See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/264785679

How We Remember the Emotional Intensity of Past Musical Experiences

Article in Frontiers in Psychology · August 2014

DOI: 10.3389/fpsyg.2014.00911

CITATIONS READS

33 247

3 authors, including:

Thomas Schäfer
MSB Medical School Berlin
62 PUBLICATIONS 1,902 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Examining Listening strategies and reward mechanisms from music in Indian population View project

All content following this page was uploaded by Thomas Schäfer on 16 August 2014.

The user has requested enhancement of the downloaded file.

 ORIGINAL RESEARCH ARTICLE
published: 15 August 2014
doi: 10.3389/fpsyg.2014.00911

How we remember the emotional intensity of past musical

experiences
Thomas Schäfer*, Doreen Zimmermann and Peter Sedlmeier
Department of Psychology, Chemnitz University of Technology, Chemnitz, Germany

Edited by: Listening to music usually elicits emotions that can vary considerably in their intensity
Andrew Kemp, Universidade de São over the course of listening. Yet, after listening to a piece of music, people are easily able
Paulo, Brazil
to evaluate the music’s overall emotional intensity. There are two different hypotheses
Reviewed by:
about how affective experiences are temporally processed and integrated: (1) all moments’
Andrew Kemp, Universidade de São
Paulo, Brazil intensities are integrated, resulting in an averaged value; (2) the overall evaluation is
Frederick Streeter Barrett, Johns built from specific single moments, such as the moments of highest emotional intensity
Hopkins School of Medicine, USA (peaks), the end, or a combination of these. Here we investigated what listeners do when
Ruth Wells, University of Sydney,
Australia
building an overall evaluation of a musical experience. Participants listened to unknown
songs and provided moment-to-moment ratings of experienced intensity of emotions.
*Correspondence:
Thomas Schäfer, Department of Subsequently, they evaluated the overall emotional intensity of each song. Results indicate
Psychology, Chemnitz University of that participants’ evaluations were predominantly influenced by their average impression
Technology, 09107 Chemnitz, but that, in addition, the peaks and end emotional intensities contributed substantially.
Germany
e-mail: thomas.schaefer@
These results indicate that both types of processes play a role: All moments are integrated
psychologie.tu-chemnitz.de into an averaged value but single moments might be assigned a higher value in the
calculation of this average.

Keywords: music, intensity, peak–end, temporal integration, duration neglect

INTRODUCTION In the following, we first discuss the theoretical background

Music is used in everyday life for numerous purposes, one of and empirical findings regarding the subjectively felt emotional
the most important of which is the regulation of moods and intensity of affective experiences in general. Afterward, we do the
emotions (e.g., Sloboda et al., 2001; Juslin and Laukka, 2004; same for studies that used music as the affective stimulus. Finally,
Saarikallio and Erkkilä, 2007; Schäfer et al., 2013). Hence, it we present the results of our own study.
comes as no surprise that music’s potential to regulate moods Note that we use the term musical experience, which does
and emotions has an influence on how pleasant people perceive a not include activities such as playing an instrument, singing, or
piece of music to be (see Sedlmeier and Schäfer, 2013). Practically, remembering or reading about music, but only the specific act of
this means that people will actively choose to buy and/or lis- focused listening. Thus, the term is comparable to what Behne
ten to pieces of music they know have emotion-regulating or (1997, p. 143) called Musikerleben: “the sum of psychic processes
mood-regulating potential. But how do listeners remember the which accompany the experience of music in situations when
emotional impact of music they heard in the past? Do they music is in the focus of interest: When a person is not only
remember single extraordinary moments or elements of pieces hearing, but listening to and appreciating music.” The feeling of
or is there something like an averaged remembered value? There emotions is one of the most important parameters of musical
has not been much empirical research on these questions. In experiences. The absolute extent of emotions felt during listen-
the present article, we present a study on the evaluation of past ing to music (regardless of whether they are positive or negative)
musical experiences. When listening to a piece of music there is defined as emotional intensity.
is a time course of emotional intensity. On the one hand, emo-
tional intensity can change from moment to moment (reach EVALUATION OF AFFECTIVE EXPERIENCES IN GENERAL
peaks and troughs) and does not need to be constant over the THEORETICAL BACKGROUND
time course of the musical experience. On the other hand, lis- Affective experiences can be of a short- or long-term nature.
teners are also able to evaluate the overall emotional intensity Regarding short-term experiences, the momentary affect can be
of the whole musical experience. Our study was intended to easily evaluated at the instant it unfolds. On the other hand—and
analyze how listeners take this time course of emotional inten- this is what the present study is about—there are experiences of
sity into account to arrive at a subsequent overall evaluation longer duration and varying affective intensity. Forming an overall
of emotional intensity. Knowledge of this process should be evaluation of these experiences requires generating a global over-
of interest to performers and composers. It could help them all value of affective intensity (Varey and Kahneman, 1992). The
arrange pieces of music that leave an emotionally intense overall evaluation of momentary affect happens with little reflection. The
memory. retrospective evaluation of past affective experiences, however,

