ArticlePDF Available

Musical Preferences Predict Personality: Evidence from Active Listening and Facebook Likes

Authors:

Abstract and Figures

Research over the past decade has shown that various personality traits are communicated through musical preferences. One limitation of that research is external validity, as most studies have assessed individual differences in musical preferences using self-reports of music-genre preferences. Are personality traits communicated through behavioral manifestations of musical preferences? We address this question in two large-scale online studies with demographically diverse populations. Study 1 (N=22,252) shows that reactions to unfamiliar musical excerpts predicted individual differences in personality-most notably openness and extraversion-above and beyond demographic characteristics. Moreover, these personality traits were differentially associated with particular music-preference dimensions. The results from Study 2 (N=21,929) replicated and extended these findings by showing that an active measure of naturally-occurring behavior, Facebook Likes for musical artists, also predicted individual differences in personality. In general, our findings establish the robustness and external validity of the links between musical preferences and personality.
Content may be subject to copyright.
Musical Preferences Predict Personality: Evidence from Active Listening and Facebook
Likes
Gideon Nave, Department of Marketing, The Wharton School of the University of Pennsylvania
Juri Minxha, Computation & Neural Systems, Caltech
David M Greenberg, Department of Psychology, University of Cambridge
Michal Kosinski, Graduate School of Business, Stanford University
David Stillwell, Judge Business School, University of Cambridge
Jason Rentfrow,* Department of Psychology, University of Cambridge
* Corresponding author, pjr39@cam.ac.uk
Abstract
Research over the past decade has shown that various personality traits are communicated
through musical preferences. One limitation of that research is external validity, as most studies
have assessed individual differences in musical preferences using self-reports of music-genre
preferences. Are personality traits communicated through behavioral manifestations of musical
preferences? We address this question in two large-scale online studies with demographically
diverse populations. Study 1 (N=22,252) shows that reactions to unfamiliar musical excerpts
predicted individual differences in personality - most notably openness and extraversion - above
and beyond demographic characteristics. Moreover, these personality traits were differentially
associated with particular music-preference dimensions. The results from Study 2 (N=21,929)
replicated and extended these findings by showing that an active measure of naturally-occurring
behavior, Facebook Likes for musical artists, also predicted individual differences in personality.
In general, our findings establish the robustness and external validity of the links between
musical preferences and personality.
Introduction
With the proliferation of Internet-based services for sharing and streaming music on demand,
personalized music is becoming a more central and prominent fixture in many people’s lives.
This increase coincides with a growing interest in understanding the psychological basis of
musical preferences. Over the past decade, several studies have investigated individual
differences in musical preferences with the aim of identifying its structure and psychological
correlates. In general, these investigations offer promising evidence that musical preferences can
be reduced to and conceptualized by a few broad dimensions, and that various aspects of musical
preferences are associated with individual differences in a range of psychological variables.
Informed by theory and research on person-environment interactions, a number of studies have
examined associations between musical preferences and personality (e.g., Miranda and Claes
2008; Greenberg et al. 2015, 2016; Langmeyer, Guglhör-Rudan, and Tarnai 2012; Schäfer and
Mehlhorn 2017; Rentfrow and Gosling 2003). The motivation for these studies has been to
develop and test the hypothesis that individuals are drawn to musical styles that satisfy and
reinforce their psychological needs. The results suggest, for example, that people who have a
need for creative and intellectual stimulation prefer unconventional and complex musical styles,
and that people who are sociable and enthusiastic prefer musical styles that are energetic and
lively (Rentfrow and Gosling 2003).
Although the results from studies on the links between musical preferences and personality
generally converge, past research suffers important limitations. One limitation concerns the way
in which musical preferences are measured. The most common method for assessing musical
preferences relies on self-reported preferences for musical genres (e.g., classical, rock, rap, etc.),
treated as proxies for musical preferences. This is problematic for three reasons. First, there is no
consensus about which genres to measure, with studies employing from a few to over 30 genres
and subgenres (e.g., George et al. 2007; Yeoh and North 2010). Second, participants may differ
in their definitions and interpretations of what type of music different genres represent, which in
turn might add undesirable noise to the measurement of their musical tastes. Third, it is not clear
to what extent findings from survey studies represent the actual preferences and behaviors of the
participants in the real world.
Another limitation of past studies is their reliance on samples of college students (e.g., Brown
2012; Langmeyer, Guglhör-Rudan, and Tarnai 2012; George et al. 2007; Palmer and Griscom
2013; Vuoskoski and Eerola 2011; Rentfrow and Gosling 2003). As music is particularly
important to young people (Bonneville-Roussy et al. 2013) and their peer-group relations
(Delsing, et al., 2008), college students may report stronger preferences for musical genres that
are popular among their peers, due to social desirability.
To overcome these limitations, we conducted two studies investigating whether the links
between musical preferences and personality generalize across different assessment methods and
across age-diversified samples. Our primary objective was to determine if individual differences
in the Big Five personality domains can be predicted from musical preferences. In Study 1 we
used a machine-learning “predictive” approach (Yarkoni and Westfall 2017) to examine whether
participants’ preference ratings following active listening to novel musical stimuli can be used
for out-of-sample predictions of their personalities. Study 2 replicates and extends Study 1 using
an ecologically valid behavioral measure of musical preferences: Facebook Likes of musical
artists.
Study 1: Preferences for novel music following active listening predict personality
Methods
Participants. We used data from a sample of 22,252 MyPersonality users from 153 different
countries.1 Majority of the participants self-reported gender (N = 20,770; 62% female); about
half reported their age (N = 10,414, median = 22, interquartile range= 7), 45% of which reported
being over 22 years of age, the typical age of a college graduate in the US. Among 17,988 users
who reported their geographical location, 25% (N = 4,517) lived in countries other than the US,
UK, or Canada. All respondents received feedback about their musical preferences (according to
1 The myPersonality Facebook app ran from 2007 to 2012. It presented the opportunity for Facebook users to take real scientific
research questionnaires and get feedback on their results. Overall, more than 6 million users took at least one questionnaire. The
raw data used in the current study is available for researchers on the project’s website, http://mypersonality.org.
the MUSIC model, further details below) and their Big Five personality following the
questionnaire. The study’s sample included all MyPersonality users who (1) completed a Big
Five personality questionnaire, and (2) completed at least one music-preference survey (further
details below).
Personality. Respondents’ personality profiles were estimated using the International Personality
Item Pool (IPIP) questionnaire measuring Five Factor Model (FFM) of personality (20 to 100
item long; Goldberg et al. 2006/2).
Musical preferences. Preferences for Western music can be reduced to a few dimensions
(Colley, 2008; George et al., 2007; Rentfrow, Goldberg, & Levitin, 2011; Rentfrow, Goldberg,
Stillwell, Kosinski, Gosling, & Levitin, 2012; Rentfrow & Gosling, 2003; Rentfrow &
McDonald, 2009; Schäfer & Sedlmeier, 2009). Analyses of the psychological, social, and
auditory characteristics of the dimensions suggests they can be defined as: Mellow,
Unpretentious, Sophisticated, Intense, and Contemporary (MUSIC). The Mellow dimension
represents music that is romantic, relaxing, and slow, and comprises soft rock, R&B, and adult
contemporary musical pieces. The Unpretentious dimension represents music that is
uncomplicated, relaxing, and acoustic, and comprises country, folk, and singer/songwriter pieces.
The Sophisticated dimension represents music that is inspiring, complex, and dynamic, and
comprises classical, operatic, world, and jazz pieces. The Intense dimension represents music
that is distorted, loud, and aggressive, and comprises classic rock, alternative rock, punk, and
heavy metal pieces. The Contemporary dimension represents music that is percussive, electric,
and not sad, and comprises rap, electronic dance music, Latin, and Euro pop pieces. Recent work
indicated that the MUSIC model accounts for 55% to 59% of the variance in preferences for
Western music (Bonneville-Roussy et al. 2013; Rentfrow et al. 2012; Rentfrow et al., 2011).
We estimated musical preferences using surveys designed according to the five-factor MUSIC
model (Rentfrow et al., 2011; Rentfrow et al. 2012).2 Each survey comprised 25 different 15-
second musical excerpts, with five excerpts representing each factor. Overall, there were 6
different musical surveys (Rentfrow et al., 2011; Rentfrow et al. 2012). Two surveys (Mix_A, N
= 17,904; Mix_B, N = 10,840), consisted of excerpts from a multitude of genres and subgenres,
the copyrights of which were purchased from Getty Images; thus, it was unlikely that
participants had previous exposure to them. Four other surveys included commercially released
music by known artists, of which two surveys consisted of only rock excerpts (Rock_A, N =
2,758; Rock_B, N = 1,748), and two surveys included only jazz excerpts (Jazz_A, N = 1,590;
Jazz_B, N = 8,887). All of the excerpts were used as stimuli that represent the 5-factor MUSIC
model in previous work.3 Each participant was assigned to one of three conditions [Mix, Jazz,
Rock] and took its corresponding survey “A”. Then, participants were given the opportunity to
take the second survey (“B”), always in the same condition as survey A. Surveys with missing
responses were excluded from further analysis.
