Everyone has a song which irritates the hell out of them – but Achy Breaky Heart by Billy Ray Cyrus was found by one 2005 study to have been nominated most often as “the worst song ever”. The authors, academics from New Zealand and the US, listed a few reasons: awful lyrics, an overly simple melody, negative personal associations, but they also found that their respondents “wrestled – unsuccessfully – with the problem of providing a reasoned, rational analysis of a visceral response”.

In other words, they found it hard to put in to words just why, or how much, they hated the song. From songs and films to universities, baking and mortgages, it seems as though everything is now ranked and rated. Consumers want to know what to choose, and organisations want to know what to back. Getting the rating right is important, so can we objectively distinguish the good from the bad?

In some cases there is a clear, objective criterion. When two football teams play each other, the better one scores more goals. When choosing between two mortgages, the better one costs less money. Sometimes we want to know which will be better in the future. Which football team is going to win next weekend, and which mortgage will cost less in ten years? We can guess, or we can make an objective prediction based on past data. So, for example, we can usually say with some confidence that Manchester City will probably beat Southampton.

So, how about songs? There have been claims that machine learning can use data from past chart performances to predict, from its acoustic characteristics, a song’s likelihood of success. Tests have yielded mixed results. Research which has been successful in predicting success has mostly been in limited domains. A larger study found machine learning methods could not distinguish what acoustic characteristics led to success.

This is hardly surprising. Although many hit songs have characteristics in common, there are always oddities that succeed when in theory they should not – remember Crazy Frog? In the wider world of music, acoustic characteristics seem to have little impact on whether a piece classes as music at all, let alone whether or not it is successful. There is the John Cage piece, Organ²/ASLSP (As Slow as Possible), which is scheduled to last 639 years, Gÿorgy Ligeti’s Poème Symphonique which solely constitutes sounds from ticking metronomes, and an entire composition – once again from John Cage – in which no sounds are played at all. All three of them regularly bring in audiences (including me).

It is clearly difficult to predict musical popularity, but judging the characteristics of a song – such as mood or “danceability” – has been much more successful. As with most things we choose different types of music for particular purposes. The tunes that ease the morning commute may not help you get your groove on in the evening.

Specific characteristics of a song contribute to its effectiveness for certain uses: a clear beat around 120 beats per minute if you want to dance – or something with no sudden changes in tempo if relaxation is what you want. The most successful song ever, by number of times it has been performed, is almost certainly Happy Birthday by Patty and Mildred J Hill. It is superbly suited to its sole purpose: a public and often spontaneous celebration. It is short, easy to remember and easy to sing. I doubt, however, that anyone would claim Happy Birthday was the best song ever.

Although objective characteristics can teach us something about how suitable a song is for a given situation, the notion of a song being “good” or “bad” in an absolute sense is much more problematic. But anyone who has ever switched off Radio 1 in disgust – or wrenched the sound system away from a friend playing just the wrong part of Madonna’s early work – has had the experience of recognising whether a song is good or bad.

How is it that we can be so confident in our own judgement and yet incapable of designing an objective means of explaining why? “Ultimately”, concluded the 2005 study, “the songs that we dislike depend as much upon ourselves as upon characteristics of the songs.” The characteristics of the songs are fixed. The characteristics of the listeners can change.

So here is my hypothesis. Really great songs are those which transcend the purpose for which they seem intended and make a change in us. On hearing a song like Leonard Cohen’s “Hallelujah” (performed well) we become a different person, a person who loves that song. Bad songs are not those which just leave us cold and unchanged, they make us actively hate them.

And bad songs are no use for anything except annoying our friends. Remember Crazy Frog again?

This article was originally published on The Conversation