘d,’ there is
no uncertainty whether the
stimulus is present
• With a weak stimulus, the ‘d’
becomes much smaller
• d' = separation / spread
• This number, d', is an estimate of the
strength of the signal.
• its value does not depend upon the
criterion the subject is adopting,
• it is a true measure of the internal
response.
16.
Where does confusionoccur?
• Since the curves
overlap, the internal
response for a noise-
alone trial may exceed
the internal response
for a signal-plus-noise
trial. Vertical lines
correspond to the
criterion response
17.
Subject criterion and
Decisionprocess
• Lax Criterion
• Lax: Indicate a
stimulus even with a
great deal of
uncertainty (e.g., eye
screening).
• A Lax criterion results
in a substantial
number of false
positives, but very few
misses.
Lax Criteria (sensitivity)
High: Hits, False Positives
Low: Misses, Correct Rejects
18.
Subject Criterion and
Decisionprocess
• Strict Criterion
• Strict: Do not
indicate a stimulus
until they are
certainly present.
• A Strict criterion
results in fewer
hits, but a lower
number of false
positives. Strict Criterion (specific)
High: Misses, Correct Rejects
Low: Hits, False Positive
19.
The Receiver
Operating
Characteristic
• Capturesthe various alternatives
that are available to the examiner
in a single graph.
• ROC curves are plotted with the
false alarm rate on the horizontal
axis and the hit rate on the vertical
axis.
• For any reasonable choice of
criterion, the hit rate is always
larger than the false alarm rate, so
the ROC curve is bowed upward
values closer to 1 indicate the
reliable screening measure
whereas values at .50 indicate
the predictor is no better than
Varying the noise
•For stronger signals, the
probability of occurrence curve
for signal-plus-noise shifts right
and detection is easier.
• The spread of the curves:
Clearly, the signal is much more
discriminable when there is less
spread (less noise) in the
probability of occurrence
curves.
The separation between
the peaks is the same but
the second set of curves
are much thinner.
22.
When does criterionnot affect?
• d' = z(FA) - z(H);
• Where, FA = false alarm, H = hit
• d’ = 0
• Stimulus is so weak, no signal is
produced
• Regardless of criteria, the
proportion of hits will match the
proportion of false positives.
• d’ = infinity
• Stimulus is easily distinguished and
will always be seen by the observer
(No false positives)
d’ = 0
23.
How do wedetermine thresholds?
•Method of Ascending Limits
•Method of Descending Limits
•Staircase Method
•Method of Constant Stimuli
•Method of Adjustment
•Forced Choice Method
24.
Method of AscendingLimits
• Stimulus is initially presented below
threshold
• Stimulus is presented at increasingly
intense levels from presentation to
presentation until visible by observer
• Advantage:
• Relatively quick method
• Disadvantage:
• Participant Anticipation
• How to Avoid: Start each trial with
stimulus of a different intensity
25.
Method of DescendingLimits
• Stimulus initially presented
clearly visible and reduced
until no longer seen
• Example: Visual Acuity
• Disadvantage:
• Patient Anticipation
• How to Avoid: start each trial a
different level of visibility
26.
Staircase Method
• Acombination of Ascending and Descending method
• How Does It Work?
• Stimulus starts below threshold
• Presented in discrete steps of increasing visibility until observer reports
stimulus
• Visibility is reduced in discrete steps until stimulus can no longer be
detected
• Staircase is again reversed
• Threshold is defined after three or four reversals
• Advantage: Quick and Reliable
• Example: Frequently used in Visual Field Testing
Method of ConstantStimuli
• Stimulus is randomly varied from
presentation to presentation
• Large number of stimuli
presented at each level of
visibility
• Advantage:
• No patient anticipation how should
they respond
• Disadvantage:
• Time Consuming (not typically
used clinically)
29.
Method of Adjustment
•Participants adjust intensity
until the stimulus is barely
visible
• Advantage:
• Relatively quick
• Disadvantage:
• Patient criteria skews results
30.
