|
On June 12 2012 22:45 BlueSpace wrote: 1) Cause and effect:
The OP has presented data which shows that a 10% gold lead by 12 minutes leads in 90% of the cases to win for the leading team. People here interpret this in a way which indicates that the rest of the game is "meaningless". You could also interpret this in way that the stronger team will manage to pull ahead early and this shows.
This is...false.
Randomness decreases as gametime increases. It also decreases as the number of decisions a team, as a whole, makes. The problem? Early game is the part of the game where teams make the fewest real decisions. In reality, a few (think 3-4) are turning the outcome of the game.
Particularly in pro games. Last hitting is pretty standard, so that isn't going to be a huge advantage (unless you got a kill, which is what probably happened to create this lead 1 kill, or 2), team comp could be giving you an advantage (your jungler could be faster), but usually its a bunch of RNG and highly guess-y things determining who has the lead at this point. Good examples of what I'm saying:
A successful or failed invade; a successful gank, perhaps 2; a countergank, or bait bottom.
Most of these are based on where the jungler is, and are very dependent on the fog of war, and getting good ward placement at this point in the game is nearly impossible. So, not only are you determining the outcome of the game during a small portion of the game, you are also determining the outcome during the period of the game that is most fundamentally random.
Thats why you see so many safe picks in pro games, particularly top and mid.
|
OP back from a long and boring business trip here and collecting some notes from those criticisms both constructive and otherwise in the backlog.
First thing's first: I am not a statistician. My mother is, oddly enough, but I am definitely far far from it. I welcome all helpful instructions on how to go about pursuing and collating this information. Rackdude in particular has been very very helpful getting me a rough outline of equations that will account for skill.
Next thing next: Perhaps my thesis was unclear. 12 minutes was an arbitrary choice of time to end the early game. IF the gold earned disparity was within the (ridiculously low) 10% tolerance at 12 minutes, I would note the time that the disparity diverged past the 10% point, hence the "lead taken at 14:30.... lead taken at 18min" comments. My secondary implied thesis was that the decisive, game-changing lead is taken BEFORE half the game is even complete. Therefore, the point in which the lead was taken should be AFTER half the game is done to be considered a tie... That secondary circumstance is awfully convoluted and confusing, so I am thinking of removing that stipulation completely, or clarifying it somehow. But to people looking at the date and being like, "a TIE!? WTF? what is a tie omgooses!"... ties and losses are what I was hoping for. They represent even or back-and-forth, and therefore interesting game.
Something that interests me further (particularly if I can get the VODs for reference in the future) would be to go back and take hard benchmarks of gold disparity at either every minute mark (oh my god the work!) or every 3 minutes. Were I to do this, I would definitely gameplan with a statistician as to how to incorporate this information efficiently. Sitting down with an old-fashioned just to watch LoL and take benchmarks is not my idea of a relaxing afternoon. Given that there are currently 10 series archived on the MLG site, going back and re-referencing might be impossible.
Basically, I began this because something seemed pretty predictable about watching LoL in a way that I don't often see in other pro e-sports. Thank you all for helping me in the pursuit of answers to this question! I welcome your help in trying to polish an end result for all this information (if there can be something that would hold up to scrutiny). In the coming days, this thread and whatever PMs you all want to send will be my primary resource for cleaning this information. I respond to and welcome any and all PMs so keep them coming! + Show Spoiler + even those wonderful "LOL NOOB. Thanks for proving the better team wins 90% of the time" ones
|
MLG spring finals game 5 shows how money isnt as important as good gameplay. TSM was leading by a big margin until they screwed up 1-2 teamfights and CLG got the first Baron. Up until the 26 minute mark TSM was leading by a solid amount in kills (6 vs 3), towers (5 vs 2) and with a 7k gold advantage (40k vs 33k). Then Jax got killed twice and it simply spiralled down from that. A big contributing factor might also be the fact that Soraka on TSM got two of those kills and she doesnt scale as well as a more offensive focused champion does, thus part of the money advantage was not as well invested as it could have been.
