Sources of the Continued Influence Effect: When Misinformation in Memory Affects Later Inferences

Abstract

Several lines of research have found that information previously encoded into memory can influence inferences and judgments, even when more recent information discredits it. Previous theories have attributed this to difficulties in editing memory: failing to successfully trace out and alter inferences or explanations generated before a correction. However, in Exps 1A and 1B, Ss who had received an immediate correction made as many inferences based on misinformation as Ss who had received the correction later in the account (despite presumably having made more inferences requiring editing). In a 2nd experiment, the availability of the misinformation within the comprehension context was tested. Results showed that Ss continued to make inferences involving discredited information when it afforded causal structure, but not when only incidentally mentioned or primed during an intervening task. Exps 3A and 3B found that providing a plausible causal alternative, rather than simply negating misinformation, mitigated 1 effect. The findings suggest that misinformation can still influence inferences one generates after a correction has occurred; however, providing an alternative that replaces the causal structure it affords can reduce the effects of misinformation. (PsycINFO Database Record (c) 2012 APA, all rights reserved)

Full text

Journal of Experimental Psychology:
Learning, Memory, and Cognition
1994,
Vol. 20, No. 6,1420-1436Copyright 1994 by the American Psychological Association, Inc.
0278-7393/94/S3.00
Sources of the Continued Influence Effect:
When Misinformation in Memory Affects Later Inferences
Hollyn M. Johnson and Colleen M. Seifert
Several lines of research have found that information previously encoded into memory can
influence inferences and judgments, even when more recent information discredits it. Previous
theories have attributed this to difficulties in editing memory—failing to successfully trace out and
alter inferences or explanations generated before a correction. However, in Experiments 1A and
IB,
Ss
who had received an immediate correction made as many inferences based on misinforma-
tion as Ss who had received the correction later in the account (and presumably had made more
inferences requiring editing.) In a 2nd experiment, the availability (Tversky
&
Kahneman, 1973) of
the misinformation within the comprehension context was tested. The results showed that Ss
continued to make inferences involving discredited information when it afforded causal structure,
but not when only incidentally mentioned or primed during an intervening task. Experiments 3A
and 3B found that providing a plausible causal alternative, rather than simply negating
misinformation, mitigated the effect. The findings suggest that misinformation can still influence
inferences one generates after a correction has occurred; however, providing an alternative that
replaces the causal structure it affords can reduce the effects of misinformation.
One may often learn "facts" about an event that later turn
out to be false or unfounded. Ideally, one might want a
correction to diminish or eliminate subsequent effects from
such misinformation so that one understands and reasons
about it like those not exposed to the misinformation. How-
ever, previous research (e.g., Carretta & Moreland, 1983;
Ross,
Lepper, & Hubbard, 1975; Wilkes & Leatherbarrow,
1988;
Wyer & Budesheim, 1987) has found that discredited
information can continue to influence reasoning and under-
standing, under some conditions, despite an instruction to
disregard it. As a result of the prevalence of misinformation, it
is important to investigate how one reasons about accounts
containing corrections and what conditions could lead to a
continued influence effect.
Some research has found influence effects, even when two
simple explanations for them are ruled out: First, that subjects
do not notice the discrediting, and second, that they fail to
make the connection between the disregard instruction and
the information it refers to. Much research (Carretta &
Moreland, 1983; Ross et al., 1975; Wilkes & Leatherbarrow,
1988) has found that subjects do recall or acknowledge the
disregard instruction when asked about it directly, suggesting
Hoilyn
M.
Johnson (now at Washington State University—Tricities)
and Colleen M. Seifert, Department of Psychology, University of
Michigan.
This work was supported by the Office of Naval Research under
Contract N00014-91-J-1128 to the University of Michigan. We thank
Jennifer
Coe,
Andrew
Jeung,
Srinika Narayan, and Christy Marshuetz
for assistance in conducting the experiments, and extend special
thanks to Julie Bush for help with
materials.
We also thank Hal Arkes,
Brian Ross, A. L. Wilkes, Arthur Glenberg, and other, anonymous
reviewers for comments on a draft of
this
article.
Correspondence concerning this article should be addressed to
Colleen M. Seifert, Department of Psychology, University of Michi-
gan, 330 Packard Road, Ann Arbor, Michigan 48104. Electronic mail
may be
sent to seifert@umich.edu.
that they have encoded it and can retrieve and potentially
comply with it. Wilkes and Leatherbarrow also manipulated
whether a correction directly stated what one was to disregard
or required subjects to infer this, but they found no difference
in the amount of influence from the misinformation. This
suggests that influence can occur even when subjects have
made the connection between the disregard instruction and
the information it refers to.
Several theories have proposed that misinformation can
influence judgments because one integrates it into a memory
representation before a correction occurs. Hastie and Park
(1986) argued that one makes an immediate personality
assessment when one hears a trait or behavior attributed to a
person, and others have proposed that one integrates new
knowledge about a person into one's previous representation
(Wyer & Srull, 1986). In this case, corrected information may
continue to influence personality judgments because one may
still retrieve and use assessments based on the initial informa-
tion (Wyer & Budesheim, 1987; Wyer & Unverzagt, 1985).
Similarly, belief perseverance research has proposed that the
more prediscrediting elaboration or explanation one does the
more influence the discredited information has (Anderson,
Lepper, & Ross,
1980;
Ross et
al.,
1975).
These prior processing
theories raise a couple of issues about what occurs when one
must understand an account containing a correction.
One issue is whether misinformation can influence infer-
ences made after a correction occurs, as well as those made
before. When making on-line inferences (Graesser, 1981;
Hastie & Park, 1986) while comprehending an account, one
may involve misinformation in them before learning that the
information is incorrect. Subjects may then fail to go back and
successfully trace out and alter these inferences when the
correction occurs. Thus, one might retrieve and report such
unedited inferences from memory at a later assessment, rather
than generating new ones. In this case, continued influence
would reflect problems in editing what subjects already have m
1420

CONTINUED INFLUENCE OF MISINFORMATION1421
memory. However, inferences showing influence from misinfor-
mation could also originate, in two different ways, after a
correction occurs. First, one may still need to make on-line,
backward inferences after the correction to link postcorrection
information to what one read previously, thereby maintaining
coherence in the account. Second, one may make postcorrec-
tion inferences in response to questions that arise after one has
fully comprehended the account (on request). In both of these
cases,
continued influence could reflect subjects retrieving and
using the misinformation itself to make new inferences. Previ-
ous work has not addressed whether the continued influence
effect only stems from "unedited" precorrection inferences or
may also occur through postcorrection inferencing.
A second issue is whether continued influence effects occur
because misinformation is merely more available than other
information, as a result of its having appeared within the
understanding context. The mention of misinformation may
activate key concepts in memory for subjects who hear it,
whereas control subjects, who do not hear the target informa-
tion, would not have these concepts activated. Encoding a
correction later may not sufficiently reduce this activation,
leaving these concepts more available than other alternatives.
Tversky and Kahneman (1973) have proposed that informa-
tion one can easily retrieve from memory, because it is more
available,
can bias
frequency judgments. Similarly, misinforma-
tion may influence one's inferences about an account because
the activated concepts seem more subjectively probable and
lead subjects to produce plausible, easy inferences without
evaluating them within the context of the account. In contrast,
subjects who do not have these concepts activated would
produce fewer inferences consistent with them.
On the other hand, continued influence from misinforma-
tion in an account may occur as a result of the causal role it
plays,
which provides a coherent structure to the event
described. When making an inference, one may search one's
representation of the account for an antecedent that satisfies
causal constraints, such as operativity and sufficiency (Tra-
basso & van den Broek, 1985). One may retrieve a candidate
antecedent that would meet these constraints, were it not
discredited, and then may either fail to note the correction
while making the inference or find the account hard to
structure or extend without using the misinformation. Thus,
one may retrieve this information and use it because it
provides connections that structure the account, which might
otherwise remain fragmented.
This causal explanation predicts a larger continued influ-
ence effect when a correction merely negates earlier informa-
tion and does not provide any causal alternative.
Previous
work
suggests that people rarely generate alternative interpretations
spontaneously when a correction negates prior information
(Anderson, 1982; Massad, Hubbard, & Newtson,
1979),
which
may leave only the misinformation to provide causal structure.
Van den Broek (1990) has argued that accurate memory for
text depends on constructing a connected causal chain. If a
negation would seriously disrupt such a
chain,
then the need to
maintain coherence may override the correction instruction.
However, misinformation may have less influence when a
correction presents an alternative that also provides a causal
structure to the account. Such information could take over the
misinformation's role in the account, so one could incorporate
this alternative and still maintain a coherent understanding.
To investigate these hypotheses about how continued influ-
ence from misinformation arises, we used a fire investigation
scenario from a text comprehension paradigm developed by
Wilkes and Leatherbarrow (1988). In Experiments 1A and IB,
we examined whether misinformation influences inferences
only
as
a result of problems in editing precorrection inferences
stored in memory or whether postcorrection inferences can
also show influence. These experiments manipulated whether
the correction statement occurred several messages after the
information it discredited (as in Wilkes and Leatherbarrow,
1988) or, instead, immediately following it, thus limiting the
opportunities subjects had for making precorrection infer-
ences based on the misinformation. In Experiment 2, we
manipulated the role that target concepts played in an account
to test whether mere mention and activation of concepts
present in misinformation lead to influence effects. If these
effects depend on the information's role in the account's
structure, rather than its mere activation in memory, causally
implicated concepts should promote more "influenced" infer-
ences than concepts mentioned incidentally. Finally, in Experi-
ments 3A and 3B, we examined whether subjects show less
continued influence from misinformation when a correction
presents a plausible causal alternative, rather than just negat-
ing the target information.
Experiments 1A and IB
In Experiments 1A and IB, we manipulated how soon a
correction appeared after a piece of misinformation to test the
"editing difficulties" explanation for the continued influence
effect. Each group read a series of messages describing a
warehouse fire, which were based on materials from Wilkes
and Leatherbarrow (1988). In correction conditions, subjects
read that a short circuit occurred near a closet reportedly
containing
volatile
materials (cans of paint and pressurized gas
cylinders), whereas control subjects read that the short circuit
occurred near an empty closet. In a delayed correction
condition, subjects read five additional messages before learn-
ing that the closet did not contain volatile materials. In an
immediate correction group, the correction message directly
followed the message about the volatile materials. For all
conditions, the early messages in the series were written to
limit opportunities for forward and backward inferences that
would link information about the volatile materials to other
contents in the account. Experiments 1A and IB differed only
in how the correction messages were worded.
If influence effects occur only when subjects make infer-
ences based on the misinformation before the correction and
then must edit them, one would expect more influence in
delayed correction
groups
than in immediate correction groups.
Subjects in the delayed conditions would have
a window
within
which they could potentially make additional, strategic, causal
and coherence-maintaining inferences before the correction.
When the correction then occurs, they might not be able to
track down and alter them successfully. Several theories have
presumed that editing difficulties underlie continued influence
effects (e.g., Ross et al.,
1975;
Wilkes & Leatherbarrow, 1988;

1422
HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
Wyer & Budesheim, 1987). In contrast, subjects in an immedi-
ate correction condition would have less opportunity to make
precorrection inferences involving the volatile materials and
less concrete information (e.g., features of the fire) to make
volatile-materials inferences. This would limit precorrection
inferences to those that subjects could randomly generate
(Kintsch, 1988) on the basis of the volatile-materials message.
In this case, one would not expect as much influence from
misinformation when it is immediately corrected. Alterna-
tively, if subjects use the misinformation to generate new,
postcorrection inferences, one would expect influence even
when subjects receive an immediate correction. In Experi-
ments 1A and IB, we tested the locus of the continued
influence effect (whether subjects make new, postcorrection
inferences based on misinformation), and do not address
whether any of the inferences made arise as a result of on-line
processes.
Method
Subjects.
The
subjects were University
of
Michigan undergradu-
ates who received course credit in an introductory psychology class for
participating. Subjects were run in groups
of
8
to
10,
in single sessions
lasting approximately
50 min. We
conducted
the two
experiments
separately, with
32
subjects participating
in
Experiment
1A and 33
subjects in Experiment IB.
Design. Experiments
1A and IB
each featured three groups:
a
delayed correction condition,
in
which five messages about
a
ware-
house fire intervened between the misinformation and the correction;
an immediate correction group,
in
which
the
correction message
appeared immediately after
the
misinformation;
and a
no-reference
control group,
in
which subjects did not receive the message that was
later corrected
in the
other groups.
The
delayed correction
and
no-reference control groups replicated groups used
in
Wilkes
and
Leatherbarrow (1988).
Materials.
The
materials were modified versions
of a
series
of
reports used by Wilkes
and
Leatherbarrow (1988) that described
the
investigation of a warehouse fire. The series consisted
of
13
individual
messages, each 2 to 4 sentences long (see Appendix
A).
The messages
were combined into
a
booklet, with
1
message
per
page.
For the
no-reference control
groups,
the 5th message in the series stated that a
closet
in the
warehouse
was
empty,
and
later messages
did not
controvert this information.
For the
discrediting groups,
the 5th
message stated that
the
closet contained cans
of oil
paint
and
pressurized
gas
cylinders,
and the
correction appeared
as a 2nd
message from
a
police investigator stating that
the
previous message
regarding
the
closet's contents was incorrect
and
that
the
closet was
empty. For the immediate correction groups, the correction appeared
as Message
6
(immediately after
the
information
it
discredited),
whereas for the delayed correction groups, this statement appeared
as
Message
11
(allowing
for 5
intervening messages).
For the
control
group, Message
11 was a
notice from
the
police investigator that
several
firefighters
had been released from the hospital. For all three
conditions, references
to
features
of the
fire were deleted from
Messages
1 to 4 so
they would
not
become involved
in
backward
inferences once one had encountered Message
5.
This would also limit
subjects' ability
to
form
a
schematic
or
stereotypical view
of
the fire
and thus could limit predictive, forward inferences.
The text
of the
correction messages from
the two
experiments
is
shown
in
Table
1. The
correction message used
in
Experiment
1A
essentially replicated that
in
Wilkes
and
Leatherbarrow (1988).
However, previous research (Baker & Wagner, 1987) has found that
subjects have more trouble detecting contradictions when presented in
Table
1
Correction
Messages Used in Experiments 1A
and IB
Message/ experiment
Complex/Experiment
la
10:40
a.m. A second message received from Police Investi-
gator Lucas regarding the investigation into the
fire.
It
stated that the closet reportedly containing cans of oil
paint and gas cylinders had actually been empty before
the fire.
Direct/Experiment
lb
10:40 a.m. A second message received from Police Investi-
gator Lucas regarding the investigation into the
fire.
It
stated that there were no cans of
oil
paint or
gas
cylinders
in the closet that had reportedly contained them;
the
closet had actually been empty before the fire.
subordinate, rather than main, clauses.
In our
experiments, subjects
could potentially misinterpret this phrasing
as
stating that the cans and
cylinders are empty, but not the closet
itself,
which would contribute to
the continued influence effects reported
by
Wilkes
and
Leatherbar-
row.
To
avoid this problem, we used
in
Experiment
IB a
correction
that directly asserted that
the
closet contained no volatile materials,
rather than embedding this information in a subordinate clause.
Two memory tests were also prepared: first,
a
summary sheet,
in
which subjects wrote
a
free recall
of the
reports' contents
and
what
they thought was responsible
for the
fire (what caused
it to
happen),
and second,
an
open-ended questionnaire adapted from Wilkes and
Leatherbarrow (1988).
The
latter included
10
questions
on
facts
directly presented
in the
messages, 10 other questions requiring
the
subjects
to
make inferences about
the
event,
and 2
final questions
assessing whether subjects were aware
of
any correction
or
contradic-
tion
in the
series (questions from
the
open-ended questionnaire
are
shown
in
Appendix B). All questions appeared
in
the same order
for
each subject, with
all
fact questions appearing before
any
inference
questions
to
prevent
the
latter from introducing biases.
The 2
contradiction questions always appeared after the inference questions.
As
an
intervening task, subjects
did a
series
of
unrelated similarity
ratings.
Procedure. Each subject received
a
booklet
of
reports
and was
instructed to read through
it at
his or her own pace, but they were not
to
go
back and reread any of the
messages.
Subjects were also told that
they would
be
asked
to
recall
the
information later. When each
individual subject had
finished
reading, the experimenter collected his
or her booklet of
reports,
and the subject received the summary sheet.
When everyone
had
finished this test,
all
subjects worked
on the
intervening task
for 10 min.
After this time
had
elapsed, subjects
received the open-ended questionnaire and were instructed to answer
each question on the basis
of
their understanding
of
the reports. The
experiment ended when all subjects had completed the questionnaire.
We chose
the
instructions
and
order
of
tasks
to
replicate those
of
Wilkes and Leatherbarrow (1988).
Results
Scoring. A coder who
was
uninformed about the experimen-
tal conditions scored both the summary sheet and the open-
ended questionnaire (including fact and inference questions).
Three inference measures were computed: a
thematic-inference
measure, a
direct-reference
measure, and a
global-cause
mea-
sure.
The thematic-inference measure consisted of negligence
theme responses made on the 10 inference questions from the
open-ended questionnaire. These were responses consistent

CONTINUED INFLUENCE OF MISINFORMATION1423
with believing that the warehouse had contained carelessly
stored volatile materials that caused or contributed to the fire,
as would be reasonable if the information about the volatile
materials had not been discredited. This theme encompassed
references using key words from the discredited message (e.g.,
oil,
paint, gasfes], cans, or
cylinders),
mentions of the closet
itself without indications that it
was
empty, and attributions of
carelessness or negligence (see Table 2 for sample answers
consistent with the negligence theme). One might expect
subjects who received no information about the volatile mate-
rials to make some inferences consistent with a negligence
theme, but most
answers
would likely reflect themes other than
negligence, such as intentional setting of the fire, stored
stationery at the warehouse, and the structure of the building.
The direct-reference measure of inferencing consisted of a
count of all direct and uncontroverted references to the
volatile materials themselves (paint cans or gas cylinders),
including all measures: the free recall, in answer to fact
questions, or in answer to the inference questions. A reference
was considered controverted if the subject also stated that a
later message had shown this to be incorrect or untrue, or the
subject provided the gist of
the
correction
message.
Expressing
uncertainty about the presence of the volatile materials alone
was not considered a controverted reference. This direct-
reference measure indicates the influence of discredited
information occurring anywhere in subjects' protocols and
provides an estimate that does not rely on "thematic" interpre-
tations. Finally, a global-cause measure was determined by
scoring whether answers to the fire-cause question from the
summary sheet involved uncontroverted reference to the
volatile materials.
Three recall measures were also computed. First, a
story
recall
measure consisted of the number of noncorrection idea
units that subjects reported in their free recall on the summary
sheet. The series of story reports was broken down into
component idea units by using an adaptation of procedures
described in Kintsch
(1974).
A story unit
was
scored as recalled
if the subject reproduced a recognizable portion of
its
content;
otherwise the unit was scored as absent. Only messages
common to all three conditions were scored; thus, the story
recall measure does not include the message about the closet's
contents (empty or storing volatile materials) or the message
from the police investigator (correction or update on firefight-
ers) because these did not appear in the story for all condi-
tions.
Second, a fact recall measure was computed by scoring
accuracy on the 10 fact questions from the open-ended
Table 2
Sample Negligence Theme Answers to Inference Questions
QuestionAnswer
What was the possible
cause of the toxic
fumes?
Why do you think the fire
was particularly
intense?
For what reason might an
insurance claim be
refused?
Burning paint.
Oil fires are hard to put
out.
Because flammable items
were not kept in a safe
place.
questionnaire. Finally, a correction recall measure was deter-
mined on the basis of whether subjects in the correction groups
accurately referred to it in either the summary or the open-
ended questionnaire.
Inferences. The analyses reported below exclude idiosyn-
cratic responses and items left blank, which made up 14.2% of
the responses to the inference questions. Analyses of variance
(ANOVAs) were done using the thematic-inference measure.
For both experiments, Table 3 shows the mean number of
negligence theme inferences per subject within each group.
The results show a significant difference on this measure: F(2,
31) = 13.30,;? < .0001,M5e = 41.01;F(2,31) =
7.47,p
< .002,
MSC = 14.96, respectively. Planned comparisons revealed
significant differences between the correction groups and the
control group, with the control group making significantly
fewer inferences consistent
with
the negligence theme: t(29) =
5.00, p <
.0001,
SE = 1.31; t(29) = 3.84, p <
.001,
SE = 1.08,
respectively, but no significant difference between the two
correction groups (fs < 1.5 in both experiments).
Table 3 shows the mean number of direct references to the
stored volatile materials for each group, in both experiments.
An ANOVA of this direct-reference measure of influenced
inferences showed a significant difference in both experiments:
F(2,31) =
U.71,p
<
.0001,
MSe =
35.63;
F(2,31) =
5.74,/>
<
.008,
MSe = 12.98, respectively, with the correction groups
showing more influence than the no-reference control groups.
In Experiment 1A, post hoc analyses using Tukey's honestly
significant difference (HSD) procedure showed significant
differences between the no-reference control group and each
of the correction groups, but no differences between the two
correction groups (p < .007). This pattern also occurred in
Experiment IB, with a significant difference (p < .006) be-
tween the immediate correction group and the no-reference
control
group,
and
a
marginally significant difference (p < .08)
between the control and the delayed correction
group.
Overall,
95%
and
91%
of subjects in the correction groups, in the two
experiments respectively, made at least one direct and uncon-
troverted reference to the volatile materials. In the control
groups for the two experiments, 27% and 30% of the subjects
made reference to similar materials.
In contrast to the findings above, only 24% and 9.1% of
subjects in correction groups, in the two experiments respec-
tively, reported the volatile materials in response to the
global-cause question from the summary sheet. The rest of the
responses to this global-cause question were similar to those
given by the no-reference control group, such as faulty wiring
and arson, or were left blank.
Recall. There were no group differences on the story recall
or the fact recall measures (Fs < 1 for each, in both experi-
ments).
The overall means for story recall were 14.3 and 12.2
units in Experiment
1A
and
IB,
respectively; subjects correctly
recalled a mean of 9.2 and 8.3 facts, respectively. The correc-
tion recall measure showed high levels of recall in the
correction groups in both experiments (100% of the delayed
correction groups from both experiments and 90% and 91.7%
of the immediate correction groups from the experiments,
respectively). Analyses omitting subjects in the correction
groups who did not report noticing the correction showed no
difference from the results found when including all subjects.

1424
HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
Table
3
Number of References
and
Inferences Consistent With Misinformation,
by
Group, From
Experiments
1A and IB
Reference/inference
Direct references
M
SD
Negligence inferences
M
SD
Delayed
3.27
1.56
3.82
1.78
Experiment 1A
Immediate
4.40
2.01
4.90
1.60
Control
0.82
1.66
1.09
1.87
Delayed
2.10
1.20
3.10
1.37
Experiment IB
Immediate
2.75
1.96
3.25
1.60
Control
0.60
1.87
1.10
1.20
Discussion
The findings that subjects show influence from misinforma-
tion, whether corrected early
or
late
in the
sequence, replicate
and extend those
of
Wilkes
and
Leatherbarrow (1988)
and are
consistent with work on belief perseverance (Ross
et
al., 1975).
The results
of
both Experiments
1A and IB
suggest that
influence from misinformation
can
also occur
as a
result
of
postcorrection inferencing processes, rather than only when
subjects may have generated
the
inferences before
the
correc-
tion. Subjects
in the
immediate correction group presumably
made
few
precorrection inferences
and
thus
had
less need
to
edit their memories; this suggests that subjects
may
retrieve
the misinformation itself
to
generate postcorrection infer-
ences.
This could occur either on-line, as one makes backward
inferences that link postcorrection information
to
what
one
read previously
or
reconstructively,
in
answering questions
that arise after
one has
fully comprehended
the
account.
The
latter view
is
consistent with previous research
on
reconstruc-
tive memory (Conway
&
Ross, 1984; Spiro, 1980), which
had
found that subjects, perhaps inappropriately,
use
more recent
information
to
reconstruct earlier information and attitudes.
Memory differences cannot account
for the
continued influ-
ence effect found here:
The
groups
did not
differ
in
recall
of
the reports,
and
over 90%
of
those
in the
correction groups
recalled
the
correction. Thus, these subjects
had the raw
materials available
to
avoid inferences influenced
by the
misinformation,
yet
they
did not do so.
Further, over 75%
of
the subjects
in
both experiments failed
to
report
the
volatile
materials when directly asked what
was
responsible
for
the fire
(the global-cause question). This suggests that results
are not
due
to
simple lack
of
belief
in the
correction.
The
results also
show
a
continued influence effect whether
the
correction
message
was
complexly worded,
as in
Experiment
1A, or
directly worded, as in Experiment
IB.
However, the correction
groups
in
Experiment
IB
made fewer negligence inferences
(M
= 3.18)
than those that occurred
in
Experiment
1A
(M
=
4.36), suggesting that
a
more direct message
had
some
mitigating effect.
One could argue that
the
editing hypothesis could still
account
for
these results because
the
experiments
do not
provide direct evidence that subjects
in the
immediate correc-
tion condition
did not
make precorrection inferences.
How-
ever, there
are
several reasons
why
this interpretation
is not
plausible. First, the early messages in the series did
not
contain
references
to
features
of the
fire that
the
volatile materials
could explain. Without these references,
one
would need
to
make
few if any
backward inferences
on
encountering
the
volatile-materials message. Thus, subjects
in the
immediate
correction group would
not be
likely
to
have generated such
precorrection inferences, which would require editing after the
correction. Second,
it is
unlikely that subjects would make
many forward inferences because
the
earlier messages just
mentioned
the
existence
of a
fire.
Van den
Broek (1990)
argued that readers
are
more likely
to
make forward infer-
ences when necessary
and
sufficient causal conditions
con-
strain those inferences;
in
this case,
the
mention
of the
fire
gives
few
clues
to its
specific characteristics,
and
so one might
not expect many predictions until more information came
in.
Third,
the
open-ended questionnaire asked about characteris-
tics
of the
fire mentioned after Message
6
(postcorrection
for
the immediate correction group) and
thus
would be unlikely
to
tap
and
assess inferences based
on
precorrection information.
In interpreting these results, however,
one
cannot conclude
that postcorrection inferencing accounts
for all
continued
influence found
in
prior research.
The
editing hypothesis may
still account
for the
continued influence that occurs
in
studies
that actively promote inference
or
attribution making before
a
disregard instruction occurs (e.g., Anderson
et
al., 1980; Schul
& Burnstein, 1985). Further, Experiments 1A
and IB did not
assess
whether the influenced inferences
in
the delayed correc-
tion condition reflect precorrection
or
postcorrection infer-
ences;
both could contribute
to the
influence observed.
How-
ever, even
if
subjects make few precorrection inferences in this
condition,
as
suggested
by a
minimal inferencing account
(McKoon
&
Ratcliff,
1992), this would still support
the
main
claim
of
these experiments: Postcorrection inferencing involv-
ing misinformation
can
account
for
some
of the
continued
influence effect observed.
Thus,
these findings address
the
locus
of the
continued
influence effect
and
suggest that
it can
occur
as a
result
of
processes operating after
the
correction,
as
well
as
those
occurring before.
One can
also
ask
whether misinformation
must originally play
a
causal role
in the
account
to
influence
later
reasoning.
Previous research on comprehension (Graesser
& Clark, 1985; Haviland
&
Clark,
1974) has
suggested that
causal relations give
an
account structure
and
provide
a
basis
for further inferences. Subjects
may
have trouble finding
an
alternative
way to
structure
an
account coherently when
causally central information
is
discredited,
so
they
may con-
tinue
to
rely on
it
and use
it
in later inferencing. Some research

CONTINUED INFLUENCE OF MISINFORMATION1425
in belief perseverance (Anderson et al., 1980; Fleming &
Arrowood,
1979;
Ross et al., 1975) has proposed that misinfor-
mation has more influence when subjects
try
to explain reasons
for their performance, which would presumably involve the
misinformation in a causal structure; however, other work has
not reported this explanation effect (Anderson, 1982, 1983;
Jennings, Lepper, & Ross, 1981).
Alternatively, perhaps the mere mention of misinformation
activates concepts that suggest plausible inferences. Prior
work
in decision making (Tversky
&
Kahneman, 1973) has proposed
that more readily available information in memory can bias
judgments. If a similar activation hypothesis can account for
the continued influence of misinformation in accounts of
events, this would suggest that any mention of
the
information,
regardless of contextual support, could show similar influence.
It would also suggest that the continued influence effect may
be an artifact of the assessment, in some cases. If activation is
solely responsible, one would expect influence effects to
diminish over time, as activation lessens for key concepts
introduced in the misinformation. In Experiment 2, we tested
whether information merely mentioned in an account influ-
ences one's inferences or whether the information must also
afford causal structure for it.
Experiment 2
In Experiment
2,
we tested whether the continued influence
effect depends on the sheer availability of the misinformation
in memory, regardless of its role in the account. First, in a
corrected causal reference group, subjects read that a short
circuit occurred near a closet reportedly containing volatile
materials. Here causal inferences are likely because conditions
of operativity and sufficiency (Trabasso & van den Broek,
1985) are met. Subjects later learned that the closet did not
contain volatile materials. In a second causal reference group,
subjects read the same volatile-materials information as in the
corrected causal reference group, but did not receive a
correction
later.
The third condition was a noncausal reference
group, which presented the volatile materials within the
context of the account but as products in a store across the
street from the fire location. In this condition, the information
does not meet conditions conducive to involvement in causal
inferences, such as operativity and sufficiency (Trabasso
&
van
den Broek, 1985). The presence of volatile materials in a store
across the street is not sufficient, in the context of the reports,
to account for the warehouse fire; one would need to make
many inferences beyond the story context (e.g., that someone
moved the materials into the warehouse, that someone in-
tended this, and so forth) to establish these materials in a
causal role. For this group also, the series of reports did not
include a correction. Finally, in a fourth, external reference
condition, subjects read a series of reports that did not present
information about the volatile materials or a correction.
However, in an intervening task, they generated associates to
key concepts introduced as misinformation in the corrected
causal reference group. Previous research on implicit memory
(Schacter, 1987) has found that processing a word list can
increase the probability that subjects will produce those words
later on in an ostensibly unrelated task (here, the postcorrec-
tion assessment).
If the causal structure of information affects its influence on
inferences, one would expect little influence in the external
and noncausal reference
groups,
in which
the volatile materials
information contributes little to one's causal understanding of
the account. One would expect greater influence in the causal
and corrected causal reference
groups.
At the other
extreme,
if
the continued influence effect results simply from activation of
plausibly related concepts in memory, one would expect the
external reference group and the corrected causal reference
group to show similar influence on inferences reported in the
postcorrection assessment, even though the activation of the
volatile material
key
words originates outside the story context
for the former. Finally, if one considers the postcorrection
assessment as an explicit-memory test (Schacter, 1987), sub-
jects may only rely on activation of concepts that occur within
the episodic representation of the report series. Thus, if the
influence effect is due to activation of plausibly related
concepts within a particular episodic context, one would
expect few influenced inferences in the external reference
group, but a higher number of influenced inferences in both
the noncausal and the corrected causal reference groups.
Method
Subjects. Sixty-one University of Michigan undergraduates partici-
pated
as
partial fulfillment of
an
introductory psychology requirement.
Subjects were run in single, 45-min sessions, in groups of
8
to 10.
Materials. The materials were modified versions of the report
series used in the previous experiments. For the causal reference and
the corrected causal reference
groups,
the
fifth
message stated that the
closet contained volatile materials. The correction appeared as Mes-
sage 11 for the corrected causal reference group; in the causal
reference group, this message stated that several
firefighters
had been
released from the hospital. In the noncausal reference
group,
Message
3 stated that the fire department was called by the manager of a store
located across from the warehouse, which sold oil paint and pressur-
ized gas
cylinders.
In
this
group,
Message
5
stated that the closet in the
warehouse was empty, and Message
11
was the same as in the causal
reference group. Thus, the noncausal reference condition mentioned
the volatile materials within the story context, but as a peripheral fact.
Finally, the external reference group used the same series of reports as
in the causal reference group, except that Message 5 stated that the
warehouse closet was empty rather than containing volatile materials.
An intervening task was also constructed, made up of
102
pairings
of
category names, such as
city,
with initial letters, such as
F.
For each
pairing, subjects were to think of a category exemplar that begins with
that letter.
Two
versions of
this
task were used. Subjects in the external
reference group saw the priming version, which presented the words
paint,
gas,
and
container as
categories and also presented categories for
which
paint,
gasoline,
and oil are typical examples. Thus, by reading
and generating these items, subjects in the external reference condi-
tion would activate
key
concepts introduced
in
the misinformation that
the corrected causal reference group received. These items appeared
among the 20 items presented on the first page of the task. Subjects in
the other three groups received a second control version of the task,
which
was
the same except that neutral items replaced those involving
the key concepts from the misinformation message.
The
two
memory tests
were
the same
as
in the previous experiments.
For the noncausal reference group, the open-ended questionnaire
contained an additional question about the contents of the store across
the street, appearing after the
two
contradiction questions.

1426HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
Procedure. We used the same procedure as in the previous experi-
ments.
Results
All dependent variables were coded as in the previous
experiments.
Inferences. The following analyses exclude idiosyncratic
responses and inference questions left blank (18%).
Table 4 shows the mean number of negligence thematic
inferences and of direct references to the volatile materials.
An ANOVA of the number of negligence thematic inferences
showed a significant difference: F(3, 57) = 9.16, p <
.0001,
MSe = 26.34. A post hoc analysis using Tukey's HSD proce-
dure found no significant difference between the two causal
reference groups or between the external and noncausal
reference groups. However, both of the causal reference
groups significantly differed from the other two groups
(p < .05).
An ANOVA of the direct-reference measure, with group as
a variable, also showed a significant difference, F(3, 57) =
30.34,
p <
.0001,
MSe = 62.77. A post hoc comparison using
Tukey's HSD procedure again showed no significant differ-
ence between the two causal reference groups, which heard
about the volatile materials, or between the two other groups
that did not. However, both of the causal reference groups
again significantly differed from the noncausal and the external
reference groups (p < .001). Overall, 97% of the subjects in
the two causal reference groups made at least one direct and
uncontroverted reference to the volatile materials, versus 30%
of the subjects in the noncausal and external reference groups.
Analysis of the global-cause question from the summary
sheet showed that 22.6% of the subjects in the two causal
reference groups attributed the fire to the volatile materials
when directly asked what
was
responsible for
it.
In comparison,
only 3% of subjects from the external-noncausal reference
groups made such attributions. The rest of the responses
attributed the fire to such causes as faulty wiring or arson or
were left blank.
Recall. As in Experiments 1A and IB, the groups did not
significantly differ on the fact recall measure (F < 1), with an
overall mean of 8.4 facts recalled correctly, or on the story
recall measure (F < 1.2), with an overall mean of 13.02 units
Table 4
Number of Inferences and References Consistent With
Misinformation, by Group, From Experiment 2
Inference/
reference
Negligence
inferences
M
SD
References to
volatile
materials
M
SD
Causal
reference
3.88
1.96
4.44
1.71
Group
Corrected
causal
3.40
1.88
3.47
1.81
Noncausal
reference
1.13
1.19
0.73
1.22
External
reference
1.73
1.62
0.33
0.72
recalled. In the corrected causal reference group, 80% of the
subjects recalled the correction. Analyses omitting subjects in
this group who did not report noticing the correction showed
no difference from the results found when including all
subjects.
Other manipulation
measures.
In the noncausal reference
condition, 80% of the subjects correctly recalled the incidental
information about the store's contents. Analyses omitting
subjects who did not recall this showed no difference from
results found when including all subjects.
A coder who was uninformed about the experimental
conditions scored the intervening task. Subjects received a
point for each associate written in response to one of the three
volatile-materials words presented
(oil,
paint,
and
gas)
and for
each volatile-materials key concept generated as an associate
to a presented item. The results show that the external
reference group made an average of 2.5 such responses.
Among the other groups (who were not presented with any of
these target items), only 2 subjects spontaneously generated
such a response.
Discussion
The results of Experiment 2 show that misinformation is
more likely to influence inferences when it affords causal
structure for the account and not simply when its key concepts
remain activated and are therefore easily retrievable. Subjects
in the external and noncausal reference groups made few
inferences consistent
with
or mentioning the volatile materials.
However, subjects in the external reference group processed
and generated the key concepts in the intervening task, and
80%
of the subjects in the noncausal reference group recalled
the incidental information about the store's contents when
directly questioned about it. These results suggest that both
groups had the key concepts available, but that they failed to
use them in making inferences, as a result of the lack of causal
relevance. In contrast, subjects showed a substantial number
of
inferences involving the volatile materials in both the groups in
which this information was mentioned within a causal context.
This suggests that the causal structure that the information
affords, and not its mere availability, can lead to influence on
inferences about an account, even when a correction appears
later.
There are a couple of potential explanations for why the
present results do not replicate those of previous research
(Golding, Fowler, Long, & Latta, 1990; Wyer & Budesheim,
1987;
Wyer & Unverzagt, 1985), which has found that a
disregard instruction at least partially mitigates effects of
mistaken information. First, a majority of those studies have
used an impression formation task, rather than a text compre-
hension task, and this may involve a different discounting
procedure. Wyer and Budesheim (1987) proposed that people
may have theories on how misinformation could influence
their personality judgments and may use them to adjust their
ratings. Subjects in our study may not have a comparable
theory on how misinformation about the event would become
incorporated into their inferences, and so they
may
not be able
to adjust their responses accordingly. Second, subjects in the
Wyer & Budesheim study knew their goal was to form an

CONTINUED INFLUENCE OF MISINFORMATION1427
impression, which they would base judgments on later. In
contrast, in the present experiment, subjects were given recall
instructions before the task, but no explicit instructions about
comprehension. Thus, subjects in the person impression task
may process the correction differently because they can moni-
tor their discounting efforts in terms of that task. Without the
ability to monitor their discounting efforts, our subjects'
memories may contain blatantly contradictory beliefs about
the event, which they might not realize until something draws
their attention to it.
Alternatively, previous research (Golding et al.,
1990;
Wyer
& Budesheim, 1987; Wyer & Unverzagt, 1985) also differs
from what is reported in this article in how they motivated the
disregard instruction. In previous work, subjects received a
disregard instruction because the experimenter mistakenly
read information about a completely different entity (a differ-
ent person or work
contract).
Subjects showed more mitigation
of the misinformation's influence in mistake conditions than
when simply told to disregard information because it was
confidential (Golding et al.,
1990;
Wyer
&
Unverzagt,
1985).
In
the present research, the correction negated information
appearing earlier but did not address why the mistake oc-
curred. Thus, subjects do not know why someone volunteered
the misinformation in the first place, when providing it would
violate conversational norms (Grice, 1975) of quality and
relevance. They may accept the correction, in a limited or
literal sense, but still believe that there is something to the
original information, or else the reports would not have
mentioned it. For example, subjects might accept the correc-
tion's literal content and believe that the closet was empty
before the fire, but they might still believe that the volatile
materials existed somewhere else or at some other
time.
In this
case,
the continued influence effects may reflect subjects'
reasonable attempts to make sense of why the report originally
mentioned the misinformation and to address the error's
implications. Such a process would still result in comprehen-
sion differences between subjects who encounter misinforma-
tion and those who do not.
Overall, the present results provide stronger evidence that
subjects come to understand the same event differently on the
basis of whether an account contains a correction. Such
findings would have implications for one's understanding of
events in the world. Often, as one obtains more information
about events over time, many initial speculations may turn out
to be unfounded. However, even explicitly stated corrections
often merely indicate that certain previous information is
incorrect, and require the listener to reason out the implica-
tions of the change, or
why
someone provided what turned out
to be misinformation. These experiments demonstrate that
one cannot achieve complete discrediting, if that is one's
intent, by merely negating the literal content of information.
However, there may be other ways to provide corrections that
would lead subjects to accept the initial report as a meaning-
less error (that the materials were not in the closet, or
anywhere else in the warehouse, at any time).
Our findings are also consistent with other research that has
examined similar effects, despite differing definitions of avail-
ability and activation. In research on belief persistence, Ander-
son, New, and Speer (1985) proposed that availability contrib-
utes to influence from discredited information. However, that
work manipulated whether one had alternatives to the discred-
ited information available, but it did not make claims about, or
directly manipulate, the relative activation level of
the
discred-
ited information
itself.
Subjects were also likely to have made
connections linking the information to other world knowledge
before it was discredited; this would more closely parallel the
causal reference groups in this experiment than a mere-
mention condition. Thus, our results are consistent with this
work. Other previous research, on co-reference (O'Brien &
Albrecht, 1991; O'Brien & Myers, 1987), has found that it is
easier to reinstantiate a concept that has causal connections to
other text material than one that is less causally implicated.
Thus,
causal connections may lead to more activation than a
mere mention would, and therefore result in more influence.
However, this is not inconsistent with the findings here, as it
still supports the idea that both the causal and the corrected
causal reference groups show influence due to the causal
structure that the volatile-materials information affords.
On the basis of the causal hypothesis, one might also expect
subjects to show less influence from misinformation when a
correction provides a causal alternative, rather than simply
negating earlier information. When a correction merely ne-
gates information, many aspects of the event are left unac-
counted for. One may make inferences, and therefore re-
sponses, that maintain a coherent story versus having an
account that contains many loose ends. However, often a
correction provides an alternate fact, for example, "No, it
wasn't a blue car, it was a red one." In this case, one has a
concretely stated alternative that one can believe and under-
stand in the same causal role that the misinformation held.
Thus,
one would not need to maintain the misinformation to
have
a
coherent account. The alternative could allow one to, in
essence, "drop" the misinformation from one's representation
of the event or could compete and win out as a causal
antecedent for use in inferences. In Experiments 3A and 3B
we examined inferences made when a correction provides a
new causal alternative along with a negation by using different
story scenarios.
Experiment 3A
To determine whether misinformation has less influence
when a correction provides a causal alternative, we used in
Experiment 3A modified versions of the materials from the
earlier experiments. For two groups, the series of reports
implied an accident theme, which encompassed both the
negligence theme from the earlier experiments and a theme
involving
the stationery supplies stored in the
building.
For two
other
groups,
the reports implied an arson theme.
The
variable
of implied theme was crossed with whether a correction
occurred or not. Thus, the four groups were: (a) a negation
group, as in previous experiments, in which a correction stated
that the earlier information about volatile materials was not
true,
leaving one with a stationery supplies subtheme if one
took the negation into account; (b) a mention control
group,
in
which the reports presented information about the volatile
materials and did not correct this later, leaving one with the
negligence subtheme; (c) an arson control group, in which the

1428HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
reports made no mention of the volatile materials (paint and
gas) but instead mentioned that suspicious rags and steel
drums (arson materials) were found in another location; and
(d) an alternative group, in which subjects initially read about
the volatile materials stored in the closet and later learned that
this was not true, but arson materials had been found else-
where. Throughout the rest of the article,
volatile materials
refers exclusively to the oil paint and gas cylinders reportedly
stored in the closet, and
arson materials
refers to the gasoline-
soaked rags and emptied steel drums used to intentionally set
the fire.
If subjects can discredit information when the correction
presents an alternative instead of just a negation, one would
expect the alternative group to show fewer negligence infer-
ences,
and fewer references to the volatile materials, than the
negation group would; at best, it would be no different from
the arson control group, which never heard mention of the
volatile materials. Subjects could create and report a coherent
account of the event by relying on the information about the
arson materials and thus would not use the information about
the volatile materials in their inferences. However, if having a
causal alternative does not limit the influence from misinforma-
tion, in the worst case, one would expect the alternative group
and the mention control group to show a similar number of
negligence inferences and references to volatile materials. To
validate the arson manipulation,
we
performed comparisons of
the arson control and mention control groups, which will show
whether the implied themes are distinct enough to warrant
interpretation of the other conditions or whether mention of
these arson materials leads to responses involving the volatile
materials by association, even though the reports do not
directly mention them.
Method
Subjects. Eighty-one University of Michigan undergraduates par-
ticipated as partial fulfillment of an introductory psychology require-
ment. Subjects were run in single, 45-min sessions, in groups of 10
to 12.
Design. The design was a 2 x 2 berween-subjects factorial, with
correction (yes vs. no) and implied theme (accident vs. arson) as
variables. The mention control and negation conditions systematically
replicated the causal reference group of Experiment 2 and the
negation groups in the previous experiments, respectively. Two other
groups read reports that suggested arson as the fire's cause. The arson
control group received only information suggesting arson, and no
contradictory information, whereas the alternative group first read
about the volatile materials carelessly stored in a closet, learned this
was an error, and then received information suggesting arson.
Materials. The materials were essentially the same as those used in
the previous experiments, except that the series of reports contained
15 messages, with 2 new messages added to allow for further
inferences that could discriminate between the negligence and arson
themes. Appendix C shows the text of the new messages; Table 5
indicates how messages were combined to manipulate theme and
correction in the different conditions.
Message 6 either read that a storage closet near where the fire
started contained cans of paint and gas cylinders or read that it was
empty. Message 13 either stated that the previous message regarding
the closet's contents was incorrect and the closet was empty or stated
that two
firefighters
had been released from the hospital. Message 14
Table 5
Content of
Critical Messages
by Group From Experiment 3A
GroupMessage
6
Message 13 Message 14
Volatile Firefighters Warehouse
materials supplies
Volatile Empty closet Warehouse
materials supplies
Volatile Empty closet Arson materials
Mention
control
Negation
Alternative
Arson control Empty closet Firefighters Arson materials
either gave a list of supplies normally stocked
by
the warehouse or
gave
indications that gasoline-soaked rags had been found in suspicious
circumstances. The summary sheet and open-ended questionnaire
were the same as in the previous experiments, except that the
questionnaire contained several additional questions, on topics like
motives of the owner and the cause of
a
previous fire on the premises,
to assess inferences consistent with the different themes used here.
Procedure. We used the same procedure as in the previous experi-
ments.
Results
Scoring. The dependent variables used in the previous
experiments were coded as described in Experiment 1. As
before, the negligence thematic-inference measure included
responses that explicitly used key words from the message
about volatile materials in the closet (e.g., oil, paint, gasfes],
cans,
or cylinders). In addition, a coder who was uninformed
about the experimental conditions scored the inference ques-
tions for responses consistent with an arson theme. This arson
theme encompassed responses that explicitly used key words
from the message presenting the arson materials (e.g., gaso-
line,
rags, drums, or barrels) or expresses arson intent or
motives for arson. The strict coding of key words from the
original "volatile-materials" message versus the later "arson
materials" message allowed the coding scheme to distinguish
between the volatile and arson materials and themes. Answers
containing a feature from both the arson and the negligence
theme were coded as a "mixed reference." The following
analyses do not include idiosyncratic responses or inference
questions left blank (20.4%).
Inferences. In 2 x 2 ANOVAs, we tested whether the
groups differed in number of thematic inferences consistent
with either negligence and arson (see Table 6). There was a
significant main effect of implied theme for both the negli-
gence and the arson themes: for negligence, F(l, 76) = 15.02,
p <
.0001,
MSe =
48.81;
for
arson,
F(l, 76) = 27.98,p <
.0001,
MSe
= 95.90. For both of these analyses, the main effect of
correction and the interaction of correction and implied theme
were not significant (F < 1). The two control conditions
differed significantly on these
two
measures
in
planned compari-
sons:
for negligence, f (76) = 3.29, p < .002, SE = .57; for
arson, <(76) =
3.54,
p <
.001,
SE = .59. Overall, the mention
control group made few arson inferences and the arson control
group made few negligence inferences.
To assess specific influences of the discredited information,
we
compared the direct-reference measure in a
2
x
2
ANO VA,
with implied theme (accident
vs.
arson) and correction
(yes
vs.

CONTINUED INFLUENCE OF MISINFORMATION1429
Table 6
Number of Thematic Inferences and References Consistent With
Misinformation, by Group, From Experiment 3A
Reference/
inference
Negligence
inference
M
SD
Arson inference
M
SD
Direct references
M
SD
Mention
Negation
4.14
2.17
1.33
1.32
3.71
2.43
Control
4.32
2.40
1.47
1.84
3.47
2.01
Arson
Alternative
2.60
1.50
4.00
1.69
3.25
1.89
Control
2.10
1.41
4.33
2.54
0.29
0.90
no) as the variables. This measure showed a significant
interaction between groups, F(l, 76) =
9.51,
p <
.003,
MSe =
33.45 (see Table 6). A post hoc comparison using Tukey's HSD
procedure showed a significant difference between the arson
control group, which never heard about those materials, and
each of the three groups that did (/? < .001). In the three
groups that heard mention of the volatile materials, 91.5% of
the subjects made one or more direct and uncontroverted
references to them, whereas in the arson control group, only
14.3%
made spontaneous reference to similar materials.
Responses to the fire-cause question from the summary
sheet were also scored as before in a global-cause measure.
In the mention control group, in which information about
volatile materials was not controverted later, 78.9% of the
subjects attributed the fire to those materials. In the negation
and alternative groups, in which information about the
volatile materials was discredited, 28.8% and 22.2% of
the subjects attributed the fire to those materials, respectively.
On the other hand, subjects in the alternative and arson
control groups attributed the fire to arson (77.8% and 81.0%,
respectively).
Recall. As in the prior experiments, the four groups did not
differ on story recall or in fact recall measures (F < 1.7 and
F < 2.61, respectively). The overall mean for story recall was
11.19 units, and subjects recalled a mean of 8.2 facts. On the
correction recall measure, both of the correction groups
showed high levels of recall of the correction, with 95.2% of
the negation group and 75.0% of the alternative group
recalling it. Analyses omitting subjects in the correction groups
who did not report noticing the change showed no difference
from the results found with all subjects.
Discussion
The thematic inference results show that misinformation
has less influence when subjects receive a correction that
provides an alternate causal explanation, rather than one that
only negates the information. Subjects in the alternative group
made as many arson inferences as a control group for whom
arson was the only theme given, and the two groups did not
differ in negligence inferences, despite the alternative group
having heard mention of the volatile materials. However,
subjects in the negation group made as many negligence
inferences as a control group for whom the information on
volatile materials was valid. These findings also support the
claim that the continued influence effect depends on the causal
structure that the misinformation affords an account and are
consistent with proposals that one may reason by using
discredited information when other alternatives do not come
to mind (Anderson et al.,
1985;
Wegner, Coulton, &
Wenzlaff,
1985).
The results also suggest that a correction accompanied by a
causal alternative works better than one entailing only nega-
tion, at least when the alternative adequately fills the causal
role the misinformation would hold. This could occur for
several reasons. First, misinformation that is only negated may
interfere
with
subjects' ability
to
generate alternatives, in effect
fixing them in one interpretation of the event. Providing an
alternative may break this set and allow subjects to consider
other interpretations. Second, the alternative and the misinfor-
mation in memory may compete when one must make an
inference. If both pieces of information come to mind, one
might have to deliberate between the two, which could make
one more likely to heed the correction. Finally, the alternative
may help one explain why the misinformation was mentioned
in the first place. In this case, subjects may reason that
someone volunteered the original information about volatile
materials because they had mistaken them for the arson
materials, the presence of which was confirmed later. With a
simple negation, however, subjects do not know how the
misinformation originated and so may infer that the materials
must exist somewhere to explain their mention.
These results also replicate those in the previous experi-
ments and provide more evidence that continued influence
effects can occur despite subjects noticing and accepting the
correction in some sense. First, over 90% of subjects in the
negation group recalled the correction, suggesting that the
effect is not just due to subjects failing to detect a discrepancy
in the account. Second, only 29% of the subjects in the
negation group mentioned the volatile materials in response to
the
fire-cause
question, as compared with 79% in the mention
control group, in which the information was valid. One would
not expect such a large difference if subjects simply did not
believe the correction. However, over 90% of the subjects in
both of these groups made at least one uncontroverted
reference to the volatile materials, which provides evidence
that the effect
is
not merely due to responses from subjects
who
did not accept the correction.
However, a causal alternative does not necessarily prevent
all influence from misinformation. The results also show that
subjects made more direct references to the volatile materials
if they were mentioned at all, regardless of the type, or
existence, of
a
correction. This may occur because a correction
does not expunge the discredited information from
memory,
so
one still has some chance of retrieving it. It may then beat out
the alternative, or provide a better or more plausible explana-
tion, so one may involve it, rather than information about
arson materials, in one's inferences. Or, subjects in the
alternative condition may be more likely, in this case, to link
the volatile materials and the arson materials and see them as
being the same thing, thus interpreting the earlier message as
correcting the location of the volatile materials and not

1430HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
whether those materials existed at all. This might be especially
likely because the two alternatives share a causal role and
many of the properties that allow them to fill that role. Some
support for this option comes from the finding that subjects in
the alternative group showed lower recall of the correction
(75%) than did the negation group (over 90%).
Experiment 3B
One could argue that the continued influence effects found
in the previous experiments are an artifact of the particular
warehouse fire scenario used, and thus that
findings
on how to
mitigate the effect might not generalize to other stories. A
number of studies have found similar continued influence
effects by using different story materials in comprehension
tasks (Hertel, 1982; Spiro, 1980; Wilkes & Leatherbarrow,
1988) and in a variety of tasks other than comprehension:
Subjects have been found to involve discredited information in
making judgments about task ability in others and oneself
(Ross et al., 1975), social theories (Anderson et al., 1980),
person impressions (Schul & Burnstein, 1985; Wyer &
Budesheim, 1987; Wyer & Unverzagt, 1985), decisions about
likely activities (Schul & Burnstein, 1985), and acceptability of
a work contract (Golding et al., 1990). However, few other
studies (cf. Anderson et al., 1985) have directly examined
influence effects when one is presented with a causal alterna-
tive as part of the correction. In Experiment
3B,
we attempted
to extend the results of the previous experiment to additional
materials.
Method
Subjects. A total of 40 University of Michigan undergraduates
participated in a single session lasting approximately IS min. Thirty-
three of the subjects received course credit for participating, and 7
were paid. Each experimental condition contained both paid and
credit subjects. Subjects were run in groups of
2
to 6.
Design. In three conditions, we varied the information presented
to subjects. The no-mention control group saw no information about
the son as a suspect. The negation and alternative conditions were
given information about the son as a suspect, which was later negated
in the negating correction group and corrected with an alternative in
the alternative correction group.
Materials. The materials were a series of reports describing a
police investigation into a theft at a private
home.
The series consisted
of 14 individual messages, each 2 to 4 sentences long (see Appendix
D),
which were combined into a booklet, with
1
message per
page.
For
the no-mention control
group,
Message
6
stated that the homeowner's
son had been out of town and was unable to watch the house. This
condition did not include a correction. For both the correction groups,
Message 6 stated that the son was suspected of taking the homeown-
er's jewelry box to pay off gambling debts, and Message 12 corrected
this,
stating that the police no longer considered the son as a suspect
because he had been out of town when the robbery occurred. Then, in
the alternative correction
group,
Message
13
stated that the police had
caught another suspect with the jewelry. In the negating correction
group (and the no-mention control group), Message
13
stated that the
homeowner would offer a reward for return of the missing jewelry.
A memory test was also prepared, similar to the open-ended
questionnaire used in the previous experiments. It included an initial
question about the cause of the incident,
10
questions on facts directly
presented in the messages,
11
other questions requiring the subjects to
make inferences about the event, and 2 final questions assessing
whether subjects were aware of the updated information (questions
from this measure are shown in Appendix
E).
All questions appeared
in the same order for each subject, with all fact questions appearing
before any inference questions to prevent the latter from introducing
biases. The 2 questions about noticing the update always appeared
after the inference questions.
Procedure. We used the same procedure as in the previous experi-
ments, except that subjects did not fill out a summary sheet or do an
intervening task before answering the open-ended questionnaire. The
summary sheet was omitted because of time constraints and because
results from the fact recall and story recall measures were redundant
in the previous experiments.
Results
Scoring.
A
coder who
was
uninformed about the experimen-
tal conditions scored the responses to the inference questions
for a thematic-inference measure and for a global-cause
question about responsibility for the theft. Direct references
were not coded in this experiment. Two themes were consid-
ered: an inside job theme, in which the son
was
implicated, and
a
stranger-burglar
theme, in which the son was not thought to
be involved with the theft. The inside job theme encompassed
references to the son's potential involvement, using a key to
enter the house, familiarity with the homeowners or their
home, knowing where to find the jewelry, and attempts to
conceal the crime from the homeowners. References to a
stranger-burglar theme were also coded. These were refer-
ences that indicated a burglar other than the son, entering the
house through a broken window, and not having enough time
to steal further items. Table 7 shows sample thematic answers
to inference questions from the questionnaire. The analyses
reported below exclude idiosyncratic responses and items left
blank, which made up 7.7% of the responses to the inference
questions.
Inferences. Table 8 shows the mean number of thematic
inferences per subject for the inside job and stranger-burglar
themes by group. A one-way ANOVA of the number of
inferences consistent with the inside job theme showed a
significant group difference, F(\, 37) =
3.41,
p <
.045,
MSe =
14.88.
Planned comparisons showed a significant difference
between the negating correction group and the other two
groups, F(l, 37) = 5.94, p < .02, M5e = 25.89. A planned
Table 7
Sample Answers to Inference Questions in Experiment
3B by Theme
Answer
QuestionInside jobStranger-Burglar
What could the Har-
ters have done to
better avoid the
problem?
Why
wasn't the TV
taken?
Not trusted their son.
Harters would notice
it
was
missing.
How might the thief Through the door;
have gotten into the with a key.
house?
Used a safe deposit
box.
It
was
too heavy; not
enough time to take
it.
Through the broken
basement window.

CONTINUED INFLUENCE OF MISINFORMATION
1431
Table 8
Number of Thematic Inferences Consistent With Misinformation,
by Group, From Experiment 3B
Inference
Inside job
M
SD
Stranger-Burglar
M
SD
Negation
3.18
3.09
4.09
2.51
Control
1.70
1.75
6.08
1.94
Alternative
1.06
1.39
7.87
1.75
comparison between the no-mention control and the alterna-
tive
groups was not significant (F <
1).
Conversely, an ANOVA
of the number of inferences consistent with the stranger-
burglar theme also showed a significant difference, F(l, 37) =
11.31,
p <
.0001,
MSe = 46.96. Planned comparisons showed
significant differences between the negating correction and the
no-mention control group, F(l, 37) =
5.66,
p <
.023,
MSt =
23.50,
and between this control group and the alternative
group, F(l, 37) =
5.59,
p <
.023,
MSe = 23.19. As shown in
Table 8, the alternative group endorsed the stranger-burglar
theme more than the no-mention control group did, even
though they received the same initial misinformation as the
negating correction group did.
For the global-cause measure, responses to the direct
question about what was responsible for the theft were also
scored. No subject in any of the groups answered that the son
was responsible, the information initially provided before the
correction.
Recall. Fact questions from the questionnaire were scored
for accurate content. There were no group differences on this
fact recall measure (F < 1.5). Subjects recalled a mean of 8.9
facts correctly. For the correction recall measure, a subject was
scored
as
noticing the correction if
it was
referred to accurately
in the open-ended questionnaire. Only two of the subjects
failed to report the information that the son was out of town
and thus was no longer a suspect.
Discussion
The results replicate the findings in Experiment 3A and
suggest that the effects found earlier are not just an artifact of
the particular scenario used. As expected, subjects showed a
significant influence effect when the correction merely negates
previously presented information, but the discredited informa-
tion has less influence when the correction provides an
alternative explanation for events in the scenario. In fact, the
group given an alternative after the discredited information
appeared similar to control subjects who never heard the
original information. This experiment demonstrates that the
effect of adding a causal alternative to the correction is not
specific to the story used in the earlier studies because the
materials in Experiment 3B present a different story context
and different features of the account that need explaining.
Further, the type of causality differs in Experiment
3B
because
the scenario deals with an intentional agent (the son) causing
events rather than a physical substance (volatile materials)
causing events. Thus, the continued influence effect and the
mitigation of the effect when a causal alternative is included in
the correction are not due to any specific features of the
warehouse story used in Experiment 3A.
The two scenarios used in these experiments do share some
similarities that may suggest constraints on when the contin-
ued influence effect might occur and when corrections with
causal alternatives may be effective. First, both stories contain
many details that are consistent with the misinformation, even
though one can potentially explain those details in another
way. One might expect less influence in a story that contains
little information other than what was corrected or when the
story contains more information that is clearly inconsistent
with the misinformation. Second, the events in the stories used
here tended to avoid stereotypical default explanations that a
subject might readily generate. For example, an auto accident
story might tend to be associated with specific schemas in
memory (such as drunk driving). If another cause for an
accident were discredited, subjects could potentially use this
alternative schema already in memory to reconstruct the event
and avoid use of misinformation. On the other hand, one might
expect more of a continued influence effect if the event is one
for which subjects have little structured background informa-
tion available in memory. Finally, the stories in these experi-
ments do not explain the origin of the discredited information.
One generally believes that a communicator is following
conversational maxims (Grice, 1975) and thus providing accu-
rate information. If information is corrected later, and one
does not decide that the initial communicator was lying, one
may still believe that the communicator had a reason for
providing the initial information. One may then reinterpret
either the correction or the initial information and show
continued influence as a result of these reinterpretations. A
correction that also explains why a communicator might have
provided inaccurate information (e.g., honest mistake, deceit,
and
so
forth) could also reduce the continued influence effect.
General Discussion
The results of the present experiments suggest a reexamina-
tion of previous theories about how one comprehends correc-
tions in complex contexts. The results show that misinforma-
tion
is
more likely to influence inferences when it plays a causal
role in the account. In Experiment 2, subjects showed many
inferences involving volatile materials when the reports pre-
sented them as causally sufficient to account for a fire's
starting, regardless of whether a correction occurred. How-
ever, when the story context made an incidental (noncausal)
mention of the same materials, subjects recalled that informa-
tion but made very few inferences referring to those materials.
This provides evidence that a story's causal affordances can
determine the degree of influence misinformation may have.
Experiments 3A and 3B demonstrate further effects of
causal affordances. Subjects made fewer inferences involving
misinformation when the correction also presented a causal
alternative to
it.
This suggests that one
may involve
misinforma-
tion in inferences when it is the only material that affords
causal structure to the account, and without it, one could not
construct as good a representation of the event. This is
consistent with other proposals that subjects use discredited

1432HOIXYN
M.
JOHNSON AND COLLEEN M. SEIFERT
information when they do not have other "relevant" informa-
tion available (Anderson et al.,
1985;
Wegner et al., 1985), but
because the mere presence of plausible information does not
lead as readily to influence, the current studies specify rel-
evance in terms of the causal role in which the misinformation
was presented. These influence effects may be specific to texts
in which many of the details support a causal interpretation
consistent with the misinformation and the alternative pro-
vides plausible causal affordances.
The present experiments also show that two previously
proposed accounts of continued influence did not account for
the effects found here. First, the results were inconsistent with
an availability theory, which predicts that the mere presence of
misinformation in the context of the account leads to influ-
ence.
Subjects in Experiment
-2
made few inferences involving
the volatile materials information when it was presented
incidentally or primed. However, subjects in the noncausal
condition recalled the information, suggesting that it was at
least available for use. Similarly, subjects in the external
reference condition read and generated key words from the
volatile materials message other subjects saw. Previous re-
search (Schacter, 1987) has found that exposure to items
makes them more likely to appear later on tests that do not
require explicit recall of previous information. However, exter-
nal priming
was
not sufficient to result in influenced inferences
in this comprehension task. Thus, theories that focus on
availability of information (e.g., Anderson et al., 1985) do not
account for all the influence found here.
Second, the results show that theories based on an "editing
failure" assumption (e.g., Ross et
al.,
1975) also do not account
for all the continued influence found in this
case.
Such theories
suggest that one elaborates on misinformation and integrates it
into a representation before a correction occurs. Subjects
might then have trouble "editing out" these elaborations once
the discrediting occurs, and thus could retrieve them from
memory later, thereby showing the influence effect. However,
another main finding of these experiments is that influence
from discredited information can occur in inferences originat-
ing after the correction occurs. In Experiments 1A and IB,
subjects who received the correction in a message immediately
following the misinformation still showed more influence from
it than a control group did and showed the same level of
influence as a group with a longer delay between receiving the
original message and the instruction to disregard it. However,
subjects in the immediate correction group had less opportu-
nity to make precorrection inferences than those in the group
with the longer delay, who could potentially make causal and
coherence-maintaining inferences before the correction oc-
curred. This work adds to previous research that has suggested
that influence occurs as a result of problems in editing
prestored inferences (Ross et al., 1975; Wilkes & Leatherbar-
row,
1988;
Wyer
&
Budesheim,
1987) by
showing that misinfor-
mation can also influence postcorrection inferences.
One may also need to consider more complex explanations
of how people understand accounts containing corrections. In
addition to considering a story's causal affordances, as dis-
cussed above, one may want to look at the conversational
implications (Grice, 1975) an original presentation of the
information has, given that it is corrected later. One tends to
assume that communicators have reason to believe any mes-
sage they
provide,
before it
is
shown to be misinformation. One
may then have to reconcile one's belief that the communicator
was attempting to be accurate with the fact that he or she was
actually wrong. This could lead one to reinterpret either the
initial information or the correction in a way that leads to
continued influence. The materials used in this research do not
provide an explanation for how the misinformation originated
(e.g., honest mistake, deceit, and so forth); a correction doing
so might also be more effective than one that just negates
literal content presented earlier. However, our experiments
presented the stories in the context of on-line news reports
coming in from an investigation scene; consequently, subjects
may have more readily accepted updated information that
contradicted earlier reports.
Our research extends that of traditional verbal-learning
paradigms dealing with directed forgetting (e.g., Bjork &
Woodward, 1973), despite seemingly contradictory findings.
Our research parallels these directed forgetting paradigms in
the emphasis on how people deal with irrelevant information
and in subjects' initially assigned task of recall. However, the
research here finds continued influence from misinformation,
in contrast to the robust forgetting effects found in much of the
directed forgetting literature (e.g., Bjork & Woodward, 1973;
Horton & Petruk, 1980; MacLeod, 1975). The effects found
here may depend on having a coherent, causally related
account in which a single or minimal correction has a signifi-
cant impact on the construal of meaning. Directed forgetting
work using sentences (Geiselman, 1974, 1975), as opposed to
lists of unrelated
words,
has also found that subjects forget the
forget-cued sentences less effectively when they are themati-
cally related to remember-cued sentences or to each other.
The present results suggest further examination of conditions
in which people can and cannot seem to deal with irrelevant
information successfully.
Our work also differs from that on postevent misinformation
and eyewitness testimony (e.g., Loftus, Miller, & Burns, 1978).
In that paradigm, subjects witness a complete event, and then
receive new, misleading information. In the present research,
the information subjects learn initially turns out to be invalid,
and they receive new, accurate information during their
comprehension of the event. More critically, the postevent
misinformation paradigm usually presents misleading informa-
tion about a peripheral detail in the situation, resulting in
subjects being unaware that they have been misled; otherwise,
subjects do detect and reject misinformation that blatantly
contradicts their perceptions (Loftus, 1979). In contrast, the
work in this article shows influence from misinformation that is
central to one's understanding of the event, and subjects do
openly detect, accept, and accurately report the correction on
memory measures.
These results have several implications for theories of
comprehension involving
corrections.
They suggest that assert-
ing information can result in its propagation through later
inferences despite even direct, immediate corrections. This in
turn suggests that one cannot completely discredit information
in memory by merely negating the literal content expressed
earlier; instead, an effective correction must go further if it is
to result in a successful discrediting. The results also suggest

CONTINUED INFLUENCE OF MISINFORMATION1433
that the more likely a piece of information is to become
involved in causal inferences, the more likely it is to lead to
continued influence on inferences after
a
correction. Thus, one
could have more trouble eradicating a central and important
piece of information from memory if incorrect, but this would
also be the information most important to correct accurately.
The correction message should be directly stated and should
optimally provide an alternative to the misinformation in order
to improve the likelihood of accurate correction in memory.
Other potential methods for enhancing the processing of
corrections, such as ways of addressing the conversational
implications of someone initially providing misinformation,
remain to be explored.
In summary, our results support the hypothesis that contin-
ued influence from misinformation depends on its propensity
for becoming involved in causal inferences and does not
depend on simple heightened availability in memory. Further,
having a causal alternative associated with the correction of
misinformation decreases, but does not completely eliminate,
its influence. The results also show that subjects' inferences
about an event can show influence from misinformation, even
when they encounter an immediate, direct instruction to
disregard it. Finally, the results suggest that such inferences
can originate after the correction occurs in which subjects use
the misinformation at a later test time to generate new
inferences. This contrasts with previous work that has posited
continued influence due to failure to edit prestored inferences
(Anderson et
al.,
1980;
Wyer
&
Budesheim, 1987) and suggests
that two different processes may result in continued influence
from misinformation.
References
Anderson, C. A. (1982). Inoculation and counterexplanation: Debias-
ing techniques in the perseverance of social theories. Social Cogni-
tion,
1, 126-139.
Anderson, C. A. (1983). Abstract and concrete data in the persever-
ance of social theories: When weak data lead to unshakeable beliefs.
Journal of Experimental Social
Psychology,
19, 93-108.
Anderson, C. A., Lepper, M. R., & Ross, L. (1980). Perseverance of
social theories: The role of explanation in the persistence of
discredited information. Journal of Personality and Social
Psychology,
39, 1037-1049.
Anderson, C. A., New, B. L., & Speer, J. R. (1985). Argument
availability as a mediator of social theory perseverance. Social
Cognition, 3, 235-249.
Baker, L., & Wagner, J. L. (1987). Evaluating information for
truthfulness: The effects of logical subordination. Memory & Cogni-
tion,
15, 247-255.
Bjork, R. A., & Woodward, A. E. (1973). Directed forgetting of
individual words in free recall. Journal of Experimental Psychology,
99, 22-27.
Carretta, T. R., & Moreland, R. L. (1983). The direct and indirect
effects of inadmissible evidence. Journal of Applied Social Psychology,
13,
291-309.
Conway, M., & Ross, M. (1984). Getting what you want by revising
what you had. Journal of Personality and Social Psychology, 47,
738-748.
Fleming, J., & Arrowood, A. J. (1979). Information processing and the
perseverance of discredited self-perceptions. Personality & Social
Psychology Bulletin, 5, 201-205.
Geiselman, R. E. (1974). Positive forgetting of sentence material.
Memory
&
Cognition, 2(4), 677-682.
Geiselman, R. E. (1975). Semantic positive forgetting: Another
cocktail party problem. Journal of
Verbal
Learning
&
Verbal Behavior,
14(1),
73-81.
Golding, J. M., Fowler, S. B., Long, D. L., & Latta, H. (1990).
Instructions to disregard potentially useful information: The effects
of pragmatics on evaluative judgments and recall. Journal of Memory
and Language, 29, 212-227.
Graesser, A. C. (1981). Prose comprehension beyond the
word.
New
York: Springer-Verlag.
Graesser, A. C, & Clark, L. F. (1985). Structures and procedures of
implicit
knowledge.
Norwood, NJ: Ablex.
Grice, H. P. (1975). Logic and conversation. In P. Cole, & J. L. Morgan
(Eds.),
Syntax and semantics, 3: Speech acts (pp. 41-58) San Diego,
CA: Academic Press.
Hastie, R., & Park, B. (1986). The relationship between memory and
judgment depends on whether the judgment task is memory-based
or on-line.
Psychological
Review,
93, 258-268.
Haviland, S. E., & Clark, H. H. (1974). What's new? Acquiring new
information as a process in comprehension. Journal of Verbal
Learning and Verbal Behavior, 13,
512-521.
Hertel, P. T. (1982). Remembering reactions and facts: The influence
of subsequent information. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 8, 513-529.
Horton, K. D., & Petruk, R. (1980). Set differentiation and depth of
processing in the directed forgetting paradigm. Journal of Experimen-
tal
Psychology:
Human Learning and Memory, 6(5), 599-610.
Jennings, D., Lepper, M. R., & Ross, L. (1981). Persistence of
impressions of personal persuasiveness: Perseverance of erroneous
self-assessments outside the debriefing paradigm. Personality &
Social
Psychology
Bulletin, 7, 257-263.
Kintsch, W. (1974). The
representation
of meaning in memory. Hillsdale,
NJ: Erlbaum.
Kintsch, W. (1988). The role of knowledge in discourse comprehen-
sion: A construction-integration model. Psychological Review, 95,
162-182.
Loftus, E. F. (1979). Reactions to blatantly contradictory information.
Memory
&
Cognition, 7, 368-374.
Loftus, E. F., Miller, D. G., & Burns, H. J. (1978). Semantic
integration of verbal information into a visual memory. Journal of
Experimental
Psychology:
Human Learning and
Memory,
4,
19-31.
MacLeod, C. M. (1975). Long-term recognition and recall following
directed forgetting. Journal of Experimental Psychology: Human
Learning and Memory, 104(1), 271-279.
Massad, C. M., Hubbard, M., & Newtson, D. (1979). Selective
perception of events. Journal of Experimental Social
Psychology,
15,
513-532.
McKoon, G., &
Ratcliff,
R. (1992). Inference during reading. Psycho-
logical
Review,
99, 440-466.
O'Brien, E. J., & Albrecht, J. E. (1991). The role of context in
accessing antecedents in text. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 17, 94-102.
O'Brien, E. J., & Myers, J. L. (1987). The role of causal connections in
the retrieval of text. Memory
&
Cognition, 15, 419-427.
Ross,
L., Lepper, M. R., & Hubbard, M. (1975). Perseverance in
self-perception and social perception: Biased attributional pro-
cesses in the debriefing paradigm. Journal of Personality and Social
Psychology,
32, 880-892.
Schacter, D. (1987). Implicit memory: History and current status.
Journal of Experimental
Psychology:
Learning, Memory, and Cognition,
13,
501-518.
Schul, Y., & Burnstein, E. (1985). When discounting fails: Conditions
under which individuals use discredited information in making a
judgment. Journal of Personality and Social
Psychology,
49, 894-903.

1434HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
Spiro,
R. J. (1980). Accommodative reconstruction in prose recall.
Journal
of Verbal
Learning and
Verbal
Behavior,
19,
84-95.
Trabasso, T.,
&
van den Broek, P. W. (1985). Causal thinking and the
representation of narrative events. Journal of
Memory
and Language,
24,
612-630.
Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for
judging frequency and probability.
Cognitive
Psychology,
5, 207-232.
van den Broek, P.
(1990).
Causal inferences and the comprehension of
narrative texts. In Graesser, A. C, & Bower, G. (Eds.), The
psychology
of
learning and
motivation:
Inferences
and
text
comprehen-
sion
(Vol.
25,
pp.
175-196). San Diego,
CA:
Academic Press.
Wegner, D. M, Coulton, G. F., &
Wenzlaff,
R. (1985). The transpar-
ency of denial: Briefing in the debriefing paradigm. Journal of
Personality and Social
Psychology,
49,
338-345.
Wilkes, A. L., & Leatherbarrow, M. (1988). Editing episodic memory
following the identification of error. Quarterly
Journal
of Experimen-
tal
Psychology,
40A,
361-387.
Wyer, R. S., & Budesheim, T. L. (1987). Person memory and
judgments: The impact of information that one is told to disregard.
Journal of
Personality and
Social
Psychology,
53,
14-29.
Wyer, R. S., & Srull, T. K. (1986). Human cognition in its social
context.
Psychological
Review,
93,
322-359.
Wyer, R. S., & Unverzagt, W. H. (1985). Effects of instructions to
disregard information on its subsequent recall and use in making
judgments.
Journal
of
Personality
and Social
Psychology,
48,
533-549.
Appendix A
Materials Adapted From Wilkes and Leatherbarrow (1988)
Message 1
Jan. 25th 8:58
p.m.
Alarm call received from premises of
a
wholesale
stationery warehouse. Premises consist of offices, display room, and
storage hall.
Message 2
A serious fire
was
reported in the storage hall, already out of control
and requiring instant response. Fire engine dispatched at 9:00 p.m.
Message 3
The alarm was raised by the night security guard, who had smelled
smoke and gone to investigate.
Message 4
Jan. 26th 4:00 a.m. Attending fire captain suggests that the fire was
started by a short circuit in the wiring of a closet off the main storage
hall. Police now investigating.
Message 5 (Mention and Discrediting Groups)M
4:30 a.m. Message received from Police Investigator Lucas saying
that they have reports that cans of oil paint and pressurized gas
cylinders had been present in the closet before the fire.
Message 5 (Control)*1
4:30 a.m. Message received from Police Investigator Lucas saying
that they have reports that the closet was empty before the fire.
Message 6
Firefighters attending the scene report thick, oily smoke and sheets
of flames hampering their efforts, and an intense heat that made the
fire particularly difficult to bring under control.
Message 7
It has been learned that a number of explosions occurred during the
blaze, which endangered
firefighters
in the vicinity. No fatalities were
reported.
Message 8
Two
firefighters
are reported to have been taken to the hospital as a
result of breathing toxic fumes that built up in the area in which they
were working.
Message 9
A small fire had been discovered on the same premises, six months
previously. It had been successfully tackled by the workers themselves.
Message 10
10:00 a.m. The owner of the affected premises estimates that total
damage will amount to hundreds of thousands of
dollars,
although the
premises were insured.
Message 11 (Control)*1
10:40 a.m. A second message received from Police Investigator
Lucas regarding the investigation into the fire. It stated that the two
firefighters taken to the hospital had been released.
Message 11 (Discrediting Groups)*1
10:40 a.m. A second message received from Police Investigator
Lucas regarding the investigation into the fire. It stated that the closet
reportedly containing cans of oil paint and gas cylinders had actually
been empty before the fire.
Message 12
The shipping supervisor has disclosed that the storage hall con-
tained bales of paper; mailing and legal-size envelopes; scissors,
pencils, and other school supplies; and a large number of photo-
copying machines.
Message 13
11:30 a.m. Attending fire captain reports that the fire
is
now out and
that the storage hall has been completely gutted.
Note. The materials are adapted from "Editing Episodic Memory
Following the Identification of Error" by A. L. Wilkes and M.
Leatherbarrow, 1988,
Quarterly Journal
of
Experimental
Psychology,
40A,
pp. 381-383. Copyright 1988 by The Experimental Psychology
Society. Adapted by permission
A1 These are the messages that changed in the different conditions.

CONTINUED INFLUENCE OF MISINFORMATION
1435
Appendix
B
Comprehension
Questions
Fact Questions
1.
What
was
the extent of the
firm's
premises?
2.
Where did an attending
firefighter
think the fire started?
3.
Where on the premises
was
the fire located?
4.
What features of the
fire
were noted
by
the security guard?
5.
What business
was
the
firm
in?
6. When
was
the fire engine dispatched?
7.
What was in the storage hall?
8. What
was
the cost of the damage done?
9.
How was
it thought the fire started?
10.
When
was
the fire eventually put out?
Inference Questions
1.
Why did the fire spread
so
quickly?
2.
For what reason might an insurance claim be refused?
3.
What
was
the possible cause of the toxic fumes?
4.
What
was
the relevance of the closet?
5.
What aspect of the fire might the police want to continue
investigating?
6. Why
do
you think the
fire
was particularly intense?
7.
What is the most likely cause of the fire that the workers
successfully put out earlier?
8. What could have caused the explosions?
9. Where
was
the probable location of the explosions?
10.
Is there
any
evidence of careless management?
Manipulation Check Questions
1.
What
was
the point of the second message from the police?
2.
Were you aware of
any
corrections in the reports that you read?
Appendix
C
New Messages
Added to Story
New Message Added in Experiment 2
Message
3
(noncausal reference
group)01
The alarm was raised by
the manager of
a
store located across the street, which sells cans of oil
paint and pressurized gas cylinders. He had smelled smoke and
become concerned.
New
Messages
Added in Experiment 3
Message 5. The fire marshal had recorded several fire code
violations on the premises at a surprise inspection two months earlier.
Message
7.
The display room
was
reported to contain display cases,
catalogs, and the sales staffs' desks. It was only staffed from
11
a.m. to
2
p.m.,
due to diminishing sales.
Message
14
(alternative
and arson
control
groups).01
11:08 a.m.
Firefighters have found evidence of gasoline-soaked rags near where
the bales of paper had been stored in the storage hall, as well
as
several
emptied steel drums of suspicious nature. The owner denies any
knowledge of these materials.2
clThese are the messages that changed in the different conditions.
Appendix
D
Jewelry
Theft Story
Message 1
3:00 p.m., May 2nd. Police respond to a call made from a home on
Acorn St., in a middle-class residential neighborhood.
Message 2
The homeowner,
Ms.
Harter, reports that her jewelry
box is
missing.
Contents are reported to include gold chains, gold and silver earrings,
rings,
and pendants of precious stones.
Message 3
She discovered that the box was missing when she returned from a
vacation and wanted to put a new necklace she'd bought in it. It had
been stored in a locked drawer in her bedroom dresser.
Message 4
She swears she had checked the box before leaving on vacation, and
everything
was
in order. A tall tree arches near the bedroom window,
but police have found no evidence of tampering with the window.
Message 5
The Harters report that they had asked their son, Evan, to check in
on the house periodically during their absence. The son also did other
odd
jobs
for many of the neighbors.
(Appendixes continue on
next page)

1436HOLLYN M. JOHNSON AND COLLEEN M. SEIFERT
Message 6 (Control)™
Police have found that Evan had been called away on business and
had not been in town to look after the house during the Harters'
vacation.
Message 6 (Correction)01
Police suspect that Evan may have taken the box from the house to
help
pay
off large gambling debts.
Message 7
The neighborhood has been hit with a number of thefts recently.
There are no arrests or leads in these cases so far.
Message 8
The Harters' next-door neighbor reports that she noticed a light on
in the house, after her dog suddenly began barking late Saturday
evening, April 28th. An unfamiliar dark-colored car had been parked
in a nearby alley.
Message 9
A search for footprints and tire tracks has turned up inconclusive,
due to a recent rainstorm. In the course of the investigation, an officer
noted a broken latch on a basement window.
Message 10
Police are still attempting to determine whether other valuables are
missing from the house. The television and a home computer had not
been disturbed, however.
Message 11
The Harters have contacted their insurance company about the loss.
The last appraisal showed the box's contents to be worth several
thousand dollars.
Message 12 (Control)01
A second message from the police investigators about the incident.
It stated that several officers had checked the main pawn shops in
town, and confirmed that no jewelry matching the Harters' descrip-
tions has surfaced yet.
Message 12 (Correction)01
A second message from the police investigators about the incident.
It stated that the Harters' son is no longer a suspect, because several
independent sources confirm that he had been out of town on business
during the Harters' vacation.
Message 13 (Control)01
Ms.
Harter is considering offering a reward for return of several of
the pieces, because they have great sentimental value for her. There
would be no questions asked.
Message 13 (Alternative)01
Police have now arrested excon Dan Fowler, who had tried to pawn
some of the jewelry. His girlfriend, Ana, had told him about Ms.
Harters' jewels, which she had heard about while working as a new
maid in the neighbor's house.
Message 14
Detectives will look for similarities between this case and the other
thefts reported in the neighborhood recently.3
D1These are the messages that changed in the different conditions.
Appendix E
Comprehension Questions
Cause Question
1.
What caused the box to be missing from the Harters' home?
Fact Questions
2.
How much did an appraisal show the box's contents to be
worth?
3.
Where
was
the Harters' home located?
4.
Where was the jewelry
box
normally kept?
5.
Why did
Ms.
Harter consider offering a reward?
6. What did the Harters' next-door neighbor notice?
7.
What kinds of jewelry did the
box
contain?
8. What arrangements did the Harters make for checking up on
the house?
9. When did
Ms.
Harter discover that the jewelry
box was
missing?
10.
What did police notice about the bedroom window?
11.
When did the neighbor's dog suddenly start barking?
Inference Questions
12.
Why might the neighbor's dog have been barking?
13.
Whose car might the neighbor
have
noticed, parked in the alley?
14.
Why might the son feel bad about the incident?
15.
What could the Harters have done to better avoid this problem?
16.
Who, if anyone, should be questioned more thoroughly by the
police?
17.
Why
wasn't the television taken?
18.
How might the thief
have
gotten into the house?
19.
Why
might the Harters be angry with their son?
20.
What might be responsible for the other thefts in the neighbor-
hood recently?
21.
What steps should the police take next?
22.
Where
was
Evan Harter
on
the evening of April 28th?
Manipulation Check Questions
23.
What did the police investigators report about where Evan
Harter
was
during the Harters' vacation?
24.
What facts about the case did the police change their minds
about, based on information they discovered later?
Received July 27,1992
Revision received December 23,1993
Accepted December 30,1993 •