Content uploaded by Kongmeng Liew
Author content
All content in this area was uploaded by Kongmeng Liew on Aug 14, 2023
Content may be subject to copyright.
Psychology of Aesthetics, Creativity, and the Arts
Groovin’ to the Cultural Beat: Preferences for Danceable Music Represent
Cultural Affordances for High-Arousal Negative Emotions
Kongmeng Liew, Alethea H. Q. Koh, Noah R. Fram, Christina M. Brown, Cheslie dela Cruz, Li Neng Lee, Romain Hennequin,
Amanda E. Krause, and Yukiko Uchida
Online First Publication, July 13, 2023. https://dx.doi.org/10.1037/aca0000599
CITATION
Liew, K., Koh, A. H. Q., Fram, N. R., Brown, C. M., dela Cruz, C., Lee, L. N., Hennequin, R., Krause, A. E., & Uchida, Y. (2023,
July 13). Groovin’ to the Cultural Beat: Preferences for Danceable Music Represent Cultural Affordances for High-Arousal
Negative Emotions. Psychology of Aesthetics, Creativity, and the Arts. Advance online publication.
https://dx.doi.org/10.1037/aca0000599
Groovin’to the Cultural Beat: Preferences for Danceable Music Represent
Cultural Affordances for High-Arousal Negative Emotions
Kongmeng Liew
1, 2
, Alethea H. Q. Koh
1, 3
, Noah R. Fram
4, 5
, Christina M. Brown
6
, Cheslie dela Cruz
7
,
Li Neng Lee
7
, Romain Hennequin
8
, Amanda E. Krause
9
and Yukiko Uchida
3
1
Graduate School of Human and Environmental Studies, Kyoto University
2
School of Psychology, Speech and Hearing, University of Canterbury
3
Institute for the Future of Human Society, Kyoto University
4
Center for Computer Research in Music and Acoustics, Stanford University
5
Department of Otolaryngology —Head & Neck Surgery, Vanderbilt University Medical Center
6
Department of Psychology, Arcadia University
7
Department of Psychology, National University of Singapore
8
Deezer Research, Deezer
9
College of Healthcare Sciences, James Cook University
Music is a product of culture. Cross-cultural examinations of music features can reveal novel information
about the cultural psychological processes involved in shaping music preferences. In Studies 1 and 2, we
first identified differences in music preferences through machine learning of East-Asian and Western popular
music on Spotify (combined N=1,006,644). In interpreting these results, we developed a theory on dance-
ability as a music feature, that represents cultural affordances for high-arousal emotions. Subsequent confir-
matory studies (Studies 3–5, combined N
songs
=3,343, N
participants
=495, N
countries
=60) tested this theory
by examining danceability and the role of emotion in music preferences. Specifically, we found that dance-
ability represents cultural affordances for high-arousal negative (HAN) emotions: societies with greater
HAN emotion prevalence generally prefer listening to more danceable music. Consistently, this was also
observed more in independent individuals and culturally looser countries. Using evidence from Japanese
and American participants (Study 5), we propose a mechanism through discharge regulation in music: cul-
tures with looser cultural norms would also have more experiences of HAN emotions in daily life. Discharge
regulation, which is listening to music to cathartically release HAN emotions, would then skew music pref-
erences toward high-arousal (danceable) music to facilitate this cathartic HAN downregulation. These find-
ings have implications for cross-cultural research by demonstrating that music features, being widely
accessible and almost universally perceived, can quantify cultural tendencies toward affective (HAN emo-
tion) norms beyond commonly used self-report paradigms.
Keywords: cross-culture, music preference, emotion regulation, danceability, machine learning
Music is powerful, emotive, and prevalent across cultures and his-
tory (Dunbar, 2012). Despite some cross-cultural variations in the
experience of music, there is considerable agreement that statistical
universals exist in the way music is perceived (Cowen et al., 2020;
Fritz et al., 2009;Purves, 2017;Stevens & Byron, 2016). Thus,
cross-cultural variation may often imply differences in music prefer-
ence: functions and inclinations that afford preference for specific
types of music over others (Cross, 2001;Juslin et al., 2016). In
this article, we explore cultural differences in music preferences as
areflection of wider sociocultural trends. Specifically, we posit
that cultural differences in preferred musical features can be
explained by the psychological affordances for those features, and
the value of those affordances varies by culture.
Music as a Cultural Product
Preferences for different products—defined here as objects or con-
structs with a physical basis—vary across cultures. This is evident in
consumption behavior, such as choosing music consistent with cultur-
ally determined affect values (Tsai et al., 2007) and choosing products
based on societal values of uniqueness versus conformity seeking
(Kim & Markus, 1999). This is also evident in production, where prod-
ucts are created to fit shared values and esthetics (e.g., analytic vs.
Kongmeng Liew https://orcid.org/0000-0002-0755-7173
Christina M. Brown is nowat Meta, Inc. in Seattle, Washington, United States.
All online supplementary material is located in our OSF (Open Science
Framework) repository: https://osf.io/hg7sc/?view_only=9bb77cab0c7642
af98c4fc37980627ac. This includes code for data collection and analysis
from the R programming environment, as well as raw Jamovi output for
analyses in Jamovi. Where applicable, data are available online. Data are
also available upon request to the corresponding author.
This project was supported by a Grant-in-Aid for Research Fellows of the
Japan Society for the Promotion of Science (19J14431), and a National Arts
Council Capability Development Grant (FY 16/6).
Correspondence concerning this article should be addressed to Kongmeng
Liew, School of Psychology, Speech and Hearing, University of Canterbury,
Private Bag 4800, Christchurch 8140, New Zealand. Email: kongmeng
.liew@canterbury.ac.nz
Psychology of Aesthetics, Creativity, and the Arts
© 2023 American Psychological Association
ISSN: 1931-3896 https://doi.org/10.1037/aca0000599
1
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
holistic attention styles, Wang et al., 2012). These “cultural”products
are public, tangible, and shared, reflecting the intersubjective
consensus of a given culture (Lamoreaux & Morling, 2012;Morling
& Lamoreaux, 2008). Cultural products are commonly used in
research to quantify real-world cultural differences beyond self-report
questionnaires to support and reinforce current theory (e.g., individual-
ism–collectivism: Morling & Lamoreaux, 2008; analysis–holism:
Masuda et al., 2008). Such research allows for examinations of culture
beyond the laboratory setting, with real-world consequences in con-
sumption and behavior. The cultural values of a society may thus
afford preferences for certain styles or characteristics of music.
Music, in particular, may hold several benefits over other frequently
used products (e.g., news articles, tweets, and children’s books) due
to the widespread availability of related data and universality in percep-
tion, that provides usable data for cross-cultural (products) research.
Firstly, due to the recent growth of the field of music infor-
mation retrieval, and the prevalence of music streaming technol-
ogy, huge databases have become available to researchers (e.g.,
Bertin-Mahieux et al., 2011;Meseguer-Brocal et al., 2017). New
music is constantly being released and ranked, capturing contempo-
rary and historical trends across cultures. Knowing how cultural val-
ues are mapped on to music may thus allow for convenient, online
tracking of cultural change and comparisons through these massive
cultural (music) product databases. Moreover, these databases con-
tain information about the structural make-up of songs, through
music features such as rhythm and tempo. Some low-level features
can also be combined to form more complex features, such as dan-
ceability (rhythmic salience) or dissonance, that quantify music.
Secondly, these musical features reliably reflect individuals’percep-
tions of music (see Fricke et al., 2018), and are strongly consistent
with music-emotion recognition across cultures (Balkwill &
Thompson, 1999;Balkwill et al., 2004). For example, music with
strong rhythmic and percussive features is almost universally per-
ceived as highly arousing (Mehr et al., 2019), and is strongly asso-
ciated with dance across cultures (Savage et al., 2015). Yet, much
less is known about cultural differences in music preferences and
their implications on everyday emotional experiences, which is an
area we explore with the present research.
The cross-cultural consistency between features and perception
enables us to assume that significant cultural variation in musical
products (features), can then be attributed primarily to preference,
rather than perception. Countries that produce and consume more of
a particular style of music, for example, would also reflect their pref-
erences for those styles. These differences in musical styles can be
quantified in terms of low-level musical features, which should differ
across cultures in a consistent manner. Thus, music features may offer
a standardized way to quantify the cultural affordances behind these
preferences, without low-level confounds that comparisons of text-
based cultural products may face (e.g., differences in translation or
language structures). Accordingly, the present research starts with
data-driven analyses on cultural products (i.e., music features) to gen-
erate novel theories involving sociocultural affordances for music
preference, that are later examined by theory-driven approaches.
How Does Music Preference Reflect Emotions Across
Cultures?
Schäfer and Sedlmeier (2009) found that individuals’music prefer-
ences were attributable to music’s functional ability to regulate/
influence participants’affective states. Yet, desired affective states
are not equivalent across societies and cultures: Affect Valuation
Theory (AVT) argues that individuals are motivated to seek affective
experiences that match their preferences for ideal affect: emotions that
people ideally want to feel (Tsai, 2017). Accordingly, this impacts
cultural variations in music preference. Generally, Westerners prefer
high-arousal positive (HAP) emotions (e.g., excitement, elation) and
East Asians prefer low-arousal positive (LAP) emotions (e.g., calm,
serenity), and this extends to preferred arousal levels in music. Tsai
et al. (2007: Study 4) found that when asked to select music to listen
to by themselves in preparation for a later cooperative building task,
those in Hong Kong were more likely to choose music that had
lower arousal descriptions, while Caucasian-Americans were more
likely to choose music that had higher arousal descriptions. This is
consistent with cross-cultural research on composite Spotify features
that indicated arousal: users residing in Asia tended to play songs
that had lower arousal than users residing in Oceania, Europe, and
the Americas (Park et al., 2019).
The preference for high HAP emotions in Western societies is
consistent with a greater trend of affordance for HAPand HAN emo-
tions that are prevalent in these societies. For example, Western cul-
tures that are typically individualistic and independent, foster
socially disengaging emotions like pride or frustration (Kitayama
et al., 2006) that are usually high-arousal (Lim, 2016). By contrast,
collectivistic, interdependent societies may prioritize social har-
mony, thereby favoring socially engaging emotions like guilt or
sympathy, which are usually low-arousal (Kitayama et al., 2006).
In sum, preferences and prevalence for high-arousal emotional
states in the West seem congruent with high-arousal music prefer-
ences. However, to the best of our knowledge, comparatively less
research has sought to understand the reasons behind this phenom-
enon: how does high-arousal music preference relate to cultural
affordances for high-arousal emotions? One possibility could be
that high-arousal music directly induces the desired HAP emotion,
implying that listeners use music as mechanisms to upregulate
HAP emotion. Additionally, this could also reflect the downregula-
tion of high-arousal negative (HAN) emotions: cultures with social
norms that encourage HAP emotions also tend to accept displays and
experiences of HAN emotions. Consequently, these HAN emotion
states are cathartically released by listening to matching high-arousal
music (i.e., discharge regulation: Saarikallio, 2012;Sharman &
Dingle, 2015). For collectivistic cultures that prefer low-arousal
music, such high-arousal regulatory mechanisms may not be as rel-
evant, as initial evidence suggests that emotion regulation by music
occurs through low-arousal relaxation and comfort-related mecha-
nisms instead (Saarikallio et al., 2021). We propose that elucidating
these mechanisms could aid our understanding of music preference
as a reflection of sociocultural affordances for emotions. These stud-
ies have also focused primarily on emotion-arousal as an avenue for
cultural influences, but music may also reflect other aspects of emo-
tion through cultural mechanisms.