2 The excerpts are available for download on the project’s page in the open science framework (OSF):
https://osf.io/nfqb9/?view_only=dff0271c8e0049bc88738e5a7b51ec2f
3 The mixed genre excerpts were used in (Rentfrow, Goldberg, and Levitin 2011; Rentfrow et al. 2012; Greenberg et al. 2015,
2016). A sub-sample (about 5%) of the “mix” respondents in the current study were also included in (Rentfrow et al. 2012). The
rock and jazz excerpts were used in (Rentfrow et al. 2012; Greenberg et al. 2015) and (Greenberg et al. 2016).
Prediction algorithm. For each of the Big Five personality traits, we conducted out-of-sample
predictions based on 1) preference ratings for the 25 musical excerpts; 2) survey responses +
gender and age; 3) gender and age alone. Predictions were carried out using the following nested
cross-validation procedure4:
1. We randomly split the entire data set into ten groups of participants.5
2. For each of the 10-holdout groups, we trained a linear model to predict each of the Big
Five personality traits by fitting a linear regression with a Least Absolute Shrinkage and
Selection Operator (LASSO) penalty to the remaining 90% of the data (Tibshirani 1996;
Camerer, Nave, and Smith 2017). The tuning parameter, l was optimized via 10-fold
cross validation (Stone 1974), performed within each training set.6
3. Using that trained model, we conducted out-of-sample predictions for the remaining 10%
of the data (i.e., the holdout group). We estimated the predictive accuracy by calculating
the Pearson’s correlation between the actual and predicted personality-trait scores.7
Results
Preferences for novel music predict personality traits
Here, we report personality predictions based on the responses to the survey with the largest
number of responses, Mix_A (N = 17,904), and discuss further replications in the section that
follows. The results are summarized in Fig. 1A and Table 1. For all the personality traits, we
found reliable correlations between the music-based personality predictions and the actual traits
(all p-values <0.001). The highest correlation was observed for openness (r(17,904)=0.25, 95%
confidence interval (CI)=[0.23 0.26]), followed by extraversion (r(17,902)=0.16, 95% CI=[0.14
0.17]), agreeableness (r(17,903)=0.15, 95% CI=[0.14 0.17]), neuroticism (r(17,905)=0.12, 95%
CI=[0.10 0.13]), and conscientiousness (r(17,174)=0.11, 95% CI=[0.10 0.13]). The music-based
predictors of openness, extraversion and agreeableness were significantly better than a baseline
model that predicted personality solely using gender and age (we rejected the null hypothesis of
equality in out-of-sample predictive accuracies at the p<0.01 level, Steiger’s z-test (Steiger
1980)).8 Adding musical preferences to the baseline model (gender and age) significantly
increased the predictive accuracy for all of the Big Five traits (all p’s<0.012, Steiger’s z-test). To
put these results in perspective, knowledge of one’s musical preferences reveals nearly as much
about their personality trait of openness as their behavior at work reveals to a work colleague; for
the remaining traits, predictive accuracy ranged between 41% (conscientiousness) and 66%
(neuroticism) of a colleague’s accuracy (Youyou, Kosinski, and Stillwell 2015).
4 Analyses scripts and pre-processed data are available on the project’s OSF page:
https://osf.io/nfqb9/?view_only=dff0271c8e0049bc88738e5a7b51ec2f
5 The random partition of the data into ten hold-out groups was conducted independently for each of the models (i.e., there was a
different partition for every combination of personality trait and dependent variables).
6 Optimization was performed using the ‘lassoglm’ function implemented in ‘Matlab’, under its default setting. Thus, we first
estimate l_MAX - the largest value of the penalty parameter l that gives a non-null model, and perform optimization by
exploring a geometric sequence of 100 values between 0.0001l_MAX and l_MAX. The values of the tuning parameter l for the
main models (averaged across the 10 folds estimating personalities from the survey responses) are available on Table S3. As the
LASSO procedure is sensitive to the scale of the inputs, all independent variables were standardized (z-scored) prior to model
training.
7 We estimated the correlations and confidence intervals on all of the data collapsed.
8 The results hold when limiting the training and testing samples to participants who self-reported age and gender, ruling out the
possibility that sample size differences underlie the improved predictive accuracy of the music-preferences based models.
These results indicate that preferences for short musical excerpts contain predictive information
about personality traits. However, they do not allow us to tease apart whether this information
arises from our participants’ unique musical tastes (represented by the liking of individual
excerpts) or from their tendencies to like music in general. To further investigate this issue, we
constructed, for each of the Big Five traits, an additional “general baseline model”, that included
(i) participants’ general evaluative tendencies (i.e., mean preference rating from all the musical
pieces), (ii) a general music liking factor, calculated by fitting a bi-factor model (Reise, Moore,
and Haviland 2010) to the liking ratings9, and (iii) gender and age.
Contrasting the predictive accuracies of the model that includes responses to individual survey
items, gender and age (Table 1, row 2) with the general baseline model (Table 1, row 4), allows
us to disentangle the predictive accuracies arising from specific versus general musical
preferences. We find that the additional predictive accuracies obtained by including the
individual survey responses (above the general baseline model) was highest for openness (Dr =
0.09, 55% increase) and extraversion (Dr = 0.08, 79%). However, they were less pronounced for
the three other traits (Dr = 0.03, 17% for neuroticism, Dr = 0.02, 15% for agreeableness, and only
Dr = 0.01, 7% for conscientiousness).10 Thus, both specific and general musical preferences
underlie the capacity to predict personality from musical preferences, where the former play a
substantial role for the cases of openness and extraversion, and the latter underlie the capacity to
predict the other three traits.
Finally, we explored the generalizability of our findings to two sub-populations that are typically
under-represented in laboratory studies conducted in college students. First, we found that all of
the results hold when limiting the estimates of predictive accuracy to participants who self-
reported residing outside the US, UK, or Canada (N=1,596, see Table S1): adding musical
preferences significantly increased the predictive accuracy of the baseline model that included
only age and gender in this sub-population (for neuroticism p=0.039, for all other traits p’s <
0.01, Steiger’s z-test). The results hold when restricting the analyses to participants that self-
reported being over 30 years of age (N=1,528, see Table S1): adding the musical preferences
survey increased the predictive accuracy of the baseline demographic model for all traits
(openness, extraversion and agreeableness: p’s < 0.01; for conscientiousness p=0.047, for
neuroticism p=0.060, Steiger’s z-test).
9 The general music liking factor was estimated by fitting a bifactor model to Mix_A survey responses. The model included a
general factor and five additional orthogonal dimensions (representing the MUSIC factors), and was fitted using the ‘psych’
library of the statistical software ‘R’. The general factor significantly loaded on 18 of the 25 response items, but did not
substantially load on the “intense” excerpts (see Table 3). The empirical question of whether a valid general music liking factor
exists is beyond the scope of this paper.
10 The predictive accuracy of the general baseline model was significantly smaller that the model that also included the ratings of
individual items, for four of the traits (all p’s<0.028, Steiger’s z-test). For conscientiousness, predictive accuracy was greater in
the model that included the ratings of individual items, though the difference in predictive accuracy did not reach statistical
significance (p=0.29).
Table 1: Predictive accuracy of musical preference based personality predictors (out of
sample), for all Big Five traits, test Mix_A (N=17,904, N=8,100 with age and gender
information).
Figure 1A: Correlations between music preference-based big five personality predictors
(out of sample) and actual personalities, test A (N=17,904). Error bars denote 95% CIs.
Figure 1B: Correlations between music-based big five personality predictors (out of
sample) and actual personalities across tests and genres. Error bars denote 95% CIs.
Replication across tests and genres
To evaluate the robustness of the predictive accuracy results, we carried out the same analyses
again for the other five musical preferences surveys. It is important to bear in mind that the
sample sizes for these surveys were significantly smaller (between 5% and 45% of Mix_A’s
sample size), and therefore (1) predictive accuracy was expected to decline, as the models’
parameter estimates were less stable, and (2) the capacity to detect effects decreased due to
reduced statistical power, especially for the traits for which the associations between preferences
and personality were expected to be smaller (conscientiousness and neuroticism).
