Can someone write a psychology paper on memory lost in APA format including special instructions?

The impact of post-event information on study-related memories: An exploration of the roles of judgemental anchoring, specific expectations about change, and

motivational influences

Peter Sedlmeier and Sonia Jaeger

Chemnitz University of Technology, Germany

We explored how well common theories about the impact of post-event information on memories explain recollections that occur naturally in university students’ study routines. Instead of starting from a familiar research paradigm, such as those used in hindsight-bias research, the present study used a situation common to university students, and examined how well three candidate explanations *judgemental anchoring, implicit theories of change, and motivational influences *could explain the results we obtained in a long-term memory study that included three sessions, six months apart. We found that about two thirds of the memories of study-related issues were indeed biased, and that the impact of post- event information being used as an anchor is the most plausible explanation for the results. There were also some indications that memory biases might have been due, at least in part, to motivational factors.

Our memories are continuously updated by new experiences. Many if not most of the experiences

we have on a daily basis are quite similar to ones

we have had in the past. What effect does this

have on the memories of the original experi-

ences? Consider a simple hypothetical example.

Six months ago you were drinking coffee and

eating a delicious piece of cake with an old friend.

You were a little overweight at that time, and in

the course of your conversation you made a

prediction about how many kilograms of weight

you intended to lose in the next six months. Now

you meet your friend again and she asks you

whether you remember your prediction. In the

meantime, you might have had similar conversa-

tions (including the predictions) with other peo-

ple, talked to your spouse or to other friends

about your conversation, read in a magazine

about the probability of predictions of this kind

coming true, or watched a documentary on the

success of dietary measures on TV. You might

also be aware of how many kilograms you have

actually lost in the last six months, or you might

just now have made another prediction about

your intended weight loss in the next six months.

Will any of these kinds of information, which are

obviously all related in some way to the original

memory content*your prediction*have a sys- tematic impact on its recollection? If so, will

information that more directly relates to the

original event (e.g., knowing that you have

actually lost 4 kg of weight over the last six

months) have a stronger impact than a more

indirect piece of information (e. g., your new

prediction that you will lose 5 kg in the six months

to come)? Will additional assumptions about

Address correspondence to: Peter Sedlmeier, Chemnitz University of Technology, Department of Psychology, 09107 Chemnitz,

Germany. E-mail: [email protected]

We would like to thank Tilmann Betsch, Hartmut Blank, Anita Hewer, Anita Todd, and two anonymous reviewers for their

helpful comments on an earlier version of the paper.

# 2007 Psychology Press, an imprint of the Taylor & Francis Group, an informa business

MEMORY, 2007, 15 (1), 70 �92

http://www.psypress.com/memory DOI:10.1080/09658210601087068

some mechanisms of change involved (e.g., as- sumptions about the time course of your weight loss) also influence the memory of the original prediction? Will the strength of your motivation to lose weight have any impact? Will the recollec- tion of the original event get worse if it is repeatedly tested? All these questions deal with the impact of different kinds of subsequent or post-event information*that is, information re- ceived, generated, or experienced after the origi- nal judgement or prediction*on the recollection of that judgement or prediction.

The main aim of the study reported here was to explore how well common theories about the impact of post-event information explain natur- alistic memories. Instead of starting from a familiar research paradigm, such as those used in hindsight-bias research, we begin with a situa- tion common to university students, and explore the impact of different kinds of post-event information. In this paper we first review current theoretical approaches to explaining memory distortions that occur after post-event informa- tion is received or generated. Then we give an outline of our study, which included the repeated collection of students’ memories over a period of one year. We examine how well different expla- nations can account for our results. Finally, we discuss the limitations and possible extensions of our exploratory approach.

IMPACT OF POST-EVENT INFORMATION

There is abundant evidence that post-event in- formation of different sorts can indeed bias memory contents (e.g., Ayers & Reder, 1998; Christensen-Szalanski & Willham, 1991; Guil- bault, Bryant, Brockway, & Posavac, 2004; Ross, 1989). However, exactly why, how, and to what extent this happens in specific circumstances is still open to debate. Explanatory approaches differ in respect to the underlying memory models, the processes held responsible, and also in the role conceded to motivational processes. Some of the theoretical accounts only apply to the impact of certain kinds of post-event information, whereas others have not been specifically devel- oped to explain the biasing impact of post-event information or other false-memory effects, but instead cover a wide variety of issues related to memory.

Varieties of design

Before discussing the theoretical accounts, it might be helpful to look at how the impact of post-event information is usually studied. A large number of studies dealing with this topic have used variants of the following design, consisting of three stages. In stage 1, participants make an observation or a judgement: This is the original memory content. In stage 2, they receive informa- tion related to the original memory content. The nature of this post-event information varies, however. For instance, in studies on the ‘‘mis- information effect’’ (e.g., Loftus, Hoffman, & Wagenaar, 1992), this information is potentially misleading (e.g., a picture of a traffic situation that is identical to a picture shown in stage 1, except that a stop sign is replaced by a yield sign), whereas in ‘‘hindsight bias’’ studies (e.g., Hawkins & Hastie, 1990), the postevent information is usually the actual outcome of a prediction or the true state of affairs. Finally, in stage 3, participants are tested about their original memory contents.1

This test can be of the ‘‘yes�no’’ variety (mostly in research on the misinformation effect), but can also consist of making numerical judgements (common in studies on the hindsight bias). In a typical study of this type, the items used are of the ‘‘almanac type’’ (e.g., participant’s judgement: ‘‘The height of the Eiffel tower is 260 metres.’’), the new information is provided by the experi- menter (e.g., ‘‘The actual height of the Eiffel tower is 300.51 metres.’’), and the intervals between the three steps are rather short.2 The usual result in such studies is that recollection of the original memory content is biased by the post- event information (e.g., ‘‘I said that the height of the Eiffel tower is 285 metres.’’).

In other designs, especially outside the mis- information-effect and hindsight-bias paradigms, the ‘‘new information’’ is often provided by the participants themselves, and is usually quite self- relevant, and the interval between new informa- tion and original memory content can be sub-

1 Sometimes, in so-called hypothetical designs, the first step

is omitted, and effects of the new information are examined by

comparing the hypothetical recollections (‘‘What would have

been your judgement back then?’’) of participants who

actually received the new information with ones who did not. 2 There are, of course, also some exceptions to the typical

procedure in hindsight bias studies that have used more

personally relevant questions and longer intervals, such as, for

instance, Renner (2003), Stahlberg, Eller, Romahn, and Frey

(1993), or Mark and Mellor (1991).

STUDENT MEMORIES 71

stantial (see Ross, 1989, for a collection of such studies). Again, the new information is thought to bias the recollection of earlier related memories, but usually some additional expectations or (mostly implicit) theories are assumed to con- tribute to and, therefore, bias the recollection of older memories.

The processes elicited by these postulated expectations can be quite diverse. For instance, when spouses are asked about how their relation- ship satisfaction developed over time, they tend to recall that satisfaction declined early on in the relationship but improved more recently, although a common empirical pattern is a mono- tonic decrease (and not a U-shaped pattern) in the level of relationship satisfaction (e.g., Vaillant & Vaillant, 1993). In this case, participants’ current level of relationship satisfaction (the ‘‘postevent information’’) influenced their recol- lection of satisfaction over time. There are now several possible explanations of how the biased recall of prior levels of satisfaction might arise, based on either cognitive or motivational pro- cesses (Karney & Frye, 2002, p. 235): People might, for instance, use an implicit theory about change, and expect changes to proceed in a U-like manner; they might be motivated to selectively remember specific events that occurred during the time in question, or they might accurately remember past events but change their standards (e.g., the standard by which an event is considered unpleasant) to evaluate those events over time.

Representational assumptions

A sound theory about how memory contents can be biased requires some assumptions about how experiences are represented and accessed in memory. One might argue that there are basically three different views on this issue (for detailed overviews, see Reyna, Holliday, & Marche, 2002; Reyna & Lloyd, 1997). The first view, sometimes termed constructivist, holds that new information is immediately assimilated with existing know- ledge, thus constructing a mix of old and new (e.g., Fischhoff, 1975; Loftus, 1975). Only the modified memory traces can be accessed, and if these differ from the original ones, one obtains biased memory contents. Thus, the source of the bias is attributed to a storage failure.

The second view, which can be summarised under the term ‘‘source monitoring’’ (e.g., John- son, Hashtroudi, & Lindsay, 1993), contends that

both the content and the source of a memory trace are crucial. According to this view, biased recollections mainly arise from confusions about the source of the memory content. The new information might bias the retrieval process for those memory contents that have been forgotten or are not accessible at the moment of retrieval, although they might be accessible later either spontaneously or with a different kind of prompt (e.g., McCloskey & Zaragoza, 1985; Schwarz & Stahlberg, 2003). Thus, in this view, the source of the bias is attributed to a retrieval failure.

Whereas the constructivist and the source- monitoring views rely on a global kind of memory with different properties, a third view postulates the existence of different kinds of memories or judgement processes. A prominent example of such a view is fuzzy-trace theory (e.g., Brainerd & Reyna, 2002; Reyna & Brainerd, 1995), which holds that the surface forms and the meaning contents of experience are stored in parallel, with the representations of the former*verbatim traces*dissociated from the latter*gist traces. Obviously, verbatim traces (e.g., ‘‘at 9:35 p.m.’’) become inaccessible more rapidly than gist traces (e.g., ‘‘after dinner’’). A verbatim trace can also contain source information. Because in fuzzy- trace theory forgetting is seen as the disintegra- tion of a trace, the content and source of a verbatim trace might eventually become sepa- rated, with the source being forgotten more rapidly than the substance of events (e.g., Reyna & Lloyd, 1997). This is one account of how new information (a new verbatim trace) might bias memory contents: it might cue a wrong (old) verbatim trace. More often, however, the basis for biased recollections might be that people answer questions that ask for (non-accessible) verbatim traces by retrieving gist traces, which they mis- takenly attribute to experience. As gist traces, which are much more durable than verbatim traces, usually deviate at least slightly from verbatim traces, this results in biased recollec- tions.

