On the relevance of music genre-based analysis in research on musical tastes

Abstract

The investigation of the link between personality and musical tastes has led certain psychology researchers to examine the latent dimensions of musical tastes. In this area of research, investigators have largely relied on genre-based analysis, the relevance of which remains unclear. In this study, we examined the impact of changes in the selection of musical items on the identification of musical taste dimensions. Indeed, investigators have employed heterogeneous sets of music genres in prior research. Such a heterogeneity may partly explain why no clearly reproducible structure of musical tastes has emerged in the literature. Based on principal component analysis, our results indicate that the apparent structure of musical tastes is highly affected by even subtle variations in the items selected. Our findings also suggest that the identified structure of musical tastes strongly depends on the social background and cultural capital of respondents. Finally, our results highlight the limitations of the models that interpret the dimensions of musical tastes in terms of intrinsic musical properties.

Keywords

cultural capital musical tastes music genres music dimensions principal component analysis

To date, research on the link between musical tastes and personality has mainly relied on genre-based analysis. Stimulated by the creation of the Musical Preference Scale (Litle & Zuckerman, 1986) and the Short Test of Music Preferences (STOMP; Rentfrow & Gosling, 2003), the use of music genres in this research area has developed greatly since the early 2000s and has led psychology researchers to investigate the underlying dimensions of musical tastes (Brown, 2012; Colley, 2008; Delsing, ter Bogt, Engels, & Meeus, 2008; Gardikiotis & Alexandros Baltzis, 2012; George, Stickle, Rachid, & Wopnford, 2007; Gouveia, Pimentel, Santana, Chaves, & Rodrigues, 2008; Langmeyer, Guglhör-Rudan, & Tarnai, 2012; Schäfer & Sedlmeier, 2009; Zweigenhaft, 2008). Researchers have used dimension-reduction techniques such as principal component analysis (PCA) in order to compute correlations between the dimensions underlying musical tastes and personality scale scores. While some authors have suggested that the dimensions of musical preferences are rather inconsistent from one to study to another (e.g., Dunn, de Ruyter, & Bouwhuis, 2012), others perceived a “considerable degree of convergence between these studies” (Rentfrow, Goldberg, & Levitin 2011, p. 1141).

However, some investigators—including the creators of the STOMP themselves—have progressively pointed out methodological flaws in genre-based analysis. As a result, these investigators have recommended, or opted for, the use of other musical taste indicators, such as excerpts or artists (Rentfrow et al., 2011; Rentfrow et al., 2012; Ferrer, Eerola, & Vuoskoski, 2013; Greenberg, Baron-Cohen, Stillwell, Kosinski, & Rentfrow, 2015; Greenberg et al., 2016). It has been suggested, indeed, that (a) music genres constitute ill-defined categories; (b) the quantity and the quality of relevant music (sub)genres are difficult to specify; (c) music genres’ “ecological” validity is questionable, since artists and songs do not always fall within a unique genre; and (d) the artists and musical pieces ascribed to a given genre are likely to vary as a function of respondents. For instance, the STOMP and its revised form, the STOMP-r, include problematic categories that contravene Rentfrow and Gosling’s intention (2003, p. 1241) to circumscribe the analysis to music genres (i.e., to exclude both sub- and super-genres). As an illustration, the “alternative” category may be viewed by participants as a subtype of rock, as non-mainstream music, or even as a kind of music that transcends the notion of genre. In a similar vein, so-called “religious music” involves a large range of styles, including traditional and contemporary Christian, Hindu, or Islamic musical forms. The “oldies” category includes several music genres (e.g., folk, jazz, rockabilly). “Soundtracks and theme songs” gather pieces that greatly differ with each other: Nino Rota’s “Godfather Waltz” has little in common with Cliff Martinez’s compositions for Soderbergh’s adaptation of Solaris. The “electronica/dance music” category is also questionable, since it associates a super-genre that encompasses a large array of music styles that are not necessarily geared for dancing (e.g., ambient, new wave, trip hop) with a music genre specifically made for dancing. Because of these methodological problems, certain researchers investigated the connections between personality and musical tastes by mobilizing music excerpts, artists, and musical or emotional attributes (e.g., “low tempo,” “sad”; see Finnas, 1987; Schwartz & Fouts, 2003). For example, Rentfrow et al. (2011) chose to rely on excerpts, thereby renewing with the method of investigation employed in the seminal studies of the field (Cattell & Anderson, 1953; Cattell & Saunders, 1954). In these authors’ view, excerpts have higher “ecological” validity than genres and do not require participants to possess label knowledge.

Despite the criticisms addressed to genre-based analysis and the availability of alternative musical taste indicators, the use of music genres has persisted in recent years (see, e.g., Franken, Keijsers, Dijkstra, & ter Bogt, 2017; Fricke & Herzberg, 2017; Vella & Mills, 2017). Such a methodological option is rarely justified, though. In fact, very few articles have empirically tackled the issue of the reliability of genre-based measures (Ferrer et al., 2013). Moreover, investigators have relied on heterogeneous sets of music (sub)genres (Schäfer & Mehlhorn, 2017), the variety of which impedes between-study comparisons. For example, mobilizing several religious music categories enables the identification of religious music dimension(s). By definition, mobilizing a unique religious music category does not. The studies by George and colleagues (2007) and Rentfrow and Gosling (2003) illustrate this point. Whereas the former used 30 (sub)genres, including four religious music categories, and identified two religious music dimensions, the latter employed 14 genres including one religious music category and found religious music to be associated with country, soundtracks, and pop. Furthermore, some researchers have modified the STOMP or the STOMP-r in order to adapt those tests to the specificities of their national context. For instance, Fricke and Herzberg (2017) excluded from the STOMP-r the bluegrass, new age, and reggae categories and added Musical and Volksmusic, two categories that cannot be considered equivalents to the excluded ones. Coupled with the use of different dimension-extraction techniques (e.g., Kaiser rule, parallel analysis), such differences in item selection may explain why no consistent structure of musical tastes emerged in past research.

In the present study, we examined the extent to which subtle changes in the selection of musical items affect the identification of musical taste dimensions. To this end, we carried out a series of PCAs involving various sets of musical (sub)genres from an original list of 40 items. We focused on PCA because it is the statistical test that has been the most frequently used by psychology researchers in their attempt to identify the dimensions underlying individuals’ musical tastes. We relied on two different samples in order to increase the external validity of our study. Our first sample consisted of university students and our second sample of vocational secondary school students. Both samples involved teenagers and young adults to allow us to make comparisons with the existing literature, which has mainly focused on students (Schäfer and Mehlhorn, 2017).

