Content uploaded by Clark Mccauley
Author content
All content in this area was uploaded by Clark Mccauley on Nov 09, 2015
Content may be subject to copyright.
JOU R N
Al_
OF
Personality
Social
Psychology
Volume
37
May
1979
Number
5
The
Popularity
of
Conspiracy
Theories
of
Presidential
Assassination:
A
Bayesian Analysis
Clark McCauley
and
Susan Jacques
Bryn
Mawr College
Journalist
Tom
Bethell
has
advanced
the
hypothesis that conspiracy explana-
tions
of
Presidential assassination
are
popular because people have
an
irrational
need
to
explain
big and
important events with proportionately
big and
impor-
tant
causes. This
is a
species
of
consistency hypothesis
and
clearly predicts that
a
shot
that
kills
the
President
is
more likely than
a
miss
to be
attributed
to a
conspiracy. Four studies
are
reported that support this prediction. Three
of the
four
studies provided
a
check
on
whether conspiracy
was
overly
favored,
in the
case
of
successful assassination,
by
comparison with
the
normative Bayesian
formulation.
No
evidence
of
this kind
of
departure
from
rationality
was
found.
It
appears that people associate conspiracy with
successful
assassination,
not
because
of any
kind
of
special need
for
proportionality
of
cause
and
effect,
but
because
of a
belief
that conspiracies
are
more
effective
and
successful
than lone
assassins.
John
F.
Kennedy
was
shot dead
in
Dallas
in
1963,
and the
Warren Commission reported
in
1964 that
the
assassination
was the
work
of
a
lone
assassin,
Lee
Harvey Oswald.
In
1979,
the
issue
is
evidently
not yet
settled.
The
House Select Committee
on
Assassina-
tions
has
concluded
its
work, still uncertain
about
a
fourth
shot. Polls indicate
that
the
majority
of
Americans, around
80%
in
fact,
believe
that
others besides Oswald were
in-
volved
in the
assassination (Gallup,
1976).
Books
and
articles
theorizing
about
the as-
sassination
are
still appearing regularly.
Re-
cent examples
of the
genre include They've
Requests
for
reprints should
be
sent
to
Clark
Mc-
Cauley, Department
of
Psychology, Bryn
Mawr
Col-
lege, Bryn Mawr, Pennsylvania 19010.
Killed
the
President
by
Robert
Sam
Anson
(1975),
Coincidence
or
Conspiracy?
by
Ber-
nard
Fensterwald
(1977)
and the
Committee
to
Investigate Assassinations,
and
Legend:
The
Secret
World
of Lee
Harvey
Oswald
by
Edward
Jay
Epstein
(1978).
Clearly,
Ameri-
cans
are not
satisfied
with
the
conclusion
of
the
Warren Commission.
Whatever
the
merits
or
defects
of the
War-
ren
Commission report,
the
continuing popu-
larity
of
conspiracy theories
is
itself
a re-
markable fact. Most
events,
no
matter
how
traumatic,
do not
last
in the
public awareness
as the
Kennedy assassination has. Most
events,
no
matter
how
great, quickly drop
from
headlines
to
history. Accidents, scan-
dals,
great men,
and
even wars
are
left
be-
hind,
forgotten
or at
least much
faded.
But
Copyright 1979
by the
American
Psychological
Association,
Inc.
0022-3514/79/3705-0637$00.75
637
638
CLARK
McCAULEY
AND
SUSAN
JACQUES
new
books about John
F.
Kennedy's assassina-
tion
are
selling
IS
years
after
the
event, sell-
ing
in
supermarkets
and in
drug stores against
competition
from
pop
psychology
and sex
books.
It is the
premise
of
this article that
the
continued popularity
of
conspiracy the-
ories
of the
Kennedy assassination
is a
sur-
prising
social
fact
that
is
worthy
of
investiga-
tion.
The
studies reported here
are
aimed
at
test-
ing
one
hypothesis about
the
popularity
of
conspiracy theories, namely, that people
ir-
rationally
seek
big
causes
to
explain
big
effects.
According
to
this hypothesis,
ad-
vanced
by
journalist
Tom
Bethell
in The
Washington Monthly
(1975),
a
lone assassin
is
too
small
and
insignificant
a
cause
to
pro-
vide
a
satisfactory explanation
for
such large
effects
on
policy
and
people
as
follow
a
Presi-
dent's death.
We
are
expected
to
believe, according
to the of-
ficial
explanation,
that
the
Johnson
Administration
and
all
that
it
entailed, possibly including
the
debacle
of
Vietnam,
was set in
motion
by one man
who
had
quarreled with
his
wife;
who
had,
as it
were,
gotten
out of bed on the
wrong side
that
morning,
and
found
a gun
lying there.
The
cause
doesn't
fit the
effect.
But the
fact
is,
when
great
power
is
vested
in one
man,
as in the
President
of the
United
States,
it is
always
pos-
sible
that
a
small cause
(a
microbe
in his
blood,
for
example, leading
to a
fatal disease, leading
to
a new
President,
leading
to a
"Vietnam")
can
trigger
a
large
effect.
In
such cases many people will seek
a new
cause
that
is
commensurate
with
the
effect—seek,
in
other
words,
large
and
global
explanations that
thereby imbue
the
event with
appropriate
meaning.
In the
case
of the
Kennedy
assassination,
of
course,
this means looking
for a
conspiracy—preferably
a
large one. (Bethell, 197S,
p. 39)
BethelPs hypothesis
is
recognizable
as a
species
of
consistency hypothesis,
of
which
dissonance theory
is the
most prominent pre-
vious
example (Brown,
1965,
chapter 11).
A
need
for
consistency
of
cause
and
effect
clearly
implies
that
the
need
for a big
cause should
be
greater,
the
greater
the
effect
to be ex-
plained. Thus
the
Bethell hypothesis predicts
that
the
perceived probability
of a
conspiracy
should
be
much higher when
the
President
is
shot
at and
killed than when
he is
shot
at but
missed. Study
1 was
designed
to
test this
prediction.
Study
1
Method
Subjects. Subjects were
20
undergraduate students
of
Bryn Mawr
and
Haverford Colleges,
10
female
and 10
male.
Questionnaire.
The
questionnaire consisted
of two
pages.
On the top of
each page
was the
following
introduction:
"News
reports
of
violent events
and
their causes
are
sometimes surprising
and
sometimes
not. This study aims
to
measure your personal feel-
ing
about
the
likelihood
of
several stories
of
several
events."
Below
the
introduction
was
typed
a
head-
line
in
capital letters:
"A MAN
SHOOTS
AT THE
PRESI-
DENT
AND
MISSES"
or "A MAN
SHOOTS
AT THE
PRESI-
DENT
AND
KILLS
HIM."
On
each page,
after
the
head-
line,
the
same
two
questions appeared:
"What
is
the
probability that this
man is
acting alone
and
unaided?"
and
"What
is the
probability that this
man is
acting
as a
member
of a
group organized
to
kill
the
president?"
Procedure.
The
order
of the two
pages
was re-
versed
for
half
the
subjects.
Two
female experi-
menters each obtained
10
completed questionnaires.
Results
In
this
and
succeeding studies,
it is the
relative
probabilities associated with group
and
individual explanations
of
assassination
that
are of
interest. These relative probabil-
ities
are
naturally expressed
as
ratios,
and we
report
our
data
in the
form
of
median ratios.
Mean ratios
do not
properly represent
the
cen-
tral tendency
of
distributions
of
these ratios,
since
a few
ratios
in
every distribution
are
likely
to be
very extreme values. When most
subjects
are
giving
ratios such
as
2:1,
5:1,
or
1:4,
one
subject giving
a
ratio
of
100:1
can
make
the
mean ratio totally unrepre-
sentative. Thus, means
and
parametric sta-
tistics
are
inappropriate with
our
data,
and
we
use
medians
and
nonparametric statistics
instead.
The
data
of
Study
1 did not
appear
to de-
pend
on the
order
of the
pages
of the
ques-
tionnaire
or on the
experimenter,
so the
data
of
all 20
subjects were pooled
for
analysis.
