Skip to content
Prev 14824 / 20628 Next

Anova II table, df, drop1 and very complex regression models!

I'd probably use Mplus (www.statmodel.com) for that, because it implements a very flexible generalized latent variable modeling framework that handles random effects via latent variables. It supports FIML estimation, plus a number of other estimators, including Bayesian estimation via MCMC. While expensive and not open source, it's probably the leading software package for people who do structural equation modeling; it's heavily used in the social sciences and educational research. 

Fortunately, the lavaan package in R replicates some of what Mplus can do. It may be capable of handling the problem, but I haven?t been keeping up with recent development of lavaan to see how much of the more recent Mplus features it is now supporting. 

However, if one goes the imputation route instead, there are several R packages that offer sophisticated support for imputation (Amelia, mi, mice, mitools, mix, pan, VIM, and likely more that I haven't seen/tried yet). Then you don?t need FIML estimation, just whatever normal modeling tools you already use. 


Steven J. Pierce, Ph.D.
Associate Director
Center for Statistical Training & Consulting (CSTAT)
Michigan State University

-----Original Message-----
From: Ben Bolker [mailto:bbolker at gmail.com] 
Sent: Tuesday, August 30, 2016 2:15 PM
To: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] Anova II table, df, drop1 and very complex regression models!


  Can you recommend a convenient full-information ML estimation method
that handles a wide range of cases (hopefully nested+crossed, GLMM +
LMM, etc, hopefully implemented in R) ?

  Ben Bolker
On 16-08-30 08:42 AM, Steven J. Pierce wrote: