[R-sig-ME] Post-hoc analysis of logistic mixed effects model
Yasuaki SHINOHARA
y.shinohara at aoni.waseda.jp
Tue Mar 1 10:14:46 CET 2016
Dear list,
I am analyzing data with a logistic mixed effects model.
I created the following model.
mod1<-glmer(binomial_resposne ~ FactorA*FactorB*FactorC
+(1+FactorA|subject)+(1+FactorA|speaker/item),family=binomial,data=mydata,control=glmerControl(optimizer="bobyqa",
optCtrl=list(maxfun=1000)))
FactorA is a categorical variable with 2 levels, "A1" and "A2".
FactorB is a categorical variable with 3 levels, "B1", "B2", and "B3".
FactorC is a continuous variable.
To analyze the main effect of each factor and their interactions, I
used Anova() with the car package, and reported the results of main
effects of each factor in a paper (e.g., χ2(2) = 20.11, p < 0.001).
To do post-hoc analyses, I made orthogonal contrasts for each factor,
and used the summary() function. I reported the b-value, SE, z-value
and p-value for the significant effects (e.g.,β = 0.04, SE = 0.01, z =
2.76, p < 0.01).
Is it possible to stick to one method of the analyses, rather than
using two different methods (chi-square and regression coefficients)?
So my questions are as follows.
1. If I stick to the regression coefficient method with the summary()
function, how could I get the results of the main analysis for FactorB
with 3 levels?
2. If I stick to the chi-square method with the Anova() function, how
could I do post-hoc tests with the same method (not lsmeans or glht)?
Thank you very much for your help in advance.
Best wishes,
Yasu
More information about the R-sig-mixed-models
mailing list