Joint significance F-test

[Michael McCracken]

Hello fellow statisticians,

I have a very general question. In survival/duration analysis is it viable to use command -test to test joint significance of few variables?

If yes, given that I am doing -stcox regression ( the coefficient on estimates are above/below 1 depending whether they increase/decrease survivability) do I need to specify in -test that the null hypothesis should be that variables are equal to 1 to check their actual significance? (I think stata by default assumes hypothesis 'equal to 0')


Thank you for any feedback


Michael McCracken

[Jorge Eduardo Perez Perez]

Yes, it is feasible to use the test command, and no, you should not set the nulls to equality to 1. This is because stcox shows the hazard ratios by default, but what is stored (and called by test) is the actual coefficients.

webuse kva
stset failtime
* Shows hazard ratios
stcox load bearings
* But estimates coefs
mat list e(b)
* p-value here is same as table, Hazard ratio=1 or coef=0
test load