The Present Research
We thus needed to start with an exploratory approach to under-
stand broad links between music, emotion, and culture. In this arti-
cle, five studies were conducted to investigate the aspects of music
preferences, defined primarily by features, that reflect sociocultural
differences and affordances for emotion. Study 1 examined music
LIEW ET AL.2
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
features from music produced within cultures, reflecting the individ-
ual’s (producer) esthetics for music making. Studies 2 and 4 used
music features from songs in the Top 50 charts, reflecting the collec-
tive preferences for music within a culture. For generalizability, we
also examined music preferences from self-reported ratings (Studies
3 and 5). Our research structure is as follows: Studies 1 and 2 iden-
tified music features that reliably differed across cultures for theory
generation. Studies 3–5refined that theory and tested potential
underlying mechanisms. This allowed us to identify features that
reliably reflected regional differences in cultural affordances for
emotions, which would be narrowed down by the theory generation
Studies 1 and 2.
Exploring Cultural Differences in Music Features
Study 1
We compared musical features in popular music that were pro-
duced predominantly by either the United States (West) or Japan
(East). As cultural products, music that is locally produced should
reflect that culture’s preferences as indicative of broader societal
trends (e.g., Golder & Macy, 2011). We used machine learning clas-
sification (gradient boosted decision trees [GBDTs], Friedman,
2001) on data from Spotify (total N=1,006,644 songs) based on
afinal list of artists (Japan =2,515; United States =1,834) obtained
from Spotify’s recommendation systems on initial Top 50 playlists.
We classified songs as produced by Japanese or Western
1
artists
based on all 13 features from Spotify’s application programming
interface (API; see Table 1). Feature importance was evaluated
using relative variable importance (RVI) and partial dependence
plots (PDPs; Friedman, 2001). Our goal was to identify features
from the RVI list that could explain United States–Japan differences
in terms of affect.
Method and Analysis
Study 1 was conducted on Spotify data using the Spotify
(Thompson et al., 2019) wrapper for the Spotify API, and a total
of N=503,322 Japanese songs and N=678,218 Western songs
(later down-sampled to N=503,322 songs for balanced compari-
son) were collected. This was based on a list of artists (Japan =
2,515; Western =1,834), determined via a pseudo-snowball sam-
pling strategy: where Spotify’s recommendation systems on initial
Top 50 playlists from Japan and the United States were repeatedly
used for up to three iterations (two for the United States), and
non-Western/non-Japanese artists were manually removed from
the respective data by the authors. All songs released before 2010
were excluded, and duplicate songs (such as in rereleases or compi-
lation albums) were removed based on Spotify track identifiers (ID).
A list of artists, genres, songs, and release years are available in our
OSF repository. Japanese songs were obtained in July 2020, and
Western songs were obtained in November 2019.
For each song, a list of scores for various musical features was
obtained from Spotify’s API. Here, we note that in the list of publicly
available features provided by Spotify, most were low-level features
(e.g., duration, loudness, tempo), some were composites of a number
of low-level features (e.g., danceability, energy), and some compar-
atively high-level estimates (e.g., valence) were likely confidence
measures determined through predictive (machine learning) algo-
rithms previously trained on expert ratings (Van Buskirk, 2013),
and are more of an approximation of the variable of interest.
To increase the robustness and generalizability of our exploratory
findings, we used a machine learning approach (GBDT). The
Spotify features were entered as predictor variables, and the cultural
membership of the song (Japanese or Western) was entered as the
outcome variable in a binomial classification model. Using the val-
idation set approach (see James et al., 2013), we divided our data into
training and testing sets along a 3:1 ratio. All predictor variables
were mean-centered and standardized. A model was first developed
on the training set using GBDTs, before generating a set of predicted
outcomes (classification to Western or Japanese) based on the pre-
dictor variables of the testing set. These predicted outcomes were
then compared to the actual scores of the testing set, for a measure
of prediction accuracy as an indicator of the predictive power of
the model. A high accuracy score would suggest that the model is
capable of predicting if a song was Japanese or Western based on
its Spotify features, which would imply that strong differences
exist between both groups from within the predictor variables.
GBDT refers to a decision tree-based ensemble method for
machine learning that utilizes sequential iterations of multiple deci-
sion trees. In this method, successive trees are fitted to minimize the
residuals, resulting in a model that is often capable of high-accuracy
predictive learning. For more information on GBDTs, see Friedman
(2001). To minimize the risk of overfitting while training the model,
we used fivefold cross-validation on the training set to determine
the optimum parameters: number of decision trees in the ensemble
(ntrees) and the interaction depth (orders of interactions allowed
in each tree). This resulted in the following parameters: ntrees =
150, interaction depth =3. All procedures were conducted in R
Table 1
A List of Song-Level Audio Features Obtained From Spotify (https://
developer.spotify.com/documentation/web-api/reference/tracks/get-
audio-features/)
Audio feature Description
Duration The duration of a song (ms).
Key The estimated main key of a song.
Mode If a song is major or minor (in modality).
Time signature The estimated main time signature of a song.
Acousticness A confidence measure on whether a song is acoustic.
Danceability The suitability of a song for dancing. This is based on
several musical features, such as tempo, rhythmic
stability, regularity, and beat strength.
Energy A measure of the intensity and activity of a song. This is
based on several musical and spectral features, such as
dynamic range, loudness, timbre, onset rate, and
entropy.
Instrumentalness A confidence measure of whether a song contains no
vocals.
Liveness A confidence measure on the presence of audiences in the
recording.
Loudness The overall intensity of the song in decibels (dBFS).
Speechiness A confidence measure on the presence of spoken words
(e.g., audiobooks) in a song.
Valence An estimate of whether a song conveys positive or
negative affect.
Tempo The estimation of the main tempo of a song.
1
Artists collected from the US charts were considerably more diverse
in cultural origin, but still mainly from Western, predominantly English-
speaking cultures.
DANCEABILITY AND CULTURAL AFFECT 3
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
(R Core Team, 2017), using the “caret”wrapper (Kuhn, 2008) for
the “gbm”package (Greenwell et al., 2019). The analysis scripts
are also available in our OSF repository.
Results and Discussion
The model achieved a moderate accuracy of 0.76, 95% CI
[0.75, 0.76], area under the curve =0.83,
2
above a no information
rate of 0.5: the predicted classifications matched well with actual
cultural membership in the testing set. This suggested the existence
of systematic differences within the data in differentiating between
Japanese and Western songs on Spotify-based music features. In
interpreting the model, we checked the RVI of each predictor in
the model (in OSF). RVIs indicate the number of times (percentage)
a predictor is chosen for stratification from all the individual decision
trees in the model, as a measurement of the importance of that feature
in contributing toward the model’s accuracy. We focused on
predictors that had RVIs of more than 10.0, as these indicated com-
paratively stronger effects: “speechiness”(RVI =17.9), “duration”
(RVI =17.1), “instrumentalness”(RVI =16.4), “danceability”
(RVI =12.5), and “energy”(RVI =12.0). An estimate of their
differences across cultures can be inferred through the PDPs in
Figure 1. PDPs visualize the effect of a single predictor on the
logit probability of the classification (outcome variable) from
the model, while holding all other predictor variables constant.
The descriptive means and standard deviations of the predictor var-
iables are also included in our OSF repository.
Our goal was to identify features that could potentially indicate
cultural affordances for affect from the RVI list that could explain
the United States–Japan differences. We evaluate features from the
RVI list on the basis of their possible theoretical links to emotion,
but acknowledge that emotion is just one approach to examine cul-
tural differences in music. Of the selected feature set, danceability
and instrumentalness had the most accessible explanations grounded
in emotion-arousal. Danceability, a composite feature quantifying
rhythmic salience, was higher in American than Japanese music,
and could be indicative of HAP valuation and corresponding
upregulation, or HAN downregulation (Sharman & Dingle, 2015).
To a smaller extent, Japanese preferred music that was higher in
instrumentalness compared to Americans, suggesting that popular
music produced by Japanese artists may have had stronger instru-
mental elements. Instrumental music (e.g., classical music) often
has a calming, relaxing effect on the listener, reducing negative emo-
tions (NEs; such as anxiety and stress; Labbé et al., 2007) through
solace or distract regulation strategies (Saarikallio, 2012). This
would be consistent with AVT, in that East Asians prefer LAP emo-
tions. Energy, which encompasses a variety of different low-level
features like spectral density, intensity, and entropy, can also be
seen as linked to arousal, but was excluded from further analysis
in this article due to the vagueness in Spotify documentation
and difficulties with defining it.
3
Duration and speechiness did not
appear to be related to emotion and were excluded from this
interpretation.
Music features quantify the styles and types of music, and their
variation across cultures signify that the styles of music produced
are different between these cultures. However, what is produced in
a culture may not necessarily be what is widely consumed in that cul-
ture (see Liew, Mishra, et al., 2022). To examine if the sociocultural
affordances (such as function, Schäfer & Sedlmeier, 2009) for music
preference underlay the current findings (on music produced), we
would expect consistent differences to be found also in the (pre-
ferred) music listened to by these cultures, as measured by their
respective top charts.
Study 2
Study 2 used a more generalized Western (United States, United
Kingdom, Canada, Australia)–East Asian (Japan, Taiwan, Hong
Kong, Singapore) cultural comparison (N=800 songs) from official
Spotify Top 50 charts. Charts are frequently used in cultural product
research (e.g., Pérez-Verdejo et al., 2021) as they function as selec-
tion processes that capture the prevalent musical preferences of a cul-
ture. Following Study 1’s results, we hypothesized that East Asian
charts would have higher instrumentalness, but Western charts
would have higher danceability. Using charts from Spotify also
ensured consistency in the ranking mechanics regardless of country.
Method and Material
Songs were obtained from the Spotify Top 50 Playlists (charts) for
four East-Asian cultures (Japan, Taiwan, Hong Kong, Singapore)
and four Anglo-Western cultures (United States, United Kingdom,
Canada, Australia). Data were obtained at two time points 5 months
apart (December 2018 and April 2019) for a total N(songs) =800.
All data were obtained through the Spotifyr package in R. All mate-
rial (data set and Jamovi analysis files) are available in our OSF
repository. Pairwise comparisons were conducted using Student’s
ttesting for danceability, and Mann–Whitney Utesting for instru-
mentalness, due to the violation of normality assumptions.
Results and Discussion
The pattern of results from Study 1 was replicated for danceability.
Danceability scores were significantly higher in Western than
East Asian charts (t(798) =−7.49, SE =0.01, 95% CI [−0.099,
−0.058], p,.001, d=−0.53). However, instrumentalness was
inconsistent as it significantly differed between cultures in the oppo-
site direction from Study 1: instrumentalness was significantly higher
in Western than in East Asian charts (U=71,629, [−0.00005,
−0.00003], p,.001, d=−0.17). This raises the possibility that it
was a result of sample bias in Study 1, such as an oversampling
of instrumental music (e.g., soundtracks) from Japan, which could
be specific to the Spotify platform. As danceability was also easy to
define and based on a larger body of previous research, we thus
decided on examining danceabilityand its usefulness as a distinguish-
ing feature of music preferences between cultures.
Theory Generation: Danceability and Cultural
Differences in Affect
Our findings from the exploratory section revealed danceability as
a suitable differentiating feature between Western and East Asian
musical preferences in accordance with affect-based affordances.
2
An AUC score of 0.83 would be approximately Cohen’sd=1.37 or an
odds-ratio =11.5 (Salgado, 2018), for correct (true positive/true negative)
versus incorrect classifications.
3
For an updated discussion on how energy differs from danceability in
representing culturally-based affect, see Liew, Uchida, et al. (2022).
LIEW ET AL.
