I'm working on a data set modeling road kills (0 = random point, 1 = road kill) as a function of a number of habitat variables. Following Hosmer and Lemeshow, I've examined each continuous predictor variable for linearity, and a couple appear nonlinear. I'd like to try a fractional polynomial transformation for each, also following Hosmer and Lemeshow, and have looked at the R package mfp, but I'm having trouble coming up with (and understanding) the R code that will correctly transform the variable. Can anyone suggest R code that would help me accomplish the concepts on p. 101 - 102 of Hosmer and Lemeshow's Applied Logistic Regression (2000).
Here is some R code with an example taken from an example data set included in package
library(MASS) library(mfp) data(birthwt) vignette("mfp_vignette",package="mfp") # Read this! mfp_mod <- mfp(factor(low) ~ fp(age,df=4)+fp(lwt,df=4)+factor(race)+factor(smoke)+ptl+factor(ht)+factor(ui)+ftv, family=binomial,data=birthwt) summary(mfp_mod)
(I did not include output). An alternative would be to use splines.