Phoenix Wright
He's a Capcom character. Capcom already has a character. He's unconventional. He's kinda been in Marvel vs. Capcom so they wouldn't be breaking new ground by making him a zany fighter.
He's also not that legendary yet.
Phoenix Wright chances: 0.1%
I doubt Capcom would get a second rep. And if they did, I think Zero would probably be more likely. But Zero's about as likely as Tails, Knuckles or Shadow (i.e. not very likely at all).
Phoenix Wright want: 40%
(50% = indifference)
I've never played his games. I'd want other characters.
Masked Link prediction: 2%
I dunno. Everyone is expecting low because Toon Link. Just how low is the question.
Nominations:
5x Sheik
Rate their chances statistics:
Oh yeah, and I did do some statistics, although I'm still working on coming up with a simple predictive model.
If you saw my post in the Franchise sales/representation thread, you saw that the number of representatives that a series gets in a Smash game can be modeled with a linear regression using Franchise Sales and Previous # Slots (how many reps they had in the previous game) with an R^2 value of a little over .75.
Previous number of slots accounts for a large amount of the variation on its own (although so does sales, to a lesser degree).
At any rate, I figure you take the number of slots that predicts for the game and combine that with the predictions people made by series might get a reasonable performance. I want something I can apply to our current predictions to make some for Smash 4. But I've been busy playing Wind Waker instead, so yeah.
Anyway, here's how the previous iterations of the game did compared to some very simple and obviously wrong baselines:
Actual Brawl: 35/39 slots/movesets, 4 Mario characters, 4/6 Pokemon, 4/5 Zelda, 2 Donkey Kong, 2 Fire Emblem, 4 Retro, and 2 third-party characters.
GameFAQs: predicted a roster with 85.8 characters, including 11 Mario characters, 7.6 Pokemon, 8.6 Zelda, 3 Donkey Kong, 6 Fire Emblem, 7 retros and 13 third-party characters.
SmashBoards: predicted a roster with 52.7 characters, including 7.3 Mario, 5.9 Pokemon (not bad there), 6.9 Zelda, 2.6 Donkey Kong, 2.6 Fire Emblem, 5.1 Retro, and 5.2 third-parties.
My sales-based model predicts a roster with 34.8 first-party characters (I did not include 3rd-parties in the model), and 38.4 movesets. It is interesting that it overestimates almost exactly 2 slots/movesets more than it really was (the model does not know how many slots Brawl had overall), since that implies that Sonic and Snake took slots that would've gone to first-party characters. It predicts 5.2/5.9 Mario, 4.3/4.9 Pokemon, 4/5 Zelda, 1.9/2.1 Donkey Kong, 2.1/2.3 Fire Emblem and 4.4/4.6 retro characters.
Brier score (lower is better, 1 is the worst possible, 0 is perfect accuracy):
GameFAQs top 10 newcomers: 0.222
SmashBoards top 10 newcomers: 0.233
As I said, I'm still working on a model for comparing on newcomers, but yeah. It will have to exclude third-parties.
GameFAQs top 100: 0.143
SmashBoards (all predictions since there were less than 100): 0.185
Notice that you can improve your Brier score by making a larger number of obvious predictions. Predicting low numbers for obscure characters increases the score but doesn't really mean you were a prescient predictor just because you gave low likelihood to things that most people would give low likelihood to.
On the same characters as GameFAQs Top 100, simply predicting 100% likelihood for every Melee veteran and 0% for any newcomer (E3 newcomers not included since as they were revealed before the game started, they cannot be part of any predictions) would get you a Brier score of 0.14.
The same thing on the SmashBoards gets a Brier score of 0.182, about the same.
So, they'd have been just as, if not more accurate, simply by predicting no more newcomers and that every Melee veteran would return.
Later I'll do some similar statistics on the current iteration of Rate Their Chances to provide a summary of what kind of roster size and series distribution people seem to be expecting. I intend to re-calibrate my ratings based on a sensible roster size.
Based on Fire Emblem, we're probably being a bit more sensible than last time. Our combined prediction seems to be about 3.1 Fire Emblem characters, which is not ridiculous like the 6 GameFAQs predicted for Brawl.
