(NOTE: Updated and Expanded)
by Marshall Flores
Welcome to the final episode of Awards Daily’s Statsgasm for the 2013 Oscar season. We are now in the home stretch, and it’s time for everyone to fill out their ballots and enter their pools. If you haven’t done so already, do enter AD’s Predict the Winner Contest – as always, there will be prizes and bragging rights for the best predictor!
Before we unveil the final slate of predictions made by our Statsgasm models, we would like everyone to keep in mind three things:
1. Correlation is *not* causation. AD’s forecasting models are first and foremost built to predict outcomes as accurately as possible, not explain them. We really can’t make strong assumptions or inferences on how AMPAS voters behave because, outside of a few anecdotes and polls from voters self-selecting to talk under anonymity, we simply don’t have any good data on them (what they like, what they know, what they consider when voting, etc.). And we’ll probably never know any of that.
That being said, the strongest predictors in our models tend to be guild/BAFTA outcome-related precursors, and the reasons for this do make sense intuitively: these groups, with their overlapping membership with the Academy, reflect AMPAS to a limited degree while also acting as its official “gatekeepers.” So we could apply a certain amount of causative analysis in some cases.
2. We *don’t* expect AD’s models to bat 21 for 21, and neither should you. Trends are often broken, history gets made, statistics is never having to say you’re certain. These models make forecasts that may be correct over the long run, with a large enough sample. But in the long run we’re all dead, and the Oscars are a yearly one-and-done deal. I know I’m making predictions in a few categories that are contrary to what the models are indicating.
But I’m regularly checking and validating the models, researching new methods that could help build better mousetraps, considering other predictors which could be significant. All while trying to keep it as simple as possible. As Billy Beane says to his Oakland Athletics in Moneyball, “It’s a process, it’s a process, it’s a process…”
Similar prediction models (devised by smarter guys than I) have batted for 75-80% – on par with the typical Oscar pundit. I’ll be more than happy if our models can match that overall performance in their rookie season. Anything above par is just sweetener.
3. With limited data, math can only do so much – Intuition, creativity, a strong passion for film + Oscars history are keys to making this type of applied statistical analysis enjoy any successl in the long-term.
As far as I’m concerned, I had the best teacher one could ever have, a teacher who absolutely taught me everything I know about the Oscars and more. That teacher is no other than our patron saint of lost causes here at Awards Daily, the Dr. Frankenstein who more or less invented the Internet Oscar-watching industry when she founded Oscarwatch.com nearly 15 years ago. Hopefully enough of her insights and perception have been imbued into these otherwise soulless constructs. This work simply wouldn’t exist without her.
Tribute aside, let’s take a look at our final batch of predicted winners, starting with the major and feature categories. I will provide some commentary as well:
Best Picture:
The BP model reflects the narrative of this season being a hotly contested contest between Gravity and 12 Years a Slave, with American Hustle lurking around as a possible spoiler. Gravity does have a lead in the model, a result of its DGA win (which remains the strongest BP predictor historically). However, its lead is by no means an insurmountable one – if I were a betting man, I’d definitely take the approximately 3:1 odds against winning for 12 Years a Slave.
On a more personal note, either 12 Years a Slave or Gravity would be incredibly worthy, excellent BP winners that should survive the test of time. Both are ground-breaking masterpieces. Concurrently, *not* winning BP will absolutely not affect either film’s ultimate legacy in my eyes. The Oscars may be a zero-sum popularity contest, but our own love of cinema should not have to be.
Best Director:
Statsgasm is predicting that there’s an 83.73% chance that Ang Lee will no longer be the only non-Caucasian Best Director winner after Sunday night.
Like with Best Picture, either Alfonso Cuaron or Steve McQueen winning will be a momentous, history-making event for incredible, masterful work – something that we should all be celebrating. There is still a lot left to be desired regarding the Oscars and diversity, but progress, slow and frustrating as it may be (especially with an insular old boys club like AMPAS), is progress.
Best Actress:
All hail the great Cate Blanchett – Statsgasm’s biggest lock of the night among the major categories.
Best Actor:
What was initially supposed to be a very competitive category at the start of the season shifted rather quickly to Matthew McConaughey’s favor in the span of a couple of weeks with his Globe, BFCA, and SAG wins. As such, he’s the odds-on favorite. Not a lock by any means, but he does have a solid lead, and McConaughey winning would cap off quite the career revival he’s been having over the past two years. Dallas Buyers Club being MIA at BAFTA shouldn’t really affect his chances, as BAFTA hasn’t actually had all that great of a track record in Best Actor since 2000, being a total non-factor in Denzel Washington’s and Adrien Brody’s upsets, missing on Sean Penn twice, missing on Colin Firth in 2009.
In fact, I’ve had multicollinearity issues if I attempt to include the BAFTA in this particular model, i.e. the model would end up estimating BAFTA with a negative coefficient, which basically means that winning the BAFTA somehow *reduces* the odds of winning the Oscar. So I’ve elected to leave it out completely. I suspect many of you will be aghast by this revelation, but this is yet another example of how statistical analysis has its limitations. In the end, quite a bit depends on the abilities and judgement of the model builder.
Statsgasm’s Best Actor model does factor in previous acting nominations and wins, which gives Bruce Dern and Leo DiCaprio a slight boost while also penalizing Christian Bale’s chances. As for Chiwetel Ejiofor, I do believe he has an outside chance of “pulling a Brody”, and it would be an equally amazing moment. But there are some key differences between then and now; most importantly, Brody was against 4 previous winners and The Pianist was riding a serious jolt of momentum with its surprise BAFTA wins in Best Film and Director.
Supporting Actress:
Although the model is giving Jennifer Lawrence a sizable lead over Lupita Nyong’o on account of her Globe + BAFTA wins (a very potent combo that has batted 1.000 for Oscar in this category since 2000), this is the first category in which my prediction (Nyong’o) will diverge from the model’s. But I also wouldn’t completely rule out the possibility that Lawrence and Nyong’o end up splitting support from younger voters, giving Squibb an avenue to the statuette if older voters unite behind her.
Look for this to be a *huge* early inflection point with regards to the BP race on March 2nd – if Nyong’o wins, 12 Years a Slave will be on the traditional path to victory.
Supporting Actor:
Jared Leto has a very sizable lead on BAFTA surprise winner Barkhad Abdi. Much of this lead is a result of his Globe win, which is the strongest predictor in this category and is estimated to improve the odds of winning the Oscar by a factor of 43.
Adapted Screenplay:
This will very likely be a guaranteed miss by the model – 12 Years a Slave should have this in the bag despite not winning either the WGA (which it was ineligible for) or the BAFTA. But then again, most were predicting 12 Years to win the BAFTA too, only to see Philomena ride a wave of popularity and end up winning.
In terms of pure adaptation, as someone who read Solomon Nortrup’s harrowing memoir in my AP US history class in high school, 12 Years gets my vote as the winner, hands down. That being said, it’s a fine set of nominees.
