Home  /  Resources & support  /  FAQs  /  Stata 7: Estimating stepwise Cox models
Note: This FAQ is for users of Stata 7. It is not intended for more recent versions.

Stata 7: How can I estimate stepwise Cox models?

Title   Stata 7: Estimating stepwise Cox models
Author William Gould, StataCorp
Mario Cleves, StataCorp

The Stata stepwise estimation command sw can be used with cox to estimate Cox proportional hazards models. sw cannot be used with stcox, however.

Thus, if you want to estimate stepwise models, we advise you to use cox in place of stcox. While it is true that stcox and cox estimate the same model, you want to be sure that you type the right cox command. Imagine the following situation:

  1. You have stset your data in a complicated way.
  2. You have an stcox model that you know how to run.
  3. You now want to use sw to estimate a stepwise model. sw cannot be used with stcox; you must use cox.
  4. You want to be sure you type the right sw cox command.

The way to be sure is to type the stcox command you know is right with the cmd option. For instance, if

        . stcox x1 x2, cluster(i_clm)

is an example of a command you know is right, type

        . stcox x1 x2, cluster(i_clm) cmd
                              ^^^

Rather than estimating the model, stcox will show you the cox command it would have run on your behalf. You might see something like

        . stcox x1 x2, cluster(i_clm) cmd
        -> cox _t x1 x2, t0(_t0) dead(_d) cluster(i_clm)

Now take what stcox reported and put sw in front of it:

        . sw cox _t x1 x2, t0(_t0) dead(_d) cluster(i_clm)