Music listening: a tough test for marketing laws
Common sense tells us that different music appeals to listeners with different demographic (e.g., age, gender) and psychographic (e.g., personality type) characteristics. For instance, stereotypically we expect white males to be Rock listeners and old-rich people to be Classical listeners. These stereotypes seem logical. With the rise of music streaming services, there is now on-demand access to millions of songs. In this sense, any physical availability advantages held by large artists are diminished, so less famous artists should be able to compete more easily. Streaming services also use sophisticated algorithms to make music suggestions for each individual. This may create partitioning in the market, or accentuate it further. For these reasons, we might expect marketing laws to not hold for music listening.
If music genres are listened to by different people, then they will not have overlapping listening bases (or minimal overlap). An alternative possibility is that the degree of listener overlap between two genres depends more on their relative popularity, a pattern known as the law of Duplication of Purchase or ‘DoP’ (Goodhardt & Ehrenberg, 1969). This law-like pattern states that brands share their customers in line with penetration — brands share many customers with large brands and far fewer with small brands. So two popular genres (e.g., Rock and Electronic) would very much have an overlapping listener base. And any genre would share many of its listeners with a popular genre like Rock or Pop.
We can also compare the demographic and psychographic listener profiles of genres to determine the presence and extent of differences. If different genres appeal to different people, the demographic and psychographic listener profiles will not be the same. On the other hand, if people listen across genres, the same people will appear in the listening bases of multiple music ‘brands’. Therefore, genres will have a similar demographic and psychographic listener composition. This is an important, and somewhat counter to ‘common sense’, marketing law, which is that competing brands rarely differ in their demographic or psychographic customer profiles (Hammond, Ehrenberg & Goodhardt, 1996; Uncles et al., 2012).
Does Duplication of Purchase (DoP) apply to streamed music listening?
To examine whether DoP (see Corporate Reports #51 and Report #53) holds in an online music context, we analysed music listening data from an online music aggregator and personalised radio station, last.fm. This captures music played by individuals across iTunes, Spotify, YouTube, and a large number of other sources. The data is from 30,000 users and 84 million listening occasions in four datasets, over one year (2013-14). We analysed the largest 99 brands by penetration and an aggregated “all other” brand at genre, artist, album and song levels.
Finding 1: People listen across a spectrum of music genres, artists, albums, and songs
In line with the expected DoP patterns, we find that music brands have overlapping listener bases. The degree of overlap largely follows the popularity of the music brands. That is, for any music brand’s listeners, many also listen to the very popular brands, but few also listen to less popular brands.
Table 1 shows Rock (the most popular genre) shares 58% of its listeners with Electronic, 61% with Metal, right through to 24% with House (the least popular genre in the top 99). In return, all genres share on average 89% of their listeners with Rock, but only 29% with House. There are a number of genres that share more or fewer listeners than expected. For example, Metal and Death Metal, Soul and Jazz, and Indie and Indie-Rock share quite a lot more listeners with each other than other genres. Conversely, Death Metal shares far fewer listeners with Electronic, Pop and Indie. Many of these individual deviations from the expected DoP pattern are unsurprising. Importantly, listener sharing still generally declines from left to right as the popularity of the genre decreases, which is consistent with the law of DoP.
Table 1: Sample results for genre listener sharing (1-month)
Whether we look at competing artists, albums or songs, we see the law of DoP. A sample of results for competing artists is shown in Table 2.
Despite an eclectic mix of artists, we see the overriding DoP pattern, where artists share more of their listeners with more popular artists than they do with less popular artists. In this data, all artists share on average 44% of their listeners with the most popular artist, Johann Sebastian Bach, but only 23% with the least popular artist, The Cranberries. However, given that this is only a small subset of competing artists, we see some variability from the DoP pattern, with a number of artists over-and-under sharing listeners. For instance, Bach, Beethoven and Mozart share more listeners with each other than we would expect for their penetration. Similarly, Lana Del Ray, Florence and the Machine, Lorde and The XX overshare listeners, while The Beatles, Queen and David Bowie share more listeners with Led Zeppelin and vice versa. It could be that artists operating within the same genre or era, share more listeners with each other. It could also be that this oversharing has been accentuated by music recommendation — streaming services may have successfully recommended music by similar artists to these listeners.
We also see cases of listener under sharing, predominantly between classical and non-classical based artists (e.g., Beethoven and Radiohead, Mozart and Lana Del Ray). However, it is important to note that the magnitude of under sharing deviations is far smaller than it is for over sharing deviations. This indicates that there are groupings of artists, where listeners may be more likely to listen to other artists within the same grouping. However their listening is not isolated or partitioned, and penetration is still the primary driver of multi-artist listening. These types of groupings are commonly seen in any normal market – for example, within the soft drink market, we commonly see diet soft drinks share more customers with other diet soft drinks, but we do not see clear isolation from non-diet soft drink brands.
We observed similar patterns of sharing for competing albums and songs, where DoP is evident, but not as neat as listener sharing between genres. Our overall findings reveal that while DoP describes music listening across genres, artists, albums and songs, the optimal level of fit is for genres. This is similar to SKUs, where the duplication pattern is somewhat messier than the more aggregate brand level.