There is increasing evidence that a pure con- structivist view on how new information can bias original memory contents cannot explain a vari- ety of findings. Memory traces can still persist despite an inability (even if transient) to access them. The source-monitoring approach accounts for this, as does the fuzzy-trace theory. An advantage of the latter can be seen in its distinc- tion between verbatim and gist memory that captures a wide variety of false-memory effects

72 SEDLMEIER AND JAEGER

(Brainerd & Reyna, 2002). In addition, it seems to connect well with a widely used explanation in judgement research that has also been used concerning hindsight-bias effects: judgemental anchoring.

Judgemental anchoring

The concept of judgemental anchoring was made popular by Tversky and Kahneman (1974, p. 1128) as a process in which ‘‘people make estimates by starting from an initial value that is adjusted to yield a final answer [and] . . . adjustments are typically insufficient.’’ This explanation has been applied to numerous judgement phenomena (see Chapman & Johnson, 2002, for an overview) and also to memory biases after post-event information, such as the hind- sight bias (e. g., Hawkins & Hastie, 1990). Mean- while, however, there is growing evidence that insufficient adjustment cannot be upheld as the central mechanism of the anchoring effect. In- stead, the anchor seems to influence a subsequent judgement by selective associative processes; that is, by increasing the accessibility of features that the anchor and the target judgement have in common (e.g., Chapman & Johnson, 1999; Muss- weiler & Strack, 1999a). Taking these results into account, Mussweiler and Strack (1999b) proposed the selective accessibility model of judgemental anchoring, which is based on two hypotheses. The selectivity hypothesis states that participants compare the target judgement to the anchor value, thereby generating knowledge that is con- sistent with the notion that the target’s value is equal to the anchor value. The accessibility hypothesis then postulates that generating such knowledge makes it more accessible, so that it is used for the target judgement.

Although, to the best of our knowledge, the selective accessibility model has not been applied to memory biases thus far, it seems to be a suitable candidate explanation for the influence of post-event information on the recollection of memory contents. Usually in judgement research, the anchor value is introduced by a comparison procedure (e.g., ‘‘Is the percentage of African countries in the UN lower or higher than 65%?’’). In studies on memory biases there is almost never such a comparison, but anchoring can also be achieved without any comparison process if participants devote sufficient attention to the anchor (Chapman & Johnson, 2002). This should

be the case for most kinds of post-event informa- tion used in the respective studies.

What might one expect to happen according to the selective accessibility model? If one assumes the distinction between verbatim and gist traces made by fuzzy-trace theory, nothing should hap- pen if the verbatim trace asked for (the original specific memory) is very strong. However, if forgetting has had an effect, the post-event information (the anchor) should activate know- ledge consistent with it, and if source and content information are already dissociated, the wrong contents might be more strongly activated than the original memory trace. The more the verba- tim traces have decayed, the higher are the chances that gist memory consistent with the post-event information will be activated and reported as the original memory. A general effect of judgemental anchoring would be that recol- lected original memories become more similar to the post-event information.

Expectations about specific changes

Judgemental anchoring can be expected to change original memories in a systematic direc- tion: towards the anchor. However, particularly with personally relevant issues, people seem to possess quite elaborate expectations or theories about how things change over time. In this case, they might use the status quo or some other piece of new information as a starting point, but adjust their memories with the help of a specific expectation or theory, which might not be explicit to them.

Implicit theories about what changes in one’s life and what remains stable seem to play an important role in our memories. Ross (1989) surveyed relevant research and found that if there were discrepancies between implicit theories and facts, theories had a good chance of being accepted as the truth. For instance, if participants in an inefficient study skills programme (fact: no change) have to remember their skills before the programme, and if they believe in the efficiency of that programme (implicit theory: positive change), they take the status quo (the new information: not really doing so well) as a starting point, and may tend to exaggerate how poorly they did before (Conway & Ross, 1984). Implicit theories also work for hypothetical scenarios. For instance, Hirt (1990) provided participants with information about ‘‘JW’’s mid-term grades

STUDENT MEMORIES 73

(original information). Then, participants’ expec- tations (their implicit theories about change) were manipulated in that some were led to expect JW’s grades to improve and others to expect his grades to decline or remain the same. When JW’s final grades were given (the new information), participants’ memories about the mid-term infor- mation were systematically influenced by their implicit theories. Those who expected improve- ment recalled lower mid-term scores, whereas those who expected decline remembered higher scores.

Motivational influences

Expectations about changes over time can be modified by differences in personality variables or motivational concerns. For instance, Story (1998) found that recall accuracy for feedback was dependent on level of self-esteem: Indivi- duals with high self-esteem recalled unfavourable feedback less accurately than favourable feed- back, whereas for individuals with low self-esteem the reverse was true. In another study (Safer & Keuler, 2002), clients for whom psychotherapy was least successful were particularly likely to overestimate its success by recalling more pre- therapy distress than they actually had, whereas those who improved the most were more accu- rate. A similar effect was found in the recollec- tions of grief following the death of spouses: Participants whose grief diminished only little over time tended to overestimate prior grief more (Safer, Bonanno, & Field, 2001).

In general, ‘‘failures’’ seem to lead to stronger memory distortions than ‘‘successes’’ (Renner, 2003; Tykocinski, 2001). For instance, Haslam and Jayasinghe (1995), who in their study on the hindsight effect used recollections of predicted grades as the dependent measure, found a marked discrepancy between the judgements of particip- ants who had made overly optimistic predictions (and experienced failure) and the judgements of their more pessimistic counterparts. If students’ predictions about their mid-term grades were better than their actual grades (optimistic predic- tion), most of them exhibited the usual hindsight bias; that is, they remembered their predictions as lying somewhere between their actual predictions and their actual grades. If, however, students’ grades were better than predicted (pessimistic prediction), the majority of them exhibited a so- called reverse hindsight bias; that is, they recol-

lected predictions of even better grades. Motiva- tional processes in the recollection of grades already seem to arise when only the status quo is known: Bahrick, Hall, and Berger (1996) found that the academic achievement of university students (the status quo) correlated substantially with biases in the recollection of their high-school grades. A common pattern in all the studies cited above seems to be that for highly personally relevant variables or variables that pertain to self- appraisal, the status quo (the ‘‘postevent informa- tion’’) or the discrepancy between the status quo and the original memory contents covary with the size of the memory distortion.

Length of time intervals and repeated feedback

In everyday life, the amount of time between the encoding of a memory content and its recollection can vary considerably, and so can the frequency of some kinds of feedback. (Henceforth, we will use the term feedback to refer to post-event informa- tion that relates to the original memory content.) The general finding about the impact of the length of time between the encoding of the original memory and feedback seems to be that a longer interval results in stronger memory biases (e.g., Belli, Windschitl, McCarthy, & Winfrey, 1992; Blank, 1998; Blank, Fischer, & Erdfelder, 2003; Chandler, 1993; Hirt, Erick- son, & McDonald, 1993; McDonald & Hirt, 1997).

With repeated feedback, however, the expec- tations are not so clear-cut (see Reyna & Lloyd, 1997). What would one expect given the theor- etical approaches discussed above? If the (deviat- ing) repeated feedback is consistent or identical over time, a constructivist memory update should lead to a stronger effect with repeated than with a single feedback, because it would be expected to modify the original memory content more strongly towards the feedback. If feedback is inconsistent, predictions would depend on the order of the different instances of feedback, and the latest feedback could be expected to have the strongest effect. Fuzzy-trace theory seems to make similar predictions if there is still some access to the original verbatim memory. However, if only gist memory is accessible, the spacing of feedback might be decisive. If, for instance, the first feedback was given very shortly after the original memory content was encoded, the gist

74 SEDLMEIER AND JAEGER

memory might include both original memory and feedback. If, however, feedback is given after long intervals, it should not make a big difference whether feedback is repeated or consistent.

THE CURRENT STUDY

Instead of examining the predictions of a given memory research paradigm, we explored what happens with naturalistic memories. We focused on the achievements and opinions of university students during the first three semesters of their studies. Apart from finding out more about students’ prevalence of memory biases for study-related issues, we wanted to explore the roles of judgemental anchoring, specific expecta- tions, and motivational influences on their mem- ory recollections.

To examine the impact of anchoring processes, we had participants generate feedback that we expected them to use as anchors. Because we also wanted to find out about the impact of specific expectations, we had to choose kinds of judge- ments that very likely deviated from the anchors in different directions. This is necessary to make differential predictions: If, for instance, particip- ants expected a small improvement in some skill, and if the feedback value was consistently higher than the original memory content, a recollection of the level of skill lying in between original memory and feedback would be consistent with both a simple assimilation process due to judge- mental anchoring and the working of the specific expectation (small positive change). If, however, the feedback value was lower than the original memory, different predictions could be made: judgemental anchoring should lead to results lying above the feedback value, whereas the specific expectation of a small improvement should yield results below that value.

Because we also wanted to study possible motivational influences, we looked for judge- ments that we expected to differ in personal relevance. For variables with higher personal relevance, one would expect higher correlations between the amount of memory distortion on the one hand and the discrepancy between feedback and original memory on the other hand than for variables with lower personal relevance. Bearing these constraints in mind, we decided to examine students’ judgements and predictions about study-related issues that we expected to vary in their degree of personal relevance.