Original Screenplay:
The model accurately reflects the dead heat between American Hustle and Her among the experts and pundits. Her has beaten Hustle head-to-head twice this season at the Globes and the WGA, but AMPAS is a far larger body with a sizable acting bloc that has adored Hustle by all accounts. That reason alone nudges me into predicting Hustle, despite my (strong) preference for Her.
If you’re stuck on making a decision here, save yourself a headache and just flip a coin. 🙂
Animated Feature:
Documentary Feature:
A very tight race through and through. All 5 are excellent nominees, and there were plenty more that were unfortunately left out of the lineup, especially Stories We Tell and Blackfish. My preference is for the absolutely horrifying and surreal Act of Killing, but I do feel Act, The Square, and Dirty Wars will all split votes from one another, enabling the more populist and undeniably rousing 20 Feet from Stardom to win.
Foreign Language Film:
I will be perfectly honest: the precursors in this category have been very poor overall in predicting Oscar since 2000. As such, this is the one model out of the 21 Statsgasm models in which I am not all that confident in regarding its construction.
Still, The Great Beauty did win the Globe and the BAFTA, defying the consensus picks to boot. It’s a marvelous, dazzling film, and I’m comfortable in sticking with it. But The Hunt and (especially) Broken Circle Breakdown have excellent shots at winning as well.
That’s all for now. Return tomorrow for Part 2 of our final Statsgasm predictions in the tech categories.
Happy predicting!
Email: marshall(dot)flores(at)gmail(dot)com
Twitter: IPreferPi314
🙂
Marshall et al. might have the right mathematical formula, but choosing the right variables for each category is just as if not more important.
To build an effective model, I think you have to be an astute observer of the Academy history. You have to understand the nuances and the psychology of the AMPAS’s voting pattern. You have to keep an open mind.
Yup, your numbers (and my stats, which had 12 Years as the clear mathematical favorite all along) were better than Marshall’s. But his model couldn’t POSSIBLY be improved upon with the help of any of the stats we used, even though they did much better… 🙂 No, his model – which he himself admits, however, CAN be improved – can only be improved by whatever stats HE deems fit to use, because he’s the math genius, right?! His theoretical knowledge is better than ours, so he must be right about everything else too!
(I’m sorry, Marshall, but since you don’t think us worthy of an explanation as to why you can’t/won’t include all/any of the stats I mentioned in my earlier post in your model, you’re opening yourself up to such criticism – explain, and I will gladly withdraw my suggestions, if they are mathematically unusable. Don’t give me homework instead, if you want to be taken seriously!)
Thanks for the compliment – I return it with equal confidence!
I did not win anything, though. 🙂 Makes sense, I guess, that, out of 800+ participants, enough (13) would get perfect scores, 21/21, with so few upsets in play. Besides, there were plenty with 20/21 as well, of course.
I’m not unhappy either – I got BP right again, as always. That’s the one I really care about. My stats held up (in a very competitive year) and proved once more that the SAG and Screenplay nominations ARE vital, no matter the reasons various movies are snubbed. Plus, my favorite movie won and I really enjoyed the ceremony. 🙂
And it’s not like the prizes were particularly impressive (a Gravity DVD and some t-shirts and pens and such for 1st place and, obviously, not much better for the rest). Won’t be losing any sleep over it…
I hope you won something. You’re too intelligent of an analyst to walk away with nothing. I decided not to participate in the various contests this year because my passion for 12YAS winning was sufficient suspense for the evening.
But I am satisfied that the two categories, BP and BSA, that I decided to do the math for. My numbers turned out to be better than Marshall’s, right? 🙂
Again, thanks for making me reconsider that one, Alan! It would have turned out the same anyway (1 out of 2 right), as it seems I was incapable of picking Nyong’o/Her, because I didn’t believe in AH getting nothing, but still…
So, I got 19/21 (missed Original Screenplay and Editing) in the contest I was most interested in (they didn’t have the short categories). I don’t know what prize I’ll get, as there weren’t many surprises this year, so who knows… But last year 19 was good enough for 1st place there. I doubt it will be this year, but I imagine I’ll get one of the other prizes.
Here at AD I went with the Gravity scenario (Picture, Editing, Production Design etc., plus Lawrence/Her and U2 for Song), as insurance in case my favorite, 12 Years, did in fact end up losing BP (and other stuff). Thankfully, that didn’t pan out, so I only got 18/24. I’m curious how many the winner will get. I imagine it’ll be 22 or 23.
Very good arguments for screenplay! Yeah, I’m quite unsure too. I could EASILY get both of these wrong (this + supporting actress). I’m really just going with the scenario I think is slightly more likely right now, stats-wise. I have little confidence it’ll go this or that way…
I’m cool with that. Besides, my stats and reasoning already pointed toward Lupita. If Jennifer wins, it would be a suprise, not an expected event.
It’s close enough for an “upset”; and the reason for that would be June Squibb taking votes away from Lupita because she’s another alternate to Jennifer.
Screenplay for AH is reasonable, but there’s also compelling argument for “Her”. I’m really unsure about this category. With 10 nods, it’s hard to see AH walking away with nothing, but perhaps voters will spread their support all over 10 categories. There isn’t a consensus category to focus on. But screenplay is the place for it. This is the consolation prize for BP co-frontrunner(s). It would have gone to Gravity if that film were strong enough for a nomination. Since AH is next in line, it would go there.
Look, you might just be disappointed if I tell you what it is anyway. You might think it’s highly circumstantial. I happen to believe that the rather strong mathematical relation between this particular stat and Nyong’o being the favorite is real and relevant, but it’s probably something easy to dismiss by someone who feels that the cause-effect relationship is demonstrable – it is, I think, but my reasoning is certainly questionable and, in part, based on personal evaluations and interpretations.
Anyway, I didn’t need much after you added your stat (which is also very good), seen as the race is so close in this category, and, with this one I discovered on my own, I feel it’s enough for me to switch – even though, don’t get me wrong, my intuition hasn’t changed and is still strongly telling me that Lawrence will win. But I have to go with the facts, now that I believe they’re compelling enough to make Nyong’o the (slightly, but definitely) preferable bet. The stat doesn’t completely rule out Lawrence as the winner – there are precedents (percentage-wise, about the same as the precedents for your stat) -, but it does make her look a far less likely winner.
Plus, like I said, if I do reveal it, and enough people think it is relevant, like I do, it could conceivably affect the stat itself in the future. Perhaps even short-term… I really don’t want to do that, as I worked hard to find this, I spent hours centralizing and comparing data and I would hate to have that work be rendered useless in the future due to my not having been able to keep it under wraps when I should have.
Of course, I’ve also switched to American Hustle for screenplay, partly based on the same study, and partly because, like I said, I definitely don’t think AH will walk away empty handed.
I hope you won’t think I’m being unreasonable – you’ve probably read some of my posts and you know I’m completely open about 99% of the stats I work with (probably too open, actually, thus inviting much unnecessary criticism), so you can imagine that, if I didn’t actually think it was substantially better to keep this one to myself, I would tell you in a heartbeat!
