Sample size calculation when comparing two coefficients of variation

sirgeensnor

New Member
Hi,

I am a medical student, wanting to start a clinical study between two devices. Since I want to compare especially on precision and reproducibility, I am going to use the coefficient of variation. I already have the CV of one of the devices from another published article (14,6%) and have calculated the other device will give me a CV of 11%.
Which method do I use to calculate my sample size that is needed to power this study? At this moment I use G*Power for my calculations.

Thank you and have a great weekend.

fed2

Active Member
is the measurement continuous, count, normal, lognormal or what? Need to have some idea of that. I'd suggest that this will work out best under log-normal model.

katxt

Active Member
Assuming you want to show that the second is more precise than the first, you can use a Monte Carlo approach. Choose a possible sample size n. Generate n normal(100, 11). Calculate CV for your sample. Record 1 if the sample CV is less than 14.6% else 0. Repeat many times. The average of the 0, 1s is the power. If it is too low, increase the sample size and try again. This is a one tailed test as it stands.

katxt

Active Member
This little known formula may also help. For reasonable values of CV and n, SE(CV)=CV/sqrt(2n)

fed2

Active Member
im pretty sure the OP just wanted to know what button to push on G*Power. Poor bugger....