How to interpret tvc after stcox?

[There Scho]

Hello everyone,

I have a peace agreement dataset and I am looking at peace duration afterwards. I have one control variable ( female share in parliament) that is time-varying, so I want to use tvc, however, I am not sure how to interpret the results.
And another question: Do I need the log of time, thus ln(_t), or in which cases do I have to use it?

This is my command:
stcox gender wdi_lifexp gdp fh_polity2 wdi_fertility intensity_level wdi_wip, tvc(wdi_wip)


These are my results:

[There Scho]

If anything is not clear, I can also specify it.

[Paul Dickman]

Here's a brief answer. This question is quite standard so you will be able to find details elsewhere (including other posts in this forum).

The HR for wdi_wip is 0.96 at time 0 and is multiplied by 1.04 for each 1 unit increase in time.

Do I need the log of time, thus ln(_t), or in which cases do I have to use it?

There are an infinite number of ways to have non-proportional hazards; texp() allows you to specify one way. You should fit a model that is most appropriate for your subject matter and/or data.

I have one control variable ( female share in parliament) that is time-varying, so I want to use tvc ....

Be aware that time-varying covariates and time-varying coefficients are very different concepts. tvc() and texp() are for time-varying coefficients. Having time-varying covariates does not imply the need for time-varying coefficients.

[There Scho]

Thank you so much!

[Andrea Discacciati]

The fact that tvc() stands for time-varying covariates in stcox's help file can be source of confusion, in my opinion (although the [ST] Reference manual (v15, page 79 "Cox regression with continuous time varying covariates") explains why tvc = time varying covariates).
Arguably, it's more common to refer to those models as "Cox regression models with time-varying (or time-dependent) coefficients".

[Paul Dickman]

The fact that tvc() stands for time-varying covariates in stcox's help file can be source of confusion, in my opinion (although the [ST] Reference manual (v15, page 79 "Cox regression with continuous time varying covariates") explains why tvc = time varying covariates).
Arguably, it's more common to refer to those models as "Cox regression models with time-varying (or time-dependent) coefficients".

Thanks Andrea.

Yes, you're quite right about this being a source of confusion (based on my experience) and you are also quite right that my statement "tvc() and texp() are for time-varying coefficients" is not strictly correct. As you point out, these options can also be used for time-varying covariates.

In the version 16 help it's example 4 on page 79 where tvc() is used for time-varying covariates. That example contains the statement "Above we used tvc() and texp() to demonstrate fitting models with time-varying covariates, but these options can also be used to fit models with time-varying coefficients." In my practical experience I have only ever used these options, or seen them used in practice, for time-varying coefficients. The important point, however, is to understand the difference between time-varying coefficients and time-varying covariates.