Hello All,
I'm using a panel dataset to investigate the effect of R&D
collaboration diversity on innovation performance at the firm level.
The nature of my data is such that there is not that much variation
over time with respect to the number and type of one company's set of
partners, while there is substantial variation among different
companies. In other words, the composition of one firm's partners
usually doesn't change considerably over time, while comparing
different companies with each other reveals considerable differences.
I'm using Stata to run panel data regression models. Depending on the
type of my dependent variable (censored or not), I use -xtreg- or
-xttobit-. Here it comes the problem:
1) FE, RE, or BE?
When using -xtreg- I can go for FE, RE, or BE (or is there any other
versions?). As expected, the FE results in very poor, disappointing
and unexpected results, with many variables dropped from the
regression, or showing insignificant estimates. I believe, this is
mainly due to the nature of the data explained above. Am I right?
Alternatively, RE and BE result in better and more understandable
results. However, when I run the Hausman test, two things could
happen: either a) the null hypothesis is usually rejected, stating
that FE is the preferred method and that RE results in biased
estimations, or b) I get an error stating that the fitted model on the
data fails to meet the asymptotic assumptions of the Hausman test.
***What should I do? What is the valid approach to pursue? How should
I justify using RE or BE? Is there any alternative tests or methods
that can be used? What specific conditions should I check (and how?)
to be sure about using RE for my estimations?
2) Robust standard errors?
There is a lot of heterogeneity in my data with respect to
firm-specific or industry-specific characteristics. The results show
that heteroskedasticity is inevitable. To correct for that, I can use
the "robust" option when using -xtreg-. Am I right? But how about when
using -xttobit- as there is no "robust" option available for
-xttobit-?
***What would you suggest? How would you correct for
heteroskedasticity? Is there any other important characteristics that
I need to check before I can be sure about the validity and
reliability of my results? What pre- or post-tests do you suggest?
3) SYS-GMM method?
I know that SYS-GMM is a good way to deal with the problems of
omitted variables bias and simultaneity (endogenous variables).
***How can I successfully implement this method in Stata? Is there any
alternatives that you would suggest? In general, how would you correct
for simultaneity problem, if you don't have access to good
instruments?
I know I asked too much but I decided to post my questions all in one
email rather than bothering you with multiples. I hope you have time
and willingness to support me and in any case, I thank you for reading
and possibly responding to this post. Thank you so much all.
Good luck All,
Fardad
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
