[R-sig-ME] Truncated Negative Binomial Model Unexpected Marginal Means
Alex Waldman
@|ex@w@|dm@n @end|ng |rom @jc@ox@@c@uk
Tue Feb 15 16:04:50 CET 2022
Dear All,
Hope all is well! This may be a naïve question but I am running a hurdle negative binomial model to look at the differences in counts of differing types in different locations. My major interest is the conditional model (ie when counts are above 0).
I run the following code:
model<-glmmTMB(Count ~ Location*Type + (1 | ID), zi=~Location*Type + (1|ID), data=data, family="truncated_nbinom1",control=glmmTMBControl(optimizer=optim, optArgs=list(method="BFGS")))
var.corr <-VarCorr(model)
Conditional model:
Groups Name Std.Dev.
ID (Intercept) 0.37105
Zero-inflation model:
Groups Name Std.Dev.
ID (Intercept) 1.3207
emmeans <- emmeans(model, ~ Location*Type, type="response", sigma=0.37105, bias.adjust=TRUE)
Location Type response SE df lower.CL upper.CL
0 0 1.117 0.277 631 0.687 1.82
1 0 0.940 0.251 631 0.556 1.59
2 0 0.893 0.266 631 0.498 1.60
0 1 1.325 0.254 631 0.909 1.93
1 1 1.090 0.248 631 0.698 1.70
2 1 1.452 0.300 631 0.967 2.18
Confidence level used: 0.95
Intervals are back-transformed from the log scale
Bias adjustment applied based on sigma = 0.37105
However, I’m not sure why the estimated means and confidence intervals will include values below 1 in the conditional model as I anticipated these values would represent the average number of non-zero counts? Is there something I may be doing wrong or not understanding?
Thanks in advance for your help!
Warm Regards,
Alex
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list