1

I'm trying to do a Dunnett's test on a linear mixed model using lme4 and glht. I set up and ran the model as below

Untransformed.lmer <- lmer(Sum ~ Treatment + (1|Block), data = EggCounts_poolSUM)
anova(Untransformed.lmer)
summary(glht(Untransformed.lmer, linfct = mcp(Treatment = 'Dunnett'), alternative = 'less'))

And when I run that I get the following output

     Simultaneous Tests for General Linear Hypotheses

Multiple Comparisons of Means: Dunnett Contrasts


Fit: lmer(formula = Sum ~ Treatment + (1 | Block), data = EggCounts_poolSUM)

Linear Hypotheses:
             Estimate Std. Error z value Pr(<z)  
75 - 0 >= 0    -914.2      911.6  -1.003  0.372  
150 - 0 >= 0  -1207.4      911.6  -1.325  0.243  
300 - 0 >= 0  -2162.2      911.6  -2.372  0.030 *
600 - 0 >= 0  -1446.3      911.6  -1.587  0.160  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Adjusted p values reported -- single-step method)

Can someone explain how all treatments could end up with the same Std. error? Is there something I'm doing wrong?

HylaRestuans
  • 11
  • 2
  • Is it possible to share your data, most likely there's something weird about your predictors. you can also check the variance covarience with vcov(Untransformed.lmer) – StupidWolf Apr 05 '20 at 12:42

1 Answers1

0

The Dunnett standard error for treatment i is sqrt(s2) * sqrt(1/ni + 1/n0) where s2 is the pooled variance estimate, ni is the number of observations for treatment i and n0 is the number of observations for the reference group. So the standard errors are all the same in the case when the ni's are equal. This is likely the case for your data.

Stéphane Laurent
  • 75,186
  • 15
  • 119
  • 225