I would like to make an adjustment on a baseline variable (which has a small imbalance).This is a post hoc adjustment (I recognize, however, the limits of this adjustment which has not been pre-specified). It is a randomized clinical trial. I am wondering if analysis of variance is more appropriate here than linear regression. What factor should be taken into account to decide which type of analysis to use?

However I have some missing value for the baseline variable that I shoud adjust for. I hypothesize missing data randomly.The variable to be adjusted for is a continuous variable. Is it a good idea to categorize the variable and transform into a non-responding category, the individuals for whom we have NA ? or should we do an imputation by the mean and leave the variable in numeric?


1 Answer 1


It is not permissible to adjust for observed imbalances. This causes a bias and inflates the standard error for treatment. Covariate adjustment needs to be pre-specified. Failure to pre-specify covariate adjustment as the primary analysis is a failure to understand how to maximize statistical power for an RCT. Details are here.

NEVER categorize a continuous variable.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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