Another important aim of the study was to find

out more about the impact of different kinds of

feedback. Whereas in hindsight-bias research, the

feedback is usually the actual outcome of a

prediction or the true state of affairs, in daily

life (and in other research traditions) feedback is

often only indirectly related to the original

memory content, as shown in the introductory

example. We wanted to find out whether feed-

back that referred to the outcome of a prediction

(the original memory content), which we term

direct feedback, yielded a stronger memory bias

than feedback that only indirectly referred to the

prediction, such as a new prediction, which we

call indirect feedback. What should one expect

here? From a constructivist perspective one might

expect outcome information to elicit a stronger

memory distortion, because it directly refers to

the prediction, whereas judgemental anchoring in

conjunction with the assumption that verbatim

and gist memory are separately stored could lead

us to the rather surprising opposite expectation.

Because a new prediction about a related issue

might be more similar to the old prediction than

information about the outcome of the old predic-

tion, the biasing effect of the indirect feedback

might be expected to be stronger if only gist

memory was accessible. Finally, because we were also interested in

memory changes over time in such a natural

environment, we had participants repeatedly

produce self-generated feedback, and we also

tested their memories repeatedly. In particular,

we wanted to find out whether students’ recollec-

tions improved over time; that is, whether stu-

dents were able to disregard feedback better

when it was given repeatedly in a naturalistic

context; and we were interested in whether

the length of the interval between the creation

of the original memory contents and the self-

generated feedback systematically influences re-

collections. We begin with a description of the design of

our study, which consisted of three sessions

altogether, scheduled six months apart. Then we

present the results of a pilot study, which served

to generate predictions for students’ implicit

theories about change. Finally, we report and

discuss the results of the main study, for what

participants recollected in the second and third

sessions.

STUDENT MEMORIES 75

GENERAL PROCEDURE

We were interested in finding out more about the error-proneness of personally relevant memories in a naturalistic environment and the role of potentially biasing mechanisms. To simplify mat- ters, we looked for a context with a relatively high degree of structure and a good availability of data. These conditions are met in students’ first two years of university.

General schedule

Students’ judgements and self-generated feed- back were collected in three sessions at the beginning of three consecutive semesters, with about six months in between. In each of the three sessions, participants first gave judgements or predictions about their current opinions that all pertained to peculiarities and expectations about their studies. In the second session (second semester), participants were then asked to recall the judgements they had given in the first semester. In the third session (third semester), participants were asked to recollect both the judgements given in the first semester and those given in the second. The order of recollections about these two sessions was completely counter- balanced.

General method

Choice of participants. In all parts of the study we concentrated on one cohort of psychology students; that is, those who began their studies at Chemnitz University of Technology in the winter semester of 2003/2004. Because we did not want to sensitise this group*henceforth termed the ‘‘naturalistic memory group’’*by selectively ask- ing only them to attend the different parts of the study, we also gathered data from students in later semesters who happened to be present in the respective settings. We will, however, only report the data for the naturalistic memory group. For reasons of feasibility, we concentrated on stu- dents’ study-related memories in courses that dealt with the topic of psychological methods. The first two sessions took place at the beginning of two successive statistics classes, and the third session was held at the beginning of a class on experimental practice. Most of the first-semester students who took the methods course could also be expected to attend the two further courses,

taught mainly to second- and third-semester students, that were to take place in the following two semesters.

Basic judgements. The judgements or questions used in this study were part of larger question- naires that also contained other questions used to obtain data to illustrate statistical procedures. In all three sessions, judgements and answers were given anonymously, but participants were asked to use the same four-character ID: second letter of mother’s first name, third letter of birth place, number of siblings, and last letter of own first name. This ID allowed us to relate participants’ recollections to their earlier judgements. Table 1 contains the judgements solicited in session 1 and referred to in the other two sessions.3

Basic analyses. In our analyses we always first determined the percentage of correct recollec- tions. For the incorrect responses (that is, the memory distortions), we examined how well each of the three mechanisms described above*jud- gemental anchoring, specific expectations, and motivational factors*explained the results.4

Our analyses of the results of sessions 2 and 3 were based on all the participants of the natur- alistic memory group whose data were available for the respective session. We also conducted additional analyses with only the participants who gave judgements in all three sessions. However, the results of the latter analyses did not differ systematically from the results for all available participants, and therefore we report only the former.

STUDENTS’ IMPLICIT THEORIES ABOUT CHANGE

To be able to make predictions about students’ implicit theories about change in respect to the topics dealt with in this study (Table 1), we conducted a separate study outside the classroom context. This procedure was intended to prevent

3 The material used in all parts of the study was originally

in German. 4 We counted a response as correct if it matched the

original judgement or prediction exactly. One might object

that this criterion is too strict, because it is difficult to

remember an arbitrary number. However, an inspection of the

judgements or predictions revealed that 96% of all judgements

in session 1 were multiples of 10 or 5, and so were 94% of all

judgements in session 2. Thus, numbers were relatively easy to

remember.

76 SEDLMEIER AND JAEGER

students from making an explicit connection between their implicit theories of change and their judgements and memory recollections, which were always measured in classroom set- tings. The separate study was done in between sessions 2 and 3, to get an up-to-date version of participants’ implicit theories about change. This choice of timing also ensured that students who switched to another university after the first semester*which quite a few did that year as a result of administrative events*were not in- cluded in the predictions.5

Participants and procedure

A total of 74 psychology students from Chemnitz University of Technology participated in this study, but only the data of the 52 members of the specific cohort (then in their second semester) were used later on (82.7% females, mean age�/23.4 years). At the end of an un- related computer-based experiment, participants were given a questionnaire. The topic of the study was introduced to the students by informing them

that expectations and opinions sometimes change

over time, but at other times remain stable.

Participants were told that they would be asked

about changes in students’ expectations and

opinions on several aspects of the first two years

of their studies. In particular, they were required

to choose the pattern of change (or stability) out

of those shown in Figure 1 (also included in the

questionnaire) that they found to be typical for a

given aspect.6 They were instructed to choose the

pattern that was most similar to the pattern they

had in mind if they thought that the most typical

pattern was not among the three shown. The specific study-related aspects concerned

the probability with which students in this stage of

their studies expected to switch to another uni-

versity7 (i.e., does this probability typically re-

main stable over time, increase, or decrease?), as

well as several questions about courses in differ-

ent subdisciplines of psychology. The subdisci-

plines included ‘‘motivation and emotion’’,

‘‘methods’’, ‘‘learning and memory’’, and ‘‘per-

sonality psychology’’, which are commonly taught

by different groups in German psychology de-

partments. In the main study, participants were

only asked questions about the methods course.

The questions about the other courses were

TABLE 1

Judgements solicited in session 1 and referred to in the other two sessions

Topic Judgement

Switch The probability that I will later (e.g., after the first 2 years) switch to another university is: ______%

(please insert value between 0 and 100)

Grade I expect the following grade on the exam for this course: _____ a

Interest My interest in this course: _____ (please insert number between 0 and 100: 0�/ I find this course totally uninteresting, 100�/ I find this course extremely interesting)

Workload To prepare for a session of this course and afterward to review the contents of the session, I assume

that I need on average ____ minutes (please insert number)b

Relevance In my opinion, the relevance of this course for my studies in comparison to the other courses I am

taking this semester is _____ (please insert number between 0 and 100: 0�/ this course is totally irrelevant compared to the other courses, 100�/ this course is extremely relevant compared to the other courses)

Session 1 was winter semester 2003/2004. a Possible values range from 1 (best) to 4 (worst grade for passing) or 5 (failed). The basic grades (1 �3) can be qualified by .3 and

.7 decimal points (e.g., 1.3 and 1.7). bOnly used in sessions 1 and 2.

5 In Germany, places for psychology students are restricted

and are mainly distributed by a federal agency. Because there

are many more applicants than free spaces, many students who

are not granted a space try to get into the psychology

programme of any university by petititioning these

universities, because once admitted to any programme,

students can change universities. For the semester in

question, about 90 students petitioned Chemnitz University

of Technology to get admitted to the psychology programme.

A large number of the plaintiffs were, in fact, admitted, but

left the university again after completion of the first semester.

6 We restricted ourselves to three linear patterns because

the period concerned is relatively short, and opinions are

therefore assumed to change (if they change) only in a way

that can be roughly described as linear. 7 In Germany, the best occasion to change universities is

after the ‘‘Vordiplom ’’, which is roughly comparable to the

bachelor’s degree, and which is usually obtained after two

years of study.

STUDENT MEMORIES 77

included so as not to direct the participants to use their implicit theories explicitly in their memory judgements in the main study. For each of the courses (in the different subdisciplines), partici- pants were asked to choose the respective pattern (Figure 1) for the following topics: interest (i.e., does it typically remain stable, increase, or decrease over time); relevance of the course in question in comparison to other courses; work- load in a given course; and grades (i.e., do grades remain the same, get better, or get worse over time).

Results and discussion

Because later we will only be concerned with participants’ implicit theories about methods courses, only those results are given in Table 2. The table shows that whereas the opinions about the changes in the probability of switching to another university after two years are about evenly distributed among students, there are more pronounced preferences for the other four topics. For instance, a majority of students ex- pected their interest in methods courses to remain stable, and for the workload as well as the relevance of these courses to increase over time. If one looks at the patterns for all students asked (including later-semester students), the picture is quite similar: The mean absolute deviation from

the percentages shown in Table 2 is only 2.1%. Thus, the implicit theories about stability and change shown in Table 2 seem to be quite constant across different cohorts of students.

FIRST RECOLLECTION: SIX MONTHS LATER

As already mentioned, we had students give study-related judgements (Table 1) in a methods course at the beginning of their first semester. At the beginning of the following semester, in the follow-up methods course, we collected students’ memories about the judgements they had given in the first semester. We were interested in how well they could remember their original judgements, and how their memory biases, if any, could best be explained. The three candidate explanations ex- amined here were the use of feedback as an anchor, the impact of implicit theories, and motivational influences.

