Transcript

Voting biases in music polls

Audio

You can also listen to the interview [MP3 15.2MB].

iTunes

Visit this channel at La Trobe on iTunesU.

Transcript

Matt Smith:

Welcome to a La Trobe University podcast. I would be your host Matt Smith and my guest today is Dr Liam Lenten from the School of Economics at La Trobe University. He's here today to talk about what factors could make a song successful in a music poll, specifically the Triple J Hottest 100, an annual music poll on Australian music station Triple J, in which listeners determine the favourite song of the year.

Liam Lenten:

The Hottest 100 was originally the brainchild of Lawrie Zion, who back in those days was a Triple J presenter. These days he's one of our colleagues here at La Trobe University in Journalism and Media Studies. He started the Hottest 100 in the late 1980s as an all-time music poll, ie the hottest 100 of all time. But then it became an annual poll from 1993 onwards, charting the hottest one hundred songs of the previous calendar year. Since then, it's become the world's most popular on-line music poll.

Matt Smith:

It's very much an alternative music station, isn't it? You won't get what we'll call mainstream music.

Liam Lenten:

No, because Triple J has somewhat of a charter to promote independent and alternative acts and they also have an emphasis particularly on local and emerging talent, so we would expect to see somewhat of an Australian bias in their playlists.

Matt Smith:

You're more known as a sports economist. How did you become interested in the music poll?

Liam Lenten:

It was Australia Day 2007, and I was adamant that Eskimo Joe was going to carry off the big prize with their song Black Fingernails, Red Wine. Much to my surprise they were beaten into number 2 by Augie March's One Crowded Hour and I'd started wondering whether this was the case merely because Augie March had the advantage of having a band name that began with the letter A and whether they attracted a lot of votes merely because a lot of people who went in to vote browsed the list of songs by artist's name and subsequently voted for that particular song because that was when they were most susceptible to ticking a box. There was also a controversial incident in the 2009 Hottest 100 of All Time poll where no solo female artist appeared in the Hottest 100 and a lot of listeners were outraged that this had occurred. And that also got me thinking about the possibility of a gender bias with respect to artists. After that I teamed up with Geordie McKenzie from the University of Sydney, who has done a lot of work previously on modelling box office revenues from the motion picture industry, and we began discussing various ideas for other factors that may lead to a song performing better than expected in the poll.

Matt Smith:

So what were you looking for within the data?

Liam Lenten:

Well, we started out with the presumption that songs that resonate more with the music listening public will perform better in terms of all different ways that we can capture the concept of demand. So one way would be to look at unit sales – that's via the ARIA charts. The Hottest 100 is another measure of how well a song resonates with the music listening public, in fact, the correlation between the two is quite significant at 40%. So what we're looking at doing here is trying to explain some of that remaining 60% as best as we can. And in terms of varying from that correlation, what we're particularly interested in is, what causes a song to do better in the Hottest 100 relative to what you would expect given how well it's done in the charts.

Matt Smith:

So how were you looking for this?

Liam Lenten:

The main issues that we faced were in terms of identification of the factors that we thought might make a difference here and also measurement of these factors. That's not to mention having to build the database and that was highly problematic because we were dealing with songs from up to twenty years ago. If I could focus on two examples of these factors – we started out by looking at broad ideas, and one was that promotional aspects should make a difference to how the song performs in the Hottest 100. A lot of this stuff is driven by the station itself. For instance, if they get more air play during that particular year on Triple J, then we would think that that song would do better, all other things being equal in the Hottest 100. We don't have data on how often songs were played on the station but one little way that we can capture that is via a segment that Triple J has called The Feature Album Segment, and this is where for a given week, they take one album and they will play a lot of the tracks on that album a lot more often than they would during the remainder of the year. So we can construct a simple dummy variable, whereby a song that was on an album that was feature album at some time during the previous year was given a value of 1 and zero otherwise. Another segment that they have on the station is called The Unearthed competition, whereby new emerging artists are given the opportunity to submit a song, and then a winner is selected from this and then that song is generally given greater airplay. So there are a couple of ways that we can kind of identify those promotional aspects. Another example that we also think is important is whether that band had a touring presence at approximately the time that voting is taking place. Now, we certainly can't go back twenty years and identify who was touring during the voting period over the past twenty years, but a very rough way that we can get at that is to look at the bills of the Big Day Out, because that's a festival that's been running virtually every year that the poll has existed, similarly to before we construct a dummy variable if a song was by an artist that was on the Big Day Out because the Big Day Out festival in most of those years has been highly contemporaneous with the voting period itself. Now, those issues aside, ultimately we use standard statistical regression techniques to actually estimate these models.

