Exploring geographic and cultural differences in preferences for valence and arousal in music using regional music charts

Abstract

Music can evoke and enable emotional expression. Yet few studies have investigated which aspects of music preference are universal, and which aspects vary geographically and culturally. To investigate, we used a large, ecological dataset to assess the valence and arousal of songs in music charts from 64 countries. Furthermore, we explored how differences between music charts were predicted by income inequality, GDP, Hofstede’s cultural dimensions, and tightness-looseness. First, we used a top-down approach whereby we grouped countries into five global regions—Western and North-West Europe, Eastern and Southern Europe, Asia, Central and South America, and North Africa and The Middle East. Second, we used a bottom-up approach where we grouped countries through the unsupervised learning algorithm, k-means clustering. Using both approaches, we found that, broadly, Central and South American countries preferred more positive and arousing music than countries from other global regions. However, some countries (e.g., Japan, Chile, Switzerland, and Spain) showed music preferences that did not align with other countries in their geographic region. Furthermore, we found that uncertainty avoidance, cultural tightness, and income inequality were the strongest predictors for regional differences in music preferences. This study provides robust evidence for both the universality and diversity of preferences for valence and arousal in music across geographic and cultural groups.

Music appears in all known societies (Mehr et al., 2019). The way people use, perceive, and respond to music is largely universal (Cowen et al., 2020; Greenberg et al., 2022; Jacoby et al., 2019; Mehr et al., 2019). However, there are small but consequential geographic and cultural differences between music preferences (Greenberg et al., 2022; Liew et al., 2021; Park et al., 2019). This indicates the importance of understanding the aspects of musical preference that have universal characteristics and reveal cultural or geographic diversity.

People listen to music for a variety of reasons that can differ across cultures. People use music for nostalgia, entertainment, celebrations, to express their identity, to self-regulate, and to improve cognitive performance (Boer & Fischer, 2012; Hargreaves & North, 1999). One of the most salient functions of music cross-culturally is its ability to evoke and express emotion (Boer & Fischer, 2012; Juslin et al., 2008; Juslin & Västfjäll, 2008; Koelsch, 2014; Mehr et al., 2019; Pannese et al., 2016; Schäfer et al., 2012). However, few large-scale studies have considered how preferences for certain kinds of affect in music—in our case, valence (how positive or negative an emotion is) and arousal (how calming or exciting an emotion is)—differ geographically and cross-culturally. Here, we used top-down and bottom-up analytical approaches to investigate differences in valence and arousal of songs in the Spotify music charts from 64 countries. Furthermore, we also investigated how differences between music preferences were associated with cultural values and economic parameters. We suggest that regional music charts may offer a means through which to understand universality and diversity of preferences for valence and arousal in music (hereafter referred to as affective music preferences).

Preferences for valence and arousal

Affect is often measured through dimensional models, such as Russell’s circumplex model (Russell, 1980). This model focuses on two axes of emotion, valence and arousal. Some theories use this model to understand affective preferences. For instance, affect valuation theory suggests that the affective preferences individuals seek are primarily culturally learned, with research largely focusing on East-West or individualist-collectivist cultural comparisons (Ruby et al., 2012; Tsai et al., 2006; Tsai, Miao, et al., 2007). For example, people in Western and Latin American countries typically seek high-arousal positive valence emotions, whereas people from East Asian countries seek low-arousal positive valence emotions (Chentsova-Dutton et al., 2007; Ruby et al., 2012; Tsai, Louie, et al., 2007; Tsai, Miao, et al., 2007).

Affective preferences can be explored through cultural products, which often embody the shared values and collective aesthetics of a society (Lamoreaux & Morling, 2012). Indeed, differences in affective preferences are often expressed in children’s book characters, social media, and music (Askin & Mauskapf, 2017; C.-M. Huang & Park, 2013; Tsai et al., 2016; Tsai, Louie, et al., 2007; Tsai, Miao, et al., 2007). With music, for example, European Americans are more likely to choose a CD of perceived high-arousal positive music, whereas Hong Kong Chinese and Asian Americans are more likely to choose a CD of perceived low-arousal positive music (Tsai, Miao, et al., 2007). Furthermore, collectivist countries (Brazil, Kenya, Portugal) express more nostalgic and spiritual reactions to music than individualist countries (Australia, USA, Sweden; Juslin et al., 2016).

