Generally speaking, playlist selection was done as described in the first post about representation of composers in classical streaming playlists:
For this analysis, I examined top classical playlists — four from Apple Music, and four from Spotify — that you might find by searching “classical” or “orchestral” in each service and picking from the top results.
In a handful of cases, there were some additional considerations, as noted below:
- Classical Essentials - top result for “classical” (downloaded 2020-12-30)
- Classical Sleep - high ranked result for “classical” (downloaded 2021-01-03)
- Classical New Releases - not a top-ranked result, but chosen to compare with Apple Music “A List” playlist (downloaded 2021-01-03)
- Orchestra 100: Spotify Picks - top result for “orchestra” (downloaded 2020-12-30)
- Classical Essentials - high ranked result for “classical” (downloaded 2020-12-28)
- The A-List: Classical - top result for “classical” (downloaded 2020-12-28)
- Relaxing Classical - high ranked result for “classical” (downloaded 2021-01-16)
- Classical Kids - not a top-ranked result, but chosen because of notable target audience (downloaded 2020-12-28)
- Some of these playlists are updated fairly frequently, so the songs on them now may be different than when this analysis was done 1This sort of begs the question, “why not do an analysis over time?” I mean maybe… the problem is, that especially for playlists that completely change every week, this would be even more time consuming than this project. However, I’ve kept the door open to the later possibility by continuing to download the data for these playlists every Sunday. So who knows… maybe over holiday break 2021 I’ll get bored again..
- Apple Music did not have an official “Orchestra” type playlist
- Spotify did not have an official “Classical Kids” type playlist
- In both cases, I used a brand-new account, which should have helped to minimize the influence of personalization on the search results
Data Sources and Preparation
For Spotify, I exported the playlist information using Exportify, a tool created by Howard Wilson. (Github Repository for Exportify) The options “Include artists data”, “Include audio features data”, and “Include album data” were all selected.
For Apple Music, I exported the playlist information from iTunes, as described in this help page: Save a copy of a playlist in Music on Mac 2I’m running Mojave, so Apple Music still runs via iTunes on my computer..
Generally speaking, I determined gender and race information for composers by regular web searching, in particular paying attention to:
- Pronouns for gender - ideally on their own bios or Wikipedia pages, or in media coverage;
- Self-descriptions (bios, interviews) for race, or 3rd party descriptions of cultural heritage when self-descriptions were not available.
The Apple Music metadata that you can download includes a composer field, so identifying the right composer(s) was mostly straightforward for this platform. Although Spotify has a “songwriter” field, it is not yet available in the API, so this is not a field that can be downloaded using Exportify.
To save time, I cross-imported several of the Spotify playlists into Apple Music and downloaded the equivalent Apple Music metadata to fill out the composer fields. I used Soundiiz to achieve this. (I normally wouldn’t use a platform like Soundiiz because of privacy concerns, but since both accounts were created specifically for the purpose of this analysis, I wasn’t bothered in this case.) This correctly filled the composer fields for Spotify playlists about 90-95% of the time.
In both cases, if the composer name(s) was/were missing or unclear, I conducted additional research to identify the composer(s) as best as possible.
Information about the “Average US Orchestra” comparison based on Institute for Composer Diversity data can be found on this page: Methodology: Using Data from the Institute for Composer Diversity 2019/20 Season Analysis.
I mostly followed the Canadian “Visible Minority” framework for noting race of composers in the data, although I used a broader interpretation that is more inclusive of mixed-race composers than under the official Stats Canada methodology, as explained on this page.
Additional Special Cases
Ensembles and Multiple Composers
Most of the tracks I analyzed had a singular composer listed. However, some had an ensemble or multiple people listed under the composer field. This raises an interesting question of “what question are you looking to answer?” in the data analysis, somewhat similar to how asking different questions about orchestra seasons would lead you to measure different things.
For the purpose of this series, I chose to handle multiple composers on a fractional basis - a track composed by a male artist and a female artist, for example, would count as 0.5 female artists. Since the vast majority of tracks have a singular composer, this is probably good enough?
However, if you were going to analyze playlists where it’s common to have multiple composers (or songwriters), you would probably want to use a couple different metrics to demonstrate both the frequency and prominence of artists from certain demographic groups. I think Liz Pelly’s Discover Weakly article gives a good example of how to do this:
On Today’s Top Hits, I found that over the course of one month, 64.5 percent of the tracks were by men as the lead artist, with 20 percent by women and 15.5 percent relying on collaborations between men and women artists. When all features were taken into consideration, I found that 85.5 percent of tracks included men artists, while only 45.5 percent included women. This was one of the highest percentages of women artists out of all the playlists I examined.
This begs an interesting question about classical music - why is so much classical music written by singular people when other forms of music are often written collaboratively? Maybe I’ll write a post about this at some point.
Some pieces of music don’t have known composers, e.g. folk songs and nursery rhymes. In these cases, the composer usually gets written in as “Traditional”. I chose to include these for the purposes of the percentage calculations for each demographic category. All of the folk songs and nursery rhymes that I came across were from European countries, but in theory if there were a Japanese folk song, for example, I would have counted that as track written by a composer of colour.
When it comes to gender, this gets a bit tricky. Because I only counted demographic categories I could reasonably verify, this means that folk songs default to male in the analysis. Which doesn’t quite seem right? Perhaps you could include folk songs in the percentage calculation for race, but not for gender, but this complicates the figures in a way that could end up being confusing. I suppose you could argue that this does make sense because representation depends on recognition.
I don’t really know what the right answer is here. But this is how I calculated it, and you can decide (or let me know!) if there’s a better way it should have been done.
Composers with Unknown Demographic Attributes
One of the quirks of being able to have an entirely-online career as a musician is that you can create whatever identity you like. Especially on the mood-themed playlists, it was pretty common to find artists with no identifiable link to a specific, real person. You’ll see these as “No Sources Found” in the data. I included them in the percentage calculations — I’ll note again that this leads to a similar problem to folk songs and nursery rhymes where the data effectively defaults to counting these composers as white and male.
There’s a lot more to this subject of effectively-anonymous artists, and I’ll almost definitely write a post about it at some point.
Arrangers vs Composers
For this analysis, I didn’t count any arrangers for the summary statistics produced.
The fun question to ponder here is this: when is an arranger’s influence on the music so strong that they should be considered one of the “authors” of the piece? You could ask the same question about producers, or even people who write lyrics or libretto. Because my girlfriend studies these kinds of things, I can tell you there’s a whole part of translation theory that looks at questions like this.
Composers aren’t the only artists involved in making music - representation of musicians, producers, etc. also matters. There are musicians who are women, Black, Indigenous, or people of colour that are not noted in this analysis because of its singular focus on composers (e.g. Sheku Kanneh-Mason performing traditional English folk song Blow the Wind Southerly on Spotify’s Classical Sleep playlist.)
However, I think composers are a good role to focus in on for a couple reasons:
- In classical or orchestral music, composers often have the highest stature, compared to any other role 3To prove my point, I quickly counted the number of times that composers, soloists, and conductors were mentioned in concert titles for the TSO 2019/20 Masterworks season. Composers were mentioned 28 times, conductors 5 times, and soloists 4 times..
- Relevant benchmarks for composer diversity are easiest to find.
- If there is clear underrepresentation of composers from certain demographic groups, then it is highly likely that there is underrepresentation in other artist roles as well.