Re: A letter to a statistician. Emailed and airmailed and no reply


"Ray Johnstone" <ray@xxxxxxxxxxxx> wrote in message
news:emom03di9v20dgu9t9f313biisldniotj2@xxxxxxxxxx
http://members.iinet.com.au/~ray/DeVeaux.htm

Professor R.D. De Vaux,

Bronfman Science Center,

Williams College,

Williamstown, MA01267

Dear Professor De Veaux,

In your book "Intro Stats", you say (p126) "Ironically, the proof
that smoking indeed is the cause of many cancers came from
experiments conducted following the principles of experimental
design and analysis that Fisher himself developed - and that
we'll see in Chapter 13." Chapter 13 is about controlled
intervention trials. Seven such trials have been conducted
to determine the association between smoking and health, with
a hundred thousand test and control subjects followed for seven
years - Whitehall, MRFIT, Goteburg, Finnish Businessmen, Oslo,
WHO collaborative and North Karelia. (References are available
on my website.) The results of all are uniform, forthright and
unequivocal: they show no association between smoking and life
expectancy, deaths from cancer or deaths from any other cause.
The "proof" that smoking causes cancer and other diseases comes
from trials of the kind Fisher had condemned - uncontrolled,
nonrandomised, nonintervention trials such as the British Doctors
and American Cancer Society studies: it is these which have been
used to calculate the number of deaths supposedly caused by
smoking.

For real irony read the words of A.B. Hill, a principal author
of the British Doctors study. On p250 of his Principles of Medical
Statistics" (1971) he uses the words "inferior" and "second-best"
to describe such work. But perhaps that is more hypocrisy than
irony.

This is surely a unique episode. So far as I am aware,
nowhere in medical science are the results of controlled
intervention trials rejected in favour of uncontrolled
non-intervention trials. But it is more than unique.
It is scandalous.

It is clear that Fisher loved the science of statistics.
He would have been appalled to witness a generation of
statisticians degrade that science.

See this article on Fisher by Sir Walter Bodmer who was a student of his and
was later Director of Research at the Imperial Cancer Research Fund.

http://ije.oxfordjournals.org/cgi/content/full/32/6/938#SEC6

On "Smoking and Lung Cancer"

"It is clear from the correspondence starting towards the end of 1958, that
the relationship between Fisher and Bradford Hill soured as a result of
Bradford Hill and Doll's studies on the relationship between smoking and
lung cancer. Fisher's attack on Bradford Hill arose not from any intrinsic
animosity, indeed quite the opposite, but at least in part because of his
libertarian views as applied to the then smoking and lung cancer
controversy. He himself was an inveterate pipe smoker, but did not, to my
knowledge, smoke cigarettes.

Fisher's antagonistic response to the data on the relationship between
smoking and lung cancer was influenced by two strong beliefs. The first, and
perhaps foremost, came from his firmly developed views on the nature of
statistical and scientific inference. Correlation must not be taken as proof
of causation. He thus sought common causes such as genetic factors, but
without at the time really being aware of the strength of the association
and the unlikelihood of it being due to a common cause rather than a direct
effect. His second response was based on his libertarian views. He believed
strongly that people should be left freely to come to their own conclusions
and simply be given the data. He was strongly against the publicity that was
put out about the dangers of smoking. In an article for the Centennial
Review published in September 1959,14 he talks about:
.... an annotation published by the British Medical Association's journal,
leading up to the almost shrill conclusion that it was necessary that every
device of modern publicity should be employed to bring home to the world at
large this terrible danger.
He was perfectly happy for individuals to make their own decisions that they
should consider giving up smoking, but this in the absence of what he
considered convincing data of a causal effect. Many have accused Fisher of a
conflict of interest in these arguments because in 1956 he had already
accepted the invitation of the tobacco manufacturers standing committee to
be their scientific consultant. However, as Joan Box (Fisher's daughter)
points out in her biography, Fisher was naive enough to believe that he was
free to give advice in whatever way he wanted and that receiving payment
for so doing should not influence the nature of this advice. "