Beyond individual-level studies, it may be also beneficial to use large-scale data to assess changes in affective music preferences. For instance, an analysis of recorded music from 1967 to 2017 showed changes in cultural preferences for various genres and the overall increase in music diversity over this period (Negro et al., 2022a). Further research has found that music of social or cultural relevance (e.g., music that has won a Grammy) can lead to changes in music styles and influence the heterogeneity of cultural products (Negro et al., 2022b).

Cultural and geographic differences in valence and arousal

Research has also been conducted to assess the cultural and geographic differences in music listening. An analysis of global music streaming data for one million individuals found that music played in Latin America is higher in arousal and valence, whereas music in Asia is typically lower in arousal and valence (Park et al., 2019; Uchida & Kitayama, 2009). The above research uses a top-down approach to understand broad differences between geographic groups. However, one limitation of classifying countries by their regions is it assumes that groups of countries within these regions inherently share similar cultural characteristics (Liew et al., 2021; Taras et al., 2010). Through effects such as globalization, geography alone may not suffice when understanding differences in affective music preference across cultures (Figge & Martens, 2014; Greenberg et al., 2022). Many researchers note the importance of looking beyond large cultural or geographic categories when assessing global variation on psychological variables (Krys et al., 2022; Liew et al., 2021; Miller, 2002; Tsai et al., 2006; Vignoles et al., 2016). This is because investigating large, geographic based cultural groups may miss important within-group differences (Muthukrishna et al., 2020). Liew et al. (2021), for example, found variance in music preference within non-Westernized cultural groups, with songs originating from Japan and Japanese music charts showing higher song arousal than songs from Taiwan or the USA.

Research assessing cross-cultural and geographic differences affective preferences have largely focused on East Asian, Euro-American, and Latin American cultures due to clear differences in individualism-collectivism (Ruby et al., 2012; Tsai et al., 2006). Individualist cultures more often aim to influence others, and therefore prefer high-arousal states to meet this aim, whereas collectivist cultures more often aim to adjust to others, and therefore desire lower arousal states (Tsai, Miao, et al., 2007). However, many researchers note the importance of looking beyond cultural dichotomies, particularly of individualism and collectivism. This is because such dichotomies may not adequately capture the diversity between different global regions (Krys et al., 2022; Miller, 2002; Vignoles et al., 2016).

As music charts are widely accessible through music streaming services such as Spotify, they can be used to assess these differences across cultures and global regions, rather than making East-West or individualist-collectivist comparisons (Vignoles et al., 2016). Assessing several countries from various global regions means we can investigate an array of cultural and economic values that may account for cultural variation in music preferences. Cultural values consist of the norms, beliefs, and attitudes that distinguish one group or culture from another (Kaasa, 2021). There are several elements that make up culture, and many scholars have tried to develop various sets of cultural dimensions. Indeed, music is essential for the expression of cultural and national identity, with many countries holding culture-specific music styles (Boer et al., 2013). As such, assessing the cultural values across countries may provide additional insight into why cultures differ in affective music preference (Kirkman et al., 2006).

In this paper, we chose to use the six cultural dimensions mentioned by Hofstede et al. (2010): individualism, masculinity, long-term orientation, power distance, uncertainty avoidance, and indulgence. Hofstede’s cultural dimensions have been shown to be theoretically and empirically similar to other cultural value frameworks, including Schwartz’s (1992) cultural dimensions. For example, the autonomy, power distribution, and mastery in Schwartz’s dimensions are theoretically similar to individualism, power distance, and masculinity in Hofstede’s dimensions (Nardon & Steers, 2009). However, we decided to use Hofstede’s framework as it includes the dimension uncertainty avoidance (a countries intolerance for ambiguity), which is associated with how likely a country follows cultural scripts in response to certain events and situations to avoid emotional ambiguity (Petro et al., 2018; Ruby et al., 2012; Triandis et al., 1984). Potentially, the avoidance (or lack thereof) of emotionally ambiguous situations may be an important predictor in understanding geographic differences in affective music preferences.

Additionally, we included the cultural dimension tightness-looseness, which measures the strength of social norms in a society (Gelfand et al., 2011). Past research has found that tightness-looseness was associated with emotional expression, with culturally tight countries being more likely to express positive emotions. As such, we believed it could be an additional relevant predictor in understand affective music preferences across countries (Liu et al., 2018).

The present research

We used regional music charts to assess cultural differences in preferences for valence and arousal in music. We also investigated how cultural values predict differences in these preferences. We assessed the Spotify music charts of 64 countries and applied both top-down and bottom-up analytical approaches. First, we assessed differences in preferences for valence and arousal across five global regions: Central and South America, Asia, Southern and Eastern Europe, North Africa and The Middle East, and Western and North-West Europe. Second, we performed a bottom-up, data-driven approach to assess how countries cluster in terms of the valence and arousal of songs in their music charts. In line with past research on ideal affective preferences, we predicted that Asian countries would listen to lower arousal music, whereas countries from Central and South America, and Western and North-West Europe would listen to higher arousal music (Ruby et al., 2012). We expected there be no differences in valence preferences across these regions. We did not have any a priori predictions for how countries from Southern and Eastern Europe, and North Africa and The Middle East would differ in song valence and arousal preferences. For both analytical approaches, we explored how cultural values (e.g., Hofstede cultural dimensions, tightness) and economic parameters are associated with regional differences in affective music preference. We report how we collected our data and all data exclusions. Analyses used in this study were not pre-registered. All data, analysis code, supplemental materials, and research materials for are accessible here: https://osf.io/6f8bx/?view_only=906354b62df541c38b47c5562d0a0edf. We analysed all data using R, version 4.2.1 (R Core Team, 2022). We cite specific packages in the method section of each study.

Method

Data collection

Music charts for each country were collected from Spotify. Spotify is a large music listening platform with over 406 million monthly active users from a diverse age range. For instance, 55% of users are under the age of 35, and 19% of users are over the age of 55 (Spotify, 2021).

Spotify—along with researchers of Music Emotion Recognition (MER)—uses computational models that aim to automatically detect the elicited emotion from music (Panda et al., 2020, 2021; Thompson et al., 2021). To do this, Spotify uses a music intelligence service called The Echo Nest (Hern, 2014). The Echo Nest is a machine-learning, deep learning, and digital signal processing algorithm that estimates and continually updates song information on a variety of audio features (Askin & Mauskapf, 2017; Panda et al., 2021). As Spotify is a private company, not all details on the Echo Nest are publicly available. The Echo Nest used expert-annotated data to initially develop the algorithm to detect these audio features. Following this classification, a machine-learning algorithm was developed to extend those results to all music on the platform (Dredge, 2013). Spotify uses this information—along with cultural knowledge scraped from the internet—to cluster artists into genres and moods (Johnston, 2018). It represents the current gold standard in music information retrieval (Askin & Mauskapf, 2017; Park et al., 2019).

Automatically generated features can similarly capture latent dimensions of human-perceived attributes, like affect (Fricke et al., 2018; Park et al., 2019). Two audio features captured by Spotify are titled “valence” and “energy”. According to Spotify, valence refers to the musical positiveness conveyed by a track, where songs with higher valence are more happy and cheerful. Energy refers to a measure of intensity and activity, including dynamic range, perceived loudness, timbre (e.g., characteristics of a musical sound), onset rate (e.g., how quickly a note is played), and general entropy (e.g., the balance between musical patterns and chaos; Krols et al., 2023; Thompson et al., 2021). Lyrics are excluded from valence and energy ratings as musicologists argue that audio features such as arousal have better cross-cultural applicability without the constraints of lyrics (Park et al., 2019). Finally, past research has found the Spotify algorithm shows no cultural bias to English songs and conclude that the current algorithm is a good objective proxy for human judgements, at least within pop music, which largely constitutes music charts (Lee et al., 2021).

The dimensions of valence and energy stem from human input and, thus, likely reflect the valence and arousal concepts in music and affect literature (Eerola & Vuoskoski, 2013; Russell, 1980). These two dimensions are most comparable to valence and arousal in Russell’s circumplex model of emotion, with energy serving as a proxy for arousal (Panda et al., 2021; Russell, 1980). As such, from this point on, we will refer to song energy as arousal.

Audio features and listening data from Spotify are potentially an ecologically valid tool for understanding how people listen to and use music. Indeed, past research has found that human ratings of song valence and arousal are positively correlated with Spotify’s ratings (Vidas et al., in press). As such, Spotify audio features have been used in variety of research, including investigating the relationship between music preferences and pain management (Howlin & Rooney, 2021), nostalgia (K.-J. Huang et al., 2023), anxiety (Pyun et al., 2020), listening to relaxing music (Baltazar & Västfjäll, 2020), anger down regulation (Liew et al., 2023), time of day (Park et al., 2019), the COVID-19 pandemic (Vidas et al., 2021), and listening preferences between countries such as Taiwan, America, and Japan (Liew et al., 2021).

Two studies have been conducted to assess the accuracy of Spotify against previously validated music using the circumplex model (Russell, 1980). Panda et al. (2021) used a 704-song dataset, with songs annotated in terms of Russell’s quadrants using mood descriptors from online music service, AllMusic. ¹ Krols et al. (2023) used Spotify to predict valence and arousal scores on the Deezer Mood Detection Dataset, which includes 18,000 songs rated on valence and arousal. Songs were rated using mood descriptors from online music service LastFM. ² Mood descriptors from both websites are user generated. For both studies, mood descriptors were matched using a validated dataset which associates 14,000 English words into the two-dimensional model of emotion (Russell, 1980; Warriner et al., 2013). Both studies found valence and arousal to be the strongest predictors. This indicates that these higher-level audio features are measuring their appropriate underlying construct. As such, both studies conclude that these features are highly relevant to MER.

Additionally, Panda et al. (2021) found acousticness—whether the song includes acoustic features—to be a highly relevant feature, being strongly (negatively) correlated with arousal. Whereas Krols et al. (2023) found that danceability, instrumentalness, mode, and speechiness were also predictors of valence and arousal. These studies conclude that although the publicly available API does not perform as well as some state-of-the-art MER software, Spotify can provide desirable higher-level emotionally relevant features that are interpretable to human concepts. Based on the above research, along with past psychological literature using the Spotify as a tool, we believe that Spotify is an effective tool to assess the emotional valence and arousal of music charts across countries and regions.

We collected weekly Spotify chart data for all available countries (N = 64) at six time points between January 1st and December 31st, 2019, using https://charts.spotify.com/. Six time points allowed us to collect data at 2-month intervals, reducing the possibility of seasonal effects. Spotify provides country-specific playlists of the top 200 most played songs for a specific week. In total, 78,000 songs were obtained for this analysis. We obtained the valence and energy (which we refer to as arousal) values for each track using the R package SpotifyR which queries the API provided by Spotify (Thompson et al., 2021).

Cultural values and economic parameters

We assessed six cultural values described by Hofstede et al. (2010). These include uncertainty avoidance (a countries intolerance for ambiguity), power distance (a countries acceptance of unequal power distribution), individualism (a countries value of individualist social ties), masculinity (a countries propensity to adopt more masculine behaviours), long-term orientation (a countries prioritization of connection to future actions), and indulgence (a countries need to fulfil human desires). We assessed cultural tightness using data collected by Eriksson et al. (2021). Data for gross domestic product (GDP) and income inequality are from The World Bank, World Development Indicators (The World Bank, 2021).

Top-down analytical approach

We categorized global regions using the Australian Bureau of Statistics Standard Classification of Countries (Australian Bureau of Statistics, 2016). In our dataset, Central and South America is comprised of 17 countries. Asia is comprised of ten countries from Eastern, South-Eastern, and Central Asia. Southern and Eastern Europe is comprised of 12 countries. North Africa and The Middle East is comprised of six countries. Finally, Western and North-West Europe is comprised of 18 countries from North-Western Europe and the Anglosphere. For a full list of countries in each region, see Supplemental materials. Data could also be collected for South Africa; however, we could not classify it with the other global regions used in this study. Therefore, we excluded it from our top-down analytical approach.

Statistical approach

We performed two separate linear mixed-effects models to assess the effects of global regions on both valence and arousal of songs using the lme4 package (Bates et al., 2015). Specifically, we classified global region as a predictor variable, and song arousal and song valence as outcome variables. We used Asia as the reference variable in both models. We modelled country as a random effect. We performed Tukey adjustments to control for Type 1 error when assessing simple effects.

Results

We found that valence and arousal were positively correlated (r = .70, p < .001). Our mixed-effects regression revealed significant differences in global region when predicting song arousal (F_4,59 = 11.08, p < .001, ηp² = .43). We found that Central and South American (M = 0.68, SD = 0.15) countries listened to significantly higher arousal songs than Western and North-West European countries (M = 0.63, SD = 0.16, estimate = 0.05, 95% CI = 0.02–0.07, p < .001), Southern and Eastern European countries (M = 0.64, SD = 0.15, estimate = 0.04, 95% CI [0.01–0.07], p = .018), North African and Middle Eastern countries (M = 0.63, SD = 0.16, estimate = 0.06, 95% CI [0.01–0.09], p = .003), and Asian countries (M = 0.60, SD = 0.18, estimate = 0.08, 95% CI [−0.12 to −0.05], p < .001). Furthermore, East Asian countries listened to significantly lower arousal music than Southern and Eastern European countries (estimate = −0.05, 95% CI [−0.08 to −0.01], p = .009). No other follow-up analyses were significant (see Figure 1). ³

Figure 1.

Differences in song arousal and song valence scores for North Africa and The Middle East, Southern and Eastern European, Western and North-West European, Asian, Central and South American Countries.

Our second mixed-effects regression showed significant differences in global region when predicting valence in songs (F_4,59 = 41.08, p < .001, ηp² = 0.74). This analysis showed that Central and South American (M = 0.62, SD = 0.21) countries listened to significantly higher valence songs than Western and North-West European countries (M = 0.49, SD = 0.21, estimate = 0.12, 95% CI [0.08–0.16], p < .001), Southern and Eastern European countries (M = 0.48, SD = 0.21, estimate = 0.14, 95% CI [0.08–0.16], p < .001), North African and The Middle Eastern countries (M = 0.48, SD = 0.21, estimate = 0.13, 95% CI [0.08–0.18], p < .001), and Asian countries (M = 0.45, SD = 0.20, estimate = −0.16, 95% CI [−0.20 to −0.12], p < .001). No other analyses were significant (see Figure 1).

Bottom-up analytical approach

In line with past research, we found that the music charts of Central and South American countries contained higher arousal and more positively valanced music than any other global region (Park et al., 2019). One limitation of classifying countries by their regions is it assumes clusters of countries represent cultural groups (Liew et al., 2021; Taras et al., 2010). Through effects such as globalization, geography may not be the only way for understanding differences in affective music preferences across cultures (Figge & Martens, 2014). As a result, we tested how countries cluster together on the valence and arousal of songs in their respective music charts.

To assess this, we used K-means clustering. K-means clustering is an unsupervised algorithm that clusters similar data-points to a pre-determined number of groups. This approach has been used to understand cross-cultural differences in music preferences in Taiwan, Japan, and the USA (Liew et al., 2021), and as such, it may provide a more accurate representation of cross-cultural differences in affective music preferences. In our case, countries were clustered by how similar they were on scores of song arousal and valence.

To determine the best number of clusters we used the R package NbClust, which provides 24 indices for determining the optimal number of clusters in a data set (Charrad et al., 2014). This approach indicates that the ideal number of clusters is two. As a result, we used the two cluster centres generated by the k-means algorithm to assess the countries that are grouped together with respect to arousal and valence. We set the algorithm to attempt 25 initial configurations for the centroids and set the maximum number of iterations to 5,000. Cluster one includes 19 countries (M_valence = 0.62, M_arousal = 0.69), and cluster two includes 46 countries (M_valence = 0.47, M_arousal = 0.62). Cluster one includes all Central and South American countries, except for Chile. In addition, cluster one includes Spain, Switzerland, and Japan. Cluster two includes Chile, and all remaining countries from Asia, North Africa and the Middle East, Western and North-West Europe, and Southern and Eastern Europe (see Figure 2).

Figure 2.

Song arousal and song valence among countries, regions, and clusters.

Cultural values and economic parameters

To determine which cultural and economic parameters are associated with song arousal and song valence, we used a model-selection approach using the MuMIn package in R (Bartoń, 2022; Burnham & Anderson, 2002). Model selection is the process of selecting the most appropriate model among a suite of models. This selection is based on an information-theoretic criteria. Under this approach, we do not rely on a single model and the combination of parameters and interactions we select for that model. Instead, we assess all possible combinations of parameters and interactions with several models in an unbiased way. This approach allows for more robust and reliable inferences.

