I have some data on the presence of disease pathogens in edible crabs. I wish to know if the presence of a given pathogen is affected by crab width, sex, body condition, external disease, biofouling and if this differs between sampling sites (n = 7). As site has created dependency, it has been suggested that I run a random-effects model on this data instead of a Bernoulli GLM.
Here is what my model looks like for each pathogen in R:
model <- glmmTMB(y ~ X+.....Z+(1|Location), data = crab, family = binomial)
Where y is the pathogen, X+...Z+ are sex, condition, disease, biofouling and width.
For one pathogen, Paramikrocytos canceri, I have found a significant effect of width and body condition on the probability of being infected
Single term deletions
Model:
Paramikrocytos ~ Width + Condition + (1 | Location)
Df AIC LRT Pr(>Chi)
<none> 199.08
Width 1 210.48 13.4016 0.0002514 ***
Condition 1 200.96 3.8816 0.0488167 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
However, when I try and run DHARMa model validation, pretty much all validations are invalid, that is there is a significant deviation from uniformity
Also there are large deviations in the scaled quantile residuals and zero-inflation
It appears this model won't work for my data as I have a non-linear relationship. My question is what kind of model can I run on non-linear data with a random effect? I am new to mixed effect models and DHARMa so any help is greatly appreciated.