Method

Study sample and recruitment procedure

The present study involved two samples. The first sample comprised 522 students from a Swiss university (M_AGE = 22.72, SD_AGE = 4.05; 68% female). We sent an email that contained a URL to an online survey to all students. Students participated on a voluntary basis. The response rate was 14%. The second sample involved 185 high schoolers from a vocational secondary school located in France (M_AGE = 17.20, SD_AGE = 1.01; 59% female). We surveyed all eleventh- and twelfth-grade classes. We administrated our questionnaire in the classrooms. Both samples included French-speaking participants only.

Musical taste inventory

Respondents reported their degree of appreciation of 40 music genres and subgenres (e.g., rock, alternative rock) using a five-point rating scale (from 1 for I dislike very much to 5 for I like very much). A supplementary response option allowed participants to indicate that they did not know the music (sub)genre in question. We selected those 40 (sub)genres based on a preliminary survey in which 15 French and 15 Swiss undergraduate students were asked to specify which music genres and subgenres they regularly listened and never listened to. We found that 47 categories were mentioned at least three times (i.e. by at least 10% of the pilot sample). Since some participants explicitly discriminated American from French rap, we used such a distinction and included those two categories in our music inventory. We amalgamated “black metal,” “death metal,” and “trash metal” into the category “extreme metal.” Because of its generality and its overlap with “dance,” “electropop,” “house,” and “techno,” we excluded the category “electro.” We neglected the categories “chill” and “minimal” (three occurrences for each) in order to not over-represent electronic music in our inventory. We also excluded the categories “indie” and “commercial” because they cover several music genres. The Appendix displays the list of items included in our inventory and the corresponding means and standard deviations in both samples. It also reports, for each (sub)genre, the rate of participants that indicated that they were not familiar with the corresponding label. The same inventory was administered to both samples. We note that our pilot study did not involve high-schoolers. Because there is evidence that musical tastes remain consistent from adolescence to early adulthood (Delsing et al., 2008; Mulder, ter Bogt, Raaijmakers, Gabhainn, & Sikkema, 2010), we assumed that involving high-schoolers in our pilot study was not needed. Nevertheless, we asked high-schoolers to indicate which music genres and subgenres they regularly listened to and never listened to. A vast majority of the responses referred to music categories that were already included in our inventory. Although other categories appeared, they were mentioned by a very small proportion of respondents. For instance, “flamenco,” “Guggenmusik,” “traditional Turkish music,” and “tribe” were cited once, “dancehall” and “slam/spoken words” were cited twice, and “Afro-trap” was cited thrice.

Music-genre sets

In order to examine the extent to which applying subtle modifications in the item selection influences PCA results, we relied on six different music-genre sets.

The first set involved the genres included in the STOMP (Rentfrow & Gosling, 2003), with one exception: “soundtracks.” This category was not mentioned by the respondents to our pilot survey. Moreover, its vagueness was potentially problematic and led us to neglect it. In addition, we did not employ the “electronica/dance” category, which problematically combines non-dance-oriented music (e.g., ambient, post-punk) with dance. Instead, we created a score for electronic dance music (EDM) appreciation by calculating the mean level of appreciation of dance, house, and techno. We focused on EDM because our data indicate that non-dance-oriented electronic genres (e.g., new wave) were unknown to a large part of our two samples (see the Appendix). Finally, since our inventory mobilized the categories “American rap” and “French rap,” we created a global score for rap appreciation by computing the mean level of appreciation of these two subtypes. We used the STOMP as point of comparison because this test has been abundantly mobilized in the literature.

The second set was similar to the first one but involved taste for dance instead of taste for EDM. This set is more consistent with Rentfrow and Gosling’s goal to focus on musical genres rather than on musical sub-genres or musical super-genres. It allowed us to assess how a very subtle modification in the selection of music genres influenced the emerging structure of musical tastes.

In the third set, we excluded the “alternative rock” category, which is redundant with the “rock” category. We instead added the “R&B” category, which refers to a fashionable genre among teenagers and young adults.

The fourth set adapted the STOMP to local specificities and systematized the inclusion of pairs of (sub)genres involved in the STOMP with the “alternative” and “rock” categories. We thus replaced country and US folk by French variété and international variété, two popular genres that can be considered equivalent to soft adult contemporary music. The retained pairs of (sub)genres were alternative rock and rock, blues and jazz, classical and opera, conscious rap and rap, dance and house, extreme metal and metal, and the two subtypes of variété.

The fifth set gathered the (sub)genres (n = 30) that were known by at least two-thirds of the participants of both samples. We created this set in order to observe whether (potentially) consistent between-item associations previously identified were retrieved when enlarging the item selection.

The sixth set of (sub)genres (n = 20) was used only for the high-schooler sample. To reduce the number of items (compared to the previous set) and provide sufficient musical variety, this set included pairs of electronic, so-called “highbrow,” Latin, metal, rap, rock, and variété (sub)genres. The corresponding items were selected on a twofold basis: the degree of label knowledge and the degree of appreciation. For instance, we neglected opera because it was the most unknown “highbrow” genre. We included electropop and house because they were the most and the least appreciated electronic subgenres. Moreover, because a non-negligible part of the members of the high-schooler sample were French Arabs and Blacks, we also included African music, raï, and zouk in the count. The sociological literature indeed highlighted that ethnic features modulate musical tastes (Robinson et al., 1985). Finally, we added pop, reggae, and R&B, because these genres were among the most appreciated in our high-schooler sample. The reasons that led us to perform this supplementary PCA are described below, in the corresponding section.

Data analyses

We carried out PCAs with promax rotation—a type of rotation used when between-component independence is not assumed. In order to estimate the number of components to be extracted, we performed parallel analyses (Horn, 1965). Parallel analysis enables investigators to avoid both under- and over-extraction and to optimize the reliability of the components (O’Connor, 2000; Zwick & Velicer, 1986). This technique is considered more reliable than the Kaiser criterion (Kaiser, 1960), which selects components based on eigenvalues higher than 1, and than the scree test (Cattell, 1966), which consists in graphing the eigenvalues and retaining those that appear to precede the point of inflexion (Costello & Osbourne, 2005; O’Connor, 2000; Thompson & Daniel 1996). Because our survey allowed participants to respond “I do not know this (sub)genre,” we treated such cases with the pairwise-deletion technique (Van Ginkel, Kroonenberg, & Kiers, 2014). We used the Bartlett’s test of sphericity and the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy as indicators of suitability. All Bartlett’s test results were significant. All KMO measures indicated that our data were suitable for PCA (Hutcheson & Sofroniou, 1999; Kaiser, 1974). Those values are reported in the following section.

Results

Given the high number of analyses involved in our study, we limited ourselves to concisely reporting hereafter the main findings related to each performed PCA.

