Jerry Dallal wrote:
>
> Alan McLean ([EMAIL PROTECTED]) wrote:
>
> : There is certainly an argument that when trialling a new treatment (I
> : initially used the word 'testing' here, but figure that it may be
> : confused with the statistical test of the resultant data) it is
> : presumably expected to work. Consequently, if a person in the trial is
> : given a placebo, there is a clear expectation that he or she is being
> : disadvantaged - given an inferior treatment.
>
> Placebo controlled trials are unethical and often illegal when placebo
> means withholding standard medical care.
>
> : On the other hand, if a placebo is not used, the results of the trial
> : will be unclear. This will presumably disadvantage Society. The ethical
> : choice is then between disadvantaging a number of individuals by giving
> : them a treatment which is expected to be inadequate (rather than a
> : treatment which is expected to be better - but may not be!) and
> : disadvantaging society by reducing the increase in knowledge - which is
> : expected to advantage many people in the future.
>
> Controls need not be placebos. The usual control is standard medical
> care. Standard medical care changes. For example, it olden times (like
> 10 years ago) you could do a placebo controlled trial in subjects with
> cholesterols of 240 mg/dl. Today, they get referred to their physicians!
Hi Jerry,
I was using 'placebo' as an example - although not a medical researcher,
I realise that the control is in some sense or other the 'not different'
treatment. My point however is still true - that the person who receives
the control treatment is presumably getting an inferior treatment. You
certainly don't test a new treatment if you think it is worse than
nothing, or worse than current treatments!
As far as I can make out, this is in line with what you were saying
anyway.
>
> : This is certainly an ethical question (though I might argue that neither
> : choice is unethical if the choice is made ethically!) But I don't see
> : how the type of statistical test done in analysing the resultant data
> : can be ethical or not.
>
> A one-tailed test (as usually proposed) presumes that the difference, if
> there is one, can only be it a specified direction.
> This violates the principle of equipoise.
>
I am not sure what you mean by the principle of equipoise. However, my
statement that the lack of ethics, if it is there, is in giving the
control treatment to a person, not in the statistical test, remains
true.
A one tailed test does not 'presume a difference can only be in a
specified direction'. What it does is to consider only differences in
one direction to be significant. If I am trialling a new treatment, and
I measure some variable such that the mean mu of that variable is
positive if the treatment 'works'. By 'works' I mean that this new
treatment is better than past ones, or that it does something different
from current ones, or whatever is relevant. If there is a control group,
mu will be the difference in means.
For the sample data I compute xbar (the difference of sample means if
there is a control group). There are three possibilities.
1. xbar is negative
2. xbar is positive, but small
3. xbar is positive, and large
Given that we would normally only trial a treatment if we expected it to
improve things, either 2 or 3 is likely to be the result. But 1 can
happen!
If 1 does happen, we would conclude either that the new treatment is no
better than the control, and may be worse. In either case we junk the
new treatment.
If 2 happens, we would conclude that the new treatment is no better than
the control (so we might as well stick with current practice - and junk
the new treatment.
If 3 happens, we would conclude that the new treatment is better than
the control - so it might well replace current treatments (maybe after a
lot more testing....)
Nowhere in this statistical testing is there an ethical issue. (Except,
if you like, that it is ethically correct to require the new treatment
to be shown to be better than the current treatment before it is
accepted - that is, take the 'no better' as the null hypothesis.)
Regards,
Alan
=================================================================
--
Alan McLean ([EMAIL PROTECTED])
Department of Econometrics and Business Statistics
Monash University, Caulfield Campus, Melbourne
Tel: +61 03 9903 2102 Fax: +61 03 9903 2007
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
