How do you compare skewness and kurtosis levels between data sets?

  • Level: Undergrad 
  • Thread starter member 428835
  • Start date
1 reply · 3K views · last reply
member 428835
hey pf!

i am wondering, if you're looking at two data sets and each set has different skewness levels (i.e. perhaps set 1 has a skewness of .4 and set 2 has a skewness of .5) do we say that these two are relatively un-skewed or highly skewed (or perhaps one of each)?

in other words, how do i compare levels of skewness?

i have the same question for kurtosis, if you could explain?

thanks a ton!

josh
 
You may be trying to get more out of these measurements than they deserve, but that's OK. Here is my two bits. Maybe someone can clarify more:

The skewness of a PDF is definitely influenced by its standard deviation, σ. For this reason, I think that the skew values of two PDFs can only be meaningfully compared if they have identical standard deviations. Of course, any symmetric PDF will have skewness=0. Other than that, I would normalize the skewness number by dividing it by σ3. I would look at some well known unsymmetric distributions like Chi2 and normalize their skewness values to get a feel for the meaning of some values. I have never done this exercise, but maybe someone else can add some insight.