Discussion
and
Evaluation
Discussion
This research project allowed us to clearly see how popular songs have evolved throughout the years. Some of our findings yield some unsurprising information whilst others contradict our hypothesis.
Length
As shown in our analysis, the length of popular tracks has been decreasing. This is likely due to the rise of streaming services which has allowed artists autonomy over their song length contrasting with when they used to be forced to follow the typical length of the radio slot. This, coupled with shorter attention spans in societies has led to shorter songs (Bemrose, B. 2019). There is an interesting question on whether by appeasing shorter attention spans, this encourages them to shorten songs further. The continuous decrease in song length would seem to initially support this hypothesis but more research is likely needed.
​
Danceability, Energy & Tempo:
The increase in danceability in popular songs over the years seems logical, especially considering the rise of dance-based songs. The rise of TikTok and other social media sites support the production of songs which are very 'danceable'. Additionally, the rise in clubbing culture which predominantly features songs with high danceability has contributed to the continued importance of this audio feature. Therefore, it comes as no surprise that danceability is on the rise and should be expected to remain a prominent audio feature.
​
Songs have remained high in energy in the last 50 years, but this feature has recently been decreasing which may be a testament to evolving tastes. It is clear that different decades have undergone different evolutions of music tastes and although current music is still energetic, it may not be as energetic as previous eras. This may even be the after-effects of changing tastes a few decades ago. For example, highly energetic songs, such as punk rock, started to phase out (University of Minnesota Libraries, 2016) and songs are still adapting to this change.
​
The tempo of songs has followed many different trends likely reflecting the diverse tastes in music in the population. The tempo levels also have strong links to emotional intensity and valence. Higher tempo tends to be linked to emotional valence rather than emotional intensity, whereas medium tempo songs are the opposite. Therefore, music can clearly shift to cater to society’s needs (Fernández-Sotos, et al., 2016).
​
Liveness and Loudness:
The Liveness and Loudness trends (which have been increasing in the last 50 years) are unsurprising as recording technology has drastically improved. There is now a larger focus on the recording studio and the potential ability of listeners to raise a song's volumes, rather than waiting for a performance in front of a live audience. An additional explanation for the increase in loudness could be the production trick used by the music industry since the late 1980s in order to make each track more impactful than the following, in a sort of ‘sonic arms race’. The trick is called ‘dynamic range compression’, which boosts quieter parts of the song so that the music appears louder overall (Cox, 2016).
​
Speechiness & Instrumentalness
The low level of speechiness is also not a surprise especially considering it has remained constant throughout the decades. It seems that popular songs have low levels of spoken words in their track, which is natural because of the distinction between a song and a poet and the mainstream nature of songs in comparison. It should be noted there has been a slight increase over the years which could be attributed to a rise in rap music. Instrumentalness has also remained at low levels and seems to never have been an important audio feature.
​
Time Signature & Acousticness
Time signature has been a constant likely due to our innate preference for binary metric beats (Potter, Fenwick, Abecasis and Brochard, 2009). Acousticness has experienced some fluctuation, reflecting a need for less electronic-heavy productions as the 90s and 00s saw a surge in electronic-based songs. The decrease in confidence of a song being acoustic in the 1980s and in following decades can be explained by the emergence of technology in the opportunity to create electronically produced sound (Feinstein and Ramsay, n.d.).
​
We decided to only study the Top 20 songs on the US Billboard charts each year. This can be considered be too restrictive or still too broad. Choosing a different definition of what is a popular song might have completely changed our study as our data values and thus, the results of the analysis may have changed. However, as there is no set definition of popularity and after consideration, 20 songs seemed to be a fair estimation, which resulted in just under 1000 songs in our total sample.
​
Additionally, some songs throughout the years were not available on Spotify and were therefore unable to be included for our data analysis. However, only a little more than five songs were unavailable, limiting the impact these absences might have had on our research.
​
This project relied on audio features as defined by Spotify. Although Spotify gives a short description of each of their audio features, the algorithms used to develop them are not open for all to see. Therefore, we cannot be completely sure of their significance and the reliability of the definitions of the audio features. Whilst there is little that can be done to overcome this limitation, it should be noted that these audio features are supplied by a global, world-renown company which arguably adds credibility.
The environmental factors and context at the time which may have affected the popularity of a song are not taken into account. Although there may be songs which are naturally popular because of their audio features which can often reflect the different musical tastes over generations, environmental factors can influence the popularity of songs rather than their audio features. Such songs would skew our results and limit our analysis as we have not factored those conditions in. However, we have mitigated this by taking a large data sample of the Billboard chart to reduce any possible imbalance in the data resulting from this effect. We were unable to factor in environmental factors which could have helped elevate our analysis.