I hope I steered you in the right direction. I am just dying with curiosity with your revelation. Can you reveal it privately?
If not, I can live with mystery hanging. 🙂
I can’t reveal it. I think there’s a good chance it might be detrimental (to various people, myself included) in the long run.
But yes, you’re right, I do owe you if Lupita wins! If you hadn’t made the case for her like you did, I probably wouldn’t have reconsidered and might not have checked what I ended up checking.
But I’m not saying she’s a lock! She could still lose. What I discovered in my analysis just makes me think she’s the (clear) favorite again. And your statistic gives me extra confidence, of course…
Will you at least reveal this after the show? If Lupita wins, you owe me. 🙂
OK, I’ve done another study (I’d rather not reveal). I think it’s powerful enough for me to conclude you’re definitely right. Nyong’o is the best bet for Supporting Actress.
By ‘on their next nomination after winning Lead’ I mean that they weren’t nominated for Lead again after winning that Leading Actor/Actress Oscar, because that’s what we have with Lawrence and I think it makes a difference.
Interesting… So 2/12 won Supporting on their next nomination after winning Lead (which is what Lawrence’s situation is) – I checked. That’s not a huge sample size, plus the fact that, for consecutive years, there is only Jennifer Jones. I don’t know what to say… Is it conclusive? Isn’t it?…
The thing is, I kind of have to predict American Hustle for something (due to other, stronger stats that say it won’t go 0/10, and I personally think it’ll get at least 2, most likely), and I just don’t think it’ll beat Gatsby for Costumes. So I have to predict either Lawrence, or the screenplay. I have more confidence in Lawrence beating Nyong’o than I do in Russell beating Jonze, but it’s definitely close. I’m not sure which way I’ll end up going…
And even if AH does get Costumes, that’s not a profitable prediction to make right now. If I predict that and am wrong, too few other people will be wrong with me. That strategy is OK for AD’s contest, where only 1st place gets a prize, and there isn’t much of a chance of getting it unless you predict enough stuff that most of the others didn’t predict, but there’s another contest on this other site where there are far more prizes, and, therefore, I have to go for maximum expected value in every category there. And others will be predicting Lawrence, but not as many will be predicting Hustle for Costumes (at least that’s what I think). Plus, like I said, I think it’s getting two, most likely, so, out of Screenplay and Lawrence, one is nearly guaranteed. Plus, I think Gatsby is far more of a favorite for Costumes than Lupita/Lawrence for Supporting Actress.
If nothing was at stake, I’d say Lawrence in a heartbeat. On this, my intuition is clear.
Forgot to add. Waltz’s supporting actor rivals last year ALL had won the Oscars before. Note how he beat the two BA Oscar winners in Hoffman and the perceived frontrunner DeNiro.
This year there are 3 first time nominees in Lawrence’s category.
Claudiu,
Right. Both are supporting. Jason Robards won twice in consecutive years. FOR SUPPORTING. Academy can be generous with the supporting Oscars if they hadn’t given you the lead.
Had Christoph won as leading actor for IB, trust me, he wouldn’t have won supporting again for Django (in Django he was on screen for a whopping 97 minutes, making his win very suspect, and in some way it was like graduating to lead). Best leading Oscar is a BIG DEAL. It’s like once you’re crowned a queen, you’re not going to get crowned again as a princess.
Actors who won their subsequent Oscar as supporting have many years in between, it’s as though they are different actors playing a very different character in their old age. And in some cases it’s a tribute to their long career.
Ingrid Bergman (1956 & 1974)
Helen Hayes (1931 & 1970)
Gene Hackman (1971 & 1992)
Jack Nicholson (1975 & 1983)
Maggie Smith (1969 & 1978)
The following actors/actresses who received Oscar nods for supporting after winning for lead(s) and didn’t win:
Meryl Streep
Lawrence Olivier
Marlon Brando
Jane Fonda
Robert DeNiro
Robert Duvall
Alec Guinness
Maximillian Schell
Paul Scofield
Jon Voight
William Hurt
Anthony Hopkins
Geoffrey Rush
Phillip Seymour Hoffman
Sally Field
Kathy Bates
Holly Hunter
Frances McDormand
Helen Hunt
Jennifer Jones **
Jennifer Jones was also nominated for supporting the year after she won for lead. So take their first name as an omen. 🙂 Also, I’m uncertain that the above list is exhaustive. I might have missed a couple.
And another actor to be added to this list (aside from Lawrence) after this year is Julia Roberts.
“That may be intuitively appealing, but theoretically and practically it is not the case in regression analysis. Issues of overfitting, loss of degrees of freedom, and multicollinearity are some of the legitimate issues that can/will result from using too many variables in a regression-based model. I won’t go over what those terms mean here, so I would invite you to read up on these by yourself.”
OK, I believe you. 🙂 I just wanted a quick example of how one of the aforementioned variables does that to the model, that’s all. But if it’s not easy to give one without my having studied the theory of it all intensively, I’ll not insist…
“I know it’s tedious thinking, but Academy is quite stingy with their acting Oscars. They’re not going to reward you with a lesser prize when they had already given you the ultimate prize the year before.”
I’m too lazy to check (unless there’s a table somewhere), but how many times has an actor/actress even been nominated for supporting after just having won for lead? Can’t be too many times…
“JLaw would have to do something extraordinary, play a completely different kind of role convincingly, to win again. If they are going to give you another Oscar, they want *more* from you, not less.”
Christoph Waltz. (I know, I know, both supporting – so what? It’s basically the same performance.)
^ I know it’s tedious thinking, but Academy is quite stingy with their acting Oscars. They’re not going to reward you with a lesser prize when they had already given you the ultimate prize the year before.
JLaw would have to do something extraordinary, play a completely different kind of role convincingly, to win again. If they are going to give you another Oscar, they want *more* from you, not less.
Claudiu,
I’m aware of that pattern/stat. But none of the winners had won the year before. A few actresses have won before, a few to many years before, but their latest win was a lateral win, not a downward win. Downward wins are rare in Oscar history. Consecutive year wins are always lateral wins.
American Hustle will be in great company with Sunset Boulevard and My Man Godfrey.
That may be intuitively appealing, but theoretically and practically it is not the case in regression analysis. Issues of overfitting, loss of degrees of freedom, and multicollinearity are some of the legitimate issues that can/will result from using too many variables in a regression-based model. I won’t go over what those terms mean here, so I would invite you to read up on these by yourself.
“No, it is not so obvious. There are theoretical *and* practical rationales in attempting to adhere to Occam’s Razor when model building. Bigger is not always better (and can often actually be *worse*). If two models lead to the same predictive performance, the simpler one is almost always preferable. Again, it’s up to the builder’s judgement.”
Well, I meant as long as there are no damaging effects to the model. That was my claim, really – that all of the variables in question (screenplay/director/editing nomination, WGA nomination, PGA/DGA/SAG win, DGA/PGA/SAG nominations) can only improve the accuracy of any BP predicting model, overall, no matter the approach, because they all have outstanding matching records, they rarely, if ever, conflict with each other, and don’t reduce the sample size in any way.
