edit: To clarify, this post was not about the upcoming MSL final, let alone making a prediction. I was just curious about whether Jaedong really does play better in bo5's or whether it only seemed that way at a glance.
I often see people talk about Jaedong's Bo5 record as if it is something particularly impressive, like he steps up his game another notch when he is in a best-of-5.
People like to make storylines like that. They do it all the time; in baseball, for instance, they talk about clutch hitters. They even do it, half-jokingly, with patently ridiculous correlations like the various curses you hear about (Kim Carrier, Sports Illustrated cover, the Bambino) that last until they don't. Even the more plausible ones, though, like "clutch hitting", don't usually stand up to statistical analysis. That's why I wonder whether Jaedong's supposed prowess in best-of-5's is actually a fact.
Fact: the longer a series goes, the more likely the better player is to win it. It's easier for the worse player to get lucky once in one game than to get lucky three times in five games. Jaedong, of course, is better than nearly every other player, so we'd expect his record in best-of-5's to be better than his record in best-of-1's, for no other reason than that.
Specifically, there are 32 possible outcomes for a 5 game series between X and Y. (2^5--each of the 5 games can go one of two ways). In 16 of them, X wins. It can be 5-0 (1), 4-1 (5), or 3-2 (10). Say that X has an x chance of winning each individual game and a y (=1-x) chance of losing. The odds of his winning the series 5-0 are 1 * x^5. The odds of his winning 4-1 are 5 * x^4(y). Winning 3-2: 10 * x^3(y^2). Add these up and you have the odds of his winning the series.
Jaedong's overall winning percentage according to TLPD is 68.29%. Using that as an approximation for his chance of winning any given game he plays, and substituting that for "x" above, we would expect Jaedong to win 81.36% of his Bo5's.
Anyway, it's just a thought. Maybe the Jaedong bo5 effect is a perception created by the natural tendency for any really good player to have a better series win-percentage than individual game win-percentage.
Here's the interesting thing, though: According to Liquipedia, Jaedong's overall best-of-five record is 17-3 (85%), which is quite a bit higher than the 81.36% than his overall record would predict. Add in the fact that we would expect his individual game-winning percentage to be lower in best-of-five's since he only faces those players who have been playing well enough to make it to the end of tournaments, and his series-winning percentage starts looking even more impressive. In fact, over the course of compiling that 17-3 series record, Jaedong won 71.6% of individual games in those series (which makes the 85% series-winning record unsurprising).
In short: it doesn't seem that Jaedong plays particularly clutch in deciding games of series or anything, but it does seem that when he is in best-of-5's, his overall gameplay is kicked up a notch.
Also, you only generally play Bo5s vs great opponents (semis and finals) of tournaments. His 68% win percentage includes everyone, from qualifiers to bo1 proleague games, etc.
Thanks for crunching the numbers. The phenomenon that you mention --- where a player who's simply better than the field will have a very high Bo5 win percentage by the very nature of a Bo5 --- is sometimes overlooked. It seems like Jaedong really is the Bo5 master though --- as the games progress, his opponents get more and more demoralized and Jaedong himself plays better and better. Jaedong has incredible mental strength (see: Batoo finals) and that makes him especially dangerous.
edit: I see the math is a little wonky but I think the main point stands: Jaedong's success in Bo5's is greater than you'd expect from his normal win ratio, especially since Bo5's tend to be against a higher class of opponent than random proleague matches.
In short: it doesn't seem that Jaedong plays particularly clutch in deciding games of series or anything, but it does seem that when he is in best-of-5's, his overall gameplay is kicked up a notch.
I also read the rest of the text, but this is contradicting yourself. Kicking your gameplay up a notch in best-of-5 games, is playing particularly clutch in deciding games.
On January 21 2010 00:57 timmeh wrote: don't care much for this. but why are you conidering 5-0 if a best of 5 is defined by a player winning 3 out 5 matches?
On January 21 2010 00:58 Kaniol wrote: There's SLIIIGHT problem with your calculations. Odds of his winning the series 5-0 are 0, same for 4-1, same for losing 1-4 and losing 0-5.
BO5 = till one of the players has 3 wins
Right, should have clarified that, but my post was already getting so long that I didn't want to bloat it even further.
Technically you guys are right that those phantom games (say, the last two in a potential 5-0 victory) are not actually played, but it's just convenient to think of them that way. Otherwise, I would have to distinguish between winning the first 3 games, winning 3 of the first 4 or winning 3 of the first 5.
As far as the odds go, it boils down to the same thing. Think of it this way: if I have X chance of winning a series that is played until decided, would playing the extra matches (as in a showmatch) have any effect on my chance of winning the series?
In short: it doesn't seem that Jaedong plays particularly clutch in deciding games of series or anything, but it does seem that when he is in best-of-5's, his overall gameplay is kicked up a notch.
I also read the rest of the text, but this is contradicting yourself. Kicking your gameplay up a notch in best-of-5 games, is playing particularly clutch in deciding games.
You can say that he plays clutch in games of a bo5, compared to his usual play. I just meant that he doesn't play "clutcher" in deciding games of the series than in the series overall. IOW, it's not like, say, he plays like regular Jaedong until he is faced with losing the series and then he goes into overdrive. Just an observation.
On January 21 2010 01:05 qrs wrote: Right, should have clarified that, but my post was already getting so long that I didn't want to bloat it even further.
Technically you guys are right that those phantom games (say, the last two in a potential 5-0 victory) are not actually played, but it's just convenient to think of them that way. Otherwise, I would have to distinguish between winning the first 3 games, winning 3 of the first 4 or winning 3 of the first 5.
As far as the odds go, it boils down to the same thing. Think of it this way: if I have X chance of winning a series that is played until decided, would playing the extra matches (as in a showmatch) have any effect on my chances of winning the series?
The problem is this: you have 5 possible ways to win 4-1 and 10 possible ways to win 3-2.
I'm assuming your 5 possible 4-1s are: LWWWW WLWWW WWLWW WWWLW WWWWL
You can't collapse the 4-1 group into 3-1 wins, because the latter 2 cases would technically be identical to 3-0 wins. Similarly, with 3-2s, your 10 possible combinations are:
While these are 3-2s with regard to the combination of 5 games, the bolded sets are actually resolved as 3-1s and the italicized one as a 3-0, because the losses are never reached. Your chance of winning a Bo5 overall is the same, but your chance of winning with a particular result is different.
On January 21 2010 00:57 timmeh wrote: don't care much for this. but why are you conidering 5-0 if a best of 5 is defined by a player winning 3 out 5 matches?
On January 21 2010 00:58 Kaniol wrote: There's SLIIIGHT problem with your calculations. Odds of his winning the series 5-0 are 0, same for 4-1, same for losing 1-4 and losing 0-5.
BO5 = till one of the players has 3 wins
Right, should have clarified that, but my post was already getting so long that I didn't want to bloat it even further.
Technically you guys are right that those phantom games (say, the last two in a potential 5-0 victory) are not actually played, but it's just convenient to think of them that way. Otherwise, I would have to distinguish between winning the first 3 games, winning 3 of the first 4 or winning 3 of the first 5.
As far as the odds go, it boils down to the same thing. Think of it this way: if I have X chance of winning a series that is played until decided, would playing the extra matches (as in a showmatch) have any effect on my chance of winning the series?
No, but 4-1 could be 3-0 and then he drops the (nonexistent) 4th game.
On January 21 2010 01:14 Pokebunny wrote: No, but 4-1 could be 3-0 and then he drops the (nonexistent) 4th game.
Strictly speaking, it's still a win, so it doesn't affect his overall likelihood of winning a Bo5, it just factors in when you're trying to analyze his chance of winning in 3/4/5 games.
On January 21 2010 01:01 Hot_Bid wrote: Also, you only generally play Bo5s vs great opponents (semis and finals) of tournaments. His 68% win percentage includes everyone, from qualifiers to bo1 proleague games, etc.
Right, I mentioned that at the end. Sorry if my OP is a bit schizoid: the fact is that I started writing it expecting to find that there was no significant Jaedong bo5 effect, but when I discovered that page of stats on Liquipedia, I had to change my mind.
On January 21 2010 01:01 RaGe wrote: you can't consider 4-1 or 5-0 cause that can't happen.
On January 21 2010 01:03 freelander wrote: there are 6 possibilities, not 32 wtf
winning 3-0, 3-1, 3-2 or losing 0-3, 1-3, 2-3
OK, so many people are bringing this up that I see it was a terrible idea to gloss over it in the OP.
You're right, there are 6 actual possibilities. But here's what I meant: suppose you build a tree of possibilities:
G1 /\ 1-0 0-1 /\ /\
etc. If you extend that to the bottom, there would be 32 leaves. It's just that we cut off the branches that don't make a difference. So, for instance 3-0 is the equivalent of 4 different outcomes, all of which result in the same player's victory. As far as odds go, though, if you want to count every possibility for a series win as being equally likely (which is how I was doing it), then you have to think of a 3-0 victory as four possibilities in one. If you prefer, you can say that there are only 6 outcomes, but some are more likely than others (3-2 is much more likely than 3-0). It boils down to exactly the same thing.
edit:
On January 21 2010 01:13 TheYango wrote: ...While these are 3-2s with regard to the combination of 5 games, the bolded sets are actually resolved as 3-1s and the italicized one as a 3-0, because the losses are never reached. Your chance of winning a Bo5 overall is the same, but your chance of winning with a particular result is different.
Exactly right. All I was interested in was the chance of winning a bo5 overall. Sorry for going at it in such an unclear way.
