Musical Preference: Role of Personality and Music-Related Acoustic Features

Music Classification

The inherent ambiguity of genre classification has been a persistent concern in the field of music preference and personality research. Aucouturier and Pachet (2003, p. 83) have stated that genre is “intrinsically ill-defined”, and described genre as “intentional and extensional” concepts that are mismatched in the real world—how we interpret genre (intentional) and how we expect genre to consist of qualitative properties (extensional) do not align. Genre classifications may refer to music by the period in which it was written (e.g., baroque), associated location or culture of origin (e.g., Indian, Balinese), topic of lyrical content (e.g., love song), type of dance (e.g., waltz), instrumentation (e.g., piano), language of lyrics, and so on. However, the music of an artist, from an album, song, or even part of a song frequently does not conform to a single genre category, belonging, instead, to a combination of definable categories (e.g., folk rock) or specialized subcategories (e.g., drumstep). This phenomenon highlights the abstract and subjective (and possibly inaccurate) nature of music as classified by genre.

Furthermore, the conception of musical genre can be influenced by age, gender, and culture, among a variety of other factors. For instance, the frequency of listening to pop and folk music increases with age (Dunn et al., 2012). Is this a result of changes in preference with age, or the differing conceptions of genres by old and young listeners? We do not know the answer to this question; however, using principle component analysis, Rentfrow and Gosling (2003) determined four music-preference dimensions along which music is categorized, while Zweigenhaft (2008) suggested that there are five dimensions, and noted that different populations and the timing of studies may have influenced the discrepancies between their analyses. Subsequently, Rentfrow, Goldberg, and Levitin (2011) included a fifth factor, in what they refer to as the MUSIC model. Music-preference dimensions in the MUSIC model include mellow, unpretentious, sophisticated, intense, and contemporary (Rentfrow et al., 2012).

While genre classification provided a starting point for the analyses cited, other researchers have begun to explore alternative methods. Acknowledging the constraints of “broad and illusive genres or styles,” Greenberg et al. (2016, p. 597) assessed preference for perceived attributes of music. These attributes described emotional and cognitive aspects of music (e.g., aggressive, intelligent, or dreamy) and were organized into three dimensions: arousal, valence, and depth. In turn, each dimension was correlated with the Big Five personality factors, which we discuss next (John, Donahue, & Kentle, 1991). In contrast, Dunn et al. (2012, p. 426) noted that “preference for objectively measured music characteristics might provide greater accuracy for measuring music preferences.” This approach led Eerola, Friberg, and Bresin (2013) to manipulate musical cues (e.g., mode and tempo) in their study of emotional expression in music; such cues were important for the expression of scary, happy, sad, and peaceful sentiments in music. Arguably, this type of systematic manipulation of musical cues has, to date, received relatively little attention in the music-preference literature.

Personality

In general, the assessment of personality attempts to determine variation among individuals with respect to specific dimensions (Buss, 1987). Although a number of methods have been advanced to define personality dimensions (e.g., the Myers–Briggs Type Indicator, NEO Personality Inventory (NEO-PI), and Minnesota Multiphasic Personality Inventory), consensus has converged, in many fields of research, on the Big Five taxonomy (John, Naumann, & Soto, 2008). With development rooted in lexical analyses, the Big Five categorizes personality traits based on commonly used language and, after many variations, has arrived at the following factors: (1) extraversion (vs. introversion), relating to traits that are described as sociable, outgoing, energetic, enthusiastic, adventurous, and forceful; (2) agreeableness (vs. antagonism), described as warm, forgiving, sympathetic, and not stubborn, demanding, or showing off; (3) conscientiousness (vs. lack of direction), traits related to being efficient, organized, thorough, and not lazy, careless, or impulsive; (4) neuroticism (vs. emotional stability), related to shy, moody, irritable, tense, and not confident or content; and (5) openness (vs. closed to experience), related to being curious, artistic, imaginative, excitable, having wide interests, and unconventional. Each factor has been shown to be stable within individuals over time and has been verified as predictive. For example, Big Five personality factors predict juvenile delinquency and school performance in adolescents, and job performance and health outcomes for older adults (John et al., 1991, 2008).