He's a Capcom character. Capcom already has a character. He's unconventional. He's kinda been in Marvel vs. Capcom so they wouldn't be breaking new ground by making him a zany fighter.
He's also not that legendary yet.
Phoenix Wright chances: 0.1%
I doubt Capcom would get a second rep. And if they did, I think Zero would probably be more likely. But Zero's about as likely as Tails, Knuckles or Shadow (i.e. not very likely at all).
Phoenix Wright want: 40%
(50% = indifference)
I've never played his games. I'd want other characters.
Masked Link prediction: 2%
I dunno. Everyone is expecting low because Toon Link. Just how low is the question.
Nominations:
5x Sheik
Rate their chances statistics:
Oh yeah, and I did do some statistics, although I'm still working on coming up with a simple predictive model.
If you saw my post in the Franchise sales/representation thread, you saw that the number of representatives that a series gets in a Smash game can be modeled with a linear regression using Franchise Sales and Previous # Slots (how many reps they had in the previous game) with an R^2 value of a little over .75.
Previous number of slots accounts for a large amount of the variation on its own (although so does sales, to a lesser degree).
At any rate, I figure you take the number of slots that predicts for the game and combine that with the predictions people made by series might get a reasonable performance. I want something I can apply to our current predictions to make some for Smash 4. But I've been busy playing Wind Waker instead, so yeah.
Anyway, here's how the previous iterations of the game did compared to some very simple and obviously wrong baselines:
Actual Brawl: 35/39 slots/movesets, 4 Mario characters, 4/6 Pokemon, 4/5 Zelda, 2 Donkey Kong, 2 Fire Emblem, 4 Retro, and 2 third-party characters.
GameFAQs: predicted a roster with 85.8 characters, including 11 Mario characters, 7.6 Pokemon, 8.6 Zelda, 3 Donkey Kong, 6 Fire Emblem, 7 retros and 13 third-party characters.
SmashBoards: predicted a roster with 52.7 characters, including 7.3 Mario, 5.9 Pokemon (not bad there), 6.9 Zelda, 2.6 Donkey Kong, 2.6 Fire Emblem, 5.1 Retro, and 5.2 third-parties.
My sales-based model predicts a roster with 34.8 first-party characters (I did not include 3rd-parties in the model), and 38.4 movesets. It is interesting that it overestimates almost exactly 2 slots/movesets more than it really was (the model does not know how many slots Brawl had overall), since that implies that Sonic and Snake took slots that would've gone to first-party characters. It predicts 5.2/5.9 Mario, 4.3/4.9 Pokemon, 4/5 Zelda, 1.9/2.1 Donkey Kong, 2.1/2.3 Fire Emblem and 4.4/4.6 retro characters.
Brier score (lower is better, 1 is the worst possible, 0 is perfect accuracy):
GameFAQs top 10 newcomers: 0.222
SmashBoards top 10 newcomers: 0.233
As I said, I'm still working on a model for comparing on newcomers, but yeah. It will have to exclude third-parties.
GameFAQs top 100: 0.143
SmashBoards (all predictions since there were less than 100): 0.185
Notice that you can improve your Brier score by making a larger number of obvious predictions. Predicting low numbers for obscure characters increases the score but doesn't really mean you were a prescient predictor just because you gave low likelihood to things that most people would give low likelihood to.
On the same characters as GameFAQs Top 100, simply predicting 100% likelihood for every Melee veteran and 0% for any newcomer (E3 newcomers not included since as they were revealed before the game started, they cannot be part of any predictions) would get you a Brier score of 0.14.
The same thing on the SmashBoards gets a Brier score of 0.182, about the same.
So, they'd have been just as, if not more accurate, simply by predicting no more newcomers and that every Melee veteran would return.
Later I'll do some similar statistics on the current iteration of Rate Their Chances to provide a summary of what kind of roster size and series distribution people seem to be expecting. I intend to re-calibrate my ratings based on a sensible roster size.
Based on Fire Emblem, we're probably being a bit more sensible than last time. Our combined prediction seems to be about 3.1 Fire Emblem characters, which is not ridiculous like the 6 GameFAQs predicted for Brawl.