Multilevel Model (Individuals Within Firms) -- How to Account for Multiple Membership?

[Andrew Lee]

Hi there.

I have a cross-sectional dataset of individuals nested within firms. Some individuals appear in only one firm, while others appear in multiple i.e., this is a case of 'multiple membership'.

I'm not seeing anything in the documentation on MLM STATA commands like "mixed" / "xtmixed" that specifically address a multiple membership situation. There's information on 'crossed' data, but my understanding is that such data situations, while related, are not equivalent to multiple membership ones. .

Posts like this, however, lead me to believe that some strategy exists in STATA for setting up MLM coding for multiple membership data by treating it as if it were crossed data. I'm not partial to any specific STATA command, and it's no trouble to restructure my data if need be. And, supposing any 'weighting' determinations were needed for such multiple membership cases, I'd want to impose equal weight across a given individual's multi-firm data. (And, again, this is cross-sectional / time-invariant data).

Given my circumstances, is there a STATA-based route forward? Many thanks for any advice.

[Erik Ruzek]

There are some good resources for this online and it is worth creating an account to utilize the excellent LEMMA from U Bristol, which has a multiple membership module. To get a sense of what a data setup for a multiple membership looks like, you can download the Tennessee STAR data file that is part of Rabe-Hesketh & Skrondal's Multilevel and Longitudinal Modeling in Stata volumes.

use https://www.stata-press.com/data/mlmus3/star_mm.dta, clear

* Weights constructed based on the proportion of observations in which the student was in each classroom/teacher
list wttr1-wttr30 in 1/1

* Model using those weights
mixed reading || _all: wttr1-wttr30, cov(identity) nocons || schid: