“If I had said it I would have remembered it”: Reducing false memories with a distinctiveness heuristic

Abstract

We examined the contributions of decision processes to the rejection of false memories. In two experiments, people studied lists of semantically related words and then completed a recognition test containing studied words, unrelated lure words, and related lure words. People who said words aloud at study were less likely to falsely recognize related lures on the test than were those who heard words at study. We suggest that people who said words at study employed a distinctiveness heuristic during the test whereby they demanded access to distinctive say information in order to judge an item as old. Even when retrieving say information is not perfectly diagnostic of prior study, as in Experiment 2, in which participants both said and heard words at study, people persist in using the distinctiveness heuristic to reduce false memories.

Full text

Psychonomic Bulletin & Review
2001,8 (1), 155-161
Expectations, metamemorial beliefs, even the format
of a memory test can influence the kind and amount of
information that is required for making a memory judg-
ment (see, e.g., Dodson & Johnson, 1993, 1996; Lindsay
& Johnson,1989; Marsh & Hicks, 1998; Multhaup,1995;
Schacter, Israel, & Racine, 1999). Theoretically, decision
processes are a basic part of both the source-monitoring
framework (SMF) of Johnson and colleagues (Johnson,
Hashtroudi, & Lindsay, 1993) and the constructive mem-
ory framework of Schacter and colleagues (Schacter, Nor-
man, & Koutstaal, 1998). A central assumptionof the SMF
is that memories are not tagged with an abstract label spec-
ifying their origin. Instead, decision processes evaluate the
information that is retrieved and attribute it to a source,
such as assessing whether enough pictorial information
is remembered to respond that an item was seen earlier.
Metamemorial beliefs can affect this decision process
(Anderson, 1984; Johnson & Raye, 1981). Johnson and
Raye noted that people generally “expect self-generated
information to have an advantage in memory” (p. 80). A
number of studies have shown that people exhibit a bias
on memory tests that is consistent with this expectation
(e.g., Conway & Gathercole, 1987; Foley, Johnson, &
Raye, 1983; Johnson, Raye, Foley, & Foley, 1981; Hash-
troudi, Johnson, & Chrosniak, 1989). For instance, Foley
et al. (1983) had participants hear words and say words
aloud. On a subsequent source memory test, participants
were biased to respond that new words, incorrectlyjudged
old
, had been heard earlier rather than said. According to
Foley et al., the participants used a test strategy based on
their belief about the relative memorability of the study
items. Because the participants seemed to expect that they
would be more likely to remember said than heard infor-
mation, test items that were familiar were more often
judged to have been heard rather than said.
Schacter et al. (1999; see also Israel & Schacter, 1997)
have shown that decision processes can have a powerful
impact on the occurrence of false memories. They demon-
strated that people can use what they called a
distinctive-
ness heuristic
to reduce the high false recognition rate of
related lure words in a paradigm initially developed by
Deese (1959) and recently extended by Roediger and
McDermott (1995). In the Deese/Roediger–McDermott
(DRM) paradigm, people study lists of words (e.g.,
tired
,
dream
) that are related to a nonpresented lure word (e.g.,
sleep
). On a subsequentold–new recognitiontest contain-
ing studied words (e.g.,
tired
), new unrelated words (e.g.,
butter
), and new related lure words (e.g.,
sleep
), partici-
pants frequently judge that they previously studied the
related lure words. However, studying the same items as
pictures, instead of words, dramatically reduces false
recognition of the related lures. Schacter et al. (1999) ar-
gued that rejecting related lure words after picture encod-
ing, as compared with word encoding, stems from partic-
ipants’ metamemorial belief that they ought to remember
this distinctive pictorial information. By this view, par-
ticipants use a distinctiveness heuristic whereby they de-
mand access to pictorial information as a basis for judg-
ing items as previously studied; the
absence
of memory
for this distinctiveinformation indicatesthat the test item
is new. In contrast, participants who studied words would
not expect detailed recollectionsabout studied items and,
thus, would not base recognition decisions on the pres-
ence or absence of memory for distinctive information.
155 Copyright 2001 Psychonomic Society, Inc.
This research was supported by National Institute on Aging Grant
AG08441 and by a grant from the Human Frontiers Science Program.
We appreciate the helpful comments of Reed Hunt, Marcia Johnson,
Kathleen McDermott, and John Wixted. Correspondence concerning
this article should be addressed to C. S. Dodson, Department of Psy-
chology, Harvard University, 33 Kirkland Street, Cambridge, MA
02138 (e-mail: cdodson@wjh.harvard.edu).
“If I had said it I would have remembered it”:
Reducing false memories with
a distinctiveness heuristic
CHAD S. DODSON and DANIEL L. SCHACTER
Harvard University, Cambridge, Massachusetts
We examined the contributions of decision processes to the rejection of false memories. In two ex-
periments, people studied lists of semantically related words and then completed a recognition test
containing studied words, unrelated lure words, and related lure words. People who said words aloud
at study were less likely to falselyrecognize related lures on the test than were those who heard words
at study. We suggest that people who said words at study employed a distinctiveness heuristic during
the test whereby they demanded access to distinctive
say
information in order to judge an item as
old
.
Even when retrieving
say
information is not perfectly diagnostic of prior study, as in Experiment 2, in
which participants both said and heard words at study, people persist in using the distinctiveness
heuristic to reduce false memories.

156 DODSON AND SCHACTER
Although the studies of Schacter et al. (1999) are con-
sistent with the notion that participants can use a distinc-
tiveness heuristic to reduce false recognition, it is possi-
ble that the effects are specific to pictorial stimuli. In the
present experiments, the generality of the distinctiveness
heuristic as a tool for reducing false recognition is ex-
plored. These experiments draw on the findings from
Johnson and colleagues, especially Foley et al. (1983),
that participants’ expectations about the relative memo-
rability of generated and perceived items can produce a
response bias on the test. Participants in the say encoding
condition of Foley et al.’s study appeared to use a kind of
distinctiveness heuristic, inferring that the lack of mem-
ory for distinctive
say
information indicated that the test
item was earlier heard. We reasoned that a say–hear en-
coding manipulation within the DRM false memory par-
adigm might reveal the use of a distinctiveness heuristic.
If so, saying words at study should yield fewer false recog-
nition responses to the related lure items than does hear-
ing words at study. Importantly, observing such effects
with these purely verbal materials would extend the gen-
erality of the distinctivenessheuristic as a tool for reduc-
ing false recognition beyond the domain of pictorial en-
coding initially examined by Schacter et al. (1999).
EXPERIMENT 1
Method
Participants. Forty-two Harvard University undergraduates par-
ticipated in this experiment. Each was paid $8.
Materials and Design. We created 24 study lists that were based
on the lists used by Roediger and McDermott (1995); each list con-
tained 15 semantically related words. For counterbalancing pur-
poses, the 24 lists were divided into three sets of 8. The participants
studied 16 lists and then completed a recognition test containing
items from each of the 24 lists. During both study and test phases,
all the items appeared centered on an Apple Power Macintosh G3
computer in bold lowercase letters and in 24-point Geneva typefont.
For both say and hear encoding groups, all the study words were seen
for 1.5 sec, and each word was separated by a 1.5-sec delay. The
participants in the hear encoding condition also listened to items at
conversational levels through headphones. Approximately 400 msec
elapsed between the visual and the subsequent auditory presenta-
tions of the word. In the say encoding condition, the items were seen
only, although the participants spoke the word aloud upon seeing it.
Following the study phase, the participants completed a 96-item
recognition test. The test contained 48 targets or studied words
(from Positions 1, 8, and 10 of the 16 studied lists), 24 target con-
trols or unrelated lures (from Positions 1, 8, and 10 of the 8 non-
studied lists), 16 false targets or related lures (the critical lure that
is semantically related to the items in each of the 16 studied lists),
and 8 false-target controls or unrelated lures (the critical lure that is
semantically related to the items in each of the 8 nonstudied lists).
The test was constructed so that the items were randomly inter-
mixed and also obeyed the following constraints: (1) Each type of
item (target, target control, false target, and false-target control) ap-
peared equally often in each quarter of the test; (2) no more than
two items of the same type appeared consecutively; and (3) items
from the same study list were at least eight positions apart.
In sum, the main design consisted of one between-groups vari-
able, the say versus hear encoding condition, and one within-groups
variable, the type of item.
Procedure. The participants were tested individually and were
told to pay attention to the words that they would see because their
memory for these words would be tested later. In the hear encoding
condition, the participants were told that they would hear each of
the words in addition to seeing it. The participants in the say en-
coding condition were instructed to say aloud each word that they
saw. All the participants were told that they would see 16 different
lists of words, with 15 words in each list. They were instructed that
each list was separated by a 1-min break, during which time they
would work on a puzzle. After studying the 16 lists, the participants
worked on puzzles for 3 min before receiving the recognition test.
For the recognition test, the participants indicated whether each
test item was old (i.e., it had been seen earlier on 1 of the 16 study
lists) or new (i.e., it had not appeared on the study lists) by pushing
the “a” or the “;” key, which had been labeled “old” and “new,” re-
spectively. All test words were presented visually in the center of the
screen, with the response options “Old” or “New” appearing 2 cm
below the word. After each response, the screen cleared and was
followed by a 1.5-sec delay before the presentation of the next test
item.
Results
Table 1 displays the central finding of the experiment:
False targets (related lures) are incorrectly judged
old
much less often in the say than in the hear encoding con-
dition, as was confirmed by an analysis of variance
[ANOVA;
F
(1,40) = 7.90,
p ,
.01]. In contrast, there was
no difference between the say and the hear encoding con-
ditions in the recognition rate of target items, target con-
trol items, and false-target control items [all
F
s(1,40)
,
1.43]. This pattern was confirmed by a 2 (say/hear en-
coding)
3
2 (target/false target) ANOVA, which yielded
significant main effects of encoding condition[
F
(1,40) =
5.13,
MS
e
= 0.041,
p ,
.05], and item type [
F
(1,40) =
4.56,
MS
e
= 0.043,
p ,
.05], and, importantly, a signifi-
cant interaction [
F
(1,40) = 5.30,
p ,
.05]. Recognition
rates of targets (.71) and false targets (.72) were nearly
identical after hearing words at study [
F
(1,40)
,
1], in
line with other studiesusing the DRM paradigm (Mather,
Henkel, & Johnson, 1997; Norman & Schacter, 1997;
Payne, Elie, Blackwell, & Neuschatz, 1996; Roediger &
McDermott, 1995;Schacter, Verfaellie, & Pradere, 1996).
After saying words, however, false targets (.51) were less
likely to be called
old
than were targets [.76;
F
(1,40) =
9.86,
p ,
.01].
To identify the contributions of sensitivity and re-
sponse bias to our findings, we performed signal detec-
tion analyses, using
A¢
as an estimate of sensitivity and
B²D
as an estimate of bias (Donaldson, 1992; Snodgrass
Table 1
Proportion of Items Judged Old in Experiments 1 and 2
Experiment 1 Experiment 2
(Between Groups): (Within Groups):
Encoding Condition Encoding Condition
Item Type Hear Say Hear Say
Targets .71 .76 .70 .79
Target controls .16 .16 .12 .12
False targets .72 .51 .57 .58
False target controls .29 .28 .14 .14

IF I’D SAID IT I WOULD’VE REMEMBERED IT 157
& Corwin, 1988).
1
These analyses are based on those used
by Koutstaal and Schacter (1997), Schacter, Verfaellie,
Anes, and Racine (1998), and Schacter et al. (1999). Val-
ues of
A¢
can vary between 0 and 1, with higher values
indicatingbetter sensitivity and chance performance cor-
responding to a value of .5. Scores for the response bias
measure,
B²D
, can vary between
2
1 (extremely liberal
responding) and 1 (extremely conservative responding).
Followingother studies from our laboratory (Koutstaal
& Schacter, 1997; Schacter et al., 1999; Schacter, Verfael-
lie, et al., 1998), we analyzed the data, as is shown in
Table 2, in three different ways. First, the say and hear en-
coding conditionsdid notdiffer in either sensitivityor bias
associated with discriminating targets (studied items)
from target controls (unrelated lures) [
F
s(1,40)
,
1].
Second, the participants who said words were better able
to discriminate targets from false targets (
A¢
= .69) than
were those participants who heard words, who were at
chance in this comparison [
A¢
= .50;
F
(1,40) = 14.79,
MS
e
= 0.027,
p ,
.001]. In this target/false-target compar-
ison, however, there was generally more conservative re-
spondingafter say than after hear encoding,althoughthis
difference was not significant [
F
(1,40) = 2.43]. Finally,
we analyzed the sensitivity and bias scores for the com-
parison between responses to false targets (related lure
words) and false-target controls. Hear encoding produced
generallygreater
A¢
scores [
F
(1,40) = 3.74,
MS
e
= 0.035,
p
= .06], and more liberal responding [
F
(1,40) = 3.21,
MS
e
= 0.280,
p ,
.08], than did say encoding, although
both of these effects were only marginally significant.
Discussion
Experiment 1 demonstrates that saying, as compared
with hearing, words at encoding reduces the false recog-
nition rate of semantically related lures on the test. These
results extend those of Israel and Schacter (1997) and
Schacter et al. (1999) by showing that reductions of false
recognition produced by distinctive encoding are not re-
stricted to pictorial materials (see also Smith & Hunt,
1998).
There are a number of aspects of our data that suggest
that participants use a distinctiveness heuristic to reject
false targets. First, suppressing the occurrence of false
recognition in the say encoding condition was specific
to false targets and was not a by-product of an overall re-
duction in recognition of all the items: The studiedwords
were recognized at nearly identical rates in the two en-
codingconditions.Second, the pattern of results from the
signal detection analyses paralleled the recognition data.
Say and hear encodingyielded comparable levels of sen-
sitivity (i.e., values of
A¢
) for distinguishing targets
(studied words) from target controls (unrelated lures).
However, say encoding produced greater sensitivity than
did hear encoding for distinguishing the false targets
from the targets. In general, people tended to be more
conservative,as indexed by
B²D
, after say than after hear
encoding. Overall, we observed a selective reduction in
false-recognition responses to the false targets in the say
encoding condition.
The previous analyses indicate that the participants
used a distinctiveness heuristic after say encoding to re-
ject related lure items. The more conservative respond-
ing in the say encoding condition, we argue, is the prod-
uct of participants’ metamemorial beliefs about the kind
of information they feel they ought to remember. After
saying words at study, the participants employeda heuris-
tic during the test, whereby they demanded access to dis-
tinctive say information in order to judge an item as
old
.
In other words, the participants’ reasoning may be char-
acterized as “If I had said it I would have remembered it.”
Thus, the absence of memory for saying target items in-
dicated that the word was not encountered during the
study phase.
There are two curious aspects to our results, however.
First, in both encodingconditions,there was a high false-
recognition rate for the false-target controls. This seems
to be a typical finding and probably occurs because the
false-target controls tend to have higher word frequen-
cies than do the target controls (see, e.g., Arndt & Hirsh-
man, 1998; Israel & Schacter, 1997; Seamon, Luo, &
Table 2
Values of Sensitivity (A
¢
) and Response Bias (B
²
D)
for Items in Experiments 1 and 2
Experiment 1 Experiment 2
(Between Groups): (Within Groups):
Encoding Condition Encoding Condition
Comparison Hear Say Hear Say
Values of Sensitivity (A
¢
)
Target versus target control .84 .86 .86 .88
Target versus false target .50 .69 .62 .68
False target versus false target control .77 .66 .76 .77
Values of Bias (B
²
D)
Target versus target control .40 .33 .54 .45
Target versus false target 2.57 2.37 2.30 2.46
False target versus false target control .13 .43 .63 .62

158 DODSON AND SCHACTER
Gallo,1998). Second, saying words during encodingpro-
duceda subsequentreduction in
old
responses to false tar-
gets, but not in
old
responses to other new items. If the dis-
tinctivenessheuristicis a strategy that is applied throughout
the test, we would expect lower false-recognition rates
for all of the new items—the unrelated new items (i.e.,
the target controls and false-target controls), in addition
to the false targets. Indeed, Schacter et al. (1999) used a
similar design and found that picture encoding, as com-
pared with word encoding (using a between-groups de-
sign), produced lower false-recognitionresponses to both
the false targets and to the true- and false-target controls.
We suggest that say encoding yields less salient or dis-
tinctive recollections than does picture encoding. Con-
sequently, after saying words aloud, people may apply
the distinctiveness heuristic selectively, such as for test
items that are very familiar. For these items, subjects
may respond
old
only when they can remember infor-
mation about having said this item at study. Therefore,
although the false targets may feel very familiar, the ab-
sence of memory for say information is evidence that
they were not studied previously. This strong sense of fa-
miliarity would not be associated with unrelated new
items, and hence, the distinctivenessheuristic may not be
needed or used.
However, consider an alternative account of the data
that does not require a distinctiveness heuristic. Specif-
ically, suppressing false recognition following say en-
coding may depend on remembering list-specific infor-
mation aboutstudied items. For example, when presented
with the related lure item
sleep
on the recognition test,
the participants in the say encoding condition may have
recalled saying related words, such as
tired
,
bed
,
dream
,
and so forth. Because they could remember saying these
words but could not recall saying
sleep
, the participants
may have concluded that
sleep
was never presented dur-
ing the study phase. This list-specific account hinges on
participants’recollectingitems from particularstudy lists
and using this list-specific information to avoid false
recognition responses. On this account, saying, as com-
pared with hearing, words increases the probability of
recollecting list-specific information.
EXPERIMENT 2
One way to evaluate the influence of a distinctiveness
heuristic is to construct a situation in which it is difficult
to use it. Following the logic of Schacter et al. (1999), we
implemented this strategy in Experiment 2 by using a
within-groups design, instead of the between-groups de-
sign used in Experiment 1. That is, the participants heard
some words and said others during the study phase and
then completed a recognition test. In this design, the dis-
tinctivenessheuristic should be ineffective,because there
is no longer a particularkind of information that is solely
diagnostic of a test item’s oldness or newness. Whereas
in Experiment 1 the absence of information about having
said an item suggested that the item was new, this was not
the case in Experiment 2. Therefore, if suppressing false
recognitionis primarily attributableto the distinctiveness
heuristic, a within-groups manipulation of say versus
hear encoding should eliminate the suppression effect.
On the other hand, if false-recognition suppression is
based on recalling list-specific information for the spo-
ken lists and not for the heard lists, we should observe a
selective suppression effect for the spoken lists with the
within-groups manipulation.
Method
Participants. The participants were 27 Harvard University un-
dergraduates, who were each paid $8.
Materials and Design. The study and test materials were iden-
tical to those used in Experiment 1. The only difference was that
each participant said eight of the studied lists and heard eight of the
lists. No more than two lists were studied consecutively in the same
manner.
Procedure. The one difference from Experiment 1 was that the
participants were instructed that they would say aloud some of the
lists of words and hear other lists of words. Before each list, in-
structions appeared on the screen indicating whether the list was to
be spoken aloud or heard.
Results
As can be seen in Table 1, hit rates were higher for
spoken words (i.e., targets) than for heard words. Impor-
tantly, false-recognition rates were nearly identical to the
false targets related to spoken lists and to heard lists. A
2
3
2 ANOVA, with factors of encoding (say vs. hear)
and item type (target vs. false target), yielded a margin-
ally significant effect of encoding [
F
(1,26) = 3.04,
MS
e
=
0.021,
p ,
.10], a significanteffect of item-type [
F
(1,26) =
13.86,
MS
e
= 0.051,
p ,
.01], and importantly, a signifi-
cant interaction [
F
(1,26) = 4.08,
MS
e
= 0.01,
p
= .05].
Although the participants showed higher recognition
rates of spoken words than of heard words [
F
(1,26) =
10.32,
p ,
.01], there were no differences in the false-
recognitionrate of false targets in the say and the hear en-
coding conditions [
F
(1,26)
,
1].
Table 2 presents the estimates of sensitivity (
A¢
) and
bias (
B²D
) for our three critical comparisons. An ANOVA
of the
A¢
and
B²D
scores associated with discriminating
targets from target controls revealed higher sensitivity
scores and more liberal responding for said words than
for heard words [
F
(1,26) = 9.50,
MS
e
= 0.025,
p ,
.01,
and
F
(1,26) = 5.38,
MS
e
= 0.021,
p ,
.05, respectively].
For the comparison of the rates at which targets and false
targets were called
old
, saying words produced greater
discriminability (i.e., higher
A¢
scores) than did hearing
words [
F
(1,26) = 6.37,
MS
e
= 0.008,
p ,
.05]. However,
the participantswere more conservative after having heard
words than after having said words, although this differ-
ence was marginally significant [
F
(1,26) = 3.57,
MS
e
=
0.093,
p
= .07]. Finally, there were nearly identical
A¢
and
B²D
scores in the two encoding conditions for the com-
parison between false targets and false-target controls
[
F
s(1,26)
,
1].

IF I’D SAID IT I WOULD’VE REMEMBERED IT 159
Discussion
Saying words, as compared with hearing words, pro-
duced higher hit rates and higher
A¢
scores associated
with discriminating targets from target controls. Despite
higher levels of memory for spoken than for heard words,
proportions of
old
responses to false targets were nearly
identical in these two encoding conditions. The false-
recognition rates of the false targets and the
A¢
scores as-
sociated with distinguishing between false targets and
false-target controls were no different after having said
or heard words. These results support our prediction that
when the distinctiveness heuristic is rendered ineffective
by having people both hear and say words at encoding,
there is no difference between the two encoding condi-
tions in the false-recognition rate of the false targets.
Moreover, these results are evidence against the alterna-
tive list-specific account of the suppression effect in Ex-
periment 1. If say encoding allowed people to recall par-
ticular list items (e.g.,
tired
,
bed
,
dream
, etc.) as a way of
ruling out the false target (i.e.,
sleep
was not on the list),
there should have been a selective reduction in this ex-
periment in the false-recognition rate of the false targets
related to the said lists. Since there was no selective re-
duction, we conclude that when it is, to some degree, di-
agnostic of an item’s oldness, the distinctivenessheuristic
is the primary mechanism for suppressingfalse memories
in this paradigm.
However, if having people hear some lists of words
and say other lists renders the distinctiveness heuristic
ineffective, we would expect the false-recognition rates
to the false targets in both encoding conditionsin Exper-
iment 2 to rise to the level of the hear encodingcondition
in Experiment1—the conditionin which people show ro-
bust false memories. As can be seen in Table 1, the false-
recognition rates to the false targets in both conditionsof
Experiment 2 are somewhat suppressed and are compa-
rable with the false-recognition rate in the say condition
in Experiment 1. Schacter et al. (1999) observed a simi-
lar finding (see their Figure 1) across their two experi-
ments. Taken together, these results suggest that people
still attempt to use the distinctivenessheuristic even though
it is not perfectly diagnostic of an item’s
oldness
,and
does not produce selective suppression of false alarms
for the say encoding condition, as compared with the
hear encoding condition.If people are heavily weighting
their memories for say information, this would contribute
to the higher recognition rates for the said words than for
the heard words (whereas there were no differences in
the recognition rates of said and heard words in Experi-
ment 1, which was a between-subjects manipulation).