Instruments designed to measure Big Five factors, such as the lengthy 240-item NEO-PI questionnaire, have been extensively tested and validated. The NEO-PI questionnaires break each personality factor into six unique facets. For example, extraversion can be broken down into warmth, gregariousness, assertiveness, activity, excitement-seeking, and positive emotions. While the full NEO-PI questionnaire is recommended, the Big Five Inventory (BFI) is a shorter, efficient alternative. In sum, the BFI reliably determines an individual’s position along each of the five personality factors using only 44 questionnaire items and a five-point Likert scale; however, facet scores per dimension are not produced.

Although preference for music has been linked to personality a number of times in the literature, interpreting the extent to which these factors correlate depends on the classification methodology used in each study. Rentfrow and Gosling (2003) used the BFI to investigate preference for various music dimensions extrapolated from 14 genres. For example, the reflective and complex dimension describes music that is slower in tempo, using acoustic instruments, and with little singing; the intense and rebellious dimension describes music of a faster tempo, using electric instruments, and with moderate singing. Zweigenhaft’s (2008) study used the NEO-PI and investigated preference for 21 musical genres. While his results overlapped with Rentfrow and Gosling’s (2003, 2006) analyses using music dimensions, differences between these studies prompted Zweigenhaft to continue his analysis at the genre level, as opposed to using solely music dimensions. Lastly, using the International Personality Item Pool, Greenberg et al. (2016) found that preference was linked to 38 perceived psychological attributes in music. These attributes, in contrast to music dimensions and genres, instead refer to the level of arousal, type of valence, and emotional depth that is perceived in music. The foregoing underscores the point that many approaches and interpretations have been undertaken by researchers with respect to music preference and personality, leading to a plethora of interpretations and conclusions.

In addition to the categorization methods used in these studies, the interpretation of results must consider the manner in which preference ratings are recorded. Rentfrow and Gosling (2003) measured self-reported preferences with the absence of auditory stimuli. They developed the Short Test of Music Preference (STOMP), a questionnaire that used a seven-point Likert-type scale to rate 14 music genres. In contrast, Greenberg et al. (2016) presented participants with musical excerpts to evaluate. While these approaches may be consistent within themselves, categorizing music via questionnaires is susceptible to subjective interpretation and may not be consistent between studies that measure preference for perceived auditory stimuli. Rating auditory stimuli in terms of low-level features may represent preference for music more accurately, since participants are presumably not conflating their responses with high-level subjective genre knowledge. That said, as with all music-personality preference studies, participants’ responses may be prone to short-term fluctuating influences, such as mood and environment.

Behavioral analyses offer additional insight into this issue. Music streaming services acquire large amounts of data, such as the frequency and duration of played songs on users’ playlists, allowing the preferred songs and genres of listeners to be inferred. Since these data are often recorded over many months or years, preferences may reflect personality more reliably as the variability of mood is averaged over time. Corroboration of self-reported preferences and preferences inferred from behavioral data has been shown in the literature. Dunn et al. (2012) measured self-reported preferences and analyzed listening behavior from a music streaming database over a 3-month period. Their results showed a positive correlation between these two measures, an important finding that supports the validity of existing research (that used self-report measures of music preference) and inferences made in studies, such as that of Bansal, Flannery, and Woolhouse (2020), who deduced musical preference based on user’s music collections.

Despite the thoroughness of the investigations, results from these studies have been somewhat mixed, with only a few consistent relationships found. In general, openness often correlates with reflective and complex and intense and rebellious music dimensions, consisting of blues, classical, folk, jazz, alternative, heavy metal, and rock genres. These dimensions and genres include a broad range of tempi, and acoustic and electric instruments, with little to moderate singing. Extraversion correlates with upbeat and conventional and energetic and rhythmic dimensions, and the genres of country, pop, religious, soundtracks, dance, rap, and soul. These dimensions and genres are more narrowly focused on music that is of moderate tempo, using electric instruments, and with moderate singing (Dunn et al., 2012; Rentfrow & Gosling, 2003; Zweigenhaft, 2008). In relation to perceived attributes, agreeableness and conscientiousness correlate with preference for low arousal attributes, which are gentle, calming, and mellow, and negatively with music that is intense, forceful, abrasive, or thrilling. Neuroticism is correlated with negative valence attributes, which are depressing and sad, and negatively with fun, happy, lively, enthusiastic, and joyful music. In addition, openness correlates with attributes of positive valence and depth, attributes that are perceived as intelligent, sophisticated, inspiring, complex, poetic, deep, emotional, and thoughtful, and negatively correlates with party and danceable music (Greenberg et al., 2016). While these studies provide insight into how personality relates to music, in sum, the use of abstract categorization methods, such as genre and musical attributes, limits the conclusions that can be made about the specific features of music responsible for our likes and dislikes. Thus, there is a need for further research.

The Study

This study aims to provide a deeper understanding of how features of music, independent of genres, dimensions (e.g., reflective and complex), or psychological attributes, are preferred by individuals. While factor analysis can determine dimensions that are more or less orthogonal to one another, the precise composition of the resulting factors, in terms of musical attributes, remains unclear. Or, to put it another way, if an individual has a preference for a specific genre or perceived psychological attribute of music, how can we describe that preference objectively? To address this issue, and in contrast to the studies cited previously, we systematically varied musical attributes of the stimuli in a quantifiable manner. Referred to as music acoustic feature (MAF) manipulation, the factors we chose to manipulate were dynamics, mode, register, and tempo; a detailed description of these manipulations is provided in the Methods section. In brief, dynamics were varied using intensity attack and decay rates (the speed at which a waveform increases and decreases in intensity); mode was effected by (re)composing the stimuli so as to conform to music-theoretic major and minor scales; register reflected the octave transpositions the stimuli underwent; and tempo was varied by changing the rate of note onset and thus the perception of beats per minute (bpm). With respect to the findings of the research discussed previously, in which a number of associations between personality and broad aspects of music were found, we hypothesized that the manipulation of MAFs would significantly affect musical preference ratings and interact with personality dimensions. Specifically, based on research by Rentfrow et al. (2011, 2012) and Nave et al. (2018), we predicted that there would be a preference for slower-tempo music by participants high in extraversion and a preference for soft music by participants high in openness. However, it should be noted that the work of Rentfrow et al. (2011, 2012) and Nave et al. (2018) involved genre, and thus a direct relationship to our study, which involved MAFs, cannot be determined. As a result, we formulated our hypothesis broadly, supported by a series of predictions. Although other features are, of course, present in music (for example, melody, harmony, articulation, timbre), and acoustic features vary widely between and within pieces, we chose to isolate and manipulate the four features described here, based on their prevalence and use within the literature outlined next.

Much research has explored the use of acoustic-feature manipulation and analysis, and its links to affect-related responses in music listening. For instance, in order of importance, Eerola et al. (2013) found that mode, tempo, register, dynamics, articulation, and timbre significantly contribute to the perception of emotional expression. Furthermore, extracted acoustic features have been used in data analysis with respect to people’s playlist choices. Barone et al. (2017) analyzed acoustic features that were algorithmically defined and extracted by Spotify’s Web API. Broadly speaking, the acoustic features used in Barone et al. (2017) were described as follows: danceability, related to tempo, beat strength, and rhythm stability; energy, related to fast, loud, and noisy music; loudness, related to amplitude in decibels (dB); valence, related to the detection of positive and negative affect; and acousticness, related to the use of acoustic instruments, including voice. Many of these features were in agreement with the elements of Eerola et al. (2013), such as tempo, dynamics, and timbre. Returning to Barone et al. (2017), their study revealed that acoustic features of acousticness, danceability, energy, loudness, and valence were correlates of primary and secondary preferred genres. For example, dance music consists of relatively high proportions of the acoustic feature energy; listeners who preferred dance music also preferred non-dance tracks with high levels of energy. Their findings suggest that listeners’ choices are not only guided by genre, but that listeners also gravitate to music with higher proportions of specific acoustic features, irrespective of style category. That said, the use of Spotify’s feature-extraction method arguably poses a similar problem to that of genre classification—at root, the relationship between Spotify’s acoustic features, the auditory signal, and musical perceptions are ill-defined.

The acoustic features in our experiment were presented as audible stimuli, which participants responded to directly using a preference rating. Therefore, in contrast to other studies that ask for self-reported genre preference (e.g., the STOMP), measurement of preference for acoustic features did not require the listeners to have a conceptual understanding of the features they were rating, in turn, minimizing the myriad subjective factors that could affect their interpretations.

Participants

Undergraduate students ( N = 90 , 76 female) from the host institution participated in a listening experiment. Participants’ ages ranged from 17 to 21 years ( M = 18.5 , S D = .68 ). Demographic data were collected through a short questionnaire, along with general musical training and basic music-preference information. Musical training ranged from no training ( n = 26 ) to 10 or more years of training ( n = 15 ); the mean number of years of musical training was 4.1, with a standard deviation of 3.89. Each participant completed the 44-item version of the BFI (John et al., 2008), which established scores for each of the five personality factors. There were no exclusion criteria other than those relating to normal audition and all participants were included in the analysis.

Materials

The open-source music composition and notation software MuseScore (Version 3.3.4) was used to create the stimuli. First, three major-key excerpts of solo keyboard music were selected from pre-existing compositions: (1) Mozart, Piano Sonata no. 9 in D Major, K. 311, measures 17–24; (2) Beethoven, Piano Sonata No. 1, Op. 2 No. 1, Adagio, measures 1–8; and (3) Bach, Prelude No. 3 in C-sharp Major, BWV 848, measures 1–16. These excerpts were transcribed into MuseScore and then modified to create two levels within the following four factors: dynamics, mode, register, and tempo (described next); the excerpts formed a fifth factor with three levels, which we referred to as piece. These resulted in a full-factorial design consisting of 48 stimuli (2 × 2 × 2 × 2 × 3 excerpts), which were then exported from MuseScore as .wav files using the software’s “grand piano” instrument selection (included as the general MIDI SoundFont: MuseScore_General.sf3). A graphical program was designed (using Python 3.8 and Kivy 1.9.1) to administer the pre-experiment questionnaire, the BFI, and present the 48 stimuli in a randomized order to each participant. The entire experiment was completed in approximately 35 min using a desktop computer. Stimuli were presented through AKG K 172 HD headphones, with a frequency range of 18 Hz–26 kHz. Each participant adjusted the headphone volume to a comfortable level prior to the first trial of the experiment session.

Piece (Bach, Beethoven, Mozart)

The three excerpts were selected owing to the ubiquity and popularity of the composers (i.e., they are quintessential exemplars) and because their medium tempi and registers allowed them to be effectively manipulated. All the excerpts were well-formed from a music-theoretic perspective, contained regular four- or eight-measure phrases, and followed established principles of functional harmony (Figures 1, 2, and 3).

OPEN IN VIEWER

Figure 1.

Example of register and dynamics manipulation in the Bach piece. Stimulus A is high register and piano dynamics; stimulus B is low register and forte dynamics. Other MAFs are consistent in both stimuli: mode, major; tempo, slow (90 bpm). Audio of Figure 1 is provided as a supplementary file labeled “Audio_Figure_1.wav.”

OPEN IN VIEWER

Figure 2.

Example of mode manipulation in the Beethoven piece. Stimulus A is in F major; stimulus B is in F minor. Other MAFs are consistent in both stimuli: dynamics, piano; register, low; tempo, slow (45 bpm). Audio of Figure 2 is provided as a supplementary file labeled “Audio_Figure_2.wav.”

OPEN IN VIEWER

Figure 3.

Example of tempo manipulation in the Mozart piece. Stimulus A is fast (120 bpm); stimulus B is slow (60 bpm). The remaining MAFs are consistent in both stimuli: dynamics, forte; mode, major; and register, high. Audio of Figure 3 is provided as a supplementary file labeled “Audio_Figure_3.wav.”

