What is statistical inference?

[Truman Prevatt <truman.prevatt@netsrq.com>]

There has been a lot of gnashing of teeth as of late on RC about
"sceintific studies" proving this or that, call it X.  Then there are
people that reply that "it ain't true cuze my rat does well on the
opposite of X.  Rice brand is the latest but not the only issue by far.
There is the rider weight issue, as another one.  

Before we all get our panties in a twist we should sit back and
understand just what these scientific studies are and are not.  

First and formost you can not prove the truth of a hypothesis with a
statisticial analysis of any data - no matter how well the experiment
was designed.  At best you can disprove a hypothesis.  Even it
mathematically precise sciences such as physics, we keep learning at our
laws really are not laws only hold in special cases.  They are at best
theories.

What you can say is that with a confidence of (say) 95% X is true.  Say
X is horses will gain more weight supplementing with a cup of corn oil
than they will supplementing with a pound of rice brand.  What does that
mean.  That basically means that if you take a random sample of 100
horses, then 95 will gain more weight on oil than rice brand.  It does
not say every horse will.  Nor does it say any endurance horses will. 
If one wants to test this hypothesis for endurance horses, one needs to
run the experiment with random samples of endurance horses.  

This still would not say that X is true for the consistent top 10
endurance horses.  One again would have to use random samples from
consistent top 10 endurance horses.

If the original experiments were done for a random sample under a fixed
set of conditions, it does not that X is true for a random sample under
different conditions.  For example the environment of the dry climates
of the U.S. west coast are different than those of the southeastern
U.S.  In order to say X for the Southeast then the experiment would have
to be performed using random samples of horses in the Southeast.

There is no magic in the scientific method - and there is a lot of
work.  One problem is that I for one would like to see the conditions
under which the experiments were done and the confidences intervals
used.  I would also like to see people lighten up and be less
defensive.  If the confidence interval is 80% for X then it is 20% for
not X.  Inferences using statistical methods are not absolute, but they
do provide valuable information.  

Just remember there are people that still live to 110 smoking two packs
a day.  Does that mean the studies that link cancer, heart attacks, etc
to smoking are false.  No, absolutely not.  What they do say is out of a
random sample of 100, 97 will die early. That is all.

Truman


-- 
Truman Prevatt
Mystic “The Horse from Hell” Storm with one in the oven
Rocket a.k.a. Mr. Misty
Jordy a.k.a. Bridger (when he is good)
Danson Flame - hey dad I'm well now and ready to go!

Brooksville, FL