The results are summarized in Fig. 1B and SOM Table S2. The most similar replication used
survey Mix_B, that was taken by a subpopulation (about 45%) of Mix_A respondents, and like
Mix_A, consisted of excerpts from multiple genres. The predictive accuracies of the models
trained using Mix_B were significantly greater than zero (all p’s < 0.001), and their point
estimates were greater than the lower bounds of the 95% CIs of the predictive accuracies
obtained from Mix_A responses, for all of the Big Five traits. Furthermore, adding Mix_B
survey responses to the baseline demographic model (constructed from age and gender)
significantly improved the predictive accuracies for openness, extraversion and agreeableness
(p’s < 0.001, Steiger’s z-test), providing a successful replication of survey Mix_A for these traits.
For neuroticism the improvement was marginally significant (p=0.11), and for conscientiousness
we did not detect a reliable improvement (p=0.47).
Next, we repeated the analyses for the rock and jazz surveys. These surveys were designed to
capture the dimensions of the MUSIC model, while containing excerpts from exclusively one
genre. For the two rock surveys (Rock_A, N = 2,758; Rock_B, N = 1,748) the predictive
accuracies of all ten models (5 personality traits x 2 surveys) were reliably greater than zero (all
p’s<0.01), and adding the responses for these musical surveys to the baseline model (gender and
age) increased the predictive accuracy of the models for all traits except neuroticism (openness,
extraversion and agreeableness: p’s<0.01; conscientiousness p<0.10).
The models using responses to Jazz_A (N=1,590) had statistically significant predictive
accuracies (p’s<0.01) for all traits except extraversion. Adding the responses for these musical
surveys to the baseline model (gender and age) increased the predictive accuracy of all traits,
though the improvement was not statistically significant, perhaps due to the small sample. For
Jazz_B (the smallest survey, N=887) we detected a reliable predictive accuracy only when
predicting openness (p<0.001), and marginally significant (p<0.10) predictive accuracies for
agreeableness and neuroticism, likely because the small sample (about 20 times smaller than
Mix_A) might have been insufficient for model training.
Robustness of the 5-factor MUSIC model in a large diverse sample
Apart from examining the capacity to predict personality from liking of music, our data provide a
unique opportunity to estimate the robustness of the 5-factor MUSIC model (Rentfrow et
al.,2011; Rentfrow et al. 2012), and the capacity of our musical surveys to capture it. To do this,
we subjected the survey responses of the participants who answered both surveys A and B of the
mix genre excerpts (i.e., Mix_A and Mix_B, N = 10,840) to Principal Component Analysis
(PCA).11 Investigating the projections of the different musical excerpts onto each of the principal
components, revealed that each group of excerpts, that was a-priori selected to represent a
MUSIC dimension, mapped into a unique principle component, for which the average projection
was an order of magnitude greater than the projection onto the four other components (Table 2,
see supplementary Fig. S1, A-E for projections of individual excerpt). Further, the first five
principal components explained 59% of the variance in the data (Fig. S1 F). Similar results were
obtained for the responses to the Jazz and Rock surveys, and are published elsewhere (Rentfrow
et al. 2012).
As the musical surveys used by the current investigation were specifically designed to capture
the MUSIC model, examining these results in isolation would not allow concluding that all types
of Western music are captured by the five-factor framework. However, it is important to keep in
mind that the MUSIC model was originally constructed based on exploratory research that used a
wide variety of musical pieces that are different from the ones used in the present research
(Rentfrow, Goldberg, and Levitin 2011). The current results corroborate that the MUSIC model
is a robust framework for organizing individual differences in preferences for music, and
demonstrate its generalizability to a large and diverse population.
11 A similar analysis using the same participants who took the jazz and rock excerpts, and a small sub-sample (about 5%) of the
mixed survey respondents was previously published in (Rentfrow et al. 2012).
Table 2: Average loadings of the excerpts’ liking ratings on the first five principal
components of the data. Each row represents the 10 excerpts from surveys Mix_A and
Mix_B that a-priori represented each of the five MUSIC dimensions.
Table 3: Average loadings of the excerpts’ liking ratings on the general factor and the first
five principal components of the data, extracted using a bi-factor model. Each row
represents the 5 excerpts from surveys Mix_A that a-priori represented each of the five
MUSIC dimensions.
Links between the Big Five and MUSIC dimensions
The results indicate that the MUSIC model can be recovered from preference ratings for novel
musical stimuli, and that personality traits can be predicted from these ratings. We now turn to
investigate whether, and to what extent, systematic associations between the Big Five and
preferences for specific MUSIC dimensions exist.
In order to tease apart the different MUSIC components from general liking tendencies, we
performed a bi-factor analysis on the individual responses to survey Mix_A. The analysis
resulted in five factors that captured the (a-priori defined) MUSIC dimensions, as well as a
general liking factor (see Table 3). We then calculated the partial-correlation between the Big
Five traits and the projections of participants’ preference on (i) the general liking factor, and (ii)
the lower dimensions capturing the MUSIC dimensions. These partial correlations controlled for
gender and age, and for the low-order MUSIC dimensions they also controlled for the general
liking factor.
The results are summarized in Table 4, and show that two personality traits are associated with
preferences for specific MUSIC dimensions, above demographics and the general liking
tendency. In particular, openness is associated with greater liking of sophisticated music
(r(8,097) = 0.16, 95% CI = [0.14 0.18], p<0.001), and less liking of mellow (r(8,097) = -0.12,
95% CI = [-0.10 -0.14], p<0.001) and contemporary music (r(8,098) = -0.11, 95% CI = [-0.09 -
0.13], p<0.001), where extraversion is associated with preference for unpretentious music
(r(8,096) = 0.13, 95% CI = [0.11 0.15], p<0.001). Openness and extraversion are also associated
with general liking of music (openness: r(8,098) = 0.14, 95% CI = [0.12 0.16] p<0.001;
extraversion: r(8,097) = 0.10, 95% CI = [0.08 0.12], p<0.001). For the remaining three traits,
none of the specific correlations exceeded r = 0.06, and agreeableness was the only trait
associated with general liking of music (r(8,098) = 0.14 95% CI = [0.12 0.16] p<0.001).
We further explored the links between personality and preferences for the individual excerpts
representing the MUSIC dimensions in all of the six musical surveys, by estimating the
univariate correlations between responses to the different survey questions (i.e., specific
excerpts) and personality traits. In Fig. 2, each 6 x 5 framed square represents a different
combination of a Big Five trait (row) and a MUSIC factor (column). For example, the top left
square represents the correlations between openness and the different excerpts capturing the
Mellow dimension. Each row within this square represents one of the six different surveys, and
contains the five different excerpts that correspond to the mellow factor in the survey.
Several patterns emerge in the correlation map. Most notably, the correlations are typically small
in size (none was greater than r = .21), and are positive for all of the traits except neuroticism. In
line with the partial-correlations reported above (for survey Mix_A), openness most strongly
correlated with liking the sophisticated excerpts, and extraversion was most strongly correlated
with evaluating the unpretentious excerpts more positively.
Table 4: Partial correlations between the Big Five traits and the general music liking factor
as well as the lower-order MUSIC dimensions, extracted by performing a bi-factor model
on the responses to survey Mix_A. The correlations control for gender and age, and for the
lower dimension they also control for the general factor.
Figure 2: correlations between big five personality traits and liking of different musical
excerpts.
Study 2: Musical Facebook Likes predict personality traits
The results from Study 1 indicated that preferences for unfamiliar musical stimuli contain some
valid information about personality. The aim of Study 2 was to replicate and extend these results
to real-world behavior by investigating whether naturally-occurring statements of musical
preferences, as represented by Facebook Likes of musical artists, also predict personality traits.
The Like feature is a mechanism used by Facebook users to publicly express their positive
association with online content, by generating a digital record that is accessible to their friends,
Facebook, software developers who provide services to users, as well as outside parties,
including governments and industries. Facebook Likes represent a very generic class of digital
records, similar to Web search queries or credit card purchases, and are used to signal positive
associations with many different types of content, including photos, friends’ status updates, and
Facebook pages of products, sports, books, restaurants, popular websites, and musicians. There is
evidence that Facebook Likes, in general, contain information about many personal attributes,
from religiosity and political views to sexual orientation and personality (Kosinski, Stillwell, and
Graepel 2013). However, that work examined Likes in general, irrespective of content, so it is
not clear whether Likes for specific types of content are reliably associated with personality.12
Thus, Study 2 not only evaluates the generalizability of Study 1 to behavioral indicators of
musical preferences, it also examines whether, and to what extent, Likes of musical artists alone
betray information about the personalities of Facebook users.
12 Various Facebook likes were shown to contain personality-related information, but many of the findings are difficult to
interpret. For a general list of facebook likes that are most predictive of personality traits, see
http://www.pnas.org/content/suppl/2013/03/07/1218772110.DCSupplemental/st01.pdf
Methods
Participants. We used data from a sample of 21,929 MyPersonality users (65% females), with a
median age of 21 (interquartile distance = 5).13 The study included all of the participants in the
MyPersonality database who (1) completed a Big Five personality questionnaire (2) had at least
20 “Likes” of musical artists that were used for personality prediction (further details below) and
(3) shared information about their age and gender.
