Xtpoisson and xtnbreg questions

[Ioannis Siskos]

Dear Stata community,
After reading many topics around the challenges I am facing, I still have some questions, the answers to which are not clear to me.

I have count panel data. My outcome is number openings of plants for a number of countries and a number of years (balanced). My predictors are several macroeconomic indicators of the countries.
I came to the conclusion that I am going to use either xtpoisson or xtnbreg
a) in order to determine which one is better, is it enough to check the alpha parameter at xtnbreg (given that model specification is the same) and if alpha is 0 or very small it is safe to go for xtpoisson? Is this the test for over-dispersion?
b) does the xtpoisson - fe specification suffer from the same problem as xtnbreg -fe? I am talking about the problem described here: http://statisticalhorizons.com/fe-nbreg
c) do I need to check for multicollinearity? I know that multicollinearity is a problem for linear models, but I also read that poisson and negative binomial belog to generalised linear models family, so I am bit confused whether multicollinearity can be considered as a problem
d) do I need to check for autocorrelation?
e) in my specification I use dummies to control for years and countries (i.e i.year and i.country). Is there any rule against or for doing this in relation to xt and -fe. What I mean is whether the fact that I use a panel command (xt) already controls for year (i.e no need to include i.year) or -fe controls for the panel variable (i.e no need to include i.country).
Thank you in advance for your response.

[Joao Santos Silva]

Ioannis,

a) Just for with xtpoisson fe because it is much more robust (unless you want to compute probabilities of certain events).
b) No.
c) There is not need to check for multicollinearity; you know it is there!
d) Just use clustered standard errors.
e) If you use fe, there is no need to include i.country.

Best wishes,

Joao

[Ioannis Siskos]

Hi Joao,
Many thanks for covering everything!

[Onur Taskin]

Dear Professor Santos Silva,

If you allow, I would have a question for the subsections c) for my clarification:
Since you said, that there is no need to check for collinearity, because "you know it is there".
I feal like I have missed an important point.
Could you maybe clarify why it is "implied", that there is collinearity or suggest a source for me, to read and understand it.

Thank you for your time and and suggestions in advance!

Best wishes,
Onur

[Joao Santos Silva]

Dear Onur,

About collinearity, there is no better source than Chapter 23 in this book.

Best wishes

Joao

[Onur Taskin]

Dear Professor Santos Silva,

Thank you for the recommendation. Also Professor Schechter suggested to have a look into it.
I just borrowed a printed version

Thank you for your time!

Best wishes,
Onur