Article: Friday, 2 May 2014
Knowing which party a consumer votes for can potentially predict what kind of car they will drive, the brand of clothes they wear, and the name of the cologne on their bathroom shelf. It certainly predicts the kind of movie they will pay to see.
I read an article in "The Economist" ten years ago that said pornography sales in the United States “bore an eerie resemblance” to an electoral map*. That is to say, it is theoretically possible to predict which party a person would vote for by tracking their consumption of purchased pornography.
I discussed this idea with my co-author, Ron Shachar of IDC Herzliya, and we wondered if the reverse would also hold true. So many variables about people are used to predict election results, from basic demographics right down to how an individual uses Twitter.
So, what if we were to switch the regression equation around and see if election results could predict these variables? Namely, can you look at consumers’ voting decisions and use them to predict their purchasing decisions? As it turns out, you "can", at least with a certain kind of product. Our paper, "When Kerry met Sally: Politics and perceptions in the demand for movies", examines this correlation, revealing some interesting results for marketers.
We chose to look at movie sales rather than pornography or even blue jeans for two reasons:
First is the practical importance of predicting whether a movie will be popular. The movie industry is very well-studied in the marketing literature because it's an industry where every product is unique. Predicting a movie's success is something like predicting the success of a completely new type of yoghurt, in which the flavour has never been used before, the consistency has never been used before, the packaging has never been used before, and so forth. Nobody really knows for sure whether the movie will be good until they go to see it. As making movies is neither cheap nor easy, a great deal of time and effort is spent trying to predict how well audiences will receive them.
The second reason is more psychological. We think movies have the potential to appeal to consumers’ self-images and aspirations. Unlike more functional products (think plastic spoons), movies call for emotional and intellectual engagement; they encourage a viewer to identify with – or in opposition to – the main selling points of the product (the characters within the story). This type of emotional engagement can be – and often is – engineered in the marketing of other products that appeal to a consumer’s sense of “who you want to be.” In this way, the idea of using electoral data to predict consumption can be extended to many more product categories.
While genres are routinely used to classify movies, this method can overlook important similarities between seemingly disparate films. What we call “perceived attributes” are the consumer’s own perceptions of a movie. These attributes are much more subtle than standard genres, particularly when factors like the ethnicity or gender of the lead actors are concerned. Many moviegoers will happily acknowledge that they like romantic comedies, but few will realise or acknowledge that they consistently choose young white female lead actors over African-American male leads. Even fewer will realise that these choices correlate closely with their political preferences.
Compared to the list of genres that typically describe movies, perceived attributes should be more meaningful to marketers, because they are measured directly based on the movies consumers choose to watch. Genres, on the other hand, are defined in a top-down fashion by reviewers. The usefulness of perceived attributes (when compared with the genre system) in grouping movies and predicting their fit with consumers is so pronounced that we were able to show a US$93 million improvement in yearly revenue forecasting in the United States film industry "before" we even factored in the political data.
Perceived attributes group films in ways that would be impossible under the standard genre system. And it wasn’t until we laid out the groupings in visual form that the striking nature of these similarities became apparent. For example, although the films "Crouching Tiger, Hidden Dragon" and "Ocean’s Eleven" would be classified into very different genres (action/drama/romance vs. crime/thriller), consumers perceived them to be quite similar.
Over the course of our study of movie sales, six significant latent attributes became apparent, as did their correlation with consumers’ voting preferences.
For instance, in markets where votes favour the Democratic Party in congressional races, voters prefer movies with African-American male leads. At first glance, one might expect this to be due to the popularity of the Democratic Party among African-American voters, but the numbers stayed true even after we adjusted for a large number of demographic variables. Congressional Democrat voters like to see movies with African-American male leads and congressional Republican voters prefer movies starring young white women. It’s that simple.
Because the identified attributes come out of the collected data (rather than pre-defining categories and trying to make the product fit into one of these categories), it is conceivable that this approach can be adapted to other products that also appeal to consumers’ self-images. The attributes will obviously be different for each product category, but the principle should remain the same.
Our paper uncovered two separate findings. The first relates to the relationship between electoral results and movie sales. The second is the use of perceived attributes in predicting the success of a movie. Each of these is useful in their own right and when combined create a highly accurate model of consumer behaviour.
Unlike census data, electoral results are “refreshed” every couple of years. Not only do these data give an updated map of political views that can be used to predict the sale of a certain kind of product, they have the added benefit of giving a more accurate reflection of the changing demographics within a geographical region. These data are an untapped marketing resource.
The method we used to predict movie sales is applicable not only to two-party systems like the one used in the United States. In fact, it is fair to expect that the information gathered within a multi-party system would be even more detailed, and therefore more useful. Let’s say we live in a world that has only two soft drinks - Coke and Pepsi. If all I know about you is which of these drinks you prefer I might accurately predict if you like potato chips. But in a world with Coke, Pepsi, and 7-Up, I might do even better, perhaps even predicting which brand of potato chips you like.
Perceived attributes go beyond typical product classifications. In this paper, we believe we have tapped into something fundamental about what people see in movies. Extrapolated to other settings, this approach might reveal fascinating associations across seemingly unrelated product categories. By looking back at sales of past products and lines and applying a model of perceived attributes to explain their failure or success, we could conceivably provide a foundation of knowledge that increases future sales and directly influences product development.
Science Communication and Media Officer
Corporate Communications & PR Manager