Re: resampling methods are serious procedure?

On Feb 3, 2:00 pm, illywhacker <illywac...@xxxxxxxxx> wrote:
On Feb 3, 6:38 pm, Old Mac User <chendrixst...@xxxxxxxxx> wrote:

When I mentioned "sampling" I meant sampling.

I am not attacking you personally, and I do not understand why the
tone of your post is hostile; obviously I hit a button. In fact, you
never used the word 'sampling', only 're-sampling', and you first
mentioned it right before you described sampling. You never described
re-sampling. Perhaps you can see why I was confused.

I take exception to saying they "should never be taught at all".

Why would you take exception? I am sure you are an excellent teacher.
However, you are teaching an approach to statistics, however
implicitly, and whether you use the words or not, that is wrong. Take,
for example, confidence intervals. People can never explain them
because they are nonsensical: theoretically wrong and practically
incorrect, even incoherent.

illywhacker;

Illywhacker,

I apologize for my hostile tone. I thought I casually mentioned
sampling early on. Aside from that, the concept of "sampling" a
physical material or "sampling" a population of people is something we
(or at least I) tend to take for granted. That's usually a mistake.
You never know.

Yes I can see why you were confused. My bad, and I'm sorry for that.

Well, I'm still not lined up with you re: confidence intervals.
I profess they can be explained and they do purposes. I've watched
statisticians wax and wane over this one. For a while we reported
comparative data as "there really is a difference" (i.e.,
"significant") or not. That left our clients with a sense of "yes or
no" when we know it's not that simple. Confidence intervals at least
give folks a sense of "how good is this estimate".

The correct expression for describing a confidence interval is... I
admit... ponderous. "95% of the intervals computed in this fashion
will encompass (or cover, as some day) the true difference. 5% will
fail to do so. So the odds are 10/20 that the interval we have
reported does in fact encompass the true difference of averages" etc.
That really is ponderous, though it is accurate.

I'm going to search for a link (stored somewhere in my computer"
titled something like "Is Statistics Difficult". I'll post that when/
if I find it. The author goes into the matter of inference and how
difficult it is to teach and to learn.

My fascination with words stems in part from dealing with certain
legal matters. Sitting in the big oak chair with a steely-eyed judge
and 12 of our peers (who don't know a thing about statistics,
engineering, or law) is a challenging experience. As one lawyer I
worked for put it... "Don't every use an equation in front of a jury.
Someone in the jury will think 'you are starting to sound like my old
algebra teacher... you know... the one who flunked me.' " He was
right. Juries are much more convinced by depth of your conviction
than by the height of your knowledge. Yes, I've explained
"significance" and "confidence intervals" and things more esoteric
than those in that environment. "Words, like parts, should fit with
precision".

So I profess to you that confidence intervals (1) have a valid purpose
and (2) they can be explained.


.