And, again, I’m not saying you should include stuff just for the sake of it. I think all of these are necessary – each of them helps make sense of various (otherwise inexplicable) BP defeats/wins in recent years, while not taking anything away from the winners in any of the other years. I get the Occam’s Razor (I hate that name) principle; I just think that, when it comes to BP, the ideal model needs to include all or nearly all of these. I might be making some simple error in judgment here, missing some simple point, due to my not having tried to actually build a model myself yet – there might be some practical reason for not including some of these -, I don’t know; but this is how I see it. I’d gladly admit to being wrong, if someone would just point out what this error is. I’m always eager to learn.
@Alan D – I’m predicting Lawrence now, actually. 🙂 I think it’s too close to call (the stats mostly confirm this), but my intuition is telling me Lawrence is the more likely name to be announced on Oscar night, and Lupita the more likely also-ran, so I’m going to go with that. I might be wrong, but I strongly feel this is how it’s going to go down. Also, if you look here http://en.wikipedia.org/wiki/List_of_films_with_all_four_Academy_Award_acting_nominations, you’ll see some interesting patterns that, I think, favor a win for Lawrence. It’s not clear enough for me to count this as actual evidence, but a win for Lawrence fits the patterns there significantly better than a loss, if you ask me. Just saying…
Best Supporting actress, Jennifer Lawrence? ARE YOU FUCKIN CRAZY?
And thanks, Bob! I’ve really appreciated your substantive and incisive feedback on my work. Always nice to talk with someone who can speak the same language. 🙂
I had considered using a Bayesian approach to Statsgasm, although admittedly I was never really that comfortable with it in my classes. But then again, I was skeptical about the utility of using *any* formal statistical methods for Oscar predicting up until last summer. I’ll definitely consider integrating William Bayes again if Statsgasm gets renewed for another season.
(And if you’re wondering where I fall on the frequentist/Bayesian debate, although my demeanor/perspective leans Frequentist, I’m a utilitarian pragmatist lol. I’ll use anything if a) I’m comfortable with it, and b) I’m comfortable with its ability to explain the data I observe).
That being said, Nate Silver is a huge Bayesian and the models he has used to predict Oscar haven’t really performed any better than the simpler logistic regression models I’ve seen being utilized.
thank you for the response, Marshall– I just wanted to see what your rationale was, and it seems acceptable to me (I understand that you’re trying to write for a more general audience and so wouldn’t want to explain all of this in the actual post). And of course you’re right, it’s not the case that one should include everything just because– the tolerance/VIF is one reason you would want to remove a variable even when the goal is overall model fit. Did you look at what happened when you kept the BAFTA in but removed something else to address the issue? I’m guessing you did and saw that the model was less accurate…
Well good luck anyway, I hope it performs well! I have an inkling though that a Bayesian approach is likely to do much better at this whole enterprise…
No, it is not so obvious. There are theoretical *and* practical rationales in attempting to adhere to Occam’s Razor when model building. Bigger is not always better (and can often actually be *worse*). If two models lead to the same predictive performance, the simpler one is almost always preferable. Again, it’s up to the builder’s judgement.
And Claudiu, I appreciate your input. If you feel your approach is superior, then more power to you. But I’ll stick with applied regression analysis.
Give ’em hell, Claudiu! 🙂
I have numbers for the competitive Best Supporting Actress category. I forgot to bring my Excel file with me earlier so I had to wait until I get home.
By simply plugging in the season’s precursors, Jennifer Lawrence led with a 38.5% chance, while second place Lupita landed with a 34.5%. I thought to myself, it’s almost a tie and Lupita should win the Oscar because JLaw just won last year, then I thought why haven’t I include this variable into the model?
Looking back in the last 18 races, not only all the winners were first timers, a majority of them were even first time nominees. There’s a “ingenue” quality to this category that is unique from the other acting categories. All other acting categories have repeat winners in their category. The odds are against you if you won, but in Supporting Actress category it’s particularly even more difficult.
I then realize I should include variables like age, being in the Best Picture nominee, Best Picture frontrunners, so on. So the end result is as follow:
Lupita Nyong’o – 33.8%
Jennifer Lawrence – 26.8%
June Squibb – 21%
Sally Hawkins – 11.3%
Julia Roberts – 7.2%
The adjustment didn’t improve Lupita’s so much as it really negatively affected JLaw’s chance, with Squibb and Hawkins benefiting from JLaw’s dip.
“if the model fit is improved, you should just keep all of the variables— you’re not trying to make any claims of significance.”
Obviously.
@datSTATISTIKSDoe:
Theory isn’t everything, you know… That’s why it’s constantly evolving, in every field. Rest assured that studying whatever it is you studied to get so ‘knowledgeable’ about statistics and how they work is NOT as hard as you probably think it is, and doesn’t make you some sort of genius. Smart people can still quite easily figure out the basic inner-workings of statistical analysis without needing to have actually studied the subject scholastically. Plus, it isn’t exactly the most complicated branch of mathematics.
So, I’m sure you know your theory quite well, but you’re definitely applying it wrong here. The variables we’ve been bringing up (except for critics’ groups) ALL WORK TOGETHER without having to reduce the sample size. I prefer to work with the SAG era, 18 years, which might not be a good enough sample size for you, but is plenty fine for me and produces pretty much impeccable results. I have a lot of variables that complement (and pretty much never contradict) each other, and never ever make the issue messier, instead of clearer, like you’re suggesting. And no, I do NOT include BAFTA. Why should I? It’s not insignificant, even for BP alone, but I simply have no need for that variable. 🙂
I’m talking almost exclusively about predicting BP because I don’t really care about the other categories as much as I do BP – they’re not as easily predictable and don’t really interest me as much either, except for, at most, 1-2 weeks each year.
I believe you know what you’re talking about, theory-wise, but, when it comes to BP stats, you’re way off if you think using fewer variables is better (except for critics’ awards which, indeed, shouldn’t be used at all, most likely). That theory is so easy to disprove empirically, it’s not even funny! If you want to challenge this point, you’re welcome to try! The variables I use (there are about 10) don’t negate/contradict each other at all. And the explanation is that most of them are in the 90-95% matching range. They just hardly ever fail, and rarely even disagree enough to cause any kind of conflict. I know this, because I’ve been using them (successfully) for a long time.
Again, I’ve not compiled an actual model, but I have a table that I use for my BP predictions very much like a mathematical model would use the same data. It’d probably be much easier than I’ve been assuming to convert it into a model, come to think of it…
P.S.: If you’re going to claim I don’t know what I’m talking about, please have the decency to take some time to actually explain yourself, no matter how superior you consider yourself to be. I always have, whenever I’ve made any kind of claim on this site.
two thoughts .. J Law was much better than Lupita and people are staring to acknowledge it ( Sally hawkings was better than both).. I finally watched Gravity and it was meh.. A special effects marvel indeed but it was meh.. People hate Hustle but love Gravity seriously? I’m now convinced that 12 years was best picture of the year( Chiwetel and Fassender were the stand outs)..