The first
column
of
Table
1
shows that
the
median
odds
for
conspiracy were
1:1
when
the
President
had
been killed,
but
were
1:2
when
the
President
had
been missed.
That
is,
the
likelihood
of
conspiracy relative
to the
likelihood
of a
lone assassin
was
typically
seen
as
greater when
the
assassination
was
CONSPIRACY
THEORIES
OF
ASSASSINATION
639
successful.
We can use a
sign
test
to
test
the
significance
of
this tendency.
Of the 20
subjects
in
Study
1, 13
subjects gave odds
of
conspiracy higher when
the
President
was
killed
than when
he was
missed.
Six
subjects
indicated
no
difference
in
these odds,
and one
subject reported
the
reverse
difference
in
odds.
These
data
indicate
(p < .05 by
one-tailed
sign
test
for
correlated samples)
that
a
suc-
cessful
assassination
is
more likely
than
a
failure
to be
attributed
to a
conspiracy.
Discussion
Study
1
supported
the
prediction
that
in-
formation
about success
or
failure
of an as-
sassination attempt makes
a big
difference
in
the
popularity
of a
conspiracy explanation.
This
prediction came
from
the
general
hy-
pothesis that people need
a big
cause
to ex-
plain
a big
effect.
A
special need
for
con-
sistency
in the
size
of
cause
and
effect
is
not,
however,
directly demonstrated
by the
data
of
Study
1. For
instance,
it
might well
be
that
the
effect
of
information
about success
comes
rationally
from
a
judgment that groups
are
more
effective
and
likely
to
succeed than
individuals.
In
order
to
demonstrate
the hy-
pothetical consistency need,
it
must
be
shown
that people systematically exaggerate
the
probability
of a
conspiracy beyond what
the
news
of a
successful
assassination rationally
calls for. Clearly, this demonstration requires
a
formulation
of the
rational impact
of in-
formation,
and
Bayes' rule provides just this
normative
formulation.
In
probability
form,
Bayes'
rule requires
that
^(conspiracy/President
killed)
=
^(conspiracy)
p
(President killed/conspiracy)
/((President
killed)
'
Note
that
this formulation calls
for
revision
in
the
prior probability
of a
conspiracy
to the
extent
that
the
ratio reflecting
the
efficacy
of
a
conspiracy—p
(President
dead/conspiracy)/
/•(President
dead)—is
greater than 1.0. More
useful
for
present purposes
is
Bayes'
rule
in
odds form:
p
(conspiracy/President
killed)
/•(individual/President
killed)
_
^(conspiracy)
/•(individual)
p
(President killed/conspiracy)
/•(President
killed/individual)'
(1)
This
form
indicates
that
the
odds
favoring
a
conspiracy over
an
individual
assassin,
given
that
the
President
is
killed, need
to be re-
vised
and
increased over
the
prior
odds
favor-
ing
a
conspiracy
to the
extent
that
a
con-
spiracy
is
seen
as
more likely than
an
indi-
vidual
to
succeed
in
killing
the
President.
If
the
posterior odds
are
found
to be
systemati-
cally higher
than
called
for by the
prior odds
and the
efficacy
ratio, then
the
hypothesis
of
a
consistency need
to
explain
the
departure
from
rational prescription would
be
strongly
supported.
Studies
2 and 3
Method
Subjects.
The
subjects
of
Study
2
were
six
males
and six
females
recruited
individually
by a
female
experimenter
in the
environs
of the
Bryn
Mawr
train
station.
The 12
subjects
ranged
in
(estimated)
age
from
early
20s to
late 50s. Subjects
of
Study
3
were
15
males
and 9
females
recruited
by a
different
female
experimenter
in a
restaurant
and bar
near
Bryn
Mawr College. These subjects appeared
to be
in
their
20s and 30s and
were
generally approached
as
same-sex groups
(though
subjects
filled out the
questionnaire
without discussion with their
friends).
Questionnaire.
The
questionnaire
for
Studies
2
and 3 was
composed
of
four
pages,
two of
which
asked
questions much like
the two
questions used
in
Study
1. At the top of
these
two
pages
was the
instruction:
"Imagine
the
following
news headline":
There
followed,
on one
page, "MAN
SHOOTS,
KILLS
PRESIDENT,"
and on the
other page,
"MAN SHOOTS
AT
PRESIDENT,
MISSES."