Personality. Study2 used the same IPIP measure described in Study 1 (Goldberg et al., 2006).
Musical Facebook Likes. In order to focus our analysis on the predictive power of musical
artists Likes, we first filtered out all the Likes that were not categorized by Facebook as music-
related. We then searched all of the remaining Likes in EchoNest (http://the.echonest.com/) - a
major online musical database containing over 3 million artists, and excluded all Likes that did
not appear in the database. Next, we excluded all users that had less than 20 Likes, and included
only artists that had at least 20 Likes.14 This resulted a large, sparse logical matrix L, in which
each row r represented a participant and each column c represented an artist, such that L(r,c)
equals one if participant r Likes artist c and zero otherwise. The matrix L has dimensions 21,929
(users) x 62,036 (artists).
Prediction algorithm. For each of the Big Five personality traits, we conducted three out-of-
sample predictions based on 1) the musical Likes matrix; 2) the Like matrix, gender and age; and
3) gender and age alone.
Predictions were carried out using the following procedure:
1. We randomly split the participants into ten groups, in a similar fashion to Study 1.
2. For each of the 10-holdout groups, we reduced the dimensionality of the liking matrix L
to N users x 500, by performing sparse singular value decomposition (SVD) on the
remaining 90% of the data.
3. For each of the 10-holdout groups, we trained a linear model to predict each of the Big
Five personality traits, by fitting a linear regression with a LASSO penalty to the
remaining 90% of the data. The tuning parameter, l was optimized via 10-fold cross
validation, performed within each training set, in a similar fashion to Study 1. All of the
independent variables were standardized priori model training, as penalized regression
models (such as the LASSO) are sensitive to the scale of the inputs.
4. Using the trained model, we conducted out-of-sample prediction on the 10% of the data
that comprised the holdout group.15 We estimated the goodness of fit by calculating the
Pearson’s correlation between the actual and predicted personalities.
13 A small proportion of the participants of study 2 (roughly 3.5%) were also in the subject pool of study 1.
14 The study’s sample was constructed in the following manner: (1) users with less than 20 musical artist likes were excluded (2)
artists with less than 20 likes among the remaining users were excluded (3) the first two steps were repeated iteratively until
convergence, to ensure that each user had at least 20 likes and that each artist was associated with at least 20 users.
15 In order to generate predictions, the liking matrix of the hold-out group was first projected onto the first 500 dimensions of the
training data, calculated in step 2.
Results
Musical Facebook Likes predict personality traits
The results are summarized in Fig. 3. and Table 5. We found reliable correlations between the
music Likes-based personality predictors and all of the Big Five personality traits (all p-values
<0.001). Like in Study 1, the highest predictive accuracy was for openness (r(21,929)=0.30, 95%
CI=[0.29 0.31]), followed by extraversion (r(21,929)=0.21, 95% CI=[0.20 0.22]),
conscientiousness (r(21,929)=0.19, 95% CI=[0.17 0.20]), neuroticism (r(21,929)=0.18, 95%
CI=[0.17 0.20]), and agreeableness (r(21,929)=0.17, 95% CI=[0.15 0.18]). To put our results in
perspective, the predictive accuracy of the music Likes-based model for openness and
neuroticism was roughly the same as a personality prediction made by a co-worker. For the other
traits, accuracy ranged between 55% (agreeableness) and 77% (conscientiousness) of the
accuracy of a work colleague’s prediction (Youyou, Kosinski, and Stillwell 2015).
For all the traits, the musical Likes-based predictors were substantially more accurate than the
baseline demographic model (all p-values <0.001, Steiger’s z-test). Adding Likes to the baseline
model significantly improved the results for all traits but neuroticism (see Fig. 3, (all p’s<0.001,
Steiger’s z-test).16 For neuroticism the predictive accuracies of the Like-based model and the
demographic model were similar, suggesting that the former model’s predictive accuracy stems
from information that is also captured by age and gender.
Personality inferences based on Facebook Likes were more accurate, on average, compared to
inferences based on active listening. As Facebook Likes contain meta information about the
performing artist that goes beyond the pure auditory content, this finding is somewhat
unsurprising. However, this finding should be interpreted with caution, as we cannot rule out the
possibility that other factors, that are not directly related to metadata (e.g., differences in
measurement error between these two types of variables), underlie the differences in predictive
accuracy between the models.17
The results indicate that Facebook Likes of musical artists carry personality-relevant
information. However, they do not allow us to tease apart the different contributions of particular
musical tastes (i.e., liking of specific artists) from more general tendencies, such as a general
tendency to like musical Facebook pages (e.g., high openness individuals tend to like more
artists), or an inclination to like popular pages (e.g., agreeable individuals tend to like artists that
are liked by others). To investigate this issue, we constructed an additional “general baseline
model”, predicting each of the Big Five traits, using (i) # of Likes: a single scalar denoting the
total amount of musical artists that each participant liked, and (ii) popularity score: a scalar
16 Somewhat surprisingly, the predictive accuracy of the “likes only” model was slightly greater than the accuracy of the full
model containing likes, gender and age, for some of the traits (Fig. 3). A possible account (apart from sampling error), is that age
and gender are highly predictable from facebook likes (Kosinski, Stillwell, and Graepel 2013), which might have generated
multicollinearity in the full model (Chong and Jun 2005). Note that the “likes only” model contains orthogonal features
constructed using SVD.
17 When repeating the analysis of study 2 using a subsample of 17,904 participants (the sample size of study 1 for the Mix_A
survey), the predictive accuracy of the Likes-based models only slightly decreased, and was still greater than the excerpt-based
models (see Table 5, bottom row). This suggests that the larger training set was not the main cause for the superior performance
of the Like-based models.
denoting the average popularity across all artists liked by the user, such that popularity was
defined as the logged number of Likes that an artist had across the study’s participants,
normalized by the total number of users, and (iii) gender and age.
Contrasting the predictive accuracies of the model that include individual Likes, gender and age
(Table 5, row 2) with the general baseline model (Table 5, row 4), allows us to disentangle the
predictive accuracies arising from specific, versus general, musical liking tendencies. We find
that the additional predictive capacities obtained by including the individual Likes were
substantial for four of the traits. For openness, we found Dr = 0.20 (200% increase), followed by
extraversion (Dr = 0.15, 260%), agreeableness (Dr = 0.12, 240%), and conscientiousness (Dr =
0.07, 70%). The increase was not pronounced for neuroticism (Dr < 0.01, only 2%).18 In
conclusion, the majority of predictive personality information can be attributed to individual
Likes rather than general tendencies, with the exception of neuroticism.
Figure 3: Correlations between Facebook music Likes-based big five personality predictors
(out of sample) and actual personalities (N=21,929). Error bars denote 95% CIs.
18 The general baseline model was inferior to a model that also included the ratings of individual items, for all of the traits except
neuroticism (all p’s<0.001, Steiger’s z-test).
Table 5: Predictive accuracy of music artist Like based personality predictors (out of
sample), for all big 5 traits (N=21,929).
Discussion
Recent research has suggested that individual differences in musical preferences and personality
traits are linked. Using a diverse sample composing tens of thousands of participants, we
corroborated these findings and further extended them in four important ways.
First, our results show that affective reactions to 15-second excerpts of novel musical pieces,
which lacked metadata information (e.g., artist name), are sufficient for predicting individual
differences in personality. This finding replicated across- and within-genres, and demonstrates
that preferences for the musical content itself, rather than the name of the artist, or a genre,
contains sufficient information for personality inference.
Second, our study corroborates the MUSIC model’s capacity to capture individual differences in
preferences for Western music in a large and diverse population. Extensive research and
discussions devoted to estimating the replicability of laboratory experiments in psychology, and
social sciences in general, have highlighted the critical importance of replication efforts (Open
Science Collaboration 2015; Nosek et al. 2015; Camerer et al. 2016; Simons 2014; Nave,
Camerer, and McCullough 2015; Lane et al. 2016; Carter and McCullough 2013). Our results
show that the 5-factor MUSIC model is highly replicable, providing a solid foundation for the
future investigations of musical preferences and their links with other psychological constructs.
Third, we find that preferences for specific dimensions of the MUSIC model are associated with
two of the Big Five traits. Preferences for sophisticated musical excerpts were related to
openness to experience, whereas preferences for unpretentious excerpts were associated with
extraversion.
Fourth, the present research helps to establish the external validity of the link between musical
preferences and personality, by showing that personality traits can be reliably predicted both
from liking ratings that follow actual listening, and also from digital records of naturally-
occurring, real-world behaviors. Previous research has shown that personality can be inferred
from Facebook Likes in general (Kosinski, Stillwell, and Graepel 2013), yet the mechanisms at
work are poorly understood. Here, we have shown that focusing on musical preferences alone
reveals valid information about users’ personalities. With the growing presence of services for
streaming and sharing music online, this finding has direct implications for the music industry,
recommendation algorithms, and marketing practitioners (Ogden, Ogden, and Long 2011/3;
Bruner 1990; Matz et al. 2017; Matz, Gladstone, and Stillwell 2016).
Our study has several important limitations, leaving open questions for future research. First,
while our results demonstrated that musical preferences carry personality information that goes
beyond age, gender, and general liking tendencies, we recognize that there are likely other
unmeasured person-level variables (e.g., geo-location, socio-economic status, culture, and
preferences for different leisure activities) that would capture at least some of this incremental
variance. Moreover, our results are correlational, and therefore cannot address questions of
causality. For example, it is possible that common environmental factors (e.g., peer influence)
influence both personality and musical preferences.
Second, although Facebook Likes are active, naturally-occurring behaviors, they do not
automatically reflect what music people actually listen to. Moreover, liking of an artist might be
driven by factors other than musical taste, such as peer influence or self-image. The increasing
use of music streaming services (e.g., Last FM, Spotify) is expected to allow further
investigations of the links between personality and active ecologically-valid music listening
behavior.
Third, although our study generalizes previous findings to populations that are beyond college
students, our sample is composed of Facebook users. It is thus an open question whether our
findings generalize to populations that are not represented in the current work, such as non-
Western societies (Henrich, Heine, and Norenzayan 2010). Moreover, the musical pieces used in
Study 1 were entirely Western in origin, so for now, the conclusions we can draw from the
current findings are restricted to predominantly Western societies.
Finally, while our data-driven predictive analyses provide strong evidence supporting the link
between musical preferences and personality, the results call for further development of
theoretical models for identifying the mechanisms at work. One such candidate mechanism is
based on self-identity motives (Abrams 2009). That is, people are drawn to musical styles that
validate their self-perceptions and convey that information to others (e.g., listening to avant-
garde music can serve to simultaneously reinforce and communicate the belief that one is
creative and unconventional). A second mechanism is based on emotion regulation (Saarikallio
and Erkkilä 2007). That is, people prefer styles of music that reinforce their mood or emotional
state (e.g., listening to uplifting music may help to maintain a positive mood). A third possible
mechanism is based on activity congruence, or the idea that people prefer auditory content that
complements the activities they regularly pursue. For example, fast and upbeat music
complements various energetic activities, from dancing to socializing, that are likely to appeal to
extraverted people.
While some of our exploratory findings demonstrate associations between personality traits and
components of the five-factor MUSIC model that are consistent with the above mechanisms
(e.g., high openness is associated with liking of sophisticated music), and are also in accord with
previous research (Schäfer and Mehlhorn 2017), the magnitudes of these associations are
generally small in size. Thus, a considerable amount of variance in musical preferences remains
unexplained. Future investigations concerned with musical preferences should illuminate the
underlying mechanisms by investigating how preferences relate to identity motives, emotion
regulation processes, and activity preferences, and also by exploring how preferences for
particular auditory features (e.g, rhythm, time signature, frequency components) may correspond
to different personality traits (see Logan and Others 2000; Lindenbaum et al. 2010).
In summary, we have shown that preference ratings for unfamiliar musical stimuli and naturally
occurring statements of musical preferences in online social media, allow for making reliable
inference of personality traits. These results corroborate that music - a form of self-expression
that is ubiquitous across human cultures - communicates meaningful information about basic
psychological characteristics.
Author Contributions
All authors developed the study concept and contributed to the study design. Data collection
were performed by M.K and D.S, data analysis and interpretation was performed by G.N. and
J.M under the supervision of J.R.; G.N. drafted the manuscript.
References
Abrams, Dominic. 2009. “Social Identity on a National Scale: Optimal Distinctiveness and
Young People’s Self-Expression through Musical Preference.” Group Processes &
Intergroup Relations: GPIR 12 (3 Spec). Sage Publications: 303–17.
Bonneville-Roussy, Arielle, Peter J. Rentfrow, Man K. Xu, and Jeff Potter. 2013. “Music
through the Ages: Trends in Musical Engagement and Preferences from Adolescence
through Middle Adulthood.” Journal of Personality and Social Psychology 105 (4).
psycnet.apa.org: 703–17.
Brown, R. A. 2012. “Music Preferences and Personality among Japanese University Students.”
International Journal of Psychology: Journal International de Psychologie 47 (4). Taylor &
Francis: 259–68.
Bruner, Gordon C. 1990. “Music, Mood, and Marketing.” Journal of Marketing 54 (4).
American Marketing Association: 94–104.
Camerer, Colin F., Anna Dreber, Eskil Forsell, Teck-Hua Ho, Jürgen Huber, Magnus
Johannesson, Michael Kirchler, et al. 2016. “Evaluating Replicability of Laboratory
Experiments in Economics.” Science 351 (6280). science.sciencemag.org: 1433–36.
Camerer, Colin F., Gideon Nave, and Alec Smith. 2017. “Dynamic Unstructured Bargaining
with Private Information and Deadlines: Theory, Experiment, and Outcome Prediction via
Machine Learning.” Manuscript Submitted for Publication.
Carter, Evan C., and Michael E. McCullough. 2013. “Is Ego Depletion Too Incredible? Evidence
for the Overestimation of the Depletion Effect.” The Behavioral and Brain Sciences 36 (6):
683–84; discussion 707–26.
Chong, Il-Gyo, and Chi-Hyuck Jun. 2005. “Performance of Some Variable Selection Methods
When Multicollinearity Is Present.” Chemometrics and Intelligent Laboratory Systems 78
(1). Elsevier: 103–12.
Collaboration, Open Science, and Others. 2015. “Estimating the Reproducibility of
Psychological Science.” Science 349 (6251). American Association for the Advancement of
Science: aac4716.
Colley, Ann. 2008. “Young People’s Musical Taste: Relationship With Gender and Gender-
Related Traits.” Journal of Applied Social Psychology 38 (8). Wiley Online Library: 2039–
55.
Delsing, Marc J. M. H., Tom F. M. ter Bogt, Rutger C. M. E. Engels, and Wim H. J. Meeus. 2008.
“Adolescents’ Music Preferences and Personality Characteristics.” European Journal of Personality
22 (2). John Wiley & Sons, Ltd.: 10930.
George, Darren, Kelly Stickle, Faith Rachid, and Alayne Wopnford. 2007. “The Association
between Types of Music Enjoyed and Cognitive, Behavioral, and Personality Factors of
Those Who Listen.” Psychomusicology: A Journal of Research in Music Cognition 19 (2).
Music and Arts Publications: 32.
Goldberg, Lewis R., John A. Johnson, Herbert W. Eber, Robert Hogan, Michael C. Ashton, C.
Robert Cloninger, and Harrison G. Gough. 2006/2. “The International Personality Item Pool
and the Future of Public-Domain Personality Measures.” Journal of Research in Personality
40 (1). Elsevier: 84–96.
Greenberg, David M., Simon Baron-Cohen, David J. Stillwell, Michal Kosinski, and Peter J.
Rentfrow. 2015. “Musical Preferences Are Linked to Cognitive Styles.” PloS One 10 (7).
journals.plos.org: e0131151.
Greenberg, David M., Michal Kosinski, David J. Stillwell, Brian L. Monteiro, Daniel J. Levitin,
and Peter J. Rentfrow. 2016. “The Song Is You Preferences for Musical Attribute
Dimensions Reflect Personality.” Social Psychological and Personality Science. SAGE
Publications, 1948550616641473.
Henrich, Joseph, Steven J. Heine, and Ara Norenzayan. 2010. “The Weirdest People in the
World?” The Behavioral and Brain Sciences 33 (2-3). Cambridge Univ Press: 61–83.
Kosinski, Michal, David Stillwell, and Thore Graepel. 2013. “Private Traits and Attributes Are
Predictable from Digital Records of Human Behavior.” Proceedings of the National
Academy of Sciences of the United States of America 110 (15): 5802–5.
Lane, A., O. Luminet, G. Nave, and M. Mikolajczak. 2016. “Is There a Publication Bias in
Behavioural Intranasal Oxytocin Research on Humans? Opening the File Drawer of One
Laboratory.” Journal of Neuroendocrinology 28 (4). Wiley Online Library.
doi:10.1111/jne.12384.
Langmeyer, Alexandra, Angelika Guglhör-Rudan, and Christian Tarnai. 2012. “What Do Music
Preferences Reveal About Personality?” Journal of Individual Differences 33 (2).
econtent.hogrefe.com: 119–30.
Lindenbaum, Ofir, Shai Maskit, Ophir Kutiel, and Gideon Nave. 2010. “Musical Features
Extraction for Audio-Based Search.” In Electrical and Electronics Engineers in Israel
(IEEEI), 2010 IEEE 26th Convention of, 000087–000091. IEEE.
