Skip to content
Prev 300132 / 398506 Next

significance test interquartile ranges

On Jul 14, 2012, at 12:25 , Rui Barradas wrote:

            
That's just wrong, is it not? Just because things were suggested by someone semi-famous, it doesn't mean that they actually work...

Take two normal distributions, equal in size,  with a sufficiently large difference between the means, so that there is no material overlap. The quartiles of the pooled sample will then be the medians of the original samples, and the test will be that one sample has the same number above its median as the other has below its median.

If it weren't for the pooling business, I'd say that it was a sane test for equality of quartiles, but not for the IQR.
To wit:
Westenberg-Mood test for IQR range equality

data:  rnorm(100), rnorm(100, 10, 3) 
p-value = 0.2248
alternative hypothesis: the IQRs are not equal
[1] 0.2248312 0.2248312 0.2248312 0.2248312 0.2248312 0.2248312 0.2248312
 [8] 0.2248312 0.2248312 0.2248312