Negative Stereotypes Associated with Heavy Metal and Hip Hop

Heavy-Metal and hip-hop music are two representative genres of so-called "problem music" (Bodner & Bensimon, 2015; Lozon & Bensimon, 2014), often thought to be associated with emotion words such as anger and disgust, high arousal (energy) and negative valence emotions (that is, occurring in the extremes of the second quadrant of a two-dimensional emotion space, see Russell, 1980). Barongan and Hall (1995) studied the effects of misogynistic lyrics in rap (i.e., hip-hop) music on White American participants. Two groups of listeners were assigned to one of two conditions. Participants in Group 1 were asked to listen to misogynous rap music and Group 2 were asked to listen to non-violent rap (i.e., lyrics were non-misogynous). After listening to the music, the listeners were asked to choose a short film vignette portraying either a neutral, sexually-aggressive, or assaultive scene. The group of participants who listened to the misogynistic music chose the sexually-aggressive vignette significantly more than the participants who listened to the non-violent music. This suggests that misogynous rap music facilitates sexually aggressive behavior, consistent with the derogatory stereotypes of rap music (Neguţ & Sârbescu, 2014; Reyna, Brandt, & Viki, 2009).

Barongan and Hall's (1995) study suggests that sexually aggressive association with the music was informed by the semantic content of the lyrics. However, overall, these participants reported that they disliked the music, with mean scores for liking of less than 3 on a scale from 1 (strongly disliked) to 5 (strongly liked). The listeners therefore had low fandom for rap music, and, as STEM proposes, may have been more disposed to stereotyping rap with sexual aggression. Several studies investigating misogynistic and violent lyrics in "problem music" have argued that lyrics induce negative effects in a listener (Cundiff, 2013; Lennings & Warburton, 2011; Martino et al., 2006; Mast & McAndrew, 2011), including negative stereotyping effects regardless of the individual's relationship with and knowledge about the genre. However, the role of expertise has, to our knowledge, not been explicitly controlled.

Dunbar, Kubrin, and Scurich (2016) investigated whether Americans' perception of rap music was influenced by genre-specific stereotypes. Replicating a study conducted by Fried (1999), Dunbar and colleagues tested whether listeners presented with lyrics labelled as rap were perceived as more threatening, dangerous and in need of regulation in comparison to identical lyrics labelled as country. Participants were randomly assigned to either lyrics labelled as rap or labelled as taken from a country song. They were then asked to respond to 14 different items on a 9-point Likert scale from 1 (strongly disagree) to 9 (strongly agree), measuring the offensiveness of the lyrics, their threatening nature and if the song should be regulated. The findings suggested that when the lyrics were labelled as taken from a rap song, they were considered more negative and in need of more regulation, compared to when the same lyrics were labelled as taken from a country song, consistent with Fried's (1999) results. The authors concluded that rap music may overall be negatively stereotyped.

Associations with lyrics can therefore be influenced by their genre membership, suggesting that stereotypes of the music genre might inform other associations, including emotional responses to music. Furthermore, it may be that the individual's relationship with and knowledge of the music (e.g., fan versus non-fan) will influence the extent to which stereotyping occurs (e.g., because of the more filtered—or stereotyped—emotions availed to the non-fan). Heavy-metal and hip-hop music are ideal genres to study because of their identification in some literature as "problem music" (Lozon & Bensimon, 2014) and often assumed expression of negative, high-arousal emotions, based on the foregoing review of the literature. Consequently, these two genres provide material for investigating possible influence of fandom predicted by STEM and the influence of lyrics in particular.

Participants

Two hundred and thirty-eight participants were recruited. Their ages ranged from 18 to 42 years (M = 22.3, SD = 3.76; 151 females and 87 males). Participants were distributed at random into one of two sets, A or B (explained in the Procedure section). Set A comprised 115 participants (M = 20.7, SD = 2.89 years; 86 females and 29 males) and Set B comprised 123 participants (M = 20.67, SD = 2.0 years; 65 females and 58 males). All participants were undergraduate students at a tertiary education institution in a multicultural city of Australia. Participants received course credit for participating in the study.