4
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Danceability measures the rhythmic salience of a song, which tends
to be universally perceived as high-arousal and associated with
dance activities (Mehr et al., 2019;Savage et al., 2015). Dance itself
is also a high-arousal activity (Bernardi et al., 2017), and has been
shown to enhance positive emotional experiences with “happy”
music (Christensen et al., 2014). Accordingly, we predicted that dan-
ceability in music reflects the extent to which high-arousal emotions
were valued or experienced in various cultures. While the research
primarily focuses on music features, in this section, we also quantify
music-emotion preferences through self-report surveys for a more
holistic examination.
Study 3
Following our hypothesis, cross-cultural differences in danceabil-
ity, and by extension, arousal, should also be present in other forms
of measurement. Study 3 examined individual levels of cross-
cultural differences in arousal and dance preferences through inves-
tigating music functions and preferences via a questionnaire study on
Singaporean (N=127) and American (N=141) undergraduates.
4
As functions of music are closely related to music preferences
(Schäfer & Sedlmeier, 2009), both affect preference and music-
listening functions should reflect the mechanisms predicted by dan-
ceability in a consistent manner: (a) preferences for (positively and
negatively valenced) high-arousal music should be stronger in the
United States than Singapore; (b) individuals in the United States
would prefer more dance functions, but individuals in Singapore
should prefer low-arousal regulatory functions (i.e., mood regula-
tion). Past research has elaborated on the link between affect and
independent-interdependent self-construal, so we also examined
cultural effects through self-construal at the individual level.
Given that the United States is typically more independently oriented
than East Asia (including Singapore; Markus & Kitayama, 1991),
we posit that (c) individuals with higher independent orientation
would use music for more dance functions and prefer high-arousal
music.
Music listening may not always be a main activity, in that recrea-
tional music listening is rarely devoid of secondary tasks or distrac-
tions, often accompanying other activities, like working, studying,
or driving (Krause & North, 2014;Sloboda & O’Neill, 2001).
Accordingly, we examined music preferences for arousal in music
in two contexts. Participants were asked to create an imaginary play-
list for the context of “relaxing”and “studying,”corresponding to
active versus background listening, respectively. Participants then
Figure 1
Partial Dependence Plots Indicate the Relationship Between Each Music Feature and Culture, in Computing the Unidirectional
Classification Probability of Songs as Japanese (1) or Western/United States (0) Across the Range for the Target Feature, with all Other
Features Held Constant
Note. The bottom left panel, for example, shows that higher danceability was more predictive of a song being Western than Japanese. PDP =partial depen-
dence plots.
4
Sample size was limited by the availability of the student sample in
Singapore, and the US sample size was set to match the Singapore sample.
DANCEABILITY AND CULTURAL AFFECT 5
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
rated the arousal for these playlists, and provided information on
musical reward styles and cultural self-construal. However, as
reported in the Results section, an equivalence test showed arousal
ratings across both contexts to be equivalent, so scores were aver-
aged into a single measure for arousal.
Procedure
An online questionnaire (in English) was administered to 268 par-
ticipants (Singapore: N=127, M
age
=21.4, SD =1.8, Females =
93, Males =34; United States: N=141, M
age
=20.0, SD =2.4,
Females =105, Males =33, Others =3) who were recruited
from respective undergraduate participant pools in Singapore
(National University of Singapore) and the United States (Arcadia
University) for course credit. Participants from Singapore completed
the web-based questionnaire on their own time and participants from
the United States completed the questionnaire within the psychology
laboratories on campus. This study obtained approval from the
respective institutional ethics review committees.
Material
To examine arousal levels of music preferences, participants were
asked to create an imaginary playlist for the context of “relaxing”
and “studying,”that measured different aspects of music listening:
attended recreational listening (relaxing) and unattended background
listening (studying). Following this, they were asked to rate each
playlist on a list of 24 descriptors (see North & Hargreaves, 1996).
Following stable factor structures identified by Krause and North
(2018), these descriptors were then grouped into three latent factors:
arousing (attention-grabbing, invigorating, loud, exotic, can dance
vigorously to it, exciting/festive, and strong rhythm), serene (beau-
tiful, inspiring-majestic, natural-fresh, romantic, and relaxing-
peaceful), and melancholy (sad and moody). For the purpose of
our study, we assumed the arousing factor to reflect high-arousal
preferences.
5
Participants also completed the Barcelona Music
Reward Questionnaire (Mas-Herrero et al., 2013), which consists
of five facets of possible rewards from music listening: emotion evo-
cation (HAP emotion), mood regulation (low-arousal affect regula-
tory functions), social reward (prosocial functions), sensory motor
(dance-tendencies), and musical seeking (music enjoyment).
Finally, participants provided other information pertaining to cul-
tural self-construal (Singelis, 1994) and demographics (age, gender,
and everyday music listening habits). Participants in Singapore also
completed questionnaires on social withdrawal and subjective well-
being for a different research project.
Results and Discussion
Descriptives and full results are in the OSF repository. Using a “two
one-sided test”of equivalence with α=.05, and equivalence bounds
(smallest effect size of interest) of Cohen’sd=0.3, we noted no sig-
nificant differences in arousing scores between the “relaxing”and
“studying”contexts, t(534) =−1.39, p=.166, and that the 90% CI
was fully contained within the equivalence interval—upper: t(534) =
−4.9, p,.001; lower: t(534) =2.1, p=.019, suggesting that the
arousing scores between the “relax”and “studying”context were
not significantly different and statistically equivalent. Accordingly,
for the rest of the analyses, these scores were averaged for a single
“arousing”score that represented both contexts.
One-way analyses of variance (ANOVAs) were conducted to
examine cultural differences for arousal, music reward, and indepen-
dence/interdependence. A significant difference was found for
arousing, F(1, 236) =5.53, p=.019, ω
2
=0.017, and sensory
motor, F(1, 266) =20.6, p,.001, ω
2
=0.068 with post hoc tests
revealing significant differences (Tukey HSD) for sensory motor
(t(266) =−4.53, p,.001) and arousing (t(266) =−2.35,
p=.019). U.S. participants preferred dance functions of music
more than Singaporean participants, and they also preferred more
arousing music. Post hoc power analyses using G*Power (Faul
et al., 2007) reveal an observed power =0.64 ( f=0.14, α=.05) for
arousing and power =0.99 ( f=0.28, α=.05) for sensory motor.
No other significant differences were observed. This supports our
first hypothesis that U.S. participants preferred high-arousal music
and dance functions.
With culture, age, and gender as controls, we examined the effect of
independence on music emotion and function. Again, we report only
significant effects, but the full results are available in our OSF repos-
itory. Independent self-construal significantly predicted arousing
(b=0.09, SE =0.05, 95% CI [0.001, 0.18], t=2.00, p=.047);
social reward (b=0.03, SE =0.02, [0.003, 0.07], t=2.14,
p=.033); sensory motor (b=0.07, SE =0.02, [0.04, 0.11], t=
4.08, p,.001); musical seeking (b=0.03, SE =0.02, [0.001,
0.06], t=2.06, p=.041); and mood regulation (b=0.03, SE =
0.01, [0.003, 0.06], t=2.21, p=.028). In short, across cultures, inde-
pendent individuals preferred more arousing music, and were more
likely to listen to music for social reasons, dance, enjoyment, and
mood regulation. Interestingly, listening to music for intense emo-
tional experiences (emotion evocation) was not more prevalent in
the United States than Singapore, nor higher in independently oriented
(than interdependently oriented) individuals, suggesting that musical
HAP upregulation may not significantly reflect cultural differences for
HAP emotions. Also, listening to music for low-arousal mood regula-
tion purposes was not significantly stronger in Singapore than the
United States, which suggests that low-arousal listening may notdiffer
between cultures as strongly as high-arousal listening.
Independent orientation appeared to predict music preference in
the direction that we initially anticipated (e.g., arousal; sensory
motor/dance). Yet, despite known cultural differences in self-
construal (e.g., Wee et al., 2021), the United States did not score
higher than Singapore on independence, possibly due to the refer-
ence group effect (Heine et al., 2002): this underrepresents cultural-
level differences in independence and interdependence when mea-
sured at the individual level. Moreover, self-construal showed poor
measurement invariance across cultural groups (see the online sup-
plemental material in our OSF repository). Accordingly, we needed
additional measures to examine if independence, at a macro/cultural
level, would show a relationship with danceability across countries.
Study 4
A macro-level cross-cultural comparison study was conducted to
address the following concerns: Firstly, we elucidate relationships
5
We did not use the Serene and Melancholy factors, as given our interest in
arousal, the individual emotion terms in the Serene factor showed differing
arousal patterns (e.g., inspiring-majestic and romantic descriptors may
imply high-arousal emotional experiences), and Melancholy was addition-
ally confounded by a clear negative valence.
LIEW ET AL.
6
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
between danceability, affect, and independent orientation, that were
found in Study 3. Studies 1–3 relied on dichotomous East–West cul-
tural comparisons, that are not sufficiently representative of global
variation in cultural contexts and values (Triandis, 1995).
Furthermore, effects in Study 3 were observed at a participant level,
which may not necessarily generalize to cultural-level effects. A
macro-level cross-cultural study was thus needed to investigate if
country-level relationships on music preferences support our hypoth-
eses on danceability as reflective of cultural affordance for emotion
and independence (from Study 3). To measure cultural variation in
independence, we used two commonly used indices that encompass
independent self-construal at the country level: individualism–collec-
tivism and tightness–looseness. Tightness–looseness measures the
strength of social norms, and residents of “tight”cultures often have
higher impulse control and cautiousness than “loose”cultures
(Harrington & Gelfand, 2014) that overlap with interdependent orien-
tations. Similarly, collectivistic societies value social aspects of soci-
ety, like family integrity, that encompass interdependent orientations
(Triandis & Gelfand, 1998).
Secondly, we model cultural differences at a macro level to test for
alternative explanations. For example, rather than our hypothesized
affect regulation and independence explanations, relationships
between culture and danceability could also be due to structural
issues, such as socioeconomic status (SES). Societies with lower
SESs have generally narrower music preferences (Chan &
Goldthorpe, 2006). One possibility could be through limited channels
of access to global or new music, which may in turn drive preference
for homegrown, localized music. Accordingly, cultures with wider
income gaps (income inequality) may show more diverse preferences,
which may manifest in danceabilityscores. Geopolitical and historical
reasons could also play a role: if certain cultures are historically more
linked toward dance music, these could spread within their spheres of
political or cultural influence, for example, through colonialization,
migration, or investment inflows and outflows. By including these
variables as covariates in the model, we can also test the robustness
of the danceability and affect hypotheses above and beyond these
alternative explanations.
Finally, we included additional measures of danceability with
open-source documentation. Without a publicly released computa-
tional breakdown of Spotify’s danceability, we are unable to ade-
quately define danceability, and future research targeting specific
aspects of danceability may have difficulty with interpreting, replicat-
ing, and reproducing this line of research. As such, we included two
other measures of danceability in this study. These were based on
open-sourced danceability models from the Essentia (Bogdanov
et al., 2013) audio-analysis library, that were executed on the audio
files for all songs in the data set. We hope to demonstrate that these
findings are robust across measures of danceability, and to facilitate
future research on specific components of danceability and culture.
We used Top 50 playlist information from 60 countries that
Spotify operated in to obtain scores for danceability, and combined
these song-level data with country-level data. Our priority was to
establish a link between individualism, positive and negative affect,
and danceability at a country level. Subsequently, we examined the
robustness of these relationships amid the inclusion of covariates for
colonialization, trade, migration, political structure, and income
inequality. We finally examined the robustness of the danceability
construct, by comparing Spotify’s scores for danceability with
scores from two other open-sourced models of danceability.
