Faster, Louder, More Explicit: How Music Has Evolved Over the Years (and How My Own Musical Taste Compares)

My partner and I discuss music frequently. Some of these conversations are about our shared interest in playing guitar, going to live concerts, and unashamedly blasting Justin Timberlake on road trips. We played “Man of the Woods” a good five or six times on our Yosemite vacation last year because, you know, woods. Others of our conversations are out of disbelief. I can’t relate to his love of EDM and trap. He is still in awe (not in a good way) of my fondness toward folk, Americana, and yes, country music.

This got me thinking. How can data help us bring more nuance into our music conversations? How, more precisely, is my music taste different from his — and how do both of our tastes compare with what’s “popular”? Speaking of popular music: what attributes set popular songs apart from songs that can’t get no love? And how has music evolved over the years, anyway?

As a fantastic starting point, I discovered that Spotify has helpfully assigned a variety of musical attributes to every track in its database. Here are some of those attributes. Definitions below provided by Spotify, abridged by me.

  • Popularity: The popularity of a track is a value between 0 and 100, with 100 being the most popular. The popularity is calculated by algorithm and is based, in the most part, on the total number of plays the track has had and how recent those plays are.
  • Danceability: Danceability describes how suitable a track is for dancing based on a combination of musical elements including tempo, rhythm stability, beat strength, and overall regularity. A value of 0.0 is least danceable and 1.0 is most danceable.
  • Acousticness: A confidence measure from 0.0 to 1.0 of whether the track is acoustic. 1.0 represents high confidence the track is acoustic.
  • Energy: Energy is a measure from 0.0 to 1.0 and represents a perceptual measure of intensity and activity. Typically, energetic tracks feel fast, loud, and noisy.
  • Tempo: The overall estimated tempo of a track in beats per minute (BPM). In musical terminology, tempo is the speed or pace of a given piece and derives directly from the average beat duration.
  • Loudness: The overall loudness of a track in decibels (dB). Loudness values are averaged across the entire track and are useful for comparing relative loudness of tracks. Values typical range between -60 and 0 db.
  • Instrumentalness: Predicts whether a track contains no vocals. The closer the instrumentalness value is to 1.0, the greater likelihood the track contains no vocal content.
  • Speechiness: Speechiness detects the presence of spoken words in a track. Values above 0.66 describe tracks that are probably made entirely of spoken words. Values between 0.33 and 0.66 describe tracks that may contain both music and speech, either in sections or layered, including such cases as rap music. Values below 0.33 most likely represent music and other non-speech-like tracks.
  • Valence: A measure from 0.0 to 1.0 describing the musical positiveness conveyed by a track. Tracks with high valence sound more positive (e.g. happy, cheerful, euphoric), while tracks with low valence sound more negative (e.g. sad, depressed, angry).
  • Explicit: Whether or not the track has explicit lyrics.
  • Key: The estimated overall key of the section. The values in this field ranging from 0 to 11 mapping to pitches using standard Pitch Class notation (E.g. 0 = C, 1 = C♯/D♭, 2 = D, and so on).

I guess when you’re Spotify, you can make up all kinds of words. I’m going to start saying “speechiness” all the time. As in, “this guy has too much speechiness.”

Through Kaggle, I was able to get a hold of a massive dataset from Spotify. With songs from 1921 to 2020, a full century’s worth. Almost 170,000 rows, each featuring a distinct track and its core info and musical attributes.

For example let’s take the very deep 2013 hit song “TKO” from Justin Timberlake. This row looks something like: Artist = ‘Justin Timberlake’ | Track = ‘TKO’ | Duration = 424147 (in milliseconds — BTW that’s over 7 minutes, for goodness’s sake JT!) | Popularity = 57 | Danceability = 0.611 | Energy = 0.678 | Explicit = true | Instrumentalness = 0.000 | Key = 3 (matching to D#) | Loudness = -6.723 | Speechiness = 0.243 | Tempo = 137.864 | Valence = 0.493 | Release_year = 2013.

And of course, Philosophical_deepness = 0.999.

In other words, the song is moderately danceable, contains explicit content, is very vocal-dominant (vs instrumental-dominant), sits right around the midpoint between feeling positive and negative, and is apparently not particularly popular. Sorry JT. Go cry us a river.

Armed with this dataset, my SQL knowledge, a Postgres tool called pgAdmin, and a tool called Metabase that I used to help with visualizations, I went about playing with music.

By the way, I embedded many songs via Spotify into this blog post, so for the full 20/20 Experience (Deluxe Version and 2 Of 2), you may want to turn your audio up.

(Oh in case you’re curious: JT’s most popular songs are “SexyBack” and “Mirrors,” which are tied for first place. Least popular song: “Nothin’ Else.” Yea, I haven’t heard of it either.)