The reason for selecting the particular MAFs for this study relates in part to the ecological validity of the stimuli. For example, the Mozart excerpt occurs in both the major and minor mode versions within the larger piece, an example of him engaging in the practice of mixture (Aldwell & Cadwallader, 2018, p. 435). Register, too, is frequently manipulated within real-world compositions. For instance, the second subject material in the recapitulation of a sonata form movement—melodies are invariably transposed up or down a perfect fifth or fourth in accordance to the unfolding key structure of the piece. Moreover, exposition material is often treated to registral and mode manipulation within the development sections of sonatas. And, naturally, performers change the tempi and dynamics of their playing, resulting in any one piece being interpreted differently along these parameters. As a result, the MAF manipulations that we deployed—mode, register, tempo, dynamics—are all subject to a high degree of variability within music. Which is to say, none of the manipulations created seemingly unrealistic sounding stimuli from a stylistic perspective. These factors are described in detail next.

Procedure

Participants attended this experiment in person (rather than online) and were compensated with partial course credit. Prior to the experiment, an outline of the study was explained, and participants voluntarily provided written consent before continuing with the rest of the procedure. Participants completed a demographic questionnaire, then the BFI of personality (John et al., 1991, 2008). The inventory is a five-point Likert scale that consists of 44 items. Participants responded to items phrased as “I am someone who…[statement]” (e.g., “is talkative”), where answers ranged from disagree strongly (1) to agree strongly (5). On scoring, participants were positioned along five personality dimensions: extraversion, agreeableness, conscientiousness, neuroticism, and openness (as discussed in the Introduction). Scores ranged from 1 (low) to 5 (high) for each factor.

Stimuli consisted of 48 manipulations, such as the examples shown in Figures 1, 2, and 3. After each presentation, participants were asked to rate the stimulus based on the following prompt: “You will hear different versions of several short piano pieces. Using the slider on the screen, please tell us how much you like or dislike each version.” The sliding scale recorded responses from 0 (dislike) to 100 (like).

Data Analysis

To measure the effects of MAF manipulation on the dependent variable of preference ratings, separate mixed-factorial analyses of variation (ANOVAs) were performed (using R statistical software; Bates, Mächler, Bolker, & Walker, 2015; R Core Team, 2020) for each personality factor. The between-subject factor related to personality, and was either agreeableness, conscientiousness, extraversion, neuroticism, or openness; each had two levels, high and low. These levels were determined by organizing participants in relation to the median score of each factor, resulting in two groups with equal numbers of participants. The high level within each factor consisted of participants above the median; the low level consisted of participants below the median. This did not result in the skewing of demographic attributes between the groups. For example, male participants were not disproportionately low in agreeableness. Within-subject factors were the five MAFs, which included dynamics (forte, piano), mode (major, minor), piece (Bach, Beethoven, Mozart), register (high, low), and tempo (fast, slow).

Music Acoustic Features

There were significant main effects for four of the five MAFs on preference ratings for stimuli shown in Figure 4. There was a main effect of dynamics, F 1 , 4218 = 50.001 , p < .001 , η p 2 = .012 ; piano was preferred over forte. For mode, F 1 , 4218 = 34.238 , p < .001 , η p 2 = .008 , major was preferred over minor. For piece, F 2 , 4218 = 61.488 , p < .001 , η p 2 = .028 , Bach was preferred over both Beethoven and Mozart. For tempo, F 1 , 4218 = 110.156 , p < .001 , η p 2 = .025 , fast was preferred over slow. Register did not have a significant main effect.

OPEN IN VIEWER

Figure 4.

Notched box plots of preference ratings for the two levels within each MAF and the three levels within piece. Notches represent the standard error around the median; *** p < .001, ** p < .01, * p < .05, ns = not significant.

