st: Re: question for fixed effect model/random effect model

["Rodrigo Alfaro" <ralfaro@bu.edu>]

Hi Lijun: xtlogit with fe option is the as clogit, which means conditional 
logit... this kind of procedure is common called fixed-effects logit but it 
is not a "true" fixed-effects, created (I guess) by McFadden (read carefully 
the help for clogit in the manual). You will loose the observations that do 
not change, for example Karen always attend the class then you cannot 
compute the conditional probability for Karen, you don't have information 
where she is at home. Also, time-invariant will be omitted because there is 
not within-group variance. For random effects, you need that the 
unobservable is uncorrelated with all x (time-variant and time-invariant)... 
and you will use the entire sample (in particular Karen will be in the 
sample). In other words, in random-effects you will use all the variables 
(after collinearity and all the sample). A easy reference is Chamberlian 
1980, but also you can start with this hand-written note 
http://people.bu.edu/vaguirre/courses/bu709/ec709_binary_1104.pdf. Finally, 
I don't understand what do you mean with two-wave, but if you think that 
some left-hand variable is endogenous, you have to move onto instruments 
estimators. Good luck, Rodrigo.




> Hi all,
>
> I come to three questions.
>
> First, I am trying to use ramdom effect model or fixed effect model to 
> predict
> a dummy variable, the high wage using education and job experience.
>
> The simple command look like
> xtlogit highwage educ jobexp, fe i(id)
>
> I got a note "note: multiple positive outcomes within groups encountered.
> note: 644 groups (1288 obs) dropped due to all positive or all negative 
> outcomes."
>
> What does this mean?
>
> Second, for variables that are constant overtime within individuals,
> do I still need to include them into my model? If I include such 
> variables,
> the fixed effect model will drop them while the random effect model with 
> consider them.
>
> Third, if my data is two-wave cross-sectional time series data,
> I wonder whether I still could not infer causality since dependent 
> variables
> and independent variables are measured at the same time. Is that right?
>
> Best
>
>
> --
>
> Lijun Song
> Department of Sociology
> Box 90088,Duke University
> Durham, NC 27708
> Phone (919) 660-5604
> Cell  (919) 724-3811
> FAX (919) 660-5623
> lsong@soc.duke.edu
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/