Sample 1: University students

Table 1 displays the results pertaining to the PCA involving the genres included in the STOMP, soundtracks excluded and with electronic dance music (EDM) replacing electronica/dance. We found a five-component solution accounting for 69.73% of the variance. The first component (C1) accounted for 23.78% of the variance. One between-component correlation higher than .25 was found, between C2 and C4 (r = .264). While all main loadings were positive and higher than .6, metal presented another high (negative) loading on C3 and rap showed problematic cross-loadings on C1 and C5. C1 clustered “classic” Afro-American genres, and C2, rock and metal styles. C3 is hardly interpretable, especially for a Swiss sample, since it gathers pop, country, and US folk music. C4 combined classical and religious music, which might suggest that respondents considered as “religious music” the sacred forms of European Christian music, not styles such as Gospel or Hindu music. Finally, C5 involved EDM and rap, two relatively recent and so-called “urban” genres.

Table 1.

Musical taste component scores: First set, university student sample.

	C1	C2	C3	C4	C5
Blues	.867	.050	−.022	.019	−.064
Jazz	.825	−.014	−.061	.139	.071
Soul	.817	−.027	.107	−.089	.150
Rock	.016	.897	.150	−.121	.076
Alt rock	.059	.868	.115	−.175	.043
Metal	−.113	.728	–.403	.210	−.002
Pop	−.212	−.057	.785	−.134	.291
Country	.151	−.002	.612	.255	−.240
Folk	.210	.206	.601	.080	−.143
Religious	−.062	−.167	.034	.880	.155
Classical	.151	−.003	.025	.724	.054
EDM	−.129	.160	.156	.314	.821
Rap	.386	−.058	−.183	−.119	.679

Notes: Loadings > .5 are in bold; loadings between .4 and .5 are in italics. Bartlett’s test of sphericity: χ²(78) = 1235.66, p < .001; KMO = .663. EDM = Electronic Dance Music.

Table 2 reports the results pertaining to the PCA involving the genres included in the STOMP, soundtracks excluded and with dance replacing electronica/dance. We found a four-component solution accounting for 61.69% of the variance. C1 accounted for 23.8% of the variance. Two between-component correlations higher than .25 were found, between C1 and C4 (r = .258) and between C2 and C4 (r = .332). Thus, using dance instead of EDM resulted in finding a more synthetic component solution. Compared with Table 1, Table 2 involved two other main differences: contrary to EDM, dance clustered with pop and not with rap, which in turn appeared to be (weakly) associated with blues, soul, and jazz. In addition, two genres showed problematic cross-loadings. US folk did not load well, since all the corresponding loadings were lower than .4 and the loadings on C2 and C3 were very close. Country exhibited one loading slightly higher than .4 (C3) and one loading slightly lower than 0.4 (C4). In brief, C1 clustered the Afro-American trio and rap; C2, rock and metal styles; C3, pop and dance; C4, classical and religious music.

Table 2.

Musical taste component scores; Second set, university student sample.

	C1	C2	C3	C4
Blues	.862	.092	−.169	.045
Soul	.844	−.015	.056	−.093
Jazz	.822	−.026	−.141	.137
Rap	.465	−.264	.029	−.269
Rock	−.005	.910	.076	−.142
Alt rock	.045	.886	.015	−.191
Metal	−.163	.632	−.461	.146
Pop	−.154	.020	.827	−.104
Dance	−.090	−.061	.704	.052
Country	.123	.178	.405	.386
US folk	.201	.367	.390	.188
Religious music	−.103	−.252	−.003	.883
Classical	.101	−.048	−.060	.745

Notes: Loadings > .5 are in bold; loadings between .4 and .5 are in italics. Bartlett’s test of sphericity: χ²(78) = 1255.15, p < .001. KMO measure of sampling adequacy = .671.

Table 3 reports the results pertaining to the PCA involving the genres included in the STOMP, soundtracks and alternative excluded, R&B added. We found a four-component solution accounting for 61.25% of the variance. C1 accounted for 22.65% of the variance. One between-component correlation higher than .25 was found, between C3 and C4 (r = .285). Table 3 reveals that removing the “alternative” category severed the rock-metal component that we found thus far. Here, rock was associated with folk and country, but not with metal, which negatively loaded on a component combining pop, R&B, and dance. In addition, rap (see C1 and C3) and pop (see C2 and C3) exhibited problematic cross-loadings. C1 associated, again, blues, jazz, and soul with rap, which did not load very well. C2 reflected taste for pop, R&B, dance, and distaste for metal. C3 clustered rock, US folk, and country; C4, classical and religious music.

Table 3.

Musical taste component scores; Third set, university student sample.

	C1	C2	C3	C4
Blues	.848	−.177	.222	−.026
Soul	.805	.133	.120	−.074
Jazz	.796	−.104	.062	.151
Rap	.420	.269	−.355	−.070
Pop	−.210	.753	.376	−.088
R&B	.218	.699	−.224	−.056
Dance	−.143	.666	.231	.131
Metal	−.036	–.615	.300	.051
Rock	.084	−.206	.754	−.225
US folk	.204	.167	.693	.026
Country	.129	.244	.606	.180
Religious music	−.073	.043	−.116	.881
Classical	.125	−.105	.036	.752

Notes: Loadings > .5 are in bold; loadings between .4 and .5 are in italics. Bartlett’s test of sphericity: χ²(78) = 1391.97, p < .001. KMO measure of sampling adequacy = .665.

Table 4 reports the results pertaining to the PCA adapting the STOMP to local specificities and including seven pairs of allegedly close (sub)genres. We found a five-component solution accounting for 70.15% of the variance. C1 accounted for 19.92% of the variance. One between-component correlation higher than .25 was found, between C1 and C3 (r = −.291). Interestingly, the electronic genres were the only ones to load on different components. Dance loaded on the variété component and house on the metal component. Importantly, metal and rock loaded separately, here. It should also be noted that blues, classical, jazz, and opera clustered into a single component, that could be considered to involve “highbrow” genres. Finally, metal (see C3 and C5) and house (see C4 and C5) exhibited problematic cross-loadings. In sum, we found a variété component including dance (C1), a “highbrow” component (C2), a rock component (C3), a rap component (C4), and a metal component including house (C5).

Table 4.

Musical taste component scores: Fourth set, university student sample.

	C1	C2	C3	C4	C5
Inter. variété	.908	−.036	.176	.001	−.205
French variété	.894	.015	.243	.006	−.156
Dance	.565	.035	−.056	.184	.112
Jazz	−.190	.767	.079	.180	−.225
Classical	.174	.761	−.076	−.194	.189
Opera	.193	.731	−.161	−.168	.297
Blues	−.190	.650	.223	.180	−.243
Rock	.230	.008	.891	−.007	.159
Alt rock	.146	.004	.852	−.002	.091
Rap	.108	−.026	−.107	.884	.057
Conscious rap	.029	−.014	.110	.859	.196
Extreme metal	−.247	−.010	.278	.001	.758
Metal	−.208	−.017	.425	.004	.749
House	.149	.036	−.229	.393	.577

Notes: Loadings > .5 are in bold; loadings between .4 and .5 are in italics. Bartlett’s test of sphericity: χ²(91) = 1916.43, p < .001; KMO = .600.

