Abstract
A
central
issue
confronting
both
philosophers
and
practitioners
in
formulating
an
analysis
of
causation
is the
question
of
what
constitutes
evidence
for
a
causal
association.
From
the
1950s
onward,
the biostatistician
Jerome
Cornfield
put
himself
at
the
center
of
a controversial
debate
over
whether cigarette
smoking
was
a
causative
factor
in
the
incidence
of
lung
cancer.
Despite
criticisms
from
distinguished
statisticians
such
as
Fisher,
Berkson
and
Neyman,
Cornfield
argued
that
a
review
of
the scientific
evidence
supported
the
conclusion
of
a
causal
association.
Cornfield's
odds
ratio
in
case‐control
studies — as
a
good
estimate
of
relative
risk — together
with
his
argument
of
''explanatory
common
cause''
became
important
tools
to
use
in
confronting
the
skeptics.
In
this
paper,
I
revisit
this important
historical
episode
as
recorded
in
the
Journal
of
National
Cancer
Institute
and
the
Journal
of the
American Statistical
Association.
More
specifically,
I
examine
Cornfield's
necessary
condition
on
the minimum
magnitudes
of
relative
risk
in
light
of
confounders.
This
episode
yields
important
insight
into the
nature
of
causal
inference
by
showing
the
sorts
of
evidence
appealed
to
by practitioners
in supporting
claims
of
causal
association.
I
discuss
this
event
in
light
of
the
manipulationist
account
of causation.