Materials

The stimuli were comprised of eight music excerpts each of approximately 24 seconds in duration and sharing lyrics across combinations of genre pairs, the genres investigated being heavy metal, hip hop, and, as a control, pop (see Table 1).

For each excerpt, we selected a continuous passage of lyrics that was either an entire verse or refrain (i.e., chorus). The lyric passages were of varying word length (range 22-57 words) with an average of 40.50 words per excerpt.

Design

Participant responses were collected using an online survey designed using Key Survey software (www.keysurvey.com). The music excerpts were divided into two sets of four. Each set consisted of two test excerpts and two control excerpts. Set A consisted of two hip-hop stimuli (test excerpts), and two pop stimuli (control excerpts). Set B consisted of two heavy metal stimuli (test), and two pop stimuli (control). The lyrics of the two pop control excerpts in Set A were matched with the lyrics of the two heavy-metal test excerpts in Set B. The lyrics of the two pop control excerpts in Set B were matched with the lyrics of the two hip-hop test excerpts in Set A. Using this design, participants heard excerpts with different lyrics, but the lyrics across the sets were identical. Only genre was manipulated across the sets.

Table 1. Music excerpt, test and control genre, and lyrics sentiment analysis

Music excerpt number and name	Control stimulus, music genre, and set membership	Test stimulus, music genre, and set membership	Lyrics	LIWC Positive Emotions %	LIWC Negative Emotions %	iFeel Most Frequent Emotion Polarity
1. Lay Lady Lay	Bob Dylan (1969) Pop (Set A)	Ministry (1996) Heavy metal (Set B)	Why wait any longer for the world to begin You can have your cake and eat it too Why wait any longer for the one you love When he's standing in front of you	2.86	0	Positive
2. Bitches Ain't Shit	Ben Folds (2005) Pop (Set B)	Dr. Dre (feat. Jewel, Tha Dogg Pound and Snoop Doggy Dogg) (1992) Hip hop (Set A)	I'm back to the motherfuckin' county jail Six months on my chest, now it's time to bail I get released on a hot sunny day My nigga D.O.C. and my homey Dr Dre Scooped in a coupe, Snoop we got news Your girl was trickin' while you was draped in your country blues I ain't been out a second And already gotta do some motherfuckin' chin checkin'	1.75	1.75	Positive Negative Neutral
3. In-A-Gadda-Da-Vida	Iron Butterfly (1968) Pop (Set A)	Slayer (1987) Heavy metal (Set B)	Oh, won't you come with me And take my hand Oh won't you come with me And walk this land Please take my	0	0	Positive
4. Sleazy	Ben Folds (2010) Pop (Set B)	Kesha Sebert (2010) Hip hop (Set A)	I don't need you or your brand-new Benz Or your bougie friends I don't need love, looking like diamonds	4.17	0	Positive

Sentiment Analysis of Lyrics

To document the nature of the emotion expressed by the lyrics (which were identical across Sets A and B), sentiment analysis software was used to examine the prevalence of positive emotion words and negative emotion words, generated by the Linguistic Analysis Word Count (LIWC) software 2007 (Pennebaker, Boyd, Jordan, & Blackburn, 2015; Pennebaker, Francis, & Booth, 2007). LIWC cross-analyzes text with a dictionary of almost 6,400 words, assessing the target text according to the degree to which it expresses positive and negative emotion (among other aspects of sentiment) (Tausczik & Pennebaker, 2010). To cross-check LIWC's output, we also used the output of 18 sentiment analysis tools generated by iFeel Sentiment Analysis Framework (http://blackbird.dcc.ufmg.br:1210). iFeel categorizes each of these outputs into positive, negative or neutral emotion polarity from which the most frequent polarity can be determined (Araújo, Gonçalves, Cha, & Benevenuto, 2014). Table 1 summarizes the LIWC and iFeel emotion valence results of the lyrics. Overall, three of four passages rated 0 on the LIWC negative emotion scale, while 1 passage (Stimulus 2, see Table 1) achieved equal percentages for both positive emotions and negative emotions (1.75%). Both analyses show that the lyrics in the stimuli were overall positive.