Music has gotten more in-your-face (in-your-ears?) over time

Over the past century, music has become a lot less acoustic, higher energy, and somewhat more danceable.

I would imagine the sharp plummet in average acousticness is likely related to the rise of rock ‘n roll music in the late 1940s and early 1950s. For reference, Bob Dylan’s “Blowin’ in the Wind” has an acousticness of 0.914, versus 0.090 for “Sweet Child O’ Mine” by the Guns N’ Roses.

Music also seemed a lot more variant in the 1920s–1940s (the wider spikes up and down) than from the 1950s to now, when there appears to be less fluctuation year over year in average acousticness, energy, or danceability — or many of the other attributes I’ll go into below.

Music has also become increasingly louder and faster.

This inconsistency in loudness is probably why we feel like we have to constantly adjust our volume when listening to different types of music on Spotify.

The loudest song in the dataset I looked at is the rock ‘n roll band The Stooges’s “Raw Power — Iggy Pop Mix.” Loudness score: 1.3 (remember that based on the way Spotify identifies loudness, most songs are between 0 and -60, so 1.3 is very loud). Give it a listen. Then, contrast with Brian Eno’s “Signals — Remastered 2005,” which has a loudness score of -41.8. I know which one I’d pick if I were looking for peace and calm.

Songs have also become a lot less instrumental (AKA more vocal-heavy) over the past decades. And since around 1960, “speechiness,” or presence of spoken words, has been on a slow, gradual incline.

Since the 2000s, the average instrumentalness of songs has been just 0.073, a 79% drop from the 1920s–1940s when the average instrumentalness was 0.348. Apparently vocals are winning the war against… musical instruments?

Music has also become significantly less kid-friendly since around the 1950s. Fully 49% of tracks in my dataset from this year, 2020, were marked as containing explicit lyrics, versus literally 0% for much of the 50s and 60s. Jay-Z, Eminem, Missy Elliot, and Lil Wayne are just a few examples of artists with an extremely high percentage of explicit songs. JT comes in at about 22% explicit, so they should really be playing my road trip playlists at elementary school holiday parties (do elementary schools have holiday parties?). JT is basically Barney the Purple Dinosaur by 2020 standards.

Song length, mood, and key have stayed fairly consistent

The length (duration) and “mood” (valence) of music have actually stayed relatively flat. It seems we had a wave of “let’s see how far we can push it” in the 1970s–2000s. Case in point is Prince’s “Purple Rain”: amazing song, probably didn’t need to be almost nine minutes long, just sayin’. Thankfully, average song duration has dropped back down to about three and a half minutes in 2020.

As for how happy or sad a song sounds, or what Spotify calls “valence,” we’ve stayed shockingly balanced on average over the past century, with a dip around 2017 but picking up since then. I guess humans are pretty resilient!

Finally, musicians’ choice of musical key has also stayed highly consistent. C continues to be the most popular key, followed closely by G, D, A, and F. The D key has very gradually gained popularity over time, while D# has slightly fallen out of favor.

Honestly, most “pop” songs use the same chords in general. It’s why, as I once demonstrated while busking at a subway station, you can play all of the following songs and many, many others with literally the exact same chord progression of C-G-Am-F (or G-D-Em-C, as one way to transpose it):

  • Don’t Stop Believing
  • Yesterday
  • Let It Go
  • Someone Like You
  • Call Me Maybe
  • Wagon Wheel
  • With or Without You
  • Auld Lang Syne

For those of you who play music, give it a try. So many mashup opportunities abound here.

Popular music: faster, louder, higher energy, electronic-heavy, vocal-heavy

According to Spotify, “popularity is calculated by algorithm and is based, in the most part, on the total number of plays the track has had and how recent those plays are.”

What makes popular music popular? Not surprisingly, it matches the over-time trends we saw above.

We love songs that are fast, loud, high-energy, and (relatedly) danceable. We hate songs that are acoustic and instrumental-heavy. Speechiness is apparently divisive: the most popular and least popular groups of songs are both the speechiest (now that’s a word!) on average. Valence doesn’t seem to affect our choices very much.

Here’s how Billie Eilish’s “Bad Guy,” the most popular song in Spotify’s list of top-streamed songs from the 2010s, compares. Popularity: 91 | Danceability: 0.701 | Tempo: 135.128 | Loudness: -10.965 | Energy: 0.425 | Acousticness: 0.328 | Instrumentalness: 0.130 | Speechiness: 0.375 | Valence: 0.562. So pretty danceable, super fast, very vocal-heavy.

And here is The Weeknd’s “Blinding Lights,” the absolute most popular song in my entire Spotify dataset, with a popularity score of 100! Danceability: 0.51 | Tempo: 171.01 | Loudness: -5.93 | Energy: 0.73 | Acousticness: 0.001 | Instrumentalness: 0.00 | Speechiness: 0.06 | Valence: 0.33. Not as danceable, but extremely fast, very high energy, super non-acoustic / electronic-heavy, super non-instrumental / vocal-heavy.