Logan, Beth, and Others. 2000. “Mel Frequency Cepstral Coefficients for Music Modeling.” In
ISMIR. cs.swarthmore.edu.
http://web.cs.swarthmore.edu/~turnbull/cs97/f09/paper/logan00.pdf.
Matz, Sandra C., Joe J. Gladstone, and David Stillwell. 2016. “Money Buys Happiness When
Spending Fits Our Personality.” Psychological Science 27 (5). pss.sagepub.com: 715–25.
Matz, S., M. Kosinski, G. Nave, and D. Stillwell. 2017. “Psychographic Targeting as an
Effective Approach to Digital Mass Persuasion.” Proceedings of the National Academy of
Science, in press.
Miranda, Dave, and Michel Claes. 2008. “Personality Traits, Music Preferences and Depression
in Adolescence.” International Journal of Adolescence and Youth 14 (3). Taylor & Francis:
277–98.
Nave, Gideon, Colin Camerer, and Michael McCullough. 2015. “Does Oxytocin Increase Trust
in Humans? A Critical Review of Research.” Perspectives on Psychological Science: A
Journal of the Association for Psychological Science 10 (6): 772–89.
Nosek, B. A., G. Alter, G. C. Banks, D. Borsboom, S. D. Bowman, S. J. Breckler, S. Buck, et al.
2015. “SCIENTIFIC STANDARDS. Promoting an Open Research Culture.” Science 348
(6242). ncbi.nlm.nih.gov: 1422–25.
Ogden, James R., Denise T. Ogden, and Karl Long. 2011/3. “Music Marketing: A History and
Landscape.” Journal of Retailing and Consumer Services 18 (2). Elsevier: 120–25.
Palmer, Stephen E., and William S. Griscom. 2013. “Accounting for Taste: Individual
Differences in Preference for Harmony.” Psychonomic Bulletin & Review 20 (3). Springer:
453–61.
Reise, Steven P., Tyler M. Moore, and Mark G. Haviland. 2010. “Bifactor Models and Rotations:
Exploring the Extent to Which Multidimensional Data Yield Univocal Scale Scores.”
Journal of Personality Assessment 92 (6). Taylor & Francis: 544–59.
Rentfrow, Peter J., Lewis R. Goldberg, and Daniel J. Levitin. 2011. “The Structure of Musical
Preferences: A Five-Factor Model.” Journal of Personality and Social Psychology 100 (6):
1139–57.
Rentfrow, Peter J., Lewis R. Goldberg, David J. Stillwell, Michal Kosinski, Samuel D. Gosling,
and Daniel J. Levitin. 2012. “The Song Remains the Same: A Replication and Extension of
the MUSIC Model.” Music Perception 30 (2). mp.ucpress.edu: 161–85.
Rentfrow, Peter J., and Samuel D. Gosling. 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). psycnet.apa.org: 1236–56.
Saarikallio, Suvi, and Jaakko Erkkilä. 2007. “The Role of Music in Adolescents’ Mood
Regulation.” Psychology of Music 35 (1). SAGE Publications: 88–109.
Schäfer, Thomas, and Claudia Mehlhorn. 2017. “Can Personality Traits Predict Musical Style
Preferences? A Meta-Analysis.” Personality and Individual Differences 116 (October).
Elsevier: 265–73.
Simons, Daniel J. 2014. “The Value of Direct Replication.” Perspectives on Psychological
Science: A Journal of the Association for Psychological Science 9 (1): 76–80.
Steiger, James H. 1980. “Tests for Comparing Elements of a Correlation Matrix.” Psychological
Bulletin 87 (2). American Psychological Association: 245.
Stone, Mervyn. 1974. “Cross-Validatory Choice and Assessment of Statistical Predictions.”
Journal of the Royal Statistical Society. Series B, Statistical Methodology. JSTOR, 111–47.
Tibshirani, Robert. 1996. “Regression Shrinkage and Selection via the Lasso.” Journal of the
Royal Statistical Society. Series B, Statistical Methodology. JSTOR, 267–88.
Vukasović, Tena, and Denis Bratko. 2015. “Heritability of Personality: A Meta-Analysis of
Behavior Genetic Studies.” Psychological Bulletin 141 (4). psycnet.apa.org: 769–85.
Yarkoni, T., and J. Westfall. 2017. “Choosing Prediction over Explanation in Psychology:
Lessons from Machine Learning.” Perspectives on Psychological Science, in press
Yeoh, Joanne P. S., and Adrian C. North. 2010. “The Effect of Musical Fit on Consumers’
Memory.” Psychology of Music 38 (3). SAGE Publications: 368–78.
Youyou, Wu, Michal Kosinski, and David Stillwell. 2015. “Computer-Based Personality
Judgments Are More Accurate than Those Made by Humans.” Proceedings of the National
Academy of Sciences of the United States of America 112 (4): 1036–40.
... Another future avenue for research is to understand the role of individual differences in how personality impacts MIA. Research building on interactionist theories 37 has shown that individual differences in musical preferences regarding both broad styles and specific attributes are associated with personality traits and cognition 17,[38][39][40] . Furthermore, recent research has shown evidence for the self-congruity effect of music, whereby listeners prefer the music of artists who have similar personal characteristics as themselves 41 . ...
Article
Full-text available
Music-induced analgesia (MIA) is a phenomenon that describes a situation in which listening to music influences pain perception. The heterogeneity of music used in MIA studies leads to a problem of a specific effect for an unspecified stimulus. To address this, we use a previously established model of musical preferences that categorizes the multidimensional sonic space of music into three basic dimensions: arousal, valence and depth. Participants entered an experimental pain stimulation while listening to compilations of short musical excerpts characteristic of each of the three attribute dimensions. The results showed an effect on the part of music attribute preferences on average pain, maximal pain, and pain tolerance after controlling for musical attributes and order effects. This suggests that individual preferences for music attributes play a significant role in MIA and that, in clinical contexts, music should not be chosen arbitrarily but according to individual preferences.
... Musical preferences and personality traits are linked (103). Psychology, marketing Facebook likes Bystanders will help in public conflict (104). ...
Article
Full-text available
Social science is entering a golden age, marked by the confluence of explosive growth in new data and analytic methods, interdisciplinary approaches, and a recognition that these ingredients are necessary to solve the more challenging problems facing our world. We discuss how developing a “lingua franca” can encourage more interdisciplinary research, providing two case studies (social networks and behavioral economics) to illustrate this theme. Several exemplar studies from the past 12 y are also provided. We conclude by addressing the challenges that accompany these positive trends, such as career incentives and the search for unifying frameworks, and associated best practices that can be employed in response.
... Musical preferences have been linked to personality factors (Dunn et al., 2012;Katz-Gerro, 1999;Nave et al., 2018). Most people's musical preference dimensions fall into one of four broad categories: a) reflective and complex -classical, blues, jazz, and folk music; b) intense and rebellious -rock, alternative, and heavy metal music; c) upbeat and conventional -country, sound track, religious, and pop music; and d) energetic and rhythmic -rap/hip-hop, soul/funk, and electronica/dance music. ...
Article
In the current digital era, we continuously create records of our activities, that are accumulated in a variety of data-storages. One common way to protect our privacy is to remove identifiers (e.g., ID, name) from the records. This approach is known to be naive, as in many cases re-identification is enabled based on quasi-identifiers (e.g., age, gender). In this research we examine an interesting and unexpected new quasi-identifier – music selections of an individual which represents their musical preferences. In the current era we consume music mainly on-demand by streaming (e.g., Spotify, YouTube, Apple Music) rather than as broadcast. The prosperity of the various music platforms is immense, and so is the sharing of beloved music, for example on online social networks. Thus, the creation of records that represent music selections is prevalent. In this paper we introduce a methodology to re-identify users based on their music selections, and prove the efficiency of the methodology empirically in four experiments (n=22,38,35,30). We discuss the social and emotional benefits of the current way we listen to music, against the threat of privacy disclosure.
... Musical preferences have been linked to personality factors (Dunn et al., 2012;Katz-Gerro, 1999;Nave et al., 2018). Most people's musical preference dimensions fall into one of four broad categories: a) reflective and complex -classical, blues, jazz, and folk music; b) intense and rebellious -rock, alternative, and heavy metal music; c) upbeat and conventional -country, sound track, religious, and pop music; and d) energetic and rhythmic -rap/hip-hop, soul/funk, and electronica/dance music. ...
