[R-sig-ME] glmer with/without intercept gave different results

[Chang, Yu-Mei]

Dear Thierry,

I understood the hypotheses were different between the two models. What
surprise me were the different estimated variances for the random
effects and also the estimated differences between fixed effects levels.


Ruby

-----Original Message-----
From: ONKELINX, Thierry [mailto:Thierry.ONKELINX at inbo.be] 
Sent: 18 November 2010 15:43
To: Chang, Yu-Mei; r-sig-mixed-models at r-project.org
Subject: RE: [R-sig-ME] glmer with/without intercept gave different
results

Dear Ruby,

The hypotheses of those models are different. Hence the diference in
p-values.

Fit1:
H0: Capsule 1 = 0
H0: Capsule 2 - Capsule 1 = 0
H0: Control - Capsule 1 = 0

Fit2:
H0: Capsule 1 = 0
H0: Capsule 2 = 0
H0: Control = 0

However, the predictions of both model should be the same.

Best regards,

Thierry


------------------------------------------------------------------------
----
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek
team Biometrie & Kwaliteitszorg
Gaverstraat 4
9500 Geraardsbergen
Belgium

Research Institute for Nature and Forest
team Biometrics & Quality Assurance
Gaverstraat 4
9500 Geraardsbergen
Belgium

tel. + 32 54/436 185
Thierry.Onkelinx at inbo.be
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
  

> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces at r-project.org 
> [mailto:r-sig-mixed-models-bounces at r-project.org] Namens Chang, Yu-Mei
> Verzonden: donderdag 18 november 2010 15:22
> Aan: r-sig-mixed-models at r-project.org
> Onderwerp: [R-sig-ME] glmer with/without intercept gave 
> different results
> 
> Dear all,
> 
>  
> 
> I have fitted two glmer models (with or without intercept 
> term). I thought the results should be similar if not 
> identical, but they are quite different. I suspect it's 
> related to the random effects. Any suggestions on how to 
> proceed is greatly appreciated.
> 
>  
> 
> Kind regards,
> 
> Ruby Chang
> 
>  
> 
> fit1 <- glmer(Cyptoplasmic.vacuolation ~ Group +  (1|Villus) 
> + (1|Cell),family=binomial(link = "logit"))
> 
> fit2 <- glmer(Cyptoplasmic.vacuolation ~ Group -1 +  
> (1|Villus) + (1|Cell),family=binomial(link = "logit"))
> 
>  
> 
> > table(Cyptoplasmic.vacuolation, Group)
> 
>                         Group
> 
> Cyptoplasmic.vacuolation Capsule 1 Capsule 2 Control
> 
>                        0       560       340    1230
> 
>                        1       190       160      20
> 
>  
> 
> > fit1 <- glmer(Cyptoplasmic.vacuolation ~ Group +  (1|Villus) +
> (1|Cell),family=binomial(link = "logit"))
> 
> > summary(fit1)
> 
> Generalized linear mixed model fit by the Laplace approximation 
> 
> Formula: Cyptoplasmic.vacuolation ~ Group + (1 | Villus) + (1 | Cell) 
> 
>    AIC   BIC logLik deviance
> 
>  138.9 168.1 -64.47    128.9
> 
> Random effects:
> 
>  Groups Name        Variance  Std.Dev.
> 
>  Cell   (Intercept) 1983.5708 44.5373 
> 
>  Villus (Intercept)    5.9475  2.4387 
> 
> Number of obs: 2500, groups: Cell, 250; Villus, 50
> 
>  
> 
> Fixed effects:
> 
>                Estimate Std. Error z value Pr(>|z|)
> 
> (Intercept)     -11.945      9.123  -1.309    0.190
> 
> GroupCapsule 2   -1.409     14.343  -0.098    0.922
> 
> GroupControl    -17.792     33.251  -0.535    0.593
> 
>  
> 
> Correlation of Fixed Effects:
> 
>             (Intr) GrpCp2
> 
> GroupCapsl2 -0.636       
> 
> GroupContrl -0.274  0.175
> 
> > fit2 <- glmer(Cyptoplasmic.vacuolation ~ Group -1 +  (1|Villus) +
> (1|Cell),family=binomial(link = "logit"))
> 
> > summary(fit2)
> 
> Generalized linear mixed model fit by the Laplace approximation 
> 
> Formula: Cyptoplasmic.vacuolation ~ Group - 1 + (1 | Villus) + (1 |
> Cell) 
> 
>    AIC   BIC logLik deviance
> 
>  132.9 162.0 -61.43    122.9
> 
> Random effects:
> 
>  Groups Name        Variance   Std.Dev.  
> 
>  Cell   (Intercept) 5.9933e+03 7.7417e+01
> 
>  Villus (Intercept) 1.2025e-07 3.4677e-04
> 
> Number of obs: 2500, groups: Cell, 250; Villus, 50
> 
>  
> 
> Fixed effects:
> 
>                Estimate Std. Error z value Pr(>|z|)
> 
> GroupCapsule 1   -15.08      17.70  -0.852    0.394
> 
> GroupCapsule 2   -14.72      19.29  -0.763    0.445
> 
> GroupControl     -18.23      54.54  -0.334    0.738
> 
>  
> 
> Correlation of Fixed Effects:
> 
>             GrpCp1 GrpCp2
> 
> GroupCapsl2 0.000        
> 
> GroupContrl 0.000  0.000
> 
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list 
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>