GENERAL DISCUSSION
In our experiments, we investigated the contributions
of a distinctiveness heuristic to reducing false memories
in the DRM paradigm. In Experiment 1, we found that
the participantswho said words at study showed less false
recognitionof related lure words than did the participants
who heard words at study. Saying words aloud—like
studying pictures in Schacter et al.’s (1999) study—pro-
vides a basis for employing a distinctiveness heuristic
during the test. That is, saying words at study leads peo-
ple to expect to remember information about having said
the word; the absence of this expected information sug-
gests that the test word is new. Even when retrieving say
information is no longer diagnostic of prior study, as in
Experiment 2, in which participants both said and heard
words at study, people persist in attempting to use the
distinctiveness heuristic to reduce false memories. The
absence of a difference between the two encoding condi-
tions in the false-recognition rates of the false targets in
Experiment 2 indicates that say encoding does not allow
people to remember list-specific information as a way of
rejecting false targets, disconfirming an alternative list-
specific account of the suppression effect in Experi-
ment 1. In short, the distinctiveness heuristic seems to
be the main mechanism underlying false-recognition
suppression after picture or speak-aloud encoding.
The distinctiveness heuristic is consistent with one of
the basic tenets of the source-monitoring framework of
Johnson and colleagues—namely, that people use various
decision strategies to evaluate activated information and
attribute it to a particular source (Johnson et al., 1993).
Hicks and Marsh (1999) proposed an account similar to
our distinctiveness heuristic to explain conditions in
which they observed reduced levels of false recall of crit-
ical lures in the DRM paradigm. They found that people
were less likely to recall a critical lure after studying a
list of related words from two different sources, such as
hearing some words and seeing other words as anagrams,
than after studying a list of related words from the same
source, such as hearing all of the items. Hicks and Marsh
argue that “presenting DRM items from multiple sources
can potentially create diagnostic evidence that can be
used to edit out false memories” that lack this quality
(p. 1206). These studies, together with our findings and
those of Johnson and colleagues(e.g., Foley et al., 1983),
indicate that an inference process based on the absence
of memory for expected distinctive characteristics is a
general retrieval strategy that is used during recall, recog-
nition,and source monitoring (for a different perspective
on retrieval strategies and criterion setting in the DRM
paradigm, see Miller & Wolford, 1999; cf. Roediger &
McDermott, 1999; Wixted & Stretch, 2000).
It is worth noting that Smith and Hunt (1998) report
lower false recall and recognition of the false targets
after visual than after auditory presentation of the study
words. Because people saw the words in the say condition
of our experiments, it is conceivablethat our suppression
effect is attributable, at least in part, to the visual pre-
sentation of the study words. However, two lines of evi-
dence indicate that this suppression effect is due to say-
ing the words, rather than to visual presentation of the
words. First, visual presentation of the study words was
held constant across our study conditions.For instance,in
Experiment 1, people either heard and saw the words or

160 DODSON AND SCHACTER
said and saw the words. If visual presentation were en-
tirely responsible for the suppression effect, we should
have found no difference in the false-recognition rate of
the false targets in these two conditions. Second, Israel
and Schacter (1997) compared visual and auditory pre-
sentation with auditory-only presentation and found no
difference in subsequenttrue- and false-recognitionrates.
Seeing the words at study did not affect subsequentrecog-
nition performance (see also Arndt & Hirshman, 1998;
Seamon et al., 1998, who find robust false-recognition
rates after visual presentation of the study words). In sum,
our experiments indicate that saying words at encoding,
as opposed to seeing them, leads people to expect to re-
member these words at test, thus yielding the use of the
distinctiveness heuristic.
A metacognitiveheuristic similar to the distinctiveness
heuristic has been used to explain the pattern of false-
recognition data reported by Strack and Bless (1994).
Strack and Bless required participants to study related
words, such as different tools, together with a few words
that represented unique instances of other categories,
such as a book or a shoe. On a subsequent recognition test,
the participants exhibited lower false recognition of
salient distractor items (e.g., a bouquetof flowers) than of
nonsalient distractor items (e.g., a kind of tool). Strack
and Bless argued that the salience of the distractor item
was diagnosticof its
newness
and, thus, served as a basis
for its rejection on the recognitiontest. However, Rotello
(1999) noted that salient distractors are more dissimilar
to studied items, and consequently less familiar, than are
nonsalient distractors. Therefore, the similarity of the dis-
tractor to studied items may explain the pattern of greater
rejection rates of salient than of nonsalientdistractors. In
a series of experiments, Rotello provided evidence in sup-
port of this similarity account of the false-recognition
data in the Strack and Bless paradigm.
