[R-sig-ME] problem with lme4 glmer

Thierry Onkelinx th|erry@onke||nx @end|ng |rom |nbo@be
Mon Oct 19 19:58:29 CEST 2020


Dear Justin,

First of all you are comparing two different algorithms: GEE vs mixed
models. GEE estimates 'population average' estimates for the fixed effect.
The mixed models fixed effect refers to an average individual. Those will
be by definition different.

Very large estimates and standard errors indicate (quasi) complete
separation, leading to numerical instability. Rather a problem with the
data / model formulation than with the algorithm.

Best regards,

ir. Thierry Onkelinx
Statisticus / Statistician

Vlaamse Overheid / Government of Flanders
INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
FOREST
Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
thierry.onkelinx using inbo.be
Havenlaan 88 bus 73, 1000 Brussel
www.inbo.be

///////////////////////////////////////////////////////////////////////////////////////////
To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
///////////////////////////////////////////////////////////////////////////////////////////

<https://www.inbo.be>


Op ma 19 okt. 2020 om 19:46 schreef Rhodes, Justin S <jrhodes using illinois.edu>:

> Dear R project mixed models users:
>
> We used "lmer4", "glmer" function see below.  I attached the data set.
> The programs and results for SAS and R are shown below.  Results are
> incredibly different, and seem impossible to explain by differences in
> computational algorithms.  The estimates from SAS are reasonable, but the
> estimates from R are clearly wrong, based on looking at the simple data.
> We realize that we are underpowered to estimate the random effect here, but
> it still should give reasonable estimates if it converges, right?  Can
> someone please help us figure this out?  Thanks very much for any
> information
>
> R code:
>
> #import data
> Pole<-read.table("Pole.txt",header = TRUE)
>
> #define factors
> Pole$Color<-as.factor(Pole$Color)
> Pole$Treatment<-as.factor(Pole$Treatment)
> Pole$ID<-as.factor(Pole$ID)
>
> #model statement
> fm1<-glmer(Outcome ~ Eggs + Color + Treatment + (1|ID), family=binomial,
> data=Pole)
>
> #results
> summary(fm1)
>
>                                                Estimate            Std.
> Error            z value                 Pr(>|z|)
> (Intercept)                         -23.4673             12.4832
>       -1.880                                0.06012 .
> Eggs                                    -0.0496                3.4036
>            -0.015                  0.98837
> Colorspotted                     36.0295               8.4055
>     4.286                   1.82e-05 ***
> Treatmentsharp                12.4964              4.1453
>  3.015                    0.00257 **
> ---
>
>
> SAS code:
>
> proc genmod data=temp.Pole;
> class ID Treatment Color;
> model Outcome= Eggs Color Treatment/d=bin link=logit;
> repeated subject=ID/type=cs;
> run;
>
> Analysis Of GEE Parameter Estimates
> Empirical Standard Error Estimates
> Parameter
>
> Estimate
> Standard
> Error
> 95% Confidence Limits
> Z
> Pr > |Z|
> Intercept
>
> -1.0491
> 1.3990
> -3.7911
> 1.6928
> -0.75
> 0.4533
> Eggs
>
> -0.1071
> 0.4632
> -1.0151
> 0.8008
> -0.23
> 0.8171
> Color
> blue
> 1.5046
> 0.8476
> -0.1567
> 3.1660
> 1.78
> 0.0759
> Color
> spot
> 0.0000
> 0.0000
> 0.0000
> 0.0000
> .
> .
> Treatment
> blunt
> 0.3908
> 0.2841
> -0.1660
> 0.9476
> 1.38
> 0.1690
> Treatment
> sharp
> 0.0000
> 0.0000
> 0.0000
> 0.0000
> .
> .
>
>
>
>
> Thanks very much for your help!!
>
> Justin Rhodes
> Professor
> Department of Psychology
> Beckman Institute
>
> 405 N Mathews Ave
> Urbana, IL 61801
>
> Affiliations:  Neuroscience Program, Program for Ecology, Evolution and
> Conservation Biology, Institute for Genomic Biology, Division of
> Nutritional Sciences
>
> Email: jrhodes using illinois.edu<mailto:jrhodes using illinois.edu>
> Phone: 217-265-0021
>
> Website: http://rhodeslab.beckman.illinois.edu/
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]



More information about the R-sig-mixed-models mailing list