Rick,
I understand barely a word of the 'alpha trim mean' numbers. To this
liberal-arts type it suggests alpha(male) who has trimmed off some weight
and is ornery because of it. But I digress.....
"Visual Display of Quantitative Information" , written by Tufte, a
professor of statistics at Yale, is a fascinating read. It's graphs,
basically; what makes them effective, ineffective, informative and very
misleading, legible and illegible, etc. Who would ever have thought that
it could be an interesting subject? But it can be, and the illustrations
are simply amazing. F'rinstance, a graph that shows railroad schedule
between Paris and Lyon. Vertical lines represent time. Horizontal lines
represent distance. Towns are noted on left edge at distance from Paris.
Rate of speed is shown by slope of line. Time in depot is shown by
horizontal line. Really something to see. But the most fantastic graph
(who would ever have thought of it) show the size of Napoleon's army from
the Elbe to Moscow to the Elbe. Width of line represents # of men.
Horizontal represents distance. Major rivers are plotted on the
horizontal axis. Temperature scale is below the bottom line. What it
shows in a dreadfully fascinating way is the # of men lost on the
retreat due to freezing temps and the crossing of each river. Trust
me---a graph can really make you catch your breath! Maybe numbers can
be fun.
Bob
On Tue, 20 Aug 2002 14:42:29 -0500 "Rick Lindsay"
<ROLindsay@Emeraldgrc.com> writes:
(snip)
>
> In geophysics -- a science whose data is frequently
> contaminated with noise -- we frequently use what is called
> an 'alpha trim mean', especially in data scaling. In that case,
> the 'alpha' parameter is similar to the 15% number you quote.
> In our case, using your numbers, the top 15% and the bottom
> 15% of the observations are not representative of realistic data;
> Therefore, they are 'trimmed off before the arithmatic mean is
> found. This explanation probably helps no one. Sorry.
>
> rick
/// or try http://www.team.net/cgi-bin/majorcool
/// Archives at http://www.team.net/archive
|