Signal Detection Theory

Question 1

SDT

Answer 1

-provides simple, mathematical model for quantifying the transmission of signals and general decision making processes in humans and animals

Question 2

Applications of SDT

Answer 2

• detection, discrimination, classification of stimuli
• memory (recognition)
• communication: optimal decisions in uncertain sensory contexts
• higher-order cognition: decision making, diagnoses etc

Question 3

Basic elements of SDT

Answer 3

1. Signal: the stimulus, object, target
2. Response: the action taken or decision made
3. Noise: the uncertainty factor, interference (intrinsic or extrinsic)
4. Response bias: bias from the decision maker, responder

Question 4

Detection/discrimination parameters

Answer 4

• accuracy in detection/discrimination measured by d’
• observers bias is called the criterion and is measured by c or B (beta)
• both can be represented with ROC (receiver operating characteristics) curves or plots
  
  • the curves show the relationship between hits and false alarms

Question 5

D’

Answer 5

D’=Z(H)-Z(FA)

Question 6

Errors

Answer 6

• type 1 error: false alarm (detecting signal that is not present)
• type 2 error: miss (not detecting signal this is present)

Question 7

Liberal decision making

Answer 7

• lenient
• say yes easily
• higher percentage of false alarms than misses
• p(hit)~0.90+
• p(FA)~0.70

Question 8

Conservative decision making

Answer 8

• says no easily
• strict
• higher percentage of misses than false alarms
• p(hit) ~ 0.50
• p(FA)~0.1

Question 9

If no errors (misses or FAs)

Answer 9

• d’ will be abnormally high (uninformative/misleading)
• we can apply a correction to approximate d’
• can transform the proportions of 0 to 1/(2N), and the proportions of 1 to 1 - 1/(2N), where N=number of trials that the proportions are based on
• OR add 0.05 to all data cells

Question 10

Criterion

Answer 10

C=-(Zh+Zfa)/2

• when c is positive, the subject is very conservative
• when c is negative, the subject is liberal

Question 11

D’ for detection vs discrimination

Answer 11

• detection: noise vs stimulus + noise
  
  • d’=Zh-Zfa
• discrimination: stimulus 1 vs stimulus 2
  
  • d’=(Zh-Zfa)/SQRT2

Question 12

Sensitivity

Answer 12

• About the hits and hit rate

- how often the decision maker can identify the stimulus accurately

Question 13

Specificity

Answer 13

• about the correct rejections
• correct rejection rate
• how well the decision maker can decide what the stimulus is NOT
• doesnt take into consideration errors (FAs and misses)

Question 14

Errors and specificity/sensitivity

Answer 14

• Type 1 error:
  
  • high type 1 = low specificity
  • low type 1 = high specificity
• type 2 error:
  
  • high type 2 = low sensitivity
  • low type 2 = high sensitivity

Question 15

Accuracy

Answer 15

-validity
-constant error
-

Question 16

Precision

Answer 16

• reliability

- variable error

Question 17

Repeatability

Answer 17

• conditions are kept constant
• same instruments
• same experimenters

Question 18

Reproducibility

Answer 18

• measurement process is kept the same
• different instruments
• different experimenters

Question 19

Basic parameters

Answer 19

-sensitivity: proportion of correct detentions = true positive rate
   Hits/(hits+misses)
-specificity: true negative rate
   CR/(CR+FA)
-accuracy: (hit+CR)/(hit+miss+FA+CR)
-precision: hits/(hits+FA)