I wish to compare three or more models using AICc. How do I compute their relative probabilities of being correct?

This method comes from page 75 of Burnham and Anderson¹.

1. Compare the models two at a time, each vs one standard or control model (it really doesn't matter which model you pick for this). Prism will calculate the difference in AICs scores for each pair of models. Be sure to track the sign properly, so the sign is negative when the standard model is worse, and positive when the standard model is better.
2. For each model, compute exp(-.5*delta), where delta is the difference in AICc scores, and exp() means e to the power. For the first (standard) model, delta is zero by definition, so this term equals 1.0. These values are proportional to the likelihood that the model is correct.
3. Add up those values. 
4. For each model, divide the value computed in step 2 by the total computed in step 3.
5. Check that the values computed in step 4 sum to 1.0. (This is just a check on calculations)
6. Interpret each value as the weight of evidence for that model. If you assume one of the models must be correct (not always a valid assumption), then that weight of evidence is also a probability -- the probability that that particular model is correct.

 

¹ KP Burnham and DR Anderson, Model Selection and Multimodel Inference, Second Edition, Springer-Verlag, 2002.