Agreeableness

There was a main effect for agreeableness, F 1 , 88 = 5.13 , p = .026 , η p 2 = .001 ; participants with high levels of agreeableness rated stimuli higher overall (Figure 5). In addition, there were several significant interactions. For piece and agreeableness (Figure 6, row 4, column 1), Bach excerpts were rated disproportionately higher by participants with high agreeableness, F 2 , 4218 = 3.316 , p = .036 , η p 2 = .002 . For tempo and agreeableness (Figure 6, row 3, column 1), slow tempo was disproportionately less preferred for low agreeableness, F 1 , 4218 = 4.723 , p = .03 , η p 2 = .001 .

OPEN IN VIEWER

Figure 5.

Notched box plots of preference ratings for the two levels within each personality factor. Notches represent standard error around the median; *** p < .001, ** p < .01, * p < .05, ns = not significant.

OPEN IN VIEWER

Figure 6.

Significant interactions between personality and MAF. Personality factors are grouped into high and low levels on the x-axis, and mean preference ratings are shown on the y-axis. The MAF levels are represented by line shading. Error bars indicate standard error around the mean.

Conscientiousness

There was no significant main effect of conscientiousness, but there were significant interactions of conscientiousness with mode, piece, and tempo. For mode and conscientiousness (Figure 6, row 1, column 3), major mode was rated highest with low conscientiousness, and there was no difference in ratings for minor mode, F 1 , 4218 = 7.264 , p = .007 , η p 2 = .002 . For piece and conscientiousness (Figure 6, row 4, column 2), the low conscientiousness group rated Bach over Beethoven, and Beethoven over Mozart. For high conscientiousness, Bach was rated lower than the low conscientiousness group, and Beethoven and Mozart were not significantly different, F 2 , 4218 = 7.852 , p < .001 , η p 2 = .004 . For tempo and conscientiousness (Figure 6, row 3, column 2), slow tempo was rated lowest in the low conscientiousness group and there was no difference in ratings for fast tempo, F 1 , 4218 = 4.174 , p = .041 , η p 2 = .001 .

Extraversion

There was no significant main effect of extraversion. Significant interactions were found, however, with dynamics, mode, register, and tempo. With extraversion and dynamics (Figure 6, row 1, column 1), forte was rated lowest with low extraversion, while there was a difference in ratings for piano, F 1 , 4218 = 4.263 , p = .039 , η p 2 = .001 . For extraversion and mode (Figure 6, row 1, column 4), major mode was rated highest with high extraversion, and there was no difference in ratings for minor mode F 1 , 4218 = 3.998 , p = .046 , η p 2 = .001 . For the interaction between extraversion and register (Figure 6, row 2, column 1), high register was rated highest with high extraversion, while there was no difference in ratings with low extraversion F 1 , 4218 = 6.071 , p = .014 , η p 2 = .001 . There was an interaction between extraversion and tempo (Figure 6, row 3, column 3), with slow tempo least preferred by participants with low extraversion; there was no preference for fast tempo, F 1 , 4218 = 5.809 , p = .016 , η p 2 = .001 .

Neuroticism

There was no significant main effect of neuroticism. Interactions were found with register and neuroticism, and tempo and neuroticism. For register and neuroticism (Figure 6, row 2, column 2), low register was rated lowest with high neuroticism, while there was no difference between ratings for register by participants with low neuroticism, F 1 , 4218 = 5.664 , p = .017 , η p 2 = .001 . For tempo and neuroticism (Figure 6, row 3, column 4), fast tempo was rated highest with low neuroticism, while there was no difference in ratings for slow tempo, F 1 , 4218 = 5.552 , p = .019 , η p 2 = .001 .

Openness

There was no significant main effect of openness, but there were interactions with dynamics and openness, and piece and openness. For dynamics and openness (Figure 6, row 1, column 2), piano dynamics was rated disproportionately higher with high openness, F 1 , 4218 = 4.893 , p = .027 , η p 2 = .001 . For piece and openness (Figure 6, row 4, column 3), Beethoven was rated above Mozart (and below Bach) with high openness, while there was no difference in ratings between Beethoven and Mozart with low openness , F 2 , 4218 = 3.906 , p = .02 , η p 2 = .002 .