Measures
Scores for danceability were obtained from 60 countries that had
access to Spotify (at the point of data collection in December 2018).
As with Study 2, Top 50 lists for each country were used to obtain
songs for analysis. Wecombined these song-level data with country-
level data for log-transformed GDP (gross domestic product) per
capita, NE experience, and individualism. Scores for individualism
were obtained from Hofstede’s Cultural Insights (Hofstede et al.,
2010), and NE and positive emotion (PE) experience, and GDP
were obtained from the World Happiness Report 2019 (Helliwell
et al., 2019). NE experience scores were the sample prevalence of
the presence of NE experiences (averaged score of anger, worry,
and sadness) experienced by participants on the preceding day.
This was similar to positive emotional experience for happiness,
laughter, and enjoyment. GDP was obtained from log-transformed
2018 or 2017 GDP per capita where applicable. We also obtained
scores for migration (immigrants as a percentage of the population)
in 2015, Gini coefficients (for income inequality) from 2016, 2017,
or 2018 where applicable, foreign direct investment (FDI) inflows
and outflows (as a percentage of GDP) in 2019, and various gover-
nance indicators (rule of law, political stability, regulatory quality,
accountability, and corruption control) from the World Bank
Database. Domain-general scores for cultural tightness–looseness
were obtained from Uz (2015), and the colonial history of a country
was obtained from the International Correlates of War (ICOW) data
set (Hensel & Mitchell, 2007).
For additional estimations of danceability, we used the dance-
ability (Streich & Herrera, 2005) and danceability-musicnn-mtt
(Correya et al., 2021) models from the Essentia library. The former
estimates danceability through detrended fluctuation analysis
(DFA), a low-level acoustic/audio feature that tracks the stability
and strength of the beat of a song. The latter uses predictions
from a convolutional neural network (CNN) model, trained ini-
tially through “danceable”tags from the MagnaTagATune data
set (Law et al., 2009;seePons & Serra, 2019). These are high-level
probability scores, that are predicted by a deep learning model pre-
viously trained on human annotations of “danceable”music, and
may not necessarily map on to low-level acoustic features of
music. These estimates were applied on the raw audio files for
songs belonging to the Top 50 playlists. We used string matching
(Levenshtein’s distance) to match the Spotify artist names and song
titles, with recorded artist names and song titles from the Deezer
music catalog, and manually confirmed the match. Scores for dan-
ceability (DFA danceability) and danceability-musicnn-mtt (CNN
danceability) were computed and provided by Deezer with the
Essentia models. All country-level data are publicly accessible in
our OSF repository.
Results
Danceability Correlations
We first examined correlations between Spotify’s and Essentia’s
versions of danceability at the song level. After removing duplicate
songs, we counted N=945 unique songs. Spotify’s danceability
was positive and significantly correlated with both CNN danceability,
r=.62, 95% CI [0.58, 0.65], p,.001, and DFA danceability,
r=.51, [0.46, 0.55], p,.001, and CNN danceability and DFA
DANCEABILITY AND CULTURAL AFFECT 7
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
danceability were also similarly correlated, r=.52, [0.47, 0.56],
p,.001.
Affect, Individualism, and Tightness
Spotify Danceability. We examined these three effects in a
series of random-intercept mixed regression models. Confidence
intervals were calculated using Wald’s method, and degrees of free-
dom were estimated using Sattertwaith’s method in jamovi (The
Jamovi Project, 2020). Note that, as the number of countries
included in each regression analysis differed according to the data
set we used, we were generally unable to compare differences
between models through commonly used fit indices. For this reason,
we were conservative about the inclusion of variables in each model,
to maximize the sample size of the data set. For affect, we fitted a
model predicting Spotify’s danceability at the song-level (see
Table 2: Model S1) and country-level predictors of PE and NE expe-
rience, and controls for log-transformed GDP per capita (country-
level SES). With N=61 countries, a restricted maximum likelihood
(REML) model (R
2
Marginal
=0.072, R
2
Conditional
=0.153) showed a
significant effect of NE, but not PE nor GDP per capita. We then fit-
ted a second model (Table 2: Model S2) that included an addition of
Hofstede’s country-level scores for individualism to the model
above. With N=60 countries, a REML model (R
2
Marginal
=0.073,
R
2
Conditional
=0.152) showed a significant effect of NE, but not indi-
vidualism, PE, or GDP per capita.
We then fitted a separate model (Table 2: Model S3) using
country-level scores for tightness–looseness (Uz, 2015), that also
reflect independence-interdependent social norms, and corre-
sponding emotional expression for PE and NE (Liu et al., 2018).
NE, PE, and GDP per capita were also included as covariates.
Using Uz’s (2015) domain-general estimations of tightness–loose-
ness from N=34 countries that had Spotify data available (from
our data set), a REML model (R
2
Marginal
=0.064, R
2
Conditional
=
0.117) showed a significant effect of tightness–looseness, and
NE. No significant effects were observed for PE, and GDP per
capita.
Essentia Danceability. We then replicated the affect and
tightness–looseness models with DFA danceability and CNN dance-
ability. For DFA danceability, the affect (NE, PE, and GDP per cap-
ita) model (Table 2: D1; N=61 countries; R
2
Marginal
=0.122,
R
2
Conditional
=0.264) revealed a significant effect of NE, PE, and
GDP per capita. However, with the inclusion of tightness–looseness
(Table 2: D2), no significant effects were found (N=34 countries;
R
2
Marginal
=0.037, R
2
Conditional
=0.174) for tightness–looseness,
NE, PE, and GDP per capita.
For CNN danceability, the affect (NE, PE, and GDP per capita)
model (Table 2:C1;N=61 countries; R
2
Marginal
=0.078,
R
2
Conditional
=0.203) revealed a significant effect of NE, but not PE
or GDP per capita. With the inclusion of tightness–looseness,
(Table 2:C2;N=34 countries; R
2
Marginal
=0.037, R
2
Conditional
=
0.174) we found similar results to the original Spotify measure:
CNN danceability was significantly predicted by tightness–looseness,
and NE. No significant effect was observed for PE and GDP per capita.
In sum, we found a robust effect of NE: the higher a country’sNE
score, the higher the danceability of its Top 50 popular songs. Even
when controlling for tightness–looseness and individualism across dif-
ferent estimation methods for danceability, this relationship was largely
observed. In contrast, PE showed only localized effects in specific
Table 2
Significant Effects for Regression Models Predicting Danceability
Danceability Model Terms bSE
95% CI
df t pLL UL
Spotify S1
NE 0.41 0.10 0.21 0.60 57.0 4.10 ,.001
PE 0.13 0.07 −0.002 0.26 57.0 1.93 .058
GDP −0.02 0.01 −0.04 0.0008 56.8 −1.88 .066
S2^
NE 0.41 0.10 0.23 0.60 57.1 4.10 ,.001
Individualism −0.00005 0.0003 −0.001 0.0001 54.8 −1.45 .154
S3^
NE 0.26 0.11 0.05 0.48 29.1 2.42 .022
Tightness–looseness 0.001 0.003 0.0005 0.002 28.9 4.10 ,.001
Essentia—DFA D1
NE 0.39 0.14 0.11 0.67 56.9 2.75 .008
PE 0.27 0.09 0.09 0.46 56.9 2.93 .005
GDP −−−−−0.04 0.01 −−−−−0.07 −−−−−0.01 56.8 −−−−−2.94 .005
D2^
Tightness–looseness 0.0006 0.0004 −0.0001 0.001 28.9 1.49 .147
Essentia—CNN C1
NE 0.96 0.26 0.44 1.47 57.1 3.63 ,.001
PE 0.33 0.17 −0.01 0.67 57.0 1.93 .063
GDP −0.04 0.03 −0.09 −0.01 56.9 −1.45 .151
S3^
NE 0.74 0.29 0.17 1.31 29.1 2.54 .017
Tightness–looseness 0.003 0.0007 0.003 0.004 29.0 4.12 ,.001
Note. Covariates are only reported for base models, and only significant effects and new terms are reported for subsequent models (marked by a “^”), but
outputs of full models are available in our OSF Repository. Bold values indicate statistical significance, p,.05. CI =confidence interval; LL =lower limit;
UL =upper limit; NE =negative emotion; PE =negative emotion; GDP =gross domestic product; DFA =detrended fluctuation analysis; CNN =
convolutional neural network.
LIEW ET AL.
8
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
conditions. We also found that tightness–looseness, and not individual-
ism, predicted danceability across cultures: countries with higher dance-
ability scores in their Top 50 songs also had looser social norms.
Additional Covariates
Income Inequality. We added the Gini coefficient (at a coun-
try level) into the model, alongside covariates for NE, PE, GDP per
capita, and tightness–looseness. In the model (Table 3: II-S; N=31
countries; R
2
Marginal
=0.09, R
2
Conditional
=0.116), the Gini index and
tightness–looseness, showed significant effects in predicting dance-
ability, but no significant effects were observed for NE, PE, and
GDP per capita. Using the DFA danceability measure, the model
(Table 3:II-D;N=31, R
2
Marginal
=0.078, R
2
Condit ional
=0.180)
showed a significant effect of income inequality, but no significant
effects were observed for NE, PE, tightness–looseness, and GDP
per capita. Finally using the CNN danceability measure, the model
(N=31 countries, R
2
Marginal
=0.117, R
2
Condit ional
=0.174) resembled
the Spotify danceability model, in showing a significant effect of
income inequality, and tightness–looseness, but no significant effects
were observed for NE, PE, and GDP per capita.
In sum, countries with greater income inequality have the Top 50
songs with higher danceability scores. Depending on the measure
of danceability, despite the inclusion of the Gini index into the
regression model, tightness–looseness still significantly predicted
danceability. However, NE no longer significantly predicted dance-
ability across all three estimates. Our interpretation is that countries
with high-income inequality would also have greater affordance
for NEs (country level; Pearson’sr=.53, 95% CI [0.28, 0.71],
p,.001), which can be reflected through greater consumption of
danceable music (Table 3).
Colonial History, Migration, FDI Inflows/Outflows,
and Governance. Next, we examined if the cultural variation
in danceability could also be predicted by geopolitical and historical
factors. First, we examined colonial history: countries may be influ-
enced by the lasting values, trends, and esthetics of their colonial
past, where countries with similar colonial histories may share pref-
erences for certain types of music, that may then be reflected in dan-
ceability preferences. As there were 19 levels for this variable, we
report only the significant effects in this section. The full results
are available in our OSF repository.
Overall, the model fitted to Spotify danceability scores (N=57
countries; R
2
Marginal
=0.140, R
2
Conditional
=0.171) with covariates
for PE, NE, and GDP per capita (tightness–looseness was removed
from the model due to its reduced sample size of 31 countries),
revealed a significant effect of NE. No significant effects of PE
and GDP per culture were observed. As colonial history was devia-
tion coded, we observed a significant effect of colonizer for Brazil,
Central America, Colombia, Czechoslovakia, Haiti, Russia, Spain,
Sweden, and the United States. In sum, compared to the mean across
all levels, the Top 50 songs from countries that were historically col-
onized by Brazil, Central America, Colombia, Haiti, and Spain had
higher danceability, and the Top 50 songs from countries colonized
by Czechoslovakia, Russia, Sweden, and the United States had
lower danceability. Countries with no colonization history also
had significantly lower danceability than the average. Even after
controlling for this effect, NE significantly predicted danceability
(Table 4). Due to space constraints in the manuscript, results from
the DFA and CNN danceability models are in our OSF repository.
We also examined the proportion of migrants within a population,
as well as FDI inflows and outflows of a country, relative to GDP.
GDP per capita, NE, and PE were included in the model with
Spotify danceability scores (N=57 countries; R
2
Marginal
=0.076,
R
2
Conditional
=0.164). Aside from NE (b=0.37, SE =0.11, 95%
CI [0.14, 0.59], t(50.0) =3.24, p=.002), no other significant
effects were observed. We then fitted a model with governance indi-
cators (rule of law, political stability, regulatory quality, accountabil-
ity, and corruption control) with GDP per capita, NE, and PE, but
only NE significantly predicted danceability (b=0.28, SE =0.13,
Table 3
Effects for Income Inequality in Predicting Danceability
Danceability Model Terms bSE
95% CI
df t pLL UL
Spotify II-S
NE 0.11 0.10 −0.07 0.30 24.9 1.18 .247
PE 0.05 0.06 −0.06 0.16 25.0 0.91 .371
GDP 0.01 0.02 −0.01 0.05 24,9 1.05 .303
Tightness–looseness 0.0009 0.0002 0.0004 0.001 24.8 3.44 .002
Gini 0.006 0.001 0.003 0.008 24.7 4.09 ,.001
Essentia—DFA II-D
NE 0.05 0.16 −0.26 0.36 25.0 0.32 .748
PE 0.14 0.09 −0.04 0.33 25.0 1.53 .140
GDP 0.01 0.03 −0.03 0.07 25.0 0.65 .519
Tightness–looseness 0.0004 0.0004 −0.0004 0.001 25.0 1.06 .297
Gini 0.007 0.002 0.002 0.01 24.9 2.65 .007
Essentia—CNN II-C
NE 0.38 0.27 −0.16 0.92 25.0 1.39 .176
PE 0.02 0.16 −0.29 0.34 25.0 0.15 .881
GDP 0.07 0.04 −0.02 0.15 25.0 1.53 .142
Tightness–looseness 0.002 0.0007 0.0009 0.004 24.9 3.25 .003
Gini 0.01 0.004 0.006 0.02 24.8 3.46 .002
Note. Bold values indicate statistical significance, p,.05. CI =confidence interval; LL =lower limit; UL =upper limit; NE =negative emotion; PE =
negative emotion; GDP =gross domestic product; DFA =detrended fluctuation analysis; CNN =convolutional neural network.
DANCEABILITY AND CULTURAL AFFECT 9
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
[0.03, 0.54], t(48.8) =2.15, p=.037). Full results are reported in
the OSF repository.
Discussion
Our results showed a strong and stable effect of NE experience in
predicting danceability of Top 50 charts across cultures. Even when
controls were included for cultural differences in values (individual-
ism–collectivism), social norms (tightness–looseness), migration,
foreign investment, and geopolitical (colonial) history, results
showed that higher prevalence of NEs on a cultural level predicted
the danceability of songs on their Top 50 charts.
6, 7
While associa-
tion with NEs was not significant after the inclusion of income
inequality, this may still be consistent in that income inequal socie-
ties also have higher prevalence of NEs.
Danceability and HAN Affect
This is further contextualized by the significant relationships
between danceability and tightness–looseness (which was also
robust across analyses and evident in both Spotify’s and Essentia’s
CNN danceability estimates). Looser cultures, where social norms
are less strict and displays of emotion more encouraged, prefer
music with higher danceability. As dancing is often seen as an out-
ward expression of one’s emotional states (Schwender et al., 2018),
music that can facilitate dancing (high danceability) would naturally
be more popular in looser cultures than tighter cultures.
Our results also showed that cultures with higher prevalence of
income inequality have increased danceability. Moreover, when
income inequality was included in the model, the variance explained
by NE experience was drastically reduced. Past research has shown
that low SES creates the environment for increased NEs (Gallo &
Matthews, 2003) and that income inequality is itself also associated
with more NEs (Godoy et al., 2006). Accordingly, in our data, we
found a similar correlation between NE experience and income
inequality. Countries with higher income inequality experience
more prevalent negative affect, so the corresponding relationship
with danceability scores may consistently be reflective of a functional
usage of music listening to downregulate these HAN emotions.
Here, our hypothesis is that danceability may be representative of
tendencies to downregulate HAN affect. Societies with high-income
inequality, and/or societies with loose social norms, may have more
prevalent experiences of HAN emotions that result in the use of
danceable music as a downregulation strategy. Yet, what role does
danceability play in HAN downregulation? Given that dance is a
high-arousal activity (Bernardi et al., 2017), and that danceability
may quantify the level of arousal in a song, we propose that cathartic
music listening for discharge regulation of HAN emotions could be
the missing link, which we examine in Study 5.
Study 5
We theorized that individual in cultures where HAN emotions are
more widely experienced and expressed could be cathartically listen-
ing to high-arousal, danceable music to downregulate these emotions,
by matching the arousal of the listener with the arousal of the music
(Sharman & Dingle, 2015). Danceability could thus reflect the
HAN emotion prevalence within a culture, by representing its cultural
tendencies toward discharge listening in music to downregulate it. If
Table 4
Significant Effects for Colonial History Predicting Spotify’s Danceability
Terms Coloniser bSE
95% CI
df t PLL UL
NE 0.27 0.10 −−−−−0.07 0.47 34.0 2.66 .012
PE −0.10 0.07 −0.24 0.04 33.8 −1.35 .185
GDP 0.01 0.01 −0.01 0.03 33.9 0.97 .338
Colonial history
Brazil 0.001 0.01 0.01 0.14 34.1 2.35 .025
Central Am 0.09 0.02 0.05 0.13 33.8 4.16 ,.001
Colombia 0.08 0.02 0.03 0.12 34.1 3.12 .004
Czechoslovakia −−−−−0.07 0.02 −−−−−0.12 −−−−−0.03 33.9 −−−−−3.09 .004
Haiti 0.08 0.03 0.02 0.15 34.1 2.72 .010
Russia −−−−−0.04 0.01 −−−−−005 −−−−−0.005 34.3 2.39 .040
Spain 0.07 0.01 0.04 0.10 34.0 5.04 ,.001
Sweden −−−−−0.11 0.03 −−−−−0.17 −−−−−0.05 34.1 −−−−−3.52 .001
United States −
−−−−0.08 0.03 −−−−−0.14 −−−−−0.008 34.1 −−−−−2.19 .035
None −−−−−0.03 0.01 −−−−−0.05 −−−−−0.005 34.3 −−−−−2.39 .023
Note. Full model outputs (alongside DFA and CNN danceability) are available in our OSF repository. Note that colonial history was deviation coded,
comparing the mean of each level against the mean of all levels. Bold values indicate statistical significance, p,.05. CI =confidence interval; LL =lower
limit; UL =upper limit; NE =negative emotion; PE =negative emotion; GDP =gross domestic product; DFA =detrended fluctuation analysis; CNN =
convolutional neural network.
6
This was most evident in Spotify’s danceability and Essentia’s CNN dan-
ceability, but the comparatively lower-level DFA danceability scores were
only significantly replicated in the base (positive and negative affect and
GDP per Capita) model. One possibility could be that danceability, as pre-
dicted by machine learning algorithms trained on human annotations, was
more sensitive and reflective of danceability as a construct. By contrast,
DFA may not be as sensitive to smaller changes and minute fluctuations in
danceability (Streich & Herrera, 2005).
7
As data were collected in 2019, we updated it with a newer round of data
collection following the same analysis scripts in 2022. Here, NE continued to
predict danceability in the base affect models, but not in the subsequent mod-
els. Full results are available in our OSF repository. This may arise from our
covariate data being outdated (i.e., several variables, like GDP and NE were
obtained from 2019, despite playlists (songs) from 2022, and more research is
needed to replicate and extend the findings.
LIEW ET AL.
10
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
so, cultures with more HAN emotion prevalence would thus engage
more with discharge regulatory strategies in music listening. By con-
trast, cultures that do not frequently experience HAN emotions may
not need discharge regulation, but can make do with solace or distract
regulation methods that are better suited to downregulate low-arousal
negative (LAN) states (Saarikallio, 2012).
Specifically, we hypothesized that discharge regulation would be
higher in looser cultures, and associated with higher HAN emotion
prevalence: HAN prevalence should mediate the relationship
between culture and the usage of discharge regulation in music lis-
tening. To examine this hypothesis, we examined these tendencies
at an individual participant level, and conducted a survey of N=
227 adults from Japan (tight culture) and the United States (loose
culture), and examined their self-reported experiences of emotions,
and emotion regulation strategies (including discharge regulation)
from music.
Procedure
An online questionnaire was administered to participants from
Japan (N=116; M
age
=43.5, SD =8.8, Females =34, Males =81,
Rather not say =1) and the United States (N=111, M
age
=33.5,
SD =12.9, Females =52, Males =53, Others =4, Rather not say =2),
recruited from the Lancers (Japan: https://lancers.jp) and Prolific
(United States: https://prolific.co) crowdsourcing platforms for
¥500 and $10.00, respectively. Participants completed the Affect
Valuation Index (AVI; Tsai et al., 2007), a self-report measure that
examines participants’ideal or desired emotions, and more rele-
vantly, the actual emotions experienced over the course of a week.
These were broadly grouped into four categories: HAN (e.g., fearful,
hostile), HAP (e.g., enthusiastic, excited), LAN (dull, sluggish), and
LAP (rested, calm), across “ideal”and “actual”emotion experi-
ences. Next the Brief Music in Mood Regulation Scale (B-MMR;
Saarikallio, 2012) was administered. We focused mainly on the dis-
charge regulation subscale (e.g., “When I’m angry with someone, I
listen to music that expresses that anger”), but additional analyses
were also conducted on the diversion (e.g., “For me, music is a
way to forget about my worries”) and solace (e.g., “When everything
feels bad, music understands and comforts me”) subscales that com-
prised strategies toward NE downregulation. Participants provided
demographic information, such as age, gender, music experience,
SES, and ethnicity, and was approved by the respective
Institutional Review Boards of Stanford University and Kyoto
University for data collection in the United States and Japan. As
the survey was administered in Japanese in Japan, we used the
Japanese versions of the AVI and B-MMR (Shoda et al., 2019).
Demographic questions were translated into Japanese, and then
back-translated into English for checking.
Results and Discussion
After controlling for age, gender, and country, a linear regression
model (R
2
=0.28) showed that HAN-actual significantly predicted
discharge regulation, where higher HAN-actual scores were associ-
ated with higher discharge regulation scores. Country was also sig-
nificant in predicting discharge regulation (Japan as reference),
where discharge regulation was stronger in the United States than
in Japan. To control for possible effects of arousal and valence,
we conducted a separate analysis with all four emotion terms
(HAP, HAN, LAP, LAN) in the “actual”context. With controls
for age, gender, and country, a linear regression model (R
2
=
0.29) showed than HAN-actual was still predictive of discharge reg-
ulation, as with country, in similar directions. Furthermore,
HAP-actual also significantly predicted discharge regulation, albeit
to a smaller extent than HAN-actual. Post hoc power analyses
using G*Power (Faul et al., 2007) reveal an observed power =
0.99 ( f=0.39, α=.05) for the HAN-actual model and power =
0.99 ( f=0.41, α=.05) for the full (HAP, HAN, LAP, LAN) model.
Next, we examined “ideal”emotion scores. With the same con-
trols for age, gender, and country, a linear regression model (R
2
=
0.24), HAN-ideal significantly predicted discharge regulation.
However, when all four ideal emotions (HAP, HAN, LAP, LAN)
were included in the model (R
2
=0.23), no significant effects
were observed for any of these terms. Regression results are summa-
rized in Table 5. Post hoc power analyses reveal an observed power
=0.99 ( f=0.32, α=.05) for the HAN-ideal model.
