Use of tvc and smooth.formula with factor variables

[mikesweeting]

Hi Mark, I think there may be a bug using stpm2 with factor variables. I get an error message when trying to change the degrees of freedom for a time-varying factor variable, but the error message goes if we use dummies instead. See example below

data(brcancer)
# Create a 3-level factor variable
brcancer$x4 <-  factor(brcancer$x4)

# This works... 
summary(stpm2(Surv(rectime,censrec==1)~x4,data=brcancer,df=3, tvc = list(x4=3)))
# but the following do not fit
summary(stpm2(Surv(rectime,censrec==1)~x4,data=brcancer,df=3, tvc = list(x4=2)))
summary(stpm2(Surv(rectime,censrec==1)~x4,data=brcancer,df=3, tvc = list(x4=2), control=list(robust_initial=T)))
summary(stpm2(Surv(rectime,censrec==1)~x4,data=brcancer,df=3, tvc = list(x4=1)))

# We can avoid the error using dummy variables instead 
library(fastDummies)
brcancer <- dummy_cols(brcancer, select_columns = "x4", remove_first_dummy = T)
summary(stpm2(Surv(rectime,censrec==1)~x4_2+x4_3,data=brcancer,df=3, tvc = list(x4_2=3, x4_3=3)))
summary(stpm2(Surv(rectime,censrec==1)~x4_2+x4_3,data=brcancer,df=3, tvc = list(x4_2=2, x4_3=2)))

[mclements]

Mike: my apologies for this very late reply.

There is a long-standing issue with using interactions between factor variables and splines in rstpm2 -- the design matrix is often not what you want. The simplest approach is to use indicator variables.

Sincerely, Mark.

[mikesweeting]

Thank you Mark for informing me about issues with using factor variables in rstpm2. I'll bear this in mind and will use dummy variables when needed.

Best wishes Mike