Rescorla-Wagner Calculator for Equilibria
This applet allows you to quickly compute equilibria of the Rescorla-Wagner
model for a particular probability distribution. To use the applet, you:
- Click on the button below to start.
- Enter the number of variables (not including the context or the effect) in
your system, and press "Continue"
- Enter the names of the variables and the effect, and the values for
beta when the outcome is present, beta when it is absent, and lambda (the
maximum associative strength supported by the effect)
- Note: the betas and lambda must all be greater than 0.0
- Enter the probabilities to complete the matrix. The cells start with
descriptions of the probabilities that belong in them. The column and row names
provide an additional guide for entering the probabilities: you need to enter
P(ColumnName | RowName). So, for example, the cell in the third column,
second row, should contain P(Variable2 | Variable1). The cells that are
already filled in (with 1.0) cannot be changed.
- If you have differing values for the two beta parameters (and you have
more than one cue variable), you will also need to input the probabilities
of certain conjunctions. Again, the cells start with descriptions of the
probabilities for those cells (and the column and row names also provide a
way of determining which probabilities go in which cells).
- When you have finished filling in the matrix, press "Compute R-W
Equilibrium", and a window will display with the unique equilibrium (if there
is one), or else tell you if there are infinitely many equilibria.
Note: all parameters and probabilities can be entered either as
decimals (e.g., 0.5) or as fractions (e.g., 2/3).
This applet does not allow you to set the alpha (salience) parameters for
the variables. If there is a unique equilibrium, then the alphas are
irrelevant. If there are infinitely many equilibria, then although the alphas
determine the unique equilibrium, I do not know, in general, how to take the
alphas into account. (If someone else can figure out the general solution, I will
gladly incorporate those formulae into this applet.)
The derivations of the formulae used in this program can be found in my
paper, "Equilibria of the Rescorla-Wagner Model," currently submitted to
Journal of Mathematical Psychology.
Any comments/errors/suggestions are quite welcome. They should be sent to
david@danks.org (David Danks).