cph fails with rhs in the form of factor(x)

[tonigi]

Using a formula like "surv ~ factor(x)" produces the error X[, mmcolnames, drop = FALSE] : subscript out of bounds.

set.seed(10)
n <- 500
ds <- data.frame(
  ftime = rexp(n),
  fstatus = sample(0:1, size = n, replace = TRUE),
  x = sample(LETTERS[1:4], size = n, replace = TRUE) )

library(rms)
dd <<- datadist(ds)
options(datadist="dd")

fit <- cph(Surv(ftime, fstatus) ~ factor(x), data=ds)

Passing an already-factored x works as expected. Not sure if this is the intended behavior, and/or the error message may be made more explicit. The debug argument clarified the problem.

> fit <- cph(Surv(ftime, fstatus == 1) ~ factor(x), data=ds,debug=T)

sformula

Surv(ftime, fstatus == 1) ~ factor(x)
     colnames(X)  mmcolnames Design colnames
[1,] "factor(x)B" "xB"       "x=B"          
[2,] "factor(x)C" "xC"       "x=C"          
[3,] "factor(x)D" "xD"       "x=D"          
Error in X[, mmcolnames, drop = FALSE] : subscript out of bounds

This may be analogous to #15 .

Thanks!

[harrelfe]

As described in issue 29, factor() was never supposed to work with rms. It only worked in the past by accident. Now it always fails. Make the variable a factor when constructing the dataset or use the scored() function in the model formula if the variable is ordered and might possibly have a linear effect.