A Karaoke App Is Using Musical Data To Predict the Next US President

Even if you don't know the parent company by name, you're probably familiar with many of the apps created by San Francisco developer Smule. Founded in 2008 by former IBM software engineer Jeff Smith and Stanford faculty member Ge Wang (the latter left the company in 2013), the company has been responsible for a number of music-focused apps on the quirkier end of the spectrum.

There's the Sonic Lighter, an animated flame that you can make sway in the (digital) wind by blowing air into your microphone. There's also Ocarina, an app that lets users mimic different musical instruments by again, blowing into their mics, as well as I Am T-Pain, which allows you to transform your natural voice into that of the famed auto-tune crooner merely by signing into your phone. But their flagship, and by far most popular, product however is Sing! Karaoke. Released in 2012, the app allows users to choose songs from a library and sing them, solo or simultaneously, with friends across the product's network of over 50 million monthly users. It's essentially a karaoke bar that you can fit into your pocket.

The reason I was speaking with Smith on a rainy afternoon wasn't to hear about how I could transform my nasally New Jersey twang into that of a vocoder sex god (a chat for another day perhaps). Instead, I called him up when I heard that the company is trying to use their karaoke app in a curious new way—to gather mass amounts of user data that they believe can help predict the outcome of the 2016 Presidential Election.

While it might seem farfetched, but collating and studying user data is at the core of what Smule does as a company. Smith has a PhD in in Computer Music from Stanford, documenting cultural differences of music performance interpretation. The point of Smule's studies is to find out more about why people engage in music, and how those findings can be applied to understanding the relationship between music and culture, which is where their research about the election come in.

Their first task in the study was to figure out what songs users of their app were singing in traditionally blue and red states. After collecting data for three months, they learned that, for example, that people in states that traditionally vote Democratic are more likely to sing Latin music and showtunes on the app, while those in Republican states often go with country and Christian music.

Next, in order to predict who will win the election, they needed to calculate the "blue"-ness and "red"-ness of the various swing states in the United States. By studying what kind of songs users in specific swing states were singing, they believe they were able to narrow down the political affiliations of these "schizophrenic states," as Smith puts it. Things got more complicated when they dug deeper into these swing states, noting certain outlier genres not necessarily always associated with Democrats and Republicans, like punk, and 80s and 90s hits across genres.

We caught up more with Smith to hear about the company's unique study, as well as what they ultimately figured out after sifting through millions of people singing "Careless Whisper." (Hint: Hillary wins).

THUMP: Did you have any plans to use user data for these types of studies when Smule was first founded?

Jeff Smith: I think the premise behind Smule is to bring music back to its roots as a creative expressive medium. We believe that music kind of lost its way during the past 100 years during the modern era of recording and has become more of this more leaned back medium. But there's more to music than just listening. Music is part of where we are [as a society]—it's intrinsic to who we are—and at Smule we're trying to open it up so more people can participate in creating and discovering music and not just listening.

A lot of what we're trying to do with our data is find out why people are inhibited in engaging with music, and then asking: how do we take that inhibition down so we have more people singing, rapping, or playing our piano product? Data's been at the core of who we are, but really for us it's a means for us to better understand the relationship of people to music. In the case of this study we looked at 100 million performances of people singing songs through their phones.

Can you tell me a bit more about how that applied to this specific study based around the election? Smith: After looking through the performances we then analyzed them by location and created this normative behavior that predicted what musical preferences were in terms of genre between red and blue states. We noticed there really were distinctive preferences between traditionally red and blue states. In essence, what we did was, swing state by swing state, forecasted from a cultural standpoint if the musical preferences we measured were culturally aligned. We took a snapshot of the data really just from the past few months, so we really didn't go back too far.

You did something similar in 2014 around the mid-term election, correct? Smith: Yes, but there we were taking a look at a different data set, studying performances through our piano product. In the last analysis we used more detailed techniques that came out of my dissertations on how specifically people treated time, and that ended up correlating strongly with what culture was in place in the study. We did call the election in [2014]—we missed Florida—but the result came out pretty close to the results of our study.

Can you explain to me some of the findings about swing states this year? Smith: If you look at a state like Ohio you can see it's really a schizophrenic state. You have a significant percentage of country music in that state—5% according to our data—but you also have 6% of punk rock and alternative music coming out of Ohio as well. So here's a state that's swinging from both sides. When we're looking at a statistical model to predict where Ohio was going to go, we found that the alternative and country were both kind of awash, so we had to look at other signals. What was most interesting in Ohio was all of the 80s and 90s music we saw, like Bryan Adams, Whitney Houston, and Lionel Ritchie type stuff. It's kind of funny, but that music and demographic we found was going to be the best predictive demographic on where that state was going to end up. It turned out the people most aligned with those genres was more closely aligned with blue states versus red states.

How do you think someone's music taste can be connected to how they'll vote? Smith: One of the things about this election is that it's especially confusing to a lot of us. If you look at platforms, candidates, and what's being said it's hard for most people to reconcile that back to the polls and the percentages you're seeing. If you go one level deeper beyond the political platforms and ask what are the real values, how do those diverge, and how can we measure, those values for culture, you'll have a different picture that emerges and you can see that you have a divided country.

You have vastly different cultural biases between these different regions, so that's almost the starting point of the election—the culture, those values. If that's the starting basis, what we're really looking at with the election are marginal swings around inherent biases, and which candidate, perhaps, is more able to cross some cultural divide. I think that's more or less what we were seeing in our study. It was inevitable whoever was running on the Republican ticket was going to carry most of those red states from a cultural standpoint—the question is, was there some common cultural expression that was going to predict how some of the swing states would vote.

Mark Dunning (VP of Data Analytics): The musical preferences across states are very consistent. The interesting things are how in certain states, especially within the undecided states mentioned, you have punk music, and different genres of music that have come up. So I think as you see music become more diverse, in my view it can be interpreted as people becoming more diverse, too. As their music tastes broaden, probably so does their spectrum in terms of their political thoughts.

There's something in the study you've called "The Kevin Costner" factor. What does that mean exactly? Smith: This is one of the fun parts about pulling this data. If you take some place like Ohio and look at the music preferences there you can see that there is a strong preference towards 80s and gospel music, and that's contrasted by preferences towards showtunes. They offset each other. What we found in the undecided states is where we have factors that offset each other, what was common was the 80s-90s genre being the key component that could decide this election [because those songs were sang more often in places affiliated with democratic voters]. When we looked into that further we noticed that the songs that really popped a lot, and were statistically more significant, were one's like Bryan Adams' "I Do it For You" as well as Whitney Houston's "I Will Always Love You," both from movies featuring Kevin Costner. So it's slightly tongue and cheek but we found that maybe [Kevin Costner] could be the difference between Hillary and Donald being in the White House.

David Garber is on Twitter.