See also "When Genius Errs: R. A. Fisher and the Lung Cancer Controversy" by
Paul D. Stolley

http://aje.oxfordjournals.org/cgi/content/abstract/133/5/416

More importantly

"editorial To smoke or not to smoke?" in The Cancer Journal Vol 5 No 2 Mar -
Apr 1992

http://www.tribunes.com/tribune/edito/5-2z.htm

"While hardly anybody doubts that smoking is hazardous to our health, it is
not at all clear whether this can be proven by epidemiological arguments.
This issue that splits the epidemiological community was recently raised in
its prestigious journal (1-4). Does smoking cause lung cancer? "Of cause it
does not! " exclaimed the late R.A. Fisher, one of the greatest
statisticians of our century, creator of statistical genetics, analysis of
variance and maximum likelihood. The observation that lung cancer is
associated or correlated with smoking can be interpreted in two ways. Either
smoking causes lung cancer, or lung cancer causes smoking. In other words,
patients carrying lung cancer may tend to smoke more than others. One ought
to also consider the person's constitution. For some, smoking may be
dangerous, while others may benefit from it. Fisher was particularly
suspicious of case-control studies and refused to accept their evidence.

Half a century later it appears as if the issue has been finally proven, and
one wonders what were the reasons for Fisher's error. "When Genius Errs",
exclaims Stolley in his commentary on Fisher, attributing failure to his
personality. A "political conservative", "taking fee from the tobacco
industry", "holding certain ideologic objections to mass public health
campaigns" etc. Finally, Stolley concludes that "a genius can lack insight
and fail to develop a sense of proportion about evidence", failing to
realize that this may also be applicable to himself.

"Were we really wrong?" replies Eysenck apologetically (3). Should we
abandon Hume's philosophy according to which proof of causation in the
absolute sense is impossible? Referring to epidemiological studies on the
relationship between smoking and lung cancer, Eysenck says: "We did not
demand absolute proof... What we have found are serious methodological
weaknesses in the design of these studies quoted in support of these
theories, statistical errors, and unsubstantiated extrapolations."
"Epidemiology is a complex discipline", "Risk factors interact in complex
ways and univariate analysis is quite inappropriate to the elucidation of
causality".

As in the previously described "Cancer Wars" (5), eminent epidemiologists
contradict each other on the interpretation of their observations. If one is
ready to adopt case-control studies, one may prove that "smoking causes lung
cancer". Yet with case-control design it is possible to prove virtually
anything, since the design uses retrospective observations that cannot be
randomized. More rigorous statistical methods are less conclusive.

It is not epidemiology that is complex, as conceived by Eysenck, but the
human organism. Cancer and smoking are complex processes evolving with time.
Instead of considering their entire complexity, epidemiology reduces them to
two observations, cancer mortality and smoking frequency. When confronted
with each other they are correlated. This is where epidemiology should stop,
since the interpretation of this relationship cannot be done with
epidemiological reasoning. Either smoking initiates cancer or cancer
triggers an urge to smoke, or the two may never interact and their observed
association results from a third process that was not considered in the
study. Epidemiology lacks the means to distinguish between the three and
should leave the decision to medicine. Only medicine is capable of
interpreting complex biological processes and their interaction. Instead,
epidemiology borrows medical reasoning, applying it to two arbitrary points
of a complex process, and finds this practice even justified to denounce its
geniuses. _Fisher did not err! He realized the limitation of the method that
he created._"

So much for these issues being hidden under the carpet. You are clearly a
monomaniac and that is probably why you don't get a response.

Nick

PS You posted this to sci.math in August 2006.

http://mathforum.org/kb/thread.jspa?threadID=1439301&messageID=5067984

Rather surprisingly you don't give the date of the letter.


.

Quantcast