[H] the campaign to discredit homoeopathy

[Luise Kunkle]

Well...

In order for statistics to mean anything, the goal of what they are
supposed to show has to be clearly defined.

Our goal is to show that homeopathy, more specifically the use of
diluted/potentized remedies, will show results above placebo.

None of the trials I have read - and I have read a stack about 10
inches high - are designed to show that. They are designed to show
that a specific remedy X will be better than placebo in curing
condition Y. Sometimes, but rarely, the prescriber can
choose between more than one pre-determined remedie, say 2 - 4. An
example might be: Bell (or possibly Bell, Apis or Phyt) for
tonsillitis). This, to start out with, a very bad base on which to
perscribe single remedies - probably the reason why almost all
remedies that proved significantly better than placebeo (e.g.
Traumeel) were complexes. One of the very few trials that showed
significantly better for the "homeopathic remedy" was a migraine trial
in Italy, where the homeopath had quite a few remedies to choose from.

Actually, all trials but one showed the verum remedy superior or at
least equal to placebo, by far the most showed superior - but not to
an extent that made it statistically significant.

The reason why they did not show "significantly" superior is very
simple - it is the smallness of the samples!! (A few days ago, I
presented the matter to my son , who is a mathematician with just
basic courses in statistics. He told me that with such small samples
there is hardly ever any chance to prove anything statistically
significant - something that in a MATH course on statistics is one of
the basics they learn - it seems that in the statistics courses taken
for other majors this is not impressed upon the students - or they
tend to forget it)

In small samples the "margin of error" is just too big!

For a sample of 10 it is 35 % of the sample size - i. e. for a verum
to show significantly better, the difference between it and placebo
would have to be at least 4 patients, i. e. the verum outcome would
have to be 5 as compared to only one for placebo or 4 for verum
compared to none for placebo.

For a sample of 20 the margin of error would still be roughly 23% (of
20, i. e. the outcome would have to be e. g. at least 9 patients cured
by verum compared to 5 placebo at most, or 8 verum to 4 placebo etc.

With a sample of hundred, things start looking better, there the
margin of error is 10%, i. e. 90 verum to 80 placebo.

(At thousand it is only 3% margin of error).

Now we come to the meta studies.

In those, the outcomes of the individual trials were considered
without really taking into account this basic law of statistics, i. e.
that those trials just could not answer the question of whether
homeopathy as a whole is superior to placebo as a whole!

In order to do that, one has to add up the numbers of patients that
were cured by homeopathy of the various conditions and compare it to
the number of patients that were cured by placebo of those various
conditions. This way the sample gets to be much bigger and the margin
of error for the big sample would apply.

When I studied those trials I found out that by using this method the
outcome of verum against placebo was very, very, very.. highly
significantly better!

According to what I found out from the two people I mentioned (the
statistics professor and the medical professor - both, as I said,
absolutely skeptic of homeopathy) this can be done and is often done
in statistics, e.g.

Testing light bulbs ( those are often used to demonstrate in stat) of
whether there are more "duds" in brand A or Brand B, at least 100 of
them would have to be tested to make sense, the best sample is 1000.
Those 1000 test need not be done at the same place at the same time,
but can be spread out over time and place using say 50 lightbulbs each
occaasion. The margin of error is then taken for the entire 1000
bulbs.

Of course, what I have said above is not acceptable to journals like
the Lancet, coming from a lay person and being expressed in lay terms.
In order to get them to publish it, it would have to be an
expertise from a prestigious Statistician. This is why I have been
urging again and again, on the various lists, that the homeopaths
(societies or what ever) spend the money to get one.

I do appreciate that they would also be leery of trusting a lay person
like myself enough to spend the money on my say-so. I do not see,
however, why they cannot spend just a few bucks to get the opinion of
a math student majoring in statistics on a masters level, just to find
out whether I am right.

Of course that would not get rid of all the objections by the
scientific community as far as design etc. of the individual trials is
concerned. Those of really bad design might have to be weeded out. It
would also not get rid of the objectiion based on "publicity bias" -
meaning that many non-successful trials may never have been published.
But it would give as a much better starting point for our arguments.

IMO

Regards

Luise
--
One thought to all who, free of doubt,
So definitely know what's true:
2 and 2 is 22 -
and 2 times 2 is 2:-)
==========> ICQ yinyang 96391801 <==========

[Shannon Nelson]

Thanks Luise, I'm very happy to have this information; it does put
some things in a much better perspective. So basically, a small trial
needs a *much* bigger positive effect in order to prove *any* effect.
It makes sense, tho I'd never thought of it... (Having pretty
"successfully" avoided learning anything about statistics thus far...
I'll try to change that.)
Shannon