Frosty wrote:
> Statistics is also calculated guess work. You can make statistics give any
> answer you want.
>
>
> Charles F. Frost
> Frosty K5LBU
> frosty1@pdq.net
>
not really.
You can create misleading statistics or leave something out, but in
general, the whole point of statistics is to characterize something and
more important, to characterize the variability.
Take tower engineering... we talk about using metal of some particular
yield strength, but it's not like we actually test the metal being used
(if for no other reason than the test is destructive).. no, you rely on
the manufacturer of the material to have sufficiently good process
controls (using statistics) that their prediction of the strength of
what they ship you will be higher than the spec (most of the time).
If you're a bit more paranoid, you might do a coupon test (that is,
take a sample of what you got shipped and do an analysis on it).. but
that is still depending on statistics: you're using a sample to
represent the whole.
The concrete tests alluded to this morning are the same.. And there, the
advice was to use the fact that a sample is being done to "keep the
vendor honest"
I think the problem is that understanding how statistics work isn't
necessarily intuitive, nor is it really covered in school (although I
will say that both my children, ages 13 and 17, have had fairly decent
coverage of statistics in their math and science classes..), and, more
to the point, people who have the public ear (e.g. writers and
politicians; Mark Twain attributed the famous "lies, damn lie, and
statistics" to Benjamin Disraeli, one time prime minister of the UK)
tend NOT to have the background in statistics.
Add to this that statistics can be grossly misused, relying on the "it
has numbers so it must be right" sort of thing to bolster a
fundamentally unsound argument (correlation vs causality, and things
along the lines of 100% of murder victims were born, so birth causes murder)
_______________________________________________
_______________________________________________
TowerTalk mailing list
TowerTalk@contesting.com
http://lists.contesting.com/mailman/listinfo/towertalk
|