Method

Participants. A total of 61 psychology students participated in this part of the study. Of these, 10 were students who had studied for more than two semesters and taken the course as a refresher. Only the data of the 51 students in their second semester of study were used in further analyses.

time

1

time

2

time

3

Figure 1. The three theories about change over time among which students could choose: stability (1), increase (2), and

decrease (3).

TABLE 2

Implicit theories about how study-related opinions and results change over time for methods courses

Pattern

Topic Stability Increase Decrease

Switch (n�/46) 34.8 28.3 37.0 Grade (n�/50) 42.0 26.0 32.0 Interest (n�/52) 50.0 25.0 25.0 Workload (n�/52) 23.1 61.5 15.4 Relevance (n�/51) 37.3 51.0 11.8

Percentages of respondents who chose the respective patterns.

78 SEDLMEIER AND JAEGER

Of these 51 students, quite a few were replace-

ments for the students who had left Chemnitz

University of Technology after the first semester

(see Footnote 2). So in the end, 31 students

(80.6% females, mean age�/21.2 years) who had filled in the first questionnaire also completed the

second one.

Procedure. Participants began by filling in the first page of a questionnaire that included the

judgements shown in Table 3. Note that there

were two kinds of self-generated feedback. For all

five topics, participants generated indirect feed-

back by making new judgements or predictions

about the respective topics; that is, judgements

that pertained to the current (second) semester.

In addition, for two of the topics*grades and workload*they also produced feedback that was more directly connected to their original mem-

ories*that is, direct feedback*by having to remember their actual grades and their actual

workload in the first semester. This was done to

find out whether indirect feedback had as strong

an effect as direct feedback. After finishing the

judgements on the first page of the questionnaire,

participants turned to the second page, where

they were first reminded that at least some of

them had given answers to the same questions in

the previous semester. They were then asked

whether they had participated in the first session

and, if so, to write down what they had answered

then (see Table 1). Otherwise*if they had not participated in the first session*they were asked to give the judgements that they would have given

if they had participated back then (these data

were not used in the analyses).

Results

Correct recollections. Participants’ memories about the judgements they had given in the first session were apparently best for workload (41.9% correct recollections) and worst for interest (22.6% correct recollections), with a mean of 32.3% (see Figure 2). However, these values are upper boundaries of true memory recollections, because original judgements and recollections might be the same for other reasons. Even if participants did not remember their original judgements (e.g., interest in the course), they might on both occasions think about plausible values for such judgements, and could arrive at identical results. Moreover, nondifferences might just be lucky guesses. But even if the proportions

TABLE 3

Judgements asked for in the first part of session 2 and session 3

Topic Feedback Judgement

Switch Indirect The probability that I will later (e.g., after the first 2 years) switch to another university

is: ______%

Grade Indirect I expect the following grade on the exam for this course: _____

Direct My grade in the first course was: ____ a

Workload Indirect To prepare for a session of this course and afterward to review the contents of the

session, I assume that I need on average ____ minutesa

Direct Last semester, I needed on average ____ minutes to prepare for a session of the course

and afterward review the contents of the session a

Interest Indirect My interest in this course: _____

Relevance Indirect In my opinion, the relevance of this course for my studies in comparison to the other

courses I am taking this semester is ____

Session 2 was summer semester 2004. Session 3 was winter semester 2004/2005. For permissible answers see Table 1. a Asked in session 2 only.

Switch (25.8%)

Grade (35.5%)

Interest (22.6%)

Workload (41.9%)

Relevance (35.5%)

0.00

0.10

0.20

0.30

0.40

0.50

P ro

p o

rt io

n o

f co

rr e

ct r

e co

lle ct

io n

s

Figure 2. Proportions of correct recollections in session 2 for

the five judgements given in session 1.

STUDENT MEMORIES 79

of nondifferences shown in Figure 2 had all been due to intact memories, the figure makes clear that, for all five topics, the memory recollections of the majority of participants were not correct. The interesting question is whether the deviations between original judgement and recall were unsystematic or whether they were systematically influenced. The potential impact of three such plausible systematic influences on recollections* the use of feedback as an anchor, implicit theories about change, and a potential motivational influ- ence*is examined next.

Indirect feedback as an anchor. If self-gener- ated indirect feedback is used as an anchor for reconstructing lost or inaccessible memory con- tents, then the recollections should on average ‘‘move’’ from the original judgements towards the feedback values. To examine this hypothesis we used the index d (for a similar index see Hell, Gigerenzer, Gauggel, Mall, & Müller, 1988):

d� recollection � original judgment

feedback � original judgment ; if

feedback�original judgment "0 (1)

Note that d is only positive if the deviation from the original judgement is in the direction of the feedback, otherwise it has a negative value. It has a value of 1 if the recollection is identical to the feedback. The fact that d is undefined if feedback and original judgement coincide also makes sense in light of the argument about the use of feedback as an anchor: only if the feedback deviates from the original judgement*which was the case in an average of 78% of the five judgements*can assimilative memory processes be influenced by the feedback bias recollections of the original judgement. The results of this analysis*excluding the undefined results (see equation 1)*are shown in Table 4. To get a more exact impression of the effects of indirect

feedback on memory distortions, we also calcu- lated one-sample t -tests as well as effect sizes. The degrees of freedom vary across tests, because there were outliers*defined as extreme values in the respective box plots (Tukey, 1977)*that first had to be removed to establish the preconditions for calculating a t -test. Table 4 shows the results of the tests (t -values, degrees of freedom, and p -values) as well as mean ds and the effect size d , which was calculated by dividing the ds by their standard deviations. The overall effect of feed- back as measured by the ds is strong throughout the five judgements and huge for the question about students’ judged probabilities for switching universities.

Indirect versus direct feedback. Making new predictions about issues related to some original memory contents is quite indirect self-generated feedback, and might influence recollections of those memory contents less strongly than would trying to remember the actual outcome of a former prediction. The latter*remembering the actual outcome*is sometimes used as an anchor in hindsight studies, whereas to the best of our knowledge, the former*giving a judgement si- milar to an earlier one*has never been used in studies of that kind. For two of the topics, grades and workload, we had participants produce both indirect and direct feedback. Whereas indirect feedback was obtained by having them make predictions for the ongoing course, participants produced direct feedback by recollecting the actual outcomes relating to their predictions from the first semester (see Table 3). Figure 3 shows the differences in ds for direct versus indirect feedback. Contrary to what one might expect, direct feedback seemed to have had a less systematic influence on recollections than indirect feedback. These differences amount to effect sizes of d�/0.46, t (17)�/ 2.133, p�/ .024, and

TABLE 4

The use of feedback as an anchor as indicated by the size of d, as found in the second session

Topic t df p Mean d d

Switch 23.87 14 .000 .93 6.16

Grade 3.74 23 .001 .59 .76

Workload 2.81 21 .006 .38 .60

Interest 3.95 24 .001 .75 .79

Relevance 2.89 19 .005 .43 .65

Results of one-sided one-sample t -tests (t , df , and p ) and effect sizes (mean d, and d ). All calculations were done without outliers, which were determined by constructing the respective box plots. The d s were calculated by dividing

mean d by the standard deviation of d.

80 SEDLMEIER AND JAEGER

d�/0.37, t (17)�/ 1.660, p�/ .058, for grades and workload, respectively. The differences shown in

Figure 3 were confirmed by regression analyses,

with direct and indirect feedback as predictors,

and the size of the memory distortion (d) as the criterion. For the judgements about both grades

and workload, indirect feedback proved to be a

much better predictor (b�/ .702 and b�/ .791, respectively) than direct feedback (b�/ �/.236, and b�/ �/.210, respectively).

Impact of implicit theories. Consider, for in- stance, a participant who has just recorded that

she expects a grade of 2.0 in the current course.

Let us assume that she thinks that grades in

methods courses typically improve over time. If

she now cannot remember which grade she had

expected in the first methods course, she should

reconstruct her prior judgement by using her

implicit theory about change: she might recollect

that she had expected a higher (worse) grade

(e.g., a grade of 2.7) back then. (Note that in

Germany, the best grade is 1 and the worst* ‘‘failed’’*is 5). Take another example: If a participant’s implicit theory told him that his

interest in methods courses would increase over

time, his recollected judgement about his interest

in the first-semester course should be lower than

his new judgement, if he applied his implicit

theory. Table 2 summarises the predictions as

gathered from a sample that was representative of

the students questioned in this second session. If

participants had based their judgements on their

implicit theories about change, similar percen-

tages should be found for the patterns (stable,

increase, or decrease) between remembered jud-

gements (first session) and current judgements.8

As a measure of the fit between the predicted

patterns and the actual patterns, we calculated

chi-square tests, all with df�/ 2 and N�/ 31. Take, for instance, the predicted patterns for the

‘‘interest question’’ from Table 2 (third row):

50% of the participants in the separate study

thought there would be no change in their

interest, and 25% thought that interest would

increase or decrease. This is the prediction one

would expect if the natural memory group’s

implicit theories of change determined their

recollections. Here the research hypothesis corre-

sponds with the null hypothesis. In such a case it is

indispensable to have some additional informa-

tion about the size of the effect. Therefore, we

report results of these tests, along with p values

and the effect size measure w, in Figure 4. What results would one expect if participants’

recollections were determined by their implicit

theories? Proportions of participants divided into

the three patterns for the actual data should be

very similar to the predicted proportions, which in

turn would result in small chi-square values and

small effect sizes (w B/ .1), as well as large p

values. Clearly, these expectations were not

fulfilled in our results (Figure 4).

