Model selection : how to deal with fixed and random effects in the same time?

Question

I am studying the effect of various ecological variables on different plant species. I want to use GLM(M), but I am not very familiar with this approach.

After a preliminary correlation analysis and considering my background knowledge, I retained 10 out of 27 initial predictors. I then built a candidate model including these 10 fixed predictors and 1 random effect (accounting for observer bias).

I noticed that some parameters were not significant for all species, so I used the dredge() function and model averaging to analyze the averaged coefficient values and their significance. This allowed me to infer the positive, negative, or undetermined (neglectable ?) effects of each variable on my species.

At this point, I have two questions:

1. Does this approach make sense? What results or methods should I use to evaluate the robustness of my findings?

2. Since dredge() does not evaluate the importance of random effects, how can I assess their relevance? For example, some species might be harder to detect, leading to an observer bias that could influence the fixed-effect coefficients.

Thank you!