I seem to recall that one can compare AIC scores across all Huggins closed population data types, even those with mixtures or mis-id parameterizations. However, I want to verify this and can't remember or find where I read/heard this. Can someone verify? I would like to avoid having to worry about constraining the "w/ het data type" design matrix to reflect non-mixture models by first, building a set of non-mixture models with Huggins closed data type, second, change data type to Huggins w/ het and build mixture models, and finally, compare all models with AIC. In addition, would there be an problem with model averaging real parameter estimates with this approach?
Thanks
Jared