[Carlo Lazzaro]

David:
no, unfortunately.
Sorry.

[Joao Santos Silva]

Dear David,

Let me start by saying that using Poisson regression is a very wise choice ;-)

I am not sure if I fully understand your problem, but let me make some comments:

a) It may not make any difference, but your iv1 probably has very large values. I would divide it by 10000 and see if it makes a difference; sometimes Stata is sensitive to these things.

b) Your sample is far too small for the number of parameters you are estimating. I would say that you should not estimate more than 16 parameters with a sample of this size.

c) The small sample may be responsible for the instability that you find. Remember that, as Arthur Goldberger would say, multicollinearity is simply micronumerosity.

Best of luck,

Joao

[David Lu]

Dear David,

Let me start by saying that using Poisson regression is a very wise choice ;-)

I am not sure if I fully understand your problem, but let me make some comments:

a) It may not make any difference, but your iv1 probably has very large values. I would divide it by 10000 and see if it makes a difference; sometimes Stata is sensitive to these things.

b) Your sample is far too small for the number of parameters you are estimating. I would say that you should not estimate more than 16 parameters with a sample of this size.

c) The small sample may be responsible for the instability that you find. Remember that, as Arthur Goldberger would say, multicollinearity is simply micronumerosity.

Best of luck,

Joao

Dear Joao,

Thank you for your helpful suggestions. You're right and my iv1 has very large values. It does work and it made a difference after dividing it by 10000.

Thanks again,
David