2

Is there a version of the CLT for distributions that don't have a mean, but have a median, and finite moments of order > 1? (if such distributions exist)

BruceET
  • 47,896
  • 2
  • 28
  • 76
Frank
  • 1,305
  • 1
  • 12
  • 17
  • 3
    The sampling distribution of the sample median $\hat{M}$ is approximately normal with mean $m$ and variance $\frac{1}{4n(f(m))^2}$ where $n$ is the sample size, $m$ is the population median, and $f$ is the pdf for the observations i.e. $$\hat{M} \sim N(m,1/4n(f(m))^2)$$ This approximation is good for $n\geq 30$. Is this what you're looking for? – Matthew H. Sep 15 '20 at 02:52
  • 6
    There aren't any distributions with finite moments of order >1 but no mean. If any given moment exists, so do all the moments of lower order. – Thomas Lumley Sep 15 '20 at 03:40

1 Answers1

1

Example: A random variable with a standard Cauchy distribution (t distribution with DF=1) has no mean, but median $\eta = 0$ and $f(\eta)=1/\pi.$ So according to @MatthewHolder's Comment (see Ref.), the median $H$ of a sample of size $n=100$ from this distribution should have $H \stackrel{aprx}{\sim} \mathsf{Norm}(\mu=0,\, \sigma=\pi/20).$

set.seed(2020)
h = replicate(10^5, median(rt(100,1)))
summary(h); sd(h); pi/20
      Min.    1st Qu.     Median       Mean    3rd Qu.       Max. 
-0.8457575 -0.1043798  0.0006376  0.0013486  0.1067248  0.6956875 
[1] 0.1589696  # aprx SD
[1] 0.1570796  # asymptotic SD

hist(h, prob=T, col="skyblue2", ylim=c(0,2.5), 
     main="Dist'n of Medians of 100 Std Cauchy RVs")
  curve(dnorm(x, mean(h), sd(h)), add=T, col="red")

enter image description here

Ref: This relationship is Thm. 7.9.1 in Bain & Englehardt (1992): Intro. to Probability and Math. Stat. (2e). Also, in other math stat books along with discussions of order statistics. Also see this Q&A.

BruceET
  • 47,896
  • 2
  • 28
  • 76