www.frontiersin.org August 2014 | Volume 5 | Article 911 | 1

Schäfer et al. Remembered emotional intensity

requires recall and overall assessment that involves integrating but is simply not taken into account. Fredrickson (2000) has rea-
all—or only some—moments of that experience (Fredrickson, soned why people should build their overall evaluation from the
2000). There are two mental processes involved in retrospective peak and the end of an experience: These two moments usually
evaluation: memory and evaluation of past affective experiences are special carriers of personally relevant meaning. The peak indi-
(Kahneman et al., 1993). To understand the evaluation of emo- cates how enjoyable or how threatening an experience can get.
tional intensity of past musical experiences it is useful to explore The end conveys the information that the experience can be sur-
how people retrospectively remember and evaluate past affective vived. The peak–end rule is considered being used as a simple
experiences of long duration and varying affective intensity in heuristic, which can be very useful even though it might also lead
general. to mistakes.
There are two competing theories of how the storage of past
affective experiences in memory and the process of evaluation EMPIRICAL FINDINGS
occur (Fredrickson and Kahneman, 1993). The first suggestion— The questions of how affective experiences are represented in
sometimes illustrated by the metaphor of film—is that all details memory and how people evaluate them have been examined in
of an experience are comprehensively represented in memory. numerous studies. There is a common method used in these
Affective intensity is stored as a function of time, while time itself studies: Participants’ actual moment-to-moment ratings of affec-
might or might not be stored in memory, as well. Consequently, tive intensity are continuously measured during the experience.
retrospective evaluation of overall affective intensity is based on Subsequently, the retrospective evaluation of the overall affective
the temporal integration of the affective intensities of a certain intensity is assessed by participants’ ratings. Then, the relation-
number of single “moments.” This is why the model is referred ship between the recorded time course of the affective experi-
to as the temporal integration model. From this point of view, the ence and the retrospective overall evaluation is analyzed using
overall evaluation of experienced affect should strongly depend correlation or regression methods.
on the relative duration of specific strengths of affect. That is, A considerable number of studies have supported the idea
the retrospective overall evaluation of an experience is best pre- of the peak–end rule (for an overview, see Kahneman, 2011).
dicted by the average of all single moments because the longer In a study on pleasant and aversive film clips, Fredrickson and
the affect is relatively strong, the larger the average becomes over Kahneman (1993) demonstrated that peak affect and end affect
time, and the longer the affect is relatively weak, the smaller the had a remarkable effect on participants’ global retrospective rat-
average becomes over time1. ings of each film’s affective intensity. Moreover, the results were
The second theory—sometimes illustrated by the metaphor not influenced by the specific time delay (which could be shorter
of a collection of snapshots—is that the whole experience is or longer) between the end of the actual experience and the
represented in memory fragmentally. Only intensities of specific subsequent retrospective evaluation, indicating that the global
moments are stored, and time is not represented in memory at all. ratings were stable over time. In addition, sessions with and
Consequently, overall retrospective evaluation of the whole affec- without moment-to-moment ratings resulted in similar overall
tive experience is based on (the average of) just a few moments’ evaluations. This finding was an important justification for the
affective intensities. From this point of view, duration of the research method applied, because it demonstrated that moment-
affective experience should not have any influence on overall to-moment ratings did not distort salience or memories of spe-
evaluation, which has been called duration neglect. cific moments. In a study on pain induced by immersing a hand
What might these moments be that listeners remember more in cold ice water (Kahneman et al., 1993), participants had to
than other moments? There are three specific moments of a musi- endure a short trial (60 s of 14◦ C) and a long trial (60 s of 14◦ C +
cal experience that might be of specific importance: the onset 30 s of 15◦ C). The long trial was objectively more painful because
moment, the moment of highest emotional intensity (referred it included a greater amount of total pain, but it had a better
to as the peak), and the end of the experience. In addition, any end than the short trial. Surprisingly, participants evaluated the
combination of these three moments could play a role in the longer trial as less painful, even though they were able to judge
overall evaluation. Specifically, many scholars have considered the durations correctly. This also led to the conclusion that people
the combination of the peak moment and the end moment very put particularly high value on the end of an affective experience.
important—a conjecture that has become known as the peak–end Also, in a study on pain induced by colonoscopy and lithotripsy,
rule (see Kahneman, 1999, 2011): When people evaluate a past Redelmeier and Kahneman (1996) demonstrated the importance
experience they might pay attention above all to two things, how of the end of the painful experience. Stone et al. (2000) exam-
it felt at the peak and at the end; other information (e.g., net pleas- ined rheumatoid arthritis and found that peak–end was a better
antness or unpleasantness, duration of the experience) is not lost predictor for the evaluation of overall pain than the global aver-
age of all single moments. Neither peak nor end alone were as
1 We should mention that a special form of the temporal integration model
powerful predictors as their average. Schreiber and Kahneman
claims that the duration of the experience itself matters. That is, any affect, (2000) found evidence for the peak–end rule in a study on aver-
be it strong or weak, always simply increases the “proportion” of affect felt— sive sounds, as did Langer et al. (2005) in a study on payment
as long as the affect does not change its valence. This leads to the conclusion
that the longer an experience lasts, the greater the remembered pleasure or
sequences and Do et al. (2008) in a study on material gains.
displeasure should be, which has been referred to as temporal monotonic- Yet, there are a number of studies that did not confirm
ity. However, this conjecture does not appear to be very plausible and it has the peak–end rule. In a study on pleasant advertisements,
received no empirical support. Baumgartner et al. (1997) found that peak and end, as separate

Frontiers in Psychology | Emotion Science August 2014 | Volume 5 | Article 911 | 2

