I've recently learned about the Generalized Mean as an abstraction of the many different means, includeing the Geometric, Arithmetic, and Harmonic means, as well as others.

It is my understanding that different means are more meaningful for different data sets, as far as providing the true Measure of Central Tendency.

My question is this: Is there a way to use the Generalized mean as a continuous (or discrete) function in it's variable $p$ - working with the same data set for all $p$ values of course, and from the output of that function, determine which value of $p$ would provide the best Measure of Central Tendency?

(I should add that if anyone has anything to add about the even more abstracted Generalized f-mean as it relates to this question that would be much welcomed as well)

  • 1
    $\begingroup$ This may not exactly answer your question but it shows relation of generalized mean in statistics statistics4u.com/fundstat_eng/ee_mean_generalized.html $\endgroup$ – Kamster Jul 18 '14 at 20:59
  • $\begingroup$ One problem Im seeing with possibility to finding optimal p to find best measure of central tendency is that sometime the median is best measure of central tendency (especially if skew in data) and there is no p such that it equals the median for generalized mean $\endgroup$ – Kamster Jul 18 '14 at 21:05
  • $\begingroup$ Put possibly if you delved into it maybe there is an asymptotic behavior that makes it good estimate of central tendency $\endgroup$ – Kamster Jul 18 '14 at 21:12

First, if you use $f(x)=x^p$, then the $f$-mean is the same as the $p$-mean.

Second, "the best measure of central tendency" depends on the problem one is trying to solve much more than on the data set (although these aspects are of course related).

Third, a very important measure of central tendency - median - is not included in either generalization.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.