Matt Smith:

What kind of characteristics do you find of songs that do well in the music poll then?

Liam Lenten:

I did mention the feature album variable as part of trying to get a promotional aspect. It turns out that the feature album variable is quite significant. The Big Day Out variable is also significant, and the reason for these are quite sensible, but it doesn't always work like that. The Unearthed variable is insignificant across the board. For some strange reason, Unearthed winners have a very poor record in getting their song into the Hottest 100 in subsequent years. And similarly, and perhaps surprisingly again, a variable that we constructed for another Triple J promotion called The One Night Stand was also insignificant across the board. Much to my surprise, we found that the artist alphabetical ordering does not matter at all, but it turns out that the song alphabetical ordering did make quite an important difference and I put those two findings together and it may seem like a little bit of a mystery, but then again it's makes perfect sense, if, and only if, the majority of voters browse by songs rather than by artist. Anyway, all of this is based on, very much, the voting system. So if any of the listeners of this podcast have recently voted in the Hottest 100 for this year, they may recall the voting procedure in which you go into the page, and there are links to every letter of the alphabet, and you can actually sort through the list of songs in this way for browsing purposes. And I think that that's what a lot of people do. Once a lot of people get down to the Es and the Fs and the Gs and beyond, I suspect that a lot of people begin to realise how time consuming the voting process is, if indeed you want to scan every single song in the list. And that is specifically the factor that I believe explains the alphabetical ordering being significant, at least for the purposes of the song title. Also nationality is a very important one to look at, and it will come as no surprise that Australian artists do tend to have an advantage relative to chart success in the Hottest 100. This would be something that if I was Richard Kingsmill for example, as the music director of the program, would look at with much pride, since we as a station are here to promote Australian artists, but a couple of other little general features as well. We suspect that the timing of the release of the song during the calendar year can potentially make a difference. We weren't able to obtain definitive results on this one, but if during a given year, the song is getting its maximum airplay on the station earlier in the year, that that's a bit of a disadvantage. And the very sensible reason for this is that these are the songs that by the time that the following January comes around, perhaps voters have lost their connection with the song a little bit, because it has been a while since they were listening to it more regularly on the station. On the other hand, having the maximum airplay, we think, in that window between around about August to let's say, November, it is probably the peak time. And one other factor that we tried to control for was loosely how long the artist has been around. How we tried to measure this was looking at the number of songs that that artist has previously had cumulatively in the Hottest 100 polls. Initially having songs in the Hottest 100 is an advantage for doing better for the poll in future years, but beyond a certain point it becomes a liability. So, if you've already had five, six studio albums, we have this perception of Triple J listeners is that they want to be ahead of the curve and they're always looking for the new and emerging artists, rather than listening to the same artists year after year after year. But not all of the factors that we identify are an advantage. One that we identified as a disadvantage is if we knew that the song was a cover version of a previous song, but at this stage it kind of helps us to recollect that our benchmark is ARIA chart data and a lot of cover versions tend to do quite well in terms of sales, simply because a lot of people like cover versions because it's a recognisable song to them. But Triple J listeners probably don't really think that way so much.

Matt Smith:

Didn't Boy and Bear go really well last year, with a cover version of Fall at Your Feet?

Liam Lenten:

Absolutely that is the case, but we'll call that the exception rather than the rule. What you have to remember Matt is that unfortunately the models themselves don't have a very large explanatory power. And that makes perfect sense, because what we're dealing with here is a data set on musical tastes essentially, and musical tastes, as we know, are highly personal and wide-varied from one individual to the next. Finally, the controversial one that I mentioned before about the possible gender bias is that we did find, albeit very weak, evidence of a pro-male bias, but given the levels of significance that we found, I wouldn't put too much credence on that one.

