multinomial logit: suest with pweights and clustering

[Kerstin Schmidt]

Dear All,

Aiming at finding the determinants for different product choices, I am estimating a multinomial logit model and want to check whether the IIA assumption is violated. However, having survey data (pweights) and clustering at the same time doesn't allow me to run the suest command.
Background: My data are survey data (with one-stage clustered design with stratification):

svyset hhid [pweight=pweight]

This way, it shouldnt make any difference if using the prefix "svy" before the estimation command or including the pweights in the estimation command, right?

svy: mlogit Decision `x1', b(1) = mlogit Decision `x1' [pweight=pweight], b(1)

But let's come to my primary problem: Having four different unordered categories for my dep var, I run:

svy: mlogit Decision `x1', b(1)
estimates store m1

svy: mlogit Decision `x1' if Decision != 3 & Decision!=4, b(1)
estimates store m2

svy: mlogit Decision `x1' if Decision != 2 & Decision!=4, b(1)
estimates store m3

svy: mlogit Decision `x1' if Decision != 2 & Decision!=3, b(1)
estimates store m4

suest m*, vce(cluster hhid) noomitted
option cluster() is not allowed with svy results
r(198);

When using pweights in the estimation command instead of the prefix svy, the error message tells me to use iweights instead.

Why doesnt suest allow pweights/svy and clustering? What can I do instead to check the IIA assumption?

Thanks in advance!

[Kerstin Schmidt]

Just to add on this: Simply using the command written bei Long and Freese also doesnt help:

mlogit Decision `x1' [pweight=pweight], b(1) vce(cluster hhid) nolog
mlogtest, iia

Hausman tests of IIA assumption (N=963)

  Ho: Odds(Outcome-J vs Outcome-K) are independent of other alternatives
too many weights
r(198);

How can I check for the IIA assumption applying a pooled MNL and having pweights?

[Flor Carosa]

Hi, Kerstin! Have you by any chance found the solution to this problem? I've also been trying to use mlogtest to test iia and combinations, but received the "too many weights" error. Any leads would be really appreciated!

[Kerstin Schmidt]

No - sorry!