What Are Successful Spotify Songs Made Of?
Allison Chen (Project Lead), Cynthia Du, Larsen Bier, Kevin Espinas, Sia Phulambrikar, Deshna Govil
Introduction
Strolling down Bruinwalk, every student has their headphones in –listening to either carefully curated playlists, the most recent hits, or their friends’ recommendations. Have you ever wondered what songs might be most popular? This article investigates the following features of the top songs of 2023 on Spotify:
- Release Date: How did release dates affect how songs performed in 2023?
- Major, Minor, and Key: How did streams vary by song key?
- Danceability and Energy: Do people like songs that they can dance to?
- Artists: Who did people listen to the most?
- Speechiness: How does “wordiness” affect streams?
- Cross-Platform Analysis: How did the top songs perform on Spotify versus Apple Music?
When were the top songs of 2023 released?
To begin, we looked at the relationship between a song’s release dates and the number of streams. Did the time of year a song is released impact how popular it became? Our hypothesis was that songs released in certain months would have a higher probability of being popular due to holidays and school schedules. For example, we hypothesized that songs released in December would have a harder time becoming popular due to the prevalence of Christmas and holiday music during that time. Furthermore, we predicted that songs released in the summer months might be more popular as students, who are a large part of the Spotify demographic, are no longer in school and might have more time to stream music. Thus, we dove into the average number of streams per month and the number of songs that were released in that month that made it into Spotify charts.
Based on these graphs, it is clear that the average number of streams per song and the presence in Spotify charts do not align with the graphs. Our hypothesis was correct with respect to songs released in the summer months becoming more popular, as we see the number of songs present in Spotify charts spike in May and June. However, this doesn’t align with the average number of streams by the month released. One would expect that the two graphs match, so we can conclude that there is not a significant relationship between the month the song was released and how popular it is. Furthermore, there are confounding variables of which artists are releasing during the month. For example, Taylor Swift had an album released in July, and Metro Boomin had an album released in June, which could heavily influence the data.
Besides looking at the month, we also dove into the day or year a song was released. As expected, there is no relationship between the day a song was released and how popular it was. Songs released in 2022 were streamed the most in 2023, which makes sense as songs released in 2022 are likely to be popular in 2023, and users have the whole year to stream these songs. Thus, time is not a good predictor of song popularity and we should explore other variables, specifically the songs’ components.
What’s your major?
To get a better understanding of what the top trending music of 2023 sounded like, let’s explore the key and mode of the songs.
In the above graph, we plotted the total streams for each key by mode. There seems to be more variability in the number of streams within the major key with a spike of 45 billion streams of songs in C# Major and a minimum of less than 10 billion streams of songs in D# Major. For songs in the minor key, there are most streams for those in B Minor and least for those in D Minor. The average total streams by mode show that more top songs of 2023 are in Major mode than Minor mode. The graph below directly graphs the average streams for each mode and key combination.
Similarly, we observe that the average streams across top hits in the Major keys are generally higher than those in the Minor keys. This can be explained by how we, as listeners, associate songs in the Major key with happier emotions and songs in the Minor key with more melancholic ones. This may imply that listeners enjoy listening to music that is considered to be uplifting and cheerful. To confirm our findings, we continue to look at the danceability and energy attributes of songs.
How danceable are the songs?
Spotify assigns to each of its songs a “danceability” score, which is a percentage rating of how good a song is for dancing. Of course, dance norms can change quickly, especially in the internet age, so one might reasonably wonder how the ratings assigned to songs depend on the time period. Consider the average danceability of songs on Spotify as a function of the year:
1 standard deviation depicted in shaded region OLS trendline depicted in black
There are far more songs on Spotify from modern decades, however, the spread of the danceability has changed relatively little, suggesting that while songs are becoming more danceable on average, there is still great diversity in the danceability of music today (which also holds true historically).
Despite this, the mean danceability of songs increased over the past sixty years, although not monotonically (data from earlier than 1960 were omitted due to the low density of songs from that era on Spotify). But, while the mean danceability jumped by around 15 points from 1960 to 2020, during the 80s and 90s, danceability actually decreased from its peak in the twentieth century. What could be the explanation for this?
The shift of genres coincides with the Distribution of Songs by Popularity and Danceability graph to tell an interesting story. The groovy beats and upbeat mood of many 70s disco songs make them extremely danceable, however, disco’s time in the spotlight ended at the beginning of the 80s, ushering in an era of slightly less danceable pop and (especially) rock during the 80s. Moreover, hip hop had not yet bloomed into the mega-popular genre of the 90s, possibly leaving the eighties with an overall less danceable mainstream. Despite this, dance-oriented genres like house and synthpop increased in popularity this decade partially due to improved digital music technology. This likely helped set up the 90s to begin “recovering” from the danceability decline of the eighties.
So the average song on Spotify has gotten more danceable over the past sixty years. But you probably have never heard the names of 90% of those songs (over 1 million were sampled to produce the figure above). What can we say about the music that people are actually listening to?