On
both pages
the
succeeding
question
was the
same: "Which
is
more likely?
(A)
The man is
acting alone
and
unaided
or (B) the man
is
acting
as a
member
of an
organized group."
The
question
continued with
a
quantification:
"If you
checked
A, how
much
more
likely
is A?
(Twice
as
likely
as B?
Three times
as
likely? Five times?
Ten
times?)."
The
parallel alternative
was
also given:
"If
you
checked
B . .
.,"
etc.
The
prior odds
favoring
conspiracy were assessed
on
a
third page: "The next person
to try to
kill
the
President
will likely
be ...
(check
one)."
There
fol-
640
CLARK
McCAULEY
AND
SUSAN JACQUES
lowed
the
same
A
versus
B
choice
and the
same
quantification
of
that
choice
as
just described
for
the
first two
pages.
The
likelihood
ratio
reflecting
the
relative
efficacy
of
a
conspiracy
was
assessed
on a
fourth
page,
as
follows:
"Suppose
that
a man
acting
alone
and un-
aided
is
trying
to
kill
the
President.
Suppose
also
that
a man
acting
as a
member
of an
organized group
or
conspiracy
is
trying
to
kill
the
President.
Which
is
more likely
to
succeed
in
killing
the
President?"
There
followed
the
same
A
versus
B
choice
and
quantification
of
that
choice
as
already
described.
Procedure,
In
Study
2, the
prior-odds
question
was
always
on the
last
page,
and the 6
possible orders
of
the
other
three pages appeared twice each
to
form
the 12
questionnaires.
In
Study
3, all 24
pos-
sible orders
of the
four pages
appeared
once
to
make
up the
24
questionnaires.
Results
Columns
2 and 3 in
Table
1
show
that,
as
in
Study
1, the
conspiracy explanation
was
typically more popular when
the
President
was
killed than when
he was
missed (median
odds
of
2:1
vs. 1:3 for
Study
2, and
median
odds
of
2.5:1
vs.
1:3
for
Study
3).
This
result
is
confirmed
(p < .05 by
one-tailed sign
test
for
correlated samples)
by
noting
that
7
subjects
in
Study
2
judged conspiracy odds
higher when
the
President
was
killed than
when
he was
missed, whereas only
1
judged
the
reverse.
For
Study
3, the
corresponding
numbers were
13 and 2,
respectively (also
p
< .05 by
sign
test).
The two
additional questions
and the
odds
format
of all the
questions permit assessment
of
the
degree
to
which subjects exaggerate
the
chance
of
conspiracy.
For
each subject,
the
posterior odds
of
conspiracy given
that
the
President
was
killed were compared with
the
product
of the
prior odds
for
conspiracy
and
the
efficacy
ratio.
In
Study
2, 5
subjects gave
posterior odds
of
conspiracy higher than
re-
quired
by
Bayes' rule,
and
6
subjects gave
posterior odds
too
low.
In
Study
3, 8
subjects
gave posterior odds
too
high,
and 13
gave
posterior odds
too
low. Clearly, there
is no
evidence here
of
systematic departure
from
rationality; conspiracy
is not
overly
favored
when
the
President
is
killed.
Earlier
we
supposed
that
higher odds
for
conspiracy, given
successful
assassination,
might
be
rational
if
people believe
that
groups
are
more
effective
than lone
assassins.
That
supposition received some support
in
Studies
2
and 3,
since
the
median
efficacy
ratios
(third
row in
Table
1) of
2.5:1
and
2:1
indicate
that
groups
are
typically seen
as
more
likely than individuals
to
succeed
in an
attempt
to
kill
the
President.
In
Study
2, 9
subjects
thought groups were more likely
to
succeed,
2
subjects thought individuals were
more
likely
to
succeed,
and 1
subject thought
there
was no
difference
(p <
.05, one-tailed
binomial
test).
In
Study
3,14
subjects thought
groups
were more
effective,
and 10
subjects
thought
the
reverse
(p <
.27, one-tailed
bi-
nomial
test,
ns).
Thus, Study
2, but not
Study
3,
shows
significantly
more than
half
the
sub-
jects judging groups
as
more likely
than
indi-
viduals
to
succeed
in an
attempt
to
kill
the
President.
Study
4
It
appears that Studies
2 and 3
have shown
that
the
odds
for
conspiracy
are not
exag-
gerated, compared
to the
perception
of the
prior
odds
of
conspiracy
and the
diagnosticity
of
the
news
that
the
President
has
been killed.
Before
accepting this conclusion, however,
there
is a flaw to be
considered
in the
ques-
tionnaire
used
in
Studies
2 and 3.
That
is,
the aim was to ask
about prior
and
posterior
odds
of
conspiracy that
differ
only
in
giving
the
information
that
the
President
was
killed
in
the
posterior assessment. Unfortunately,
the
difference
between
our
prior ("The next
person
to try to
kill
the
President
. .
.")
and
our
posterior ("Man shoots, kills President")
question
is two
pieces
of
information:
that
the
attempt
got as far as
getting
a
shot
off
and
that
the
shot
was
successful.
Our ef-
ficacy
ratio,
on the
other hand, assessed only
the
information
value
of
success
and not the
information
value
of
getting
a
shot off.
In
order
to be
sure that this
confounding
is not
important
to our
results,
we
revised
the
ques-
tionnaire
and
used
it in a new
study.
Method
Subjects. Subjects were
IS
males
and 9
females
recruited
at a
shopping center near
Bryn
Mawr
Col-
lege
by the
same female experimenter
who ran
Study
3. The
subjects ranged
in
(estimated)
age
from
early
20s to
late
50s.
CONSPIRACY
THEORIES
OF
ASSASSINATION
641
Table
1
Median
Probability
Ratio
Associated
With
Conspiracy
(Group)
Versus
Lone Assassin
(Individual}
Explanations
of
Presidential Assassination
Probability
p
(group/President
killed)0
£
(individual/President killed)
p
(group/President
missed)"
^(individual/President
missed)
p
(President killed/group
try)b
/((President
killed/individual try)
p
(group try)
p
(individual try)
Study
1
Study
2
(n
= 20)
(n
= 12)
1:1 2:1
1:2
1:3
2.5:1
1:2
Study
3
(«
= 24)
2.5:1
1:3
2:1
2:1
Study
4
(n
= 24)
2:1
1:2
3:1
1:1
• In
each
of the
four
studies,
a
conspiracy
was
judged more likely when
the
President
was
killed than when
the
President
was
missed
(p < .05 by
one-tailed sign
test
for
correlated
samples).
b
In
Studies
2 and 4, but not in
Study
3,
more than
half
the
subjects judged
a
group more likely
to
succeed
than
an
individual
(p < .05 by
one-tailed
binomial
test).
Questionnaire.
The
questionnaire
was
identical
to
that used
in
Studies
2 and 3
except
for the
wording
of
the
prior-odds question
and the
relative
efficacy
question.
Whereas
the
previous questionnaire asked
about
"The
next
person
to try to
kill
the
President
. . .
,"
the
revision asked, "The
next
person
to
shoot
at the
President will
likely
be. . .
."
And
whereas
the
previous
form
asked about
the
success
expected
of
an
individual
or
group "trying
to
kill
the
President,"
the
revision went
as
follows:
"Suppose that
a man
acting
alone
and
unaided gets
a
shot
at the
Presi-
dent. Suppose also that
a man
acting
as a
member
of
an
organized group gets
a
shot
at the
President.
Which
is
more
likely
to
succeed
in
shooting
and
kill-
ing
the
President?"
Procedure.
The
procedure
was the
same
as in
Studies
2 and
3.
Results
Column
4 of
Table
1
shows
that,
as in
Studies 1-3,
the
conspiracy explanation
was
more
favored when
the
President
was
killed
than when
he was
missed (median
odds
of
2:1
vs.
1:2).
This
result
is
confirmed
by
noting
that
13
subjects judged conspiracy
more likely when
the
President
was
killed
than when
he was
missed, whereas only
4
subjects judged
the
reverse
(p < .05 by
one-
tailed sign
test
for
correlated
samples).
As
in
Studies
2 and 3, an
analysis
at the
level
of the
individual compared
the
posterior
odds
of
conspiracy with
the
product
of the
prior odds
for
conspiracy
and the
likelihood
ratio giving
the
relative
efficacy
of
conspiracy.
Nine subjects gave posterior odds
of
con-
spiracy higher than required
by
Bayes' rule,
and
IS
subjects gave posterior odds
too
low.
In
short,
the
results
of
Study
4 are
like those
obtained with
the flawed
questionnaire
in
Studies
2 and 3.
Studies
2-4
are
consistent
in
finding
no
systematic exaggeration, compared
to the
Bayesian prescription,
of the
probabil-
ity of a
conspiracy when
the
President
is
killed.
Likewise,
the
tendency
to see
groups
as
more
effective
than individuals
is
confirmed
in
Study
4. The
efficacy
ratios judged
by
sub-
jects
in
Study
4
were
for the
case
of an as-
sassin
who had
gotten
as far as a
shot
at
the
President:
the
relative probability
of
killing
the
President
for a
member
of an
organized
group
versus
an
individual acting alone
and
unaided.
The
median
efficacy
ratio (third
row
in
Table
1)
of
3:1
indicates
that
group
members
are
typically seen
as
better shots.
In
Study
4,
19
subjects thought group
members were more
effective,
and 5
subjects
thought
the
reverse
(p <
.05,
one-tailed
bi-
nomial
test).
Thus,
Study
4
shows signifi-
cantly more than half
the
subjects judging
groups
as
more likely than individuals
to
succeed
in
killing
the
President, once having
gotten
off
a
shot.
"Pure"
Data
From Studies
1-4
The
within-subjects
design
of the
present
studies, where each subject answered
two
642
CLARK
McCAULEY
AND
SUSAN JACQUES
(Study
1) or
four
(Studies
2-4)
questions,
raises
the
possibility
of
sensitization
effects.
Perhaps
subjects answer
a
question
differently
because
of
biases
or
hypotheses engendered
by
having already
answered
previous ques-
tions.
In
order
to be
sure that
our
results
do
not
suffer
from
sensitization problems,
we
brought
together
"pure"
data corresponding
to the first
three rows
of
Table
1,
that
is, for
the
ratios about which
we
have made claims
by
statistical
test.
The
"pure"
data
are
from
only
the first
pages
of
questionnaires,
that
is,
are
responses
to
questions
by
subjects
who
had not
previously answered other questions.
These data cannot therefore
be
prejudiced
by
any
kind
of
sensitization
effect.
Putting together
"pure"
data
from
Studies
1-4,
we had 26
judgments
of the
odds
for
conspiracy given success
(first
row in
Table
1)
and 26
judgments
of the
odds
for
con-
spiracy given
failure
(second
row in
Table
1).
The
median odds
for
conspiracy when
the
President
was
killed
were
2:1;
they were
1:2
when
the
President
was
missed.
These
two
"pure"
medians
are
numerically quite
close
to the
medians
for all
data
in
rows
1
and 2
(respectively)
of
Table
1,
indicating
that sensitization
effects
were
not a
problem
for
these data. Further, these
two
"pure"
medians
are
significantly
different
(p < .05
by
one-tailed
median test)
and
clearly
in-
dicate
that
in a
cross-groups design,
we
still
find
that
the
odds
for
conspiracy
are
higher
when
assassination
is
successful
than when
it
fails.
Similarly,
we put
together
data
from
Stud-
ies
2-4 to get a
total
of
IS
"pure"
judgments
of
the
efficacy
ratio (row
3 in
Table
1).
The
median
of
these
15
efficacy
ratios
was
2:1,
which
is
numerically close
to the
medians
for
all
data
in row 3 of
Table
1.
Unfortunately,
it is not
true that significantly more than
half
our 15
"pure"
subjects
saw
groups
as
"more
effective than individuals
(in
fact, only
8 of 15
did).
Still,
we
believe