Comparing nested Poisson regressions for survey data

[Chiara Piazzo]

Dear all,

Since IC tests are not available for svy commands, is it possible to compare the two following regressions?

svy: tpoisson y x1 x2, ll(0)

svy: tpoisson y x1 x2 x3, ll(0)

Thank you!

Sincerely,

Chiara

[Steve Samuels]

Of course, you can just test that the coefficients of variables in the full model only are zero:

test x3

You could also try to calculate your own AICs with the log pseudo-likelihoods (SAS does this apparently), but these are not correct. See Lumley and Scott (2015).

Reference:
Lumley, T., & Scott, A. (2015). AIC and BIC for modeling with complex survey data. Journal of Survey Statistics and Methodology, 3(1), 1-18.
available here

[Chiara Piazzo]

Thank you, Steve Samuels,

Appreciate the solution and the reference. I was wondering if there are any other formal tests available? Thanks again!

Sincerely,

Chiara

[Steve Samuels]

I was wondering if there are any other formal tests available?

None that I know of.

[Phil Bromiley]

I'm not sure this is legit, but you could duplicate your dataset, and run the two models at once which would allow you to test for parameter equality across the equations. You should also be able to test the significance of the change in log likelihood between the two models.