So, I want to talk about this profile of Worldwide Motion Picture Group, and their scientific process (“scientific” process) of evaluating the statistical likelihood of success for a screenplay. Here is an AVClub article about them, and about their founder, Vincent Bruzzesse.
This is actually very exciting for me! I like the idea of behavioral psychology, and you know, simple machines that are designed to churn out plots and stories have been around in Hollywood for half a century at least.
So, in the first place, a lot of creative people (or, anyway, “creative” people, like Holland) are worried that this is going to take the imagination out of major Hollywood filmmaking, which, I mean. Hahah. Right? We can all agree that major Hollywood pictures are only creative by accident; studio execs use all kinds of weird metrics to figure out if a movie is going to be popular that have very little to do with, “is this story imaginative and good” – it’s not like we were in a golden age of free-range writers, liberated to put down their brilliantly creative ideas into script form, which would then leap to the movie screen largely unchanged. Be serious! This is just an explicit form of the kind of dumb process that already existed. If it proves anything, it’s not that writers can be replaced by math; it’s that studio executives can be replaced by math.
Why, exactly, should I pay a guy to figure out whether or not The Lincoln Lawyer should go into production, when I can just use some spreadsheets? (The Lincoln Lawyer is listed as one of the successes of Worldwide MPG, and let me just remind you that The Lincoln Lawyer had a worldwide gross of $75 million over a production budget of $40 million. If a good rule of thumb is that a movie spends its budget again in marketing and distribution, then The Lincoln Lawyer actually LOST money. It definitely did worse than Ghost Rider.)
In the second place, remember when Netflix used ADVANCED PSYCHOMETRICS to figure out that they wanted to remake House of Cards? Like, a successful company spent time and money, conducting focus groups and gathering usage data, and finally compiling it all into a statistical model that told them, “People who like cerebral British TV shows also like Kevin Spacey and David Fincher.”
Jesus Christ, thank fuck we’ve got MATH to figure that one out for us.
(Compare to the traditional studio process, that would remake House of Cards starring Channing Tatum, because he’s so hot right now.)
Beyond this, though, I think – it’s hard to tell without being able to see it – but I am PRETTY SURE that Worldwide MPG’s system is actually bullshit (haha, I know, hear me out though).
When Bruzzesse gives his examples, he gives them according to the statistical elements that are featured in a plot – for example, he says, that “summoned demons” are statistically less popular than “targeted demons” (I don’t know exactly what this means, so let’s just keep in mind that using statistics to predict the value of scripts ALSO relies very heavily on being able to translate a script into mathematically actionable data). Of course, you and I know that Paranormal Activity and The Exorcist both featured Ouija boards, and they were very popular movies! But Bruzzesse’s statistics don’t favor them, and that’s because he’s comparing individual films to the entirety of the sample set – that is, to horror movies year over year.
You’ve realized right away, I’m sure, what the problem with this is, but I’m going to spell it out anyway:
That’s not how people go to the fucking movies.
You don’t make a decision about what movie you want to see based on what movies have come out historically you make a decision about what movie you want to see based on what movies are out THIS WEEK, so who cares if ten years ago, five movies came out with bowling scenes and none of them did very well? You’re seriously telling me that on a weekend in which Iron Man 3 opens against the second week of Pain and Gain, that a BOWLING SCENE is going to have a detrimental impact on its box office?
What this sounds like is basically hucksterism. Bruzesse is a statistician, so if there’s one thing he knows, it’s that correlation does not imply causation, and if there’s TWO things he knows, it’s that he has to explain that to people EVERY SINGLE TIME. I believe that what we’re looking at here is a guy who knows that studio executives are terrified all the time of losing their jobs, that they don’t really understand causal relationships, and that they’re easily impressed by confident-sounding-numbers, and so what he’s selling is not a system for predicting the success of movies per se, so much as a system for assuaging the fears of paranoid producers.
That SAID, I think that you COULD make some interesting predictions , and so I am going to posit my OWN method, which is proprietary of course, you can’t use it, but if you want to pay me an obscene amount of money I will use it to predict which scripts you should by.
