W3 Signal Detection

Question 1

How did Green and Swets (1966) discover Signal Detection Theory?

Answer 1

They recognized that any psychophysical task has both a sensory aspect (discriminability of the stimuli) and a decision aspect (bias, produced by nonsensory features of the task)

Question 2

What are the four things can happen on a signal-detection trial?

Answer 2

Hit, Miss. False alarm, Correct rejection.

Question 3

Using the example of a plane, explain how the payoffs and costs affect signal detection. What is the bias?

Answer 3

Radar operator watching for approaching aeroplanes – the cost of a false alarm is low (a bit of embarrassment), but the cost of a miss is very high (potential disaster). So predict a bias for “Yes”

Question 4

What is conditional discrimination?

Answer 4

When the correct choice response is conditional on the sample. It’s a way of asking animals “What do you think the sample stimulus was?” – if they peck left, they’re saying “I think it was red” – so is very like a signal-detection task

Question 5

What does the standard signal-detection matrix look like?

Answer 5

Question 6

What happens if S1 and S2 are indiscriminable (exactly the same as each other) ?

Answer 6

Choice between B1 and B2 won’t be affected by what the stimulus is.

Bw = By and Bx = Bz

Question 7

What happens if we make S1 and S2 more discriminable (more different from each other)?

Answer 7

Question 8

What happens if we reinforce B1 more than B2

Answer 8

Bias in favour of B1 over B2 independent of the stimuli.

Question 9

What is log d?

Answer 9

Question 10

What does log d measure?

Answer 10

log d genuinely measures the discriminability of the stimuli – it changed when the stimuli were made more different, but not when the reinforcers changed.

Question 11

What is the yellow and blue line here?

Answer 11

Question 12

What does Davison and Tustin’s model here simplify to if you [1st equation] - [2nd Equation]

Answer 12

Question 13

What is this equation called, and what does it estimate?

Answer 13

It is called a point estimate of discriminability – it estimates the value of log d that we would get if we did a complete variation of Rw/Rz like McCarthy and Davison (1980). It’s a quick but less accurate way of measuring log d.

It’s also called a stimulus function, because it measures the effect of the stimuli independent of the reinforcers and of any inherent bias, or an isobias function (equal bias)

Question 14

What happens if you add these equations together?

Answer 14

log d disappears

Question 15

What does the response bias measure?

Answer 15

Measures the effects of reinforcers and inherent bias, independent of the discriminability of the stimuliIt’s also called a bias function, or, in signal detection theory, an isosensitivity function.

See how the equation works – the two left-key responses W and Y are on the top line, and the right-key responses are on the bottom, so measures the tendency to respond left or right irrespective of the stimuli. This will be controlled by the reinforcers for left and right and by inherent bias, just as in the GML

Question 16

How does SPP (Signal presentation probability) affect bias?

Answer 16

‘If signal is highly probable, criterion for reporting “signal” becomes more lax’.

SPP does bias human SDT performance, also monkeys, rats and pigeons

Question 17

What does an uncontrolled reinforcer ratio mean?

Answer 17

Means that if, say, S1 is presented four times as often as S2 (SPP manipulation), the subject will get four times as many reinforcers for B1 responses than for B2 responses.

Question 18

McCarthy and Davison compared performance on two different controlled reinforcer ratio procedures.

One held SPP constant and varied the reinforcer ratio.

What were the results?

Answer 18

Reinforcer ratio did produce a biasbut SPP didn’t

Question 19

What are the Applications of the D&T model?

Answer 19

The model gives us a way to measure sensory(log d)and judgment (a) aspects of behaviour separately – if we just measure % correct we can’t do this.

Suppose we find that head-injured people are ‘worse’ at some task than uninjured controls (lower accuracy) – is this because they find it harder to detect the stimuli (sensory) or because they’re less sensitive to the payoffs and costs for correct and incorrect responses?

Question 20

What are the problems with the SDT (Signal Detection Theory) procedures?

Answer 20

In the usual SDT procedure you only get reinforcers for correct responses – maybe a bit artificial?

In the real world, the consequences for accurate responding are usually a bit fuzzier than that – sometimes there are reinforcers for errors and punishers for correct responses. (Sometimes, you successfully run the red light!)

The original signal detection theory can’t handle reinforcement for errors at all

Question 21

Draw a diagram of the signal detection/conditional discrimination/SMTS procedure.

Question 22

Write and explain the two basic equations of the Davison and Tustin model