When is it worth measuring a covariate in a randomized clinical trial