Re: st: Can I control for time invariant industry effects and time invariant country effects at the same time?

[Christopher Parker <christopher.parker101@gmail.com>]

Dear David,

thank for your answer! I have an indicator that runs over a time
period of  32 years, 24 countries and 7 industries , so that I have
168 units observed over time. I still have some questions, and would
be thankful for your help:

1st: I asked this question because I was worried that the indstury
dummies would be cancelled out. Similiar to the fact that you cant
include time invariant variables(as the industry dummies) when using
fixed effects (country effects). Even though intuitivley I would say
that it is possible as the industries still vary within the country

2nd: When do I have to worry about including to many dummies? 31 time,
23 country, and 6 industry dummies seems like a lot to me.

3rd: Even this is just terminology: when I control for country effects
but my dependent variable varies over an additional dimension
(industries), is this still technically referred to as "Least Square
Dummy Variable Regression"? To restate, is this term used as soon as I
include dummy variables

Again I would be very thankful for an answer ,

Chris

2013/8/24 David Hoaglin <dchoaglin@gmail.com>:
> Chris,
>
> The usual approach would use indicator (or dummy) variables for the
> countries and the industries.  You can set up those categorical
> variables as factor variables in Stata.  (You didn't mention the
> numbers of countries and industries, or whether you may need to
> include interactions between country and industry.)
>
> If the joint distribution of countries and industries is reasonably
> balanced in your data, those variables should not cause problems with
> collinearity.  If they were perfectly balanced, your model would
> resemble a two-way analysis of covariance.
>
> Similarly, if the Xct have nontrivial variation across countries and
> across time, they should not have collinearity relations with the
> country indicators or the time indicators.  Collinearity among the X's
> may be a possibility.
>
> In any event, the various coefficients should be straightforward to
> interpret.  The country effects are adjusted for the contributions of
> time and the X's, the time effects are adjusted for the contributions
> of countries and the X's, and the coefficients of the X's are adjusted
> for the contributions of countries and time.  The fixed effects will
> be relative to (i.e., differences from) the reference country and the
> reference year.
>
> David Hoaglin
>
> On Fri, Aug 23, 2013 at 7:42 PM, Christopher Parker
> <christopher.parker101@gmail.com> wrote:
>> Dear Statalists,
>>
>> I want to do a regression of the following form:
>>
>> Ycit= Ac + Bi +Xct
>>
>> Ycit is my dependent variable, that varies across countries c.,
>> industries i, and time t. Ac is a country effect, Bi an industry
>> effect and Xct are my explanatory variables that vary across countries
>> and time. I want to estimate this with a  normal OLS estimator by
>> using dummies.(LSDV approach). To restate, I want include
>> timeinvariant industry and country dummies in an OLS-regression. Will
>> I have any collinearity issues with this approach, and will the
>> coeffecients for the fixed effects be interpretable?
>>
>> I would be very thankful for your help!
>>
>> Chris
> *
> *   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/