model-averaged intercepts

[Mark W. Miller]

I am attempting to obtain model-averaged parameter estimates for all parameters in my best (lowest AICc) logistic regression model.

I have a set of 9 logistic regression models that include 0 to 4 independent variables. The best model has only one independent variable, x1, and an intercept.

I am following the cement example in Burnham and Anderson (1998:78-79, 145-151). From that, I can obtain the model-averaged slope estimate for my best model, but I am unsure how to obtain the model-averaged estimate of the intercept.

Do I average using the estimated intercepts from all 9 models; or only using the estimated intercepts from those models that include the variable x1? Someone suggested I use all 9 intercepts, but that I should first set the intercept equal to the average value of the dependent variable for all models that do not include x1.

The best model for the cement data in Burnham and Anderson (p. 80) was: 52.6 + 1.468(x1) + 0.662(x2). The model-averaged slope estimate for B1 = 1.4561; and for B2 = 0.6110. Could someone tell me the model-averaged estimate for the intercept in this model?

Thanks for any help or suggestion.