Well, the only times I see Wind Rises beating Frozen head-to-head was with the critics awards. Critics, for better and for worse, don’t vote for Oscar.
But in any case, more power to you Kevin, and you certainly have my sympathies. I’m sure I’d be strongly rooting for Miyazaki myself to win if I’d seen Wind Rises, although my prediction wouldn’t have changed lol.
Marshall,
You’re most likely right and the stars are pointing towards “Frozen,” but there’s a couple variables that are interesting still. First there were multiple times “The Wind Rises” popped up in the Foreign Language category, thus taking the two out of competition (like the Globes). Secondly is that ever time the two films were in competition, “Th Wind Rises” either tied or outright beat in, and in those scenarios “Frozen” rarely won. So call me optimistic, but I think there’s going to be a surprise come Sunday.
I’ve learned about Cuaron’s racial background? Have you? Has Marshall? Before calling him non-Caucasian?
That’s your version of Mexico. That’s not my observation.
Don’t call my logic stupid because I haven’t said anything about McQueen or Cuaron winning the Oscar. I was simply calling Marshall’s identification mistake.
WONK ALERT
Predictive accuracy of course is a primary concern, but as I hint at later on in my opening, I’m also interested in Occam’s Razor-optimality as well. I’ve had to code in so much data and possible predictors – if I can achieve more while keeping track of less, I’ll take that route whenever possible.
Anyways, If I enter the BAFTA into this model, I do get a negative coefficient *and* the p-values blow up for it and the rest of the predictors. The VIF test also confirms this multicollinearity. In the end, I feel the BAFTA was extraneous in this model and chose to eliminate it. But that absolutely doesn’t mean the BAFTA is insignificant, period.
In any case, the model was still able to correctly “predict” 11 of the past 13 winners even without the BAFTA being a factor. Its two misses were the Washington and Brody upsets, which precursors could not anticipate at all.
I will certainly reevaluate the Best Actor model, in addition to the other models, if Statsgasm returns next year.
“Cuaron is Mexican born. But he is *not* non-Caucasian, not anymore than, say, Pedro Almodovar. Neither are his fellow filmmaker friends Innaritu and del Toro.”
“Cuaron is WHITE, y’all.”
How do you know this, you silly individual? Just by his appearance or do you know his family tree going back to the Spaniards that you claim are his ancestors? Or he might (in all likelihood) be a mestizo with fair skin (there are millions in Mexico and Latin America) which by definition, makes him a non-Caucasian. I swear, this fool makes all kinds ridiculous claims. Have you really been to Mexico? Are we allowed to call him a Latino?
“If you’ve been to Mexico, you will notice that about 1/6 of Mexicans are of pure European descents”
And Alfonso Cuaron is one of them how? Because Guillermo del Toro is -according to Wikipedia- and because you say so?
“Well, at least Mexico kept the colored people alive and make them economically second class citizen. In Argentina, they killed them all.”
Did you ever read about the transformative Mexican revolution, did you ever read one of the most beautiful books about Mexican identity “El Laberinto de la Soledad”? Aren’t you aware that Mexico is the most racially pluralistic country of Latin American, unlike places like Peru. And by your stupid logic would an African American winning Best Director be less significant because his parents were upper-middle class?
Hi, Kevin, The Wind Rises is one of 4 nominees I’ve yet to see before Sunday and I’m very sad by that because I’m a *huge* Miyazaki/Ghibli groupie. So I can’t really comment on it.
But although The Wind Rises won a good number of critics awards, Frozen has won the major animation awards from the Globes, the BFCA, the BAFTA, the PGA, the VES, etc. I think it’s an easy call.
Huh, no commentary for animated film? That’s interesting, because I’m one of the few people who doesn’t think “Frozen” has this locked (even though I will in no way complain if it does), but I think “The Wind Rises” could actually take this. It’s the last Miyazaki film, it also received raved reviews, and while it has always been nominated in all the guild awards almost every time it was it won over the popular one (the Annies being the notable exception, and their rewarding 10 Annies to “How to Train Your Dragon” didn’t help that film any). Also since this film is made for adults but is accessible it would be a good way for the category to “grow up” and branch out of family films. I am likely completely wrong on this, but looking at that gives the impression of a perfect storm brewing.
You can be offended, Marhshall, but I’m calling a spade a spade.
Cuaron is Mexican born. But he is *not* non-Caucasian, not anymore than, say, Pedro Almodovar. Neither are his fellow filmmaker friends Innaritu and del Toro.
He is not a Mexican-American immigrant. He is not anywhere near what you’d called the oppressed Mexican minority in America. yes, he is a minority in Mexico, but it’s of the privileged kind.
If you’ve been to Mexico, you will notice that about 1/6 of Mexicans are of pure European descents. The majority, 5/6, are either mixed or of Native American descents. Yet, the ruling class is the 1/6. If you turn on a Mexican TV, you will see only white European descents. Actors, news anchor people, etc.
Cuaron and his filmmaker friends belong in this ruling class of 1/6. It’s funny, I just Googled and landed on the Wikipedia page on this topic, and on that page there are multiple pictures of examples of European descent Mexicans, and one of the photos is of Guillermo del Toro.
Marshall, you are just plain mistaken by labeling Cuaron as non-Caucasian.
http://en.wikipedia.org/wiki/Mexicans_of_European_descent
Well, at least Mexico kept the colored people alive and make them economically second class citizen. In Argentina, they killed them all.
1) I’ve talked about, at appropriate length and detail for a general audience, the underlying statistical methodology that forms the basis of my (and Ben’s) work in previous Statsgasm episodes. It’s called “regression analysis.”
2) I’m fully expecting 12 Years a Slave to win on Sunday. I predicted it to win the NYFCC, the LAFCA, the Globe, the Critic’s Choice, and the BAFTA. But I was caught in the event horizon of my pick (and preference) for BP ever since the first week of October. I believe it can win, and I believe that I’ve come this far to change my pick.
Finally, I’ve gotten a lot of crap this season for saying that Cuaron is either non-white or non-Caucasian. When I tweeted that that Cuaron made history by becoming the 2nd non-Caucasian to win the DGA, no less than members of the African American Film Critics Association and director Ana Duvernay had no problems in retweeting and echoing my sentiments.
Splitting hairs so finely in classifying people is frankly offensive to this Asian with some Spanish ancestry, and is also inimical to the goals of diversity. Just my opinion.
And what’s up with Cuaron being “a non-Caucasian”? Are we automatically assuming that a Mexican is non-Caucasian? That’s like saying that a South African is automatically black.
Cuaron is WHITE, y’all.
Claudiu et al.,
First – I’m willing to admit that my model for BP is still a bit “coarse” and can be improved. And if and when I have time in the coming days, I will include those mentioned variables. And I am not trying to disrespect Marshall’s model because I’m not privy to his method. And I’m sure he’s a genius of some sorts. I just find it head scratching that his (and Ben Zaumer of THR’s) model producing such numbers against 12YAS and Ben himself is predicting 12YAS to win BP. LOL. What is the point of presenting your math? He obviously hasn’t baked in the more nuanced, yet significant variables.