Table 5 reports the results pertaining to the PCA including 30 (sub)genres. We found a six-component solution accounting for 63.44% of the variance. C1 accounted for 19.17% of the variance. We found three between-component correlations higher than .25, between C1 and C3 (r = −.258), C3 and C4 (r = .297), and C3 and C5 (r = .254). C1 exclusively clustered rock and metal (sub)genres. C2 underlined taste affinities between rap styles, African music, and reggae, which all constitute genres that have ethnic and social connotations. C3 gathered variété, pop, and Latin music, which can be considered soft, light forms of music. C4 involved the already encountered trio of blues, jazz, and soul. C5 comprised all the electronic genres included in the count, with dance showing the lowest loading. Finally, C6 gathered classical, opera, and religious music. Importantly, enlarging the music-genre selection did not alter the associations between blues, jazz, and soul and between classical and religious music that we consistently found thus far. However, Table 5 involved some problematic cases. Country presented two loadings slightly lower than .4 on C3 and C4 and was not assignable to a specific component. R&B exhibited cross-loadings slightly higher and slightly lower than .4 (see C2 and C3). Latin music and reggaetón loaded on C3, but also presented a loading higher than .3 on C2. The same applied to African music (see C2 and C4) and dance (see C5 and C3).

Table 5.

Musical taste component scores: Fifth set, university student sample.

	C1	C2	C3	C4	C5	C6
Hard rock	.864	.033	−.026	−.079	−.007	.029
Metal	.830	.117	−.119	−.209	−.026	.119
Rock	.794	−.081	.161	.209	.057	−.177
Symphonic metal	.784	.074	.069	−.269	−.070	.206
Punk	.757	.154	−.002	.014	.103	−.034
Alt rock	.718	−.065	.047	.272	.074	−.240
1960–70s rock	.701	−.036	.026	.370	.051	−.148
Extreme metal	.694	.147	−.108	−.231	−.060	.183
Conscious rap	.229	.805	−.062	.108	.025	−.145
French rap	−.058	.789	.039	.024	.144	−.059
American rap	−.056	.733	−.138	.154	.217	−.218
Reggae/ska	.247	.664	.200	.232	−.082	−.057
African music	.015	.515	.203	.377	−.133	.207
R&B	−.237	.405	.392	.042	.135	−.109
International variété	.055	−.014	.843	−.195	.020	.180
French variété	.129	−.018	.759	−.170	−.033	.242
Latin music	−.174	.318	.665	−.066	−.111	.043
Reggaetón	−.174	.345	.631	−.161	−.080	−.049
Pop	−.033	−.227	.610	−.027	.272	−.108
Country	.148	−.159	.398	.349	−.155	.236
Jazz	−.150	.109	−.291	.886	.019	.303
Blues	.023	.158	−.204	.863	−.129	.174
Soul	−.075	.287	−.047	.794	−.070	.050
Techno	.042	.227	−.169	−.045	.805	.180
House	.069	.166	−.125	−.140	.785	.231
Electropop	.029	−.057	.175	−.086	.770	−.059
Dance	−.023	−.034	.366	−.020	.566	.112
Opera	−.015	−.149	.116	.275	.134	.813
Religious music	.004	−.052	.236	.055	.076	.794
Classical	−.011	−.195	.064	.368	.132	.738

Notes: Loadings > .5 are in bold; loadings between .4 and .5 are in italics. Bartlett’s test of sphericity: χ²(435) = 5529.80, p < .001. KMO = .813.

Sample 2: Vocational high school students

Table 6 displays the results pertaining to the three PCAs involving our first three sets of music categories. Those sets comprised the genres included in the STOMP: (a) soundtracks excluded and with EDM replacing electronica/dance; (b) soundtracks excluded, and with dance replacing electronica/dance; and (c) soundtracks and alternative excluded, R&B added. In each case, parallel analysis indicated that a one-component solution represented the optimal way to sum up the data. Table 6 reports the component matrix for each unrotated PCA. The only point of note, here, regards the negative, albeit low, loadings related to rap.

Table 6.

Musical taste component scores: First, second, and third sets, high-schooler sample.

First set		Second set		Third set
Alt rock	.815	Alt rock	.817	Blues	.772
Rock	.772	Rock	.776	Jazz	.712
Blues	.750	Blues	.749	US folk	.708
EDM	.722	Metal	.699	Rock	.691
Metal	.706	US folk	.694	Classical	.660
US folk	.706	Jazz	.676	Soul	.659
Jazz	.663	Classical	.643	Metal	.602
Classical	.625	Soul	.611	Country	.594
Soul	.615	Country	.572	Pop	.444
Country	.554	Pop	.380	Dance	.430
Pop	.377	Dance	.353	R&B	.241
Rap	−.173	Rap	−.154	Religious music	.085
Religious music	.026	Religious music	.036	Rap	−.059
% explained variance	38.70		35.90		31.68
KMO	.788		.764		.764

Notes: All Bartlett’s tests were significant at p < .001. EDM = Electronic Dance Music.

Table 7 reports the results pertaining to two PCAs. The first PCA (see the three columns on the left) involved the set adapting the STOMP to local specificities and including seven pairs of allegedly close (sub)genres. We found a two-component solution accounting for 50.68% of the variance. C1 accounted for 34.77% of the variance. The between-component correlation was equal to .337. Results are hardly interpretable but highlight an opposition between rock and metal (sub)genres and all the other categories, house excepted. Numerous problematic cross-loadings were found (e.g., see rap, blues, jazz, and classical). Compared with the previous sample, it is worth noting that involving pairs of allegedly close (sub)genres did not result in finding corresponding, specific components.

Table 7.

Musical taste component scores: Fourth and fifth sets, high-schooler sample.

