Adam Roslon wrote:
>
> Can anyone point me in the right direction to find some code or algorithms
> that can calculate the Kurtosis and Skew of and array of number, preferably
> in Pascal or C
>
> Thanks
--
Tim Victor
tvi...@dolphin.upenn.edu
Policy Research, Evaluation and Measurement
Graduate School of Education
University of Pennsylvania
There are a lot of sources in the web but I can recommend the following pages:
http://sigma.ire.pw.edu.pl/numrcp/
http://beta.ulib.org/webRoot/Books/Numerical_Recipes/
They are called "Numerical Recipes in C" (Fortran is also available).
The topics you are looking for, you can find in capter 14-1.
You need "Acrobat Reader" or any other postscript viewer.
Have Fun.
greetings
Erich
Adam Roslon schrieb:
the more important question is: why do you want to compute
skewness and kurtosis?
albyn
Thanks
1. The computation of accurate skewness and kurtosis values is not easy with
medium to large data sets. Most commercial software does not do it
correctly. As a rule of thumb, if the ratio of the square root of variance
to the mean value is less than 0.01 the computed skewness and kurtosis
values will have errors (i.e. LRE values may be less than 3).
2. Random sample sets from a true normal distribution, show wide ranges in
skewness and kurtosis values. To my knowledge there is no known true method
to determine confidence intervals about a computed skewnwss or kurtosis
value from a small to medium sample. The literature gives tables based on
asymototic methods for sample sets larger than 100 for normal distributions.
3. Richard A. Groeneveld "A Class of Quantile measures for Kurtosis", in the
current issue of the "American Statistician" (p. 325, Nov. 1998) describes
the problems with trying to make inferences based on computed fourth moment
(kurtosis). He quotes Balanda and MacGillivray as stating that "the
standardized fourth central moment in not a good measure of the shape of a
distribution".
4. Previous discussions on EDSTAT have concluded that sample skewness and
kurtosis values are of little value in determining whether the distribution
is normal or not.
DAHeiser
>
>the more important question is: why do you want to compute
>skewness and kurtosis?
>
>albyn
I knew someone would say that! Has anyone any experience of using
L-moments, which have been proposed as a more elegant way of describing
distribution shape? See
Hosking, J. R. M. Moments or L moments? Am example comparing two measures
of distributional shape. The American Statistician. 1992; 46:186-189.
Has anyone ever seen them used in published research? Come to think of
it, when was the last time you saw skewness or kurtosis coefficients
reported?
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
_/_/_/ _/_/ _/_/_/ _/ Ronan M Conroy
_/ _/ _/ _/ _/ _/ Lecturer in Biostatistics
_/_/_/ _/ _/_/_/ _/ Royal College of Surgeons
_/ _/ _/ _/ _/ Dublin 2, Ireland
_/ _/ _/_/ _/_/_/ _/ voice +353 1 402 2431
rco...@rcsi.ie fax +353 1 402 2329
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
I'm not an outlier; I just haven't found my distribution yet
>But I feel the knowing the skew and kurtosis of the histogram may allow me
>to recognize patterns more consistently.
I would look at a quantile-quantile plot or normal probability plot.
Numeric summaries of dispersion do not allow us to see patterned
departures from the expected distribution.