Table 2: Sample of results for artist listener sharing (1-year)
Do music genres appeal to different types of people?
For this analysis, we conducted a nationally representative online survey in the United States (n = 1,036) and then compared the audiences of each genre with the average audience of all genres. For example, if Rock’s audience is 70% male, and 52% of males listen to any genre on average, there is an +18 percentage point (pp) deviation from the ‘expected’ proportion of male listeners. We compare audiences across demographic, psychographic and music media usage variables. See Corporate Report #7 for more detail about analysing brand user-profiles.
Finding 2: music genres, artists, albums, and songs listeners seldom differ
Table 3 shows that most genres have listeners that match the average listener profile, although some genres have more male or female listeners than expected. For example, Pop skews more to females while Metal skews more to males. So, while there is some evidence to support the idea that some genres skew to a particular gender, most genres have similar profiles — with gender profiles differing by 7pp on average. This is slightly higher than industries such as Sweets, Computers, Coffee and Cars, which differ on average by 2pp (see Report #7).
Table 3: Genre listener profiles by gender
We repeated the analysis across numerous demographic, psychographic, and usage variables. Table 4 outlines that audiences are more similar than different on each of these variables – even more so than the Gender example.
Table 4: Summary MADs of Genre listener profiles for demographic, psychographic and music media usage variables
We find the largest genre listening differences for age, though slightly lower than gender on average. In particular, older listeners aged 55+ have quite different tastes to the younger population. This older age group (see Table 5) are under-represented listeners of Hip-Hop/Rap, Metal, Indie/Alternative and Electronica/Dance. On the other hand, they are over-represented listeners of Country, Classical, Jazz, Blues and Folk.
Table 5: Genre listener profiles by age, showing the larger deviations among listeners aged 55 or older.
The effect of nostalgia on music listening
One potential explanation for the larger listening differences for age is the effect of nostalgia on music listening. If nostalgia does affect listening, this older age group’s over-or-under-representation in a genres’ listening base could simply be a reflection of the popularity of that genre during an earlier period of their lives.
The same respondents analysed for genre listener profiles, also listened to 34 songs, one from every second year between 1950-2016. These songs were selected from the Billboard top 10, excluding those in the top 3 (to avoid extremely popular songs). Each respondent was exposed to these songs in a randomised order and asked to rate them out of 10 (1 = “I dislike it a lot”, 10 = “I like it a lot”). To take into account the different ways people use scales, we standardised the scores for each respondent, which squashes and stretches the ratings given by a respondent to fit neatly between -1 and 1, with their average rating becoming 0.
Finding 2: Smells like teen spirit…we most prefer the music from our teens
We find that people like popular music the most from the time when they are in their teenage years – the preference peak for when music was released is when respondents were between 15 and 20 years old (see Figure 1). Preferences are lower for music that was released before or after the mid teen years. Music from before listeners were born rates below the average (zero on the y-axis), as does music released after about 33 years, dropping rapidly with age. This suggests that an age-related nostalgia effect does occur. An alternative explanation is that teens are more heavily targeted by record labels, and so people are more likely to recognise (and therefore like) music from the time they were teens. Interestingly, music made in the later years is liked less than music made before birth, suggesting an exposure effect – in our teens we are exposed to music of the day and preceding years.
Figure 1: How much we like a song (Y-axis) versus our age when the song was popular (X-axis)
What do these findings mean?
Firstly, people do not limit themselves to a select group of music, but instead listen across the whole spectrum of music. The degree to which people listen across multiple music offerings is predictable and driven by the popularity of the genre, artist, album or song. So, people do listen to both Rock and Classical, and they do so in line with each genre’s popularity. This is the law of Duplication of Purchase (Report #51 and Report #53). Secondly, genres are more similar than different in their demographic and psychographic listener profiles (Report #7). For example, Classical listeners (on radio) are no wealthier than their head-banging counterparts. This means that listeners of Metal and listeners of Classical are actually quite similar types of people. Most music listening is best described as eclectic.
The differences found in the listener profiles of various genres do differ more than for other categories, but not enough to say that they have wildly different listener bases. So while music might differ by more than other categories, it is a relatively small difference in reality. It is perhaps the relative size of the differences (in comparison with other categories) that leads to the impression that music is deeply skewed to particular demographics.
One implication for advertisers who use music in their advertisements is that they do not need to worry much about whether a genre of music will alienate part of the target audience. Likewise, research is not needed to choose a genre that has suitable appeal to the advertised brand’s buyers – popular music is popular with a very broad audience.
On a more strategic level, our results add to the body of evidence that there is a natural limit to our ability to target small groups of consumers based on their preferences. More data, faster computers, and more complex statistics can not make up for the fact that consumers have broad, eclectic and changing preferences. This explains why programmatic targeting efforts on an individual level by music streaming services (e.g. Spotify’s Discover Weekly Playlists) are not very accurate in predicting what people will listen to. If someone listens to Bach it doesn’t mean they won’t listen to Green Day; if someone listens to Charli XCX it doesn’t mean they will near exclusively listen to pop.