On January 21 2010 00:57 timmeh wrote: don't care much for this. but why are you conidering 5-0 if a best of 5 is defined by a player winning 3 out 5 matches?
On January 21 2010 00:58 Kaniol wrote: There's SLIIIGHT problem with your calculations. Odds of his winning the series 5-0 are 0, same for 4-1, same for losing 1-4 and losing 0-5.
BO5 = till one of the players has 3 wins
Right, should have clarified that, but my post was already getting so long that I didn't want to bloat it even further.
Technically you guys are right that those phantom games (say, the last two in a potential 5-0 victory) are not actually played, but it's just convenient to think of them that way. Otherwise, I would have to distinguish between winning the first 3 games, winning 3 of the first 4 or winning 3 of the first 5.
As far as the odds go, it boils down to the same thing. Think of it this way: if I have X chance of winning a series that is played until decided, would playing the extra matches (as in a showmatch) have any effect on my chance of winning the series?
That makes complete sense to me now. I was initially confused with what you were saying.
Calculation seems right. (even though considering 5-0, 4-1 scores might seem weird to most ppl)
This is an interesting read, but the conclusion is... not really shocking. The difference between his individual/series statistics might not be huge but considering the fact that series opponents are likely a tad better, it seems enough to say he plays particularily good in series.
What I'm wondering is, isn't this true for other S-Class players?
On January 21 2010 01:03 freelander wrote: there are 6 possibilities, not 32 wtf
winning 3-0, 3-1, 3-2 or losing 0-3, 1-3, 2-3
You're both wrong. The first poster is aware of combinatorics, but forgets about one aspect of the problem. The second poster is very wrong. Me, I just counted them by hand, because that's always the safest way, if you can do it..
There are 20 possibilities
There's one way of winning 3-0, and one of losing 0-3 There are three ways of winning 3-1 and of losing 1-3 There are six ways of winning 3-2 or of losing 2-3
On January 21 2010 00:52 qrs wrote: Here's the interesting thing, though: According to Liquipedia, Jaedong's overall best-of-five record is 17-3 (85%), which is quite a bit higher than the 81.36% than his overall record would predict.
No, it's not "quite a bit higher". I haven't run the numbers on it (nor will I), but I expect this to be within acceptable error, i.e. not statisticially significant for reasonable p-values.
On January 21 2010 00:52 qrs wrote: Add in the fact that we would expect his individual game-winning percentage to be lower in best-of-five's since he only faces those players who have been playing well enough to make it to the end of tournaments, and his series-winning percentage starts looking even more impressive.
This, on the other hand, is probably a real effect.
On January 21 2010 01:03 freelander wrote: there are 6 possibilities, not 32 wtf
winning 3-0, 3-1, 3-2 or losing 0-3, 1-3, 2-3
You're both wrong. The first poster is aware of combinatorics, but forgets about one aspect of the problem. The second poster is very wrong. Me, I just counted them by hand, because that's always the safest way, if you can do it..
There are 20 possibilities
There's one way of winning 3-0, and one of losing 0-3 There are three ways of winning 3-1 and of losing 1-3 There are six ways of winning 3-2 or of losing 2-3
Actually, the OP is right, from a calculation perspective. I didn't see it until he mentioned why he did it.
You're right, that there are 20 possibilities, however, the calculations doing it this way are incredibly messy. Doing it, assuming 32 possibilities, is much easier, although a little trickier to understand if you haven't taken any statistics courses in high school/higher.
EDIT - I'll do it myself.
Okay, by forming a tree of possibilities, assuming shadow games are played (allowing results of 4-1, and 5-0, even though those extra games played do not mean anything), there are 32 possibilities.
The chance of winning a game is 68.29%. Or 0.6829. I'll represent this number with "a".
The chance to win a single game = a The chance to lose a single game = 1 - a
Assuming a win, the second round gives a chance to win of a, and a chance to lose of 1 - a.
Therefore, in 2 games, the chance to win both is (a)(a), the chance to win 1 and lose 1 is (a)(1-a), the chance to lose 2 games is (1-a)(1-a). There are two possible chances of having the 2 games go 1-1, meanwhile there is only one way of having the series go 2-0 or 0-2.
The chance to win at least 1 game is a^2 + 2a(1-a).
Extending that to all 5 rounds, the chance to win at least 3 games is
a^5(1) + a^4(1-a)(5) + a^3(1-a)(10)
Simplified, this leads to:
a^3(6a^2 - 15a + 10)
Substituting 0.6829 into the equation:
0.8136 = 81.36%
EDIT - until you have about 30 BO5 series to look at, the results won't be statistically significant anyways..... Suffice it to say, according to his win percentage, he should win around 80% of the BO5's he plays. And that shows to be true.
On January 21 2010 00:52 qrs wrote: Here's the interesting thing, though: According to Liquipedia, Jaedong's overall best-of-five record is 17-3 (85%), which is quite a bit higher than the 81.36% than his overall record would predict.
No, it's not "quite a bit higher". I haven't run the numbers on it (nor will I), but I expect this to be within acceptable error, i.e. not statistically significant for reasonable p-values.
Fair point. I was planning to ask anyone who knows statistics (unlike me) to weigh in on that, but once I saw the detailed numbers on the Liquipedia page, it became moot.
I like how almost every single post is about math and not about Jaedong/Bo5s.
I'll just say that Jaedong has a knack for winning series with a decided mental disadvantage to start (the Arena MSL series against Hwasin and Bacchus 09 OSL series against Fantasy come to mind) in which players go into the series with a current win-streak over him, having beaten him in arguably insulting fashion during Proleague.
On top of that, they win the first set of the series in a very convincing fashion (in both instances I said to myself "Jaedong is outclassed this time, gonna get so murked" etc.) And then he manages to make the other player extremely nervous/demoralized and sweep the next three games.
I'm curious to see how many Bo5's Flash has lost, because I'm positive the number is about just as small (Off the top of my head, he lost to Never_V_ in Arena and Jaedong in GOMTV Intel Classic final)
Edit: Looks like he lost to GGplay during Daum OSL, and I forgot about Jaedong beating him in MSL S4... so I guess in the grand scheme of things he's lost quite a few
This may be obvious but it hasn't been said explicitly... The Bo5 format is not JUST about 2 people playing games until one player gets 3 wins. There is also a mental battle ongoing throughout the series.
As many people mentioned, Jaedong (historically) does not get worse when he's in a losing position. When people say JD is good at Bo5s, they are referring to this mysterious ability to play well no matter the situation.
There's more to say but other people already said it
On January 21 2010 01:01 RaGe wrote: you can't consider 4-1 or 5-0 cause that can't happen.
OP needs to take a combinatorics class.
Not quite. He's right, from a statistical analysis point of view..... Although they cannot physically happen, you can assume that they do to make the calculations much easier.
There's also the fact that he may be quite good at preparation against opponents, something that would have a much greater effect when he has weeks to prepare.
Not all BW games are created equal, there may be a player who can win 65% of his games with 2 days of preparation, but the additional 5 days to give him a week means that even against stronger competition he wins 75%.
Obviously it would be very hard to get a statistically meaningful sample for this, but I think it should be recognized.
On January 21 2010 02:24 Nal_rAwr wrote: you can do all this math and try to put some fancy language in there in an attempt to sound smart and stuff, but jaedong is not going to beat flash
This isn't really relevant to the OP's point at all. Lets keep the focus on math and Jaedong and save the hype and trash talk for the appropriate threads.
On January 21 2010 02:24 Nal_rAwr wrote: you can do all this math and try to put some fancy language in there in an attempt to sound smart and stuff, but jaedong is not going to beat flash
yeah btw and not only can you not win 4-1 and 5-0, but the game doesn't go on if you lose before the fifth match
Assuming that those shadow games are played makes the math simpler, but just as accurate.
The small statistics for maps shows that T has a good advantage over Z on every map they will have to play. And Flash is on the fire now. Still I feel like the outcome of the games depends on Jaedong. Actually I would go with 60% vs 40% for him to win the series.
Its completely silly to take Jaedong's win % as a representation of his single-game win percentage as that takes into account his series wins. This biases the data. I was under the impression you were trying to separate the two categories.
If you want to analyze his Bo5 performance in terms of the expected win rate given his single game win %, you need to actually count up his single game win %, excluding all series games.
According to the "62.3%" for individual games, this gives Flash a chance of 72.1% chance of winning a BO5 against an individual.
Jaedong's BO5 % chance better than Flash's BO5 chance, it's 81% vs 72%. Due to this, I give Jaedong a 53% chance of winning. It's still close, and anyone's game, but I still have to give Jaedong a slight edge.
On January 21 2010 02:47 FieryBalrog wrote: Its completely silly to take Jaedong's win % as a representation of his single-game win percentage as that takes into account his series wins. This biases the data. I was under the impression you were trying to separate the two categories.
If you want to analyze his Bo5 performance in terms of the expected win rate given his single game win %, you need to actually count up his single game win %, excluding all series games.
Actually, the more games you take into account, the better the data.....
Keep in mind that his Bo5 opponents are generally better than the average player he plays against in his other games (that gave him the 68.29% win rate). So in this light his Bo5 win rate is more impressive.
I can see that you have tried incorporating the method of "Binomial Probability" to come up with a figure to represent JAedong's probability of winning a Bo5
First your wording is a little ambiguous. You said "Specifically, there are 32 possible outcomes for a 5 game series between X and Y". In this sentence you did not say "Best of five". Therefore your argument that the number of possible outcomes in the series is 32, is true.
