[R] Robust or Sandwich estimates in lmer2

Doran, Harold HDoran at air.org
Wed Sep 19 18:40:22 CEST 2007


This has come up before and I'll again ask the question "why would you
want robust standard errors in lmer"? Traditional econometric thinking
suggests that there is model mispecification if OLS is used and there is
a violation to the assumption of independence. So, one may still get the
point estimates via OLS but then get robust standard errors. This makes
sense.

But, mixed models are designed to account for violations to the iid
assumption via correctly specified random effects. So, if your lmer
model is correctly specified, the standard errors should yield an
accurate estimate of the true sampling variance.

> -----Original Message-----
> From: r-help-bounces at r-project.org 
> [mailto:r-help-bounces at r-project.org] On Behalf Of Sundar Dorai-Raj
> Sent: Wednesday, September 19, 2007 11:42 AM
> To: Abdus Sattar
> Cc: R-help at stat.math.ethz.ch; gsrwork at yahoo.com
> Subject: Re: [R] Robust or Sandwich estimates in lmer2
> 
> Abdus Sattar said the following on 9/19/2007 7:03 AM:
> > Dear R-Users:
> > 
> > I am trying to find the robust (or sandwich) estimates of 
> the standard error of fixed effects parameter estimates using 
> the package "lmer2". In model-1, I used "robust=TRUE" on the 
> other, in model-2, I used "robust=FALSE". Both models giving 
> me the same estimates. So my question is, does the robust 
> option works in lmer2 to get the robust estimates of the 
> standard error? If anybody could offer me a suggestion I 
> would greatly appreciate it. Thank you.  
> > 
> > Model-1:
> > 
> >> p.mle<-lmer2(ddimer~race+steroid+psi+sofa+apache + 
> (apache|subject), 
> >> method="ML", data=final, robust=TRUE, cluster="id", 
> weights=final$w)
> >> beta=fixef(p.mle)
> >> Vcov=vcov(p.mle, useScale=FALSE)
> >> se=sqrt(diag(Vcov))
> >> beta
> >  (Intercept)         race      steroid          psi         
> sofa       apache 
> >  5.826489820 -0.001920670 -0.242040171  0.005293996  0.075468340  
> > 0.009245152
> >> se
> > [1] 0.108325229 0.058921371 0.055975547 0.001285687 0.018119089 
> > 0.002559902
> > 
> > Model-2:
> > 
> >> p.mle<-lmer2(ddimer~race+steroid+psi+sofa+apache + 
> (apache|subject), 
> >> method="ML", data=final, robust=FALSE, cluster="id", 
> weights=final$w)
> >> beta=fixef(p.mle)
> >> Vcov=vcov(p.mle, useScale=FALSE)
> >> se=sqrt(diag(Vcov))
> >> beta
> >  (Intercept)         race      steroid          psi         
> sofa       apache 
> >  5.826489820 -0.001920670 -0.242040171  0.005293996  0.075468340  
> > 0.009245152
> >> se
> > [1] 0.108325229 0.058921371 0.055975547 0.001285687 0.018119089 
> > 0.002559902
> > 
> > 
> > Best Regards,
> > 
> > Sattar
> > 
> > 
> 
> The help page to ?lmer2 in the lme4 package makes no mention 
> of "cluster" or "robust" arguments. To me, that would mean 
> these arguments are ignored.
> 
> HTH,
> 
> --sundar
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide 
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 



More information about the R-help mailing list