Digging into the R-code behind AIC for gaussian family models, I see:
AIC = nobs * (log(dev/nobs * 2 * pi) + 1) + 2 - sum(log(wt)) + 2 * p
dev = sum(wt * (y - mean(y))^2
For the unweighted case, this translates directly to -2LL with the
penalty number of parameters including both intercept and err
joe,
some procs in SAs calculates log likelihood differently than what it
is supposed to be. try using proc nlmixed and specifying the LL
explicitly.
in your case, I has stronger faith in R result instead of SAS result.
On 9/26/07, Joe Yarmus <[EMAIL PROTECTED]> wrote:
> In accordance with Venable
In accordance with Venables and Ripley, SAS documentation and other
sources AIC with sigma^2 unknown is calculated as:
AIC = -2LL + 2* #parameters = n log(RSS/n) + 2p
For the fitness data:
(http://support.sas.com/ctx/samples/index.jsp?sid=927), SAS gets an AIC
of 64.534 with model oxygen = runt
3 matches
Mail list logo
