Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Cubic splines in stcox
From
Steve Samuels <[email protected]>
To
[email protected]
Subject
Re: st: Cubic splines in stcox
Date
Fri, 4 Jan 2013 20:49:40 -0500
Marc Peters <[email protected]>:
No version of spline is going to "fix" your problem. Your model is
failing precisely because you are fitting spline function(s) in year,
which is the time dimension of the -stcox- model-. -stcox- is
already fitting a flexible model in time, so the spline function(s) and
year are likely to be nearly collinear. When this happens, the partial
likelihood will not converge.
For a simple example with n =3, see Example 1 of:
Maximum Likelihood Inference for the Cox Regression Model with
Applications to Missing Covariates Ming-Hui Chen, Joseph G. Ibrahim, and
Qi-Man Shao. J Multivar Anal. 2009 October 1; 100(9): 2018–2030.
This is available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2744117/
Your spline functions probably violate at least condition C2 of their
Theorem 1.
You'd probably see your mistake more clearly if, instead of -stcox- you
fit -stpm- by Patrick Royston, or its extension,-stpm2- by Paul Lambert,
both available at SSC. These commands use splines in the the time
variable to fit the underlying survival functions. Your own functions
would be collinear with these.
You _can_ fit functions of time in -stcox- but only as interactions with
other predictors, not as main effects.
Steve
On Fri, Jan 4, 2013 at 10:42 AM, Marc Peters <[email protected]> wrote:
> Dear Statalist users,
>
>
> I am new to event history models and am estimating a Cox proportional
> hazard model, with civil war as the dependent variable. the dependent
> variable is measured at discrete intervals (yearly).
>
> My basic model is estimated using the following syntax:
>
> stset year, failure(cwar)
>
> stcox dem gdp, cluster (country)
>
> This model runs well, but I encounter problems when including cubic
> splines, using this syntax:
>
> mkspline sp_year=year, cubic nknots(3)
>
> stcox dem gdp sp_year*, cluster (country)
>
>
> When running the command I get the following message "flat region
> resulting in a missing likelihood"
>
>
> I am sure this is a very trivial question, but I would greatly
> appreciate if anyone could help me in explaining what the problem is
> and how I might fix it.
>
> Thank you,
>
>
> Marc
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/