Experiment Two
We performed the same regressions as for experiment one to predict infected, uninfected, and total snail biomass in experiment two using a categorical random term for pesticide, nested within class, and compared it to a null (intercept) model using likelihood ratio tests. To identify which herbicides have strong effects in the absence of predation, we used the fixed categorical predictor for pesticide identity to separately predict natural-log transformed infected, uninfected, and total B. glabrata snail biomass, in tanks without crayfish predators, using a Gaussian error distribution in the glmmTMBpackage (Brooks et al., 2017) in R v4.0 (RCoreTeam, 2013). Total biomass included living and dead snails to examine whether herbicides caused increases in snails early in the experiment despite those snails perishing before the end of the experiment.
To test for an interaction between crayfish predators and herbicides in experiment two, we separately predicted log-transformed uninfected, infected, and total Bi. glabrata snail biomass, using Gaussian error distributions that considered all possible interactions between a fixed categorical term for crayfish presence and fixed categorical terms for pesticide, class, or type. We also considered a random term for block, a term for zero-inflation, and whether water and solvent tanks should be treated separately or as a single control group, resulting in 20 possible crayfish interaction models. We compared all interaction models to one another, a model using only crayfish presence and a null (intercept only) model using AICc. We also used a Pearsons’ correlation coefficient to assess the relationship between infected and total snail biomass, and performed a negative binomial regression to predict total cercariae counts in tanks using shedding snail biomass.