Finally, we examined a mediation model involving country,
HAN-actual, and discharge regulation. Culture was dummy coded
as Japan =0 and United States =1. All standard errors and confi-
dence intervals were calculated on 1,000 bootstrapped iterations.
The direct effect of country and discharge regulation was significant
(b=2.53, SE =0.47, 96% CI [1.62, 3.48], B=0.33, Z=5.42,
p,.001), and was partially mediated by HAN-actual: countrysignif-
icantly predicted HAN-actual (b=0.24, SE =0.10, 95% CI [0.05,
0.43], B=0.15, Z=2.43, p=.015), and HAN-actual significantly
predicted discharge regulation (b=1.53, SE =0.31, [0.91, 2.12],
B=0.29, Z=4.94, p,.001), for a significant indirect effect of
country and HAN-actual on discharge regulation(b=0.36, SE =
0.17, [0.03, 0.69], B=0.05, Z=2.13, p=.033). This suggests a
significant partial mediation effect of HAN-actual on the relation-
ship between culture and discharge regulation. Participants in the
United States were more likely to use music for discharge regulation
purposes, and this effect was mediated by the amount of HAN emo-
tions experienced subjectively over the past week. Full results
(including models fitted to diversion and solace regulation) are avail-
able in our OSF repository.
Discharge regulation in music was significantly linked with partic-
ipants’HAN emotion experiences, and this effect was consistent with
participants’countriesof origin: participants from the United States, a
culture with looser social norms (Tightness–Looseness domain-
general scores from Uz, 2015: 71.5) and higher income inequality
than Japan (Tightness–Looseness: 39.2), were also more likely to
use discharge regulation strategies. Moreover, we noticed that this
effect was also true for ideal or desired HAN emotion experiences,
in that participants with higher desired HAN emotions were more
likely to use music for discharge regulation purposes. These are con-
sistent with our theory that individuals from cultures with higher affor-
dances for HAN emotions engage with music to downregulate these
emotions, as affordances for HAN emotions include both the circum-
stances to induce these emotions in daily life, and the social accep-
tance for experiencing and expressing them (“actual”and “ideal”
HAN, respectively). However, we also noted a smaller but significant
effect of HAP-actual in predicting discharge regulation. This could be
a consequence of the United States being more accepting of high-
arousal emotions in general, but we hesitate on speculating on a
link, given that Studies 3 and 4 did not show a consistent effect of
PE experiences on music listening across cultures. Additionally, in
comparing results with other regulatory styles for NE downregulation
DANCEABILITY AND CULTURAL AFFECT 11
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
through music (solace and distraction regulation: see OSF repository),
we noticed that discharge regulation was the only regulatory style
where culture exerted a significant effect, and that discharge regula-
tion was also most strongly associated with HAN emotional experi-
ences (no significant effect was observed for distract regulation, and
a considerably smaller effect was observed for solace regulation).
As music preferences emerge from these individual tendencies, this
may explain the cultural-level effects observed in music consumption
andNEinStudy4(Table 6).
General Discussion
Overall, these studies show that cultures differ in preferences for
music, and that the danceability of a culture’s Top 50 songs reflects
the amount of negative affect experienced by a culture, by potentially
representing the usage of discharge regulatory strategies of music
listening.
Musical Features as Representations of Culture
We highlight the usefulness of cultural product analyses for
bottom-up, cross-cultural research. Like other cultural products,
music (features) provides a more objective measure of population
behavior with less interference from survey and response biases
(see Kemmelmeier, 2016). With the advancement of big data and
computing technologies, the process for obtaining such physical rep-
resentations of culture has become simpler and faster. Music offers
an advantage in that it is closely linked to human affect (see Dunbar,
2012), yet offers features that are almost universally perceived and
understood. Beginning with bottom-up methods (Studies 1 and 2),
we were ultimately able to infer culturally based explanations and
develop hypotheses for the relationship between cultural preferences
for emotion and music based on danceability (Studies 3–5). Through
this process, we demonstrate the usefulness of music features, specif-
ically danceability, in representing sociocultural norms and values.
Danceability and Affect Regulation
For the individual, discharge regulation appears linked to subjec-
tive experiences of HAN emotions in daily life. Given that dance-
ability may signify the arousal levels of a song, we posit that
danceability likely indicates the overall prevalence of discharge reg-
ulatory strategies of music listening from a culture. Here, we discuss
this theory in terms of its arousal and valence components. For
arousal, our findings showed that loose cultures have greater cultural
affordances for high-arousal emotions: Study 4 showed that cultures
with more loose norms have more prevalent experiences of high-
arousal emotion. Dance itself is often a high-arousal activity, and
(looser) regions, where dancing is commonplace may also prefer
more danceable music. Accordingly, we found consistent results
in Study 3, where dance functions of music listening were higher
in the United States (loose) than Singapore (tight), and significantly
associated with independent self-construal.
Yet, in the same study, danceable music did not appear to be a
strategy used by independently oriented participants or U.S. partic-
ipants to upregulate HAP emotions. In considering valence, we
Table 5
Effects for Base Models (HAN-Actual and HAN-Ideal), Actual Model (HAN-Actual, HAP-Actual, LAN-Actual, LAP-Actual), and Ideal Model
(HAN-Ideal, HAP-Ideal, LAN-Ideal, LAP-Ideal) in Predicting Discharge Regulation
Model Terms bBSE
95% CI
tpLL UL
Base-actual
Country (US-JP) 1.80 0.46 0.51 0.80 2.81 3.53 ,.001
Age −−−−−0.06 −−−−−0.18 0.02 −−−−−0.10 −−−−−0.02 −−−−−2.89 .004
Gender (F–M) 0.05 0.01 0.48 −0.89 1.00 0.11 .912
HAN-actual 1.51 0.29 0.31 0.91 2.11 4.93 ,.001
Actual
Country (US-JP) 1.21 0.31 0.57 0.08 2.34 2.10 .037
LAN-actual −0.10 −0.02 0.36 −0.81 0.62 −0.27 .788
HAP-actual 0.87 0.16 0.42 0.05 1.70 2.10 .037
LAP-actual −0.62 −0.12 0.37 −1.36 0.12 −1.66 .099
HAN-actual 1.37 0.26 0.38 0.62 2.11 3.26 ,.001
Base-Ideal
Country (US-JP) 1.97 0.50 0.53 0.94 3.01 3.75 ,.001
Age −−−−−0.07 −−−−−0.23 0.02 −0.12 −−−−−0.03 −−−−−3.51 ,.001
Gender (F–M) 0.31 0.08 0.50 −0.67 1.29 0.62 .536
HAN-ideal 1.53 0.20 0.46 0.63 2.44 3.34 ,.001
Ideal
Country (US-JP) 1.86 0.47 0.59 0.69 3.03 3.15 .002
LAN-ideal 0.36 0.05 0.54 −0.70 1.43 0.67 .505
HAP-ideal 0.05 0.01 0.37 −0.68 0.78 0.14 .888
LAP-ideal 0.10 0.02 0.43 −0.76 0.95 0.22 .825
HAN-ideal 1.24 0.17 0.65 −0.03 2.52 1.92 .056
Note. While covariates (age and gender) are included in all models, they are only reported here for the base model. Full results (including models for solace and
distract regulation) are available in our OSF repository. Bold values indicate statistical significance, p,.05. CI =confidence interval; LL =lower limit; UL =
upper limit; F= female; M =male; HAN =high-arousal negative; HAP =high-arousal positive; LAN =low-arousal negative; LAP =low-arousal positive;
US-JP =United States–Japan.
LIEW ET AL.
12
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
noted a stronger effect of NE representation across Studies 3–5. As
mentioned earlier, Study 3 showed no effect on HAP emotion upre-
gulation. Study 4 showed a much weaker and inconsistent result of
danceability with PE experience across cultures, compared to a
clear and consistent relationship between NE experience and dance-
ability. Finally, Study 5 showed that the hypothesized mechanism of
discharge regulation in accounting for earlier effects, was stronger
and clearer in both ideal and actual (experienced) HAN emotions,
than ideal and actual (experienced) HAP emotions. In sum, these
studies converge to show that danceability differs across cultures,
in reflecting the societal prevalence of NEs, by indicating the extent
of discharge regulatory strategies of music listening within that
society.
This has implications for cross-cultural psychological research, in
presenting a new avenue for quantitative estimation of cultural dif-
ferences in emotion regulation, that combines the benefits of com-
puterized music information retrieval (e.g., reduced response bias)
for cultural emotion-index computations. As music is a product of
culture, automated danceability computations can be applied to esti-
mate the emotion regulation tendencies in subcultures or regions,
especially if combined with other psychosocial indicators (e.g.,
Gallup poll data, Google Trends). Additionally, we think that dance-
ability can also be used to track emotion regulation styles over time.
Given the availability of historical chart records and digitalized
music data, this approach may be useful for historical research into
emotional trends and patterns (see Muthukrishna et al., 2021), that
would be inaccessible with traditional (self-report) methods of emo-
tion assessment.
Differences in Music Produced Versus Consumed
Our article also examines differences in music produced and con-
sumed by a culture, in combining two approaches often used in cul-
tural products research. The music produced by a culture represents
the cumulative esthetics of the individual members within it: music
created by an artist embodies their esthetics and influences, and ana-
lyzing the aggregated “esthetics”of artists within a culture may
reveal traits and tendencies pertinent to members of that culture.
At the same time, music consumption (e.g., Spotify charts) reflects
the cumulative preferences of a society, and is susceptible to external
influences. For example, research has shown that familiarity affects
popularity (North & Hargreaves, 1995), so repeated playbacks of
music in commercials or entertainment programs may lead to that
music making the charts. Although, advertisers targeting a specific
culture may choose music belonging to styles that are popular within
that culture. Ranking metrics may also function as gatekeepers, in
that music that is considerably different from popular styles would
still not become popular. Music consumption may also be limited
by production; the accessibility of music within a specific cultural
group. In Study 4, we mentioned that low-SES groups may face dif-
ficulties with accessing music from different cultures, and the music
consumed may then from greater exposure to music produced by the
local society. Therefore, we conducted Studies 1 and 2 that adopted
these two different approaches, and found converging results: dance-
ability was identified as a feature that differed across cultures in both
music produced (Study 1) and consumed (Study 2), suggesting that
cultural preferences for danceability in music were generalizable
despite these constraints.
Limitations
Given the close replications between Spotify’s danceability and
Essentia’s CNN danceability in Study 4 (but not with Essentia’s
DFA danceability), this suggests that the type of danceability used
in this study may be closer to modeled human annotations (subjec-
tive evaluations) of music that suitably facilitates dance, and not so
much an acoustic (DFA), low-level measure of danceability. On one
hand, this could be problematic, as cultural biases (such as the anno-
tators’cultural background) may have influenced these notions of
danceability. On the other hand, these subjective perceptions may
still be more accurate in representing a universal notion of danceabil-
ity than the low-level DFA. Secondly, recent research has even
emerged to show that cultural differences in rhythmic feature recog-
nition and perception may exist (Jacoby et al., 2021), and this may
potentially bias definitions of danceability toward a more Western
conceptualization of danceability. Nevertheless, research has also
Table 6
A Summary of the Aims, Methods, and Main Findings of Each Study
Study Aims Method Main finding
1 Exploratory—classify Japanese/U.S. Songs
and explore features behind the model
GBDT classifier on Spotify data (as
cultural products)
High classification accuracy (i.e., strong differences), and
danceability and instrumentalness as strong, interpretable
predictors (features)
2 Exploratory—confirm cultural differences
in identified features
Comparison of East-West Top 50 Charts Danceability was significantly different between East Asian
and Western music charts
3Confirmatory–cultural differences in
arousal preferences in human participants
Survey on music preferences of
Singaporeans and Americans
Compared to Singaporeans, U.S. participants preferred
high-arousal music. However, independent orientation
predicted arousal preferences in both passive and
recreational contexts.
