Task 2 Flashcards
What is a signal detection experiment?
- present a single (low-intensity) tone -> difficult to hear in some trials
- subject answers, whether a tne was heard or not
- based on signal detection theory
what are signal and noise distributions?
- a distribution that displays the probability distribution of the signal detection theory
- the probability of a given perceptual effect can be presented using probability distributions
- where there is less noise the probability distributions are more peaked + have less spread
What are hits/ false alarms/ correct rejections/ misses=?
hit: yes-signal
False alarm: yes-noise
Correct rejections: No-noise
Miss: No-signal
What is the liberal criterion?
N: most of the distribution falls to the right of the criterion
= high (p) of saying “yes”, when S+N is presented
= high (p) of a false alarm
S+N: entire distribution falls to the right of the criterion
= high probability of saying “yes” when (s+N) is presented
= high (p) of a hit
What is the conservative criterion?
N: none of the curves to the right of the criterion
= low (p) of false alarm
S+N: only a small portion on the right side
= low (p) of hit
How is signal strength relevant ?
What are payoffs?
- can influence ßopt
- no longer defined as minimising errors, but as minimising gains / losses
- increases in the denominator -> decrease in ßopt
- increase in numerator -> should lead to conservative responding
What is “d´”?
- discriminatory index
- indicates the subject´s sensitivity
- consists of the distance of the distribution’s peak
- influenced by signal intensity and spread of noise
- does not depend on adopted criterion but is assumed to be a true measure of internal response
- = 0 would mean guessing
What is Beta?
- the ratio of neural activity produced by signal and noise at the critical value (=Xc) (threshold)
- ß + Xc define the response bias/response criterion
- it indicates the ratio between the two curves
- Shifting Xc to the right -> ß>1 = fewer yes (fewer hits/ false alarms)
- Shifting Xc to the left -> ß<1 = more yes (more hits/ false alarms)
- if ß1-> P(h)= P(cr) and P(m) = P(fa)
What is the optimal Beta?
- ßopt
- where ß should be set
- can be calculated with SDT using the likelihood of observing a signal and the cost of benefits
What is sluggish beta?
in reality, humans do not adjust beta as much as they should for optimal outcomes
- they are less conservative than they should be if ideal beta is high
- they are less liberal than they should be if the ideal beta is low
What is an ROC curve?
- ROC portrays equivalence of sensitivity across changing levels of bias
- makes use of d´as a measure of sensitivity difficult -> it uses distance from the ROC curve from the chance axis (= varies as a function the criterion setting)
How is an ROC curve read?
- lower left = conservative
- upper right = risky
- positive diagonal in the graph: chance performance
How is an ROC curve formed?
- each signal detection condition generates one point on the ROC
- if signal strength and observer´s sensitivity stay constant -> changing beta from one condition to another will produce a curved set of points
what is the signal detection theory?
- precise analysis of decision making, under uncertainty
- signal: stimulus presented to the subject (= tone)
- noise: other stimuli in the environment (= external) or internal processes (= internal)
- displayed as a probability distribution
- the decision depends on the location of the subject criterion
