Re: st: AIC in fitstat

[Nick Cox <njcoxstata@gmail.com>]

You therefore need a meta-criterion to choose a criterion. That will
make your model choice completely objective. (I ran into a
misunderstanding the other day for not making it clear that I was
joking, so consider a smiley of your choice to be inserted here. On
the other hand, it is possible that I am trying to make a serious
point.)

On Wed, Oct 24, 2012 at 1:20 PM, JVerkuilen (Gmail)
<jvverkuilen@gmail.com> wrote:
> On Wed, Oct 24, 2012 at 8:08 AM, Lynn Lee <lynn09v@gmail.com> wrote:
>>
>> I use -fitstat- to check AIC or BIC of probit or logit model. But Stata11
>> lists several statistics: AIC, AIC*n, AIC used by Stata. Which statistics is
>> usually reported in probit or logit estimation?
>>
>> And AIC usually is good statistics to compare between models, but if I only
>> have one mode, do I still need to report AIC?
>
> They're all variants of the same basic statistics. As long as you look
> up the formula and define which you are using in your paper, you
> should be fine. I don't have the formulas for each one or I'd say
> more. I suspect that it's discussed in Long and Freese (2006) but my
> copy is on campus.
> http://www.stata.com/bookstore/regression-models-categorical-dependent-variables/
>
> There are good reasons to think that AIC is more sensitive (it allows
> more complex models and puts favor on avoiding Type II errors) and BIC
> is more specific (it penalizes more complex models heavily and puts
> favor on avoiding Type I errors) so depending on which decision
> criterion is more important for you, you'll have to choose.
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/