Procedure

Ethics approval was obtained from the host university's Ethics Advisory Committee. Participants were assigned at random to one of two excerpt sets: A or B. All participants completed the questionnaire online in their own time. The study was conducted in three stages. In the first stage, participants were asked to report their fandom for heavy-metal, hip-hop and pop genres on a scale from 1 (non-fan – never listen to it) to 5 (fan – listen to it all the time).

In the second stage, participants were presented with the four stimuli (one at a time) from their assigned set (Table 1) in a different random order for each participant. After listening to a stimulus, participants were asked what emotion they thought the music was trying to express: "Please listen to the following music excerpt only once. From your first impression, what emotion do you think the music is trying to express?" The answer options given to the participants were: anger; disgust; fear; happiness; sadness; and surprise. In addition to these six emotion options, we added two further possible responses, "none" and "other", to increase the likelihood that participants were able to report an emotion spontaneously, without necessarily being restricted to the emotions listed (for a discussion, see Russell, 1994). If the music expressed an emotion not mentioned in the list provided, participants were instructed to choose the option "other", and to type which other emotion was expressed by the music. After the emotion reporting task, participants were asked if they were familiar with the excerpt they just heard, provided with a yes/no answer option, and if they could name (or guess the name of) the song by typing their response.

In the third stage, participants were presented with a surprise memory task comprising a list of 11 words. Six of the words were target words that were featured in the lyrics for the music stimuli and five were foils that were not featured in the lyrics of the music. The target words were: cake, diamonds, hand, jail, land, and world, as indicated in bold in Table 1. The foil words presented were: beach, bloodthirsty, dope, prayer, and summer. Participants were asked to select the words they thought were sung: "Please check any of the following words you think you heard in the previous music excerpts." Two of the five foil words, bloodthirsty and dope, were selected because they were both words often associated with heavy metal (Noys & Schmitt, 2010; Roccor, 2000) and hip hop (Alim, 2006; Morgan, 2017), respectively, even though they appeared in none of the stimuli used.

Data Pre-Processing for Stage-Two Responses

To reduce the effect of emotion perception through explicit knowledge of the music excerpts, we identified and removed from further analysis participant responses that correctly recalled or guessed the name of the excerpts. Fifteen responses were removed. Of the 223 remaining participants, 107 responded to hip-hop test excerpts (Set A) and 116 to heavy-metal test excerpts (Set B), in addition to the respective pop control excerpts. Table 2 summarizes these participants' Stage-Two responses to each stimulus.

Table 2. Raw counts of reported emotional responses to test and control stimuli.

	Genre Condition
Emotion	Test Condition				Control Condition
	Heavy metal 1	Hip hop 2	Heavy metal 3	Hip hop 4	Pop 1	Pop 2	Pop 3	Pop 4
Anger	87	38	71	31	5	5	6	0
Disgust	11	40	11	23	5	3	23	3
Fear	2	1	3	3	4	6	1	1
Happiness	3	6	7	8	3	32	13	21
Sadness	5	0	1	5	85	13	33	56
Surprise	1	4	4	2	1	5	3	2
None	0	6	13	17	3	23	10	7
Other	7	12	6	18	10	20	27	17
Total	116	107	116	107	116	107	116	107

Data Pre-Processing for "Other" Emotions

Responses for "other" emotions totaled 117. These were pre-processed to identify the categories of emotions used in subsequent analysis. This involved eight data steps, based on Augustin, Wagemans, and Carbon (2012) and on Susino and Schubert (2018):

1. Correction of spelling errors.
2. Extraction of task-related parts from a sentence. For example: "Happy and sad" was coded as one count of "happy" and one count of "sad".
3. Removal of function words (articles and pronouns).
4. Removal of qualifiers. For example, "mostly angry" was coded as "anger".
5. Consolidating of singulars and plurals for the same noun.
6. Consolidating nouns and corresponding adjectives (such as, aggression, aggressive).
7. Consolidating words that have the same stem (i.e., part of a word). For example, "hope" and "hopeful" have the same stem, "hope". The term chosen to represent emotions with the same stem, such as "hope/hopeful" was the one with the highest frequency. When this was not possible due to equal counts, it was based on the researchers' discretion.
8. A total of 99 emotion responses were identified by author MS and checked by author ES. Of the 99 emotion responses identified, 18 were determined to be unique emotion terms and distributed in emotion space, as described below.

