[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: calculating AIC for log-logistic model
Shige Song <firstname.lastname@example.org> asks:
> In page 231 of the book by Cleves et al. (2002) the formula for calculating
> AIC for parametric hazard model is given by:
> AIC = -2lnL + 2(k+c)
> where k is the number of model covariates and c is the number of
> model-specific distributional parameters. My questions is: how to count the
> number of k if I want to include the same set of covariates in more than one
> distributional parameters? For example, I am estimating a log-logistic
> hazard model; instead of intrducing covariates to the scale parameter only,
> I want to introduce the same set of covariates to both the scale parameter
> and the shape parameter. In this case, should I count the number of
> covariates (k) twice, when caluclating the AIC? Thanks a lot!
In short, you count them twice. Think in terms of how many parameters you are
estimating, rather than how many distinct covariates you are using.
In a standard log-logistic model, you have one shape parameter, gamma, and so
you set c=1 when calculating AIC. If you choose to model gamma (or ln(gamma)
as is the case) on a set of covariates, c changes from 1 (for the constant
term that is always there) to 1 + how many covariates you are modelling
ln(gamma) on. So, it is really c that increases while k stays the same.
Note however that k is just the number of covariates in the main equation, and
not 1 + this number. This is merely by convention, and we chose not to count
the constant term in the main equation.
* For searches and help try: