I wonder if anyone could help with the following query?
>From a small (n=13) clinical study, I have calculated various
coefficients of variation of method error as a measure of both
intra and inter-observer variability in several clinical variables.
The CVs are used to express the SD of the measurement as a % of the
sample mean and range from around 20% to 36%. However,
the actual patient values used to calculate these CVs are themselves
very skewed, varying across the patients by a factor of about 12.
Implicitly, this skewness results in large SDs and therefore
large CVs. Does anyone know if there is a non-parametric
equivalent of the CV or methodology to transform the data before
calculating the CV? Any suggestions would be gratefully received.
Thanks in advance.
Dr Amanda Lee
Public Health Sciences
University of Edinburgh
Teviot Place
Edinburgh
EH8 9AG
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|