For instance, if you measure habitat selection and you have many types of habitat. The research unit is usually small and contains only 1-3 types of habitat, the rest have zero percentage. Do this kind of data cause problems in logistic regression and how to deal with them.
Many thanks!
One approach is to use two variables to represent the continuous variable (The approach is described in Hosmer, Lemeshow, and Sturdivant. Applied Logistic Regression. 2013; Robertson, Boyle, Hsieh, et al, Epidemiology, 5(2); 1994). Suppose the continuous, zero-inflated variable is C (in their example, they consider smoking, where there are non-smokers, 0, and pack-year smokers -- any number greater than 0). Create a new dichotomous variable, X, such that X= 0 if C = 0; and X = 1 if C>0. Your model becomes Y = b0 + b1X + b2C. The overall effect of smoking is captured by b1 and the effect of pack-years is captured by b2.
-p
- Badges
- Science topic
- Similar topics
- Mathematics