So last fall I saw Moonlight when it was released, and I thought it was pretty good. It was provocative in subject matter, and the heavy focus on style got me thinking about the choices the director had made and why they were made. You know, the stuff you do when you see a “good” movie and want to make sure that you got it. Art house isn’t really my bag, but I enjoy putting on my film-major hat now and again and pretending I understand what’s happening. It was during a conversation later however, that I was informed that Moonlight had been projected by experts in the matter (whoever they are) to be a strong contender for the Best Picture Oscar. This was a bit surprising, as I wouldn’t have expected such an art house film to be up for the award. Moonlight was well crafted (I assume!) but let’s be real here, it wasn’t exactly captivating. A wise friend told me it was “so woke it forgot to have a plot,” and I would agree. When I was younger I remember Gladiator winning best picture, and I don’t think anyone would have accused that movie of being “artsy.” And come to think of it, didn’t the Lord of the Rings win in 2004?
Something weird is going on. A quick glance at the Oscar winners for best picture over the past several years seems to confirm this. Winning movies used to be broadly accessible, and now they seem artsy and niche. There could be many, many possible explanations of this, but I obviously won’t let that stop me from making an accusation: I think the Academy is picking them this way. I have some kind of hypothesis here, and so now I need to test it. So what do I do? I scrape movie box office data from here, grab the best picture winners and nominees since 1980 from trusty Wikipedia, and fire up R. The basic idea I want to test is this: is the Academy picking less popular Best Picture nominations then they used to?
So before I get into the statistics, I want to lay out a few descriptives I’ve learned about the movie industry. First, the size of the market has been relatively stable since 1980. I’m not sure what’s behind this, but a broadly comparable number of tickets were sold in 1980 as today (1.3 billion today vs 1.0 billion back in 1980). Second, more movies are produced today than 1980. 735 movies were produced in 2016 versus only 161 in 1980. However, there’s a very strange discontinuity in this series, as the number of movies jumps from 173 in 1981 to 428 in 1982, so I’m going to assume something weird is going on with my data for those two observations and drop them. So now I’m looking at 1982-2016.
What I want for my dependent variable is some measure of how popular these movies are. A cheap way to get a this was to divide the domestic gross of the movie by the average ticket price for that year to get a rough estimate of tickets sold. I used domestic gross as I didn’t want an expanding international market to skew the data. I then wanted to somehow normalize these tickets sold by the expected number of tickets sold. This presents a problem however, as I don’t have a complete dataset of movies released in a given year. Were this the case, I could construct some idea of what the distribution of ticket sales looks like for a given year and possibly base my inference on this. If you’re thinking, “hey, a Z-score comparison would be perfect here, but the central limit theorem doesn’t apply, so you have no idea what the ticket sale distribution of released movies would be, nor do you have any reason to believe said distributions are comparable year to year,” then you’d be right. I’m not going to lie to you, these data are extremely limited in what they can tell us about the structure of the movie market. So… what can I say?
Well, I can point out that in each of the three data cuts I tried (hand to God, this was a first run through so I wasn’t p-hacking), fewer people are seeing the movies the Academy is picking. In the first graph, we see all the nominees in our dataset. In the second, only the winners. In the third, all nominees since 2000. The line is a univariate linear regression of year on ticket sales, and in each case the coefficient is negative. The outcomes are significant at the p < .01, .10, and .05 levels, respectively, though this isn’t really meaningful, since we’re not trying to pin down an actual value for the coefficient on year. I just thought I’d let you know! Further, significance (and even the sign of the coefficient) becomes fairly sensitive to the cutoff year as our sample size decreases. You can eyeball the winners, for example, and see that if we include the outlier year Lord of the Rings won, plus just a few observations prior to that year, we could get a positive trendline. If you recall, my initial hypothesis only referred to the best picture movies within the past ten years or so, so we’re actually working with a small sample size.
It thus doesn’t look like I can speak with much confidence about my hypothesis, except for the fact that I can say fewer people are going to see these movies in theaters. It could be that people are watching these movies elsewhere. It could be that since more movies are being released in markets of near identical size, even if the best picture nominations are selected from the same point of distributions identical up to number of films released, we should expect fewer people to see them anyway. There are a lot of reasons this could be happening. For next steps, I hope to collect a complete dataset of all movies produced in these years so that I can begin to develop some idea of the market structure of movies released in a given year, and possibly begin to say something meaningful on changes in the types of movie Best Picture nominees are drawn from.
However, in the spirit of ultracrepidarianism (this is a blog, after all), I would like to note that the data thus far have not disproven the narrative I was trying to impose, which is one of cultural polarization. In my mental model of how the Academy is evolving in their decision making process, they appear to be putting more weight on movies that have an artistic or socially meaningful edge (Birdman, 12 Years A Slave, etc.) rather than movies that are merely well made and popular (Gladiator). This means that being a part of mainstream culture is no longer a prerequisite for a movie to win an Oscar, and we see the elite in yet another field (this time Hollywood) begin to isolate itself from the experience of the average person. Anecdotally, in my coastal elite bubble, loudly declaring that Moonlight was a great film deserving of an Oscar was a common sentiment. It was worn with such a badge of pride that one would almost suspect virtue signalling. Now, the elite have always attempted to separate themselves via their taste from the common people, and this may simply be a trend that will correct itself when the plebes develop a taste for movies like Birdman. But, what if it just represents just another way our country is segmenting into different worlds?