Schäfer et al. Remembered emotional intensity

factors of the experience, were better predictors of the overall eval- moment-to-moment measurement is that the listener is able to
uation than was their average. When investigating the enjoyment track the focus of attention on the music without speaking or
of meals, Rode et al. (2007) did not find either peak or end to writing.
be more important than any other element of the time course.
Robinson et al. (2011) found that only the peak of the moment- EMPIRICAL FINDINGS
to-moment enjoyment predicted the overall enjoyment of meals. The relationship of actual moment-to-moment musical experi-
Kemp et al. (2008) studied affective autobiographical events and ence and the overall evaluation of those experiences has not been
found that participants did not remember the peaks and troughs well examined so far. Evidence was provided first by Sloboda and
of the intensity of happiness during their holidays better than Lehmann (2001), whose study did not focus on felt emotions but
other moments. Cojuharenco and Ryvkin (2008) demonstrated on emotions participants perceived in the music. Nevertheless,
that average and peak–end are comparable in terms of their role in they found that the average of all single moments of musical expe-
predicting the overall evaluation of experiences and that neither rience correlated with the subsequent global rating by r = 0.50.
showed an advantage over the other. Duke and Colprit (2001) investigated the magnitude of musical
In sum, studies with nonmusic stimuli have left an unclear moment-to-moment intensity and found that the average of all
picture. Although most have revealed that the overall evaluation ratings is different from the overall post-hoc rating. However, they
of affective intensity can be well predicted by the average of the did not calculate the covariation of these measures.
most intense moment and the moment at the end of the experi- Rozin et al. (2004) investigated how remembered overall musi-
ence, there are also studies that did not support the validity of cal affect is derived from moment-to-moment musical affect.
this peak–end rule. Notably, almost all of the data supporting Their participants listened to various music selections of differ-
the peak–end rule come from negative experiences, so it is ques- ent durations (i.e., each song had a different number of sin-
tionable if those results can be transferred to musical experiences. gle “moments”). After measuring moment-to-moment affective
Specifically, it is not clear if the consolidation of pain experiences intensity ratings during each song, the authors measured remem-
is comparable to the consolidation of musical experiences regard- bered overall affective intensity of each musical selection. Based
ing, for instance, habituation processes or psychological coping on several predictors (average of all single moments’ intensi-
mechanisms. It is hard to tell from all the mentioned studies if the ties, sum of all single moments’ intensities, onset intensity, offset
remembrance and overall evaluation of a past experience rely on intensity, minimum intensity, peak intensity, and sum of peak
integrated moments, such as the average, or on distinct moments, and offset intensities), they found that remembered intensity of
such as peak or end. We now describe research that incorporated affect was most highly correlated with peak (r = 0.82), peak–end
music as a stimulus. (r = 0.81), and average (r = 0.80). However, the authors did not
run a regression analysis to identify which of the potential param-
EVALUATION OF AFFECTIVE EXPERIENCES WITH MUSIC eters accounted for a significant proportion of variance of the
THEORETICAL BACKGROUND overall evaluation. Nonetheless, they concluded that their data
Listening to music is characterized by varying moment-to- did not support the peak–end rule because peak–end was most
moment emotional intensity. The characteristics of an affective highly correlated with the overall judgment for only 3 of the
musical experience and the subsequent retrospective evaluation 20 participants. Not least, they found a slope effect: large, pos-
of its overall emotional intensity are consistent with the charac- itive differences in the emotional intensity between consecutive
teristics of the above-defined general process of remembering and moments were also a reliable predictor of the remembered overall
evaluating past affective experiences of long duration and varying intensity.
affective intensity. Hence, the theoretical approaches to explaining The results of Rozin et al. (2004) provide a valuable piece of
this general process have been utilized to investigate the spe- evidence of how listeners generate an overall evaluation of the
cial case of evaluating the emotional intensity of past musical emotional intensity of musical experiences. However, there are a
experiences. number of concerns about potential methodological limitations
Typically, the intensity of music-induced affect has been in the Rozin et al. (2004) study, particularly the choice of stim-
recorded with the use of a dial, slider, or pressure-sensitive but- uli and measurement, which we addressed in the present study.
ton. Data are recorded by having participants manipulate the (1) The authors used songs that were known to the participants
device according to the intensity of the emotions they are feel- as well as songs that were unknown. Regardless of whether a song
ing. Madsen (1990; see also Madsen et al., 1993) justified the was known or unknown, participants always listened to the music
use of such methods instead of asking people about their sub- one time to become familiar with the song and a second time for
jective experience: Children, handicapped people, and untrained measurement. It is questionable if this is an adequate procedure
musicians in general often simply do not have an ability for for achieving comparable familiarity with known and unknown
high-level verbal abstraction and find it difficult to express musi- songs. It may be more suitable to use only unknown music to
cal changes they hear or feel. Moreover, people might not be ensure a controlled design and to explore cognitive processes
able to verbally document their musical experience while actu- based on “first impressions.” (2) Rozin et al. (2004) provided
ally listening. The act of verbally reporting one’s own responses their participants with a fixed order of songs, which may lead
while listening may interfere with the actual experience, and to order effects as a result of participant fatigue and variabil-
the experience itself may cease or stop quickly when the focus ity of motivation. (3) They used just short extracts of the songs
of attention is drawn away from it. Hence, the advantage of with limited time frames of around 40 s, which may not represent

www.frontiersin.org August 2014 | Volume 5 | Article 911 | 3

Schäfer et al. Remembered emotional intensity

naturalistic listening and may hinder valid generalizations. (4) METHODS

The authors did not perform a regression analysis or calculate PARTICIPANTS
partial correlations but reported only the first-order correlations Participants (N = 54) were psychology students, 44 (81.5%)
between the overall judgment and the parameters that potentially female, 10 (18.5%) male. They ranged in age from 18 to 35 years
influence it, for each participant. Therefore, the specific propor- (M = 22.3 years, SD = 3.2). Nineteen (35%) were involved in
tions of variance explained by the most important parameters are some kind of musical activity (singer, choir, band, orchestra, etc.)
not known. Examining the specific impact of potential parame- and 35 (65%) were not. Music was an important part of life for
ters (while controlling for the impact of others) would allow for all of them (Min = 4, Max = 9, M = 7.52, SD = 1.37, on a scale
determining the relative importance of the potential strategies for from 1—not important at all, to 9—very important) and they var-
extracting an overall emotional evaluation from the time course ied in self-rated musicality (Min = 1, Max = 9, M = 5.48, SD =
of moment-to-moment evaluations. (5) The slope effect Rozin 2.23, on a scale from 1—not musical at all, to 9—very musical).
et al. (2004) identified in their Discussion suggests that the vari- The students received course credits for their participation.
ation of the time course of the moment-to-moment experience
might be an additional parameter that influences the overall eval-
ETHICAL APPROVAL
uation and therefore should be incorporated in the calculations.
The study was performed in accordance with relevant insti-
(6) The authors followed a common procedure when calculating
tutional and national guidelines and regulations (Chemnitz
peak–end values, that is, building the average of the peak value
University of Technology, 2002; Deutsche Gesellschaft für
(the moment of highest intensity) and the end value. However,
Psychologie [German Psychological Society], 2005). Informed
consistent with the central claim of duration neglect theory, one
consent was obtained from all participants. Anonymity of par-
might expect not a single peak but the average impression of all
ticipants and confidentiality of their data were ensured.
moments that stand out (multiple peaks) as well as the number
of these moments (number of multiple peaks) to determine the
overall evaluation. STIMULI
Participants listened to a selection of 11 complete songs of dif-
AIM OF THE PRESENT STUDY ferent durations and genres (pop, rock, instrumental rock, hip
By way of summary, the results of studies about the remembrance hop, electro, jazz, classical, emo, reggae, Latino). The songs were
and evaluation of the emotional intensity of past musical expe- thought to be unknown to them. As individual musical experi-
riences are as inconclusive as those from the nonmusic studies ences are very subjective, can clearly differ, and were not evident
discussed above. From these lines of research, it is still hard to before the end of the study, the choice of songs was based on
tell if retrospective overall affect is based on an integration of the authors’ own and two other raters’ subjective judgments.
the whole experience, on just specific single moments, or on the Each song had to fulfill the following criteria: During the song,
average of such moments. In the present study, we attempted to elicited emotional intensity should not be constant but should
answer this question while circumventing the limitations of past reach peaks and troughs in order to maximize the variance within
research we have addressed above. We took into account multi- a song and thus between the different parameters of a song (e.g.,
ple parameters of the temporal profile of emotional intensity of beginning, peak, end, peak–end, average). Moreover, different
musical experiences: beginning, peak, average of multiple peaks, songs were chosen to elicit different levels of emotional intensity;
number of multiple peaks, end, peak–end, multiple peaks–end, that is, some songs were intended to be generally more emotion-
sum, average, and variation (see below for the calculation of these ally intense than others in order to maximize the variance between
parameters). Since music listening is an experience that unfolds the songs. Songs, together with their performers and lengths, are
over time, we argue that listeners would process an evaluation listed in Table 1.
of the emotional intensity by continuously updating their felt
moment-to-moment affect. That is, we predict that listeners aver- APPARATUS
age their experiences over time and use the averaged value at the The laboratory room was equipped with six tables, each with a
end of the experience as an overall evaluation. There may or may comfortable chair, a computer with 17-in monitor, optical mouse,
not be moments such as the peak or the end of the experience and Sennheiser HL 270 stereo headphones. Up to six partici-
that are weighted more strongly in the process of averaging. As pants at a time could be seated. They were separated from each
we cannot draw a more specific hypothesis about this process of other by wooden screens. Lights were dimmed. An “emoslide”
weighting based on previous research, we treat this issue as an java program, which was designed for the study, played the songs
explorative question. Although it is plausible that the remem- in random order and simultaneously measured the moment-
brance of only some specific moments that might carry important to-moment ratings of subjectively experienced intensity of felt
information about an experience (as do the peak[s] and the end) emotions. Participants had to move a digital slider with the mouse
is a parsimonious heuristic, this is unlikely to be the whole story. to continuously rate their experienced emotional intensity on a
Specifically, there is no reason to expect that any information scale labeled with no emotion at all at the bottom and very intense
during an experience would be systematically ignored. It is more emotion at the top. The length of the scale on the monitor was
reasonable to expect that a process of continuous updating occurs 105 mm. Participants’ ratings were recorded with a sampling rate
over the course of an experience, which would most easily be of 10 data points/s (10 Hz; see Nagel et al., 2007). Data readout
gathered by the average of the experience of all moments. ranged from 0 (no emotion at all) to 100 (very intense emotion)

Frontiers in Psychology | Emotion Science August 2014 | Volume 5 | Article 911 | 4

Schäfer et al. Remembered emotional intensity

Table 1 | Songs used in the study. they had heard any of the songs prior to the study. The procedure
led to a certain time delay between listening with moment-to-
Song Performer Length (min:s) Frequency
moment rating and making the retrospective evaluation, which
Making Love Out of Bonnie Tyler 7:49 44 was desired to reduce recency effects due to participants still hav-
Nothing at All ing their rating profiles in mind. Finally, participants completed a
The Post War Dream Pink Floyd 3:01 49 questionnaire on their personal data and musical habits and were
A Two Hearts Spell Claim 4:01 47 then debriefed.
Path Apocalyptica 3:06 37
Im Herz Kubrick feat. Xavier 3:59 49 STATISTICAL ANALYSES
Naidoo Every participant generated continuous moment-to-moment
Silence DJ Tomcraft 4:19 45 emotional intensity data and a corresponding retrospective rating
Firstclass Suicide Anna Luca 4:23 48 of overall emotional intensity for each of the 11 songs. Only songs
Cloudburst—Grand Ferde Grofé 7:42 47 that participants had never listened to before were included in
Canyon Suite the analysis. The first song every person listened to was excluded
My Heart Is Empty Garda 3:34 49 because it was used for training the continuous self-reporting by
Mother and Child Sara Lugo 4:01 50 using the slider. As songs were presented in random order, the
Hiroshima Greta 3:49 43 exclusion affected each of the 11 songs with about the same fre-
quency (the resulting absolute frequency with which each song
Frequency denotes the number of occurrences of each song over all participants.
was included in the analyses is shown in Table 1). Figure 1 gives
Numbers differ because songs were presented in random order and the first
an example of the temporal profiles of three exemplary songs
song for each participant was used as a training stimulus and not included in the
analyses. Moreover, songs that were known to a participant prior to the study
from one exemplary participant. As can be seen, a profile can
were also excluded.
exhibit more than only one peak. Note that lines are of different
lengths because the songs differed in their lengths and thus in the
number of “moments.”
in 101 possible steps. Retrospective evaluation of overall emo- Beginning, peak, multiple peaks, end, peak–end, multiple
tional intensity was measured by participants’ single global rating peaks–end, number of multiple peaks, sum, average, and vari-
for each song with a pencil on the same scale printed on paper. ation variables were calculated for every song for every person.
Furthermore, for every song, participants had to indicate if they Beginning of a song was defined as the period from 5 to 15 s
had ever listened to it before. after the onset of a song. Unfolding of emotions over time and
response latencies for corresponding ratings are thought to take
PROCEDURE place within about 5 s of stimulus onset (Sloboda and Lehmann,
When participants had been seated in front of the computers, 2001; Nagel et al., 2007). In addition, the slider on the screen was
they received instructions regarding the purpose of the study, positioned at the bottom of the scale when each trial started. If
the use of headphones, volume settings, the digital slider, and the participants could not be expected to move the slider within
the questionnaires. Furthermore, participants learned from infor- the first seconds, this would produce zero values for the begin-
mation given on the computer screen that the study was about ning that should not be interpreted as an absence of emotional
the individual course of their experienced emotions. They should intensity during this period, however. Therefore, to obtain a reli-
continuously observe the intensity of emotions they felt and able measure of a song’s onset of emotional intensity, beginning
indicate their ratings on the scale on the screen by moving the was calculated by the average of those 10 s. End of a song was
digital slider, which they were able to become familiar with before defined as the last 10 s and was calculated by the average of rat-
the first experimental song started. After participants started the ings of the last 10 s. Peak of a song was defined as the maximum
session, the 11 songs were played in random order and moment- value of the whole temporal profile. The multiple peaks variable
to-moment ratings of emotional intensity were recorded. After was defined as the average of all moment ratings that repre-
each song, there was a short break of 10 s, automatically followed sented a local maximum2. Peak–end of a song was calculated by
by the next song. the average of the peak and end values. As an alternative peak–
When they had listened to all the songs, participants were told end measure, a multiple peaks–end variable was calculated by the
that the second part of the study was about the overall intensity of average of the multiple peaks and end values. Number of multi-
emotions felt for each song. As some time had passed since they ple peaks was the number of the local maxima during the whole
had listened to each specific song, they were told that short rep- piece. Sum was defined as the sum of the intensities of all the sin-
resentative excerpts of about 20 s of each song (snippets) should gle moments of the experience. Average of a song was defined
help them recollect the music. They were instructed to remember as the arithmetic mean of all single values. Variation of a song
and retrospectively evaluate the intensity of emotions elicited by
the respective song in general. After participants started the ses- 2 A value was defined a local maximum when there were no higher values
sion, the 11 snippets were played in the same individual order as within a 20 s window around that value AND that value lay above the moving
the songs were played earlier. After each snippet, participants had average, which was also calculated using a 20 s window. Twenty seconds were
time to give one overall evaluation of emotional intensity per song chosen as a reasonable range because peaks usually last between 2 and 20 s (see
on a paper version of the scale. They also were to indicate whether Grewe et al., 2007b).

www.frontiersin.org August 2014 | Volume 5 | Article 911 | 5

Schäfer et al. Remembered emotional intensity

FIGURE 1 | The temporal profiles of emotional intensity ratings for three exemplary songs from one exemplary participant.

average would exhibit the highest correlation to the overall evalu-

ation. In addition, we ran a multiple regression analysis, in order
to compare the relative impact of the most important parameters.
Since many of the parameters we calculated can be expected to be
affected by multicollinearity (e.g., sum will be highly correlated
with average), we were not able to include all of the parameters in
a simultaneous regression analysis. However, at least the two most
important parameters should be taken into account in any case:
average—as a measure of the temporal integration theory—and
peak–end—as a potential measure of the duration neglect theory.
Our prediction was that average is a better predictor of the over-
all evaluation than peak–end. Note that the multiple regression
analysis is based on variances that might originate from different
levels, that is, different participants and different songs. Data that
are organized in such a hierarchical structure might produce level
effects, leading to artificial correlations that are due to different
response patterns (e.g., participants with a restrained vs. partici-
pants with a permissive response behavior). Therefore, to control
FIGURE 2 | Boxplots showing the distributions of the
moment-to-moment parameters and the global rating of emotional
for the different variance components, we ran a multiple regres-
intensity, calculated across all participants and all songs. Note that sum sion analysis using hierarchical linear modeling (with the software
and number of multiple peaks are not included because they have a HLM 7; Raudenbush et al., 2004).
different scale.

RESULTS
CORRELATION ANALYSIS
was defined as the standard deviation of the temporal profile. Table 2 and Figure 3 show the results of the correlation analysis.
The boxplots in Figure 2 show the descriptive statistics of all the While average is most highly correlated with the global rating—
parameters and the overall evaluation (global rating). Note that which corresponds to our prediction—most of the remaining
sum (M = 104.659; SD = 60.842) and number of multiple peaks parameters also exhibit high correlations: end, peak, multiple
(M = 7.6; SD = 3.4) are not included in Figure 2 because they peaks, peak–end, multiple peaks–end and sum. Beginning, number
had a different scale. of multiple peaks, and variation were only moderately corre-
To analyze which of these parameters contribute most to the lated with the global rating. That is, these results do not clearly
explanation of variance of the overall evaluation, we first calcu- speak for or against one specific theory regarding the emergence
lated Pearson correlation coefficients. Our prediction was that of the overall evaluation. Considering these results, one might

Frontiers in Psychology | Emotion Science August 2014 | Volume 5 | Article 911 | 6

Schäfer et al. Remembered emotional intensity

Table 2 | Pearson correlations between the overall evaluation and all the parameters of moment-to-moment experience (N = 507).

Parameter Beginning End Peak Multiple Peak–end Multiple Number of Average Sum Variation
peaks peaks–end multiple peaks

Global rating 0.337 0.704 0.701 0.771 0.784 0.796 0.118 0.805 0.736 0.449
Beginning 0.276 0.362 0.440 0.354 0.382 0.297 0.478 0.442 −0.122
End 0.604 0.698 0.905 0.935 0.096 0.763 0.700 0.634
Peak 0.903 0.885 0.805 0.241 0.849 0.785 0.755
Multiple peaks 0.890 0.908 0.179 0.963 0.880 0.579
Peak–end 0.974 0.184 0.897 0.827 0.605
Multiple peaks–end 0.146 0.927 0.849 0.497
Number of multiple peaks 0.197 0.355 0.032
Average 0.926 0.483
Sum 0.421

All coefficients are significant at p < 0.001, except for global rating and number of multiple peaks (p = 0.008), beginning and variation (p = 0.006), end and number
of multiple peaks (p = 0.031), number of multiple peaks and variation (p = 0.474).

in a regression analysis simultaneously. Thus, we had to decide

which of the parameters to include. Given the theoretical argu-
ments raised in the Introduction, we were obliged to include at
least average and a peak–end measure. Regarding the peak–end
measure, it was difficult to decide whether to use the average of
peak and end values or the average of multiple peaks and end val-
ues, because both measures were highly correlated with the global
rating. We therefore calculated two separate models, one includ-
ing peak–end and another one including multiple peaks–end.
From the remaining parameters, there were only two left that were
not affected by multicollinearity: the number of multiple peaks
and the variation of the song’s profiles. So we eventually included
average, peak–end or multiple peaks–end, respectively, number of
multiple peaks, and variation in simultaneous regression analyses.
To check for potential level effects, we first calculated the
intraclass correlation, obtaining a value of r = 0.11, which indi-
cated the use of hierarchical linear modeling. Each of the two
regression models accounts for about 70% of the variance of the
overall evaluation. The specific influences of the four predictors
FIGURE 3 | Pearson correlations between the overall evaluation and all
in each model are shown in Table 3. In both models, average and
the parameters of moment-to-moment experience (N = 507).
peak–end or multiple peaks–end, respectively, turned out to be sig-
nificant predictors, with average yielding much larger regression
coefficients than peak–end or multiple peaks–end, respectively.
suppose that listeners build a moving average while listening to Variation and number of multiple peaks did not turn out to be
the music but that the average is affected by moments of specific significant predictors of the global rating.
importance—peak or multiple peaks, respectively, and end—that
can be combined to a peak–end measure. Since we were inter- DISCUSSION
ested in a direct comparison of the contributions of parameters The aim of the present study was to answer the question of
to explaining the variance of the global rating we additionally ran how people remember and evaluate the emotional intensity of
multiple regression analyses. past musical experiences. To investigate which parameters of past
In addition, we also calculated the correlation between remem- musical experiences influence retrospective overall evaluation of
bered overall intensity (global rating) and liking, yielding a strong emotional intensity most, several parameters of musical experi-
positive correlation of r = 0.63 (p < 0.001). ence’s temporal profile of emotional intensities were taken into
account: beginning, peak, multiple peaks, end, peak–end, mul-
MULTIPLE REGRESSION ANALYSIS tiple peaks–end, number of multiple peaks, sum, average, and
Due to multicollinearity, we could not include all of the param- variation. A correlation analysis showed that all of these param-
eters in a single simultaneous regression analysis. For instance, eters are significantly correlated with the overall evaluation, with
the variable peak–end is a nearly perfect linear combination of the peak, multiple peaks, end, peak–end, multiple peaks–end, sum, and
variables peak and end, so these three variables cannot be included average exhibiting large correlation coefficients. Average was the

www.frontiersin.org August 2014 | Volume 5 | Article 911 | 7

Schäfer et al. Remembered emotional intensity

Table 3 | Regression coefficients of a selection of distinct parameters on the overall evaluation. There may be two stages where this can
of moment-to-moment experience for their influence on the overall occur. One possibility is that moments of outstanding intensity
evaluation (N = 507). affect the continuous calculation of the moving average with a
higher weight than any other moments online, that is, during the
Parameter β p R2
course of the listening experience. There is a finding by Rozin et al.
Model 1 0.69 (2004) that speaks for this conjecture. These authors found that
Average 0.59 <0.001 steeper slopes in the profile of the moment-to-moment emotional
Peak–end 0.26 <0.001 intensity led to higher overall evaluations. This might indicate
Variation 0.02 0.56 that steeper slopes presage a peak that subsequently affects the
Number of multiple peaks 0.01 0.61 processing of an average more intensely. An alternative possibil-
Model 2 0.70 ity is that listeners continuously calculate an average, which is
Average 0.58 <0.001 adjusted by peak and end moments only afterward, that is, when
Multiple peaks–end 0.25 0.002 the listening experience is over. In this case, listeners would have
Variation 0.06 0.08 to keep the whole temporal profile in mind because they cannot
Number of multiple peaks 0.02 0.37 identify peaks until the experience has come to an end. So they
would calculate the average of peak–end or multiple peaks–end
only in retrospect and subsequently use this value to adjust the
parameter most highly correlated with the global rating, but that initial average. Cojuharenco and Ryvkin (2008) have also argued
correlation was only slightly larger than those between the global that moment-to-moment bits of information of an experience do
rating and other parameters such as peak–end or multiple peaks– not get lost—people use them very well for processing an aver-
end. When a selection of non-multicollinear parameters was taken age value—but that peaks[s] and end are nevertheless important
into account as simultaneous predictors of the global rating, aver- information people might pay particular attention to when build-
age emerged as the most influential variable. A second significant ing their subsequent evaluation. With the present data we are not
predictor—with a much smaller regression coefficient—was the able to adjudicate between these alternative hypotheses. One can
peak–end variable. For the peak–end variable, it did not matter question, however, how likely it is that listeners will hold a cor-
whether it was calculated from either the average of one peak (the rect and unbiased representation of the whole temporal profile in
highest value of the temporal profile) and the end value or multi- mind. It is more likely that what listeners remember is an evalua-
ple peaks and the end value. This indifference is also highlighted tion that was left at the end of a listening experience and achieved
by the non-significant influence of the number of multiple peaks through a weighted averaging of the experience.
on the global rating. Thus, for the overall impression about the The slope effect found by Rozin et al. (2004) had also led us to
emotional intensity of a musical experience, it seems to be essen- incorporate a measure of variation in the prediction of the overall
tial that there is an outstanding peak but it appears to make no evaluation. Although exhibiting a medium correlation with the
difference if there is only one peak or a series of multiple peaks. global rating, this parameter did not specifically contribute to the
Connecting these results to the theoretical approaches dis- prediction of the global rating. Thus, bringing more variation into
cussed in the introduction provides us with an interesting picture. an experience does not seem to lead to a higher evaluation of the
As we have pointed out, there are two main competing theories: experience in retrospect.
one that proposes that every single moment of an experience is If our conclusions are correct they should affect our under-
integrated when an overall evaluation is processed (temporal inte- standing of retrospective affective judgments in general, not only
gration) and one that proposes that this is the case for only some for musical stimuli. It seems reasonable to suggest that most kinds
specific single moments (duration neglect). Our results demon- of experiences are remembered and evaluated by a combination
strate that the average of all experienced moments is the best of average and a peak–end variable. Moreover, this might account
predictor for the overall evaluation. In addition, however, some for the heterogeneous findings in past research (see above) and it
specific moments of the experience appear to have an additional seems worthwhile to replicate studies with painful or pleasurable
influence on the final evaluation. We can thus conclude that some experiences while taking into account all the potential parameters
specific moments (such us peaks or the end) or an average of such that might influence an overall evaluation.
moments (such as peak–end or multiple peaks–end) do play a The memory and evaluation of affective intensity is clearly
role in building an overall evaluation but that they are neither of interest to performers and composers, whether they want to
sufficient for this evaluation nor the most important elements of arrange a single piece of music that leaves an intense overall
an experience. As we argued earlier, listeners might continuously memory, affectively powerful musical experiences consisting of
update the level of emotional intensity over the course of a listen- several movements, or a number of pieces into an album. Taking
ing experience. Longer passages of weak emotional intensity will the results of the present study into account, when composing
lower the moving average and thus the retrospective evaluation, a piece of music that is intended to be remembered as emo-
while longer passages of high emotional intensity will elevate the tionally intense one should arrange for an overall high level of
moving average and the subsequent evaluation. However, it seems emotional intensity or at least for a high peak or several high
reasonable that moments of outstandingly high intensity (such peaks, respectively, and an emotionally intense end. People’s atti-
as the peak or multiple peaks) and the moments at the very end tudes and behavior (how much they like a song, whether they
(which “benefit” from a recency effect) have an additional impact want to attend a concert or buy a CD) are likely to be influenced

Frontiers in Psychology | Emotion Science August 2014 | Volume 5 | Article 911 | 8

Schäfer et al. Remembered emotional intensity

by the overall emotional intensity of a musical experience they Duke, R. A., and Colprit, E. J. (2001). Summarizing listener perceptions over time.
have kept in mind. We have found a strong positive correlation J. Res. Music Educ. 49, 330–342. doi: 10.2307/3345616
Fredrickson, B. L. (2000). Extracting meaning from past affective experiences: the
between remembered emotional intensity and liking. Studying
importance of peaks, ends, and specific emotions. Cogn. Emot. 14, 577–606. doi:
the time course of emotional intensity during concerts or when 10.1080/026999300402808
listening to a whole set of pieces remains an interesting task for Fredrickson, B. L., and Kahneman, D. (1993). Duration neglect in retrospec-
future research, at any rate (see also Ariely and Zauberman, 2000). tive evaluations of affective episodes. J. Pers. Soc. Psychol. 65, 45–55. doi:
Note that we defined the peak as the most emotionally intense 10.1037/0022-3514.65.1.45
Goldstein, A. (1980). Thrills in response to music and other stimuli. Physiol.
moment and multiple peaks as the average of all local maxima
Psychol. 8, 126–129. doi: 10.3758/BF03326460
of moment-to-moment intensity of a musical experience. We Grewe, O., Nagel, F., Kopiez, R., and Altenmüller, E. (2007a). Emotions over time:
believe, however, that the peak concept needs some theoreti- synchronicity and development of subjective, physiological, and facial affective
cal clarification in the future, at least with regard to music. reactions to music. Emotion 7, 774–788. doi: 10.1037/1528-3542.7.4.774
Specifically, it remains to be clarified to what degree peak expe- Grewe, O., Nagel, F., Kopiez, R., and Altenmüller, E. (2007b). Listening to music
as a re-creative process: physiological, psychological, and psychoacoustical
riences are comparable to frisson or chill experiences (e.g., correlates of chills and strong emotions. Music Percept. 24, 297–314. doi:
Goldstein, 1980; Sloboda, 1991; Panksepp, 1995). There are some 10.1525/mp.2007.24.3.297
candidate musical characteristics that can elicit chills (such as Huron, D. (2006). Sweet Anticipation: Music and the Psychology of Expectation.
crescendos, the beginning of a new part, onset of a voice; see Cambridge, MA: MIT Press.
Huron, 2006; Grewe et al., 2007a), which might play an important Juslin, P. N., and Laukka, P. (2004). Expression, perception, and induction of musi-
cal emotions: a review and a questionnaire study of everyday listening. J. New
role in the memory of past musical experiences. Music Res. 33, 217–238. doi: 10.1080/0929821042000317813
Not least, we would argue that attitudes and behavior regard- Kahneman, D. (1999). “Objective happiness,” in Well-Being: Foundations of Hedonic
ing music depend not only on emotional intensity but also—and Psychology, eds D. Kahneman, E. Diener, and N. Schwarz (New York, NY: Russell
particularly—on pleasantness. Hence, a study that investigates Sage Foundation Press), 3–25.
Kahneman, D. (2011). Thinking, Fast and Slow. New York, NY: Farrar, Straus and
the time course of pleasantness and the influence of its param-
Giroux.
eters on the global evaluation of pleasantness would tie in well Kahneman, D., Fredrickson, B. L., Schreiber, C. A., and Redelmeier, D. A. (1993).
with the present work. Unlike the results we found in the present When more pain is preferred to less: adding a better end. Psychol. Sci. 4, 401–405.
study, results of an investigation of pleasantness could reveal a doi: 10.1111/j.1467-9280.1993.tb00589.x
much stronger impact of the beginning of a musical experience, Kemp, S., Burt, C. D. B., and Furneaux, L. (2008). A test of the peak-end
rule with extended autobiographical events. Mem. Cognit. 36, 132–138. doi:
since we all know that whether we like a piece of music or not is
10.3758/MC.36.1.132
decided quite quickly, after just a few seconds (see e.g., Zajonc, Langer, T., Sarin, R., and Weber, M. (2005). The retrospective evaluation of pay-
1980; Salimpoor et al., 2013). ment sequences: duration neglect and peak-and-end effects. J. Econ. Behav.
Organ. 58, 157–175. doi: 10.1016/j.jebo.2004.01.001
ACKNOWLEDGEMENTS Madsen, C. K. (1990). Measuring musical response. Music Educ. J. 77, 26. doi:
10.2307/3397835
The publication costs of this article were funded by the German Madsen, C. K., Brittin, R. V., and Capperella-Sheldon, D. A. (1993). An empirical
Research Foundation/DFG (Geschäftszeichen INST 270/219-1) method for measuring the aesthetic experience to music. J. Res. Music Educ. 41,
and the Chemnitz University of Technology in the funding pro- 57–69. doi: 10.2307/3345480
gramme Open Access Publishing. Nagel, F., Kopiez, R., Grewe, O., and Altenmüller, E. (2007). EMuJoy: software for
continuous measurement of perceived emotions in music. Behav. Res. Methods
39, 283–290. doi: 10.3758/BF03193159
REFERENCES Panksepp, J. (1995). The emotional sources of ‘chills’ induced by music. Music
Ariely, D., and Zauberman, G. (2000). On the making of an experience: Percept. 13, 171–207. doi: 10.2307/40285693
the effects of breaking and combining experiences on their overall eval- Raudenbush, S. W., Bryk, A. S., and Congdon, R. (2004). HLM 6 for Windows
uation. J. Behav. Decis. Making 13, 219–232. doi: 10.1002/(SICI)1099- [Computer software]. Skokie, IL: Scientific Software International.
0771(200004/06)13:2<219::AID-BDM331>3.0.CO;2-P Redelmeier, D. A., and Kahneman, D. (1996). Patients’ memories of painful medical
Baumgartner, H., Sujan, M., and Padgett, D. (1997). Patterns of affective reac- treatments: real-time and retrospective evaluations of two minimally invasive
tions to advertisements: the integration of moment-to-moment responses into procedures. Pain 66, 3–8. doi: 10.1016/0304-3959(96)02994-6
overall judgments. J. Mark. Res. 34, 219–232. doi: 10.2307/3151860 Robinson, E., Blissett, J., and Higgs, S. (2011). Peak and end effects on remembered
Behne, K.-E. (1997). “The development of “Musikerleben” in adolescence. How enjoyment of eating in low and high restrained eaters. Appetite 57, 207–212. doi:
and why young people listen to music,” in Perception and Cognition of Music, 10.1016/j.appet.2011.04.022
eds I. Deliège and J. Sloboda (Hove: Psychology Press), 143–159. Rode, E., Rozin, P., and Durlach, P. (2007). Experienced and remembered pleasure
Chemnitz University of Technology. (2002). Code of Conduct for Assurance for meals: duration neglect but minimal peak, end (recency) or primacy effects.
of Good Scientific Practice and for Procedures in the Case of Suspicion of Appetite 49, 18–29. doi: 10.1016/j.appet.2006.09.006
Scientific Misconduct of Chemnitz University of Technology From 26 November Rozin, A., Rozin, P., and Goldberg, E. (2004). The feeling of music past:
2002. Technische Universität Chemnitz. Available online at: https://www. how listeners remember musical affect. Music Percept. 22, 15–39. doi:
tu-chemnitz.de/forschung/grundsatz.php.en 10.1525/mp.2004.22.1.15
Cojuharenco, I., and Ryvkin, D. (2008). Peak-end rule versus average utility: Saarikallio, S. and Erkkilä, J. (2007). The role of music in adolescents’ mood
how utility aggregation affects evaluations of experiences. J. Math. Psychol. 52, regulation. Psychol. Music 35, 88–109. doi: 10.1177/0305735607068889
326–335. doi: 10.1016/j.jmp.2008.05.004 Salimpoor, V. N., Bosch, I., van den Kovacevic, N., McIntosh, A. R., Dagher,
Deutsche Gesellschaft für Psychologie. (2005). Ethische Richtlinien der DGPs und A., and Zatorre, R. J. (2013). Interactions between the nucleus accumbens
des BDP. Deutsche Gesellschaft für Psychologie e.V. Available online at: http:// and auditory cortices predict music reward value. Science 340, 216–219. doi:
www.dgps.de/index.php?id=96422 10.1126/science.1231059
Do, A. M., Rupert, A. V., and Wolford, G. (2008). Evaluations of pleasurable expe- Schäfer, T., Sedlmeier, P., Städtler, C., and Huron, D. (2013). The psy-
riences: the peak-end rule. Psychon. Bull. Rev. 15, 96–98. doi: 10.3758/PBR. chological functions of music listening. Front. Psychol. 4:511. doi:
15.1.96 10.3389/fpsyg.2013.00511