For this to work, I would argue that you need at least THREE models in order to establish opening rankings, and you need maybe a fourth one to predict potential revenues.
1. Theater-over-theater opening. You identify the unique elements of a screenplay (I wouldn’t actually bother with scenes, since in the first place, no one really knows what scenes are going to be in the movie before they see it, and in the second, I’d guess that fully 90% of scenes are never reported to friends and family anyway), and you compare them to the unique elements in movies opening around the same time to see how they rank.
IMPORTANTLY, you could very easily prove your credibility in this case, because you could calibrate your statistics using Box Office Mojo, and you could publish how your model holds up in terms of accurately predicting the rankings of previous weekends.
2. Past-six-months rankings. There are two basic things we know about people who like to go see things: the first is that they like things that are familiar. The second is that they DON’T like things that are TOO familiar. I would posit that this is a statistically predictable phenomenon, and if you identified unique screenplay elements and examined their success over the course of, say, six months to a year (and likewise compared them with TV and – maybe – books during that same time), you’d see an increase in interest in certain topics that builds up as more movies come out, and then quickly tapers off as audiences get exhausted with it.
In order to accurately predict rankings, you’re going to need to know where you are in the curve for a given idea – are we on the upswing for superheroes, or the downswing? Have we hit the zombie saturation point? (I expect if you looked at viewing numbers, you’d actually start to see not a complete LOSS of viewers, but a NARROWING of viewers: so, when you consider zombies, it’s not that fewer people want to watch zombie movies or shows, it’s that people want to watch fewer zombie movies or shows – they’re here for The Walking Dead, and a NEW zombie show is probably not going to be very interesting.
3. 2.5 years out. So, if you want to know how well your movie is going to do, and you believe MY (superior) theory that it’s got more to do with what else is out that weekend than it does the historical success of any given kind of movie, you’ve got to be able to predict the movie environment in advance – and since it can take anywhere from two to three years to make a movie, you’ve got to be able to predict it pretty far in advance. Fortunately, studios often make announcements about the kinds of scripts they buy, so that’s a lot of data, and there’s a certain amount of regularity in when different kinds of movies are going to get released (prestige pictures at the end of the year, garbage in February, blockbusters in May and June, &c).
But if you want to know what to buy FIRST, before the other studios have bought anything, you’ve got to not only know what movies people are going to make in two years, you’ve got to know what people are going to buy RIGHT NOW, and fortunately for THAT model, studios are nothing if not predictable. The strategy for buying movies is and has always been the same: see lightning strike; run to that spot as soon as possible.
Your third model identifies unique elements in movies and TV this year, and compares them to the sorts of scripts that are available; the most successful kinds of movies are the kinds that studios are most likely to pick up (probably with some…well, we’ll call it Perverted Variation – movies that are chosen specifically because of how UNLIKE they are from everything else). This is another one that’s easy to calibrate, too, you just make your predictions and compare them to the purchase announcements throughout the year.
4. OPTIONAL. General ticket sales. You’ve got to use things like inflation and general buying trends to predict outcomes if you want anything like actual REVENUES for a movie (as opposed to just relative success), you’ve got to know how many people are going to buy movie tickets two years from now. I argue that this is optional on the grounds that, while the first three models aren’t VERY predictable, this fourth one is pretty much a shot in the dark.
So, you see what you’re really looking at: you’re looking at HOW PEOPLE DECIDE WHAT MOVIES TO GO TO. They think about the sort of things they’ve seen recently and whether or not they like them, and then they think about what else is available to watch this weekend, and THAT is how they pick.
There are probably (definitely) other variables in all of this, and some of them are predictable and some of them aren’t (you could probably figure out whether people are more or less likely to see a huge blockbuster in January – though you’d have to really look, because the answer might surprise you!). And, of course, with three different models based on an already problematic translation heuristic (that is, turning “movie screenplay elements” into useful statistical data) full of individual variables which are often difficult (if not impossible) to get actual data on, this entire exercise is going to be largely academic.
Nonetheless, I am willing to offer my own (SUPERIOR) script-analysis services for $15,000 per script. That is a 25% savings over WorldwideMPG, sign up today!