RE: st: RE: Fixed Effects estimation with time-invariant variables

["Shi, Shishan (MU-Student)" <ss4x5@mail.missouri.edu>]

Hi Roman,
OK. Let me see if I follow your question. I first thought you had a Stata programming question, but now I think it is more of a modeling question.
In terms of whether to use Fixed Effect (FE) models or not, I think it depends on the assumptions. I suppose your data is on the company level. So before you do -xtreg, you will do -xtset companyid, if you want to do company fixed effect. If you believe the individual company characteristics affect IEO, and this individual company effect is also correlated with the error term, then I think FE models will give you more reliable estimates than OLS. 

I am not sure if that is what you are asking. If not, let me know.

Thanks,
Sybil
________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Roman Wörner [h0953997@wu.ac.at]
Sent: Saturday, January 12, 2013 3:46 AM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: RE: Fixed Effects estimation with time-invariant variables

Hi Sybil,

yes, StrategyA, StrategyB, and Scope are known. While the first two
change over time, the third does not. So I interact StrategyB and Scope
and regress StrategyA on StrategyB and the interaction of StrategyB and
Scope. The goal is to show that there is a positive relationship between
StrategyA and StrategyB for low values of Scope which becomes weaker (or
even negative) for high values of Scope (Scope can take on the values
0.2, 0.4, 0.6, 0.8 and 1). If I'd use a pooled OLS regression or a
random effects model things would be straight forward. What puzzles me
is if one could approach this question also with a fixed effects model
(to control for the individual effects).

What I actually do is running the following regression (IEO = StrategyA
and EEO = StrategyB):

xtreg IEO EEO c.EEO#c.Scope ....., fe vce(robust)

Also the results are as expected (correct sign; significant). However, I
don't know if the outcome is meaningful in a sense that the things I do
are "technically" correct.


----------------------------
                       (1)
                       IEO
----------------------------
EEO                 343.6**
                    (2.65)

EEOxScope          -415.2*
                   (-2.26)

...

_cons               184.8***
                   (11.53)
----------------------------
N                    1251
----------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001


Many thanks,

Roman
Am 12.01.2013 04:15, schrieb Shi, Shishan (MU-Student):
> Hi Roman,
> Correct me if I misunderstood your question. But it looks like your time-invariant variable 'Scope' is a known variable (I mean you know the value in the variable Scope for each record, correct?). Can you just create a new variable that is
>
> gen newvar = StrategyB * Scope
>
> and run OLS?
>
> Sybil
> ________________________________________
> From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Roman Wörner [h0953997@wu.ac.at]
> Sent: Friday, January 11, 2013 12:44 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: RE: Fixed Effects estimation with time-invariant variables
>
> Hi Dave,
>
> you are totally right! What I actually ment was "independept" variable.
> So one of my independent variables is time-invariant and the main
> question was if I could use this variable in a fixed effects regression
> if I interact it with another independent variable which changes over time.
>
> I am very sorry for the confusion (although I read the post five times
> before sending it, I didn't recognize the mistake)!
>
> Regards,
>
> Roman
>
> Am 11.01.2013 19:33, schrieb Jacobs, David:
>> I don't understand how you can interact anything with a dependent variable.  Do you mean the lag of such a variable used as an explanatory variable?  If so, you still cannot analyze time-invariant dependent variable with either -xtreg random-effects- or -xtreg, fe- routines.  The dependent variable must change.
>>
>> Unless your use of the term "dependent variable" was mistaken, I suggest you carefully read the beginning of the Xt manual and the chapter on -xtreg-.
>>
>> Dave Jacobs
>>
>> -----Original Message-----
>> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Roman Wörner
>> Sent: Friday, January 11, 2013 12:01 PM
>> To: statalist@hsphsun2.harvard.edu
>> Subject: st: Fixed Effects estimation with time-invariant variables
>>
>> Dear all,
>>
>> I am a doctoral student and rather new to STATA and statistics in general. I am thus struggling with a question I hope some of you are familiar with.
>>
>> My dataset is an unbalanced panel with N=328 and T=8. I plan to used fixed effects models to control for differences between the firms in my sample. I am aware that with fixed effects models one cannot use time-invariant dependent variables. Nevertheless, I've read that it is possible to include time-invariant dependent variables when you interact them with another (time-variant) regressor.
>>
>> Basically I have three variable of interest: two time-variant variables describing different firm strategies and a time-invariant variable describing the vertical scope (% of value chain steps of the industry the firm is active in; vertical scope takes on the values 0.2, 0.4, 0.6, 0.8, and 1) of the firm. I argue that the relationship of the two strategy-variables depends on the vertical scope of the firm - for focused firms the two strategies are complements, while they are substitutes for firms with a broad scope.
>>
>> I thus would run the following regression:
>>
>> StrategyA = b0 + b1*StrategyB + b2*StrategyBXScope + Controls
>>
>> I expect b1 to be positive and b2 to be negative. If that's the case I would interpret it that way, that an increase in the breadth of the scope reduces the complementarity between the two strategies (I would contrast combinations of StrategyA and StrategyB for different levels of Scope). I am wondering if this combination xtreg, fixed effects, time-variant and time-invariant variables are a valid design and allow for the conclusions I'd like to draw.
>>
>> I am very grateful for all comments and recommendations.
>>
>> Many thanks and best regards,
>>
>> Roman
>>
>> *
>> *   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/
>
>
>
> *
> *   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/