Does not compute

The idea that there's an algorithm which can predict if a record will be a chart hit often pops up; there's a new one being reported on by BBC News:

University of Bristol scientists claim to have developed software that can spot whether a song has hit potential.

The program looks at 23 separate characteristics including loudness, danceability and harmonic simplicity.

Trained using hit songs from the Top 40 over the last 50 years, the software can predict chart positions with about 60% accuracy, the scientists say.

Really? Given that most singles released in the last half century vanish without trace, I'd be a bit surprised if the University Of Bristol has managed to number-crunch them all, and without the ones which sounded like hits which never took off, you'd have to question that 60% accuracy figure.

Indeed, it turns out the 60% accuracy figure is based on working out where a new entry would end up, not on if a new release would chart.

That might have been a useful tool back in the 70s, when singles would enter the charts low and wind their way slowly up the list, like a tired man carrying shopping up a hill. But since the trend started for songs to debut at the top of the chart, it's hard to see what use this would be: "Your song which went straight in at number one sounds like a number one".

And there's this:

What regularly tripped up the equation were the unexpected hits that became popular for reasons that often had nothing to do with their musical qualities, he said.

In 2010, Surfin' Bird by The Trashmen reached number three in the UK charts, thanks to a web campaign persuading people to buy it to prevent X Factor winner Matt Cardle being the Christmas number one.

In a similar way, said Dr De Bie, the equation could not determine to what extent marketing determined whether a song was a hit.

Not being able to take account of marketing when determining likely chart outcomes is akin to predicting the outcome of a horse race without counting the legs of the horses.