One problem with Jason's suggestion (and this applies equally to the
interquartile range and the semi-interquartile range) is that the
resulting value tells you nothing about the asymmetry of the data.
Since John's original reason for considering the median (&c) at all
was that some of his variables are (or at least appear to be!) strongly
enough skewed to worry about, I would rather suppose that it were
preferable to report both quartiles (or other symmetric percentiles),
so as to make the degree of asymmetry at least retrievable if not
necessarily visible (depending on the form of the reportage).

To answer John's P.S.:
 Q == semi-interquartile range = 1/2(interquartile range)
 and can be construed as the average (over both directions) of the
median distance of data points from the median.  I believe (but am not
certain, not having taken the time to do the algebra) that Jason's
suggestion may sometimes generate Q, but only for some distributions.
(Jason's measure is constrained to be either an observed deviation or a
value midway between two adjacent observed absolute deviations;  Q is
not so constrained.)  (I myself would prefer Q to Jason's median of
absolute deviations from the median, I think.)

On 28 Feb 2003, Jason wrote in part:

> If your data is x1,x2,...,xn with mean m, then the SD is the square
> root of the mean of (x1-m)^2,(x2-m)^2,...,(xn-m)^2. It takes this
> form for computational reasons, but in principle you can think of it
> as the mean of |x1-m|,|x2-m|,...,|xn-m|.
>
> So if the median in M, the analog would be to look at the median of
> |x1-M|,|x2-M|,...,|xn-M|. Whether or not anyone actually uses this,
> I don't know.

 -----------------------------------------------------------------------
 Donald F. Burrill                                            [EMAIL PROTECTED]
 56 Sebbins Pond Drive, Bedford, NH 03110                 (603) 626-0816


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to