www.frontiersin.org August 2014 | Volume 5 | Article 911 | 9

Schäfer et al. Remembered emotional intensity

Schreiber, C. A., and Kahneman, D. (2000). Determinants of the remembered Zajonc, R. B. (1980). Feeling and thinking: preferences need no
utility of aversive sounds. J. Exp. Psychol. Gen. 129, 27–42. doi: 10.1037/0096- inferences. Am. Psychol. 35, 151–175. doi: 10.1037/0003-066X.
3445.129.1.27 35.2.151
Sedlmeier, P., and Schäfer, T. (2013). “Emotions, moods, and musical preferences,”
in Handbook of Psychology of Emotions, eds C. Mohiyeddini, S. Bauer, and M. Conflict of Interest Statement: The authors declare that the research was con-
Eysenck (New York, NY: Nova Science Publishers), 327–337. ducted in the absence of any commercial or financial relationships that could be
Sloboda, J. A. (1991). Music structure and emotional response: some empirical construed as a potential conflict of interest.
findings. Psychol. Music 19, 110–120. doi: 10.1177/0305735691192002
Sloboda, J. A., and Lehmann, A. C. (2001). Tracking performance correlates of Received: 27 March 2014; accepted: 30 July 2014; published online: 15 August 2014.
changes in perceived intensity of emotion during different interpretations of a Citation: Schäfer T, Zimmermann D and Sedlmeier P (2014) How we remember the
Chopin piano prelude. Music Percept. 19, 87–120. doi: 10.1525/mp.2001.19.1.87 emotional intensity of past musical experiences. Front. Psychol. 5:911. doi: 10.3389/
Sloboda, J. A., O’Neill, S. A., and Ivaldi, A. (2001). Functions of music in everyday fpsyg.2014.00911
life: an exploratory study using the experience sampling method. Music. Sci. 5, This article was submitted to Emotion Science, a section of the journal Frontiers in
9–32. doi: 10.1177/102986490100500102 Psychology.
Stone, A. A., Broderick, J. E., Kaell, A. T., DelesPaul, P. A. E. G., and Porter, L. Copyright © 2014 Schäfer, Zimmermann and Sedlmeier. This is an open-access
E. (2000). Does the peak-end phenomenon observed in laboratory pain stud- article distributed under the terms of the Creative Commons Attribution License
ies apply to real-world pain in rheumatoid arthritics? J. Pain 1, 212–217. doi: (CC BY). The use, distribution or reproduction in other forums is permitted, pro-
10.1054/jpai.2000.7568 vided the original author(s) or licensor are credited and that the original publi-
Varey, C., and Kahneman, D. (1992). Experiences extended across time: eval- cation in this journal is cited, in accordance with accepted academic practice. No
uation of moments and episodes. J. Behav. Decis. Making 5, 169–185. doi: use, distribution or reproduction is permitted which does not comply with these
10.1002/bdm.3960050303 terms.

Frontiers in Psychology | Emotion Science August 2014 | Volume 5 | Article 911 | 10

View publication stats