Spotify Gender Bias (Aguiar Et. Al., 2018)

1. Introduction

inclusion in todays top hits playlist adds on average 20 mio. streams to a song
music listeners gender is around 50:50
female artists however less than 25% successful performers (as of 2018)
even though digitization has democratized access to music market, access may not be enough as not everyone has the same opportunity and means

two possible causes:

maybe simply just more men than women in the industry
women receive less support from record labels, radio stations and streaming services -> we need to find the cause so we can divise a solution to the problem, here we focus on whether there is bias within Spotify

framework of how music is being created an published. the process:
- decision to even make music may depend on past discrimination -> according to bureau of labor statistics data from 2018 38% of musicians are female
- decisions by intermediaries such as record labels on which music to promote and finance, but also radio stations and program directors for what is being played -> in this study we look at spotifys playlst makers as the intermediary
existing research
- platform bias:
  - Edelman (2011) Google favors own properties in search results
  - Zhu and Liu (2016) Amazon enters markets established by its vendors
  - Lambrecht and Tucker (2018) gender bias in ad targeting
- gender bias in music industry:
  - smith et al (2018) only 22.4% female artists and 12.3% female songwrites made it on teh “600 popular songs on the bilboard hot 100 year end charts” from 2012-2017
  - pelly (2018) most top playlists are mainly male, e.g. todays top hits: 64.5% male and 20% female

based on four underlying data sets:

full list of all songs added to spotify from mid-2015 and early 2018 with algorithmically matched gender from the firstname of the first artist on the song (assigned gender to 30% of data)
second data set is combinaton of various data
1. all songs observed streaming in top 200 on any day in any of the 26 countries during 2017, due to overlap between country this are 1695 artists
2. list of all songs in major playlists from may 1st 2017 -> additional 963 artists
  - these are the most followed editorial global playlists durign 2017 -> genders for these 2758 was assigned manually through artist page, photographs and press, so its the preceived gender, which is the relevant one probably for its also the one reporters perceive
3. recorded artists 2016 total streams
4. genre of songs according to database
5. various characteristics, including bpm, key (major/minor) and seven additional measures on scale to 100 (danceablity, valence, energy, acousticness, instrumentalness, liveness and speechiness) -> removing all artists with unknown gender leaves 6650 songs (97.4% of total streams on spotify)
list of all top 20 songs in new music friday playlists through 2017, songs gender labeled in 4 binaries

in dataset 1 we have about 21.5% songs from artists with names that are characteristicaly female (table 2)
- since around 38% of musicians are female, this suggests that availability of female music plays a role in low share overall
13.0% of all major streaming songs (dataset 2) are all female with 22% being labeled as either female (table 3). the respective shares in actual streams are 12.4% and 31.8%
global editorial playlists (table 4) various low resulst which hint at different gender participation in different genres
in new music friday (figure 1) we can see teh shares are roughly equal in all 26 countries
in raw population data we have the 38% of musicians are female, however even though such a comparison may be useful, comapring that black amaricans make 13.4% of population but 31.5% of music illustrates, that this inference is not as straight forward we we intuitivley think