Solution |
The mismatch of the kurtosis (and skewness) between some statistics packages (e.g. Excel, Minitab, or Simca) and the SDL Component Suite is due to different assumptions: while Excel and the others assume the data to be a sample, the method SkewKurt of TVector (and of TMatrix) is based on the assumption that the data is a population. You can convert the population kurtosis into the sample kurtosis by using the following transformation:
KurtosisSample := (KurtosisPopulation+3)*sqr(NumData)*(NumData+1)/
(NumData-1)/(NumData-2)/(NumData-3) -
3*sqr(NumData-1)/(NumData-2)/(NumData-3);
with `KurtosisPopulation` being the kurtosis value calculated by TVector, and `NumData `being the number of values used for the calculation. |