L2 - SDT

Question 1

What is signal detection theory?

Answer 1

• Signal detection theory is a framework for understanding accuracy that makes the role of decision processes explicit
• Detection of a signal is said to occur whenever the output of the receiver exceeds a certain predetermined value – in the absence of any signal, the value can be exceeded by noise alone
  
  • Signal or noise question
• Most useful theory in psychology as can be applied to medicine EWT, lie detection etc…

Question 2

What is meant by criterion in a test and how can this impact testing results?

Answer 2

• Criterion is the threshold where above or below that value is correct/incorrect
• It can question how good the tests are – is it too favourable on one outcome etc.
• Since criterion is where there is no overlap in distribution then criterion result is almost always correct
• Overlapping distributions = Can’t be certain about whether patient falls into the cancer or no cancer group
• Evidence of errors
• If criterion too high = Cancer patients overlooked, if too low = system overwhelmed

Question 3

What are the assumptions of SDT?

Answer 3

1. The evidence can be represented by a single number
2. Evidence is subject to random variation
3. Choice of response is made by applying a decision criterion to the magnitude of evidence
   - What is the problem if the criterion is in the wrong place?
   - Misses and false alarms
   
   • If criterion in the middle, such as cancer example, you would expect equal chance of misses and false alarms but REALISTICALLY more false alarms than misses as there are fewer cancer patients actually in the population, so consider how common something is within the population
     - If criterion = Too Low → Causes issues such as setting off the smoke alarm for non-serious reasons
   • Sometimes have low criterion for attention – e.g. waiting for the phone to ring
   • Change threshold depending on the situation

Question 4

What is bias?

Answer 4

• Criterion position = Neutral → Unbiased
  
  • If criterion position away from equal likelihood = biased
• g. if criterion to the right in this example, then it is more likely for your brother to be blamed
• What about in application?
  
  • If more people on trial are innocent, need criterion to be higher to make sure less innocent are found guilty

Question 5

What is conservation vs liberal criterion?

Answer 5

• Conservative → Need strong evidence
• Liberal → Not much evidence needed

Question 6

How is SDT when applied to events in a simple perception experiment of motion?

Answer 6

Have to balance signal vs motion

• Liberal criterion → Call it motion in case of weak impression (many false alarms)
• Conservation criterion -> Don’t say motion except in case of strong impression (many misses)
  
  • Where our criterion can impact our hit rate
  • Is good or bad performance due to criterion or difference between distributions

Question 7

How do neutral, liberal and conservative criterion differ in terms of number of hits etc?

Answer 7

Neutral criterion has more hits and correct rejections

Literal criterion
- High rate of hits
- High rates of false alarms
- Very few misses

Conservative
- Few Hits
- High misses
- Few false alarms

Question 8

What does overall accuracy depend on?

Answer 8

• The criterion
• The sensitivity of the test (Sensitive = Little overlap, not sensitive = A lot of overlap)
• The mix of those who take the tests – e.g. how many who have the disease vs how many don’t

Question 9

What do we need to consider about response criterion studies?

Answer 9

• Someone may be a liberal responder and someone else a conservative responder = Threshold for stimuli may appear different but not in reality
• If measuring within the same person, don’t need to consider response criterion
• Signal detection solves this issue

Question 10

What is an ROC curve and what does it tell us about sensitivity?

Answer 10

• ROC = Receiver operation characteristic
  
  • Persons sensitivity to a stimulus is dictated by the shape of the ROC curve = can help us see if sensitivity is similar
• When a persons sensitivity is high = More bowed
• The distance between the neural & (s+n) probability distributions shape the ROC curve

Question 11

When do we use SDT?

Answer 11

When we are uncertain