Stuff: The great horror unwashed, the horror canon, the talented tenth, and a little thought experiment.

Recently, LoTT-D member and all-round stellar human being B-Sol, the legend behind the myth behind the Vault of Horror blog, posted a piece on a "top 50 horror films" poll created by the Euro-based music and movie retailer HMV. For the full list, make with the clickee clickee over to his site.

The poll hasn't been well received. B-Sol suggested that the poll showed a gap between committed fans of the genre and weekend warrior types. Final Girl said the list "makes the horror snob in me want to curl up in a little fetal wad of self-righteous indignation." (She then admits that some of her all time fave flicks are a bit cringe-worthy, so there's a bit of self-aware stone tossin' in this here glass house.) Curt, of Groovy Age fame, takes some pride in having not seen half of the top ten flicks on the list. And even I, with my strong contrarian streak regarding the baffling worship of Euro-crap directors and slasher junk among League members, must admit that my first thought was that HMV's customers must every manjack one of them be utter morons.

But now I'm thinking. Kinda.

My ill-conceived review of the fifth Saw pic started with a long (perhaps overly long) gag which basically used box office numbers and the idea of a relevant horror canon to make the claim that anybody who had followed all the Saw movies wasn't worth taking seriously as a critic of the genre. This was, actually, just a joke. The idea was supposed to be that I had created an elaborate justification for having seen all the Saw flicks, even when diminishing returns had long since put paid to the series' most vital aspects. Still, at least one reader took it seriously and, I noted, that the issue he took with it was not methodological, i.e., he had no real problem conflating box office numbers with things like unique viewers or my quite unsupported claim that Saw's ticket sales represented the significant population of the horror viewing public. Rather, it was defense of his specific context as valid grounds for producing critical judgments regarding horror films. It was a dismissal of the idea that good critics need to share certain touchstone works. His problem was with the idea that there is a canon of works a critic needed to know in order to have the context from which to critique other works. UPDATE: Sean, the reader in question, clarifies his objection in the comments section. I misstate his position here. Sorry Sean.

This response has made me think more and more about a canon, whether such a thing should exist, and, if it should, how would you make it.

Which brings us back to HMV's list.

HMV's customers might very well be morons. But their taste, or lack thereof, is not responsible for how unsatisfying the list is. The problem is not lack of context, the absence of critical faculties, the dismal state of modern horror filmmaking, or mainstream disrespect for the genre. It is just math.

The issue isn't that young people today don't watch good horror movies. Or that young, modern horror fans aren't familiar with the history of the genre. The issue is simply one of sample size and population demographics. B-Sol has proposed that selected horror bloggers should propose an alternate list made from the weighted votes of their collective top ten lists. The idea is, I guess, that the superior genre context of the bloggers will produce a more interesting, nuanced, diverse, and worthwhile list. But the HMV list doesn't show a lack of context. It just shows what happens when you expand your sample size.

Compared to any "grassroots" list that bloggers could cobble together, the HMV list pulls from an absolutely huge sample size. Consequently, any idiosyncratic tastes will get averaged out and only those films that have been seen by a huge number of recipients will even have a shot at getting on the list. This is going to make the skew heavily towards movies that came out fairly recently (within the last 20 years or so) and towards movies that, in a decade or two, will probably drop off the list.

"Explain yerself, CRwM," you might well demand.

Okay. Let's do a little thought experiment to test my point.

Statement 1: There have been fewer films released during the lifetime of any given film viewer (the set x) than are available to be watched (the larger set A of all horror films available, including subset x).

Assumption 1: To vote something as one of the best horror films, the voter must have seen it.

Statement 2: Given assumption 1, in order to qualify to make the best of list, the film must be seen by a significant proportion of viewers – i.e., unless many people have seen it, a film cannot get many votes.

Assumption 2: Being a film viewer includes watching new films.

Definition 1: New films are defined as films that come out in film viewer's lifetime.

Statement 3: There is a limited number of films that one viewer can see in their lifetime. If assumption 2 and definition 1 are correct, then some portion of those films must be from subset x.

Statement 4: As the population of the voters increases, the number share titles in the subset of x for all viewers decreases. This is because the number of shared titles among all viewers is bounded by whatever voter has seen the least. For weighted voting, the number of films that are almost shared in the subset of x for all viewers remains important, but that number still shrinks as more voters are added.

Statement 5: The forward motion of time and the work of human biology ensures that the shared subset of x will always consist mainly of newer films, as new viewers inevitably watch the limited selection of what is new (as defined above) and distribute the remainder of their limited amount of viewing time among an ever-increasing pool of set A flicks. NB: The further you get from the birth of film, the less statistically significant the viewing of any single film outside of set x becomes, as the set of A but not x is constantly growing.

From these premises we get the following three conclusions:

Conclusion 1: Even if the films viewed are distributed randomly over set A, with the condition that the number of films in set x must be equal or greater than the number of viewers, the results are that film viewers are probabilistically likely to have shared more films in the subset of x then they are in larger set of A.

Conclusion 2: Given assumption 1 and statement 2, it is more likely that films in the subset of x will make the list then films from the subset of A but not x.

Conclusion 3: As new viewers shift the shared titles of subset x ever forward in time, films that enter the larger set of A will begin to experience the same sort of viewing patterns that most set A films experience and are, therefore, likely to fall off the face of the Earth.