And I know I know about BAFTA’s and other “smaller” awards’ significance. I weigh them accordingly to their precursor’s “accuracy.” What d’y’all take me for? 🙂 And of course I know BAFTA moved their dates in 2000 and I have considered that in my model. But interestingly, since they moved their date to before Oscar, their BP matching rate with the Oscar has improved.
I’ve crunched numbers for the Best Supporting Actress category and will present the numbers later on. It’s a very interesting race. And my numbers will reflect more closely to reality than Marshall’s. 🙂 (Just teasing you, M).
@gail what makes you think that the Wind Rises will upset Frozen? I would be ecstatic if that happened.
I’m confused– a negative beta weight does not mean there is a multicollinearity issue, nor do I understand, the goal being purely prediction, why you would remove the BAFTA variable for that rationale. if the model fit is improved, you should just keep all of the variables— you’re not trying to make any claims of significance.
^ Finally, someone who speaks and understand the wonk language 100%. 🙂
LMAO, all y’all who are commenting about models and whatnot, come back to your senses and understand the core stats/econometrics rules. You’re working with very limited data (if you’re using the BAFTA as a predictor, then remember what Marshall said in his early blog that it broadcasted before the Oscars beginning in 2000, meaning your n for BP noms is somewhere < 100). I'm hella laughing at your comments of "ohh, let me add a SAG win, and DGA nom, and PGA win, and NBR top 10, and my aunt Peggy's predictions" because all you're doing is KILLING your degrees of freedom and opening up a HOT MESS of collinearity. SMH at how y'all are talking about models and not understanding that your parameters are insignificant cuz you're adding tooo much shaattttttt. haha
“Statsgasm is predicting that there’s an 83.73% chance that Ang Lee will no longer be the only non-Caucasian Best Director winner after Sunday night.”
Pardon me if I seem to be splitting hairs, but my impression is that Alfonso Cuaron is a Mexican whose ancestry is primarily or entirely Spanish. That said, the point is well taken in that he may be referred to as a person of color.
“lupita nyongo has to win, that’s the only way 12 years a slave can win best picture”
Again, I disagree… Ejiofor has a chance. Plus, BP for 12 Years with no acting win is also quite possible (with something like costume design instead, plus screenplay). The latter isn’t even a particularly unlikely scenario.
“Captain’s number is inflated due to its WGA win where 12YAS wasn’t competing. Her’s number might be too good, especially next to Hustle, but remember Her won 2 critics awards, more than any contender. And its WGA win can’t be discounted.”
Her is definitely getting too high a percentage. The critics’ awards are more or less meaningless compared to the guilds, BAFTA, Globe, BFCA, of which it’s won none. And the WGA win is important, but mostly for its screenplay chances. And it’s not winning BP with just screenplay – so what else is it winning? It’s not BD-nominated (which in itself is enough, couple with the lack of meaningful precursor wins, to make it more or less dead) and isn’t even close to being a contender in any other category.
Captain Phillips is also rated too highly. Also no BD nomination, and no BP precursor wins at all. These two, to me, are pretty much completely dead, 1-2% each, at best, and I’m quite confident their chances are no higher than that – I base this on stats, not stubbornness. But I understand how it’d be hard to make a mathematical model get this. Plus you might disagree, which is also fine. I know I’m more radical than most when it comes to outsiders’ BP percentages, but I have complete faith in my approach and conclusions in the matter.
But, again, the numbers for the top 3 are very close to being right.
“There are variables I could also include like SAG Ensemble nomination, SAG wins (any, not just ensemble), WGA nomination, DGA nomination, Oscar BD nomination, Oscar Editing nomination, Oscar screenplay nomination, etc.”
You HAVE to include those! 🙂 All of them (and probably Oscar nominations ranking as well). I’m quite serious! To be honest, I’m amazed that you’re getting 12 Years as the favorite even WITHOUT including (any of) those. Every single one of those is an absolutely vital tool. But that just goes to prove how strong 12 Years’ showing so far this season has really been, despite what some are saying – that it would be the favorite over Gravity even without taking into account its most important pluses (WGA, screenplay and SAG Ensemble nominations, which Gravity is lacking).
“I suspect that if I include the above variables in my numbers, Gravity’s (as well as Her and Captain) chance will decrease a little bit, and 12YAS and AH’s chance increase a little bit.”
A LOT… 🙂
As for including awards with less-than-great BP matching records in the model, I would most likely just drop all of those under 50% (perhaps even more), as they’d probably just needlessly complicate things. But, if you want to be comprehensive, you should give them with the absolute minimum weight possible (that’s still somewhat significant, when enough such wins add up) – otherwise you’re just reducing the accuracy of the model, methinks, by diminishing the weight of the fewer, but far more important, major awards (guilds, BAFTA, Globe, BFCA).
OR you could only use them in combinations (find stats like ‘no movie has won this+that+that etc. critics’ awards and not won BP, or ‘this+that+that critics’ awards plus the DGA’, or stuff like that), combinations that have somewhat smaller sample sizes but far more accuracy. But, again, this probably just needlessly complicates matters. The basic tools (which, however, include all of the ones you mentioned above) are, to my mind, quite enough to determine a single BP favorite each year, and be right about 90% of the time, even in competitive years.
Honestly, that’s why I’ve never bothered to build an actual model myself. I’ve not found it necessary so far, not for BP. There’s always been at least one movie that ticked all the boxes, and all of its opponents have always had this or that problem, thus inevitably making the former the statistical favorite. With the exception of the Saving Private Ryan – Shakespeare in Love year, in which SPR was probably the slight favorite, but had no undeniable trumps over SiL (I don’t consider DGA+PGA as strong a combination when up against a different SAG winner as most people, and previous results back me up). A similar situation was The Hurt Locker vs. Inglorious Basterds, but the latter was the kind of movie that never wins BP, so it was pretty easy to rule out based on that. Anyway, I can live with an unclear race popping up from time to time, like this year with American Hustle (which, however, I consider to be a bigger underdog to 12 Years a Slave than Shakespeare in Love was to Saving Private Ryan, for multiple reasons) – after all, I’m not a professional prognosticator. It’s just a hobby, for the most part.
John,
My model currently can be quite a bit more refined/detailed (what model can’t be improved?). There are variables I could also include like SAG Ensemble nomination, SAG wins (any, not just ensemble), WGA nomination, DGA nomination, Oscar BD nomination, Oscar Editing nomination, Oscar screenplay nomination, etc.
I suspect that if I include the above variables in my numbers, Gravity’s (as well as Her and Captain) chance will decrease a little bit, and 12YAS and AH’s chance increase a little bit.
There’s one element of Marshall’s theory that I disagree — he penalizes a contender for winning an award that hasn’t proven to be a great precursor. My theory is that *no* win is a bad win. It might not be a *great* win, but it can’t hurt you. How can an award hurt your chance? It can only give you more attention, it *adds* to resume’, fortifying your gravitas.