Fourth set			Fifth set
	C1	C2		C1′	C2′
Metal	.941	−.128	Metal	.890	−.196
Alt rock	.926	−.019	Hard rock	.889	−.194
Extreme metal	.888	−.166	Alt rock	.864	−.089
Rock	.814	.043	Extreme metal	.825	−.181
Rap	−.485	.403	Symphonic metal	.813	−.175
House	.425	.300	Rock	.813	−.003
French variété	−.203	.770	1960–70s rock	.790	.045
Inter. variété	−.180	.750	Punk	.604	.083
Conscious rap	−.036	.523	French rap	−.520	.389
Blues	.414	.521	House	.483	.218
Dance	−.092	.518	Blues	.462	.348
Jazz	.335	.489	Classical	.452	.206
Classical	.353	.470	Techno	.424	.283
Opera	.259	.456	Electropop	.418	.323
–	–	–	Opera	.348	.214
–	–	–	Country	.346	.244
–	–	–	Latin music	−.166	.690
–	–	–	Inter. variété	.041	.626
–	–	–	French variété	.034	.572
–	–	–	Reggaetón	.162	.567
–	–	–	African music	−.226	.519
–	–	–	Soul	.272	.518
–	–	–	R&B	−.090	.516
–	–	–	Reggae/ska	.228	.509
–	–	–	Pop	.126	.508
–	–	–	American rap	−.218	.450
–	–	–	Conscious rap	.059	.436
–	–	–	Dance	.082	.409
–	–	–	Jazz	.382	.402
–	–	–	Religious music	−.115	.315
Bartlett’s test	χ²(91) = 902.13; p < .001		Bartlett’s test	χ²(435) = 1752.72, p < .001
KMO	.697		KMO	810

Notes: Loadings > .5 are bolded; loadings between .4 and .5 are italicized.

The second PCA (see the three columns on the right) involved the set including 30 (sub)genres. We found a two-component solution accounting for 39.56% of the variance. C1′ accounted for 26.65% of the variance. The between-component correlation was equal to .226. Interpreting this two-component solution is arduous. Opera, country, and religious music appeared to be component-free. Jazz exhibited loadings oscillating around 0.4. While French rap negatively loaded on C1′ and presented a .39 loading on C2′, American rap and conscious rap positively loaded on C2′, albeit poorly. Several items loaded only weakly (e.g., blues, classical, dance, electropop). The only clear pattern here refers to the high loadings of rock and metal items on C1′. However, the loadings related to blues, classical, and electropop prevent C1′ from being considered a proxy for a hardcore, rock, and metal dimension. Finally, contrary to the corresponding PCA among university students, electronic music genres did not load on the same component.

In order to assess whether the structure of high-schoolers’ musical tastes was at best twofold, we performed a supplementary PCA involving a set of 20 (sub)genres, the selection principles of which are reported in the previous section. Table 8 displays a four-component solution accounting for 60.95% of the variance. C1 accounted for 25.50% of the variance. Two between-component correlations higher than .25 were found, between C2 and C3 (r = .314) and between C2 and C4 (r = .325). Table 8 reveals that a greater number of genres and subgenres does not automatically entail a better understanding of the structure of musical tastes. Contrary to Table 7, Table 8 highlights some clear patterns indeed. C2 gathers (sub)genres that have ethnic connotations and may be perceived by the respondents as more danceable than the others. C3 clustered “pop,” “soft,” and “unsophisticated” (sub)genres. C4 exclusively related to rap, which has ethnic connotations as well but may be considered as less danceable than the (sub)genres forming C2. C1, however, is hardly interpretable, since it associated rock and metal (sub)genres with house, jazz, and classical. In addition, R&B presented very poor loadings, and reggae and electropop exhibited problematic cross-loadings.

Table 8.

Musical taste component scores: sixth set, high-schooler sample.

	C1	C2	C3	C4
Hard rock	.898	−.098	−.063	−.124
Metal	.879	−.054	−.104	−.122
Extreme metal	.840	−.022	−.172	−.048
Rock	.783	−.091	.111	.013
House	.574	.183	.142	−.174
Jazz	.571	.155	−.010	.370
Classical	.487	−.164	.156	.201
African music	−.165	.880	−.147	−.068
Zouk	−.028	.723	.183	−.112
Reggaetón	.380	.704	−.072	.098
Raï	−.269	.689	−.183	.069
Latin music	−.146	.659	.292	−.041
Reggae/ska	.426	.555	.015	.025
French variété	−.098	−.045	.908	−.075
International variété	−.068	.099	.895	−.126
Pop	.134	−.016	.576	.223
Electropop	.413	−.141	.521	−.005
R&B	−.047	.274	.302	.266
American rap	.072	.011	−.077	.854
French/francophone rap	−.304	−.093	.026	.782

Notes: Loadings > .5 are in bold; loadings between .4 and .5 are in italics. Bartlett’s test of sphericity: χ²(190) = 1208.55, p < .001. KMO = .723.

Discussion

Our research goal was to assess the extent to which the structure of musical tastes in genre-based surveys is affected by modifications in the set of (sub)genres under study.

Our results indicate that applying even minor modifications in the selection of musical items produces important differences in the emerging latent structure of musical tastes. At least, this conclusion can be drawn from the analysis of the university student sample. As an illustration, the data pertaining to that sample showed that mobilizing the “EDM” or the “dance” category resulted in finding a different number of components (five vs. four) and different between-item associations. Given the overlap between EDM and dance, this result prompts us to interpret the obtained component solutions with great caution. In a similar vein, our data indicated that removing the “alternative rock” category—which can be considered redundant with the “rock” category—severed the association between rock and metal and revealed unseen links between rock, folk, and country. Overall, while the number of components did not massively fluctuate from one PCA to the other, numerous between-item associations substantially varied. For example, depending on the cases, rap and/or its subtypes appeared to be associated with (a) EDM; (b) blues, jazz, and soul but not with dance; (c) reggae and African music but not with blues, jazz, and soul or with any electronic music genre; or (d) to form a single component. Such variations impede our ability to interpret the obtained components in terms of underlying music dimensions. Even apparently strong associations dislocated when applying subtle modifications in the item selection. The cases of the rock/metal and dance/pop connections are emblematic in that respect. Interestingly, our data also point out that using categories such as “electronic dance music” may be too vague and may not account for nodal distinctions within that taxon. We indeed observed that dance and house did not systematically load on the same component. Importantly, we found only two consistent associations throughout our analyses related to the university student sample. They refer to the connections between classical and religious music and between blues, jazz, and soul. Notably, none of these associations was found by Rentfrow and Gosling (2003), and we did not retrieve them within the high-schoolers’ distribution of musical tastes. Thus, our findings suggest that interpreting component solutions in terms of general, underlying music dimensions may be hazardous and speculative. This being said, the results related to high-schoolers may be perceived as providing counter-evidence to that conclusion. Data pertaining to that sample appeared to be impermeable to minor changes in the item selection, since we mainly found one- or two-component solutions when examining the structure of high-schoolers’ musical tastes. In our view, however, those component solutions may reflect an inadequacy between the selected items and the participants’ tastes rather than an immunity to modifications in the item selection. The fact that we found a four-component solution when mobilizing a specific set of music (sub)genres that combined selection criteria such as main tastes and distastes, label knowledge, and ethnic features, advocates this view. In sum, data related to both samples suggest, though in different ways, that applying even subtle modifications in the music-genre selection is likely to substantially alter PCA results. As a consequence, it may be counter-productive to consider the component-solutions found in the literature general, robust indicators of underlying music dimensions. The study of the links between musical tastes and personality should therefore involve more reliable taste indicators and/or methods of analysis than those employed thus far.