Categorization of Emotion Responses Based on Emotion Space Quadrants

We first examined the distribution of emotion responses for the test genres (heavy metal and hip hop) in comparison to the control genre (pop). Six categories were formed, and these categories constituted the dependent variable, of which five were emotion categories and one was a non-emotion category. Four of the five emotion categories were based on regions of the two-dimensional emotion space in which emotions are distributed along valence and arousal dimensions in a two-dimensional space (Russell, 1980; Schubert, 1999). To ensure that the high arousal/negative valence (referred to as 'Quadrant 2') region captured a reasonably homogenous set of emotions, the criterion for membership of that category was made narrower than "anywhere within Quadrant 2". The region was set based on empirical data of the emotion word exemplars on so-called "problem music," described in the literature review and as applied in the checklist of emotion words (specifically, anger, disgust, and fear). The same approach was taken to more tightly defining the regions for the other three quadrants as shown in Figure 1.

The quantification of the boundaries within these quadrants on the emotion space are reported in Table 3, based on the Affective Norms for English Words (ANEW, Bradley and Lang, 1999) database values for the emotions from the checklist belonging to each quadrant. The ANEW database lists empirically assessed valence and arousal values for English emotion words on a scale from 1 (low) to 9 (high). We used the scores for "happy" and "surprise" to determine the boundaries within which other words would be classified as Q1, meaning that to be a member of the Q1 region, words would require a valence score not lower than 7.47 (the lowest valence value of the checklist of Q1 words), and an arousal score not lower than 6.49 (the lowest arousal value of the initial checklist of the Q1 region words). All emotion terms for which arousal and valence values did not fit any of the Q1 to Q4 regions were categorized as "other emotion" (OE), such as "hope" (note that OE is not the same as the raw, self-reported "other"-category option collected in Stage 2 of the procedure). The location of the quadrant regions and the OE category boundaries can be visualized using the emotion-space shown in Figure 1. Furthermore, words were categorized as "no emotion" (NE) for participants who reported "none" in Stage 2 or when "other" consisted of a non-emotion term, such as "criticism."

Table 3. Emotion categorization for analysis.

Category	Emotion	Emotion Space region boundary*
		Valence Mean	Arousal Mean
Q1	Happy	8.21	6.49
Q1	Surprised+	7.47	7.47
Q2	Anger	2.34	7.63
Q2	Disgusted+	2.45	5.42
Q2	Fear	2.76	6.96
Q3	Sad	1.61	4.13
Q4	Relaxed**	7.00	2.39
OE (Other Emotion)	Other emotions that do not fit into one of the four rectangular regions as defined above. On those occasions when an emotion word did not fit any of rectangular regions Q1 to Q4, but were clearly emotions, they were classified as OE.
NE (No Emotion)	No emotion or reported as "other" but could not be placed into one of the above five categories.

Using a chi-square analysis, we tested the distribution of each emotion category and non-emotion category for all the test stimuli (heavy metal and hip hop combined), using the control (pop) stimuli as the comparison distribution. There was a significant overall difference in distribution of test stimuli compared to control stimuli (χ²(5, N = 81) = 352.13, p < .001, Cramer's V = .397). As evident in Table 4, count of emotion words in Q2 (high arousal/negative valence) is considerably overrepresented in the test condition, in comparison to the other emotion categories. Note, too, that counts for each emotion category other than Q2 have similar or higher counts in the control condition, suggesting a wider range of emotions used to describe pop music in comparison to heavy metal or hip hop (combined).

Table 4. Observed counts for control and test conditions.

Category (sample emotion word)	Control counts	Test counts
Q1 (happy, surprised)	86	38
Q2 (anger, disgust, fear)	66	324
Q3 (sad)	189	13
Q4 (relaxed)	8	3
OE	77	60
NE	20	8

To further test H1, we compared the proportion of Q2 (high arousal/negative valence) counts with the remaining five categories—Q1, Q3, Q4, OE and NE— that would be expected to occur by chance for the test conditions and again for the control conditions. Using a chi-square analysis, results showed no significant deviation from the distribution expected by chance for pop music control condition (χ² (1, N = 65) = 0.542 p = 0.462, phi = .0913). However, a significant deviation was returned for the test condition (χ²(2, N = 323) = 283.553 p < 0.01, phi = .9369). Heavy-metal and hip-hop conditions had above-chance representation of high arousal/negative valence (Q2) emotions whereas pop did not, even though the lyrics were identical between the pop control and the test (heavy-metal and hip-hop) excerpts. This is consistent with H1.

