Knowledge@Wharton: What question motivated you to do this research?
Jonah Berger: We’ve been doing a lot of work around what’s called natural language processing, extracting behavioral insights from textual data. Everything we do — from this interview we’re recording, conversations we have with friends and family members, reviews we leave online, customer services calls, songs we listen to, articles we read — contains language. There’s a really exciting opportunity now to mine some of this data for behavioral insight to understand why songs or movies succeed, to understand why some customer service calls go better than others, and to use language to be more effective. Essentially, [we want]to extract wisdom from words, so we can all understand behavior better and be more effective.
In this particular case, we were interested in a question that I think many people have wondered about at one point in their lives. Why do some songs become hits? We all know hit songs. We hear them on the radio — we listen to them for years, if not decades, after they come out. Some songs become hits, others fail. Same thing with books, movies, and so on. Why do some of these things win out in the marketplace of ideas, and others fail? I think we’ve all wondered that as consumers, but as a marketing professor, this is something I’ve tried to study and to quantify.
We did a paper a few years ago where we found that atypical songs — songs that are about different things than [usual in]their genre — are more successful. Take country music, for example. Country music tends to talk a lot about things like girlfriends and cars. But looking at thousands of songs across multiple years, we found that songs about different things in their genres are more successful. In that project, we controlled for a variety of other factors: genre, artist, time period, and individual words. We controlled for pronouns. For example, I might use the word “I” or “me,” I might use the word “you,” I might talk about “we” or “us.” We controlled for these things to make sure that those weren’t driving our results.
We had hundreds of controls in that project, but we noticed one thing was particularly unusual. There was one class of words that we were using as a control variable that stood out for us. We started wondering why that might have worked out the way it did, and that’s really what started this new paper.
“We started with a simple question: Why do some songs become hits?” –Jonah Berger
Knowledge@Wharton: In your study, you hypothesized that the success of a lot of popular songs boils down to one simple word: the pronoun “you.” Can you explain to us why?
Berger: First, let’s make sure we understand what we mean by “you.” We use that word all the time. “What are you doing today? How are you feeling? You made me happy. You made me sad.” Songs use this word often. Think about Whitney Houston’s famous song, “I Will Always Love You.” Right? Think about Queen’s “We Will Rock You.” You can think about lots of songs that use the word “you” at some point.
You might be sitting and thinking, “Well, what do you mean? Of course, songs use the word ‘you.’ But I don’t really pay attention to whether songs use the word or not.” Even in our daily life, we don’t pay attention to the word very much. “You” is an example of what is called a style word. It’s sort of treated like noise. We pay very little attention to the words that go in between the content of what we talk about. Language researchers often talk about the difference between content versus style. If I have an important meeting, for example, I’m thinking about the content I’m sharing, what ideas I’m talking about within my presentation, but I don’t think a lot about the little words in between — articles like “and” or “the,” pronouns like “I” versus “we” or “you” that go between these more content-based words.
The words don’t have any meaning by themselves. They are just connectives between larger themes. But what we saw in this preliminary analysis was the word “you” seemed to be linked to success. Songs that said “you” more often — songs like Whitney Houston’s “I Will Always Love You” or Queen’s “We Will Rock You” — seemed to be more successful. We started wondering, why might that be?
We started doing different analyses to try to figure it out. For example, we started with a data set of around 2,000 songs over three years. We went to the Billboard charts, scraped what songs were popular in different years and controlled for a variety of things like radio airplay, genre, artists and the content. We found that songs with more “you” were more successful. We started to do a little bit more work to understand why and what made songs with “you” more likely to be a hit.
Knowledge@Wharton: You point out in your paper that there is a difference between the subjective and objective use of “you.” Can you explain the difference and why the latter case is the more powerful one?
Berger: Take “I Will Always Love You.” The subject is Whitney singing to someone else, saying, “I will always love you.” We might think she’s singing to us. We might think she’s singing to Bobby Brown, or whatever it might be. Same with Queen’s “We Will Rock You.” “We” is the subject of that sentence. Maybe it’s Queen, maybe it’s our sports team when we think about singing it, singing to the other team, “We will rock you.” You are the object of that sentence.
That’s different than sentences in which “you” is the subject, “you” is the driver of the actions. So, “You hurt my feelings.” “You love me.” “You make me feel so happy.” All of those are cases where “you” is the actor or the subject of the sentence.
“Natural language processing is a powerful tool to understand why some songs succeed and others fail, and how artists can impact that.”–Jonah Berger
We started this project by thinking maybe it’s the traditional sense of “you.” When Whitney Houston sings, we sit there imagining Whitney singing to us. It makes us feel really good inside, and that’s why it’s successful. It didn’t seem to be that. Maybe we’re imagining Whitney’s singing to Bobby Brown, and so we’re following a really amazing narrative. It didn’t seem to be that. It seemed to be something a little bit more nuanced.
The idea is, when we hear a song like “I Will Always Love You” or “We Will Rock You,” we think about someone in our own life that we feel that way toward. I think this is quite interesting because this gets to the core of why we like cultural products. Why do we like books and songs and movies in the first place? Sometimes, we like to be transported to other places. We watch a sci-fi movie because we want to be transported to something outside our own lives. But other times, we [consume]these things and enjoy them because they make our own lives in some way better. They help us see our own relationships, our own social connections, as deeper and different as they might be otherwise. When Whitney Houston is singing, “I Will Always Love You,” we might be thinking about Whitney, and Whitney singing to Bobby, or Whitney singing to us. But what we actually find in our study is it causes us to think about, “This is really an amazing, romantic song. Who do I love?” It helps us think of a close other in our own lives.