The distinctiveness heuristic discussed here and the
heuristic proposed by Strack and Bless (1994) both in-
voke a decision rule whereby the failure to retrieve par-
ticular information about an event can be diagnostic of
the event’s nonoccurrence. Hence, it is important to ask
whether a similarity account, and not the distinctiveness
heuristic, accounts for the results in our experiments. A
similarity account explainsrecognitiondecisions in terms
of familiarity, or the overall similarity (reflected by the
amount of activation) between a test item and the memo-
rial representation of studied items (e.g., Gillund &
Shiffrin, 1984; Hintzman, 1988). Test items that are suf-
ficiently similar to studied items are judged
old
,and
those that are not are judged
new
. By this view, the typ-
ically large false-recognition rates of related lures in the
DRM paradigm occur because of their high similarity to
studied items (Arndt & Hirshman, 1998). To explain our
false-target rejection results in Experiment 1, this simi-
larity account would contend that say encoding is suffi-
cientlydifferent from hear encoding that the false targets
are, in some sense, less similar to previously said words
than to previously heard words. Put another way, this
similarity account would have to argue that lower false
recognition of the critical lures after say encoding occurs
because the critical lures are less familiar (i.e., less sim-
ilar) after say than after hear encoding. There are prob-
lems, however, with this similarity explanation of our re-
sults. First, if say and hear encodingproduced sufficiently
different representations to account for the suppression
of false memories in Experiment 1, this similarity ex-
planation would predict differences in the recognition
rates of spoken and heard words. This prediction was not
supported by the data: In Experiment 1, the said and
heard words yielded nearly identical
A¢
old–new recog-
nition scores. Second, Experiment 2 provides further ev-
idence against the similarity account.The participants in
Experiment 2 showed slightly higher
A¢
old–new recog-
nition scores for spoken than for heard words. Nonethe-
less, there was no difference in the false-recognition rate
of false targets that were related to spoken and heard
lists. Thus, the similarity account offers no ready expla-
nation of these results. According to the distinctiveness
heuristicaccount, in contrast, the absence of expectedsay
information is diagnostic of a test item’s nonoccurrence.
In sum, beliefs and expectations about the workings
of memory affect judgments about the origins of past ex-
periences (e.g., Johnson & Raye, 1981). People are aware
that different kinds of encoding activities yield memo-
ries that vary in strength, detail, and vividness.This meta-
memorial knowledge can be used as a distinctiveness
heuristic to reject false memories.
REFERENCES
Anderson, R. E. (1984). Did I do it or did I only imagine doing it?
Journal of Experimental Psychology: General, 113, 594-613.
Arndt, J., & Hirshman, E. (1998). True and false recognition in
MINERVA2: Explanations from a global matching perspective. Jour-
nal of Memory & Language, 39, 371-391.
Conway,M. A., & Gathercole,S. E. (1987). Modality and long-term
memory. Journal of Memory & Language, 26, 341-361.
De e se , J. (1959). On the prediction of occurrence of particular verbal
intrusions in immediate recall. Journal of Experimental Psychology,
58, 17-22.
Dodson, C. S., & Johnson, M. K. (1993). Rate of false source attribu-
tionsdepends on how questions are asked. American Journal of Psy-
chology, 106, 541-557.
Dodson, C. S., & Johnson, M. K. (1996).Some problems with the pro-
cess dissociation approach to memory. Journal of Experimental Psy-
chology: General, 125, 181-194.
Donaldson, W. (1992). Measuring recognition memory. Journalof Ex-
perimental Psychology: General, 121, 275-277.
Foley, M. A., Johnson, M. K., & Raye, C. L. (1983). Age-related
changes in confusion between memories for thoughts and memories
for speech. Child Development, 54, 51-60.
Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both
recognition and recall. Psychological Review, 91, 1-67.
Hashtroudi, S., Johnson, M. K., & Chrosniak, L. D. (1989). Aging
and source monitoring. Psychology & Aging, 4, 106-112.
Hicks, J. L., & Marsh, R. L. (1999). Attempts to reduce the incidence
of false recall with source monitoring. Journal of Experimental Psy-
chology: Learning, Memory, & Cognition, 25, 1195-1209.
Hintzman, D. L. (1988). Judgments of frequency and recognition
memory in a multiple-trace memory model. Psychological Review,
95, 528-551.