We modelled GDP, income inequality, power distance, individualism, masculinity, uncertainty avoidance, long-term orientation, indulgence, and tightness, including all main effects and two-way interactions. These predictors were included as fixed effects, and country was included as a random effect. We performed Tukey adjustments to control for Type 1 error when assessing the simple slopes of significant interactions.

This approach compares all possible sub-models that could be created from the nine predictor variables, including the null. Each sub-model contains a subset of all the parameters we assessed. For example, one sub-model may comprise of GDP, masculinity, and tightness, whereas another sub-model may comprise of masculinity, individualism, and the interaction between masculinity and individualism. Each of the sub-models receives an Akaike Information Criterion (AIC) value. This value describes the likelihood that one model explains the data better than all other models. We selected the model with the lowest AIC value to identify the model that best explains the data.

After assessing all potential sub-models, based on Akaike model selection, we found that uncertainty avoidance alone was most likely to positively predict variation in song arousal (β = .15, 95% CI [.07, .21], p < .001). We found income inequality (β = .09, 95% CI [.02, .18], p = .019), cultural tightness (β = −.11, 95% CI [−20, −.06], p < .001), and the interaction (β = −.11, 95% CI [−.20, −.04], p = .004) between these two variables to predict song valence. Specifically, at high levels of inequality, countries high in looseness are more likely to listen to high valence music than countries high in tightness (p < .001). At low levels of inequality, there was no difference between tight and loose countries (p = .951).

General discussion

Using a large, novel dataset, the findings of this study provide robust evidence for both diversity and universality in preferences for valence and arousal in music across geographic and cultural groups. First, through clustering countries into their global regions, we found that Central and South American countries listen to more arousing and positively valanced music compared to other global regions. This result supports past research, which found that Latin America countries listen to more positive and arousing music (De Almeida & Uchida, 2018; Park et al., 2019). This also aligns with cultural differences in ideal affective preferences, as Latin American countries are more likely to endorse positive, high-arousal emotional states than East Asian cultures (Ruby et al., 2012). We found that Southern and Eastern European countries also listen to more arousing music than Asian countries; however, in general, no other global regions differed on song valence or arousal, indicating that there is relative universality in affective music preference.

These results were further supported using a bottom-up data-driven approach where we clustered countries by how similar they were on scores of song arousal and valence. Through this approach, we found two clusters: one that primarily featured countries from Central and South America, and one that primarily featured countries from the rest of the world. The predominantly Central and South American cluster was associated with more arousing and positively valanced music.

The advantage of using a bottom-up analytical approach is it allows us to look beyond geography-based classifications, which can miss important within-group differences (Liew et al., 2021). We found that some countries were notable anomalies in how they were clustered. These include Japan, Spain, Chile, and Switzerland. Past research has also observed anomalies from cluster analyses, with some geographically distant countries clustering together in terms of their structure of musical preferences (Greenberg et al., 2022). This indicates that clusters may be organized by cultural mechanisms that extend beyond geographical proximity. Indeed, Japan is distant in culture and music preferences to other countries in its region, like China, Malaysia, and the Philippines (Greenberg et al., 2022; Muthukrishna et al., 2020). Japanese music is also marked by high arousal, in contrast to other countries in its region, such as Taiwan (Liew et al., 2021). The inclusion of Spain in this cluster may be explained by the fact that most countries in Central and South America speak Spanish and share cultural and historical ties. Switzerland and Chile listen to music with a similar arousal to other countries in their region, yet song valence differed markedly. Why these countries clustered outside their global region remains unclear and future research should aim to assess these within-region differences.

Cultural values, economic parameters, and music preferences

We also explored economic and cultural-level mechanisms that may account for differences between countries and cultural groups. Through model section, we found that uncertainty avoidance best predicts differences in song arousal between countries. Uncertainty avoidance is a cultural value that represents a person’s intolerance towards ambiguous situations. Countries high in uncertainty avoidance desire predictability and are more likely to follow cultural scripts (i.e., a culturally specific way of expression or communication; Goddard & Wierzbicka, 2004) to avoid potential uncertainty (Lamoreaux & Morling, 2012). For instance, some cultural scripts followed by many Central and South American countries promote expressivity to avoid conflict and emotional ambiguity (Petro et al., 2018; Ruby et al., 2012; Triandis et al., 1984). As a result, Central and South American countries may seek higher arousal emotional states, and in turn, seek music that reflects these emotions.