Grades Workload

Direct feedback

Indirect feedback

–4.00

–3.00

–2.00

–1.00

0.00

1.00

2.00

3.00

4.00

Direct feedback

Indirect feedback

–1.00

0.00

1.00

2.00

3.00

S iz

e o

f “a

n ch

o r”

e ff e ct

( δ)

S iz

e o

f “a

n ch

o r”

e ff e ct

( δ)

Figure 3. Differences between the impact of direct vs indirect feedback found in session 2 for the judgements about grades and workload given in session 1.

8 A pattern was coded as stable if recollections and current

judgements were identical, as an increase if current

judgements were higher than the recollections, and as a

decrease if current judgements were lower than the

recollections.

STUDENT MEMORIES 81

Motivational influences. Prior research has shown that the actual difference between the status quo (the post-event information) and the to-be-

remembered status correlates with the size of the memory distortion in personally relevant variables

(e.g., Safer et al., 2001; Safer & Keuler, 2002).

There is also evidence that the current level of the status quo alone may be correlated with the

‘‘correctness’’ of the memory recollections (e.g., Bahrick et al., 1996). Such an effect seems to be

connected to the amount of self-relevance of a

given judgement. Thus, one should expect higher correlations between the size of memory distortion

and either the difference between status quo and recollection or the level of the status quo alone for

highly self-relevant judgements than for judge- ments low in self-relevance.

We had n�/12 additional students rate the relative self-relevance of the judgements on a

rating scale from 1 (lowest in self-relevance as compared to the other judgements) to 5 (highest

in self-relevance as compared to the other judge- ments), and they were advised that they could use

the same number for several topics. We found the following rank order (mean ratings in paren-

theses): grade (3.8), workload (3.0), switch (2.6),

interest (2.1), and relevance (2.1). The size of the correlations between ‘‘discrepancy’’ (feedback� original judgement) and ‘‘memory bias’’ (re- collection�original judgement) should follow the same pattern, if motivational concerns played a

role in the current recollections. Note that one would expect sizeable correlations for all kinds of

judgements, because one variable (original judge- ment) is contained in both ‘‘discrepancy’’ and ‘‘memory bias’’*the pattern of interest is only whether there is a systematic change in the size of correlations. The results for the judgements in their order of self-relevance are grade: r�/ .56; workload: r�/ .61; switch: r�/ .81; interest: r�/ .58; and relevance: r�/ .57. With direct feed- back, the results are grade: r�/ .27; and workload: r�/ .48. Thus it appears that, at least from this analysis, motivational factors did not play a systematic role.

We also correlated the status quo (i.e., feed- back at second session) with ‘‘correctness’’ (1�/ correct recollection, 0�/ incorrect recollection). Here the results for the variables in the order of self-relevance are grade: r�/ �.31; workload: r�/ �.37; switch: r�/ �.03; interest: r�/ .08; and rele- vance: r�/ �.58. With direct feedback, the results are grade: r�/ �.38; and workload: r�/ �.47. Thus it seems that students who had better grades, and students who predicted that they did not expect to work much, had a higher tendency to have incorrect recollections. A similar effect was also found for the judgement about the relevance of the course: Students who judged the relevance to be relatively low had a higher tendency for an erroneous recollection. However, this was not the case for judgements about interest and the prob- ability of switching to another university.

In addition to the above correlational analysis, we also explored whether ‘‘failure’’ led to stron- ger memory distortions than ‘‘success’’, as found previously (Haslam & Jayasinghe, 1995; Tyko- cinski, 2001). This had to be restricted to the question about grades and, possibly, workload, because in the other questions it is difficult to imagine a failure. Incidentally, as shown above, these two variables also received the highest ratings of self-relevance. Were the disappointed students (those who had made an overly optimis- tic prediction) more prone to exhibit systematic memory distortions? Figure 5 shows the results for the judgements of grades: LOWESS lines9

Switch Chi² = 3.29 p = 0.193

Grade Chi² = 17.79

p = 0.000

Interest Chi² = 12.00

p = 0.002

Workload Chi² = 9.58 p = 0.008

Relevance Chi² = 16.44

p = 0.000

0.00

0.20

0.40

0.60

0.80

E ff e ct

s iz

e w

Figure 4. Fit between implicit-memory predictions and

actual recollections in the second session. Note that the

smaller the chi-square and w, and the larger the p -value, the

better the fit. The horizontal line indicates an effect size of w�/ .1, which would be considered a small effect according to

Cohen’s (1992) conventions.

9 LOWESS lines (short for locally weighted scatterplot

smoother; Cleveland, 1985) are a powerful technique to

visualise all kinds of relationships between two variables. It

is a stepwise smoothing technique that in each step (along the

x axis) only takes into account a certain percentage of all data

points and, in addition, weighs close data points more strongly

than far-away ones. Thereby, the influence of any outliers on

the line is minimised. If the relationship between the two

variables is linear, LOWESS and regression lines are identical.

82 SEDLMEIER AND JAEGER

that illustrate the nonlinear relationship between

the discrepancy between feedback*either indir- ect (Figure 5a) or direct (Figure 5b)*and origi- nal judgement on the one hand and the size of the

memory distortion on the other. If feedback

works as an anchor, the discrepancy between

feedback and judgement should be systematically

related to the (absolute) size of the memory

distortion. This relationship is more pronounced

for those participants who were too optimistic in

their predictions and who therefore experienced a

‘‘failure’’. Unfortunately, the number of overly

optimistic students*that is, students who pre- dicted better grades than they actually got (or

whose original predictions were better than their

second predictions)*was quite small. Therefore, our results can only be interpreted as a trend, and

have to be treated with caution. A linear pattern was found for the judgement

about workload for indirect feedback. For direct

feedback (recollection of actual workload), the

results are consistent with the assumption of a

motivational influence (Figure 5c). However, the

results would imply that students experienced it as a failure if they had overestimated their work- load for the course (and it actually turned out to be substantially less than predicted).

Discussion

Six months after participants had given their first study-related judgements, they generated feed- back, either directly pertaining to the original judgements (by remembering the actual out- comes) or indirectly related to those judgements (by giving new judgements or predictions about the current course). They then tried to remember their original judgements. It turned out that only about one third of the recollections were correct. There were differences in the proportions of correct recollections for the different judgements*could these differences be due to more self-relevant memory contents being re- membered better? This seems not to be the case. Recall that the order of self-relevance we found

Indirect feedback - original 2.01.51.00.50.0–0.5–1.0–1.5

Ju d

g m

e n

t- o

ri g

in a

l

1.5

1.0

0.5

0.0

–0.5

–1.0

Direct feedback - original 1.00.50.0–0.5–1.0–1.5–2.0

Ju d

g m

e n

t- o

ri g

in a

l

1.5

1.0

0.5

0.0

–0.5

–1.0

“Success” (a) (b)

(c)

“Success”“Failure” “Failure”

Direct feedback- original 250–25–50–75–100

Ju d g m

e n t-

o ri

g in

a l

20

0

–20

–40

–60

–80

–100

–120

Figure 5. The differential impact of self-generated feedback for ‘‘failure’’ (overly optimistic predictions) versus ‘‘success’’ (correct

and overly pessimistic predictions). Shown are results for grade *indirect feedback (a) and direct feedback (b) *as well as workload *direct feedback (c).

STUDENT MEMORIES 83

in a little survey was grade, workload, switch, interest, and relevance. However, Figure 2 shows that judgements about grades were not remem- bered better than those about the relevance of the course. Thus, at least for the kinds of judgements we used, there seems to have been no systematic influence of the amount of self-relevance on correct recollections. The roughly two thirds of judgements not remembered correctly are close to what one could expect in laboratory studies with almanac-type questions. However, the rela- tively poor memories might in part be attributa- ble to the comparatively long delay between the original judgements and the recollections, (six months). For the incorrect recollections, we analysed whether they could be explained by any of three common theoretical explanations in the literature on memory failures: the impact of anchors, the working of implicit theories, and motivational influences.

It emerged that self-generated feedback ex- hibited a strong anchor effect for all judgements. Astonishingly, this effect was even stronger for indirect than for direct feedback. Such a result is, however, consistent with the selective accessibil- ity model of judgemental anchoring (Mussweiler & Strack, 1999b) in conjunction with the assump- tion that only gist memory is accessible (which seems to be a quite reasonable assumption with a six-month delay). After all, the indirect feedback was of the same type as the original memory: both were predictions. However, the effect does not seem to be restricted to predictions. The judge- ments about switch, interest, and relevance were not predictions but exhibited memory biases of comparable size. Indirect feedback is much more common in everyday life than direct feedback, and therefore the current results indicate that the kind of memory bias we found might be quite prevalent in natural settings.

Were our natural memory group’s judgements influenced by their implicit theories? Our results indicate that there was no strong impact, if any. However, subtle influences might have gone unnoticed, because our predictions were less detailed than those for the judgemental anchoring account. Whereas the former could be obtained easily for every participant, the latter were obtained only for the group as a whole. As already mentioned, this was done to prevent systematic sequential effects.

We also looked at the possible influence of motivational factors. There are some slight in- dications that such factors might indeed have

been working, although overall the result is mixed. Particularly in the two judgements with the highest degree of self-relevance (grades and workload), the level of the post-event information (predicted or recalled grade or workload) seems to have influenced the judgement. This result is complemented by the finding that a ‘‘failure’’ in the prediction of grades was tendentially asso- ciated with stronger memory distortions. The motivational influences might, however, be stron- ger with other kinds of judgements. Recall that the judgements of self-relevance we used were only relative judgements of relevance: Particip- ants were ‘‘forced’’ to use the whole scale between 1 and 5, but even so the range of mean judgements was not but very large. So our mixed results might also be attributable to the relative ‘‘closeness’’ in self-relevance of the five topics we used.

SECOND RECOLLECTION: ONE YEAR LATER

At the third session*12 months after the first one*we examined whether the main findings found after six months could be replicated; that is, whether the assumption of a process that used judgemental anchoring still explained the results well, whether implicit theories about change still did not contribute much to an explanation, and whether the data were consistent with the work- ing of motivational influences. We also wanted to examine whether feedback might have a detri- mental effect on the proportion of correct mem- ories: If so, omitting feedback on a given judgement should lead to better recollections for that judgement. Finally, we wanted to explore the impact of repetition and time interval. In particular, we were interested in whether the impact of anchors (if that still turned out to be the dominant explanation) decreased with repeti- tion, and whether memory distortions increased with increasing interval.

Method

Participants. A total of 49 students (all in their third semester) participated in the third session. Of these, 26 participated in both the first and the third session (76.9% females, mean age�/21.7 years), and 40 in both the second and the third session (82.5% females, mean age�/22.4 years).

84 SEDLMEIER AND JAEGER

Procedure. Participants received a question- naire that, on its first page, contained four of

the five judgements that provided indirect feed-

back, shown in Table 3. The judgements about

workload were omitted to find out whether feed-

back had a detrimental effect on participants’

proportions of correct recollections. Otherwise,

the procedure was identical to that used in the

second session, except that now participants were

asked to remember*on the second and third pages of the questionnaire*the judgements they had given in the first and in the second sessions.

The order of these judgements was counterba-

lanced across participants.

Results

The proportions of correct recollections did not

differ noticeably from those found in the second

session. Of the judgements given in the second

session, a mean of 30.5% were correctly recol-

lected six months later, and, of the judgements

given in the first session, the average proportion

of correct recollections 12 months later was

36.2%. The pattern across the five judgements

to be recollected was similar to that obtained in

the second session. Figure 6a shows the propor-

tions of correct recollections for the judgements

made in the second session, and Figure 6b those

for the judgements made in the first session. So,

the starting conditions for exploring possible

causes of the memory distortions were quite

comparable to those in the second session.

Anchoring effects. The data are again consistent with the view that participants used their self- generated feedback as an anchor for their recol- lections. Table 5 shows that, except for one judgement given in the second session*about the relevance of the current course*the effect of the anchors can still be considered substantial. The mean effect size for the feedback-induced biases in the recollections was d�/ 0.47 for the judgements given in the second session, and for those in the first session it was d�/ 0.51. This is smaller than the effects obtained in the second session (for the recollections from the first session), but it is still quite consistent over judgements and amounts to a middle-sized effect according to Cohen’s (1992) conventions.

Impact of implicit theories. Did participants rely on their implicit theories about change more in the third session than in the second? This time, the results for participants’ judgements about their probability of switching to another univer- sity are consistent with their implicit theories, when recalling both their judgements given in the second session (Figure 7a) and, at least in tendency, those given in the first session (Figure 7b). However, again, none of the other recollec- tions conforms to the implicit theory explanation (Figure 7).

Motivational influences. As for the data ob- tained in the first recollection, we again checked whether the amount of self-relevance of the five judgements covaried with the size of the correla- tions between ‘‘discrepancy’’ (feedback�original judgement) and ‘‘memory bias’’ (recollection�

Switch (37.5 %)

Grade (50 %)

(a) (b)

Interest (7.5 %)

Workload (42.5 %)

Relevance (15 %)

0.00

0.10

0.20

0.30

0.40

0.50

P ro

p o rt

io n o

f co

rr e ct

r e co

lle ct

io n s

Switch (38.5 %)

Grade (42.3 %)

Interest (26.9 %)

Workload (42.3 %)

Relevance (30.8 %)

0.00

0.10

0.20

0.30

0.40

0.50

P ro

p o rt

io n o

f co

rr e ct

r e co

lle ct

io n s

Figure 6. Proportions of correct recollections in session 3 for the judgements given in session 2 (a) and session 1 (b).

STUDENT MEMORIES 85

original judgement), and with the size of the

correlation between the status quo (i.e., feedback

at third session) and ‘‘correctness’’ (1�/ correct recollection, 0�/ incorrect recollection). Table 6 shows the results. It is evident that, again, the

correlations between ‘‘discrepancy’’ and ‘‘mem-

ory bias’’ were not systematically influenced by

the amount of self-relevance of the judgements.

Moreover, whereas the correlations between

‘‘status quo’’ and ‘‘correctness’’ for the judge-

ments given in session 1 resemble the pattern

found at the first recollection, those for the

judgements given in session 2 seem not to covary

with the self-relevance of the topics at all. Thus,

there seems to be minimal to no evidence for the

working of motivational influences in this second

recollection as far as the correlative analyses are

concerned.

Again, we also looked at the possible impact of ‘‘success’’ and ‘‘failure’’. For this second recollec- tion, this was only possible for the judgement on grades. However, for neither of the two original judgements in sessions 1 and 2 could the pattern shown in Figure 6 be replicated.

Does feedback decrease the proportion of correct recollections? In the current session, we omitted feedback for the judgement about work- load. If feedback tends to suppress spontaneously correct recollections, one would expect a higher proportion of correct memories (about the origi- nal judgement on expected workload) without than with feedback. The proper comparison here is between the recollection in the second session about the judgement in the first session, and the recollection in the third session about the judge- ment in the second session, because the interval

TABLE 5

The use of feedback as an anchor as indicated in the size of d, as found in the third session

Topic t df p Mean d d

Recollections of judgements given in the second session

Switch 5.02 22 .000 .497 1.05

Grade 2.52 23 .019 .244 .52

Interest 1.97 34 .057 .294 .33

Relevance �/.021 26 .984 �/.004 �/.004

Recollections of judgements given in the first session

Switch 1.91 16 .074 .294 .46

Grade 2.76 16 .014 .478 .67

Interest 1.75 17 .099 .376 .41

Relevance 2.13 17 .048 .312 .50

Results of one-sided one-sample t -tests (t , df , and p ) and effect sizes (mean d, and d ). All calculations were done without outliers, which were determined by constructing the respective box plots. The d s were calculated by dividing mean d by the standard deviation of d.

Switch Chi² = 0.10 p = 0.952

Grade Chi² = 9.08 p = 0.011

Interest Chi² = 33.79

p = 0.000

Relevance Chi² = 20.28

p = 0.000

0.00

0.20

0.40

0.60

0.80

1.00(a) (b)

E ff e ct

s iz

e w

Switch Chi² = 1.19 p = 0.551

Grade Chi² = 7.81 p = 0.020

Interest Chi² = 35.15

p = 0.000

Relevance Chi² = 53.01

p = 0.000

0.00

0.25

0.50

0.75

1.00

1.25

1.50

E ff e ct

s iz

e w

Figure 7. Fit between implicit-memory predictions and actual recollections, in the third session. Results are shown for the

judgements given in the second session (a) and the first session (b). Note that the smaller the chi-square and w, and the larger the p ,

the better the fit. The horizontal line indicates a small effect size of w�/ .1.

86 SEDLMEIER AND JAEGER

between recollections and judgements is equal in both cases. It turns out that, apparently, feedback did not influence correct recollections: the pro- portion of correct recollections remained stable*41.9% (see Figure 2) versus 42.5% (see Figure 6a)*whether feedback was provided or not. However, this cannot be taken as evidence against the hypothesis that anchors influence recollections. The finding indicates that feedback does not have detrimental effects on memories if their contents are still accessible. For non-acces- sible memories, however, the evidence from the other kinds of judgements clearly speaks for the influence of feedback as an anchor.

Do participants rely less on anchors over time? Learning effects over time can best be examined if conditions, such as materials or interval, do not change. In our study, we had exactly the same interval between the first and the second session and between the second and the third session. Apparently, participants’ memories did not im- prove on an absolute level over time, as evi- denced when the percentages of correct solutions are compared: Participants in the second session correctly remembered 32.3% of what they had written in the first session; in the third session, they remembered 30.5% of what they had written in the second session. However, the repetitive exposure to anchors might have prompted them not to rely as much on their self-generated anchors. This would be indicated in the data by more unsystematic error and fewer systematic biases when comparing the second six-month interval to the first. A comparison of Tables 4 and 5 shows that the data are indeed consistent with this view: The anchor effects (d), which express the amount of systematic error, are

smaller in the second six-months interval. Also

the results in Figure 8, which compares the anchor

effects for the recollections in the second session

(from the first session, 1�2 in Figure 8) with those of the third session (from the second session, 2�3 in Figure 8), are consistent with the assumption

that participants might have decreasingly relied

on self-generated feedback. However, how this

could have happened is still open to speculation.

TABLE 6

The impact of motivational influences on the recollection of judgements

‘‘Discrepancy’’ �‘‘memory bias’’ ‘‘Status quo’’ �‘‘correctness’’

Topic First session judgements Second session judgements First session judgements Second session judgements

Grade .32 .36 �/.18 .05 Workloada �/.24 .11 Switch .53 .62 �/.43 �/.02 Interest .38 .45 .23 .19

Relevance .16 .37 .16 .15

Correlations between ‘‘discrepancy’’ (feedback �original judgement) and ‘‘memory bias’’ (recollection �original judgement) and correlation between the status quo (i.e., feedback at third session) and ‘‘correctness’’ (1�/ correct recollection, 0�/ incorrect recollection) for judgements given in the first and the second session. Topics are rank ordered from top to bottom according to their

judged self-relevance. a The correlation between ‘‘discrepancy’’ and ‘‘memory bias’’ could not be calculated for workload because no feedback was

given in the third session.

M e a n “

a n ch

o r”

e ff e ct

( δ)

0

0.2

0.4

0.6

0.8

1.0

1.2

2–31–2

Switch

Grade

Interest

Relevance

Figure 8. Change in the impact of self-generated feedback over time (first six-months compared to second six-months).