The top 1000 songs of 2023 have danceability centered around a mean of 67.0 (median 69.0) and the distribution is slightly skewed left with a standard deviation of 14.6. In general, the danceability of songs on Spotify is close to normally distributed with a mean of 49.3 (median 50.1) and a standard deviation of 18.7.
This distribution indicates that the top songs are, in general, more danceable than the average song on Spotify. Based on a simple linear univariate linear regression analysis, danceability appeared to have no significant correlation with the number of streams a song received among the top songs. This indicates that for a song to become tremendously successful on Spotify, it helps for it to be danceable enough, but beyond a certain threshold, how groovy a song is has little impact. The fact that there are no songs in the top one thousand with less than 20% danceability, while in general there are many with below 20% supports this hypothesis. Unsurprisingly, people like music they can dance to!
Along with danceability, we wanted to see how the energy of a song affects its popularity. Our hypothesis was that songs that were more upbeat and lively would result in more streams, as people would want to listen to music that energizes them and makes them feel good.
However, looking at the data, we see that this is not necessarily the case. There does not appear to be a positive relationship between the energy of a song and the number of streams it receives. This demonstrates the popularity of slower, calmer songs, along with energetic ones. This trend could be because different emotions and situations result in people streaming different types of music. There is a large audience for relaxing and moving types of music. Studies indicate that sad songs can be self-soothing, as they release positive hormones like endorphins, oxytocin, and prolactin, which help people feel better and support their well-being. As a result, people may often subconsciously stream slower, sadder songs to improve their moods, instead of higher energy songs.
While energy cannot be used as a good indicator of a song’s popularity in streams, we wanted to see if there was a correlation between a song’s presence in Spotify playlists and charts.
We observe that there does appear to be a connection between the energy level of a song and its presence in both Spotify charts and playlists. Songs with higher energy levels appear in more Spotify charts and playlists, particularly with a peak at 55% to 75% energy. This suggests that upbeat, higher-energy songs are often added to playlists, likely such as for workouts. The distribution also indicates that higher-energy songs often receive more buzz and end up on Spotify charts as a result.
We additionally wanted to compare energy levels per genre of music. We analyzed how the energy and popularity of a song compare for each genre of music.
Predictably, ska and reggaeton have the highest energy levels out of all genres of music, while opera and classical have the lowest. Taking a look at the relationship between energy and popularity per song genre, we see that they roughly follow the same trend, with some notable exceptions. Despite ska music’s high energy, its popularity level is relatively low at approximately 0.3, which is on par with classical music’s popularity. From this, we see that energy is not always indicative of how successful a genre of music will be. As energy cannot solely be used to predict a song’s streams, we continue to look into whether a song’s artist affects its popularity.
Artists
Next, we explored whether the artist has a role to play in the song’s popularity. To investigate whether collaborating with other artists increases song popularity or the number of streams, track streams were plotted against artist count. As seen in the strip plot below, songs that are solos or duets are the most popular. Songs with 3+ artists are less popular, and each additional artist working on a single track reduces the number of streams. This was a surprising find since one would’ve predicted that collaborating with popular feature artists would increase artists’ reach and hence song popularity.
The above plot shows the Top 10 artists with the most appearances in the 100 Most Popular Songs for the year 2023. The Weeknd and Ed Sheeran appeared the most number of times in the list, with 6 songs each making it to the Top 100. Eminem ranked next, with 4 songs, followed by Dua Lipa, Harry Styles, Justin Bieber, and Bruno Mars with 3 songs each. Last on this list were Adele, Arctic Monkeys, and Bad Bunny, with only 2 songs making it to the Most Popular list.
How “wordy” do people like their songs?
Spotify defines “speechiness” as the presence of spoken words in a track. Percentages closer to 100% are indicative of audiobooks or podcasts, while songs in the range of 33% to 66% contain both music and speech, encompassing genres like rap. Songs below 33% represent more instrumentally focused tracks with less emphasis on speech.
As a whole, “speechiness” in popular songs has increased over time. The greatest increase came around the turn of the millennium. To try to understand this sudden rise, we can examine what musical genres helped to build up this sudden boom.
The 80s began the advent of hip hop. One important aspect of this genre is rapping, which results in more lyrics present in a song. Hip hop’s popularity would continue to grow into the 90s, but along with it, alternative rock and pop music by boy and girl bands would also flourish. This combination allowed for this large increase in “speechiness”.
In recent years, the average “speechiness” has remained around 10%. The heavy presence of hip-hop music helps to explain this average.
Now let’s evaluate individual “speechiness” and how it affects performance, both in Spotify streams and presence in Spotify playlists.
In both charts, we observe a relatively weak negative correlation between “speechiness” and either streams or presence in Spotify playlists (with R-values of -0.112 and -0.089 respectively). For streams, the majority of songs seem to cluster around the bottom left of the chart. Here, songs are below 500 million streams and have a “speechiness” of less than 10%. The same holds for songs and their presence in playlists, as most songs are in less than 5 thousand playlists.
Our most streamed song is “Blinding Lights” by The Weeknd, with a “speechiness” of 7%. The highest “speechy” song belongs to KayBlack and MC Caverinha with their song “Cartão Black,” which has 71,573,339 streams. The one rather “speechy” song with a large number of streams is Justin Bieber’s “Love Yourself,” which has a “speechiness” of 44% but boasts a stream count of 2,123,309,722.
