1
$\begingroup$

I am a student who is performing a meta-analysis and want to select the best regression model (in terms of AICc values) that explain what are the explanatory variables that can explain better the negative impacts of nitrogen fertilization on plant biodiversity. The idea then is to always keep nitrogen amounts as explanatory variable ("mods") and then include other 7 (continuous) variables (like precipitation or temperature) and categorical variables (e.g. type of nitrogen fertilizer) in the model and to see how the nitrogen fertilization's term changes the slope.

1)What is exactly the role of interaction (obtained with "*" in the moderators in a lm or rma function?). How should I use it? Is it true that I should exclude from the model the other explanatory variables that shows an high significance in the interaction term of my model results?

2)Would you raccomand to just rely on model selection based on the 'glmulti' function? Or it would be better to play with the explanatory variables by myself? Although it is not my case, I know that glmulti might drop out the variable that I always need to keep included in my model (amount of nitrogen fertilization).

I hope my question is not too broad.

$\endgroup$
1
$\begingroup$

You can use ?formula to give you some hints about what * means in the context of a formula. If you are unsure about what an interaction actually is then you need to revise your introductory statistics material or perhaps search for posts on this site about interaction. If you include the interaction between two variables like amount of nitrogen and precipitation with * then what you are including is amount of nitrogen, precipitation, and their interaction so does the effect of amount of nitrogen depend on precipitation. You should not try to delete the main effects (amount of nitrogen, precipitation) while leaving the interaction in unless you are very sure you understand what you are doing.

As to your second question I think you would be much better testing models which have a scientific, theoretical justification rather than relying on the computer to choose a model for you.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.