Recollections of first-session judgements in the second session

(left) are compared with recollections of second-session

judgements in the third session (right).

STUDENT MEMORIES 87

Do recollections get worse with longer intervals? One might expect that recollections get worse the

longer the interval between the generation of the original memory contents and their recollection.

Here the results to be compared are the two

memory collections in the third session: the

recollections of the judgements in the second session (six-month interval) and those in the first

session (one-year interval). A comparison of the

total percentage of correctly recalled judgements

indicates that judgements did not get worse with the longer interval: Whereas 32.3% of the judge-

ments given in the first session were recalled

correctly in the second, this percentage even

improved slightly in the third session to 36.2%. Again, one might ask whether the impact of self-

generated feedback lessens with increasing inter-

val. This seems to be the case, as illustrated in

Figure 9. It appears that the proportion of nonsystematic error (e.g., due to forgetting) as

compared to systematic biases (due to anchoring

effects) increases with increasing interval. An-

other explanation for the non-difference in pro- portion of correct recollections in the second and

third sessions might be that the interval between

the first two sessions led to an asymptote in the forgetting curves. If this is so, then participants should show a high probability of remembering their first judgement correctly at the second recollection, if they also remembered it correctly at the first recollection. These conditional prob- abilities for the five topics are switch: .83; grade: .70; interest: .50; workload: .60; and relevance: .55. All these conditional probabilities are higher than the unconditional ones (compare Figure 6a), thus rendering some plausibility to the latter explanation.

Discussion

The results we obtained in the third session of our study corroborate those of the second session: We found about the same percentage, that is, about two-thirds erroneous recollections. The result for the judgement on workload (for which no feed- back was given in the third session) indicates that this rather small proportion of correct memories is not due to a detrimental impact of feedback. So feedback seems not to prevent a correct recollec- tion, but rather to influence the (re)construction of memory contents once the original memories can not be correctly recollected.

As in the second study, implicit theories do not seem to be a good explanation for what particip- ants actually did. Even though the results for one judgement*about the probability of switching to another university*are now consistent with the assumption that participants used their implicit theories about change, the results for this topic are even more consistent with the assumption of a substantial impact of participants’ self-generated anchors. The results at the second recollection show even less support for the assumption that recollections are modified by motivational influ- ences than the results from the first recollection. As in the second session, the impact of the anchor feedback seems to be the most plausible explana- tion of the results, overall. However, despite the still substantial anchoring effects, the results in- dicate that participants might have been influ- enced less by their self-generated feedback than was the case in the second session. The decreased influence of the anchors might also have been, at least in part, responsible for why the increased interval (a comparison of the results in sessions 2 and 3 for remembering the judgements from the first session) did not lead to stronger biases. However, it is not totally clear why the anchors’

M e a n “

a n ch

o r”

e ff e ct

( δ)

0

0.2

0.4

0.6

0.8

1.0

1.2

1–31–2

Switch

Grade

Interest

Relevance

Figure 9. Change in the impact of self-generated feedback dependent on length of interval. Recollections after a six-

month interval (1 �2) are compared with recollections after a 12-month interval (1 �3).

88 SEDLMEIER AND JAEGER

influence in the third session was smaller. One reason could indeed be that participants became aware of the possible influence of anchors and ignored them on purpose. Another possible ex- planation involves the difference between the setting for the third and the second session. Whereas the first two courses dealt with methods in a theoretical way, the third course included real experimentation, and was also taught by different personnel. So the answers to the judgements given there might not have been perceived to be as relevant for the original memories as the answers given in session 2. This, again, would involve the purposeful neglect of the self-generated feedback. This issue definitely needs further exploration.

GENERAL DISCUSSION

We showed that the proportion of erroneous recollections can be quite high*about two thirds*even in naturalistic memory tasks such as students’ judgements about study-related is- sues. In our study, this high percentage may in part have been due to the relatively long interval of six to 12 months between the formation of the memories and their recollection. However, inter- vals of this duration are quite common in every- day life, and therefore substantial memory distortions even for self-relevant memory con- tents might not be so unusual. The results we found for the wrong recollections are most consistent with the hypothesis that participants systematically used self-generated feedback and applied some kind of judgemental anchoring. Implicit theories they held about stability or change in respect to the examined topics did not seem to play a strong role. There was, however, some indication that motivational factors might play a role in everyday life memories in that*at least after six-months*the status quo was corre- lated with the correctness of the recollection of more relevant judgements, and recollections were tendentially more distorted after failures than after successes. Obviously, none of the kinds of judgements we had students make were very highly relevant, and therefore one might obtain more support for the role of motivational influ- ences if judgements that are more heterogeneous concerning self-relevance are used.

Interestingly, the influence of self-generated feedback was stronger if it did not directly pertain to the original memory contents but rather in- volved predictions for related events. This is,

however, consistent with the view that participants used judgemental anchoring and accessed gist memories most similar to the information con- tained in the anchor (e.g., Brainerd & Reyna, 2002). After all, for the two topics used for the comparisons*grades and workload*the original judgement was of the same kind as the feedback judgement: both were predictions. In the third session, we also found results that are consistent with the assumption that participants had mana- ged to disregard the influence of anchors on their recollections. However, it is not clear whether the diminished influence of the anchors was due to the application of metacognitive skills (i.e., the in- creasing ability to disregard feedback) or was simply a consequence of participants perceiving the self-generated anchors in the third session as less relevant than those in the second session.

Methodological considerations

In psychological research there is always a trade- off between internal and external validity; that is, the extent to which one can be sure about the causal conclusions drawn, and the extent to which the results can be generalised to everyday life. To date, most studies on the hindsight bias, which usually have a high internal validity, have used items that were not particularly self-relevant, whereas studies that used very self-relevant judgements have had to make concessions in the rigour of design (e.g., Mark & Mellor, 1991). We tried to make a compromise that ensured a high degree of both internal and external validity. We placed the emphasis on more naturalistic tasks that were still controllable to a certain extent, and compared different candidate explanations for the results. This procedure assumes an exhaustive collection of the relevant theoretical approaches. Although we cannot ignore the fact that other theoretical approaches might have yielded a better fit, to the best of our knowledge there is no additional major alternative approach pro- posed in the literature.

One might argue that the implicit theory approach was at a disadvantage in our study because the relevant predictions were generated between subjects, whereas the influence of an- chors was tested within subjects. This may be true. However, generating predictions for participants’ implicit theories within subjects might have strongly influenced their subsequent recollec- tions, and therefore made them virtually unin-

STUDENT MEMORIES 89

terpretable. But this issue*how strongly judge- ments are influenced by having participants reveal their implicit theories about change per- taining to those judgements*is in principle open to empirical scrutiny, and should definitely re- ceive further attention. However, at least for the kinds of judgements we used, it seems rather un- likely that the fit would improve dramatically if predictions were generated within subjects, for two reasons. First, the patterns we found for participants’ implicit theories were quite stable over different cohorts of students, and second, results deviate so strongly from the predictions that an excellent fit seems unlikely even if predictions could be substantially improved.

Memory distortions and associative learning

One of the most interesting results we obtained was that indirect feedback (predictions about similar issues) led to stronger memory distortions than direct feedback (recollections of the out- comes of the original predictions). One possible explanation for this finding could be that partici- pants processed the two kinds of information differently. Whereas they might have deliberately juxtaposed the recollection about the outcome and the original judgement, the process of inte- grating the new prediction into existing memories about the older prediction might have proceeded in a more implicit way. The latter would, in our view, be consistent with predictions of fuzzy-trace theory (Brainerd & Reyna, 2002). There might, however, be another suitable theoretical ap- proach starting from research on frequency pro- cessing and using computational models of associative learning and memory. Research on frequency estimates has shown that variation in the way information is processed can have a substantial impact on judgements, with more pronounced systematic effects in the case of implicit information processing (e.g., Haberstroh & Betsch, 2002; Naveh-Benjamin & Jonides, 1986). Especially if judgements are rather similar, they seem prone to be highly associated in memory, which in turn might produce ‘‘average responses’’. There are already models that can mimic certain kinds of judgemental errors (Dougherty & Franko-Watkins, 2002; Sedlmeier, 2002, 2006). Until now, these models have only produced ‘‘gist-like’’ and not ‘‘verbatim-like’’ responses. However, this is not a major limitation

as, for instance, it has already been shown by McClelland and Rumelhart (1986) that represen- tations of specific, repeated exemplars can coexist in the same set of connections with knowledge of gist-like memory. If associative strength between feedback and original information is predictive of the strength of memory distortions, this could open up an interesting new line of research. The use of associative computational models might be helpful in making predictions more precise, and therefore leading to stronger tests of theories on memory distortions.

Possible extensions of the current approach

The study reported here was an attempt to look at memory distortions more from an everyday perspective than is usually done in some research on memory distortions. That is, instead of putting the emphasis on experimental control (e.g., pro- viding identical feedback to all participants, using tasks for which ‘‘correct solutions’’ exist, and concentrating on a single experimental para- digm), we had participants produce self-gener- ated and self-relevant judgements and feedback, and used the recollections to contrast different possible explanations. Our general topic was students’ study-related judgements, and our re- sults are consistent with the assumption that the memory biases we found were strongly deter- mined by judgemental anchoring and much less so by the impact of implicit theories and motiva- tional influences. However, we have pointed out some shortcomings concerning optimal tests of the latter two explanations. Even the issue of indirect versus direct feedback was only partly covered in our design. It would be interesting to examine a large variety of indirect feedback, as indicated in our introductory example. Moreover, our set-up was not so well suited to testing motivational mechanisms. Motivational explana- tions certainly deserve more attention, and one might begin to examine the effect of success and failure with items that differ to a larger extent in their relevance for one’s self-esteem. Here, one probably has to make a trade-off between amount of self-relevance and control. When dealing with highly self-relevant issues, such as aspects of partnership or traumatic experiences, one prob- ably needs to begin with single case studies. All in all, starting with everyday-life memories and exploring potential candidate explanations for

90 SEDLMEIER AND JAEGER

the resulting recollections, depending on the nature of the specific situation, seems to be a worthwhile approach that can successfully sup- plement mainstream research on memory biases.

