different results while using cluster and robust in poisson pseudo maximum likelihood regression

[Sonia Chen]

Dear all,

I just meet an issue while doing the Poisson pseudo maximum likelihood (pml) regression.

My model has six fixed effect.

I choose to use vce(robust) and cluster(varlists) for the regression respectively, but get different results.

firstly, for using cluster(varlists), I argue about four variables that should be clustered. But while I try different combination among those four, the regression results are insignificant, and the standard error are not that different when using different combination. Then I change my narrow previous six fixed effects down to three main fixed effect, the results do not change (coefficients, significance and s.e.).

Next I use vce(robust) instead, the results turn out to be pretty significant(p>|z|=0.000), and the standard error are much smaller than using cluster.

My question is:

1: why there are a such different results on significance and standard error values between using cluster and robust? As I read before from wooldridge(2013) and stock and Watson (2008), specifying Vce(robust) and Vce(cluster(varlists)) should be equivalent while using panel data.

2: why using cluster and robust leads a big difference in result significance? I know that using robust-clustered standard error might slightly affect the regression results to be less significant, but why this large effect in here?

Thank you!

[Phil Bromiley]

You didn't get a quick answer. Please follow the FAQ on asking questions.

For example, you don't tell us where you found this pml estimator. I easily find ppml what seems to do something similar, but for your kind of problem understanding exactly what the documentation says is essential to helping you.