If you meant "Best of five" then obviously your "calculations" would differ in figures, and you need to take into account that the series can be of 3 different number of sets - ie) there can be a 5game bo5(3-2,2-3), 4game bo5(3-1, 1-3), or a 3 game bo5(5-0,0-5), and you would need to consider the 3 "scenarios" seperately and add up the probabilties, which would complicate the problem
You have failed to take into consideration the psychological damages experienced by the players X and Y when they lose a game(especially if they lose to a cheese). When X beats Y with a BBS build then the chances of X winning the 2nd game would also increase.
As someone already mentioned, you have also failed to take into consideration the skill-level of the opponent. Doing a Bo5 against FLASH is surely different to doing a bo5 against Gorush.
Also the position of the players is not considered. What I mean is right now Flash is the top player, and although jaedong is better in a career sense, Jaedong is in the "challenger" position. This MSL final can be titled "Can jaedong stop Flash?", not "Can Flash stop Jaedong". This "force" emitted by the players certainly impact gameplay.
It is ridiculous to represent someone's probability of winning with a number in a series of games. Have we not already experienced on 2007/3/3, when bisu's probability of winning was calculated to be like 3% (which includes 3-0 victory, 3-1, or a 3-2), and yet he won 3-0 against the king savior?
On January 21 2010 03:22 NrG.GoD- wrote: I can see that you have tried incorporating the method of "Binomial Probability" to come up with a figure to represent JAedong's probability of winning a Bo5
First your wording is a little ambiguous. You said "Specifically, there are 32 possible outcomes for a 5 game series between X and Y". In this sentence you did not say "Best of five". Therefore your argument that the number of possible outcomes in the series is 32, is true.
If you meant "Best of five" then obviously your "calculations" would differ in figures, and you need to take into account that the series can be of 3 different number of sets - ie) there can be a 5game bo5(3-2,2-3), 4game bo5(3-1, 1-3), or a 3 game bo5(5-0,0-5), and you would need to consider the 3 "scenarios" seperately and add up the probabilties, which would complicate the problem
Actually, that is taken into consideration.....
You have failed to take into consideration the psychological damages experienced by the players X and Y when they lose a game(especially if they lose to a cheese). When X beats Y with a BBS build then the chances of X winning the 2nd game would also increase.
As someone already mentioned, you have also failed to take into consideration the skill-level of the opponent. Doing a Bo5 against FLASH is surely different to doing a bo5 against Gorush.
Also the position of the players is not considered. What I mean is right now Flash is the top player, and although jaedong is better in a career sense, Jaedong is in the "challenger" position. This MSL final can be titled "Can jaedong stop Flash?", not "Can Flash stop Jaedong". This "force" emitted by the players certainly impact gameplay.
It is ridiculous to represent someone's probability of winning with a number in a series of games. Have we not already experienced on 2007/3/3, when bisu's probability of winning was calculated to be like 3% (which includes 3-0 victory, 3-1, or a 3-2), and yet he won 3-0 against the king savior?
That is all true, which is why statistics in something like this is unreliable, at best. However, it does give us some kind of numerical backing. Going by just the individual games records, I can't give Flash the win, and by only going by the most recent 30 or so games, the amount of error involved in the numbers makes it useless to do..... There is little choice here.....
No. You did not because you have chosen to consider a "Five game series" not a "Best of Five".
Please read carefully what i wrote. I said "IF you meant bo5"
Actually, they are the same thing, it's just that once 3 wins are found, the rest of the games are not played. The math works out the same. It is simplifying the calculations.
If you want to do the messy math, and find the same result, go ahead.
So much math fail in this thread: P(Jaedong wins)= 0.6829 since it is Bo5 will lead to: X~NB(3,0.6829) First instance, 3 straight wins by the dong=(0.6829)^3 Second instance, 3 wins, 1 loss=3*(0.6829)^3(1-0.6829)^1 Third instance, 3 wins, 2 losses=6*(0.6829)^3(1-0.6829)^2 Doing this on my calculator gives me 0.8135733316
EDIT: ofc isnt this true because it depends on the opposing players skill level as well but might be used as an indicator EDIT2: Oh lol 0.6829 was the probability to win a Bo5? Cuz then it is kinda fail question in the op. Gonna approximate the change for the dongman to win a single match. brb
On January 21 2010 02:47 FieryBalrog wrote: Its completely silly to take Jaedong's win % as a representation of his single-game win percentage as that takes into account his series wins. This biases the data. I was under the impression you were trying to separate the two categories.
If you want to analyze his Bo5 performance in terms of the expected win rate given his single game win %, you need to actually count up his single game win %, excluding all series games.
Er, no. Noone is trying to separate any categories.
