Date:         Mon, 22 Aug 2011 14:28:54 -0400
Reply-To:     Rich Ulrich <rich-ulrich@live.com>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Rich Ulrich <rich-ulrich@live.com>
Subject:      Re: median of two groups are statistically significant
Comments: To: nsaha6@gmail.com
In-Reply-To:  <201108221444.p7MAlY77017117@waikiki.cc.uga.edu>
Content-Type: multipart/alternative;


The chapter cited by Bruce points out that the usual non-parametric tests
are tests on average rank, and not tests on the median.  And those tests 
do assume that the original distributions are of the same form, so that 
the tests are tests on "shift" or location in similar distributions.

For fewest assumptions, you might test the equality of medians
by a simple contingency table that reports how many in each 
group are below/above  the overall median.  (Recode every score
to 0/1 and cross-tabulate.)  This principle can be applied to other
fractions just as readily, such as, "top 1%".

People are usually more interested in the average rank.  There might be
more reports published where people mis-report a rank-test as a test
on medians than where people consciously intend a test on medians 
and use an intentional test on medians.  

-- 
Rich Ulrich 


> Date: Mon, 22 Aug 2011 10:44:00 -0400
> From: nsaha6@GMAIL.COM
> Subject: median of two groups are statistically significant
> To: SPSSX-L@LISTSERV.UGA.EDU
> 
> Hello All,
> 
> What is the right statistical test to determine whether the median of multiple
> groups are statistically significant or not?

                                          

[text/html]