0
$\begingroup$

I will apply the classical cox regression and time dependent cox regression, there are missing observations in my data. should i delete them or Should I use imputation methods?

$\endgroup$
0
$\begingroup$

Absent further information about the nature of the missingness, imputation will be the best approach. The advantages of imputation are explained in detail in Stef van Buuren's book. Even if the values are "missing completely at random" in the technical sense explained there, you are likely to lose power from omitting potentially useful cases. If there are any associations between the missingness and outcome, you run a risk of bias.

That said, sometimes the complete-cases approach can work adequately. You won't know, however, unless you try. In one early application of multiple imputation to survival data, the imputation didn't change the survival estimates much; the poor prediction of the missing covariate values pretty much outweighed the expected advantages. The trick will be developing a suitable imputation approach, which will require some careful application of your knowledge of the subject matter.

$\endgroup$
2
  • $\begingroup$ Thank you, but I wonder Is it okay if I delete individuals who have no data? $\endgroup$ – Cgdmm Jan 12 at 22:47
  • $\begingroup$ @Cgdmm that depends on what you mean by "have no data" and the reasons for the missingness. If there's absolutely no data, then what can you do? If the "no data" is just for some covariates, then impute, particularly If data are missing for reasons that are associated with outcome. If the data are "missing completely at random" in the technical sense, then omitting such individuals only lowers your power. Low precision in predicting the missing vales for cases with only limited data, however, might mean no net gain in power if you impute. Don't know until you try. Consult van Buuren's book. $\endgroup$ – EdM Jan 13 at 14:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.