Forced Choice Method
•Minimizes the role of individual’s
criterion
• Patient is forced to choose
between several alternative choices
(one contains the stimulus)
• A different number of choices can
be given:
• 2 Alternative Choice Method
• 4 Alternative Choice Method
• Typically results in lower thresholds
31.
Threshold Determination
• Threshold= Midway between
100% correct and ‘chance’
• Chance = percentage we expect
observer to guess correctly.
• 2 Alternative Choice Method
• ‘Chance’ performance = 50% correct
• Threshold = 75% correct
• 4 Alternative Choice Method
• ‘Chance’ Performance = 25% correct
• Threshold = 62.5% correct
32.
Summary
• Signal DetectionTheory (SDT) explains stimulus detection
amidst noise, influenced by stimulus intensity, observer
sensitivity, cognitive factors, and decision criteria.
• Observers make decisions resulting in outcomes: hit, miss,
false alarm, or correct rejection based on signal and noise
differentiation.
• Internal noise and fluctuating neural activity affect
detection; higher signal strength improves detectability (d’),
plotted using ROC curves to show the trade-off between hits
and false alarms.
33.
Summary
• Observer decisioncriteria impact outcomes: lax criteria
increase hits and false alarms, while strict criteria reduce
both; d’ measures sensitivity independent of the criterion.
• Threshold determination methods include
Ascending/Descending Limits, Staircase Method, Constant
Stimuli, Method of Adjustment, and Forced Choice Method
to establish detection thresholds.
• Threshold is defined as the midpoint between chance level
and perfect detection; for 2-alternative forced choice,
threshold is 75% correct (chance = 50%)
34.
Reference
1. Resources: VisualPerception, A Clinical Orientation Steven H.
Schwartz
2. Professor David Heeger. "Signal detection theory". Encyclopedia of
Psychology. FindArticles.com. 03 Jun, 2010.
Probability of chanceguessing
• p=(p'-C)/(1-C)
Where,
p=corrected probability
p'=raw probability
C=probability of chance success
(For 2AFC, C = 0.5, so p'= 0.75 leads to a p = 0.5.)
#5 They are complex biological system. Random neural noise is inherent within the visual system. This idea has led to the development of signal detection theory
During a psychophysical trial, the SAME stimulus strength will cause different answers from the same participant; what explains why our threshold detections are not discrete?
Signal detection theory
Noise detection theory
Stimulus discrimination theory
Judgemental bias
#6 In other words, a person will be able to detect more intense sounds or lights more easily than less intense stimuli. Further, a more sensitive person requires less stimulus intensity than a less sensitive person would. Finally, when a person is quite uncertain as to whether the stimulus was present, the individual will decide based on what kind of mistake in judgment is worse: to say that no stimulus was present when there actually was one or to say that there was a stimulus when, in reality, there was none.
An example from everyday life illustrates this point. Suppose a person is expecting an important visitor, someone that it would be unfortunate to miss. As time goes on, the person begins to "hear" the visitor and may open the door, only to find that nobody is there. This person is "detecting" a stimulus, or signal, that is not there because it would be worse to miss the person than to check to see if the individual is there, only to find that the visitor has not yet arrived.
#9 If a person participates in an experiment and receives one dollar for each Hit and there is no penalty for a False Alarm, then it is in the person's best interest to say that the stimulus was present whenever there is uncertainty. On the other hand, if the person loses two dollars for each False Alarm, then it is better for the observer to be cautious in saying that a stimulus occurred. This combination of rewards and penalties for correct and incorrect decisions is referred to as the Payoff Matrix. If the Payoff Matrix changes, then the person's pattern of responses will also change. This alteration in responses is called a criterion shift.
What is the Signal Detection Theory?
Decision making takes place in the presence of some uncertainty
A model that addresses the role of these factors in determining a threshold
It provides a precise language and graphic notation for analyzing decision making in the presence of uncertainty
#19 if the criterion is high, then both the false alarm rate and the hit rate will be very low. If we move the criterion lower, then the hit rate and the false alarm rate both increase.
A measure of goodness-of-fit is based on the simultaneous measure of sensitivity (True positive) and specificity (True negative) for all possible cutoff points.