Manuscript received 5 October 2005

Manuscript accepted 5 October 2006

First published online 6 December 2006

REFERENCES

Ayers, M. S., & Reder, L. M. (1998). A theoretical review of the misinformation effect: Predictions from an activation-based memory model. Psycho- nomic Bulletin & Review, 5 , 1�21.

Bahrick, H. P., Hall, L. K., & Berger, S. A. (1996). Accuracy and distortion in memory for high school grades. Psychological Science , 7 , 265�271.

Belli, R. R., Windschitl, P. D., McCarthy, T. T., & Winfrey, S. E. (1992). Detecting memory impair- ment with a modified test procedure: Manipulating retention interval with centrally presented event times. Journal of Experimental Psychology: Learn- ing, Memory and Cognition , 18 , 356�367.

Blank, H. (1998). Memory states and memory tasks: An integrative framework for eyewitness memory and suggestibility. Memory, 6 , 481�529.

Blank, H., Fischer, V., & Erdfelder, E. (2003). Hind- sight bias in political elections. Memory, 11 , 491�504.

Brainerd, C. J., & Reyna, V. F. (2002). Fuzzy-trace theory and false memory. Current Directions in Psychological Science , 11 , 164�169.

Chandler, C. C. (1993). Accessing related events increases retroactive interference in a matching recognition test. Journal of Experimental Psychol- ogy: Learning, Memory & Cognition , 19 , 967�974.

Chapman, G. B., & Johnson, E. J. (1999). Anchoring, activation and the construction of value. Organiza- tional Behavior and Human Decision Processes, 79 , 115�153.

Chapman, G. B., & Johnson, E. J. (2002). Incorporating the irrelevant: Anchors in judgements of belief and value. In T. Gilovich, D. Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive judgement (pp. 120�138). Cambridge, UK: Cambridge University Press.

Christensen-Szalanski, J. J. J., & Wilham, C. F. (1991). The hindsight bias: A meta-analysis. Organizational Behavior and Human Decision Processes, 48 , 147�168.

Cleveland, W. S. (1985). The elements of graphing data . Pacific Grove, CA: Wadsworth & Brooks/Cole.

Cohen, J. (1992). A power primer. Psychological Bulletin , 112 , 155�159.

Conway, M., & Ross, M. (1984). Getting what you want by revising what you had. Journal of Personality and Social Psychology, 47 , 738�748.

Dougherty, M. R. P., & Franco-Watkins, A. M. (2002). A memory models approach to frequency and probability judgement: Applications of Minerva 2 and Minerva DM (pp. 121 �136). In P. Sedlmeier &

T. Betsch, (Eds.), Etc.: Frequency processing and cognition . Oxford, UK: Oxford University Press.

Fischhoff, B. (1975). Hindsight "/ foresight: The effect of outcome knowledge on judgement under uncer- tainty. Journal of Experimental Psychology: Human Perception and Performance , 1 , 288�299.

Guilbault, R. L., Bryant, F. B., Brockway, J. H., & Posavac, E. J. (2004). A meta-analysis of research on hindsight bias. Basic and Applied Social Psychology, 26 , 103�117.

Haberstroh, S., & Betsch, T. (2002). Online strategies versus memory-based strategies in frequency esti- mation (pp. 205 �220). In P. Sedlmeier & T. Betsch (Eds.), Etc.: Frequency processing and cognition . Oxford, UK: Oxford University Press.

Haslam, N., & Jayasinghe, N. (1995). Negative affect and hindsight bias. Journal of Behavioral Decision Making , 8 , 127�135.

Hawkins, S. A., & Hastie, R. (1990). Hindsight: Biased judgements of past events after the outcomes are known. Psychological Bulletin , 107 , 311�327.

Hell, W., Gigerenzer, G., Gauggel, S., Mall, M., & Müller, M. (1988). Hindsight-bias: An interaction of automatic and motivational factors? Memory and Cognition , 16 , 533�538.

Hirt, E. R. (1990). Do I see only what I expect? Evidence for an expectancy-guided retrieval model. Journal of Personality and Social Psychology, 58 , 937�951.

Hirt, E. R., Erickson, G. A., & McDonald, H. E. (1993). Role of expectancy timing and outcome consistency in expectancy-guided retrieval. Journal of Person- ality and Social Psychology, 65 , 640�656.

Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin , 114 , 3� 28.

Karney, B. R., & Frye, N. E. (2002). ‘‘But we’ve been getting better lately’’: Comparing prospective and retrospective views of relationship development. Journal of Personality and Social Psychology, 82 , 222�238.

Loftus, E. F. (1975). Leading questions and the eye- witness report. Cognitive Psychology, 7 , 560�572.

Loftus, E. F., Hoffman, H. G., & Wagenaar, W. A. (1992). This misinformation effect: Transformations in memory induced by postevent information. In M. L. Howe, C. J. Brainerd, & V. F. Reyna (Eds.), Development of long-term retention (pp. 159�183). New York: Springer.

Mark, M. M., & Mellor, S. (1991). Effect of self- relevance of an event on hindsight bias: The fore- seeability of a layoff. Journal of Applied Psychology, 76 , 569�577.

McClelland, J. L., & Rumelhart, D. E. (1986). A distributed model of human learning and memory. In J. L. McClelland & D. E. Rumelhart (Eds.), Parallel distributed processing (Vol. 2, pp. 170�215). Cambridge, MA: MIT Press.

McCloskey, M., & Zaragoza, M. (1985). Misleading post-event information and memory for events: Arguments and evidence against memory impair- ment hypotheses. Journal of Experimental Psychol- ogy: General , 114 , 1 �16.

STUDENT MEMORIES 91

McDonald, H. E., & Hirt, E. R. (1997). When expectancy meets desire: Motivational effects in reconstructive memory. Journal of Personality and Social Psychology, 72 , 5 �23.

Mussweiler, T., & Strack, F. (1999a). Comparing is believing: A selective accessibility model of judge- mental anchoring. In W. Stroebe & M. Hewstone (Eds.), European review of social psychology (Vol. 10, pp. 135�167). Chichester, UK: Wiley.

Mussweiler, T., & Strack, F. (1999b). Hypothesis- consistent testing and semantic priming in the anchoring paradigm: A selective accessibility model. Journal of Experimental Social Psychology, 35 , 136�164.

Naveh-Benjamin, M., & Jonides, J. (1986). On the automaticity of frequency coding: Effects of com- peting task load, encoding strategy, and intention. Journal of Experimental Psychology: Learning, Memory, and Cognition , 12 , 378�386.

Renner, B. (2003). Hindsight bias after receiving self- relevant health risk information: A motivational perspective. Memory, 11 , 455�472.

Reyna, V. F., & Brainerd, C. J. (1995). Fuzzy-trace theory: An interim synthesis. Learning and Indivi- dual Differences, 7 , 1 �75.

Reyna, V. F., Holliday, R., & Marche, T. (2002). Explaining the development of false memories. Developmental Review, 22 , 436�489.

Reyna, V. F., & Lloyd, F. (1997). Theories of false memory in children and adults. Learning and Individual Differences, 9 , 95�123.

Ross, M. (1989). Relation of implicit theories to the construction of personal histories. Psychological Review, 96 , 341�357.

Safer, M. A., Bonanno, G. A., & Field, N. P. (2001). ‘‘It was never that bad’’: Biased recall of grief and long- term adjustment to the death of a spouse. Memory, 9 , 195�204.

Safer, M. A., & Keuler, D. J. (2002). Individual differences in misremembering pre-psychotherapy distress: Personality and memory distortion. Emotion , 2 , 162�178.

Schwarz, S., & Stahlberg, D. (2003). Strength of hind- sight bias as a consequence of meta-cognitions. Memory, 11 , 395�410.

Sedlmeier, P. (2002). Associative learning and fre- quency judgements: The PASS model. In P. Sedlme- ier & T. Betsch (Eds.), Etc.: Frequency processing and cognition (pp. 137�152). Oxford, UK: Oxford University Press.

Sedlmeier, P. (2006). Intuitive judgements about sam- ple size. In K. Fiedler & P. Juslin (Eds.), In the beginning there is a sample: Information sam- pling as a key to understand adaptive cognition (pp. 53 �71). Cambridge, UK: Cambridge University Press.

Stahlberg, D., Eller, F., Romahn, A., & Frey, D. (1993). Der Knew-it-all-along-Effekt in Urteilssitua- tionen von hoher und geringer Selbstwertrelevanz. Zeitschrift für Sozialpsychologie , 24 , 94�102.

Story, A. L. (1998). Self-esteem and memory for favorable and unfavorable personality feedback. Personality and Social Psychology Bulletin , 24 , 51 �64.

Tukey, J. W. (1977). Exploratory data analysis. Reading, MA: Addison-Wesley.

Tversky, A., & Kahneman, D. (1974). Judgement under uncertainty: Heuristics and biases. Science , 185 , 1124�1131.

Tykocinski, O. E. (2001). I never had a chance: Using hindsight tactics to mitigate disappointments. Per- sonality and Social Psychology Bulletin , 27 , 376�382.

Vaillant, C. O., & Vaillant, G. E. (1993). Is the U-curve of marital satisfaction an illusion? A 40-year study of marriage. Journal of Marriage and Family, 55 , 230�239.

92 SEDLMEIER AND JAEGER