1
$\begingroup$

I am analysing the share of a tree species at different forest locations. The data I am using is in percentage of the total species composition and I am modelling it based on the climatic probability of occurrence (continuous variable) and multiple site characteristics (categorical variables).

I was wondering which regression model would be most appropriate, as my target variable shows a lot of zeros and ones (meaning the species is not occurring at all or it is the only tree species at a certain location).

Here is a histogram with the distribution of my data.enter image description here

Improve this question
$\endgroup$
3
  • 4
    $\begingroup$ You say percent but your graph seems to imply that you are working an outcome or response which is a proportion or fraction. Buzzwords here are beta regression and logit regression with mild controversy about which is to be preferred. $\endgroup$
    – Nick Cox
    Commented May 14, 2019 at 15:14
  • 1
    $\begingroup$ I would strongly consider semiparametric ordinal regression, e.g., the proportional odds model as implemented in the R rms package orm function. $\endgroup$
    – Frank Harrell
    Commented May 15, 2019 at 10:56
  • $\begingroup$ Not sure if that fits for you, but you may want to look at tobit and hurdle models. $\endgroup$
    – Maël
    Commented Sep 29, 2021 at 7:16

1 Answer 1

Reset to default
-1
$\begingroup$

I suggest to build 2 models:

  • a classification model to 3 classes:0, 1, 2: the rest.
  • regression model,only on the data with target differ then 0 and 1.

Then apply the regression model,only on those classified 2.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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