0
$\begingroup$

I have my database in temporal blocks (so I have my occurrence sites into the year of sampling of a few years with maximum 7 years). In order to fix the complete separation problem that I also have, due to some levels of one categorical predictor I use penalized-likelihood logistic regression but here it will be difficult to use a mixed-effect model and more precisely test the significance of the random effect.

I’m thinking of only using fixed-effect models (penalized lik glm and random forest) and include the year of sampling as an explanatory variable in the fixed-effect part. Is it feasible to:

  1. include the year of sampling as a fixed effect instead of a random effect?
  2. and by using this variable (year of sampling) in the random forest.

It will be also very helpful if there are some works that include those two points.

$\endgroup$
4
  • $\begingroup$ If you are thinking of using a random forest then I assume you are only interested in prediction, not inference ? If so then why are you concerned about testing "the significance of the random effect" ? $\endgroup$ – Robert Long Mar 15 at 11:15
  • $\begingroup$ @RobertLong, My plan was to use random forest and GLM when the year of sampling is not significant in the GLMM because I'm not including the year of sampling as a predictor for the random forest otherwise I use only GLMM. $\endgroup$ – user1988 Mar 15 at 14:59
  • $\begingroup$ But what is your research quesion ? Are you interested in prediction or inference ? $\endgroup$ – Robert Long Mar 15 at 15:39
  • $\begingroup$ I'm interested in prediction but also I want to have information about the significant predictors in my regressions. But since I don't use mixed effect random forest, so that's why I do the test of the significance of the random effect in GLMM to decide if I include or not RF. $\endgroup$ – user1988 Mar 15 at 15:50

Your Answer

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

Browse other questions tagged or ask your own question.