[R-sig-ME] Interpret the output of a Tweedie GLMM

[Sarah Patricia Chisholm]

Hello,

I recently posted a question on Cross Validated<https://stats.stackexchange.com/questions/500855/interpret-the-output-of-a-tweedie-glmm> but haven't had any responses yet and figured I would try here. I've copied and pasted my CV question below.

I've fit the following glmm using the glmmTMB package with the tweedie(link = "log") distribution, where y is a positive, continuous DV with a large number of zeros, x is a continuous IV and z is a categorical IV with two levels (A & B). Random factor a and a spatial correlation term are also included:


library(glmmTMB)
df$pos <- numFactor(df$LONG,df$LAT)
df$group <- factor(1)

m <- glmmTMB(y ~ x*z + (1|a) + exp(0 + pos|group),
                 family = tweedie(),
                 data = df,
                 REML = TRUE)

Here's some simplified output:

               Estimate Std. Error z value Pr(>|z|)
(Intercept)   3.669e+01  1.007e+01   3.644 0.000269 ***
x            -1.838e-02  5.007e-03  -3.672 0.000241 ***
zB           -3.149e+00  1.695e+01  -0.186 0.852592
x:zB          1.574e-03  8.435e-03   0.187 0.851962


My question is how do I interpret these parameter estimates? For example, if IV x represents time in years, is there simply a -0.01838 response in y each year (when z = A)? Or, because of the log link, do I need to apply a transformation to these parameter estimates for interpretation?


Thanks everyone,


Sarah

Sarah Chisholm
MSc Biology Candidate
Department of Biology
University of Ottawa
Linkedin<https://www.linkedin.com/in/sarah-chisholm-422a5785/>

	[[alternative HTML version deleted]]