RE: st: xtlogit with (clustered individuals and households)

[Marco Albertini <marco.albertini2@unibo.it>]

Hola Alfonso, 

thank you very much for you quick answer. 

A feedback on that:

1. Apparently xtmelogit cannot be used together with mi est commands (yes, i am using a file with multiple imputations on a couple of income/wealth variables)
Then I selected just one of my imputed values and dropped the command mi est and I have used the following

xt set indviv waveid
xtmelogit depvar $indepvars || hhid: || individ:

but after almost 45 minutes stata is still stopped at "Refining starting values: " without any estimation produced (usually with the xtlogit commad i get the results in a couple of minutes). 
Is that normal?


2. Alternatively I have tried first to boostrap the errors by using 

xt set indviv waveid
xtlogit depvar $indepvar [w=weight], re vce(boot)

but stata keep returning the following error message
"weights not allowed"

this last thing is quite unexpected if I look at what is reported in the help xtlogit 
any suggestion?

ciao 
marco 



________________________________________
Da: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] per conto di Alfonso Sánchez-Peñalver [alfonso.statalist@gmail.com]
Inviato: lunedì 27 gennaio 2014 15.59
A: Stata List
Oggetto: Re: st: xtlogit with (clustered individuals and households)

Hola Marco,

several options here. You can always bootstrap the errors, and cluster them in the bootstrap. That normally works.You can also use the -melogit- or -meqrlogit- commands (Stata 13) (-xtmixed- previous versions) to model the household effects as random effects. Let’s assume that the other effects you are modeling are the individual effects across time, so the command for random effects on both could be

melogit depvar $indepvars || hhid: || individ:

If you still want to model the individual effects as fixed effects you can always include the dummy variables for the individuals (dropping one to avoid the perfect collinearity with the constant):

melogit depvar $indepvars $indivdummies || hhid:

Notice that melogit does allow for clustering in the variance, but I believe it is better to model the effects than to just allow for clustering in the variance.

I hope this is of use to you.

Best,

Alfonso

On Jan 27, 2014, at 9:24 AM, Marco Albertini <marco.albertini2@unibo.it> wrote:

> Dear Statalist members,
>
> I am using
>
> xtlogit depvar indepvar [w=weight], re/fe
>
> to estimate  re/fe logit models on panel data of individuals.
>
> The problem: individuals included in the sample are not independent observations. As a matter of fact, all individuals living within the same household (in each wave of the survey) have been interviewed. I would need to take this into account when calculating the standard errors.
>
> However xtlogit does not allow for  vce(cluster household_id)
>
> Any idea how can I solve this?
>
> regards
> marco
>
>
>
>
>
>
>
>
>
> *
> *   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/

*
*   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/