Matt Smith:

If Triple J wanted to rectify these voting biases, what would they need to do?

Liam Lenten:

The primary answer to that is that it depends on how these biases are interpreted. If I was Richard Kingsmill, I might be looking at something like the Big Day Out dummy variable being significant, and in one sense it might actually be a little bit worrying to me that certain artists seem to be doing in the poll just because they were part of the bill on the Big Day Out. On the other hand I might be inclined to look at it and think, no, this is actually a good thing because we are a station that is promoting light music and I would be proud to see acts that are on the Big Day Out are doing better and again, this might be sufficient evidence that, given our charter, that we're doing our job properly. And a similar argument might be made for the Australian artist bias as well. Nevertheless, there might be a couple of those biases that are not that difficult to fix. So let's go back to the song alphabetical bias. If we know that voters are swayed by songs that have an earlier alphabetical title, one option would be to simply remove the ability of them to sort through the list in this way, and only give voters a chance to browse through the list in a purely random fashion. Even still, if there are voters that prefer sorting through the list in that way, you run the risk of discouraging them from voting altogether. So any of these possible changes that they could make would have to be treated very, very seriously. Alternatively, for a couple of the other biases, the one regarding the Big Day Out and the other one that I said regarding the timing of the song peak airplay during the year, the station could consider devoting a single weekend during the voting period where they play a lot more of the songs that were getting peak airplay earlier in the previous calendar year, simply as a means of reminding listeners about some of the songs that are eligible for them to vote for, and yet they may have lost their connection with them, and that would be a policy change that would be reasonably easy to implement without potentially contaminating other facets of the poll itself.

Matt Smith:

So based on your data, what is your prediction for the top three songs in the Triple J Hottest 100 this year?

Liam Lenten:

Look, while this model can be used a little bit for forecasting purposes, I would be very hesitant to put a high degree or reliability on what I'm about to say. But if I mention a couple of songs, I wouldn't be surprised if these songs do better than expected, but I don't want to go out on a limb and say that these are the songs that are likely to finish in the top three.

Matt Smith:

OK, disclaimer made.

Liam Lenten:

Going back to what I said at the beginning, we know that there is a high correlation between how the songs do in the charts and how they do in the Hottest 100 poll. The starting point is to look at the year-end charts for the calendar year of 2011, and it turns out that the second biggest selling single of 2011 in Australia was somebody that I used to know by Gotye featuring Kimbra, and then the next one down of all the songs that are likely to be voted on in the polls was at number 23 and that was Foster the People's Pumped Up Kicks. Now, yes, I would expect that both of those songs would perform very well. I'm not going out on a limb and saying they'll finish one and two, but if I look further down that particular year-end chart, I don't necessarily think that those ones will do as well as people would expect. So, at 51, there was Lupe Fiasco with The Show Goes On, but that song doesn't really tick many of the boxes, as it were. So I wouldn't be surprised if that one didn't do quite as well as expected. At 54, was Coldplay with Paradise but one thing, and it's a major factor going against them is that Coldplay haven't managed to crack the Hottest 100 since 2005, and when you consider what I said before about how Triple J fans often tire of the same artist, I might be inclined to believe that that song may struggle comparative to expectations. I also applied this methodology to the list of ten songs that I myself have voted for in this year's poll, and the one that stands out for me is A.I.M. Fire! by Art versus Science. It has a title that begins with an A, but it also is subject to the gender bias – if it does in fact exist – it's an all-male group, they are indeed touring on the Big Day Out this summer, they're an Australian group of course, and they have had a couple, but not too many, of previous Hottest 100 appearances, so overall I wouldn't have expected A.I.M. Fire! to do very, very well, but it may come in higher than a lot of people expect. Seven of the ten songs that I selected are by Australian artists, five of them were kind of getting their peak airplay in that narrow window between August and November that I mentioned before. So, yes, it turns out that I myself are subject to a few of these biases that we've identified.

Matt Smith:

Dr Liam Lenten there, and keep your eye out for a proposed 2013 subject at La Trobe University focussing on the economics of music, film and television. If you have any questions, comments or feedback about this podcast, or any other, then send us an email at podcast@latrobe.edu.au.