So the deciding factor isnt really money but rather NOT screwing up. Obviously "knowing how to play well" will lead to a decent gold / kill advantage, but at low level of play the gameplay isnt that much "on the knife's edge" as with these top teams and the items really get important. A very important part of "not screwing up" is communication and teamwork. Being a solo-killer will only get you so far, but if you manage to set up traps and synergies with other champions you will be able to get your kills much easier. So another aspect to not screwing up is actually champion selection for the synergy aspects.
|
On June 13 2012 07:15 Kronen wrote: OP back from a long and boring business trip here and collecting some notes from those criticisms both constructive and otherwise in the backlog.
First thing's first: I am not a statistician. My mother is, oddly enough, but I am definitely far far from it. I welcome all helpful instructions on how to go about pursuing and collating this information. Rackdude in particular has been very very helpful getting me a rough outline of equations that will account for skill.
Why don't you show your mother this and see what she thinks of this sentence:
"Imagine watching game 7 tonight and seeing the Celtics go up 45-40 late in the second quarter and being able to say with 90% certainty that they are going to the finals." - You still haven't listened to what we're saying. This is just straight up wrong.
You don't need an equation for skill, you need to take results properly. If you actually wrote down the exact score for every match at 12 minutes, regardless of whether it's close or not, that would be a good start. I'd happily sort your data out from there. At the moment though it's impossible to say anything since you've ignored close games.
|
On June 13 2012 17:56 Klive5ive wrote: I'd happily sort your data out from there. At the moment though it's impossible to say anything since you've ignored close games.
Excellent! I will be sorting the games available on the MLG site today by taking percentile difference in gold at every 3minute mark and will gladly take you up on your offer to sort data.
I'll keep you posted! As far as organizing the backlog of previous games, if someone has access to VODs for all the games, feel free to recompile that info.
On June 13 2012 15:54 Rabiator wrote: MLG spring finals game 5 shows how money isnt as important as good gameplay. ... So the deciding factor isnt really money but rather NOT screwing up. Obviously "knowing how to play well" will lead to a decent gold / kill advantage, but at low level of play the gameplay isnt that much "on the knife's edge" as with these top teams and the items really get important. A very important part of "not screwing up" is communication and teamwork.
The problems with this is that it's quite hard to chart "not screwing up". Umm, in some of the games I have noted rationale for a team coming from the early deficit (see DAY 1, CLG.NA vs everyone they played). But in an effort away from speculation and subjective analysis of gameplay I'm trying to steer away from that.
|
Good job Kronen. We all appreciate your hard work. Now comes the hard part, analyzing what this means. Is the gold number the cause of snowballing, or is that gold statistic the result of snowballing.
Just as an example, lets say a team catches on to this number and starts doing whatever they can to get a 10% or higher gold lead early on. Will that increase their chances of winning to 90% or not? This is a very testable hypothesis and I surely hope that some team catches on and tests this out.
|
On June 14 2012 01:57 hacpee wrote: Good job Kronen. We all appreciate your hard work. Now comes the hard part, analyzing what this means. Is the gold number the cause of snowballing, or is that gold statistic the result of snowballing.
Just as an example, lets say a team catches on to this number and starts doing whatever they can to get a 10% or higher gold lead early on. Will that increase their chances of winning to 90% or not? This is a very testable hypothesis and I surely hope that some team catches on and tests this out.
Thank you for the kind words... This helps alleviate the stress the fricking MLG player is causing.... I can't force the damn thing into a fullscreen mode that will stay open and I can't pull the game out and maximize it on monitor #2. Also.... I cant' scroll to exact timestamps on the player... so I usually have to sit and watch 30secs to a minute of gameplay to get to the exact time I need to collect information... Super fuckign annoying. And this says nothing of the times it randomly becomes unmuted and I have to listen to Phreak talk.... uuuggh... margarita time IMO...
here's the link to the spreadsheet: LoL VOD benchmarks Please give me suggestions if you think there's a better way to chart this info.
|
anyone have any information why the fucking MLG player occasioanlly goes to complete illegible shit sporadically? It's so fuckign bad I can't read text. Reloading occasionally helps the problem but not consistently.
|
On June 14 2012 01:57 hacpee wrote: Good job Kronen. We all appreciate your hard work. Now comes the hard part, analyzing what this means. Is the gold number the cause of snowballing, or is that gold statistic the result of snowballing.
Just as an example, lets say a team catches on to this number and starts doing whatever they can to get a 10% or higher gold lead early on. Will that increase their chances of winning to 90% or not? This is a very testable hypothesis and I surely hope that some team catches on and tests this out.
I'm not sure what you're getting at here. