The formation of flashbulb memories

Abstract

A large group of subjects took part in a multinational test-retest study to investigate the formation of flashbulb (FB) memories for learning the news of the resignation of the British prime minister, Margaret Thatcher. Over 86% of the U.K. subjects were found to have FB memories nearly 1 year after the resignation; their memory reports were characterized by spontaneous, accurate, and full recall of event details, including minutiae. In contrast, less than 29% of the non-U.K. subjects had FB memories 1 year later; memory reports in this group were characterized by forgetting, reconstructive errors, and confabulatory responses. A causal analysis of secondary variables showed that the formation of FB memories was primarily associated with the level of importance attached to the event and level of affective response to the news. These findings lend some support to the study by R. Brown and Kulik (1977), who suggest that FB memories may constitute a class of autobiographical memories distinguished by some form of preferential encoding.

Full text

Memory &Cognition
1994, 22
(3),
326-343
The
formation
of
flashbulb memories
MARTIN
A.
CONWAY
and
STEPHEN
J. ANDERSON
University
of
Bristol, Bristol, England
STEEN
F. LARSEN
University
of
Aarhus, Aarhus, Denmark
C.
M.
DONNELLY
and
M.
A.
McDANIEL
Purdue University, West Lafayette, Indiana
A.
G.
R.
McCLELLAND
and
R.
E. RAWLES
University College London, London,
England
and
R.
H. LOGIE
University
of
Aberdeen, Aberdeen, Scotland
A
large
group
of
subjects took
part
in
a
multinational
test-retest
study
to
investigate
the
for-
mation
offlashbulb (FB) memories for
learning
the
news of
the
resignation
of
the
British
prime
minister,
Margaret
Thatcher.
Over 86%
of
the
U.K. subjects were found to
have
FB
memories
nearly
1
year
after
the
resignation;
their
memory reports were characterized by spontaneous,
accurate,
and
full recall of
event
details, including minutiae.
In
contrast,
less
than
29% of
the
non-U.K. subjects
had
FB
memories 1
year
later;
memory
reports
in
this
group were character-
ized by forgetting, reconstructive errors,
and
confabulatory responses. Acausal
analysis
of sec-
ondary
variables
showed
that
the
formation
of
FB
memories
was
primarily
associated
with
the
level
of
importance
attached
to
the
event
and
level of affective response to
the
news. These find-
ings
lend
some
support
to
the
study
by R.
Brown
and
Kulik
(1977), who suggest
that
FB
memo-
ries may constitute aclass ofautobiograpmcal memories distinguished by some form
of
preferential
encoding.
In
this paper we present findings from alarge-scale mul-
tinational test-retest study
of
flashbulb (FB) memories for
the abrupt and unexpected resignation
of
the British prime
minister, Margaret Thatcher. The resignation provided
aunique opportunity to examine FB memories (R. Brown
&
Kulik:,
1977) in groups
of
U.K. and non-U.K. nationals.
The primary purpose
of
the study was to investigate the
determinants
of
FB memories, and measures were taken
of
the variables affecting encoding and rehearsal.
We
planned to explore the structural relations between encod-
ing factors such as affect,
prior
knowledge, and conse-
quentiality, and the postencoding factor
of
rehearsal, for
both FB and non-FB memories. Before turning to details
of
the study, we frrst consider R. Brown and Kulik's origi-
nal proposals regarding the formation
of
FB memories
and then review other studies
of
FB memories.
This research was supported
by
each
of
the institutions to which the
individual authors are affiliated. Additional data were collected
by
Peter
Hayes, Cilia Morris, Peter Morris, and Stephen Dewhurst. Susan Gather-
cole and Philip Levy advised us on parts
of
the analyses and Gillian
Cohen commented on an earlier draft
of
the paper. We thank them for
their assistance. Address correspondence
to
M. A. Conway, Univer-
sity
of
Bristol, Department
of
Psychology, 8Woodland Road, Bristol,
BS8
ITN
England.
Consequentiality,
Surprise,
and
the
Formation
of
Flashbulb
Memories
Throughout the present account
of
FB memories, we
distinguish between an original event, such as the assas-
sination
of
apresident, and
one's
personal circumstances
when learning
of
the original event, which we will refer
to as the reception event (Larsen, 1988). Two critical com-
ponents in R. Brown and Kulik's (1977) model
of
FB
memories are that the original event should be both sur-
prising and consequential. Indeed, the central tenet
of
R. Brown and Kulik:' smodel
is
that an individual will have
adetailed and durable FB memory to the extent that they
experience surprise and perceive anews item to be con-
sequential. R. Brown and
Kulik:
did not measure the levels
of
surprise associated with the 10 events they sampled,
although because these were generally unexpected pub-
lic events
(i.e.,
the assassination
of
various political
leaders), it seems reasonable to assume that the news
of
these events probably was surprising. Brown and Kulik
did, however, explicitly assess the perceived consequen-
tiality
of
the news events and instructed their subjects that
"Probably
the best single question to ask yourself in rat-
ing consequentiality is
'what
consequences for my life,
both direct and indirect, has this event
had?'"
(R. Brown
&Kulik, 1977, p. 82). They further instructed their sub-
Copyright 1994 Psychonomic Society, Inc. 326

jects to consider how their lives might have progressed
if
the original event had not occurred. Finally, they ac-
knowledged that "personal" consequentiality
is
higWy
re-
lated
to
the personal importance that
an
individual attaches
to an event.
In addition to this self-assessed measure
of
consequen-
tiality/importance, R. Brown and Kulik (1977) also se-
lected their news events so that some
of
them would be
consequential and/or important for some
of
their subjects,
but not for others. They found that when an event was
consequential
or
important for one group, it was as-
sociated with acomparatively high incidence
of
FB
mem-
ories; however, the same event, when judged lower in
consequentiality/importance by another group,
was
found
to be associated with areliably lower incidence
of
FB
memories. For example, the assassination
of
Martin
Luther King was associated with ahigh incidence
of
FB
memories among black North Americans, compared with
asignificantly lower incidence for white North Ameri-
cans. Compare this with FB memories for the attempted
assassination
of
George Wallace, in which 50%
of
the
white North Americans had FB memories compared with
25
%
of
the black North Americans. For events
of
equal
significance to both groups, such as the assassination
of
President John F. Kennedy, there were no differences in
the generally high incidence
of
FB memories. Thus,
R. Brown and Kulik established that personal consequen-
tiality
or
personal importance
of
anews event
is
acriti-
cal determinant
of
FB memory formation. Finally, Rubin
and Kozin (1984), in awider survey ofhigWy vivid mem-
ories, found that increased personal importance was
closely associated with increased memory clarity, lend-
ing additional support to R. Brown and Kulik's empha-
sis on consequentiality/importance
as
one
of
the critical
variables in FB memory formation.
In summary then, according to R. Brown and Kulik's
(1977) account
of
FB memory formation, the original
news event must be (optimally) surprising and must en-
tail personal consequences.
If
these conditions are met,
then adetailed and stable memory
of
the reception event
is formed. But how detailed? R. Brown and Kulik found
that their subjects frequently recalled where they were,
who they were with, what they were doing, how they felt,
the reactions
of
others, and what happened after the news
was announced. Retention
of
this level
of
detail is in it-
selfunusual, because such details, even for events
ofna-
tional significance, are usually rapidly forgotten over
retention intervals measured in days or weeks (Larsen,
1992). The retention intervals in R. Brown and Kulik's
study were on a scale
of
years and decades. They also
found that many
of
their subjects frequently retained
knowledge
of
apparently trivial details from areception
event, such
as
the'
'feel"
of
the soles
of
aparticular pair
of
shoes, the brand name
of
adiscarded pack
of
cigarettes,
and so forth. However, they also noted that by no means
were all the details from areception event retained. For
example, in an account
of
his own FB memory for learn-
ing
of
J.F.K.'s
assassination, R. Brown states that he had
FLASHBULB MEMORY 327
been on the telephone to the dean's secretary about some
"forgotten"
business. Thus, although FB memories for
the reception events
of
learning surprising and consequen-
tial
or
important news are unusually detailed and durable
for
that class
of
reception event (see Larsen, 1992), they
are by no means complete records
of
such events.
Recent Studies
of
Flashbulb Memories
Studies
of
the FB memory concept must, then, focus
on memory for events that are
higWy
surprising and con-
sequential or important. Several researchers (Bohannon,
1988; Neisser &Harsch, 1992; McCloskey, Wible, &
Cohen, 1988), who have generally been critical
of
the FB
memory concept, have focused on FB memories
of
the
space shuttle Challenger disaster. McCloskey et al. and
Neisser and Harsch used test-retest designs in which a
group
of
subjects are tested close to the time
of
the recep-
tion event, and then asubset
of
these subjects are retested
after aretention interval
of
months or years. These sub-
jects typically provide afree description
of
the reception
event and then answer specific questions
as
to who they
were with, what they were doing, location, time
of
day,
and so on. The critical measure is consistency in mem-
ory details over the two test sessions. Very high con-
sistency is an indication
of
abasically accurate FB mem-
ory formed during
or
shortly after the reception event,
which
is
accessed on both test occasions. Deviation from
this very high level
of
consistency in repeated memory
descriptions
is
construed as denoting incomplete memo-
ries that have not endured over time, and which are not
FB memories.
McCloskey et al. (1988) found that the majority
of
a
group
of
29 subjects retested over a9-month retention
interval actually had FB memories. From their Table 1
(McCloskeyet al., 1988, p. 173), it appears that at least
89 %
of
their subjects could, 9months later, accurately
recall location, activity, source
of
the news, and their reac-
tion (first thoughts) upon first learning
of
the Challenger
disaster.
In
contrast, Neisser and Harsch (1992) found that
very few
of
their 44 subjects had FB memories after a
retention interval
of
32 to 34 months. Bohannon (1988),
in alarge-scale
FB
memory study
of
the Challenger dis-
aster, sampled two independent groups
of
subjects-one
close to the date ofthe disaster and another 8months later.
As with Neisser and Harsch, very few
of
Bohannon's sub-
jects actually had FB memories
of
the disaster.
If
we con-
sider only those subjects who could recall the source
of
the news, ongoing activity, and place (a very lenient cri-
teria for coding an FB memory), then it appears that less
than 35%
of
Bohannon's subjects had detailed and dura-
ble memories (see Bohannon, 1988, Table 2, p. 186).
One feature that
is
common to all three
of
these studies
is that the surprise value and consequentiality/importance
levels
of
the Challenger disaster were assumed rather than
measured. But what
if
these assumptions were unwar-
ranted? Consider the following. Suppose that, for the sub-
jects in the studies by Neisser and Harsch (1992) and
Bohannon (1988), the Challenger disaster was only mildly

328 CONWAY
ET
AL.
surprising and
of
little
or
no consequence
or
personal im-
portance. This then would completely explain why such
alow incidence
of
FB memories was observed in these
studies-the
levels
of
surprise and consequentiality that
are critical to FB memory formation were not reached
during the reception event and, therefore, no FB memo-
ries were formed. Asimilar argument can be applied to
those few non-FB memory subjects
in
McCloskey et al.
's
(1988) study. Moreover, for those subjects who did show
evidence
of
detailed and durable memories in all three
studies, presumably, the levels
of
surprise and consequen-
tiality at encoding passed some critical threshold and FB
memories were formed.
It
is even possible that these cross-study differences in
the incidence
of
FB
memories reflect group differences
that are similar to those explicitly assessed by R. Brown
and Kulik (1977). In McCloskey et al. (1988), the sub-
ject sample
of
faculty staff must have been considerably
older than the undergraduate subjects used by Neisser and
Harsch (1992) and Bohannon (1988). Perhaps for the older
group
of
subjects the space program was generally
of
high
personal relevance and amore integral part
of
their per-
sonal past. For the younger subjects, however, the space
program may have been low,
or
lower, in personal rele-
vance, more remote, and less obviously apart
of
shared
cultural experience.
If
this were the case, then it might
be expected that the Challenger disaster would be per-
ceived to be more consequential by older than by youn-
ger subjects, and so ahigher incidence
of
FB
memories
would be present in the former than in the latter group.
Whatever the case, because measures
of
surprise and con-
sequentiality/importance were not taken,
FB
memory find-
ings from the Challenger studies are ambiguous. This
event may have been surprising and consequential for
some subjects but not for others, and hence the pattern
of
disparate findings across studies.
R. Brown and Kulik (1977) proposed that levels
of
sur-
prise, consequentiality/importance, and affect are prob-
ably interrelated to at least some degree. Thus, amea-
sure
of
intensity
of
emotional reaction to learning
of
the
Challenger disaster might suffice as an indirect indicator
of
levels
of
surprise and consequentiality/importance.
McCloskeyet al. (1988) did not take adirect measure
of
intensity
of
affect and, although Neisser and Harsch
(1992) derived such ameasure, it was based on the num-
ber
of
negative-emotion words that asubject employed
when answering the question
"How
did you feel about
the news?" Bohannon (1988) did, however, employ a
direct measure
of
emotional intensity in the form
of
a5-
point rating scale and found that subjects reported only
moderate levels
of
affect in response to learning the news
of
the disaster. This finding
is
revealing and demonstrates
that, in alarge sample
of
subjects (n =687), news
of
the Challenger disaster did not cause widespread and
strong emotional reactions. To the extent that surprise,
consequentiality/importance, and affect are all associated
(i.e., share common variance), this finding implies that
the Challenger disaster would, at best, have generated only
moderate levels
of
surprise
and
consequentiality/importance
and, therefore, acorrespondingly low incidence
of
FB
memories
is
to be expected.
In contrast to the studies on the Challenger disaster,
other
FB
memory studies have been attempts to directly
assess the critical variables
of
surprise and consequen-
tiality/importance. Christianson (1989) conducted a
test-retest study (n =36)
of
the assassination
of
the
Swedish prime minister, OlofPalme. Subjects completed
an
FB
memory questionnaire, similar to those described
previously, 6weeks after the assassination and again 52
to 54 weeks later. In addition to providing memory de-
tails, the subjects also rated how
"upsetting"
and sur-
prising they had found the news.
It
was found that the
subjects generally had anegative emotional reaction to
the news and that all
of
the subjects were extremely sur-
prised. On a
"lenient"
scoring criteria (i.e., asubject's
response was "basically" rather than
"exactly"
correct),
it was found that over 90%
of
Christianson's subjects had
FB memories. However, when astricter scoring criteria
was used, which required the subjects to be exactly cor-
rect in their retest responses, the incidence
of
FB memo-
ries fell to just over 50
%.
The secondary variables
of
emotion and surprise were not generally related to the
incidence
of
FB memories, with the exception that, on
the lenient scoring criteria, the subjects who were most
surprised were reliably more consistent in their memory
reports than were those who were somewhat less sur-
prised.
One problem with this study was that Christianson's
subjects did not judge the consequentiality/importance
of
the event, and so it is not known whether
or
not this cru-
cial factor influenced FB memory formation. For instance,
aperson may have been "shocked and surprised" (see
Neisser &Harsch, 1992) by the news
of
Pal
me's
murder,
but the event may nonetheless have been
of
little personal
consequence.
If
this was the case for Christianson's sub-
jects, then
of
course FB memory formation would be at-
tenuated. Amuch more important problem relates to the
sensitivity
of
the measures
of
intensity
of
emotion and sur-
prise. Recall that the subjects were first questioned only
6weeks after learning
of
the news; therefore, the sensi-
tivity
of
these measures crucially depended upon their abil-
ity to accurately recall the degree to which they were up-
set and surprised.
If
the subjects could not do this, then
their judgments would not be accurate and would be un-
likely to be related to memory consistency. In connec-
tion with this, Pillemer (1984) found that ratings
of
af-
fect and surprise taken 6months after an FB memory
event were unrelated to memory consistency, which
strongly suggests that such measures must be taken rea-
sonably close to the time
of
the event. In Pillemer's study,
ratings taken within 1month proved to be reliably as-
sociated with memory consistency. The 6-week delay in
Christianson's study may, then, have been associated with
alowering
of
the effectiveness
of
these rating scales.
The study that comes closest to directly and effectively
assessing
R.
Brown and Kulik's (1977) FB memory con-

cept was reported
by
Pillemer (1984). Pillemer inves-
tigated subjects' memories for the attempted assassina-
tion of President Ronald Reagan
by
having asingle group
(n =44) complete an
FB
memory questionnaire Imonth
after the shooting and again
6Y2
months later. Apart from
completing free-memory descriptions
of
the reception
event and answers to probe questions, the subjects also
completed aseries
of
rating scales assessing surprise, con-
sequentiality, emotion, and opinions
of
Reagan. Pillemer
found that the majority
of
his subjects had FB memories
and that memory descriptions were highly consistent over
the two testing sessions. Moreover, the only two variables
found to
be
reliably and positively associated with
FB
memories were emotion and surprise. The subjects' rat-
ings indicated only moderate levels
of
emotion and sur-
prise. In contrast, consequentiality, which was assessed
by aquestion asking the subjects to judge the
"impact"
of
the news, was very low and unrelated to memory con-
sistency. It
is
far from clear that ratings
of
"impact"
as-
sess R. Brown and Kulik's notion
of
personal conse-
quences; nevertheless, it would appear that the attempted
assassination
of
President Reagan, although low in im-
pact, gave rise to the formation
of
FB memories and that
this process was closely associated with affect and sur-
prise so that the more intense the experience
of
affect and
surprise, the more consistent the memory.
Flashbulb Memories for the Resignation
of Prime Minister Margaret Thatcher
In the preceding section it was shown that anumber
of
methodological problems specific to individual studies
have prevented the direct evaluation
of
the process
of
FB
memory formation
as
originally proposed by
R.
Brown
and Kulik (1977). More generally, past researchers (with
the exception
of
Bohannon, 1988) have employed only
small groups
of
subjects. This
is
aproblem because it
precludes the use
of
multivariate analyses, which are es-
sential
if
latent constructs such as FB memories, affect,
consequentiality/importance, rehearsal, and the relations
between such constructs are to be assessed. Finally, pre-
vious researchers have not followed
R.
Brown and Kulik's
lead
of
employing different groups
of
subjects who can
be compared for memory consistency across different
levels
of
surprise, consequentiality/importance, and so
forth. In the present study, we attempted to rectify these
problems
by
sampling large groups
of
subjects
of
differ-
ent nationalities within
14
days
of
the target event and
again
11
months later. We based the design
of
our study
on Pillemer (1984) and included scales that directly as-
sessed the importance
of
the event both personally and
nationally, affect, prior knowledge
of
politics and orien-
tation to the event, and levels
of
rehearsal. The target
event was the resignation
of
Mrs. Thatcher from the post
of
prime minister, which was announced on Thursday,
November 22, 1990, at approximately 10:30 a.m.I
It
was
predicted, following R. Brown and Kulik, that higher
affect and consequentiality (hereafter referred to
as
im-
FLASHBULB MEMORY 329
portance) would be closely associated with FB memory
formation.
METHOD
Design
The main variable was nationality
of
the subject samples; there
were two levels. The U.K. group consisted
of
subjects sampled from
various sites
in
the United Kingdom, and the non-U.K. group con-
sisted
of
subjects who were not British and who were not residents
of
the United Kingdom at the time
of
the resignation. All the sub-
jects completed
an
FB memory questionnaire (FBQ) within
14
days
of
the resignation and again after aretention interval
of
11
months.
The main measure was each subject's FB memory score, calcu-
lated from their responses to the FBQ. In two sets
of
secondary
measures we examined the aspects
of
encoding and rehearsal. The
encoding measures were further subdivided into sections in which
anumber
of
variables assessed affect, importance, and prior knowl-
edge. Order
of
presentation
of
these sections on the FB memory
questionnaire was the same for all the subjects. At the II-month
test interval, confidence ratings for some
of
the FBQ items were
also collected. Part 1
of
the questionnaire required adescription
of
the reception event, Part 2assessed specific aspects
of
the recep-
tion event, including original affect and subsequent rehearsal, and
the purpose
of
Part 3was to gather information on importance and
prior knowledge.
Subjects
Three hundred and sixty-nine subjects took
part-215
in
the U.K.
group and 154 in the non-U.K. group.2 The U.K. group was
recruited from undergraduate populations at the universities
of
Aber-
deen, Lancaster, and London. The majority
of
subjects (over 90%)
were first-year psychology undergraduates, and
of
these over 70%
were tested as part
of
aclass exercise. Other undergraduates were
contacted by internal mail. The non-U.K. group was recruited from
North America (Purdue University), Denmark (Aarhus University),
and from avariety
of
other countries. The North American sample
constituted
95
%
of
this group, the Danish sample 4
%,
and the others
I
%.
The non-U.K. group
was
primarily composed
of
first-year psy-
chology undergraduates who were contacted and tested in the same
way
as
the U.K. group. The majority (80%)
of
both U.K. and non-
U.K. subjects were tested within 10 days
of
the resignation, and
the remainder were tested within
14
days. The subjects were retested
during late October and early November
of
1991, between 333 and
350 days after the resignation. The testing
of
the U.K. and non-
U.K. groups was evenly distributed over this 27-day interval.
The Flashbulb Memory Questionnaire
The FBQ was modeled on asimilar questionnaire designed by
Pillemer (1984); Appendix Ashows the composition
of
the FBQ
used in the present study.3The first question on the FBQ required
the subjects to respond
"yes"
or
"no"
to the question
"Do
you
recall the circumstances
in
which you first heard
of
the resignation
of
Margaret Thatcher?" The subjects who responded "yes" then
went on to complete all parts
of
the FBQ. Those who responded
"no"
skipped the memory description in Part
1,
answered
any
ques-
tions they could
in
Part 2(guessing where necessary), and answered
all the questions in Part 3. For the memory description in Part
I,
the subjects were instructed to write ashort description
of
the recep-
tion
event-about
aparagraph in length. Following this, space was
provided for them to list any other memories that they had
of
the
time that they had learned
of
the resignation. The subjects were
not given any direction about the type
of
memories that might come
to mind, and were simply asked to write ashort description
of
the
content
of
any memories that spontaneously occurred to them.4

330 CONWAY ET AL.
Procedure
All the subjects were provided with a copy
of
the FBQ either
directly
by
the experimenter or
in
the
mail
with an accompanying
letter, which requested that they complete the enclosed question-
naire and not discuss the contents
or
their answers with other stu-
dents. The front cover
of
the FBQ contained an introduction that
informed the subjects that the study focused on memory for public
events and,
in
particular, on memory for one's personal circum-
stances when learning the news
of
amajor public event. The ex-
ample
of
the assassination
of
John F. Kennedy was provided and
it was explained to the subjects that many people over the age
of
40 years could recall who they were with, what they were doing,
and where they were when they first heard the news
of
the assassi-
nation. The subjects were told that amajor public event that had
recently occurred in Britain was identified on the following page
and that their task was to try to recall their personal circumstances
when they had learned
of
this news. The subjects then continued
on with the various sections
of
the questionnaire, taking between
20 and 30 min to complete
it.
At retest they were informed that
they had completed asimilar questionnaire Iyear previously and
that the current questionnaire was afollow-up to the earlier test.
Finally, asmall subsample
of
subjects, all Lancaster University
undergraduates, were tested for athird time
in
February
1993,26
months after Thatcher's resignation. Thirty-three subjects, about
9%
of
the total retest sample, and all Lancaster University final-
year psychology students, took part in this third phase
of
the study.
These subjects again completed the retest FBQ.
RESULTS
The results are divided into four sections in the follow-
ing text. In the first section, the scoring
of
FB memories
and the variables selected for analysis are described. Anal-
yses
of
the incidence and accuracy
of
FB
memories are
reported next. In the third section, analyses
of
the sec-
ondary variables are reported, and in the fourth section
we report regression analyses and acausal model
of
FB
memory formation.
Scoring Flashbulb Memories
In scoring the memory data, we followed aprocedure
developed by Neisser and Harsch (1992). In this proce-
dure amemory attribute such as
"place"
is assigned a
score
of
0,
1,
or
2. Ascore
of
0indicates that the subject
either forgot the attribute (i.e., did not complete that ques-
tion in the FBQ)
or
entered adifferent attribute at retest
(e.g., originally answered
"in
my office at
work"
and
at retest responded "watching TV in
my
living room at
home"). Ascore
of
1 denotes a basically, but not exactly,
correct response.
In
this case, the subject might originally
respond with
"in
my
office at
work"
and at retest with
"at
work."
Ascore
of
2indicates that the subject is ex-
actly correct. In this case, the original and retest responses
mention identical information and the retest response in-
cludes all the information originally mentioned, and may
also include additional new information. The advantage
of
this system
is
that memories are graded for consistency
and, by implication, accuracy.
Memory scores were compiled from five memory at-
tributes corresponding to those identified
in
R.
Brown and
Kulik (1977) and in other studies as being critical attrib-
utes
of
FB memories: memory description as well as an-
swers to the questions for people, place, activity, and
source. The subjects were assigned ascore
of
0,
1,
or
2for each attribute, according to the correspondence be-
tween their original and retest responses.
For
each sub-
ject, the scores were assigned separately by two judges
and correlation coefficients between judges were com-
puted. For the five attributes, the correlations were all
higher than r>.92 and the comparatively
few
discrepan-
cies were resolved in discussion with the first author.
Using this scoring procedure, memory scores
fallon
ascale
of
0-10.
Throughout the remainder
of
this paper,
we
proportionalize these so that memory scores are ex-
pressed as scores between 0and
1.
Scores
of
.9 and 1
are classified as FB memories; these scores indicate that
asubject was either exactly consistent in the majority
of
his/her memory attributes or was exact on all but one at-
tribute, for which the answer was slightly more general
at retest than test.5Memory scores
of
less than .9 indi-
cate memories that
are
not considered FB memories. In
general, scores in the range
of
.1-.8
represent memories
for which some information has been omitted (forgotten).
However, scores
of
0represent memories that have either
been forgotten
or
that were radically inconsistent over
time. Memory descriptions, illustrating the scoring
scheme, are shown in Appendix
B.
In the analyses
of
the incidence
of
FB memories across
nationalities, we used the categorization
of
memory scores
into FB memories and non-FB memories. There are two
sets
of
secondary variable
data-one
set collected within
14
days
of
the resignation and the second set collected
approximately
11
months later. The secondary variables,
relating to the factors influencing encoding, are repre-
sented in the correlational analyses by the responses given
on the original test; this is because these responses are
closest in time to the reception event. The secondary vari-
ables, relating to rehearsal, are represented by the vari-
ous ratings
of
rehearsal (see Appendix A) collected
11
months after the resignation. We reasoned that because
rehearsal is aprocess that develops over time, the rat-
ings at the II-month interval were likely to be more rep-
resentative
of
this process than those taken closer to the
event.6
The Incidence
and
Accuracy
of
Flashbulb Memories
Astriking difference in the incidence
of
FB memories
across the two groups was observed; 85.6%
of
the U.K.
group had memory scores classified as FB memories com-
pared with an FB memory rate
of28.6%
in the non-U.K.
group. These differences in overall memory consistency
were analyzed in a 2 X2chi-square, in which national-
ity (U.K. vs. non-U.K.) and memory type (FB vs. non-
FB) were used as grouping factors. Asignificant chi-
square (1, N=369)
of
123.5,p
<.001, was observed,
indicating that frequency
of
FB memories varied with
group. For the U.K. group, 184 subjects were classified
as having FB memories compared with
31
who did not,
whereas the corresponding frequencies for the non-U.K.
group were FB =44 and non-FB =110. Thus, the in-

FLASHBULB MEMORY
331
0.6
0.5
~
0.4
...
o
'"
rJl
0.3
02
0.1
0.0 a
0.1
0.2
03
0.4
0.5 0.6 0.7
O.B
0.9
Memory
Score
Figure 1. The distribution
of
FB
memory scores
by
nationality.
cidence
of
FB memories was approximately three times
greater in the U.K. group; this is illustrated in Figure 1,
which shows the distribution
of
memory scores within the
two groups.
First, consider the U.K. group. Comparatively few
memories occur below the .9 category on the scale, and
by far the largest number
of
memories occur in the
"1"
category. This shows that the content
of
the majority
of
the U.K. subjects' FBQ memory responses
11
months
after the event exactly corresponded to their earlier re-
sponses, which were given within 2weeks
of
the recep-
tion event. Note that examination
of
the memory descrip-
tion responses did not reveal identical linguistic forms
of
the two descriptions. In less than 1%
of
cases was there
astriking correspondence between the actual words used
in the descriptions and the order
of
presentation
of
infor-
mation, suggesting that the subjects at retest did not re-
call their earlier description, but in fact recalled amem-
ory
of
the actual reception event. For the non-U.K. group,
the distribution
of
memories across the memory score
range is quite different; substantial numbers
of
memories
fall in the categories 0-.8. These memories are, perhaps,
more typical
of
autobiographical memories in general,
in
which after aretention interval
of
some months only frag-
mentary knowledge
of
the event characteristics are re-
tained (see Conway &Rubin, 1993).
In order to explore the retention
of
memory attributes
for
FB
and non-FB memories, we conducted amixed-
model analysis
of
variance (ANOVA)7on the scores as-
signed to each attribute. In this analysis, memory attri-
butes were treated as awithin-subjects variable with five
levels (description, people, place, activity, and source);
scores ranged from 0to 2on each individual attribute.
Nationality (U.K. and non-U.K.) and FB memory (FB
and non-FB) formed between-subjects variables. Follow-
ing R. Brown and Kulik (1977), we reasoned that asub-
ject either did
or
did not have an FB memory and that
FB memories were qualitatively different from non-FB
memories due to privileged encoding
or
some other set
variables influencing their formation. Hence, we decided
to treat the FB memory as agrouping factor in this anal-
ysis. However, no main effects
or
interactions are reported
because
of
aceiling effect in the FB memory group, which
was adirect consequence
of
the categorization
of
memo-
ries into FB and non-FB. Instead, the analyses focus on
comparisons between the pattern
of
accuracy scores for
the memory attributes within the FB and non-FB mem-
ory groups.
In Figure 2, the mean accuracy scores for each
of
the
five memory attributes are shown for both the FB and non-
FB memory groups. The subjects with FB memories
were, in the main, exactly correct on all memory attri-
butes. Only the scores for description differed reliably
from each
of
the other scores [the averaged contrast8was
F(1,908) =113, MSe=
6.3,p
<.001, E2=.091]. As
mentioned, this was because the subjects tended to be
slightly more general and briefin their retest memory de-
scriptions. For the non-FB group, description differed reli-
ably from people [F(1,560) =24.6, MSe=10.3, p<
.001, E2=.017] and activity [F(1,560) =22.8, MSe=
9.6, p<.001, E2=.015]. The attribute
of
people dif-
fered reliably from place
[F(1,56O)
=33.5, MSe=14.1,
p<.001, E2=.057] and source [F(1,560) =29.4,
MSe=12.3, p<.001, E2=.051]. However, neither

332 CONWAY ET AL.
2
o-l---
FB
Memory
•Description
0People
•Place
0Activity
§Source
Memory
Type
non-FB
Memory
Figure 2. Accuracy scores
of
five memory attributes for the FB
and
non-FB memory groups.
people and activity nor place and source differed reliably
(F
<1
in
both cases).
The means in Figure 2show that, for the non-FB mem-
ory group, memory for the reception event
of
learning
of
Margaret Thatcher's resignation fragmented over time.
The ability to provide acoherent memory description as
well as memory for other people and activity declined
markedly. Memory for location and source
of
the news
(other person
or
media) was better preserved, but even
in these cases the mean memory scores that were slightly
higher than 1indicated that this knowledge was not
higWy
specific or exact. For example, asubject at retest might
remember that he/she had originally
"been
at home
watching the television" when learning
of
the resigna-
tion. This contrasts with the high degree
of
specificity
of
the original response, which might have been
"at
home
in the living room watching the 7o'clock news on Chan-
ne14."
Such
"general"
responses
in
the non-FB group
were common and by our scoring scheme gained only one
point.
In afurther analysis focusing on the pattern
of
error9
responses on the FBQ, we found evidence
of
the recon-
struction
of
memory attributes and retrieval
of
the
"wrong"
memory.
For
the non-U.K. group these errors
were widespread, with approximately one third
of
all re-
sponses counted as reconstructions
or
recall
of
the
"wrong"
memory. The nature
of
these errors closely cor-
responds to the types
of
errors also observed by Neisser
and Harsch (1992). In the U.K. subject group, errors were
rare and accounted for less than 3%
of
responses, with
the memories in this group characterized by remarkable
consistency and detail. More generally,
FB
memories
in
the U.K. group (and when these occurred, in the non-
U.K. group) fit well with
R.
Brown and Kulik's (1977)
account
of
FB memory reports containing detailed infor-
mation concerning people, place, activity, source, and
some
"irrelevant"
details not usually retained
in
auto-
biographical memories. Our subjects, who met the FB
memory criteria, remembered details such as what they
had for breakfast, what television program they had been
watching, the exact words spoken by auniversity lecturer,
the place where someone had written the news on awall,
and the name
of
the radio broadcaster whose show had
been interrupted. Many
of
the subjects recalled even more
specific details, such as "tying
my
shoelaces," "hand-
ing a
£5
note to aticket vendor at aLondon underground
station," and "walking toward amirror
in
aroom as the
news was announced on the radio." In short, the
FB
mem-
ories identified
in
our subjects appeared to have the
"live"
quality emphasized by Brown and Kulik. Furthermore,
consistently recalling apparently
"irrelevant"
details
strongly suggests that the subjects accessed an actual mem-
ory
of
the reception event, which had originally been
formed at encoding.
One potential problem is that only 85.6%
of
the U.K.
group were classified as having
FB
memories. It
is
pos-
sible that this occurred because some
of
the subjects did
not form FB memories during the reception event. On
the other hand, it could be that memories in the U.K.
group were,
in
general, not durable, as indicated
by
an
annual forgetting rate
of
14.4
%.
If
this estimated forget-
ting rate is correct, then only about
75
%
of
our U.K. sub-

jects should have FB memories 2years after the resigna-
tion; after aretention interval
of
10
years, few
of
our
subjects would be able to recall the reception event. This
constrasts with
R.
Brown and Kulik's (1977) finding that
virtually all
of
their subjects had FB memories for the
news
of
J.F.K.
's
assassination many years after the event.
We decided to check on the extent
of
forgetting
by
retest-
ing afurther wave
of
subjects.
It
was not feasible to retest
all
of
our subjects, but
we
were able to retest
33
of
them
after adelay
of
26 months. (Note that these data are not
included in the analysis reported below.) Using the same
scoring criteria, we found that only 2
of
these 33 sub-
jects failed to have FB
memories-a
forgetting rate
of
6%
to 7% per year. Moreover, only 1
of
these subjects had
been classified as having an FB memory at the II-month
interval. The two non-FB memory subjects chose not to
guess, and
in
response to FBQ items simply wrote
"I
don't
remember."
These data, then, further demonstrate the
durability
of
FB memories over lengthy periods
of
time.
In summary, the analyses
of
the memory attribute data
show that the subjects with FB memories gave highly con-
sistent and detailed memory descriptions over aperiod
of
11
months, suggesting that the members
of
this group
had highly specific and enduring memories. In contrast,
the subjects who did not have FB memories gave incom-
plete and inconsistent memory descriptions over the 11-
month retention interval. The data suggest that members
of
this group had fragmentary memories that were not
highly detailed and specific. Thus, the subjects in the
non-FB group showed all the signs
of
individuals who are
in the process
of
forgetting
or
who have already forgot-
ten the reception event
in
which they learned
of
Thatcher's
resignation.
Secondary Measures
The secondary measures were designed to assess two
major components that are thought to influence FB mem-
ory formation-encoding and rehearsal. Three subgroups
of
measures-affect,
importance, and know1edge-
allowed us to examine factors influencing encoding, and
three measures
of
rehearsal allowed us to examine fac-
tors operating after encoding.
10
The central purpose
of
this section is to examine variations in FB and non-FB
memories as afunction
of
points on each
of
the rating
scales, so the rating scales are analyzed by means
of
2
x 3 chi-squares. Note that 3-point rating scales were em-
ployed: 1=high, 2=moderate, and 3=low (see Ta-
ble 1). Thus, FB memories (the between-subjects factor)
formed two levels, and each rating scale (the within-
subjects factor) formed three levels
ofthe
chi-squares (in
these analyses,
df
=2unless otherwise stated). The pro-
portions
of
FB and non-FB memory subjects falling on
each point
of
the rating scales are shown in Table
1.
Af-
fect was assessed by ratings
of
the intensity
of
experienced
emotion and the extent
of
asubject's surprise upon learn-
ing
of
the resignation. The distribution
of
FB and non-
FB memories across the rating scales differed significantly
for both intensity (X2=23.98, p<.01) and surprise
FLASHBULB MEMORY 333
Table
1
Proportion
of
FB
and
Non-FB Subjects Giving High,
Moderate,
and
Low Ratings
on
Each
of
the
Secondary
Variables
FB Memories Non-FB Memories
Variables High Moderate Low High Moderate Low
Affect
Intensity .219 .531 .250 .078 .454 .468
Surprise .566 .359 .075 .390 .475 .135
Importance
PI .167 .605 .228 .057 .404 .539
NI .715 .254 .031 .482 .468 .500
Prior knowledge
Knowledge .101 .618 .281 .014 .383 .603
Interest .201 .659 .140 .106 .575 .319
Rehearsal
Thought .070 .793 .137 .000 .425 .575
Spoke .044 .666 .290 .000 .345 .645
Watched .123 .746 .131 .035 .517 .448
Note-PI
=personal importance, NI =national importance.
(x
2=11.51, P<.01).
It
can be seen from Table 2that
the largest proportions
of
FB memories are associated with
high to moderate levels
of
intensity and surprise. Non-
FB memories, however, are associated with moderate to
low levels
of
intensity and moderate levels
of
surprise.
The importance variables, PI (personal importance) and
NI (national importance), were also both significant
(X
2=
39.4, P<.01, and X
2=20.2, p<.01, respectively),
and from Table 2it can be seen that FB memories are
associated with moderate levels
of
PI and high levels
of
NI. In contrast, non-FB memories are associated with
moderate to low levels
of
PI and are equally distributed
across the 3-point rating scale for NI. Ratings
of
both
knowledge
of
Thatcher's government and interest in po-
litics for FB and non-FB memories produced reliable dif-
ferences(x2=
41.2,p
<
.01,andx2
=
19.1,p
<.01,
respectively). FB memories were associ::.ed with mod-
erate amounts
of
prior knowledge and amoderate level
of
interest in politics. Non-FB memories were associated
with low levels
of
prior knowledge and moderate to low
general interest in politics.
Two other measures, not shown in Table
I-other
memories and
politics-were
also included in the knowl-
edge/interest group. Other memories were scored as 0for
no memories, and 1when any memories were named.
These data were analyzed in aone-way ANOVA. The
subjects with FB memories spontaneously and reliably re-
called more
"other
memories" than did those without FB
memories [F(1,367) =16.6, MSe=
4.0,
P<.001;
Ms
=.56 and .34, respectively]. The subjects recalled
between one and four other memories that, after exami-
nation
of
the protocols for the FB memory group, were
classified into memories featuring political events, per-
sonal autobiographical events only, and those featuring
both types
of
events. The distribution
of
other memories
across these categories was as follows: political events
44
%,
autobiographical events
18
%,
and mixtures
of
both
types
of
events 38
%.
For
the U.K. subjects, the political

334 CONWAY ET AL.
events that were recalled all related
to
occurrences dur-
ing Margaret Thatcher's
11
years
in
power. The subjects
in the non-U.K. group tended to recall more contem-
poraneous political events such
as
current wars and other
political changes, particularly in Eastern Europe. Thus,
the subjects with
FB
memories were often spontaneously
reminded
(in
Schank's, 1982, sense)
of
other events; when
this occurred, the recalled items were usually political
events, and for the U.K. subjects they related directly to
Thatcher's political career. It seems possible that this
spontaneous recollection
of
related events may indicate
the integration in memory
of
the resignation with other
knowledge
of
Thatcher's years in power. Such integra-
tion might facilitate the stabilization in long-term mem-
ory
of
knowledge structures representing thematic aspects
of
the political period dominated
by
Prime Minister
Thatcher (N. Brown, 1990).
For the three measures
of
rehearsal, the subjects judged
how often they had thought or spoken about the resigna-
tion and
how
often they had attended to media reports that
were either directly about the resignation or associated
with it (referred to as "watched"). The rehearsal data
were analyzed with the mixed-model chi-square; Table 1
shows the proportions
of
FB
and non-FB memories fall-
ing at each point on the rating scales. All three measures
of
rehearsal produced significant effects (for thought,
X2=83.2; for spoke, X2=48.5; and
for
watched, X2=
48.6; P<
.01
in all three cases). These differences arose
because
FB
memories were rated
as
receiving moderate
levels
of
rehearsal, whereas non-FB memories received
moderate to low levels
of
rehearsal.
Finally,
we
briefly mention the analysis
of
the confi-
dence ratings that had been collected for responses to
Part 2
of
the FBQ. The subjects used a3-point scale in
which 1=certain, 2=fairly sure, and 3=guess. Here
we focused on subjects' confidence in their accuracy
of
remembering at retest the following attributes: people,
place, activity, and source. Note that each variable was
analyzed separately and
FB
memory
(FB
vs. non-FB) was
used as agrouping factor in one-way ANOVAs. The rea-
son for separate analyses was simply that some
of
the sub-
jects could not remember some
of
the variables at times
and, obviously, could not provide confidence ratings for
these items. Ahighly significant effect
of
FB
memory was
observed for people [F(I,327) =
84.4,MS.
=27.5,p <
.001,
£:.2
=.203], and the subjects
with
FB
memories were
reliably more confident
in
the accuracy
of
their answer
than were those who did not have
FB
memories; the
respective means were 1.26 and 1.87. The same pattern
of
very high confidence for the
FB
memory group and
significant lower confidence for the non-FB memory
group
was
present for the other three attributes [mean
values were place
1.1
(FB)
vs.
1.7 (non-FB), F(l,356) =
84.4,
MS.
=29.96, p<.001,
('2
=.236; activity 1.27
(FB) vs. 1.9 (non-FB), F(I,346) =84.4,
MS
e=30.96,
p<.001,
£:.2
=.197; and source 1.1 (FB) vs. 1.6 (non-
FB), F(l,353) =84.4,
MS.
=20.83, P<.001,
£:.2
=
.202]. Thus, the subjects with
FB
memories were highly
confident in the accuracy
of
their memories, and those
without
FB
memories were reliably less confident, al-
though by and large they did not guess. Note that this
mean moderate level
of
confidence in the non-FB group
reflects the fact that by far the majority
of
these subjects
were,
in
fact, correct for at least some
of
the four mem-
ory attributes sampled
in
Part 2
of
the FBQ.
In summary, then, the subjects with FB memories ex-
perienced more affect, perceived the event to
be
more
im-
portant, knew more about the Thatcher administration,
and were more interested in politics than those who did
not form
FB
memories. Also, the
FB
memory group re-
hearsed the event more frequently than did the non-FB
memory group, but rehearsal was generally at moderate
to low levels. Levels
of
rehearsals were slightly higher
on the thought and watched scales than the spoke scale
(see Table 1). The subjects with
FB
memories were vir-
tually certain
of
the accuracy
of
their memories, whereas
those without
FB
memories were less confident.
The Roles
of
Encoding and Rehearsal
in Flashbulb Memory Formation
Table 2shows the intercorrelations between the primary
and secondary variables. Note that the primary variable
is
the categorization
of
memory scores into
FB
memo-
ries (scores
of
.9 and
1)
and non-FB memories (score
of
.8 or less).
11
Correlations higher than r=
.11
(df
=367)
were significant at the 5%level, and it can be seen that
only politics failed to correlate significantly with the other
secondary variables. In order to analyze these data, we
Table 2
Correlation Matrix
of
FB
Memory
Scores With Secondary Variables
I2 3 4 5 6 7 8 9
10
11
12
I.
FB
score I
2.
Intensity .240 I
3. Surprise .140 .430 I
4. Personal importance .300 .384 .208 I
5. National importance .167 .234 .218 .468 I
6. Other memories .182 .251 .099 .212 .095 I
7. Knowledge .331 .364 .211 .379 .192 .303 I
8. Interest .173 .326 .144 .372 .200 .202 .517 I
9. Politics .117 .040 .027 .083 .010 .052 .075 .081 I
10.
Rehearsal/thought .443 .283 .206 .447 .213 .247 .426 .360 .081 I
II.
Rehearsal/watched .296 .190 .191 .339 .225 .172 .329 .256 .060 .479 I
12.
Rehearsal/spoke .165 .105 .018 .166 .027 .120 .142 .162 .042 .283 .288
Note-Correlations in bold are significant, p<.05.

conducted aseries
of
hierarchical regressions
by
using
sets
of
variables representing encoding and rehearsal (see
Cohen &Cohen, 1975, chap. 4). The main finding, how-
ever, was that although both the encoding and rehearsal
sets were reliably associated with FB memory, they were
also intercorrelated with each other. 12 Thus, in order to
specify the pattern
of
interrelations between these sets
of
variables
in
amore integrated and detailed manner,
we
decided to adopt acausal modeling approach.
For
this analysis we used the structural equation ap-
proach
of
Bentler (see Bentler, 1980, 1989; Bentler &
Weeks, 1980). Note that the variable politics, which had
low intercorrelations with the other variables, was omitted
from these analyses. The strategy
we
adopted was as fol-
lows. Acausal model
of
the secondary ratings for the FB
memory group was developed; the model was then ap-
plied to the secondary ratings for the non-FB memory
group.
It
was reasoned that
if
FB memories are not a
unique class
of
memories, but rather represent an un-
usually detailed and durable set
of
"ordinary"
memories
(McCloskey et al., 1988), then the pattern
of
structural
relations between the secondary ratings
of
FB and non-
FB memories should be identical or, at the very least,
highly similar. On the other hand,
ifFB
and non-FB mem-
ories are completely disjunct classes
of
memories, then
different constructs and different relations between con-
structs should underlie the formation
of
these classes
of
memories. Athird possibility is that the same constructs
are associated with the formation
of
both types
of
memo-
ries, but that relations between constructs differ. In line
FLASHBULB MEMORY 335
with our earlier reasoning and with the findings
of
the
preceding section, we expected differences in the struc-
tural relations
of
affect and importance across the two sets
of
ratings.
We commenced the analysis with the model used in the
regression analyses. There were two latent factors in this
model: encoding and rehearsal. The encoding factor in-
cluded the following variables: intensity (INT), surprise
(SUR), PI, NI, other memories (O-MEMs), prior knowl-
edge (KNO), and interest in politics (lNT-P). The rehear-
sal factor comprised the following rehearsal variables:
thought (THO), watched
(W
AT), and spoke (SPO). Vari-
ants
of
this two-construct model, with and without direc-
tional paths, were fitted to the data. However, the model
and its variants did not provide agood fit. All chi-squares
resulting from the model and its variants were significant,
indicating that, statistically, models
of
this class are sig-
nificantly discrepant from the data. The problem with this
type
of
model is that the variables constituting the encod-
ing factor are intercorrelated, and models that treat en-
coding as asingle latent variable are not sensitive to these
relations.
Accordingly, we generated anew class
of
models in
which the subsets affect, importance, and knowledge/in-
terest formed latent factors, replacing the encoding fac-
tor. Explorations with this class
of
models led to the model
shown in Figure 3, which is astatistically excellent fit
to the data [nonsignificant X2(22) =22.4, p>.66] and
acomparative fit index
of
1.0. (Note that the compara-
tive fit index runs from 0to 1, with values close to 1in-
INT·P
Figure 3. Causal model
of
the relations
of
the secondary variables to
FB
memories. Dashed
lines denote fixed paths.
All
paths are significant, p<.05. See text for details.

336 CONWAY ET AL.
dicating good fit. See Bentler, 1989, for further details
of
fit indices.)
In Figure 3, latent constructs are shown
in
the circles
and measured variables are in the squares. All the paths
between the latent constructs are positive and significant.
Arrows originating at one construct and terminating at
another indicate that the originating construct contributes
to the determination
of
the variance ofthe terminating con-
struct. So, for example, the knowledge/interest construct
directly contributes to the variance of the rehearsal con-
struct, and on this path the value
of
.31
is
the correlation
of
knowledge/interest with rehearsal. Note that because
of
this structural relation, knowledge/interest also makes
an indirect contribution to the actual measures
of
rehear-
sal, shown in the paths from the rehearsal construct to
the measured variables.
Consider the paths between knowledge/interest and the
other three constructs.
It
is
clear from Figure 3that the
knowledge/interest construct
is
central to
FB
memories
in that this construct contributes significantly to all other
constructs and, hence, indirectly to
all
measured variables.
The greater the degree
of
knowledge/interest, the higher
the levels
of
importance, affect, and rehearsal. Impor-
tance, too,
is
positively associated with levels
of
affect
and rehearsal, so that the more important an event, the
greater the affect and extent
of
rehearsal. Affect and re-
hearsal, however, were not directly associated or recipro-
cally related to knowledge/interest and importance. Thus,
for FB memories, knowledge/interest and importance de-
termine the degree
of
affect and extent
of
rehearsal.
When the model shown
in
Figure 3was fitted to the
non-FH memory secondary ratings, achi-square was ob-
served
[x
2(26) =45.1, p<.01], indicating asignificant
discrepancy between the model and these data. It
was
found that the two paths leading from the importance con-
struct to the affect and rehearsal constructs were not sig-
nificant, and the correlations
fell
to .042 and .068, respec-
tively. Thus, the critical difference between the
FB
and
non-FH memory secondary ratings was in the role
of
im-
portance and its structural relation to the other constructs.
One other path was also nonsignificant-the path from the
rehearsal construct to the observed variable
"watched"
(r
=.1). This is not surprising because most, but not all,
of
the non-FB subjects were from the U.S.A. group,
who
had rated themselves
as
having comparatively little me-
dia exposure to the news
of
the resignation (see Table
1).
Anew model that omitted the three nonsignificant paths
was then applied to the non-FB memory secondary rat-
ings and agood fit was observed
[X
2(22) =27.2, p>
.20; comparative fit index
of
.977]. Figure 4shows the
non-FB model, where it can be seen that affect, impor-
tance, and rehearsal are independent constructs with no
structural relations to each other. As with the FB mem-
ory group, the knowledge/interest construct was again
found to be related to each
of
the other three constructs.
Next, consider the similarities between FB and non-
FB
secondary ratings.
It
can be seen from Figures 3and 4
that both types
of
memories involve the same latent con-
structs and, by inference, the same encoding and rehear-
sal
processes. Moreover, knowledge/interest plays asim-
1';',1
.94
\\.
,.---.
.54
-----~~~-~
Figure 4. Causal model
of
the
relations of the secondary variables to non-FB memories. Dashed
lines denote fIXed paths.
All
paths
are
significant, p<.05.
See
text for details.

FLASHBULB MEMORY 337
INT-P
~
.95',
,
'~---
,
,
,
,
,
,
,
,
,
,
.9,~"
B
Figure 5. Causal model ofthe relations ofthe secondary variables to memory fonnation. Dashed
lines denote fixed paths. All paths are significant, p<.05.
See
text for details.
ilar general role for both types
of
memories and
is
positively associated with affect, importance, and rehear-
sal. These findings suggest that some processes are com-
mon to the formation and maintenance
of
both FB and
non-FB memories. Perhaps the processes associated with
the knowledge/interest construct are common to the for-
mation
of
all types
of
autobiographical memories and prin-
cipally reflect the integration
of
new memories with pre-
existing knowledge in memory. The critical difference be-
tween the two models hinges on the role
of
the impor-
tance construct, which influences affect and rehearsal for
FB memories but not for non-FB memories. This key dif-
ference
is
predicted by
R.
Brown and Kulik's (1977) ac-
count
of
FB memory formation and is compatible with
broader studies
of
vivid memories (Rubin &Kozin, 1984).
Finally, we computed a
full
model across both sets
of
sec-
ondary ratings, plus the additional latent construct mem-
ory attributes calculated from the memory scores to the
five memory attributes: memory description (DES), peo-
ple
(PEa),
place (PLA),activity (ACT), and source
(SOU). Figure 5shows the best-fitting model
[X
2(88)
106.2, P>.09; comparative fit index
of
.991]. In this
model, the effects
of
knowledge/interest and importance
on FB memories are indirect and operate through the af-
fect and rehearsal constructs to which they directly con-
tribute. The striking feature in Figure 5is that there are
only two direct influences on the memory attributes con-
struct; these are from the affect and rehearsal constructs.
This part
of
the model shows that the higher the degree
of
affect and the more extensive the rehearsal, the more
detailed the memory. There are no direct or indirect paths
between affect and rehearsal, so it seems that the processes
underlying these constructs may have separate effects on
memory formation.
DISCUSSION
The striking finding
of
the present study was the high
incidence
of
very detailed memory reports provided by
the U.K. subjects, which remained consistent over an 11-

338 CONWAY ET AL.
month retention interval and, for asmaller group, over
a26-month retention interval. Memory for verbal and pic-
torial materials assessed under laboratory conditions typi-
cally decreases over aperiod
of
hours and days and rarely
persists for periods longer than 1month. Similarly, auto-
biographical memory for routine everyday events appears
to rapidly decline (Brewer, 1988; White, 1982) and most
people simply cannot remember events such as what they
had for breakfast
11
months ago
or
the joke made by the
lecturer in last year's philosophy seminar. Moreover, most
items
of
news are quickly forgotten; for the few items
that are retained, people rarely remember the personal
circumstances under which they learned the news (Lar-
sen, 1992). Against this background,
FB
memories clearly
are an exception, enduring for long periods
of
time and
retaining the types
of
details that are rapidly lost from
"or-
dinary"
everyday memories.
Our
findings indicate that
anumber
of
factors are influential in the formation and
maintenance
of
FB memories;
of
these, the importance
of
the original event appears to be critical.
R. Brown and Kulik (1977, Figure
1,
p. 83) proposed
aprocess account
of
FB memory formation in which the
effects
of
different sets
of
processes unfold over the pe-
riod
of
the reception event. They summarized their pro-
cess model in aschematic flow chart; in Figure 6we pro-
vide asimilar summary
of
the present findings. The main
features
of
Figure 6are based on the analyses
of
the struc-
tural relations between the secondary variables (see Fig-
ures 3and 4). The same constructs were present in the
FB
and non-FB secondary ratings, so it is assumed that
the same processes operate
in
the formation
of
both types
of
memories. The differences lie in the strength
of
en-
coding processes and in the structural relations between
these process. Table 2shows that subjects who eventu-
ally form FB memories have more detailed prior knowl-
edge and are more likely to
be
reminded
of
related events
than those who do not have
FB
memories. These find-
ings suggest that FB memory subjects may be more able,
or more prepared, to assimilate the news and its context
(the reception event) to preexisting knowledge structures
in
memory. Similarly, the
FB
memory subjects had higher
levels
of
affect and judged the event to be more impor-
tant than those without FB memories. These differences
in the strength
of
the relations between different encod-
ing processes and memory types are shown in Figure 6
by the
"greater"
and "lesser" dimension attached to each
of
the constructs.
The arrows in Figure 6depict the direction
of
relations
between the constructs and show the way in which aset
of
processes associated with one construct influences the
processes
of
another related construct. The dashed arrows
show aroute for the formation
of
ordinary, non-FB mem-
(ORDINARY MEMORY)
EVENT
non·FB
L
Maintenance
£..!:._--_£.
Elaboration
ENCODING
POST
·ENCODING
(REHEARSAL)
Figure 6. Sequence ofencoding
and
rehearsal processes in memory formation and
maintenance. L=lesser; G=greater.

ories. The main proposal for the non-FB memories
is
that
different sets
of
processes can make independent contri-
butions to memory formation. Prior knowledge and in-
terest are related to importance and affect, but they can
mediate memory formation independently. For example,
an event oflittle or
no
personal importance associated
with
only minimal levels ofaffect
may
nonetheless be encoded
into long-term memory
in
terms
of
the knowledge struc-
tures employed in the processing
of
that event. We as-
sume that this
is
the minimum that
is
required for the for-
mation
of
any memory. Additionally, an event judged
to
be personally or nationally important may come
to
be en-
coded
in
terms
of
the self (Conway &Rubin,
1993)
and/or
cultural knowledge (N. Brown, 1990). Finally, an event
may
be
comparatively unimportant, but nevertheless en-
gender some level
of
affect and so facilitate encoding.
These relatively independent effects
of
the encoding con-
structs must also be present
in
the formation
of
FB
mem-
ories. In this case, however, our findings indicate that
in
addition to the independent effects
of
the encoding con-
structs, there are also more integrated and coordinated
effects. In particular, the construct
of
importance becomes
associated with and influences affect and rehearsal (see
Figure 3).The relation between importance and affect
is
shown
by
the additional dark arrow on the "greater" path
in Figure 6and represents astructural difference (rather
than just aquantitative difference) in the formation
of
FB
compared with non-FB memories.
In Figure 6, then, the formation
of
amemory
is
viewed
as the culmination
of
asequence
of
processes operating
over time, either separately or in conjunction. The first
set
of
processes features the utilization
of
prior knowl-
edge (knowledge +interest,
in
Figure 6); these processes
are common to the formation
of
all memories (see Fig-
ures 3and 4). Asecond set
of
processes then evaluates
the importance
of
the event, and athird set "llediates the
experience
of
affect in response to event features. When
importance is high and comes to be associated with af-
fect, then FB memories are formed. R. Brown and Kulik
(1977) proposed that importance had to either reach or
exceed levels
of'
'biological significance," which we in-
terpret
as
meaning that the item
of
news has to be judged
as
having consequences for self that are more significant
than the consequences typically assigned to most items
of
news. This was certainly the case for the majority
of
the U.K. subjects in the present study, for whom the resig-
nation
of
Prime Minister Margaret Thatcher represented
the end
of
an era in British politics.
If,
however, impor-
tance does not reach some putative critical level, then an
association between affect and importance is not present
and these processes have separate effects on memory for-
mation. In the present study, this latter case was found
to lead to the formation
of
non-FB memories.
Not shown in Figure 6are the direct effects
of
prior
knowledge on rates
of
rehearsal, which were present for
both
FB
and non-FB memories. Rehearsal rates were, in
fact, very low for the non-FB group (mean
ratings>
2.6,
indicating that many
of
the subjects did not rehearse the
event) and at only moderate levels for the FB group (mean
FLASHBULB MEMORY 339
ratings>
2.1). Nevertheless, this aspect
of
the findings
demonstrates that the subjects with greater prior knowl-
edge were more likely to think and talk about the resig-
nation and more likely
to
follow media reports ofthe resig-
nation than were the subjects with lower levels
of
prior
knowledge and interest; this was the case irrespective
of
type
of
memory (see Figures 3and 4). For the FB mem-
ory group only, there was also an additional effect
of
im-
portance on rates
of
rehearsal (see Figure 6). It
is
possi-
ble that this effect reflects the greater general availability
of
FB memories in memory. Perhaps memories
of
im-
portant events are accessed comparatively frequently in
the period following the reception event because
of
their
associations with currently self-relevant themes, plans,
and goals. It
is
even feasible that this association between
importance and rehearsal may reflect some differential
consolidation
of
FB
compared with non-FB memories.
Certainly the
FB
memory subjects indicated that they
"thought" about the resignation more frequently than did
the non-FB memory subjects (see Table
1).
Finally,
as
indicated in Figure 6, we tentatively sug-
gest that rehearsal serves different functions for different
types
of
memories. For
"ordinary"
non-FB memories,
it
is
proposed that the main role
of
rehearsal
is
in preven-
tative maintenance, which acts to preserve the fragmen-
tary knowledge
of
the event, which
is
represented
by
non-
FB memories. In the case
ofFB
memories, it
is
proposed
that the main effect
of
rehearsal
is
in the elaboration
of
FB memory reports. Such elaboration could function to
establish multiple access routes to an FB memory and in
this way expand accessibility and raise availability. Over-
all, however, it seems unlikely that rehearsal is critical
to FB memories. This
is
because the smaller group
of
U.K. subjects, retested for asecond time after areten-
tion interval
of
26 months, had amean rehearsal rating
of
slightly more than 2.8, indicating that they had not re-
hearsed the news. The majority
of
these subjects had FB
memories, so alow level
of
continuous rehearsal cannot
be essential
to
their maintenance.
The findings
of
the present study establish that the per-
sonal and national importance
of
an item
of
news
is
criti-
cal to FB memory formation. This can
be
seen in the dif-
ferent incidence
of
FB memories for the U.K. and
non-U.K. groups and in the integrative and strong role
of
importance to FB memories, compared with its in-
dependent and weak role in non-FB memories. In these
respects, our findings support R. Brown and Kulik's
(1977) original proposal that some events, because
of
their
consequentiality, give rise to unusually detailed memo-
ries. Moreover, although the encoding
of
both FB and
non-FB memories features similar sets
of
processes, these
processes are more integrated in the formation
of
FB than
non-FB memories; in this sense the formation
of
FB
mem-
ories might be considered to be special or privileged.
REFERENCES
ALVEY,
N.
G., ET
AL.
(1983). GENSTAT: Ageneral statistical
programme. Oxford, England: The Numerical Algorithms Group.

340 CONWA Y ET
AL.
BENTLER,
P. M. (1980). Multivariate analysis with latent variables:
Causal modelling. Annual Review
of
Psychology, 31, 419-456.
BENTLER,
P. M. (1989). EQS: Structural equations programme manual.
Los Angeles:
BMDP
Statistical Software.
BENTLER,
P.
M.,
&
WEEKS,
D.
G.
(1980). Linear structural equations
with latent variables. Psychometrika, 45, 289-308.
BOHANNON,
J. N. (1988). Flashbulb memories for the space shuttle dis-
aster: Atale
of
two theories. Cognition, 29, 179-196.
BROWN,
N. (1990). Organization
of
public events
in
long-term mem-
ory. Journal
of
Experimental Psychology: General, 119, 297-314.
BROWN,
R., &
KULIK,
J. (1977). Flashbulb memories.
Cognition,S,
73-99.
BREWER,
W. F. (1988). Memory for randomly sampled autobiograph-
ical events. In U. Neisser &E. Winograd (Eds.), Remembering recon-
sidered: Ecological
and
traditional approaches 10 the study
of
mem-
ory (pp. 21-90). Cambridge: Cambridge University Press.
BREWER,
W. F. (1992). An analysis
of
the theoretical and empirical
status
of
the flashbulb memory hypothesis. In E. Winograd &
U. Neisser (Eds.), Affect
and
accuracy
in
recall: Studies
of
'Jlash-
bulb memories' (pp. 274-305). Cambridge: Cambridge University
Press.
CHRISTIANSON,
S. A. (1989). Flashbulb memories: Special, but not so
special. Memory &Cognition, 17, 435-443.
COHEN,
J.,
&
COHEN,
P. (1975). Applied multiple regression!correla-
tion analysis
for
the behavioural sciences. Hillsdale, NJ: Erlbaum.
CONWAY,
M.
A.,
&
RUBIN,
D.
C.
(1993). The structure
of
autobio-
graphical memory. In
A.
E. Collins,
S.
E. Gathercole, M. A. Con-
way, &P. E. Morris (Eds.),1heoriesofmemory(pp. 103-137). Hove:
Erlbaum.
KEpPLE,
G. (1982). Design and analysis: Aresearcher's handbook (2nd
ed.).
Englewood Cliffs, NJ: Prentice-Hall.
LARSEN,
S. F. (1988). Remembering without experiencing: Memory
for reported events. In U. Neisser &E. Winograd (Eds.), Remem-
bering reconsidered: Ecological
and
traditional approaches to the study
of
memory (pp. 326-354). Cambridge: Cambridge University Press.
LARSEN,
S. F. (1992). Potential flashbulbs: Memories
of
ordinary news
as the baseline. In E, Winograd &U. Neisser (Eds.), Affect and ac-
curacy in recall: Studies
of
'J7.ashbulb
memories' (pp. 32-64). Cam-
bridge: Cambridge University Press.
MCCLOSKEY,
M.,
WIBLE,
C.
G.,
&
COHEN,
N. J. (1988). Is there a
special flashbulb-memory mechanism? Journal
of
Experimental Psy-
chology: General, 117, 171-181.
NEISSER,
U.,
&
HARSCH,
N. (1992). Phantom flashbulbs: False recollec-
tions
of
hearing the news about Challenger.
In
E. Winograd &
U. Neisser (Eds.), Affect and accuracy in recall: Studies
of
'Jlash-
bulb memories·(pp. 9-31). Cambridge: Cambridge University Press.
PiLLEMER,
D.
B.
(1984). Flashbulb memories
of
the assassination at-
tempt on President Reagan. Cognition, 16, 63-80.
RUBIN,
D.
c.,
&
KOZIN,
M. (1984). Vivid memories. Cognition, 16,
81-95.
SCHANK,
R.
C. (1982). Dynamic memory. New York: Cambridge Uni-
versity Press.
WHITE,
R. T. (1982). Memory for personal events. Human Learning,
I,
171-183.
NOTES
1.
In the months preceding the resignation, Mrs. Thatcher's govern-
ment had been in some trouble and senior figures had resigned, appar-
ently in protest at her autocratic leadership and intransigence over key
issues-but
note
that
such resignations and dismissals
had
been ahallmark
of
her
years in power. One particular ex-colleague was standing against
her
in an internal party reelection campaign, although this was widely
perceived more as acontinuation
of
internal protest within
her
party
rather than as aserious challenge. Indeed, Mrs. Thatcher, in charac-
teristic style, had announced less than
12
hearlier that
"I
fight on. I
fight to
win."
At the time
of
her resignation, Mrs. Thatcher had been
prime minister for
II
years and had presided over (indeed, had per-
sonally initiated) major controversial changes in British society. She
had
also led the the country through the Falklands War and appeared as a
major figure in world politics.
Her
exceptionally cordial relationship
with the American president Ronald Reagan had given her ahigh pro-
file in North America. and her stance on the Cold War had led to promi-
nence
in
European politics (at one time she was known as the
"Iron
Lady"
for her opposition to the Soviet Union). Thus, the impact
of
the
resignation was quite remarkable.
Not
only was
it
surprising (in
the
sense
that
it
was wholly out
of
character and no one had expected her to re-
sign, although the context was
in
place for her resignation), but
it
marked
the end
of
an era
in
British politics and, to some extent, an end
of
an
era
in
Anglo-U.S. politics, which had been dominated by strongly out-
spoken right-wing politicians for most
of
the
1980s. U.K. media coverage
of
the resignation was intense and endured over aperiod
of
3to 4days
until attention gradually switched to the upcoming leadership contest
for the recently vacated post
of
prime minister. In North America, me-
dia coverage was far less intense and lasted for amuch shorter period,
apparently receiving major portions
of
"air
time"
and headline space
only on the day following the resignation. However, in the following
weeks and months, Mrs. Thatcher herself was not particularly promi-
nent
in
the media and kept arelatively low profile in the year immedi-
ately following her resignation. Nevertheless, she would certainly have
been more prominent in the British media than in the media
of
other
countries.
2. This retest group
of
subjects was drawn from alarger pool
of
subjects that had all been tested within
14
days
of
the resignation. In
addition, afurther large group was tested
II
months after the resigna-
tion. Memory perfomance in this latter group indicated that completing
the FBQ within 14 days
of
the original event had no additional effect
upon memory performance. Afull account
of
this and all other details
of
the study, as well as results mentioned in the text but not fully re-
ported, are available from the first author upon request.
3. Two questions that are not shown in Appendix Arelate to the
actual time and date
of
the original event and time and date
of
the recep-
tion event.
For
these questions the subjects were required to provide,
as accurately as possible, the nearest minute, hour, day, month, and
year
of
the original and reception events. The subjects were encouraged
to guess. In the section on
"feelings,"
the subjects were also asked to
specify any emotion(s) they experienced during the reception event. Per-
formance on these items does not relate directly to memory scores, so
they are not reported further.
4. Afinal U.K. version
of
the FBQ was also administered at Lan-
caster to agroup
of
second-year psychology undergraduates who com-
pleted the FBQ with reference to the Thatcher resignation and to arange
of
other outstanding public events, both political and nonpolitical, oc-
curring in the U.K. close to the date
of
the resignation.
Our
aim was
to identify apotential control event for the main study (Brewer, 1992).
However, none
of
the events we sampled, other than the Thatcher resig-
nation, were associated with FB memories; the subjects simply did not
recall in detail their personal circumstances when learning
of
other po-
litical events, even those less than Iweek old (see Larsen, 1992). More-
over, although the events we had chosen had been given wide media
coverage, many
of
the subjects (over 40%) claimed that they had not
heard
of
these events prior to the pilot study. We concluded that al-
though, in principle, memory for acontrol event would have been a
desirable design feature, in practice, identifying an event that all sub-
jects would initially know and have amemory for was simply not feasible.
On the basis
of
the pilot studies, various changes were made to the
FBQ.
In
the pilot study, we found that some subjects experienced prob-
lems with the question concerning consequentiality
of
an event. This
question was similar in wording to the one used by R. Brown and Kulik
(1977), but
our
subjects often requested additional clarification prior
to answering this question. We eventually established that two separate
questions probing the personal and national importance
of
an event did
not give rise to queries and, accordingly, in the final versions
of
the
FBQ, the consequentiality question was replaced by the two importance
questions (see also Rubin &Kozin, 1984). The question relating to the
status
of
Margaret Thatcher asked the U.K. subjects whether she had,
in their opinion, been one
of
the best prime ministers
of
the century,
an average prime Ininister,
or
apoor prime minister.
For
the non-U.K.
subjects, this question was changed to read
"Do
you think Margaret
Thatcher was asignificant figure in international politics, afairly aver-
age figure in international politics,
or
of
no significance in international
politics?" This change was introduced because the non-British students
in our pilot study had found it difficult to answer the U.K. version
of
this question. However, in retrospect, this change was aInistake be-

cause it regarded two different aspects of Mrs. Thatcher's career and
confounded these with two different subject samples. Consequently, data
from this question were not included
in
the analysis. A further change
to
the FBQ was to list the political parties appropriate
to
the
country
of
the group to be sampled. In all other respects, however, the FBQ
was the same for all the groups.
5.
In
fact, scores
of
.9 frequently arose because the memory descrip-
tion was marked as a I (see Figure 3), indicating aslightly more gen-
eral description at retest than test. This finding should be treated with
caution, because the subjects were not asked for exhaustive and detailed
descriptions or instructed
to
write the same memory description at retest
that they had written at test. The answers to the specific questions on
people, place, activity, and source are undoubtedly better measures
of
memory consistency over the two tests. However, the ability to pro-
vide acoherent account
of
a past event
is
an
important feature
of
mem-
ory, which is why we retained description as one
of
the
FB
memory
attributes.
6. Note that 0.5 %
of
the responses to the secondary variables were
classified as missing cases, either because they were unreadable or be-
cause asubject had failed
to
respond to one
of
the questions.
By
far,
the largest number
of
missing cases occurred
in
response
to
the ques-
tion on the valence
of
the reception event, which asked whether the ex-
perience
was
"good"
or
"bad."
For the retest group, 30 responses (8%)
to this question were classified
as
missing cases. In retrospect, this was
apoorly constructed question because anumber
of
subjects wrote
"neither"
next to it
or
else left it blank. When the missing cases for
valence were replaced and the variable was entered into the correla-
tional analyses, valence
was
not found
to
be
highly associated with either
FB
memories or other variables, so, consequently, we decided to omit
this variable
in
the analyses. For the purposes
of
analysis, missing cases
on the other variables were replaced
by
values that were calculated
by
using the expectation-minimization algorithm in the statistical package
GENSTAT (Alvey et aI., 1983).
7. An analysis
of
variance was used for this analysis because mem-
ory accuracy
is
acontinuous rather than acategorical variable. Percentage
of
variance accounted for
by
each comparison is shown
by
f2
(see Kep-
ple, 1982, p. 92).
8.
All contrasts were orthogonal planned comparisons.
9. Adetailed account
of
the analysis
of
errors is available from the
first author.
10. Additional analyses contrasting the same measures taken origi-
nally and at retest were also conducted. On some measures, the retest
samples were slightly lower than the
ori~
'nal measures and the event
was judged to be slightly less emotional .md important at retest than
it had been originally. For the rehearsal measures, the event was rated
as being slightly, but reliably, less frequently rehearsed at retest than
originally. This latter finding, however, may relate to ascaling prob-
lem. Originally, the subjects rated rehearsal with reference to amaxi-
mum period
of
14
days, and less in many cases. Thus, they had judged
frequency
of
rehearsal over ashort period, but at retest the rehearsal
FLASHBULB MEMORY
341
ratings referred
to
aperiod
of
II
months.
It
seems possible that these
ratings
may
reflect abriefperiod
of
intense rehearsal followed
by
alonger
period
of
sustained, but less intense, rehearsal. Because
of
this, the two
types
of
rehearsal may not be directly comparable.
II.
In
fact,
if
the full scale
of
FB memory scores is entered into the
correlational analysis, some
of
the correlations change. However, these
changes are only slight and the effects reported in later multiple cor-
relational analyses remain the same. We used the
FB
memory classifi-
cation because we were interested in the differences between
FB
mem-
ories and non-
FB
memories rather than in arange
of
memories varying
in
FB
memory qualities.
12.
Details
of
these regression analyses are available from the first
author.
APPENDIX A
Composition
of
the Flashbulb Memory Questionnaire
MEMORY ATTRIBUTES
Description*
People
Place
Activity
Source
ENCODING
I. Affect
Surpriset
Intensityt
Valence:!:
Emotion (named an emotion)
2. Importance
Personal importance (PI)t
National importance
(NI)t
3. Knowledge/Interest
Other
memories:!:
Knowledge
of
Thatcher's administrationt
Interest
in
Politicst
Politics§
REHEARSAL
Times thought aboutt
Times watched/read/listened to media
Times spoken about
*Score
0-2,
see text. tRated on 3-point scale: 1=high, 2
=moderate, 3=low. :!:ScoreOor
I,
see text. §Score 1-3,
see text.
APPENDIX B
Examples
of
Memory Descriptions
FB Memory Descriptions (Score .9-1)
Subject A
TEST
Iwandered into
my
philosophy seminar and wondered what
my
tutor
was
smiling about.
When everyone had arrived, he announced
"You've
all heard about the Thatcher resignation have you?"
Iwas the only person there who hadn't. He then asked,
"So
what do you want to do, continue with the semi-
nar or go and sit by your radios?"
RETEST
Istaggered out of bed just in time to make it to aphilosophy seminar.
My
tutor asked
us
if
we had heard
the news about Thatcher's resignation. "Thatcher's resigned to what?" Iasked. Which prompted afull ex-
planation
of
the morning's events. He then asked whether we wished to have the seminar or go and sit by
our radios. Amazingly, the group decided to have the seminar.

342 CONWAY ET AL.
Subject B
TEST
Ifirst heard the news when I
was
in
Dunbar Halls
of
residence
at
lunchtime. Iwas
in
the
first-floor pantry
of
E. Block boiling water for aPot Noodle when Icaught
the
end conclusion
of
a
TV
news programme high-
lighting the resignation. There were two other people
in
the
room whose faces Irecall but Ihave never known
their names.
RETEST
At the time Iheard, Isaw the news on television
in
apantry
in
Dunbar Halls while I
was
about to have
my
lunch (Pot Noodle), just after one o'clock p.m. on that day.
Subject C
TEST
Biology lecture
12
midday on the day she resigned. Lecturer announced that
we
would be learning Krebbs'
cycle and suggested
we
may remember it as
we
were learning it on the day Margaret Thatcher resigned.
RETEST
At the beginning
of
Cell Biology lecture
12
midday. Lecturer announced that this was
an
important day since
we
were to learn the Krebbs' cycle and that Mrs. Thatcher had resigned. Iwas sitting on the left-hand side
of
the Arts lecture theatre. I
was
surprised at her resignation and asked several other people if this was true.
Subject D
TEST
I
was
in
my
room walking over
to
the mirror above
my
sink when Iheard the news on the radio.
RETEST
I
was
walking over from
my
desk towards
my
mirror
in
my
room in Halls
of
residence when Iheard on
the radio that Margaret Thatcher
had
resigned.
Non-FB Memory Reports (Score <.9): Basically Accurate)
Subject E
TEST
It
was
during abiology lecture on cell respiration. We were about to start studying "Krebbs' Cycle" and
the lecturer said that
we
would remember this always because
we
were learning it on the very day that Mrs.
Thatcher resigned. Acheer resounded through the Arts Lecture theatre.
RETEST
I
was
in
abiology lecture at the
time-cell
biology, I
think-when
the lecturer announced it. Acheer re-
sounded through the lecture theatre-although one
or
two people shed atear.
Subject F
TEST
It
was mid-morning and Icame out
of
my
room (having just got up) and Iwent into the kitchen. As Iwas
in there
my
next door neighbour passed by and stopped to say hello. Then he said "Have you heard the news
about Maggie?" Iwas confused
as
to what he was talking about, then he explained that she had resigned
and Ican remember feeling very surprised.
RETEST
I
was
cooking
my
breakfast
in
the kitchen
of
the hall
of
residence that Iwas staying
in.
Afriend came in
and he told me that she had resigned.
Non-FB Memory Reports (Score 0): Incorrect
Subject G
TEST
Iwas taking astudy break
in
my
dorm room and was switching the TV channels. It was on the news and
Ialmost turned right past it. 1finally figured out what was being said and realised what was going on.
RETEST
Iwas walking to class with
my
boyfriend and he brought it
up.
Icouldn't believe that Ihadn't heard; how
cut off from the news Ihad been. Iwas surprised but when Ithought about how many other countries were
having achange in government Ididn't find it that strange.

FLASHBULB MEMORY 343
Subject H
TEST
1was sitting next to afriend in Meston 4when Dr. MacQuillan came in and said
"I
suppose by
now
you
will have heard that Mrs. Thatcher has resigned.
So
if
any
of
you want to be Prime Minister you can now
put your names forward!" There was ageneral hubbub
of
chatter
as
everyone started to discuss the event.
1said to
my
friend that 1was really surprised that
she
had resigned,
as
1had expected her to fight to the bitter end.
RETEST
Hear on Northsound Radio that Mrs. Thatcher had resigned before 2nd round
of
voting between her and
Mr. Heseltine, Hurd, and Major. Morning
of
aThursday
(I
think). Making tea in kitchen
(?)-No
the above
is not true. 1think 1heard from afriend in Psychology
Class?-Hearing
full story on radio came
later-friend
came into Psychology and told me.
(Manuscript received April 30, 1993;
revision accepted for publication July 23, 1993.)