7.15 Collinearity

Collinearity in a Cox regression leads to the same problems as it does in other forms of regression (unstable parameter estimates, difficulty in interpretation – see Section 5.20), and the solutions to the problem are the same (e.g., remove redundant variables). However, unlike lm and glm, the car::vif() function will not work with a coxph object. Instead, first fit a linear regression model with any numeric variable, as the outcome (here we use the event time variable) and compute the VIFs for that model.

pretend.lm <- lm(gestage37 ~ RF_PPTERM + MAGER + MRACEHISP + DMAR,
                 data = natality.complete)
car::vif(pretend.lm)
##            GVIF Df GVIF^(1/(2*Df))
## RF_PPTERM 1.006  1           1.003
## MAGER     1.162  1           1.078
## MRACEHISP 1.146  3           1.023
## DMAR      1.299  1           1.140