4Confirmatory–cultural differences in
danceability and NE experiences
60-country mixed effects modeling of
country-level emotion experience and
Top 50 Spotify danceability
Negative emotion experience (frequency) and tightness–
looseness significantly predicted Top 50 danceability
scores. We theorize an explanation through danceability in
music facilitating cross-cultural tendencies toward
discharge regulation of negative affect
5Confirmatory–cultural differences in
discharge regulation can be attributed to
differences in HAN emotions
Survey on affect valuation and experience
with discharge regulation in United
States and Japan
We show a link between negative affect, particularly HAN,
culture, and discharge regulation: greater HAN prevalence
is linked to increased discharge regulation in participants,
and is stronger in United States than Japan.
Note. GBDT =gradient boosted decision tree; NE =negative emotion; HAN =high-arousal negative.
DANCEABILITY AND CULTURAL AFFECT 13
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
argued for a certain universality in rhythmic perception, as most con-
sumers of music around the world (at least those that have access to
Spotify), may have been exposed to some form of Western music tra-
ditions (such as 12-tone equal-tempered scales). Furthermore, it is
unlikely that music traditions (that afford perceptual differences)
would evolve independently of cultural traditions (see culture-music
coevolution; Savage et al., 2021), so the structural and institutional
differences that shape society may also inadvertently shape both
music preference and perception simultaneously. A longitudinal or
experimental study would help unravel mechanisms used as such,
while clarifying some of the identified confounds (e.g., income
inequality) in influencing individuals’preferences for danceable
music.
Next, we did not account for measurement invariance in the scales
used across cultural groups for survey-based studies (Studies 3 and
5). While this was less of an issue for Study 5, we note that the scales
used in Study 3 showed poor invariance across groups (Singapore
and the United States). Results from measurement invariance tests
and reliability for these scales are available as Markdown files in
our OSF repository.
Finally, our research does not examine dance as an activity. This
limited the conclusions that were made on culture and emotion
expression and regulation: Does the society need to be approving
of dance itself for danceable music to effectively downregulate
HAN emotions, or can effective downregulation happen through a
music-processing mechanism (such as rhythmic entrainment: Trost
et al., 2017) independent of the cultural context?
Conclusion
We started out with exploratory analyses to examine how music
features could represent cultural differences in emotion, and identi-
fied danceability and its usefulness in indicating the cultural affor-
dances for HAN emotions in a society by quantifying the use of
discharge regulation strategies in music. Our research shows that
the frequency of experience of HAN emotions in a culture’spre-
ferred music can be reliably predicted by danceability in music,
and that HAN emotional experiences are associated with discharge
regulation. Going beyond danceability, we also show the usefulness
of examining digital trace data, such as music features on Spotify’s
database, as a convenient, robust, and powerful method to quantify
differences in psychological tendencies (and emotions) around the
world.
References
Balkwill, L.-L., & Thompson, W. F. (1999). A cross-cultural investigation of
the perception of emotion in music: Psychophysical and cultural cues.
Music Perception,17(1), 43–64. https://doi.org/10.2307/40285811
Balkwill, L.-L., Thompson, W. F., & Matsunaga, R. (2004). Recognition of
emotion in Japanese, Western, and Hindustani music by Japanese listeners.
Japanese Psychological Research,46(4), 337–349. https://doi.org/10
.1111/j.1468-5584.2004.00265.x
Bernardi, N. F., Bellemare-Pepin, A., & Peretz, I. (2017). Enhancement of
pleasure during spontaneous dance. Frontiers in Human Neuroscience,
11, Article 572. https://doi.org/10.3389/fnhum.2017.00572
Bertin-Mahieux, T., Ellis, D. P. W., Whitman, B., & Lamere, P. (2011,
October 24–28). The million song dataset [Conference session].
Proceedings of the 12th International Society Music Information
Retrieval Conference (ISMIR 2011), Miami, FL, United States (pp.
591–596). University of Miami.
Bogdanov, D., Wack, N., Gómez, E., Gulati, S., Herrera, P., Mayor, O., Roma,
G., Salamon, J., Zapata, J. R., & Serra, X. (2013, October 21–25).
ESSENTIA: An open-source library for sound and music analysis
[Conference session]. MM’13: Proceedings of the 21st ACM International
Conference on Multimedia, Barcelona, Spain (pp. 855–858). Association
for Computing Machinery. https://doi.org/10.1145/2502081.2502229
Chan, T. W., & Goldthorpe, J. H. (2006). Social stratification and cultural
consumption: Music in England. European Sociological Review,23(1),
1–19. https://doi.org/10.1093/esr/jcl016
Christensen, J. F., Gaigg, S. B., Gomila, A., Oke, P., & Calvo-Merino, B.
(2014). Enhancing emotional experiences to dance through music: The
role of valence and arousal in the cross-modal bias. Frontiers in Human
Neuroscience,8, Article 757. https://doi.org/10.3389/fnhum.2014.00757
Correya, A., Alonso-Jimenez, P., Marcos-Fernandez, J., Serra, X., &
Bogdanov, D. (2021, July 5–7). Essentia TensorFlow models for audio
and music processing on the web [Paper presentation]. Proceedings of
the Web Audio Conference (WAC 2021), Barcelona, Spain.
Cowen, A. S., Fang, X., Sauter, D., & Keltner, D. (2020). What music makes
us feel: At least 13 dimensions organize subjective experiences associated
with music across different cultures. Proceedings of the National Academy
of Sciences of the United States of America,117(4), 1924–1934. https://
doi.org/10.1073/pnas.1910704117
Cross, I. (2001). Music, cognition, culture, and evolution. Annals of the
New York Academy of Sciences,930(1), 28–42. https://doi.org/10.1111/j
.1749-6632.2001.tb05723.x
Dunbar, R. I. M. (2012). Bridging the bonding gap: The transition from pri-
mates to humans. Philosophical Transactions of the Royal Society B:
Biological Sciences,367(1597), 1837–1846. https://doi.org/10.1098/rstb
.2011.0217
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A
flexible statistical power analysis for the social, behavioral, and biomedical
sciences. Behavior Research Methods,39(2), 175–191. https://doi.org/10
.3758/BF03193146
Fricke, K., Greenberg, D., Rentfrow, P. J., & Herzberg, P. Y. (2018).
Computer-based music feature analysis mirrors human perception and can
be used to measure individual music preference. Journal of Research in
Personality,75,94–102. https://doi.org/10.1016/j.jrp.2018.06.004
Friedman, J. H. (2001). Greedy function approximation: A gradient boosting
machine. Annals of Statistics,29(5), 1189–1232. https://doi.org/10.1214/
aos/1013203451
Fritz, T., Jentschke, S., Gosselin, N., Sammler, D., Peretz, I., Turner, R.,
Friederici, A. D., & Koelsch, S. (2009). Universal recognition of three
basic emotions in music. Current Biology,19(7), 573–576. https://
doi.org/10.1016/j.cub.2009.02.058
Gallo, L. C., & Matthews, K. A. (2003). Understanding the association
between socioeconomic status and physical health: Do negative emotions
play a role? Psychological Bulletin,129(1), 10–51. https://doi.org/10
.1037/0033-2909.129.1.10
Godoy, R. A., Reyes-García, V., McDade, T., Huanca, T., Leonard, W. R.,
Tanner, S., & Vadez, V. (2006). Does village inequality in modern income
harm the psyche? Anger, fear, sadness, and alcohol consumption in a pre-
industrial society. Social Science and Medicine,63(2), 359–372. https://
doi.org/10.1016/j.socscimed.2006.01.021
Golder, S., & Macy, M. W. (2011). Diurnal and seasonal mood vary with
work, sleep, and daylength across diverse cultures. Science,333(6051),
1878–1881. https://doi.org/10.1126/science.1202775
Greenwell, B. M., Boehmke, B., Cunningham, J., & GBM Developers.
(2019). gbm: generalized boosted regression models (R package version
2.1.5) [Computer software]. R Foundation for Statistical Computing.
https://CRAN.R-project.org/package=gbm
Harrington, J. R., & Gelfand, M. J. (2014). Tightness–looseness acrossthe 50
United States. Proceedings of the National Academy of Sciences,111(22),
7990–7995. https://doi.org/10.1073/pnas.1317937111
LIEW ET AL.
14
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Heine, S. J., Lehman, D. R., Peng, K., & Greenholtz, J. (2002). What’s wrong
with cross-cultural comparisons of subjective Likert scales? The
reference-group effect. Journal of Personality and Social Psychology,
82(6), 903–918. https://doi.org/10.1037/0022-3514.82.6.903
Helliwell, J., Layard, R., & Sachs, J. (2019). World happiness report 2019.
Sustainable Development Solutions Network.
Hensel, P. R., & Mitchell, S. M. (2007). The issue correlates of war (ICOW)
project issue data set: Territorial claims data. Harvard Dataverse,34(2),
Article V2. https://doi.org/10.7910/DVN/E6PSGZ
Hofstede, G., Hofstede, G. J., & Minkov, M. (2010). Cultures and organiza-
tions: Software of the mind. Revised and expanded 3rd edition.
McGraw-Hill.
Jacoby, N., Polak, R., Grahn, J., Cameron, D., Lee, K. M., Godoy, R.,
Undurraga, E. A., Huanca, T., Thalwitzer, T., Doumbia, N., &
Goldberg, D. (2021, July 6). Universality and cross-cultural variation in
mental representations of music revealed by global comparison of rhythm
priors.https://doi.org/10.31234/osf.io/b879v
James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to
statistical learning. Springer.
Juslin, P. N., Barradas, G.T., Ovsiannikow, M., Limmo, J., & Thompson, W.
F. (2016). Prevalence of emotions, mechanisms, and motives in music lis-
tening: A comparison of individualist and collectivist cultures.
Psychomusicology: Music, Mind, and Brain,26(4), 293–326. https://
doi.org/10.1037/pmu0000161
Kemmelmeier, M. (2016). Cultural differences in survey responding: Issues
and insights in the study of response biases. International Journal of
Psychology,51(6), 439–444. https://doi.org/10.1002/ijop.12386
Kim, H., & Markus, H. R. (1999). Deviance or uniqueness, harmony or con-
formity? A cultural analysis. Journal of Personality and Social
Psychology,77(4), 785–800. https://doi.org/10.1037/0022-3514.77.4.785
Kitayama, S., Mesquita, B., & Karasawa, M. (2006). Cultural affordances
and emotional experience: Socially engaging and disengaging emotions
in Japan and the United States. Journal of Personality and Social
Psychology,91(5), 890–903. https://doi.org/10.1037/0022-3514.91.5.890
Krause, A. E., & North, A. C. (2014). Contextualized music listening:
Playlists and the Mehrabian and Russel model. Psychology of
Well-Being,4(1), Article 22. https://doi.org/10.1186/s13612-014-0022-7
Krause, A. E., & North, A. C. (2018). ‘Tis the season: Music-playlist prefer-
ences for the seasons. Psychology of Aesthetics, Creativity, and the Arts,
12(1), 89–95. https://doi.org/10.1037/aca0000104
Kuhn, M. (2008). Building predictive models in R using the caret package.