If that's unconvincing, you can run a little simulation using easy-to-manage numbers. Take a set of whole numbers 1 to 1,000. Take 10 volunteers and ask them to pick 50 different numbers, but tell them that at least 4 of these numbers must come from the set of numbers 1 through 20. Then tally up the votes each number got and find the list of "favorite numbers."

In this simulation, the whole set of numbers stands in for set A and numbers 1 through 20 equals set x. Picking them is the equivalent of having seen them, so there's actually a step missing from the simulation. What this shows is not how voting would go, but how the actual set of films that have a shot at the list is whittled down. Still, given assumption 1, you can assume that the films watched – or numbers selected in the simulation – will lead you to the films voted for.

Admittedly, this simplifies viewing behaviors. There are certainly people out there who never watch horror movies and folks who never watch anything but old horror movies. Still, I think it is fair to assume that these are outliers in film watching behavior. Somebody who never watches horror films wouldn't spend time messing with a list of the best fifty horror films. And the population of people who never watch anything that couldn't be defined as "new" by our criteria is tiny and will tend to vanish because their votes, cast adrift the in ever-widening set A, will do little to off-set consolidation trends. I also think the ratio of new to all available films is pretty generous, but we're trying to keep this workable and, in the model's defense, the generous ratio actually works against my theory - the fewer new films you have in comparison to the set of all films, the faster one should see vote consolidation.

Given a small sample size, you'll actually get a large amount of diversity on your final list. This is because, with a small population of voters, the threshold for entering the "best of" list is tiny. If anything outside of set x got more than one vote, it still has a shot at getting in the top 50.

But, increase the sample size and you'll notice that the selection of numbers in set x, the "hits" on the subset, begins to rapidly outpace hits for numbers outside the subset. This doesn't mean that people aren't picking things outside set x. In fact, they might be selecting tons of numbers outside the subset. The problem is that set A is freakin' huge! It swallows up votes. Furthermore, so long as there is the requirement of selecting some films from the restricted subset of x, the votes are going to consolidate in the subset. Even if you reduced that number to 1 or said that every one hundredth selector could ignore that rule, you'd still get consolidation as the sample set grew in size. Weighting, another common strategy when building a canon, eventually gets overwhelmed by sample size issues.

That all supports conclusion 1 and 2. What about conclusion 3? To test that, add a "new generation" of horror films and fans to simulation. Add a new set of numbers to the model: prime 1 through prime 20. Now add some more selectors: Group N. All the previous selectors (group O) can select any ten numbers from the set of prime 1 through prime 20 and 1 through 1,000 – but they must choose 4 numbers from the subset of prime 1 through prime 20 and 1 through 20. As for the new group N, they can choose any ten numbers from the full set, but 4 of those numbers must come from the subset of prime 1 through prime 20. As each generation gets added, votes will cluster towards the new. And, of course, eventually generations of viewers will pass away, leaving some numbers completely out of the subset.

Again, this simplifies the whole generational process, which in RL overlaps and staggers forward. Still, it gets across the same idea.

What does this all mean for B-Sol's list? That the horror snob (his term, not mine) list will essentially be bounded by the same criteria, it will simply benefit from a tiny sample size that will allow for more idiosyncratic, less widely supported films to make the list. In short, films that fewer people like will be given more weight.

Will a horror snob's list of best flicks be more interesting to a self-professed horror snob? Probably. Will it be more diverse and include better films? Maybe.

Personally, I suspect that our list will be a little heavier on slasher stuff and Euro horror. Not because these flicks are in anyway superior to what made the HMV list, but because the personal-blogger demographic is relatively old (anybody not currently in their first year of college thinks of blogs as team-written link dumps, usually tied to a corporate media entity or driven by ad revenues – they aren't venues for personal expression). The population of horror bloggers will have that stuff in their subset x when large swaths of the HMV population probably didn't.

If we wanted to make it interesting, we should create a control to limit the statistical accumulation in subset x for each voter. We should adapt the Copernican principle that says, when grossly simplified, "you are not special." Statistically speaking, your viewpoint on a system is unlikely to encompass a significant juncture. You are most likely one of those 95% of people with nothing particularly interesting in your subset x. That is to say that you're unlikely, for example, to be lucky enough to see the debut of the first horror film, catch the greatest horror film, suffer the worst horror film, or be at the last screening of the last horror movie ever shown. In all likelihood, your subset x is pretty mediocre. Don't feel bad, mine is probably crap too. Almost everybody you know has a crap set x. It is not your fault. Blame Copernicus. Damn you, Copernicus! Don't think we'll forget this, stargazer!

Anyway, to capture data that might escape the wasteland of mediocrity that is our subset of x's, we are forced to attempt to avoid the tendency to come strictly from our singular perspectives. We can't vote on movies we haven't seen (well, we could, I guess, but that's stupid), but we can at least complicate the link. B-Sol could have put the condition on voters that they could only vote for movies that debuted before the voter's lifetime. You'll still get some consolidation, but it would be less of an influence. Instead, you'd be asking voters to choose those films that aren't "special" to them, but rather indicate the shared common past of all voters, a shared common past that endures the constant forward creep of vote consolidation.

That, I think, might be as close as you could get to a popularly constructed canon.

YET ANOTHER UPDATE: A commenter on Final Girl has claimed that HMV poll was really more of a ranking exercise as you have to choose from a limited slate HMV selected films. The criteria behind slate selection seems to have been to drive sales for new editions out on DVD. This basically makes the poll useless as measure of anybody's actual tastes.