parameter likelihoods

[Kevin Hartke]

Is there a better way to calculate relative importance of variables when variables are unbalanced among the models? When there is unbalance among the variables, a variable that occurs in multiple models may have a greater likelihood than a variable that only occurs in the "best" model. A simple remedy would be to divide the sum of model weights by the number of models in which a variable occurs. I welcome any comments or suggestions? I realize that having balance among the variables in a model set is ideal but this seems to be a paradox because of the importance for an a priori selection of models.