W2 Choice Flashcards
Quantitative models – why are they useful?
Specific, testable predictions – guides experimental design and tells us something useful even if the model itself is wrong. Assessment generates knowledge even if the model is wrong.
What are the two equations we focus on?
GML (Generalised Matching) CDM (Contingency Discriminability)
What is the general matching equation?
log(B₁/B₂) = a log(R₁/R₂) + log c
What is the Contingency-Discriminability model equation?
What are the parts of the general matching equation?
log(B₁/B₂) = choice between behaviour 1 or 2 a = sensitivity log(R₁/R₂) = How the reinforcers are distributed between those behaviours log c = bias
What happens if you get the response-reinforcer relation completely wrong?
R1 and R2 impact B1 and B2 equally. Cancels to log(B₁/B₂) = log c. No control by reinforcer. Just bias.
What are the parts of the Contingency-Discriminability model?
log(B₁/B₂) = choice between behaviour 1 or 2 dbr = behavior-reinforcer discriminability
What values can Dbr be?
1 < dbr < ∞ 1 means complete indifference (complete undermatching) ∞ means perfect discriminability (strict matching)
What is a in General Matching model?
a is the extent to which reinforcer ratio impacts choice. a = 1 change in reinforcer ratio produces same size change in choice a < 1 change in reinforcer ratio produces smaller size change in choice
How does the Contingency-Discriminability model compare to the General Matching Model?
General Matching may only fit between -1 to +1 log units, while CDM will fit for the whole thing.
What happens as choices more discriminable?
Greater discriminability increased sensitivity to relative reinforcement rates
What happens as it becomes harder to tell which response produced a reinforcer?
Choice between delayed reinforcers. Reinforcer ratio varied across components within a session (rather than across conditions, as in a standard matching experiment) Across separate phases (sets of conditions) varied how the delays were arranged. Changes in choice are ogival (round tapered end), not linear.
What happens in the Contingency-Discriminability model when R₂ = 0?
- Response ratios are not exclusive (GML has to predict exclusive choice)
- Response ratios are constant and independent of R1
If Discriminability matters, how we use it with our reinforcers?
If we want our reinforcers to be effective, we have to make discriminability high – specific praise, immediate consequences, etc
In the extremes, is choice a linear function of reinforcer distribution?
No, fits the CMD more than the GML. Changes in choice are ogival, not linear.
