Is There a Formula For a Hit Song? 243
moveoverrover writes "What happens when two Rutgers grad students analyze 50 years of Billboard Top 10 hits with MIT offshoot Echo Nest's API and turn the data into visualizations for an assignment? Great looking visualizations for one, and a fascinating look at 50 years of Pop music at the data level. Posing the question, 'Is there a formula for a hit song?' the students write, 'What if we knew, for example, that 80% of the Billboard Hot 100 number one singles from 1960-2010 are sung in a major key with an average of 135 beats per minute, that they all follow a I-III-IV chord progression in 4/4 time signature, and that they all follow a "verse-verse-chorus-verse-chorus-bridge-chorus" sequence structure?' Using data extracted by Echo Nest on tempo, duration, time signature, musical key, as well as subjective criteria like "energy" and "danceability," the pair generated a number of visualizations with Google Motion Charts (warning: slow) and '(some) Tableau Results' for everyone to see and investigate. Curious about tempo and song duration trends in Pop music over 50 years? Correlation between record label and song tempo? Download the core data, the Tableau reader and look at it any way you want."
Correlation and causation (Score:5, Insightful)
Re:Correlation and causation (Score:4, Insightful)
Indeed, the trends they spotted over the years may also apply to all the songs that never quite made it to the top, or even into the charts.
Re:genres and trends (Score:3, Insightful)
Re:Is it the 1970s again? (Score:4, Insightful)
I heard a similar account from a buddy of mine who was in the recording industry back in the mid 70's. Someone did a bunch of metrics to determine the characteristics of a hit song and came up with some average: x% singing, y% cellos, z% electric guitars, a tempo between t1 and t2, etc. And then they made a song that had exactly all of that stuff.... and it sucked.
A brief skim of TFA leads me to conclude that it's rife with half-thought-out research. The question they pose, "What if we knew, for example, that 80% of the Billboard Hot 100 number one singles from 1960-2010 are sung in a major key..." is completely meaningless if 80% of the entire population of songs, hits and non-hits alike, are in a major key... with a 4/4 time signature, etc. It's like determining that 100% of all coffee drinkers have faces. 100% of people have faces, so you haven't discovered anything different about the coffee-drinking subset.
What you're looking for is what sets the "hit" population apart from the "non-hit" population. And, from what little I looked at, they don't address that at all.
They also try to slap a linear regression onto everything. They assert that song duration is increasing. Umm, no... it was increasing during the 70's, and then it stabilized. And that probably had a bit to do with the formats that the music was available in (ie, 78-rpm records vs. 33.3-rpm...). But, again, we would only know that if these jokers looked at the average duration of *non*-hit songs.