Article
In the contemporary digital era, users may take actions which may affect their privacy and security without being aware of the consequences. Knowing the consequences, may be perceived as a higher level of usability, above the basic operational skills. This phenomenon is mainly a direct outcome of the gap between users’ knowledge and the complexity of the digital systems. The ability to master those systems is known as technological literacy, which thus is a critical factor to protect privacy and security. However, the process of acquiring technological literacy is costly and not trivial. In this research, we address the problem with a qualitative approach, and introduce a methodology named OTLA (Optimising Technological Literacy Acquirement) to achieve optimality between the cost of the teaching process and its benefits. We conducted an empirical study with n = 320 participants on five technological literacy issues, and showed that optimal solutions to this problem can be found. Thus, OTLA has a significant contribution to protecting users’ privacy and security in the digital world.
... The degree of replication can also be judged on a continuous numerical scale, by the size of the effect estimated in the replication compared to the size of the effect in the original study. Our method uses machine learning to predict outcomes and identify the characteristics of study-replication pairs that can best explain the observed replication results (Camerer et al., 2018a;Hastie et al., 2009;Nave et al., 2018;Yarkoni and Westfall, 2017). We divide the objective features of the original experiment into two classes. ...
Preprint
Full-text available
We measure how accurately replication of experimental results can be predicted by a black-box statistical model. With data from four large- scale replication projects in experimental psychology and economics, and techniques from machine learning, we train a predictive model and study which variables drive predictable replication.The model predicts binary replication with a cross validated accuracy rate of 70% (AUC of 0.79) and relative effect size with a Spearman ρ of 0.38. The accuracy level is similar to the market-aggregated beliefs of peer scientists (Camerer et al., 2016; Dreber et al., 2015). The predictive power is validated in a pre-registered out of sample test of the outcome of Camerer et al. (2018b), where 71% (AUC of 0.73) of replications are predicted correctly and effect size correlations amount to ρ = 0.25.Basic features such as the sample and effect sizes in original papers, and whether reported effects are single-variable main effects or two- variable interactions, are predictive of successful replication. The models presented in this paper are simple tools to produce cheap, prognostic replicability metrics. These models could be useful in institutionalizing the process of evaluation of new findings and guiding resources to those direct replications that are likely to be most informative.
... We consider a large number of such candidate observable features in search of a small set that is predictive, using cross-validation (Stone, 1974) to control for overfitting. This machine learning approach has been used in many applications in computer science and neuroscience, and is beginning to be more widely used in economics (Krajbich et al., 2009;Belloni et al., 2012;Einav and Levin, 2014;Varian, 2014;Smith et al., 2014;Mullainathan, 2014;Bajari et al., 2015) and other social sciences (Dzyabura and Hauser, 2011;Youyou, Kosinski and Stillwell, 2015;Yarkoni and Westfall, 2016;Nave et al., 2017). ...
Article
Full-text available
We study dynamic unstructured bargaining with deadlines and one-sided private information about the amount available to share (the " pie size"). Using mechanism design theory, we show that given the players' incentives, the equilibrium incidence of bargaining failures (" strikes ") should increase with the pie size, and we derive a condition under which strikes are efficient. In our setting, no equilibrium satisfies both equality and efficiency in all pie sizes. We derive two equilibria that resolve the trade-off between equality and efficiency by either favoring equality or favoring efficiency. Using a novel experimental paradigm, we confirm that strike incidence is decreasing in the pie size. Subjects reach equal splits in small pie games (in which strikes are efficient), while most payoffs are close to either the efficient or the equal equilibrium prediction, when the pie is large. We employ a machine learning approach to show that bargaining process features recorded early in the game improve out of sample prediction of disagreements at the deadline. The process feature predictions are as accurate as predictions from pie sizes only, and adding process and pie data together improves predictions even more.
Article
Full-text available
U ovom je istraživanju provjereno replicira li se petofaktorska struktura Modela glazbenih preferencija MUSIC (Rentfrow, Goldberg i Levitin, 2011), čije dimenzije odražavaju preferencije prema blagoj, nepretencioznoj, sofisticiranoj, intenzivnoj i suvremenoj glazbi, na hrvatskom uzorku mladih. Također, ispitano je replicira li se trofaktorska struktura Inventara motiva slušanja glazbe (Uses of Music Inventory, UMI, Chamorro-Premuzic i Furnham, 2007), koji uključuje emocionalno i kognitivno slušanje glazbe te slušanje glazbe u pozadini. Istraživanjem je ispitan odnos glazbenih preferencija, motiva slušanja glazbe i crta ličnosti. U prvom dijelu istraživanja 457 ispitanika (74.2 % ispitanica) prosječne dobi od 20 godina ispunilo je Inventar motiva slušanja glazbe te Petofaktorski inventar ličnosti (Big Five Inventory, BFI, Benet-Martinez i John, 1998). U drugom su dijelu istraživanja procjenjivali stupanj sviđanja za 25 glazbenih isječaka (Rentfrow i sur., 2011) pomoću kojih se ispituju glazbene preferencije. Rezultati ovog istraživanja uputili su na zadovoljavajuću strukturu modela MUSIC kad se u ispitivanje glazbenih preferencija uključio 21 isječak. Također, zadovoljavajućom se pokazala i trofaktorska struktura Inventara motiva slušanja glazbe. Povezanosti glazbenih preferencija i motiva slušanja glazbe s crtama ličnosti uglavnom su replicirane. Crte ličnosti pridonijele su objašnjenju varijance glazbenih preferencija povrh dobi i spola, a kao samostalni prediktori uglavnom su se istaknule otvorenost i ekstraverzija. Uz crte ličnosti, motivi slušanja glazbe objasnili su preferencije prema blagoj, intenzivnoj i suvremenoj glazbi. Pri tome, emocionalno slušanje predviđalo je preferencije prema blagoj glazbi, dok je kognitivno slušanje bilo negativni prediktor iste. Kognitivno slušanje glazbe predviđalo je preferencije prema intenzivnoj glazbi, a isti je motiv slušanja glazbe bio negativni prediktor preferencija prema suvremenoj glazbi. Osim toga, preferencije prema suvremenoj glazbi pozitivno su predviđale i emocionalno slušanje glazbe i slušanje glazbe u pozadini.
Preprint
Full-text available
Our study aimed to approach validity issues that arise both in the case of surveys and Facebook data. We conducted a novel parallel data collection method, rarely used before, which combines a face-to-face survey with the collection of personal FB data archives. We show the research potential of using these alternative data sources together. We show that we can overcome some validity issues by cross-validating the two data sources by each other. We also show that if the researcher collaborates with the users instead of the companies and gets access to a part of the users' social media data archive, the problem of the leanness of social media data is eliminated. Our chosen topic is music interest, a key indicator of cultural sociology. To our knowledge, there are no previous studies on the rate at which any population is categorized by Facebook to advertisers as interested in music genres, or on the relationship between self-reported interest, digitally expressed interest and ad-interest categories. Specific genres were detected which show remarkable different pictures when measured these different ways. We studied also dynamic patterns in FB users music interest. We hope our data collection method and the presented validity approaches will initiate a future dialogue in digital data research in social sciences.
Article
Full-text available
We study dynamic unstructured bargaining with deadlines and one-sided private information about the amount available to share (the " pie size"). Using mechanism design theory, we show that given the players' incentives, the equilibrium incidence of bargaining failures (" strikes ") should increase with the pie size, and we derive a condition under which strikes are efficient. In our setting, no equilibrium satisfies both equality and efficiency in all pie sizes. We derive two equilibria that resolve the trade-off between equality and efficiency by either favoring equality or favoring efficiency. Using a novel experimental paradigm, we confirm that strike incidence is decreasing in the pie size. Subjects reach equal splits in small pie games (in which strikes are efficient), while most payoffs are close to either the efficient or the equal equilibrium prediction, when the pie is large. We employ a machine learning approach to show that bargaining process features recorded early in the game improve out of sample prediction of disagreements at the deadline. The process feature predictions are as accurate as predictions from pie sizes only, and adding process and pie data together improves predictions even more.
Article
Full-text available
Research suggests that musical preferences are linked to personality, but this research has been hindered by genre-based theories and methods. We address this limitation using a novel method based on the actual attributes that people perceive from music. In Study 1, using 102 musical pieces representing 26 genres and subgenres, we show that 38 perceived attributes in music can be organized into three basic dimensions: arousal, valence, and depth. In Study 2 (N = 9,454), we show that people’s preferences for these musical attributes reflected their self-ratings of personality traits. Importantly, personality was found to predict musical preferences above and beyond demographic variables. These findings advance previous theory and research and have direct applications for the music industry, recommendation algorithms, and health-care professionals.
Article
Full-text available
In contrast to decades of research reporting surprisingly weak relationships between consumption and happiness, recent findings suggest that money can indeed increase happiness if it is spent the "right way" (e.g., on experiences or on other people). Drawing on the concept of psychological fit, we extend this research by arguing that individual differences play a central role in determining the "right" type of spending to increase well-being. In a field study using more than 76,000 bank-transaction records, we found that individuals spend more on products that match their personality, and that people whose purchases better match their personality report higher levels of life satisfaction. This effect of psychological fit on happiness was stronger than the effect of individuals' total income or the effect of their total spending. A follow-up study showed a causal effect: Personality-matched spending increased positive affect. In summary, when spending matches the buyer's personality, it appears that money can indeed buy happiness.
