I want to fit a random effect model using the glmmTMB function in R. I specified the model to be as model = glmmTMB(Y ~ (1+x1+x2|group)). The estimation results still contain a universal intercept stored in obj$env$last.par.best[1] as beta, even though I did not put a 1 as a universal intercept outside of the random part. Anyone knows how to fit a random effect only model using glmmTMB? Thanks!
The reason I want a random effect only model is because I have a panel data with many groups, and each group has a very different relationship between y and the x's. So basically I just want one model for each group. I thought glmmTMB is able to help me achieve that.
I don't understand the motivation, but it's straightforward to implement this in any of the common packages using -1 in the formula to remove the constant from the fixed effects design matrix. Here is an example using the data from lme4::sleepstudy
library('glmmTMB')
df <- lme4::sleepstudy
names(df) <- tolower(names(df))
m1 <- glmmTMB(formula = reaction ~ -1 + (days|subject), data = df)
summary(m1)
# Family: gaussian ( identity )
# Formula: reaction ~ -1 + (days | subject)
# Data: df
#
# AIC BIC logLik deviance df.resid
# 1840.8 1853.6 -916.4 1832.8 176
#
# Random effects:
#
# Conditional model:
# Groups Name Variance Std.Dev. Corr
# subject (Intercept) 63770.1 252.53
# days 142.2 11.93 0.88
# Residual 654.9 25.59
# Number of obs: 180, groups: subject, 18
#
# Dispersion estimate for gaussian family (sigma^2): 655