Interestingly, our analyses related to the high-schooler sample revealed unusual patterns of results. The various one- and two-component solutions pertaining to that sample strongly contrast with the four-, five- or six-component solutions obtained when examining the tastes of university students and with the component solutions generally found in the literature (Brown, 2012; Colley, 2008; Delsing et al., 2008; Franken et al., 2017; Gardikiotis & Alexandros Baltzis, 2012; George et al., 2007; Gouveia et al. 2008; Langmeyer et al., 2012; Rentfrow & Gosling, 2003; Schäfer & Sedlmeier, 2009; ter Bogt, Raaijmakers, Vollebergh, van Wel, & Sikkema, 2003; Vella & Mills, 2017; Zweigenhaft, 2008). Accounting for such differences is difficult because several factors, either sociodemographic or methodological, probably contribute to these atypical structures of musical tastes. In particular, one might assume that the mean age of the surveyed high-schoolers and the administration of our questionnaire in the classrooms partly explained our unusual results. However, because psychological researchers have pointed out that (a) the structure of musical tastes is relatively age-invariant (Rentfrow et al., 2011; Bonneville-Roussy, Rentfrow, Xu, & Potter, 2013), and (b) musical tastes are consistent during adolescence and from adolescence to early adulthood (Delsing et al., 2008; Mulder et al., 2010), effects of age are rather unlikely here. Similarly, it is unlikely that administrating our questionnaire in the classrooms played a determinant role. Delsing and colleagues (2008) did the same with Dutch teenagers and found a four-component solution in their study involving “only” 11 music genres. In our estimation, the singularity of our findings may be mainly due to two factors. First, the common use of the Kaiser criterion, coupled or not with other techniques, in the definition of the number of component(s) to be extracted. Although the Kaiser criterion has long been showed to involve over-extraction (O’Connor, 2000; Zwick & Velicer, 1986), a number of psychology researchers have continued to rely on it (e.g., Gouveia et al. 2008; Vella & Mills, 2017). This state of affairs may partly explain why one- or two-component solutions have not been found in the psychological literature dedicated to the structure of musical tastes. Second, the social background and the cultural capital of the respondents. Bourdieu (1984) distinguished between the acquired dimension of cultural capital (indexed by individuals’ own education) and its inherited dimension (indexed by parents’ occupation and/or education). Within this framework, vocational high-schoolers possess a smaller volume of acquired cultural capital than general high-schoolers and university students. Furthermore, sociologists have shown that differences in acquired cultural capital were associated with differences in cultural tastes and practices (Bourdieu, 1984), especially in terms of taste scope (Peterson, 1992). Our findings related to high-schoolers echo such dynamics. Indeed, our raw data indicates that vocational high-schoolers reported to like, on average, only nine of the 40 (sub)genres included in our music inventory. Five of these nine categories were associated with a mean degree of appreciation very close to the neutral score (i.e. 3 on a five-point rating scale). Those nine categories, moreover, do not cover a large array of music styles, since they refer, inter alia, to three rap subtypes and two variété subtypes. Comparatively, university students reported to like, on average, 27 of the same 40 (sub)genres. Thus, the singularity of our results may be the consequence of the sociological singularity of our high-schooler sample. The great number of genres disliked by vocational high-schoolers and the corresponding levels of distaste may explain why (a) we mainly found one- and two-component solutions, and (b) subtle modifications in the item selection did not alter PCA results related to that sample. Because most studies in psychology of music have surveyed university students (Schäfer and Mehlhorn, 2017) or volunteers (i.e., mostly people interested in music), and because university students from the lower classes are likely to compensate their low inherited cultural capital with a relatively high acquired cultural capital (Bourdieu, 1984), people endowed with low acquired cultural capital may have been under-represented in psychological research on music. This may explain why one- or two-component solutions have not been found in the literature. Although only complementary studies would enable us to comprehensively address these issues, our study suggests that variables such as cultural capital should not be neglected in the examination of the determinants of the structure of musical tastes—and in the creation of musical-taste inventories. Given the links between openness and academic performance (Poropat, 2009, 2014), integrating psychological (e.g., personality) and sociological (e.g., cultural capital) variables may improve our understanding of musical tastes.

Importantly, results related to the university student sample (see the first two sets of music genres) do not corroborate the structure of musical preferences that Rentfrow and Gosling (2003) delineated when administrating the STOMP. While we identified, as these authors did, a component including rock, alternative rock, and metal, this is the sole finding that both studies have in common. Rentfrow and Gosling (2003) found that blues, jazz, classical, and folk formed a component that they named “reflective and complex,” because those are “genres that seem to facilitate introspection and are structurally complex” (p. 1241). We found that blues and jazz consistently constituted a component with soul, but not with classical and folk, which loaded on different components. This finding suggests that the question of intrinsic “structural complexity” may be ancillary. Moreover, we do not see reasons to consider those genres as stronger facilitators to introspection than, for instance, religious music. Similarly, we did not retrieve the “upbeat and conventional” component found by Rentfrow and Gosling (2003). This component comprised four genres that “emphasize positive emotions and are structurally simple” (p. 1241), namely country, soundtracks, religious music, and pop. Such mismatch cannot be solely imputed to the exclusion from our study of the “soundtracks” category—that we considered far too inaccurate and that refers to pieces that do not “emphasize positive emotions,” like in horror movies, and pieces that are not “structurally simple” (e.g., listen to Danny Elfman’s orchestral compositions). Indeed, we found religious music to form a two-item component with classical music. This result further questions the consistency of the authors’ rationale based on structural complexity, since this component gathers a “structurally complex” (i.e. classical) and a “structurally simple” genre (i.e. religious music). Furthermore, although our data accounted for connections between country and pop, the links in question were very weak. Notably, we found a stronger association between country and folk than between country and pop. Again, our findings cast doubts upon the validity of a rationale based on degrees of “structural complexity,” a concept that Rentfrow and Gosling (2003) did not define. Finally, our results showed that rap was associated with EDM but not with dance. Thus, we did not systematically retrieve the “energetic and rhythmic” component that Rentfrow and Gosling (2003, p. 1242) identified and that involves “genres that are lively and often emphasize the rhythm.” It should be noted, in passing, that genres such as hard rock or metal also emphasize the rhythm and that genres assignable to the “electronica” category (e.g., ambient) do not. Although the discrepancies between Rentfrow and Gosling’s results and ours are likely national-dependent, these discrepancies question the reliability of the authors’ line of interpretation. Again, our findings suggest that the component solutions obtained via PCAs should be interpreted with the utmost caution.

