Multi-way clustering with poisson regression

[Alessandro Sovera]

Good afternoon,
I am trying to run a multi-level Poisson regression (3-level) and I am trying to understand how to implement a multi-way clustering. I've found there exist commands for OLS and non-linear models but I've not found any for count data. Is there any command to deal with this problem or should I use some specific procedures?
Thanks

Alessandro

[Carlo Lazzaro]

Alessandro:
do you mean a hierarchical regression model (-mixed-)?
I would take a look at -egen-, -group- function, to create a multi-way cluster.

[Alessandro Sovera]

I mean I have variables on three levels (country, cohort and individual) and I've found that a way to do multi-level analysis is by clustering for the three levels at once. But the problem is that vce(cluster clustervar) allows for only one variable at a time.

[Carlo Lazzaro]

Alesandro:
set aside any nesting issue, you're correct about -vce(cluster clustervar)- machinery.
You may want to try something along the following dope example:

use "http://www.stata-press.com/data/r14/nlswork.dta", clear
egen A=group(idcode c_city south)
probit c_city i.race, cluster( A)

[Alessandro Sovera]

Alesandro:
set aside any nesting issue, you're correct about -vce(cluster clustervar)- machinery.
You may want to try something along the following dope example:

use "http://www.stata-press.com/data/r14/nlswork.dta", clear
egen A=group(idcode c_city south)
probit c_city i.race, cluster( A)

It works. Thank you very much!

[Carlo Lazzaro]

Alessandro:
being Italian (and graduated from the same university, unfortunately for me too many years ago) I should have not mistyped your given name.
Sorry for this and happy that what above helped you out.

[Simon Heß]

just for the sake of correctness I want to point out that I'm fairly sure that this does not what is usually meant by "multi-way clustering". Your code simply creates a unique variable IDing individuals and with this clusters at the individual level.