@Jman8 This might help:
http://object.cato.org/sites/cato.org/files/serials/files/regulation/2000/4/fortherecord.pdf
"Our regulation article “lies,
Damned Lies, & 400,000 Smoking-
Related Deaths” exposes the pseudoscientific,
antismoking emissions of the
Centers for Disease Control and Prevention
(cdc). For that service, we stand
accused by Ms. Lukachko and Dr. Whelan
of “straying from basic epidemiological
principles” and “touting opinions
that masquerade as facts.” In
response, we examine Lukachko and
Whelan’s four specific charges, then
offer some concluding comments.
RELATIVE RISK
Lukachko and Whelan claim that
we erroneously omit certain diseases on
the ground that relative risks of less
than 2 (a 100 percent increase) are
“insufficiently reliable to conclude that
a particular agent (e.g., tobacco) caused
a particular disease.” Well, consider
this cautionary statement: “Relative
risks less than 2 are considered
small.… Such increases may be due to
chance, statistical bias, or effects of
confounding factors that are sometimes
not evident.” That statement
comes not from us, but from a 1994
release by the National Cancer Institute
(nci), referring to a study of abortion
and breast cancer...
...nci goes to great lengths to
dispute the potentially harmful effects
of abortions while it trumpets the
harmful effects of cigarettes, applying
equally dubious evidence.
But do not take our word. A special
report from Science magazine illuminates
the real world of epidemiology
in practice. From Gary Taubses’s article,
“Epidemiology Faces Its Limits” (Science
269, July 14, 1995: 164), here is what
respected scientists from both the public
and private sector, within the United
States and without, have said about
low relative risks.
•Sir Richard Doll of Oxford University:
“No single epidemiological study is
persuasive unless one can be statistically
confident of at least a threefold
increase in risk.”
• Harvard researcher Dimitrio Trichopolous:
“A fourfold risk increase is
the lower limit.”
•Marcia Angell, editor of the New England
Journal of Medicine: “As a general
rule of thumb, we are looking for a relative
risk of three or more before
accepting a paper for publication.”
• Robert Temple, director of drug
evaluation at the U.S. Food and Drug
Administration: “If the relative risk
isn’t at least three or four, forget it.”
• And from interviews conducted by
Science magazine: “Most epidemiologists…
said they would not take seriously
a single study reporting a new potential
cause of cancer unless… exposure
to the agent in question increased a person’s
risk by at least a factor of three.”
Even then, interviewees warn, “skepticism
is in order.”
Yet Lukachko and Whelan assure
us that “a relative risk less than 2,
although small, can indeed be statistically
significant.” Well, yes, that is certainly
true—and completely irrelevant,
as even they concede. Statistical significance
measures chance error,
which depends in part on risk levels
for smokers and nonsmokers, and in
part on sample size.
All else equal, given a specified background
risk among nonsmokers, the
smaller the relative risk, the less likely
that the difference in risk between
smokers and nonsmokers is statistically
significant. Still, large samples can
produce statistically significant results
even when the relative risk is low. But
that is not the point at all.
A relative risk less than 2 means that
it is less probable, though not impossible,
that a relationship is statistically
significant.
Low relative risk may indicate that a
study did not adequately control for
confounding variables [
see Carl's piece]
.... or that it was
affected by bias on the part of
researchers or participants. Thus, statistical
significance is necessary to
demonstrate that a study is valid—it
denotes low probability of sampling
error—but it is not sufficient. The
potential problems associated with confounders
and bias do not disappear
merely because of a large sample size.
Lukachko and Whelan understand
that concept quite well. As they correctly
state, “Even if a result is statistically
significant, bias and potential
confounders must be addressed to
demonstrate a valid association.”
Astonishingly, having acknowledged
that principle, they wholly disregard
its implications. Relative risk is an
indicator not only of statistical significance
but also of possible confounders
and bias. That is why epidemiologists
uniformly hold that a
low relative risk goes hand in hand
with suspect validity.
Inexplicably, Lukachko and Whelan
persist in arguing that a small relative
risk can represent a serious threat to
public health. In support, they point to
the relationship between smoking and
heart disease. The relative risk of smoking
for many types of heart disease is
less than 2. But heart disease kills many
more people than lung cancer. Therefore,
according to Lukachko and Whelan,
“the number of smoking-related
deaths from heart disease rivals those
from lung cancer.” Verbal gymnastics,
but manifestly untrue. The missing link
is obvious: low relative risks mean that
deaths from various types of heart disease
have not been shown to be smoking-
related. To characterize those
deaths as “smoking-related” simply begs
the question."
There's more than that of course....