[R-sig-ME] Setting average as baseline rather than a dummy variable in a negative binomial glm

[Aitor Gastón]

You can change the contrasts type using contrasts.

I always use the default contrast in R (treatment), but I guess that you are 
looking for sum contrast, running

options(contrasts=c("contr.sum","contr.poly"))

before you fit the model will change to sum contrasts, i.e., the intercept 
shows the grand mean and the coefficients of each level are the difference 
between the grand mean and each level.

Aitor

-----Mensaje original----- 
From: Daniel Rubi via R-sig-mixed-models
Sent: Friday, August 05, 2016 6:37 PM
To: R-sig-mixed-models
Subject: [R-sig-ME] Setting average as baseline rather than a dummy variable 
in a negative binomial glm

Hi,
I first posted this in cross validated, but thought this forum is better 
suited for this question.

I have binomial data (meaning k successes out of n trials) for a set of 
conditions. I would like to fit a glm in order to quantify the effect of 
each condition on the success.Since the data are overdisperesed I thought of 
using a negative binomial glm (glm.nb from the R MASS package does 
that).Code snippet (though not really overdisperesed):set.seed(1)
df <- data.frame(k = as.integer(runif(200,1,20)),
                 n = as.integer(runif(200,100,200)),
                 cond = rep(LETTERS[1:20],10),
                 stringsAsFactors = F)
df$cond <- as.factor(df$cond)
library(MASS)
fit <- glm.nb(k ~ cond + offset(n), data = df)
Obviously cond A will be set as baseline and all effects will be relative to 
it. However, this makes interpretation very difficult for me. Therefore my 
question is how do I fit a glm.nb model where the effects are relative to 
the mean across all conditions rather than the dummy variable set as 
baseline?
Thanks a lot,Dan
[[alternative HTML version deleted]]

_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models