On January 21 2010 03:51 ArmChairCritic wrote: EDIT: ofc isnt this true because it depends on the opposing players skill level as well but might be used as an indicator
Sure it's true. It's true assuming no prior knowledge of the opposing player. Or rather, assuming the opposing player is drawn from the same distribution as the previous opponents. All probability computation depends crucially on the information available, you just can't escape that. There are no universally true probabilities, unless you want to talk quantum mechanics here, but I guess you don't.
On January 21 2010 03:51 ArmChairCritic wrote: So much math fail in this thread: P(Jaedong wins)= 0.6829 since it is Bo5 will lead to: X~NB(3,0.6829) First instance, 3 straight wins by the dong=(0.6829)^3 Second instance, 3 wins, 1 loss=3*(0.6829)^3(1-0.6829)^1 Third instance, 3 wins, 2 losses=6*(0.6829)^3(1-0.6829)^2 Doing this on my calculator gives me 0.8135733316
EDIT: ofc isnt this true because it depends on the opposing players skill level as well but might be used as an indicator
Sure, there's usually more than one way to break something down. The figure is the same.
By the way, this might surprise people, but I actually wasn't trying to predict anything about the upcoming MSL final. I was just curious about whether the claim that Jaedong does better in bo5's was backed up by the facts.
On January 21 2010 03:51 ArmChairCritic wrote: EDIT: ofc isnt this true because it depends on the opposing players skill level as well but might be used as an indicator
Sure it's true. It's true assuming no prior knowledge of the opposing player. Or rather, assuming the opposing player is drawn from the same distribution as the previous opponents. All probability computation depends crucially on the information available, you just can't escape that. There are no universally true probabilities, unless you want to talk quantum mechanics here, but I guess you don't.
Statistics FTL. I spent like 10 minutes trying to ake a text-based decision tree and it was crazy getting the spacing right... but so far, I'd say about 69.3% of the statistics in this thread have been facetious.
Because the original number was apparently his probability to win a bo5 so therefore i approxiated the chance of the dongman's winning one match is 0.600197 EDIT: or was it? anyways going to sleep now, cant barely read lol
Playing against random players, from the same pool of opponents, Jaedong has a 81% chance of winning any given BO5. Flash, on the other hand, has a 71% chance. Note that this is based off of their 1v1 stats.
Jaedong has a higher chance of winning against any given player, meaning he is, statistically speaking, the better player. This means he will have a statistical advantage in a BO5.
Dividing Jaedong's percent chance of winning against any one person, and dividing it by the sum of the two player's percentages, we find that Jaedong has a 53% chance of winning against Flash.
Why?
Take a 100 BO5 sample of Jaedong. He should win 81 of them. Take a 100 BO5 sample of Flash. He should win 72% of them.
Total BO5's won -> 153. Jaedong won 81 of them.
The chance of him winning against Flash is therefore 81/153 = 53%
It's very crude, and simple, but it shows that they are both extraordinary players, and it really is anybody's game. It does not take into account their recent history or recent skill level, it takes into account every recorded game..... Note that this is only an approximation, and once the players have significant skill level divergence, the estimate gets to be really bad..... But a solution is still possible.
You can't use random players. Use Elo, imprecise as it is. Flash has a 2320 vs Z Elo. Jaedong has a 2201 vs T Elo. By the Elo formula (look it up on Wiki) Flash has a 66% chance to win a single game vs Jaedong.
Translate that into a Bo5 and Flash has a 78% chance of winning the series. Jaedong is a huge underdog.
On January 21 2010 05:46 jalstar wrote: You can't use random players. Use Elo, imprecise as it is. Flash has a 2320 vs Z Elo. Jaedong has a 2201 vs T Elo. By the Elo formula (look it up on Wiki) Flash has a 66% chance to win a single game vs Jaedong.
Translate that into a Bo5 and Flash has a 78% chance of winning the series. Jaedong is a huge underdog.
Do you have any idea what K value they use? It could totally skew the results if Flash was simply on a decent streak against decent players.....
For a good K value, it would take many, many games to come close to your true skill, and even then, you can rise above it, or below it, but you'll hover near it.
While it makes a lot of sense to use the ELO, in this case it doesn't. When used in Chess, hundreds and hundreds of individual games are taken into consideration. And, yes, they are against "random" players..... The way that the "random" part of it is taken into consideration is different though.
On January 21 2010 05:46 jalstar wrote: You can't use random players. Use Elo, imprecise as it is. Flash has a 2320 vs Z Elo. Jaedong has a 2201 vs T Elo. By the Elo formula (look it up on Wiki) Flash has a 66% chance to win a single game vs Jaedong.
Translate that into a Bo5 and Flash has a 78% chance of winning the series. Jaedong is a huge underdog.
Do you have any idea what K value they use? It could totally skew the results if Flash was simply on a decent streak against decent players.....
For a good K value, it would take many, many games to come close to your true skill, and even then, you can rise above it, or below it, but you'll hover near it.