Analogous conclusions can be drawn from the comparison between our results and the results derived from the MUSIC model developed by Rentfrow and colleagues (Bonneville-Roussy et al., 2013; Rentfrow et al., 2011; Rentfrow et al., 2012). Used by its creators in both excerpt- and genre-based framework, the MUSIC model discriminates between five music dimensions: “mellow” (i.e. smooth, quiet, and slow; e.g., R&B and soft rock), “unpretentious” (i.e. not loud, distorted, nor fast; e.g., country and folk), “sophisticated” (instrumental, not electric, distorted, nor loud; e.g., classical and traditional jazz), “intense” (i.e. electric, distorted, loud, percussive; e.g., metal and rock), and “contemporary” (i.e. percussive, electric, not sad; e.g., electronica, Latin, and rap). Our results did not reflect the MUSIC model. In particular, we did not retrieve the “contemporary” dimension, since rap, Latin, and electronic (sub)genres loaded on distinct components. Because these (sub)genres were well known by the participants and may be more easily recognizable than soul, R&B, or alt rock, those are solid findings that question the general validity of the MUSIC model. Indeed, the sole loadings of rap, Latin, and electronic (sub)genres on different components almost totally reshape the structure that the MUSIC model delineates. In addition, depending on the cases, we found rock to be associated with country, R&B with dance, house with metal, or classical and jazz with rock. In other words, we did not retrieve the “mellow,” “unpretentious,” and “sophisticated” dimensions. Our results therefore suggest that the dimensions involved in the MUSIC model are not robust enough to be generalized.

Conclusion

The present study examined the relevance of the genre-based analyses commonly used in psychological research on musical tastes. We found that even subtle modifications in the item selection sufficed to substantially alter PCA results and identify antithetic patterns in the structure of musical tastes. Given the inconsistency of the obtained components, interpreting them as reflecting general music dimensions is problematic. Moreover, our study suggests that social background in general, and cultural capital in particular, may markedly affect the structure of musical tastes. Such variables should not be neglected in research on musical tastes. Finally, our findings indicate that interpreting the dimensions of music genres in terms of their intrinsic properties is probably misleading. All in all, the present study questions the relevance and the validity of the genre-based analyses ordinarily employed in psychological research on musical tastes. It also suggests that research on musical tastes may benefit from a concomitant examination of sociological and psychological variables.

Footnotes

Appendix

Appendix.

Music inventory statistics: Label knowledge and mean scores of appreciation (five-point rating scale).

	Sample 1 (n = 522)			Sample 2 (n = 185)
	DK (%)	M	SD	DK (%)	M	SD
1960–1970s progressive rock	17.05	3.89	1.13	22.70	1.77	1.31
African music	6.70	3.30	1.14	5.95	2.79	1.53
Alt rock	21.07	4.01	1.05	24.86	1.69	1.21
American rap/hip-hop	1.34	3.48	1.28	1.62	4.27	1.14
Blues	1.15	3.65	1.02	25.41	1.97	1.07
Classical	0.00	3.77	1.02	4.32	1.80	1.13
Conscious rap	10.54	3.44	1.36	15.68	3.03	1.55
Country	0.38	3.07	1.19	10.81	1.81	1.06
Dance	0.19	3.33	1.14	4.32	2.84	1.35
Electropop	2.30	3.34	1.21	11.89	2.87	1.43
Experimental rock/art rock	39.08	3.46	1.19	24.32	1.66	1.20
Extreme metal	5.36	1.93	1.27	15.68	1.46	0.98
French/francophone variété	2.68	3.37	1.23	5.95	3.07	1.42
French/francophone rap/hip-hop	1.72	3.07	1.36	1.08	4.44	1.09
Gypsy jazz	44.83	3.48	1.19	34.59	1.64	1.10
Hard rock	1.53	2.87	1.41	13.51	1.65	1.09
House	8.05	2.66	1.36	30.81	2.09	1.39
International variété	10.73	3.34	1.06	8.65	3.08	1.43
Jazz	0.38	3.55	1.16	7.57	1.89	1.15
Latin music	1.92	3.35	1.29	6.49	3.13	1.49
Metal	1.53	2.60	1.47	8.11	1.59	1.11
New age/atmospheric	46.93	3.12	1.20	56.76	1.74	1.17
New wave/goth/post-punk	43.68	2.87	1.31	54.59	1.60	1.04
Opera	0.96	2.89	1.21	8.11	1.48	1.01
Pop	0.00	3.95	1.00	3.78	3.44	1.37
Punk rock	8.81	3.04	1.23	17.30	1.80	1.13
R&B	3.07	3.05	1.25	9.73	3.62	1.46
Raï	61.88	2.58	1.10	18.92	3.04	1.56
Rap-metal	33.33	2.64	1.35	26.49	2.10	1.38
Reggae/ska	3.26	3.15	1.26	13.51	2.84	1.57
Reggaetón	7.09	2.76	1.32	24.86	2.76	1.56
Religious music	4.60	2.38	1.22	9.19	2.14	1.41
Rock	0.57	4.02	1.00	3.78	1.99	1.36
Soul	5.17	3.62	1.04	24.32	1.99	1.24
Symphonic metal	21.07	2.57	1.47	21.08	1.46	0.98
Techno	1.15	2.72	1.35	10.81	2.42	1.44
Text songs	7.09	3.90	0.98	42.16	2.75	1.40
Trip hop	60.34	3.10	1.40	47.03	2.01	1.42
US folk	18.77	3.30	1.10	41.62	1.93	1.20
Zouk	36.59	2.53	1.14	17.84	2.77	1.54

Note: “DK” refers to the percentage of participants who reported not to know the corresponding (sub)genre.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Romain Brisson

References

Bonneville-Roussy

Rentfrow

P. J.

K. M.

Potter

(2013). Music through the ages: Trends in musical attitudes and preferences from adolescence through middle adulthood. Journal of Personality and Social Psychology, 105(4), 703–717.

Bourdieu

(1984). Distinction: A social critique of the judgement of taste. Cambridge, MA: Harvard University Press.

Brown

R. A.

(2012). Music preferences and personality among Japanese university students. International Journal of Psychology, 47(4), 259–268.

Cattell

R. B.

(1966). The scree test for the number of factors. Multivariate Behavioral Research, 1(2), 245–276.

Cattell

R. B.

Anderson

J. C.

(1953). The measurement of personality and behavior disorders by the I.P.A.T. music preference test. Journal of Applied Psychology, 37(6), 446–454.

Cattell

R. B.

Saunders

D. R.

(1954). Musical preferences and personality diagnosis: I. A factorization of one hundred and twenty themes. The Journal of Social Psychology, 39(1), 3–24.

Colley

(2008). Young people’s musical taste: Relationship with gender and gender-related traits. Journal of Applied Social Psychology, 38(8), 2039–2055.

Costello

A. B.

Osborne

J. W.

(2005). Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Practical Assessment, Research, & Evaluation, 10(7), 1–9.