We also found that tightness, income inequality, and the interaction between the two best predict differences in song valence. Specifically, countries with greater income inequality, who were also more culturally loose, listened to more positive music than countries with tight cultures. For countries low in income inequality, this effect was not significant. Income inequality is negatively associated with happiness (Oishi et al., 2011), and as such, citizens of countries high in income inequality may, on average, listen to more positive music to navigate the challenges posed by income inequality. Alternatively, more affluent people are often higher in musicality and are given more opportunities to explore their music preferences, and as such may listen to more positive music while they enjoy their successful status (Müllensiefen et al., 2014). This may be further exacerbated in loose cultures, who are more likely to embrace flexibility and personal expression (Gelfand et al., 2011). Further research is needed to clarify the direction of these effects.

The above results were derived from a model-selection approach, which allows for more robust and reliable inferences. However, it is also important to acknowledge that such an approach is exploratory, and future research is necessary to further examine how cultural values, cultural norms, and economic parameters influence preferences for valence and arousal in music.

Implications, limitations, and future directions

This research has several theoretical and methodological implications. First, this paper extends our theoretical understanding of cultural differences in affective music preferences. Past research has debated the extent to which music preferences are universal (Mehr et al., 2019). By assessing the country-level music chats from various global regions, we find both similarities and differences in countries and cultural group preferences for positive and arousing music. Second, this research highlights the potential for using the Spotify API as a viable means of measuring cross-cultural differences in ideal affect. While music emotion recognition is not perfect (Panda et al., 2021; Vidas et al., in press), it allows researchers to investigate research questions that could not feasibly be achieved by human experts and could thus identify trends and patterns that may not be available with smaller datasets or human level data. Future research could explore how individual-level factors, such as personality, could predict variance in music preferences to achieve an ideal affective state. This approach could be extended to other areas of emotion research and psychological research.

Using Spotify offers us a highly ecological dataset to analyse broad trends in affective music preferences. However, a consequence of such an approach is it cannot address diversity from other facets of music such as music taste or style, or effectively view within-country trends. For instance, how does music preference differ between cultural groups within multicultural countries like the UK or Australia? Future research should therefore investigate these within-country possibilities. It is also important to note that this data is cross-sectional, thus we cannot determine a direction of causality between music preferences and cultural dimensions, nor assess how these preferences within cultures changes over time.

Another potential limitation of this study is the assumption that Spotify music charts reflect the most popular songs in the country. Spotify is the most popular global music streaming service; however, not everyone listens to music via this platform. More broadly, Spotify, along with all music streaming companies, try to recommend people songs they think that they will like. This process means that a countries most popular songs may be conditional on the music being distributed or recommended by organizational entities. Although this was true before music streaming services were popular (e.g., record labels recommending songs to be played on radio), it is possible that Spotify music charts are not a true reflection individuals’ favourite music in each country.

This may have led to the unexpected inconsistencies, such as Japan clustering with Central and South American countries, and not Asian countries. However, as of the 15th of March 2023, Spotify and Billboard—a music chart service that aggregates both streaming and radio play—share 17 songs in their respective top 20 playlists (Cabison, 2023), indicating that Japan’s results in this study are likely cultural and are not an artefact of the Spotify listener base. Furthermore, music is only one cultural product which could effectively capture affective preferences. Other cultural products (e.g., films, literature) could also be investigated. Finally, future research may also consider additional cultural or environmental factors that could influence the present results. For example, other models of cultural values, such as Shwartz’s cultural dimensions, as well as climate (Anglada-Tort et al., 2023), and political, economic, social and health factors could help to elucidate our findings.

Conclusions

Using a novel dataset, we find both universality and diversity in preferences for valence and arousal in music. By moving beyond traditional cultural dichotomies, our findings suggest there is cross-cultural variation in both preferred valence and arousal. Finally, we begin to determine how cultural and economic parameters, such as uncertainty avoidance, tightness, and inequality, potentially shape music preferences. These findings are an important step forward in understanding the complexities of cultural differences in music preferences.