We took 2 samples (sweep netting) of bee communities on 30 vegetable fields at 2 time steps (during early flowering and shortly before harvest). After the vegetables were harvested we asked the farmers how often they applied pesticides.
Now I want to model the effect on the number of pesticide applications on the abundance of bees. The issue is that I don't know when the pesticide application took place. But I assume that the majority of applications did happen between our 2 samples. I also have additional landscape predictors that are the same for both samples.
What do you think would be the best way to model this? I could either do 2 separate models (2 x 30 obs) for each sample that includes the landscape predictors but only the second one includes the pesticide variable. Or I fit a single model (60 obs) with sample as fixed effect and site as random effect, as well an interaction term sample:pesticides. If I use a single model with an interaction term, this is the resulting interaction plot (the pesticide variable is scaled by the way...)
It kind of makes sense. But it is confusing on first glance. So I would love to get some opinions on this. I also wondered if there is some sort of best practice for situations like this.
Cheers!
