There are no likelihoods in the survey statistics world, sorry. And there
are philosophical reasons as to why there are no likelihoods. First, there
is no way any particular model will fit an arbitrary finite population.
Second, the inference goal in survey statistics are population parameters,
the numbers characterizing a given finite population from which you took a
sample. Accordingly, the confidence probabilities and p-values are
probabilities of taking bad samples in which the CI does not cover the true
value, etc. When you combine the two concepts, what you end up with is
something like "Based on the sample information, I can estimate the misfit
in the population to be such and such, and out of gazillion possible
samples, 1.5% will have a test statistic greater than 3.5 observed in the
current sample" -- if any of these statements are informative at all. You
can estimate say the population R^2 in linear regression, or whatever fit
statistic that you would normally associate with a model for i.i.d. data,
but, to repeat, you know that your fit won't be great, since all models are
wrong (c) G. Box.
On Thu, Feb 11, 2010 at 8:54 AM, Kate Perper <[email protected]> wrote:
> Hello,
>
> I'm wondering how I can get fit statistics when running an multinomial
> logistic regression model while controlling for survey design. svy: mlogit
>
> When I run mlogit without svy, I can get log likelihood statistics, not so
> when svy is included.
>
> Thanks,
>
> Kate
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