Delsing

M. J. M. H.

ter Bogt

T. F. M.

Engels

R. C. M. E.

Meeus

W. H. J.

(2008). Adolescents’ music preferences and personality characteristics. European Journal of Personality, 22(2), 109–130.

10.

Dunn

P. G.

de Ruyter

Bouwhuis

D. G.

(2012). Toward a better understanding of the relation between music preference, listening behavior, and personality. Psychology of Music, 40(4), 411–428.

11.

Ferrer

Eerola

Vuoskoski

J. K.

(2013). Enhancing genre-based measures of music preference by user-defined liking and social tags, Psychology of Music, 41(4), 499–518.

12.

Finnas

(1987). Do young people misjudge each other’s musical taste? Psychology of Music, 15(2), 152–166.

13.

Franken

Keijsers

Dijkstra

J. K.

ter Bogt

(2017). Music preferences, friendship, and externalizing behavior in early adolescence: A SIENA examination of the music marker theory using the SNARE study. Journal of Youth and Adolescence, 46(8), 1839–1850.

14.

Fricke

K. R.

Herzberg

P. Y.

(2017). Personality and self-reported preference for music genres and attributes in a German-speaking sample. Journal of Research in Personality, 68, 114–123.

15.

Gardikiotis

Alexandros Baltzis

(2012). “Rock music for myself and justice to the world!”: Musical identity, values, and music preferences. Psychology of Music, 40(2), 143–163.

16.

George

Stickle

Rachid

Wopnford

(2007). The association between types of music enjoyed and cognitive, behavioral, and personality factors of those who listen. Psychomusicology, 19(2), 32–56.

17.

Gouveia

V. V.

Pimentel

C. E.

Santana

N. L.

Chaves

W. A.

Rodrigues

C. A.

(2008). Escala abreviada de preferência musical (STOMP): evidências de sua validade fatorial e consistência interna. Psico, 39(2), 201–210.

18.

Greenberg

D. M.

Baron-Cohen

Stillwell

D. J.

Kosinski

Rentfrow

P. J.

(2015). Musical preferences are linked to cognitive styles. PLoS ONE, 10(7). doi:10.1371/journal.pone.0131151

19.

Greenberg

D. M.

Kosinski

Stillwell

D. J.

Monteiro

B. L.

Levitin

D. J.

Rentfrow

P. J.

(2016). The song is you: Preferences for musical attribute dimensions reflect personality. Social Psychological and Personality Science, 7(6), 597–605.

20.

Horn

J. L.

(1965). A rationale and test for the number of factors in factor analysis. Psychometrika, 30(2), 179–185.

21.

Hutcheson

Sofroniou

(1999). The multivariate social scientist: Introductory statistics using generalized linear models. Thousand Oaks, CA: SAGE.

22.

Kaiser

H. F.

(1960). The application of electronic computers to factor analysis. Educational and Psychological Measurement, 20(1), 141–151.

23.

Kaiser

H. F.

(1974). An index of factorial simplicity. Psychometrika, 39(1), 31–36.

24.

Langmeyer

Guglhör-Rudan

Tarnai

(2012). What do music preferences reveal about personality? A cross-cultural replication using self-ratings and ratings of music samples. Journal of Individual Differences, 33(2), 119–130.

25.

Litle

Zuckerman

(1986). Sensation seeking and music preferences. Personality and Individual Differences, 7(4), 575–578.

26.

Mulder

ter Bogt

T. F. M.

Raaijmakers

Q. A. W.

Gabhainn

S. N.

Sikkema

(2010). From death metal to R&B? Consistency of music preferences among Dutch adolescents and young adults. Psychology of Music, 38(1), 67–83.

27.

O’Connor

B. P.

(2000). SPSS and SAS programs for determining the number of components using parallel analysis and Velicer’s MAP test. Behavior Research Methods, Instruments, & Computers, 32(3), 396–402.

28.

Peterson

R. A.

(1992). Understanding audience segmentation: From elite and mass to omnivore and univore. Poetics, 21(4), 243–258.

29.

Poropat

A. E.

(2009). A meta-analysis of the five-factor model of personality and academic performance. Psychological Bulletin, 135(2), 322–338.

30.

Poropat

A. E.

(2014). A meta-analysis of adult-rated child personality and academic performance in primary education. British Journal of Educational Psychology, 84, 239–252.

31.

Rentfrow

P. J.

Gosling

S. D.

(2003). The do re mi’s of everyday life: The structure and personality correlates of music preferences. Journal of Personality and Social Psychology, 84(6), 1236–1256.

32.

Rentfrow

P. J.

Goldberg

L. R.

Levitin

D. J.

(2011). The structure of musical preferences: A five-factor model. Journal of Personality and Social Psychology, 100(6), 1139–1157.

33.

Rentfrow

P. J.

Goldberg

L. R.

Stillwell

D. J.

Kosinski

Gosling

S. D.

Levitin

D. J.

(2012). The song remains the same: A replication and extension of the MUSIC model. Music Perception, 30(2), 161–185.

34.

Robinson

J. P.

Keegan

C. A.

Hanford

Triplett

T. A.

(1985). Public participation in the arts: Final report on the 1982 survey. College Park, MD: University of Maryland.

35.

Schäfer

Mehlhorn

(2017). Can personality traits predict musical style preferences? A meta-analysis. Personality and Individual Differences, 116, 265–273.

36.

Schäfer

Sedlmeier

(2009). From the functions of music to music preference. Psychology of Music, 37(3), 279–300.

37.

Schwartz

K. D.

Fouts

G. T.

(2003). Music preferences, personality style, and developmental issues of adolescents. Journal of Youth and Adolescence, 32(3), 205–213.

38.

ter Bogt

T. F. M.

Raaijmakers

Q. A. W.

Vollebergh

W. A. M.

van Wel

F. W.

Sikkema

(2003). Youngsters and their musical taste: Musical styles and taste groups. The Netherlands’ Journal of Social Sciences, 39(1), 35–52.

39.

Thompson

Daniel

L. G.

(1996). Factor analytic evidence for the construct validity of scores: A historical overview and some guidelines. Educational and Psychological Measurement, 56(2), 197–208.

40.

Van Ginkel

J. R.

Kroonenberg

P. M.

Kiers

H. A. L.

(2014). Missing data in principal component analysis of questionnaire data: A comparison of methods. Journal of Statistical Computation and Simulation, 84(11), 2298–2315.

41.

Vella

E. J.

Mills

(2017). Personality, uses of music, and music preference: The influence of openness to experience and extraversion. Psychology of Music, 45(3), 338–354.

42.

Zweigenhaft

R. L.

(2008). A do re mi encore: A closer look at the personality correlates of music preferences. Journal of Individual Differences, 29(1), 45–55.

43.

Zwick

W. R.

Velicer

W. F.

(1986). Comparison of five rules for determining the number of components to retain. Psychological Bulletin, 99(3), 432–442.