Likewise, the song in the most Spotify playlists, with 52,898 playlists, is “Get Lucky — Radio Edit” by Daft Punk. “Mr. Brightside” by The Killers follows closely, being in 51,979 playlists.
One specific threshold we can identify for both charts is a stream count of 500 million or a presence in 5,000 or more Spotify playlists. Songs with a “speechiness” of 35% or more seem to fall below this 500 million stream threshold, with only “Toxic” by BoyWithUke, “In Da Club” by 50 Cent, and “Love Yourself” by Justin Bieber breaking this trend. The same trend is prevalent with playlist presence. Only two songs with a “speechiness” of more than 35% are in more than 5000 playlists. These songs are again “In Da Club” and “Love Yourself”.
It seems that for a song to break the 1 billion stream threshold, its “speechiness” should be below 40% (with Justin Bieber being the exception to this trend).
Cross-Platform Analysis
Lastly, we analyze the performance of songs on different platforms.
The correlation matrix provides a cursory glance at the positive relationship between a song’s presence in playlists and streams. We hypothesize that as a song’s Spotify playlist number increases, its Spotify streams will also increase as more people listen to the song on the platform. On the other hand, we want to discover how the top songs on Spotify 2023 perform on Apple’s platform. Would songs with higher Spotify streams also have a higher “In Apple Playlist” value?
With a correlation value of 0.79 and 0.77 for Spotify and Apple playlists respectively, songs in more Spotify playlists are not much more positively correlated with higher Spotify stream values than songs in more Apple playlists. This suggests that songs that are popular on Spotify tend to also be popular on Apple Music. This is an interesting observation that motivates us to visualize the two correlations by plotting a scatter plot of the dataset.
The above scatter plot exhibits a song’s stream value against the number of playlists they are in on each platform. Though songs are generally added to more Spotify playlists than Apple playlists, both graphs showcase a positive correlation between the number of playlists and streams. The strength of the correlation is moderate with an R-value of 0.622 for Spotify playlists and an R-value of 0.604 for Apple playlists. Again, our findings from the correlation matrix are confirmed with a slightly higher R-value for songs in Spotify playlists.
Let us examine data points more closely to observe how songs vary by popularity on each platform. Point 1 is “Get Lucky — Radio Edit” by Pharrell Williams, Nile Rodgers, and Daft Punk. Though this song is in the most Spotify playlists — 52,898 playlists to be exact — it only garnered around 933 million streams compared to songs that also have a similar number of streams but only in less than 20,000 Spotify playlists. Moreover, the song is in 203 Apple playlists, which does not make it an outlier in the graph on Apple Playlist. Point 2 and point 6 are actually the same song: “Blinding Lights” by The Weekend. Being in 43,889 Spotify playlists and 672 Apple playlists, it amassed over 3.7 billion streams. Point 3 and point 7 are of Ed Sheeran’s “Shape of You.” Point 3 is close to the trendline, but point 7 is an outlier. “Shape of You” is in 32,181 Spotify playlists, but is only on 33 Apple playlists. This drastic difference is interesting and does not confirm our hypothesis. Point 5 is of another Ed Sheeran hit, “Perfect,” which is in only 7 Apple playlists, but in 16,596 Spotify playlists. While “Perfect” is slightly below Spotify’s trendline in the top graph, its position in the bottom graph insinuates again that high Spotify stream values do not guarantee good performance on other music-listening platforms. Lastly, point 4 is “One Kiss (with Dua Lipa)” by Calvin Harris and Dua Lipa. It is not far from Spotify’s trendline as it is 27,705 Spotify playlists, but is an outlier in the bottom scatter plot with it being in 537 Apple playlists.
It seems that while streams and the number of Apple playlists a song is in are positively correlated, it is not always the case that in-playlist value would be proportional across music-streaming platforms. There may be unidentified confounding variables that influence the relationship we analyzed above. For instance, perhaps some songs are streamed more individually rather than from playlists.
Conclusion
In conclusion, we have found the following attributes that correlate to successful Spotify hits:
- People tend to listen to happier songs!
- Good news for dancers: music has gotten more danceable through the years.
- Songs that are solos or duets are more successful than collaborations with 3+ artists.
- Songs that have higher energy levels are more frequently added to Spotify charts and playlists
- The most popular songs tend to be less “wordy”, but their effects on performance are not as drastic
- High-performing songs (500 million+ streams and in more than 5000 playlists) tend to stay below a “speechiness” of 35%.
Nevertheless, it is important to note that this is not the magic formula. For example, successful songs are better for dancing than average, but maximizing dancing compatibility doesn’t guarantee success. Having a top-performing hit on Spotify does not ensure success and high engagement with the same song on other platforms. With the variability of time, trends, and other confounding factors, it is difficult to obtain a grasp on the attributes of the “perfect” song. But, if you ever want to guess what your fellow Bruin is listening to on their way to class, perhaps you now have a place to start.
References
