No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Phtest based on Schoenfeld residuals and clustered standard errors with Cox PH model

    Dear Statalisters,

    I am running a Cox PH model on schooling duration with municipality fixed effects on a relatively big sample (300 000 observations) with stset and I am testing for the proportionality assumption.
    For some reason the test ran in phtest on the proportionality assumption fails drastically (all significantly different from zero on the 0.1 percent level) as soon as I cluster at municipality level. I would like to know why it does not fail locally if I do not cluster?

    Obviously I have to cluster, but I would still like to know if clustering does something with your Schoenfeld residuals that makes the Phtest fail more easily.

    Thanks a lot,
    Last edited by Max Julian; 11 Jan 2017, 14:02.