Israel,L., & Schacter, D. L. (1997). Pictorial encoding reduces false

IF I’D SAID IT I WOULD’VE REMEMBERED IT 161
recognition of semantic associates. Psychonomic Bulletin & Review,
4, 577-581.
Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source
monitoring. Psychological Bulletin, 114, 3-28.
Johnson, M. K., & Raye, C. L. (1981). Reality monitoring. Psycho-
logical Review, 88, 67-85.
Johnson, M. K., Raye, C. L., Foley, H. J., & Foley, M. A. (1981).
Cognitive operations and decision bias in reality monitoring. Ameri-
can Journal of Psychology, 94, 37-64.
Koutstaal, W., & Schacter, D. L. (1997). Gist-based false recogni-
tion of pictures in older and younger adults. Journal of Memory &
Language, 37, 555-583.
Lindsay, D. S., & Johnson, M. K. (1989). The eyewitness suggestibil-
ity effect and memory for source. Memory & Cognition, 17, 349-358.
Marsh, R. L., & Hicks, J. L. (1998). Test formats change source-
monitoring decision processes. Journal of Experimental Psychology:
Learning,Memory, & Cognition, 24, 1137-1151.
Mather, M., Henkel, L. A., & Johnson, M. K. (1997). Evaluating
characteristics of false memories: Remember/know judgments and
memory characteristics questionnaire compared. Memory & Cogni-
tion, 25, 826-837.
Miller, M. B., & Wolford, G. L. (1999). The role of criterion shift in
false memory. Psychological Review, 106, 398-405.
Multhaup, K. S. (1995). Aging, source, and decision criteria: When
false fame errors do and do not occur. Psychology & Aging, 10, 492-
497.
Norman, K. A., & Schacter, D. L. (1997). False recognition in
younger and older adults: Exploring the characteristics of illusory
memories. Memory & Cognition, 25, 838-848.
Payne, D. G., Elie, C. J., Blackwell, J. M., & Neuschatz, J. S.
(1996). Memory illusions: Recalling, recognizing, and recollecting
events that never occurred. Journal of Memory & Language, 35, 261-
285.
Roediger, H. L., III, & McDermott, K. B. (1995). Creating false
memories: Remembering words not presented in lists. Journal of Ex-
perimental Psychology: Learning, Memory, & Cognition, 21,
803-
814.
Roed
iger, H. L., III, & McDermott, K. B. (1999). False alarms about
false memories. Psychological Review, 106, 406-410.
Rotello, C. M. (1999). Metacognition and memory for nonoccur-
rence. Memory, 7, 43-63.
Schacter, D. L., Israel,L.,& Racine,C. A. (1999). Suppressing false
recognition in younger and older adults: The distinctiveness heuris-
tic. Journal of Memory & Language, 40, 1-24.
Schacter, D. L., Norman, K. A., & Koutstaal, W. (1998). The cog-
nitive neuroscience of constructive memory. Annual Review of Psy-
chology, 49, 289-318.
Schacter, D. L., Verfaellie,M., Anes, M. D., & Racine, C. (1998).
When true recognition suppresses false recognition: Evidence from
amnesic patients. Journal of Cognitive Neuroscience, 10, 668-679.
Schacter, D. L., Verfaellie, M., & Pradere, D. (1996). The neu-
ropsychology of memory illusions: False recall and recognition in
amnesic patients. Journal of Memory & Language, 35, 319-334.
Seamon,J. G., Luo, C. R., & Gallo, D. A. (1998). Creating false mem-
ories of words with or withoutrecognition of list items: Evidence for
nonconscious processes. Psychological Science, 9, 20-26.
Smith, R. E., & Hunt, R. R. (1998). Presentation modality affects false
memory. Psychonomic Bulletin & Review, 5, 710-715.
Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring
recognition memory: Applications to dementia and amnesia. Journal
of Experimental Psychology: General, 117, 34-50.
Strack, F., & Bless, H. (1994). Memory for nonoccurrences: Meta-
cognitive and presuppositionalstrategies. Journal of Memory & Lan-
guage, 33, 203-217.
Wixted, J. T., & Stretch, V. (2000). The case against a criterion-shift
account of false memory. Psychological Review, 107, 368-376.
NOTE
1. We also analyzed the data from both experiments with d
¢
an
d C.
T
his produced nearly identical results as the A
¢
and B
²
D analyses, ex-
cept for the comparison in Experiment 1 between the false targets and
the false-target controls in the two encoding groups. The say and hear
encoding groups were marginally significantly different with the A
¢
analysis, but they were not different in the d
¢
analysis. This discrepancy
between these two measures does not alter the conclusions of the paper.
Therefore, for the sake of comparability with previous papers on this
topic, we report the A
¢
scores.
(Manuscript received June 22, 1999;
revision accepted for publication December 20, 1999.)