cross-classified MLM - MCM - random and fixed effects

Welcome to the forum for runmlwin users. Feel free to post your question about runmlwin here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Go to runmlwin: Running MLwiN from within Stata >>
Post Reply
Posts: 24
Joined: Sun Jul 29, 2018 12:37 pm

cross-classified MLM - MCM - random and fixed effects

Post by johannesmueller »

Dear all

I am working with a cross-classified MLM where there are three non-hierarchically nested effects: Geo1 Geo2 and Time (repeated cross-section). My interest is with a predictor at the Geo1 level and I seek to control for all Geo2 and Time variance. Hence, I am using fixed effects (i.e. dummies) at these levels.

My questions are:
  • If I use MCMC and cross-classified models (additive) I need to specify these three levels and will have random effects at Geo2 and Time, but at the same time also fixed effects for Geo2 and Time are included. What is the consequence of this for the standard errors and estimates? Intuitively, I would -perhaps naively- think that because of the inclusion of the dummies, the random effects wont do anything?
    I have also experimented with a multiplicate cross-classified model where I have spelt out all random terms that could arise as a product of Geo1 Geo2 and Time. By mistake, I estimated a model where I included both fixed and random effects at the Geo2*Time level. The random-effects parameters for this turned out 20times as large as all others, something I have also observed in the previous point when only using Geo2 and Time fixed and (cross-classified) random effects. What's driving this surprising explosion?
What would be your recommendation?

Thank you in advance for your guidance.

Posts: 146
Joined: Fri May 21, 2010 1:21 pm

Re: cross-classified MLM - MCM - random and fixed effects

Post by billb »

Hi Johann,
Generally when we talk about cross-classified models we are assuming that each of these levels is a random source of variation and is treated as a set of RANDOM effects. It doesn't make a lot of sense to include fixed effects as well as random effects for the same terms. Generally fixed effects are used when the particular units are important rather than a source of variation to be controlled for. So I'd suggest either simply keeping random effects in the model OR removing the level that you wish to include fixed effects for.
Hope that helps,
Post Reply