Skewness and kurtosis

1. May 12, 2014

joshmccraney

hey pf!

i was wondering if someone could either direct me to a source or help supply a proof on why skewness and kurtosis, from their definitions as higher order moments, graphically affect the pdf in the "skewed" and "flat" way.

let me know if ive been unclear.

thanks!

josh

2. May 12, 2014

FactChecker

Other than examining the contributors to the integral (or summation) formulas of the moments I don't know what more can be said. The skewness is positive or negative if the contributors tend to be more from the "greater than the mean" or from the "less than the mean" respectively. Likewise the the kurtosis is large or small if the the contributors tend to be more from the "farther from the mean" or from the "closer to the mean" respectively.

3. May 12, 2014

abitslow

You seem to be asking, as x increases, why does xⁿ (n>1) increase faster.
I don't have much use for either metric, as a matter of fact, kurtosis isn't a single parameter, see the relevant wikipedia articles. I like to graph the data vs their probability, that graph really IS useful, imho.
I look at std dev as the simplest way (before computers) to measure the average distance of the population from the mean. If the mean is 0, then -1 is as far away as +1, but adding them gives you a sum of zero, so we make them all positive by squaring them, summing that, then taking the square root. Now, this is just a crude explanation, and the actual formal mathematics is far more elegant. (Its like claiming that electrons orbit atoms like the Earth orbits the Sun). So, what happen if you cube a difference? well the SIGN comes back that is -1³ = -1 and +1³ = +1, so summing the cube up will certainly go negative if most samples of the population are less than the mean, and positive if most are greater, but read the wiki article for qualifications to that. And the 4th moment is just the 2nd squared, so it weighs the larger differences even more.

4. May 12, 2014

FactChecker

I like them as good single-number indicators of the shape of the PDF. Skew is a good way to indicate if one tail of a PDF is "fatter" than the other. And kurtosis is a good way to indicate if the PDF has a thin peak with fat tails (large kurtosis) or a fat peak with thin tails (small kurtosis).

There is a good reason to distinguish a circular orbit from an elliptical orbit. I see skew and kurtosis as valuable in a similar way.