I updated the manuscript based on comments from Jim and Olli. Specifically:
I updated the theoretical details
I updated the competing-risk case study
I updated the last case study with the Stanford data
I updated the intro to mention upfront variable selection and competing risks
A couple weird things also occurred, I'm not sure why:
For CS 4, I tried switching to months, but the results were very different. We need to think more about the impact of changing the time scale: it changes the offset, and it doesn't behave linearly wrt nonlinear functions of time (e.g. log or splines). So I left it on the day scale for now.
The model that had interaction between splines and exposure was no longer the best fit... I'm not sure if the data changed recently. So I removed that model, but I kept a model that has interaction between time and exposure and adjusted the rest accordingly.
Those last two points may not play a big role if we decide to move CS 4 to a vignette anyway.
I updated the manuscript based on comments from Jim and Olli. Specifically:
A couple weird things also occurred, I'm not sure why:
logor splines). So I left it on the day scale for now.Those last two points may not play a big role if we decide to move CS 4 to a vignette anyway.