While it makes a lot of sense to use the ELO, in this case it doesn't. When used in Chess, hundreds and hundreds of individual games are taken into consideration. And, yes, they are against "random" players..... The way that the "random" part of it is taken into consideration is different though.
How does your method of winning % make more sense than Elo?
On January 21 2010 00:52 qrs wrote: Add in the fact that we would expect his individual game-winning percentage to be lower in best-of-five's since he only faces those players who have been playing well enough to make it to the end of tournaments, and his series-winning percentage starts looking even more impressive.
This could be accounted for explicitly by calculating, for Jaedong, separate ELOs for series games and non-series games. I'm too lazy to do it though.
On January 21 2010 05:46 jalstar wrote: You can't use random players. Use Elo, imprecise as it is. Flash has a 2320 vs Z Elo. Jaedong has a 2201 vs T Elo. By the Elo formula (look it up on Wiki) Flash has a 66% chance to win a single game vs Jaedong.
Translate that into a Bo5 and Flash has a 78% chance of winning the series. Jaedong is a huge underdog.
Do you have any idea what K value they use? It could totally skew the results if Flash was simply on a decent streak against decent players.....
For a good K value, it would take many, many games to come close to your true skill, and even then, you can rise above it, or below it, but you'll hover near it.
While it makes a lot of sense to use the ELO, in this case it doesn't. When used in Chess, hundreds and hundreds of individual games are taken into consideration. And, yes, they are against "random" players..... The way that the "random" part of it is taken into consideration is different though.
How does your method of winning % make more sense than Elo?
I'm using their relative records against the same random pool, and comparing their abilities against that same random pool, to find who is more likely to win.
It is just an estimate, since both players are very, very good, and both have a very high win percentage.
It also means little at this point, because there have not been a large enough number of games to determine the percentage precisely. And the error bounds for any calculations would be huge at this stage.
We always judge player's abilities relative to those around them, and things have changed greatly since Jaedong first won a Bo5. If you wanted to produce any meaningful number (rather than just a spun out percentage that doesn't really tell us anything other than... wow Jaedong kicks ass in Bo5s... which you only need to look at his titles to see), you'd have to weight later wins more than earlier wins... seperate it by matchup and so on and so forth.
Ultimately, getting their statistics from their entire record of Starcraft is not going to be accurate. No matter how good their record is through their entire career, it's pretty much how they're playing currently, not how good they played a year ago.
On January 21 2010 01:01 RaGe wrote: you can't consider 4-1 or 5-0 cause that can't happen.
OP needs to take a combinatorics class.
Not quite. He's right, from a statistical analysis point of view..... Although they cannot physically happen, you can assume that they do to make the calculations much easier.
How about we assume they don't with combinatorics? You can use set theory.
On January 21 2010 01:01 RaGe wrote: you can't consider 4-1 or 5-0 cause that can't happen.
OP needs to take a combinatorics class.
Not quite. He's right, from a statistical analysis point of view..... Although they cannot physically happen, you can assume that they do to make the calculations much easier.
How about we assume they don't with combinatorics? You can use set theory.
No offense, but do you know what you are talking about? Beyond vague references to combinatorics and set theory, you have said nothing at all.
On January 21 2010 05:40 lMPERVlOUS wrote: Okay, look at it this way.
Playing against random players, from the same pool of opponents, Jaedong has a 81% chance of winning any given BO5. Flash, on the other hand, has a 71% chance. Note that this is based off of their 1v1 stats.
Jaedong has a higher chance of winning against any given player, meaning he is, statistically speaking, the better player. This means he will have a statistical advantage in a BO5.
Dividing Jaedong's percent chance of winning against any one person, and dividing it by the sum of the two player's percentages, we find that Jaedong has a 53% chance of winning against Flash.
Why?
Take a 100 BO5 sample of Jaedong. He should win 81 of them. Take a 100 BO5 sample of Flash. He should win 72% of them.
Total BO5's won -> 153. Jaedong won 81 of them.
The chance of him winning against Flash is therefore 81/153 = 53%
It's very crude, and simple, but it shows that they are both extraordinary players, and it really is anybody's game. It does not take into account their recent history or recent skill level, it takes into account every recorded game..... Note that this is only an approximation, and once the players have significant skill level divergence, the estimate gets to be really bad..... But a solution is still possible.
Dude how are you getting Flash's numbers? He has a higher overall win rate, a higher ELO and a higher TvZ win rate than Jaedong's ZvT win rate. Based purely on probability measures excluding map statistics he should be the favourite.
How does Flash have a 71% chance in a BO5? Taking his overall win rate of 70.25% with 5 trials in a binomial distribution, the probability of him having 2 or fewer losses (i.e he wins) is 84.02% which is greater than Jaedong's 81.36%.