I was in junior high school when Whitney Houston’s “I Will Always Love You” came out. I remember singing that song in my head to my girlfriend at the time and thinking about how much I cared about her. It activates that self in our own lives that makes us feel more connected to the song. Now, it’s not just this abstract song. It’s this thing that touches our own life and makes our own life feel a little better. That’s one reason “you” is so powerful; it helps us connect with others and like the cultural product more as a result.
Knowledge@Wharton: Let’s touch a little bit on your experiments. How do you actually go about studying this kind of thing?
Berger: I know some folks are probably going, “Well, that’s cute.” Songs that have more “you’s” in them are more successful. But how do you know that’s the reason they’re more successful? There are hundreds of reasons why songs are successful. As I mentioned, we tried to control for various things, like a certain artist might be liked more, certain genres might be more or less popular, and more radio airplay is going to help songs be successful. We did that in the field, but then we also did some experiments. What we wanted to do is essentially manipulate the number of times a song had “you” in it and examine the link to success.
We started with something really simple. We asked a number of people in an experiment, “Think about a song that you’ve heard recently, and think about how much you like that song.” Then we went ahead and grabbed the lyrics to those songs and counted the number of “you’s” that appeared.
“‘The word ‘You’ can drive action, even outside of music.”–Jonah Berger
What’s neat about work like this is, we’re not sitting there manually counting the number of you’s. We’re using natural language processing, automated textual analysis, to count it for us. We use scripts that run through the data and can pick off things like pronouns, counting the number of “you’s.” We find not only that people report liking songs with “you” more, just like what we found in our field data, but we find evidence for a mechanism. When we asked people, “How connected do you feel to others in your own life when listening to songs like this? How much does it enable you to imagine a personal other in your own life?” — when songs happen to have more “you’s” in them, people report imagining a personal other more. And that leads them to like the song more.
You could say, “Well, hold on. That’s nice. But you’re still not manipulating the number of you’s.” My co-author, Grant Packard, has a great songwriting talent. He put together some amazing songs where we could manipulate the number of “you’s.” They’re not chart-toppers, let me tell you, but we made up these songs and asked people to listen to different versions of the songs and read the words from different versions of the songs.
We created one version of a song where there were lots of “you” pronouns – writing, for example, “I’ve known you for a while now,” where “you” is an object. We also created a no-personal-pronoun condition, where the song said things like, “I’ve known it for a while now.” All other lyrics are the same, but we’ve shifted that ‘you’ to become an “it.” We even did a third condition where we replaced the “you” pronouns with third-person pronouns. “I’ve known her for a while now,” or “I’ve known him for a while now.” In all these cases, the rest of the lyrics are the same. They’re identically decent, it turns out. But even doing that, people like the song more when it has more “you’s” in it. And they like it more because it encourages them to think about someone in their own life that they feel that way towards.
What I think is really nice about this is we can not only say we have good causal evidence for the role of ‘you’ in encouraging people to like things and why, but we can also show it affects real songs in the field. If you’re a music artist, for example, and you’re trying to think about how to get people to like your songs more, this is certainly one way [to do that].
Knowledge@Wharton: Are there lessons here for marketers beyond the music industry?
Berger: I think there are a few interesting things that come out of this paper. Specific for the music industry, if I’m a songwriter, the number of “you’s” may impact whether or not my song is successful. If I’m a music producer or a label thinking about investing in a particular artist, this might be useful to know and understand — mining lyrics as a source of insight. There have been lots of people over time who have argued they can understand why songs succeed. [But there has been] very little data actually looking at that. I think natural language processing is a really neat avenue to understand why some songs succeed and some fail, and how we can impact that.
Beyond the music industry, I think this has a lot of interesting implications. There’s other work showing that the word “you” can increase attention. For example, if I’m reading an ad or a piece of mail or an e-mail, and a subject line says, “You need to read this,” or “You won’t believe what happens next” — think about the clickbait world. You often see a lot of second-person pronouns used in very successful online content because it encourages us to pay attention. It draws our attention to something because, “Oh, wait. I want to know the implication of this for me.”
“Companies are sitting on a wealth of textual data, that can provide deep insights into to consumers and markets more generally.” –Jonah Berger
I’ve even done some work looking at online written content for a large consumer-facing company, looking at how the language they use impacts customer service reactions, and the likelihood that people read content and find it helpful. Second-person pronouns play a big role there as well. So, “you” can be a nice word to drive action, even outside of music.
Grant Packard has done some really nice work on this in customer services itself. It turns out we have to be really careful about using the word “you.” First, they find that “I” is more effective than “we.” Rather than saying, “We’ll take care of that for you,” or “We’ll be happy to solve your problem,” if I’m a customer service agent, using the word “I” and taking responsibility, saying, “I’ll solve that for you, I’m really sorry that happened,” rather than distancing it with “we,” is much better.
But “you” also matters. I was having an issue at home a couple of days ago with my Nest thermostat. I called up Google, and they said, “Have you tried this? You should try that. Did you think about doing this?” It turns out when you use the word “you” a lot in that customer service context, it can make people feel like you think they’re responsible. Customers can get really annoyed about that.
Second, as we talked about earlier in the conversation, I think this opens up a lot of avenues to study language and cultural items. We’re doing a bunch of other work in movies, looking at how the scripts may make them successful. We’re doing work in customer service calls and online written content. There’s a lot of opportunity to extract insight from the language data that’s out there.
Many companies now are doing some version of what we call social listening — listening to the chatter on social media about products and brands and services, and mining that for insight. Social media is a useful channel with a set of useful data. But it’s not the only useful data out there. There’s really lots of opportunity to mine more of this data for more insight.
