I have run mixed beta regressions on proportional data but I am struggling in my interpretation of the results. I understand this has been asked before but I have a categorical predictor with four levels and I want to understand exactly what my results mean.
In my example I have the proportion of particulate matter found in a sample within four different locations. I have 'site' as a random factor. I have used glmmTMB with beta_family.
CPOMmod2 <- glmmTMB(proportion ~ feature + (1|site), REML = TRUE, family = beta_family, data = cpomfilter)
I believe the model coefficients are log-odds and the exponent of these coefficients are odd-ratios. The plot below shows the odds ratios from the plot_model function (sjPlot).
My question is, is the interpretation of these results that "the odds of there being a higher proportion of particulate matter in Loc3 compared to Loc0 (the baseline) increased by 1.5 times (however, CIs were just overlapping)"?
Furthermore, is there a way to use emmeans (or another package) to get marginal means from a beta regression (I seem to get the log-odds)?
< Family: beta ( logit )
Formula: proportion ~ feature + (1 | site)
Data: cpomfilter
AIC BIC logLik deviance df.resid
-877.7 -861.4 444.9 -889.7 110
Random effects:
Conditional model:
Groups Name Variance Std.Dev.
site (Intercept) 1.8e-09 4.243e-05
Number of obs: 112, groups: site, 8
Dispersion parameter for beta family (): 49.2
Conditional model:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -4.74940 0.19507 -24.347 <2e-16 ***
featureLoc1 -0.02975 0.23711 -0.125 0.900
featureLoc2 0.08328 0.21931 0.380 0.704
featureLoc3 0.41225 0.29055 1.419 0.156
