Lecture 8 'Signal Detection Theory' Flashcards
What is d-prime (d’)?
The distance in standard deviations between the signal and the noise distributions.
How much a person is biased towards saying “yes the signal is probably there” (liberal/positive) or “no it’s probably not there” (conservative/negative) is called what?
response bias or criterion for responding - beta (B) or C
When can d-prime be negative?
Only if a person thinks that noise is “signal” and “signal” is just noise (reverse sensitivity)
When can Beta be negative?
Never - it is a ratio of heights of two distributions and ranges from 0 to +infinity
Criterion can be set so that it’s positive for what and negative for what?
Positive for positive response bias “I say yes a lot”
Negative for negative response bias “I say no a lot”
(Q47). In signal detection theory, what is the probability of hits if there are 15 hits, 25 false alarms, 15
correct rejections and 5 misses?
Number of hits / total number
15/60
.25
(Q48). In signal detection theory, if there are 10 hits, 5 false alarms, 25 correct rejections and 15 misses, then what kind of response bias is there?
More conservative or negative response bias
- In signal detection theory, if a participant says “yes” about as often when there is a signal vs. when
there is not a signal, then which is the most likely outcome for d’?
D-prime is close or equal to 0 (cannot distinguish signal from noise)
(Q 50) Identify the best method for changing response bias, in signal detection terms:
a. Change the payoffs for hits, misses, false alarms, and correct rejections.
b. Train people to discriminate signals better.
c. Introduce “false signals” to improve performance.
d. Take steps to remove the level of background noise from the signal.
c - introduce false signals to improve performance
(Q 51.) The findings of Tangen et al. (2011) on how people acquire expertise at fingerprint identification suggest the following:
People first develop a more liberal response bias and then develop a higher d-prime.
go from liberal positive response bias > learn to be more conservative (values of the profession) > start to learn discriminability
ROC curves are what?
Receiver Operating Characteristics
y-axis = hits
x-axis = false alarms
What are six ways to influence sensitivity/discrimination (d’)?
- feedback about accuracy
- slowing down rate of presentation
- amplify signal or make signal dynamic
- frequent rest breaks
- provide template for signal matching
- give experience seeing the signal
What are three ways to influence the response bias (beta)?
- insert ‘false signals’ to raise ‘yes’ responses
- provide incentives and exhortations
- change the payoffs for false alarms and misses