High Arousal/Negative Valence Emotion Responses Based on Fandom

The overall fandom for each of the genres was examined. Low overall fandom was reported for genres that produced stereotyped emotions: Heavy metal (fandom M = 1.93, SD = 1.08) and hip hop (fandom M = 3.32, SD = 1.15) had the lowest means and pop had the highest (fandom M = 4.14, SD = 0.94). To test H2 we used fandom scores as the dependent variable, with the stimulus as the independent variable. A Kolmogorov-Smirnov test indicated that fandom scores did not follow a normal distribution, D(223) = 0.25, p < .0001 making parametric testing inappropriate. The distribution of fandom ratings for pop music was negatively skewed with a very small number of participants indicating 1 (n = 4) and 2 (n = 8) fandom scores (fandom scale ranged from 1 to 5), making pooled fandom scores potentially misleading because high fandom scores may be confounded by fandom for pop music, rather than fandom in general. We therefore decided to eliminate pop music responses for this analysis and conduct a Kolmogorov-Smirnov test comparing heavy-metal and hip-hop fandom scores.

We examined fandom scores for participants who perceived heavy metal and hip hop as expressing high arousal/negative valence emotions (Q2) in comparison to other quadrants, with emotion being the independent variable having two levels, Q2 and non-Q2 (i.e., combination of Q1, Q3, Q4, and OE). A Kolmogorov-Smirnov test showed a significant deviation from normality, D(447) = 0.95, p < .05 making parametric testing inappropriate. The distribution of fandom ratings on a scale range from 1 to 5 for heavy metal and hip hop combined was 1 (n = 85), 2 (n = 77), 3 (n = 33), 4 (n = 11), and 5 (n = 12). A Mann-Whitney test indicated that fandom scores were lower for participants reporting Q2 emotions (Mdn = 2, n = 324) than for non Q2 emotions (Mdn = 3, n = 122) (U = 17.48, p = .0055, χ² = .59). The results confirm that low fandom towards a genre is better associated with Q2 (i.e. stereotyped) emotions than with any other emotions, lending support to H2.

Lyric Recall

To test if participants recalled the lyrics, we compared the number of target words and the number of the exclusive foil words selected for each of the four stimuli. We planned to list two target words for each stimulus, such that our expected counts would be divided with a ratio of 2:5 (target to foil), but we unintentionally chose one target word for stimulus 2 and stimulus 4 (with two target words for stimuli 1 and 3—see Table 1). Expected ratios were adjusted accordingly, that is 1:5 for stimuli 2 and 4. Using chi-square tests, we examined the proportion of target-to-foil words recalled. As shown in Table 5, target lyrics were recalled statistically above chance and foil words statistically below chance for each stimulus, except for Pop 2 where they were statistically recalled at chance levels only. With the exception of this excerpt, we concluded that listeners could discriminate the lyrics at above chance levels regardless of genre, but when there was no above-chance recognition, it was not due to any particular characteristic concerned with the intelligibility of the test genres.

Table 5. Chi-square test statistics for target vs. foil words memory and memory comparison for each stimulus.

Stimulus	Observed target (and foil) counts	Expected target (and foil) counts	Chi-square test
Pop 1	106	65.625	39.744, p <.0001
	(69)	(109.375)
Metal 1	37	20.333	16.394, p <.0001
	(85)	(101.667)
Pop 2	25	18.333	2.909, p = .0881
	(85)	(91.666)
Hip hop 2	26	15.833	7.834, p = .0051
	(69)	(79.167)
Pop 3	80	49.666	46.071, p <.0001
	(69)	(99.334)
Metal 3	71	44.571	21.940, p <.0001
	(85)	(111.429)
Pop 4	192	79.143	225.306, p <.0001
	(85)	(197.857)
Hip hop 4	75	24	130.050, p <.0001
	(69)	(120)