I want to evaluate the influence of some environmental variables on the richness and abundance of non-native species. For this I am using DISTLM + dbRDA. But I have some doubts. I used the Akaike selection criterion (AIC) to select the best model, but it is selecting some non-significant variables and excluding some significant variables.
My doubts:
- Do I need to first analyze the "marginal test" and select only the significant variables for DISTLM?
- After performing the DISTLM the variables that I use in dbRDA are the ones selected by the model?
You seem to be confusing two different kinds of model. Suggest you carefully look at the attached paper. Best regards, David Booth
Carolina -
First, I am not familiar with "DISTLM + dbRDA." I hope that the following addresses your question.
Galit Shmueli seems to have a no-copyright problem version of that paper on her ResearchGate pages:
https://www.researchgate.net/publication/48178170_To_Explain_or_to_Predict
If you look at the example in the appendix, you may see her point.
If you are just looking for the best prediction model, I suggest that you consider a few possibilities, based on the subject matter. You can do a graphical residual analysis to compare results, but to avoid overfitting, you should also hold out some data, and see how well you would have predicted for it.
You could research the terms "model selection" and "model assessment."
You might find something helpful here:
https://www.researchgate.net/publication/313925533_Comparing_Model_Performances_Graphically
I think it important that you consider subject matter reasons for considering the independent variables that you use. If the data drastically disagree, that should be noticeable, given enough data.
Cheers - Jim
Went through Safa's reference, and was reminded that you were considering "significance," which is something of a misnomer.
- Badges
- Science method