On 9 Apr 2004 02:14:37 -0700, [EMAIL PROTECTED] (Tim) wrote: > A basic simple question, but of quite a bit of importance to me: > > I'm dealing with a rather gaussian distribution, and I want to > calculate the kurtosis of the distribution. Is it necessary to adjust > the gaussian distribution to a standardnormal distribution to > calculate the kurtosis or not?
By standardnormal, you must mean (I take it), setting SD to 1.0. Not. This should be utter simplicity to demonstrate: divide your scores by a constant. Does the kurtosis stay the same? Of course, there are several formulas in use, so you might want to check the behavior of your own formula. Read the formula, for one thing: Doesn't it divide by the variance? - and other 'width' in order to standardize? I think that the answer to your questions is probably going to be best determined by studying the formulas that you are using. > In other words: is kurtosis dependent of the variance of the > distribution. > I expect this not to be the case. > Therefore I obtained the kurtosis without standardisation. In a > further step I correlated a few variables, and kurtosis and > standarddeviation seemed to be very significant correlated. This could > be inherent to my data, and would be very interesting, but therefore I > would have to be sure kurtosis and variance/standarddeviation are > independent! > So here comes my question again: are they independent, and if they > are: are they under any circumstance independent or are there > limitations? Do a groups.google Advanced search of sci.stat.* . >From a thread in March, 2003, Donald Burrill reminded us that no one much cares about the actual values of skewness and kurtosis. And that is one reason why several formulas are still used. My post in that thread mentioned that there are good references to be found by searching <skewness kurtosis group:sci.stat.* author:heiser > -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
