[R-sig-ME] Using multcomp in post-hoc comparisons

[Helios de Rosario]

Iker Vaquero Alba <karraspito at yahoo.es> wrote:

>    I was just wondering about the degrees of freedom of those z
scores. 

The z scores are compared with a standard normal distribution. So there
are no degrees of freedom to report.

For the rest of message, although your model does have mixed effects,
the questions are not specifically related to mixed models, but to the
interpreation of Tukey's test, so I recommend you to ask it in a more
general purpose list or forum.

Regards,
Helios

> 
>    Dear list:
> 
>    I am using "multcomp" package to make post-hoc comparisons in a
7-level 
> factor. The analysis performed is at follows:
> 
>> 
>
minclutchmodel2<-lmer(clsize2~mod+inslargest+briventralst+inslargest:briventra
> lst+(1|site/pair)+(1|year),family=poisson) 
>> cht<-glht(minclutchmodel2,linfct=mcp(mod="Tukey"))
>> summary(cht, test = univariate())
> 
>          Simultaneous Tests for General Linear Hypotheses
> 
> Multiple Comparisons of Means: Tukey Contrasts
> 
> 
> Fit: glmer(formula = clsize2 ~ mod + inslargest + briventralst + 
> inslargest:briventralst + 
>     (1 | site/pair) + (1 | year), family = poisson)
> 
> Linear Hypotheses:
>                Estimate Std. Error z value Pr(>|z|)    
> EC - E == 0  -0.5654710  0.5710626  -0.990 0.322072    
> R - E == 0   -0.2080028  0.3590288  -0.579 0.562354    
> RC - E == 0  -0.9835988  0.4836405  -2.034 0.041978 *  
> S - E == 0   -1.3622236  0.3916350  -3.478 0.000505 ***
> SC - E == 0  -0.0009705  0.4017037  -0.002 0.998072    
> U - E == 0   -0.1578700  0.2907425  -0.543 0.587137    
> R - EC == 0   0.3574681  0.5494919   0.651 0.515341    
> RC - EC == 0 -0.4181278  0.6527767  -0.641 0.521823    
> S - EC == 0  -0.7967526  0.5570890  -1.430 0.152658    
> SC - EC == 0  0.5645005  0.6019210   0.938 0.348331    
> U - EC == 0   0.4076009  0.5309090   0.768 0.442641    
> RC - R == 0  -0.7755959  0.4543557  -1.707 0.087818 .  
> S - R == 0   -1.1542208  0.3641393  -3.170 0.001526 ** 
> SC - R == 0   0.2070323  0.3462441   0.598 0.549882    
> U - R == 0    0.0501328  0.2431495   0.206 0.836650    
> S - RC == 0  -0.3786249  0.4875128  -0.777 0.437368    
> SC - RC == 0  0.9826283  0.4792105   2.051 0.040314 *  
> U - RC == 0   0.8257287  0.4180611   1.975 0.048252 *  
> SC - S == 0   1.3612531  0.4079427   3.337 0.000847 ***
> U - S == 0    1.2043536  0.3058414   3.938 8.22e-05 ***
> U - SC == 0  -0.1568995  0.2964628  -0.529 0.596640    
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’
0.1 ‘ ’ 1 
> (Univariate p values reported)
> 
>  
> 
>    I was just wondering about the degrees of freedom of those z
scores. As 
> they're pairwise comparisons, I guess it's comparing the means of the
2 
> levels involved for each caso, so I would say the degrees of freedom
for each 
> should be 1. Am I right or is there something that I'm missing?
>    Also, if I want to know the direction of the effect of the factor,
I 
> mean, which are the levels that are contributing more to the
significant 
> effect of the factor on the model, according to these data, would I
say it's 
> S>RC (both levels are significant, but S to a greater extent than
RC)?
> 
>    Thank you very much.