Signal Detection Theory (Week 1) Flashcards
White pixel detection example
No way to stop participants from lying- some may say they saw something on every slide even if they didn’t to get a perfect score.
To get rid of bias you need to add “catch” trials- trials where there is no white spots shown to see if people lie and say they saw something when there wasn’t one.
Signal detection theory
Motivation of bias for action changes by context.
Signal detection separates motivation from perceptual thresholds for detection.
This is done by analysing frequency of four response types:
- hits (say yes when present).
- misses (say no when present).
- false alarms (say yes when stimulus absent).
- correct rejections (say no when absent).
Similar to hypothesis testing
Null hypothesis is false - reject null = correct decision.
Null hypothesis is true - reject null = type 1 error (false alarm).
Null is false - fail to reject null = type 2 error (miss).
Null is true - fail to reject null = correct decision (correct rejection).
Criterion can be moved around.
D-prime = 1 - sensitivity to separate two distributions.
Hit rates and false alarms drop when bias is moved.
If criterion is far left- very high hits and very high FAs.
As it moves over to the right, hits decreases and the false alarms also drop.
D-prime- measure of sensitivity to separate noise from signal
d’ = separation/spread.
d’ = Z(hit rate) - Z(false alarm rate).
if d’ is bigger this means there is little overlap of the normal distributions.
Standard correction
One practical problem in calculating d’ occurs when a person detects all signals or doesn’t make any false alarms.
To correct this, assume that person would have missed half a signal and would have made half a false alarm.
If a person were to do more trials they would eventually make an error so the correction assumes:
- 40 signal ‘present’ trials, hit rate would be adjusted to 39.5/40 = .9875
- 40 signal ‘absent’ trials, FA rate would be adjusted to 0.5/40 = .0125
