Thank you for your response.
It is necessary for me that I estimate with weights, and also the
robust/cluster option. So I was wondering if there was any version of
the Hausman test that would allow me to do so.
Also, for the weights, the entire purpose of my test to compare the
weighted and unweighted models. It is a special property of a logit
model that the weighted and unweighted estimates should be the same
under the null hypothesis. Thus I cannot use the same weights for all
models in namelist and implement the Hausman test.
Is there another type of command besides suest and hasuman that could
help me implement my test?
Thanks again!
NB: While aweights are often used dealing with averaged data, it seems
to me they are also used for other weighted regressions, for example,
in dealing with choice-sampled data. In this case, each term in the
likelihood function will just be weighted by the square root of the
weighting variable, meaning that aweights are, it seems to me, the
appropriate weights, and other weights are not appropriate.
On Sun, Sep 21, 2008 at 12:04 AM, Maarten buis <[email protected]> wrote:
> The following two sentences from the help-file of -suest- seem to
> answer you questions:
>
> "If weights are applied, the same weights (type and values) should
> be applied to all models in namelist. The estimators should be
> estimated without vce(robust) or vce(cluster clustvar) options."
>
> -- Maarten
>
> BTW, Yulia explained to you here
> http://www.stata.com/statalist/archive/2008-07/msg01156.html that using
> -aweights- is not appropriate for -logit-. So you should another type
> of weight instead (probably pweights). Read section 11.1.6 of the
> User's Guide to see which type of weights is appropriate.
>
> --- L S <[email protected]> wrote:
>> I have a data-set with a binary dependent variable. A colleague
>> suggested doing a Hausman test to analyze whether a logit
>> specification was appropriate. I don't think I can just use the
>> Hausman command (both of the weighted and unweighted are inconsistent
>> if the null hypothesis is false)—when I do, the resulting chi-squared
>> statistic is sometimes negative. So instead I am trying to use the
>> suest command. I would like to use robust clustered standard errors.
>> The basic code I would like to run is (I use `version 9.0' so I can
>> use weights for logit):
>>
>> version 9.0
>> logit y x
>> est store unweighted
>> logit y x [aw=wt]
>> est store weighted
>>
>> suest unweighted weighted, robust cluster(country)
>> ttest, common
>>
>> I use cluster(country) with 'suest' command instead of the 'logit'
>> command as instructed on the suest help file. However, it appears
>> 'suest' does not like the robust option, returning the error:
>>
>> unweighted was estimated with a non-standard vce (Robust)
>>
>> Even when I leave out robust and cluster, STATA still doesn't seem to
>> like that I'm using suest comparing weighted and unweighted
>> estimates:
>>
>> . suest unweighted weighted, cluster(country)
>> inconsistent weighting types
>
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting address:
> Buitenveldertselaan 3 (Metropolitan), room N515
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>
> *
> * 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/
>
*
* 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/