Ok so now I know how to maximize my chances of making it onto the Billboard Hot 100. All I need to do is come up a song that’s ultra fast, ultra loud, ultra energetic, ultra vocal-heavy, ultra electronic-heavy, that has either a lot of spoken words or very few spoken words. On it.

My own music taste begs to differ

Unfortunately for my rich-and-famous dreams, what makes music popular to the world isn’t perfectly aligned with what makes music “popular” to me. Which brings me back to what I started this whole blog post with: how does my music taste compare to my partner’s, and the Spotify world’s?

To explore this, I looked at (1) Spotify’s “Most Streamed Songs of the Decade” (2010s), (2) My list of “Liked Songs,” and (3) My partner’s list of starred songs. Here’s how we shake out. For this part of the analysis, I normalized the Popularity and Tempo scores so they would reflect the 0 to 1 scale as well. (Also, getting to play with radar charts was quite the treat!)

Basically, I am less into super popular songs, but my partner is even less into it. In case you’re curious: the most “popular” song on my list of Liked Songs is “Shallow” from A Star Is Born (popularity score = 86). Not bad. For my partner, it’s System of a Down’s “Chop Suey!” (popularity score = 79). A bit more of an acquired taste, I’d say.

My songs of choice are also less danceable. My partner’s songs are even less so. The average danceability for his songs is just 0.55, versus 0.62 for me and 0.71 for Spotify’s Most Streamed Songs of the 2010s. For anyone planning to invite us to weddings: you have been warned.

My partner does like higher-energy and more instrumental-heavy songs than either me or the Spotify world— probably a reflection of his love of EDM. As for me, it’s not exactly a surprise that my musical taste leans more acoustic. It’s all the folk and Americana talking. What can I say? I love a good unplugged music session.

The average tempo of my and my partner’s favorite songs are fairly comparable to Spotify’s top songs from the 2010s, although apparently we like slightly sadder songs compared to the generic Spotify user. The saddest-sounding song on my list is Odesza’s “A Moment Apart” (one of the only EDM-ish songs that my partner got me hooked on) which has a valence score of 0.076, followed by JT and Anna Kendrick’s “True Colors” (valence = 0.100). Beautiful songs, but not exactly mood lifters.

My and my partner’s musical tastes are also generally more diverse. This chart plots the standard deviation of the three lists I mentioned earlier (Spotify’s Most Streamed from 2010s, my liked songs, my partner’s liked songs). In other words, it looks at how spread out the songs are from one another in a given list, for each of the musical attributes.

What jumps out immediately is that my partner’s song preferences are most varied, especially when it comes to how instrumental-heavy vs vocal-heavy the songs are (again probably a reflection of his EDM fandom, mixed in with a healthy dose of other genres in his playlists). My own taste is not as diverse as his, but still more so than the average Spotify user on most attributes.

To wrap it up

As an amateur guitar player, music holds a very special place in my heart. It was a blast to get to dig into this Spotify dataset.

What I am walking away from all this with:

  • In line with people’s preferences, music has become faster, louder, higher-energy, more danceable, more electronic-heavy, more vocal-heavy, and more explicit over especially the past half-century.
  • If I want to get my song played on the radio station, I’ll need to make some changes to my musical style. In particular, I need to get more danceable and more electronic-heavy. My partner basically has no shot. It’s true. He’s too far off.
  • SQL continues to amaze. It’s such a powerful yet relatively simple tool to analyze huge amounts of data. No way I could have done this analysis in one weekend otherwise. No way I could have done this analysis probably at all: Excel / Google Sheets would very likely have been unable to handle the sheer volume of this dataset.
  • I would really like to build on my (rudimentary) Javascript skills. I am grateful to Kaggle for supplying the dataset I used for this blog post, but even that massive dataset is still just a portion of all the data Spotify itself has. Being able to tap into APIs directly will allow analyses like this one to be more robust and comprehensive.

Finally, I would like to send you all off with the 25 most danceable English-language popular songs according to Spotify, to make your wedding or whatever party extra dance-ready. I only included songs that have a popularity score of at least 80, so that your guests are more likely to know them. I also filtered out songs containing explicit lyrics because, as I said, I want them to play my playlists at elementary school holiday parties. “Not explicit” doesn’t necessarily mean “kid-friendly,” though, so use your best judgement next time you play this at your neighborhood kids’ birthday party.

And with that, I bid you bye bye bye. Thank you for reading this blog post, which I know has been pretty high on speechiness. I hope you all have fun dancing in the moonlight to these tracks.

Got any feedback on this blog post? Call me maybe. Well, leave me a note please. I’d love to hear from you!

People & ops leader | data storyteller & nerd