We can also control for the effect of bilirubin in the PBC mortality data using stratification rather than adjustment. One way to categorize is to create approximately equal-size groups. In Stata, for...

We can also control for the effect of bilirubin in the PBC mortality data using stratification rather than adjustment. One way to categorize is to create approximately equal-size groups. In Stata, for example, you can categorize by quintile of bilirubin using the command xtile cat5=bilirubin, nq(5). Try fitting a Cox model for cholesterol stratified by bilirubin, stratified at 2, 3, 10, and 50 levels. What is the trade-off in increasing the number of levels? What number of levels works best? (Hint: Balance adjustment against the size of the standard error)