Timeline for Evaluating Functional Form in Cox Regression using rcs
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| May 9, 2021 at 9:36 | vote | accept | user89547235 | ||
| May 8, 2021 at 21:24 | history | edited | EdM | CC BY-SA 4.0 |
added 35 characters in body
|
| May 8, 2021 at 21:15 | history | edited | EdM | CC BY-SA 4.0 |
added about martingale residual plot
|
| May 8, 2021 at 20:47 | comment | added | EdM |
@pankevedmo this is what I expected: a very strong linear association with a very weak combined non-linear association, even though neither of the individual coefficients of the non-linear terms is distinguishable from 0. The Wald test combining the 2 non-linear coefficients takes into account the covariance in their estimates, which leads to the results you found. Apply the vcov() function to your model to see the variance-covariance matrix of the coefficient estimates. The standard errors are the square roots of the diagonal entries; the covariances are the off-diagonal entries.
|
|
| May 8, 2021 at 20:40 | history | edited | EdM | CC BY-SA 4.0 |
added in response to comments and edited question
|
| May 8, 2021 at 19:44 | comment | added | user89547235 | there you go. Can you please help me how to interpret this output? | |
| May 8, 2021 at 16:52 | comment | added | EdM |
@pankevedmo could you also include the result that suggests non-significant spline coefficients for var ' and var''? From the anova() output it looks like most of the response is from the linear term, with the "significance" of the non-linear term mostly due to the very large sample size.
|
|
| May 8, 2021 at 16:47 | comment | added | user89547235 | done! I've added the result of the anova. | |
| May 8, 2021 at 16:19 | comment | added | EdM |
@pankevedmo I'm a little concerned that there might be some confusion in how to interpret the anova() test on the Cox regression. It would help for you to edit your question to show those results. That will both give guidance to others who come across this question and help determine just why you are getting a significant "Non-Linear" anova() result with non-significant spline coefficients for var' and var''.
|
|
| May 8, 2021 at 7:40 | comment | added | user89547235 | Thank you for your answer! I performed the anova and the output showed a p <0.0001 for "Non-Linear" term of "var", although the coefficients in the Cox Regression (var', var'') were highly non-significant. I don't know how this can happen but still, I deduced that a non-linear fit may be worthy. | |
| May 7, 2021 at 18:35 | history | answered | EdM | CC BY-SA 4.0 |