Re: st: panel data regression question (Stata and statistical question)

[David Jacobs <jacobs.184@sociology.osu.edu>]

Any fixed-effects routine will drop any explanatory variables that 
remain constant within all cases.  That isn't a characteristic of 
Stata; it is a characteristic of that estimator.

One work around that may possibly (?) be theoretically applicable is 
to interact the unchanging explanatory variable with another 
explanatory variable that does change.  The resulting product will 
change, so the coefficient on this interaction term can be estimated 
in a fixed-effects model.

But, of course, using this approach means that the uninteracted 
variable that does change must be included in the model as well.  The 
other main effect captured by the unchanging component of this 
product term need not (and in fact could not) be included in the 
model.  This work around is not always applicable because theoretical 
considerations rule it out or statistical difficulties such 
collinearity (sp?) make it impossible.

D Jacobs

At 01:47 PM 3/21/2007, you wrote:

> Hi,
>
> I am new to Stata as I largely use SPSS.
>
> I will explain my problem:
> I have 2 years of data to do with the number of disclosure items 
> (TOTALITEM) a company discloses on
> its own website and the number of items that are disclosed on the 
> government website for that
> company.
>
> I have a number of explanatory variables, 2 of which are dummy 
> variables like Industry (Computing
> or Non-Computing) and Location (UK or not). Also, proportion of 
> foreign shareholders (PROPFOR) which
> does not vary over the 2 years.
> I have some continuous explanatory variables which do vary over the 
> 2 years, ie LOGSIZE (company's
> size), EQUITY.
>
> Using Stata, I have been looking at Fixed, Between and Random 
> Effects models using Panel Data.
>
> I have gone to Statistics Time Series Setup and Utilities and 
> declared dataset to be time series
> and the time is YEAR (2005 and 2006) and Panel ID is Subject so my 
> data looks like:
>
> e.g.
> Subject  YEAR     TOTALITEM Industry LOGSIZE
> 1            2005      20                  1           18
> 1            2006      30                  1           20
> 2            2005      50                  0           25
> 2            2006      60                  0           30
> etc.
>
> So to choose between Fixed effects or Random effects, I type the command:
>
> 1) xtreg totalitem propfor industry location logsize,fe
> 2) estimates store fixed
> 3) xtreg totalitem propfor industry location logsize,re
> 4)  estimates store random
> 5)  hausman fixed random
>
> 1) the fixed effects model in the output shows the coefficients for 
> the independent variables which
> vary across time, ie LOGSIZE and PROPFOR but those that do not vary 
> are dropped.
> 2) the random effects model shows coefficients for all independent 
> variables in the model.
> 3) hausman indicates no insignificant p-value
> Therefore, this indicates I should use fixed effects model instead 
> of random. My question is how do
> I now specify a fixed effects model in Stata which shows me all the 
> coefficients of all independent
> variables as running the command
> xtreg totalitem propfor industry location logsize,fe
> will drop those that do not vary over 2 years.
>
> Sorry, but I am only a beginner in Stata.
>
> Thanks,
> Paul
>
> ==================
> Paul McGeoghan,
> Application support specialist (Statistics and Databases),
> University Infrastructure Group (UIG),
> Information Services,
> Cardiff University.
> Tel. 02920 (875035).
> *
> *   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/