Originally Posted by RChung

The smaller the difference you're trying to detect, the more care you need to take in the testing. If there are other riders on the track your data won't be very clean; whether it's clean enough depends both on how many other riders there are and the size of the difference you're trying to measure. Do this just to begin: stick with one wheel and see whether you can get the same estimated CdA over a few laps. If you can, you have a hope to be able to detect a difference between that wheel and another but if you can't replicate the same CdA then your data are too noisy.