But again, it's retarded to calculate this stuff... do you think you can analyze a game played 2 years ago (which these win rates take into account) and tell me from that how well the same player is playing today?
I think most people would agree that based on how well each player has been playing recently, Flash would be the favourite.
If that's wrong, the calculations I did were wrong.....
Without taking a considerable number of games into consideration, statistical analysis does not work anyways. By taking older games into consideration, statistical analysis can be more useful, however, it is limited in the sense that the recent games are the ones that matter, not the older ones.
If that's wrong, the calculations I did were wrong..... It sucks either way.....
That's based on Bo5 games alone. The problem is that both players have each played less than 20 Bo5s ever (and less than 10 Bo5s in the relevant matchup). That's not a wide enough sample of players to say that variations in opponent skill average out. Even whole-career statistics aren't good enough.
This thread got me thinking as I had done a similar calculation to compare Flash Vs. Movie in the OSL final (except using raw TLPD elo & a formula I found on wikipedia) and came up with a figure of >95% chance of Flash winning (which I thought was phenominal). I think this is a better method than comparing Flash's Vs Protoss to Movie's Vs Terran as these are not directly comparable.
Doing the same for the MSL finals gives a figure of 67.9% (over 2 out of 3) chance for flash to win the BO5 against Jaedong. Interestingly, against anyone else, it would be >80%. I think this is a better method (though still not accurate as it fails to consider so many variables).
On a side note, isn't it true that Flash is 3-0 vs Jaedong in Bo3s, while Jaedong is 2-0 vs Flash in Bo5s? Kinda interesting... Also, Flash beat Jaedong in the Ro8 of both Starleagues that he won. Jaedong won Bacchus 09, which was the sponsor of Flash's first starleague championship while Flash won EVER 09, which was the sponsor for Jaedong's first starleague championship.
Let's remember who JD lost to in another MSL BO5. Yes, that's right folk. A Terran. That Terran was ForGG. Do you really thing Flash's TvZ isn't as good as ForGG's? I know that I can't say that just because ForGG beat Jaedong that Flash will, but IMO Jaedong's worse match-up is ZvT. The statistics say so.
On January 22 2010 09:19 peidongyang wrote: Let's remember who JD lost to in another MSL BO5. Yes, that's right folk. A Terran. That Terran was ForGG. Do you really thing Flash's TvZ isn't as good as ForGG's? I know that I can't say that just because ForGG beat Jaedong that Flash will, but IMO Jaedong's worse match-up is ZvT. The statistics say so.
this post is garbage Jaedong broke out onto the scene BECAUSE of his monsterous ZvT
On January 22 2010 09:19 peidongyang wrote: Let's remember who JD lost to in another MSL BO5. Yes, that's right folk. A Terran. That Terran was ForGG. Do you really thing Flash's TvZ isn't as good as ForGG's? I know that I can't say that just because ForGG beat Jaedong that Flash will, but IMO Jaedong's worse match-up is ZvT. The statistics say so.
this post is garbage Jaedong broke out onto the scene BECAUSE of his monsterous ZvT
His ZvT has been his weakest match for a long time(almost a year). Even players with poor mechanics like fantasy push him to his limits both physically and mentally.
I wish things were different but jaedong definately has a chink in his armor and that is world class tvz play.
I still don't think forgg is a fair indictator of how jaedong will fair against flash.
You have no idea. I've gone up to a girl and used the line "Hey, can I be your first derivative, so I can lie tangent to your curves? If not, can I be your second derivative, so I can get a good feel of all of your concavities?"
It didn't work, but still. Definitely worth the shot.
I thought these were some interesting statistics.... In BO5 matches, JD's worst matchup statistically? ZvZ! With a win ratio of 66.7% (Out of a huge sample size of 3 series...) His best matchup? ZvT... at a whopping 90% win rate, with his ZvP close behind at 87.5%.
He hasn't lost a Bo5 yet this year, and has had 3 of them so far. At this rate, He will probably play and win, far more Bo5 series this year than his 2008 high of 7-1. JD truly is the best Bo5 player of all time...
On January 24 2010 20:48 Warrior Madness wrote: I thought these were some interesting statistics.... In BO5 matches, JD's worst matchup statistically? ZvZ! With a win ratio of 66.7% (Out of a huge sample size of 3 series...) His best matchup? ZvT... at a whopping 90% win rate, with his ZvP close behind at 87.5%.
He hasn't lost a Bo5 yet this year, and has had 3 of them so far. At this rate, He will probably play and win, far more Bo5 series this year than his 2008 high of 7-1. JD truly is the best Bo5 player of all time...
Jaedong's best matchup was ZvT when he first started owning people, then ZvZ in 08-09 (JvZ) and now it's ZvP.
I'd still pick Calm over him in a Bo5. JvZ is not what it once was and Calm's ZvZ has been beastly all year.
EPT