For example, in the past 18 years, if a film wins a Globes comedy/musical BP, 17% of the time it will go on to win the Oscar BP. Does this mean we should penalize American Hustle? We should have penalized Shakespeare in Love and Chicago too, right? In my model, AH earned some points for winning that prize, but of course not as much points as 12YAS b/c that film won GG Drama (44% success rate). And of course neither would have earned as much points as Gravity when it won the DGA (78% success rate).
I think people are over thinking Best Picture way too much! 12 Years A Slave is winning, it’s that period piece that is actually deserving to win, it’s shot on a larger scale than some of the other nominees landscape and scenery wise but it never really gives in to the sweeping period piece usuals, it has a huge cast that is well liked, important subject matter, great story. It will win believe me!
Steve McQueen is likely to take Best Director. The spirits are communicating with me about this- I will post my final readings an hour before the ceremony. There are going to be some big upsets, in particular the supporting actor, animated feature and screenplay races. But right now here are my current signals:
BEST PICTURE: 12 Years a Slave
BEST DIRECTOR: Steve McQueen, 12 Years a Slave
BEST ACTOR: Matthew McConaughey, Dallas Buyers Club
BEST ACTRESS: Cate Blanchett, Blue Jasmine
BEST SUPPORTING ACTOR: Barkhad Abdi, Captain Phillips
BEST SUPPORTING ACTRESS: Lupita Nyong’o, 12 Years a Slave
BEST ADAPTED SCREENPLAY: 12 Years a Slave
BEST ORIGINAL SCREENPLAY: Blue Jasmine
BEST CINEMATOGRAPHY: Gravity
BEST PRODUCTION DESIGN: 12 Years a Slave
BEST COSTUME DESIGN: 12 Years a Slave
BEST MAKEUP: Bad Grandpa
BEST FILM EDITING: Gravity
BEST ORIGINAL SCORE: Philomena, Alexader Desplat
BEST SONG: “Happy”, from Despicable Me 2
BEST SOUND: Gravity
BEST SOUND EDITING: All is Lost
BEST VISUAL EFFECTS: Gravity
BEST ANIMATED FEATURE: The Wind Rises (big upset)
BEST DOCUMENTARY FEATURE: 20 Feet from Stardom
Those numbers look just about right Alan. Maybe you and Marshall can pick each other’s brains after this race and tailor the models even more.
please NOTE:
I inadvertently posted an earlier draft of this article. Marshall had a few hundred more words to add to help illuminate the Director, Actor, Actress, Supporting Actor categories.
So I’ve fixed that now and the UPDATED EXPANDED post is now what you see.
Claudiu – I’ve only crunched the numbers for Best Picture, the only category I’m interested in this year, though I might do Supporting Actress later because this is a competitive category, and it might somewhat dictate the BP race. We’ll see if I have the time. 🙂
I’ve revised the numbers for BP. Just got home with a calculator in hand (the above numbers were estimates b/c I was adding and dividing my head.
12YAS – 42%
Gravity – 24.5%
American Hustle – 15%
Her – 11.5%
Captain Phillips – 7%
Captain’s number is inflated due to its WGA win where 12YAS wasn’t competing. Her’s number might be too good, especially next to Hustle, but remember Her won 2 critics awards, more than any contender. And its WGA win can’t be discounted.
OT: Must read insane collection. Check it out!
http://www.empireonline.com/features/cinematographers/
lupita nyongo has to win, that’s the only way 12 years a slave can win best picture, plus jennifer lawrence won last year they should give a chance to another good performance
“Marshall shows us which way the scales are likely to tip.”
According to his model.
“12YAS 39.5%
Gravity 23.5%
AH 14.5%”
I like Alan’s model. Sounds like he approaches the problem pretty much like I would – only, unlike me, he’s actually bothered to go through with it, so credit to him for that! And his results (for BP, at least) are way closer to the truth, if you ask me. Still too much left for the other 6 (22.5%), but the percentages among the top 3 look pretty good. Alan, if you have models for other categories as well, I’d be quite interested to know what the results are – unless, of course, you want to keep them to yourself, which is also perfectly OK.
I think there will be a split between Director and Picture. Cuaron will win and 12 Years will win.
My model includes NBR, NYFC, LAFC, NSFC, Globes, BFCA, SAG, PGA, DGA, WGA and BAFTA, weighing accordingly each’s historical influences. Of course the DGA has the heaviest influence.
My model measures the last 18 races and also concentrating on the last 6 races, and averaging them, giving slightly more importance to the recent races.
The results (rounding to 1/2 percent):
12YAS 39.5%
Gravity 23.5%
AH 14.5%
My AP US teacher was a libertarian blowhard lol. And being a history/politics wonk even back then, we got into our fair share of spirited discussions. But he was smart, and he gave us quite a bit of freedom on what we could read. For some reason I chose 12 Years when it came to our Civil War-era unit.
The estimated weights for the model’s predictors will change accordingly with each new data point.
What’s the commotion about Leto’s pic? There’s a photogenic dude.
“as someone who read Solomon Nortrup’s harrowing memoir in my AP US history class in high school”
Wow. Just based on that fact alone I can conclude you had an infinitely better teacher than me for this class. Mine was a moron.
“…wasn’t there another picture of Leto? “
jeezuz, Bryce – does he take a bad picture? If so, I’ve never seen it.
this was the photo we originally used for Jared Leto. We had a black & white conversion.
The other photo didn’t give him much head room. Better to leave ample room for head.
Great work Marshall. And thanks for not making us wait til Part 2 for the top categories.
Only 369 DGA members who get to vote for the Oscars.
Is the above pic the replacement one for Jared Leto? What did the first one look like? The following one would have been best.
http://www.celebuzz.com/2014-01-13/jared-leto-poses-naked-for-photographer-terry-richardson/
I thought a supporting actress win is often a consolation prize and the film consequently does not win best picture. Melissa Leo, Penelope Cruz, Mo’Nique, Anne Hathaway, Octavia Spencer.
And I agree Lupita is a key early indicator. If she wins, I’m still not sure, but if she loses, Slave is dead. A sole script win I guess. On that note, Lupita + a random tech win like Costumes? The time for doubting is over.
If Gravity wins Editing, I also think it’s likely to take Picture, even if Lupita wins.
And on the offchance Hustle wins for Lawrence AND screenplay, and Gravity loses Editing, hell, that will be one close call. But I seriously doubt those circumstances.
Wow, this is so fun. Great writeup too. I eagerly await part two as well as the post-Oscars followup!
Check out the big bar on Cate.
I see it’s already been discussed, but yeah, it’s so difficult, mathematically risky even, to try to qualify the “soft” factors that end up overcoming the precursors in close-call races. I do wonder if some sort of binary for WGA ineligibility could enhance the model. A problem would be how rare it must be for a WGA-ineligible script to win the Oscar, meaning most of the time including a measure for it would be redundant/insignificant.