Article
Full-text available
The neurohormone oxytocin (OT) has been one the most studied peptides in behavioral sciences over the past two decades. Primarily known for its crucial role in labor and lactation, a rapidly growing literature suggests that intranasal OT (IN-OT) may also play a role in humans’ emotional and social lives. However, the lack of a convincing theoretical framework explaining IN-OT’s effects that would also allow to predict which moderators exert their effects and when, has raised healthy skepticism regarding the robustness of human behavioral IN-OT research. The poor knowledge of OT’s exact pharmacokinetic properties, crucial statistical and methodological issues and the absence of direct replication efforts may have lead to a publication bias in IN-OT literature with many unpublished studies with null results lying in laboratories’ drawers. Is there a file drawer problem in IN-OT research? If this is the case, it may also be the case in our laboratory. This paper aims to answer that question, document the extent of the problem and discuss its implications for OT research. Through eight studies (including 13 dependent variables overall, assessed through 25 different paradigms) performed in our lab between 2009 and 2014 on 453 subjects, results were too often not those expected. Only five publications emerged from our studies and only one of these reported a null-finding. After realizing that our publication portfolio has become less and less representative of our actual findings and because the non-publication of our data might contribute to generating a publication bias in IN-OT research, we decided to get these studies out of our drawer and encourage other laboratories to do the same.
Article
Full-text available
Why do we like the music we do? Research has shown that musical preferences and personality are linked, yet little is known about other influences on preferences such as cognitive styles. To address this gap, we investigated how individual differences in musical preferences are explained by the empathizing-systemizing (E-S) theory. Study 1 examined the links between empathy and musical preferences across four samples. By reporting their preferential reactions to musical stimuli, samples 1 and 2 (Ns = 2,178 and 891) indicated their preferences for music from 26 different genres, and samples 3 and 4 (Ns = 747 and 320) indicated their preferences for music from only a single genre (rock or jazz). Results across samples showed that empathy levels are linked to preferences even within genres and account for significant proportions of variance in preferences over and above personality traits for various music-preference dimensions. Study 2 (N = 353) replicated and extended these findings by investigating how musical preferences are differentiated by E-S cognitive styles (i.e., 'brain types'). Those who are type E (bias towards empathizing) preferred music on the Mellow dimension (R&B/soul, adult contemporary, soft rock genres) compared to type S (bias towards systemizing) who preferred music on the Intense dimension (punk, heavy metal, and hard rock). Analyses of fine-grained psychological and sonic attributes in the music revealed that type E individuals preferred music that featured low arousal (gentle, warm, and sensual attributes), negative valence (depressing and sad), and emotional depth (poetic, relaxing, and thoughtful), while type S preferred music that featured high arousal (strong, tense, and thrilling), and aspects of positive valence (animated) and cerebral depth (complexity). The application of these findings for clinicians, interventions, and those on the autism spectrum (largely type S or extreme type S) are discussed.
Article
A meta-analysis was performed on the results of previous studies investigating the association between personality traits and music preferences. Regarding the categorization of personality traits, the Big Five and sensation seeking were used most often and were therefore chosen as the most appropriate categories in the meta-analysis. Regarding the categorization of musical style preferences, the five-dimensional MUSIC (mellow, unpretentious, sophisticated, intense, contemporary) model was used most often and was therefore employed in the meta-analysis. Hence, we included studies in the analysis when they had investigated the relationship between at least one of the Big Five personality traits or sensation seeking and at least one of the five MUSIC dimensions. In total, there were 30 subanalyses. All weighted averaged correlation coefficients were very small, with most of them near zero. Only 6 of the 30 coefficients exceeded 0.1 in magnitude (|r| ≥ 0.1). The largest effects were observed for the openness to experience personality trait, which exhibited small correlations with preference for three musical styles. Thus, personality traits barely account for interindividual differences in music preferences. Musical functions are discussed as an alternative explanation for these differences. The predictability of musical style preferences based on individual psychological variables is questioned in general.
Article
INTRODUCTION: Reproducibility is a defining feature of science, but the extent to which it characterizes current research is unknown. Scientific claims should not gain credence because of the status or authority of their originator but by the replicability of their supporting evidence. Even research of exemplary quality may have irreproducible empirical findings because of random or systematic error. RATIONALE: There is concern about the rate and predictors of reproducibility, but limited evidence. Potentially problematic practices include selective reporting, selective analysis, and insufficient specification of the conditions necessary or sufficient to obtain the results. Direct replication is the attempt to recreate the conditions believed sufficient for obtaining a previously observed finding and is the means of establishing reproducibility of a finding with new data. We conducted a large-scale, collaborative effort to obtain an initial estimate of the reproducibility of psychological science. RESULTS: We conducted replications of 100 experimental and correlational studies published in three psychology journals using high-powered designs and original materials when available. There is no single standard for evaluating replication success. Here, we evaluated reproducibility using significance and P values, effect sizes, subjective assessments of replication teams, and meta-analysis of effect sizes. The mean effect size (r) of the replication effects (Mr = 0.197, SD = 0.257) was half the magnitude of the mean effect size of the original effects (Mr = 0.403, SD = 0.188), representing a substantial decline. Ninety-seven percent of original studies had significant results (P < .05). Thirty-six percent of replications had significant results; 47% of original effect sizes were in the 95% confidence interval of the replication effect size; 39% of effects were subjectively rated to have replicated the original result; and if no bias in original results is assumed, combining original and replication results left 68% with statistically significant effects. Correlational tests suggest that replication success was better predicted by the strength of original evidence than by characteristics of the original and replication teams. CONCLUSION: No single indicator sufficiently describes replication success, and the five indicators examined here are not the only ways to evaluate reproducibility. Nonetheless, collectively these results offer a clear conclusion: A large portion of replications produced weaker evidence for the original findings despite using materials provided by the original authors, review in advance for methodological fidelity, and high statistical power to detect the original effect sizes. Moreover, correlational evidence is consistent with the conclusion that variation in the strength of initial evidence (such as original P value) was more predictive of replication success than variation in the characteristics of the teams conducting the research (such as experience and expertise). The latter factors certainly can influence replication success, but they did not appear to do so here. Reproducibility is not well understood because the incentives for individual scientists prioritize novelty over replication. Innovation is the engine of discovery and is vital for a productive, effective scientific enterprise. However, innovative ideas become old news fast. Journal reviewers and editors may dismiss a new test of a published idea as unoriginal. The claim that “we already know this” belies the uncertainty of scientific evidence. Innovation points out paths that are possible; replication points out paths that are likely; progress relies on both. Replication can increase certainty when findings are reproduced and promote innovation when they are not. This project provides accumulating evidence for many findings in psychological research and suggests that there is still more work to do to verify whether we know what we think we know.
Article
A generalized form of the cross‐validation criterion is applied to the choice and assessment of prediction using the data‐analytic concept of a prescription. The examples used to illustrate the application are drawn from the problem areas of univariate estimation, linear regression and analysis of variance.
Article
The reproducibility of scientific findings has been called into question. To contribute data about reproducibility in economics, we replicate 18 studies published in the American Economic Review and the Quarterly Journal of Economics in 2011-2014. All replications follow predefined analysis plans publicly posted prior to the replications, and have a statistical power of at least 90% to detect the original effect size at the 5% significance level. We find a significant effect in the same direction as the original study for 11 replications (61%); on average the replicated effect size is 66% of the original. The reproducibility rate varies between 67% and 78% for four additional reproducibility indicators, including a prediction market measure of peer beliefs.
Article
Behavioral neuroscientists have shown that the neuropeptide oxytocin (OT) plays a key role in social attachment and affiliation in nonhuman mammals. Inspired by this initial research, many social scientists proceeded to examine the associations of OT with trust in humans over the past decade. To conduct this work, they have (a) examined the effects of exogenous OT increase caused by intranasal administration on trusting behavior, (b) correlated individual difference measures of OT plasma levels with measures of trust, and (c) searched for genetic polymorphisms of the OT receptor gene that might be associated with trust. We discuss the different methods used by OT behavioral researchers and review evidence that links OT to trust in humans. Unfortunately, the simplest promising finding associating intranasal OT with higher trust has not replicated well. Moreover, the plasma OT evidence is flawed by how OT is measured in peripheral bodily fluids. Finally, in recent large-sample studies, researchers failed to find consistent associations of specific OT-related genetic polymorphisms and trust. We conclude that the cumulative evidence does not provide robust convergent evidence that human trust is reliably associated with OT (or caused by it). We end with constructive ideas for improving the robustness and rigor of OT research.