Journal of Statistical Software,28(5), 1–26. https://doi.org/10.18637/jss
.v028.i05
Labbé, E., Schmidt, N., Babin, J., & Pharr, M. (2007). Coping with stress:
The effectiveness of different types of music. Applied Psychophysiology
and Biofeedback,32(3–4), 163–168. https://doi.org/10.1007/s10484-
007-9043-9
Lamoreaux, M., & Morling, B. (2012). Outside the head and outside individ-
ualism–collectivism: Further meta-analyses of cultural products. Journal
of Cross-Cultural Psychology,43(2), 299–327. https://doi.org/10.1177/
0022022110385234
Law, E., West, K., Mandel, M. I., Bay, M., & Downie, J. S. (2009, October
26–30). Evaluation of algorithms using games: The case of music tagging
[Paper presentation]. Proceedings of the 10th International Society for
Music Information Retrieval (ISMIR), Kobe, Japan.
Liew, K., Mishra, V., Zhou, Y., Epure, E. V., Hennequin, R., Wakamiya, S.,
& Aramaki, E. (2022, December 4–8). Network analyses for cross-cultural
music popularity [Paper presentation]. Proceedings of the 23rd
International Society for Music Information Retrieval Conference
(ISMIR), Bengaluru, India.
Liew, K., Uchida, Y., Domae,H., & Koh, A. H. Q. (2022). Energetic music is
used for anger downregulation: A cross-cultural differentiation of intensity
from rhythmic arousal. Journal of Applied Social Psychology. Advance
online publication. https://doi.org/10.1111/jasp.12951
Lim, N. (2016). Cultural differences in emotion: Differences in emotional
arousal level between the East and the West. Integrative Medicine
Research,5(2), 105–109. https://doi.org/10.1016/j.imr.2016.03.004
Liu, P., Chan, D., Qiu, L., Tov, W., & Tong, V. (2018). Effects of cultural
tightness–looseness and social network density on expression of positive
and negative emotions: A large-scale study of impression management
by Facebook users. Personality and Social Psychology Bulletin,44(11),
1567–1581. https://doi.org/10.1177/0146167218770999
Markus, H. R., & Kitayama, S. (1991). Culture and the self: Implications for
cognition, emotion, and motivation. Psychological Review,98(2), 224–
253. https://doi.org/10.1037/0033-295X.98.2.224
Mas-Herrero, E., Marco-Pallares, J., Lorenzo-Seva, U., Zatorre, R. J., &
Rodriguez-Fornells, A. (2013). Individual differences in music reward
experiences. Music Perception,31(2), 118–138. https://doi.org/10.1525/
mp.2013.31.2.118
Masuda, T., Gonzalez, R., Kwan, L., & Nisbett, R. E. (2008). Culture and
aesthetic preference: Comparing the attention to context of East Asians
and Americans. Personality and Social Psychology Bulletin,34(9),
1260–1275. https://doi.org/10.1177/0146167208320555
Mehr, S. A., Singh, M., Knox, D., Ketter, D. M., Pickens-Jones, D., Atwood,
S., Lucas, C., Jacoby, N., Egner, A. A., Hopkins, E. J., Howard, R. M.,
Hartshorne, J. K., Jennings, M. V., Simson, J., Bainbridge, C. M.,
Pinker, S., O’Donnell, T. J., Krasnow, M. M., & Glowacki, L. (2019).
Universality and diversity in human song. Science,366(6468), Article
eaax0868. https://doi.org/10.1126/science.aax0868
Meseguer-Brocal, G., Peeters, G., Pellerin, G., Buffa, M., Cabrio, E., Faron
Zucker, C., Giboin, A., Mirbel, I., Hennequin, R., Moussallam, M., &
Piccoli, F. (2017, August 21–23). WASABI: A two million song database
project with audio and cultural metadata plus WebAudio enhanced client
applications [Paper presentation]. Proceedings of the Web Audio
Conference 2017—Collaborative Audio #WAC2017, London, United
Kingdom.
Morling, B., & Lamoreaux, M. (2008). Measuring culture outsidethe head: A
meta-analysis of individualism—collectivism in cultural products.
Personality and Social Psychology Review,12(3), 199–221. https://
doi.org/10.1177/1088868308318260
Muthukrishna, M., Henrich, J., & Slingerland, E. (2021). Psychology as a
historical science. Annual Review of Psychology,72(1), 717–749.
https://doi.org/10.1146/annurev-psych-082820-111436
North, A. C., & Hargreaves, D. J. (1995). Subjective complexity, familiarity,
and liking for popular music. Psychomusicology: A Journal of Research in
Music Cognition,14(1–2), 77–93. https://doi.org/10.1037/h0094090
North, A. C., & Hargreaves, D. J. (1996). Situational influences on reported
musical preference. Psychomusicology: A Journal of Research in Music
Cognition,15(1–2), 30–45. https://doi.org/10.1037/h0094081
Park, M., Thom, J., Mennicken, S., Cramer, H., & Macy, M. (2019). Global
music streaming data reveal diurnal and seasonal patterns of affective pref-
erence. Nature Human Behaviour,3(3), 230–236. https://doi.org/10.1038/
s41562-018-0508-z
Pérez-Verdejo, J. M., Piña-García, C. A., Ojeda, M. M., Rivera-Lara, A., &
Méndez-Morales, L. (2021). The rhythm of Mexico: An exploratory data
analysis of Spotify’s top 50. Journal of Computational Social Science,
4(1), 147–161. https://doi.org/10.1007/s42001-020-00070-z
Pons, J., & Serra, X. (2019). musicnn: Pre-trained convolutional neural net-
works for music audio tagging. PsyArXiv. https://arxiv.org/abs/1909
.06654
Purves, D. (2017). Music as biology: The tones we like and why. Harvard
University Press.
R Core Team. (2017). R: A language and environment for statistical comput-
ing. R Foundation for Statistical Computing.
Saarikallio, S. (2012). Development and validation of the Brief Music in
Mood Regulation Scale (B-MMR). Music Perception,30(1), 97–105.
https://doi.org/10.1525/mp.2012.30.1.97
DANCEABILITY AND CULTURAL AFFECT 15
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Saarikallio, S., Alluri, V., Maksimainen, J., & Toiviainen, P. (2021).
Emotions of music listening in Finland and in India: Comparison of an
individualistic and a collectivistic culture. Psychology of Music,49(4),
989–1005. https://doi.org/10.1177/0305735620917730
Salgado, J. F. (2018). Transforming the area under the normal curve (AUC)
into Cohen’sd, Pearson’sr
pb
, odds-ratio, and natural log odds-ratio: Two
conversion tables. The European Journal of Psychology Applied to Legal
Context,10(1), 35–47. https://doi.org/10.5093/ejpalc2018a5
Savage, P. E., Brown, S., Sakai, E., & Currie, T. E. (2015). Statistical univer-
sals reveal the structuresand functions of human music. Proceedings of the
National Academy of Sciences,112(29), 8987–8992. https://doi.org/10
.1073/pnas.1414495112
Savage, P. E., Loui, P., Tarr, B., Schachner, A., Glowacki, L., Mithen, S., &
Fitch, W. T. (2021). Music as a coevolved system for social bonding.
Behavioral and Brain Sciences,44, E59. https://doi.org/10.1017/
S0140525X20000333
Schäfer, T., & Sedlmeier, P. (2009). From the functions of music to music
preference. Psychology of Music,37(3), 279–300. https://doi.org/10
.1177/0305735608097247
Schwender, T. M., Spengler, S., Oedl, C., &Mess, F. (2018). Effects of dance
interventions on aspects of the participants’self: A systematic review.
Frontiers in Psychology,9, Article 1130. https://doi.org/10.3389/fpsyg
.2018.01130
Sharman, L., & Dingle, G. A. (2015). Extreme metal music and anger pro-
cessing. Frontiers in Human Neuroscience,9, Article 272. https://
doi.org/10.3389/fnhum.2015.00272
Shoda, H., Yasuda, S., Nakahara, J., Tabei, K., & Isaka, T. (2019).
Tanjyuku-ban Ongaku ni yoru Kibun Chousei Shakudo (B-MMR)
no Nihongo-ban no Sakusei oyobi Shinraisei: Datousei no Kenshou
[Development of the Japanese version of the Brief Music Mood
Regulation scale and evaluation of its reliability and validity]. The
Japanese Journal of Psychology,90(4), 398–407. https://doi.org/10
.4992/jjpsy.90.18207
Singelis, T. M. (1994). The measurement of independent and interdependent
self-construals. Personality and Social Psychology Bulletin,20(5), 580–
591. https://doi.org/10.1177/0146167294205014
Sloboda, J. A., & O’Neill, S. A. (2001). Emotions in everyday listening to
music. In P. N. Juslin & J. A. Sloboda (Eds.), Series in affective science.
Music and emotion: Theory and research (pp. 415–429). Oxford
University Press.
Stevens, C. J., & Byron, T. (2016). Universals in music processing;
Entrainment, acquiring expectations, and learning. In S. Hallam,
I. Cross, & M. Thaut (Eds.), Oxford Library of psychology. The Oxford
handbook of music psychology ( pp. 19–31). Oxford University Press.
Streich, S., & Herrera, P. (2005, May 28–31). Detrended fluctuation analysis
of music signals: Danceability estimation and further semantic character-
ization [Paper presentation]. Proceedings of the AES 118th Convention,
Barcelona, Spain.
The Jamovi Project. (2020). Jamovi (Version 1.2) [Computer Software].
https://www.jamovi.org
Thompson, C., Parry, J., & Wolff, T. (2019). spotifyr: R wrapper for the
‘Spotify’web API (R package version 2.1.1) [Computer Software].
http://github.com/charlie86/spotifyr
Triandis, H. C. (1995). New directions in social psychology. Individualism &
collectivism. Westview Press.
Triandis, H. C., & Gelfand, M. J. (1998). Converging measurement of hori-
zontal and vertical individualism and collectivism. Journal of Personality
and Social Psychology,74(1), 118–128. https://doi.org/10.1037/0022-
3514.74.1.118
Trost, W. J., Labbé, C., & Grandjean, D. (2017). Rhythmic entrainment as a
musical affect induction mechanism. Neuropsychologia,96,96–110.
https://doi.org/10.1016/j.neuropsychologia.2017.01.004
Tsai, J. L. (2017). Ideal affect in daily life: Implications for affective experi-
ence, health, and social behavior. Current Opinion in Psychology,17,
118–128. https://doi.org/10.1016/j.copsyc.2017.07.004
Tsai, J. L., Miao, F. F., Seppala, E., Fung, H. H., & Yeung, D. Y. (2007).
Influence and adjustment goals: Sources of cultural differences in ideal
affect. Journal of Personality and Social Psychology,92(6), 1102–1117.
https://doi.org/10.1037/0022-3514.92.6.1102
Uz, I. (2015). The index of cultural tightness and looseness among 68 coun-
tries. Journal of Cross-Cultural Psychology,46(3), 319–335. https://
doi.org/10.1177/0022022114563611
Van Buskirk, E. (2013). Plotting music’s emotion valence, 1950–2013.
Retrieved January 12, 2019, from Spotify Insights. https://insights
.spotify.com/us/2013/11/05/musics-emotion-over-time/
Wang, H., Masuda, T., Ito, K., & Rashid, M. (2012). Howmuch information?
East Asian and North American cultural products and information search
performance. Personality and Social Psychology Bulletin,38(12), 1539–
1551. https://doi.org/10.1177/0146167212455828
Wee, S., Choo, W. Y., & Cheng, C. Y. (2021). The influence of tertiary edu-
cation disciplines on self-construals and conflict management tendencies.
Frontiers in Psychology,12, Article 659301. https://doi.org/10.3389/
fpsyg.2021.659301
Received March 21, 2022
Revision received April 21, 2023
Accepted April 25, 2023 ▪
LIEW ET AL.16
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
A preview of this full-text is provided by American Psychological Association.
Content available from Psychology of Aesthetics Creativity and the Arts
This content is subject to copyright. Terms and conditions apply.