And I can’t imagine how difficult it would be to conceive of a significant measure for something like whether the nominee is a previous winner, she just won last year, etc. It can go either way, and we just know it when we see it.
Still, it’s interesting that DGA overwhelms all the precursors that have gone with Slave. Maybe interaction variables between DGA and the other guilds or BAFTA? It’s been a long time since I’ve done economics! But I have to think that the effect of the DGA winner should be downplayed when the director frontrunner all season long is different from the film that’s been winning every Best Picture award (sans SAG).
Similarly, Her’s script also won the BFCA, but its eligibility for BAFTA and misses could be quite telling. You are right, this category and perhaps Documentary are the big coin tossers for me. But then I haven’t even seen the docs, and I’m the rare person who actually loved Hustle and think its script is even more deserving.
It would be interesting, though surely pointless and just for fun, to have a general “how many Oscar wins” model just to see what it says for Hustle. The big question marks on Oscar night are surely whether Hustle will go home emptyhanded, whether Slave or Gravity will win Picture, and just how few Oscars Slave can/will win and still take Picture.
Your models will need adjustment after this year’s race since it’s going to get 2 out of the 6 major categories wrong (Picture and Supporting Actress), which is not very good.
Props for putting it all together though.
Good on math but let’s not forget that Oscar voter mentality can often go against any sense of logic.
As proof: Crash, Roberto Beigni Best Actor, Shakespeare in Love, Marisa Tomei, Juliette Binoche (seriously, NO ONE thought she could beat Bacall in ’97), Marcia Gay Harden, the list goes on.
Sure, I place a lot of faith in these as Blanchett pretty much has it in the bag (do argue Lawrence such a large lead over Lupita) but another thread on this site quotes a member saying “Academy voters go with heart, not head” so if anything undoes math and records and such, it’s this.
“So 12 Years a Slave, according to these graphs, walks away with zilch on Oscar night? Hogwash!”
Jay, plus, anyone else who is disagreeing with Marshall, I say: Don’t hate the player, hate the game.
There is no standard way of doing this, and as a result, it will be highly dependent on the model builder’s (i.e. my judgement).
But *wonk alert* I do account for 12 Year’s ineligibility at the WGA by transforming the estimated historical weight of the WGA by the overall percentage of Oscar winners that received corresponding WGA nominations in the same category. In other words, I take the coefficient of the WGA variable and multiply it by 26/29 (of the 29 past Adapted Screenplay winners, 26 of them received WGA Adapted Screenplay nods, which has been a category since 1984). It doesn’t change the overall prediction, but it certainly keeps 12 Years within range.
Keep in mind that I built these models way back in September/October – once the nominees were announced, I’ve only made very minor changes. Adding/dropping variables midstream to match what everyone is predicting/expecting to win is bad form for any stats practitioner. And I’ve already stated that I don’t expect Statsgasm’s models to bat 21/21.
In 5 days time, we’ll know for sure who the Academy voted for. Marshall, your chart of having ‘Gravity’ to win Best PIcture by such a big margin is, I think, exaggerated; but to your credit, I do think it has a good chance of overtaking ’12 Years a Slave.’ I’m actually pulling for ‘Gravity’ at this point. But Jennifer Lawrence winning Best Supporting Actress I think it a bit of a toss-up, really. Lupita Nyong’o has been getting a lot of good press, lately. Blanchett and Leto are sure as sugar to win in their categories. The only upset I can see happening there is Barkhad Abdi overtaking Jared Leto (it would take a miracle, though). The two screenplay categories will be a little more interesting, and the best indicator of which film the Academy might have voted as Best Picture. If ’12 Years a Slave’ wins Best Adapted Screenplay, I think Best Picture is still up in the air. If ‘Captain Phillips’ wins, then the Academy might have decided to go with ‘Gravity.’ If Lupita Nyong’o wins Best Supporting Actress, and John Ridley wins Best Adapted Screenplay, I think it would be a safe bet that ’12 Years a Slave’ will take Best Picture. There’s usually a pattern that Best Picture winners adhere to, in that they almost never win Best Picture without winning one or more of the other top categories.
As long a we never forget, the margin of probability is not anything like the margin of actual votes. There’s no correlation there.
Marshall is covering the probability angle for us, and that’s fascinating.
Rob’s Simulated Oscar Ballot covers the mechanics of a voting experiment.
Simulations of voter behavior show us internal numbers that were reflected by the PGA in the real world: Gravity and 12 Years a Slave have an equal number of supporters. It’s a nearly perfect 50/50 split.
Rob shows us that the scales are perfectly balanced.
Marshall shows us which way the scales are likely to tip.
So 12 Years a Slave, according to these graphs, walks away with zilch on Oscar night? Hogwash!
The funny thing is that the model isn’t even close on Original Screenplay, Supporting Actress, or Director(and yes, obviously Adapted Screenplay). It was more accurate when it had Her at around 90-something percent, which is still true. Someone needs to tell the model that Lupita won SAG and that Lawrence just won Best Actress last year. Cuaron only at 60%? The DGA win alone gives him an 89% chance of victory. Throw in his other wins and it’s probably around 95%.
At least whomever is controlling the predictions to the right seems to know what’s really happening in these categories.
Marshall, what is your prediction for how many of Ellen DeGeneres’ jokes will be funny vs. uncomfortable, especially when she jokes about 12 Years a Slave?
I’m just kidding of course, there is no way to know. 🙂
Great Job Marshall!
I hope you’re right, that Gravity will win Best Picture.
Why is Ejiofor below a meagre 5%? He did win the BAFTA after all (in addition to the lions’ share of critics’ prizes, not that those matter much), but how does that make him a decidedly weaker option than DiCaprio and especially Dern who has won very little as far as I recall? I’m not arguing the stats as such, I’m just asking you to explain to me what I’m missing:)
Anyway, Ejiofor below 5% seems strange (compared to Abdi’s 14%, also a BAFTA winner in a category with another frontrunner. Did the fact that Ejiofor was on home ground at BAFTA, whereas Abdi was not, weigh against him in your model or…?)
Did you not account for 12 Years’ ineligibility at the WGA in your model? Is there no way to somehow incorporate the handicap of being ineligible for any given award, so that it won’t have a distorting effect on the model…?
I am a bit confused as to why Michael Fassbender is ranked last? Is it his lack of campaigning?
i’ll vomit if AH wins screenplay over HER…two words academy GROW UP
Well done, Marshall.
I know it’s a lot of work, but I’m really hoping you’re way off on about 8 of these 🙂
I do agree with Statsgasm on one thing: there will be no split between director and picture.
Gonna cringe if American Hustle wins screenplay over Her.
Yeah… Expect more than just one of these to be a “guaranteed miss by the model.”
Sounds legit: model and analysis.
It’s 1PM eastern time, still on an empty stomach, wasn’t there another picture of Leto?
p.s. GREAT WORK, Marshall!
“…wasn’t there another picture of Leto? “
wow, Bryce, nice micromanaging. ok ok, we’ll use another head shot of Leto that’s a little more consistent with the others, size-wise
🙂