I'm doing survival analysis with time dependent covariates, using the counting process style. I already have a set of models and I want to test de residuals. I'm having trouble with the lack of information about this subject, it makes sense to test Schoenfield residuals for time-dependent covariates? I'm using R.
You can plot the martingale residuals for each subject. For me deviance residuals are more clear, both can be computed in R.
dev <- resid(fit,type = "deviance")
In particular, these plots will enable outlying observations to be identified.
However, diagnostic plots for covariates are not so useful because there are a number of tdc for any, and it is not clear what the residuals for the null model should be plotted against.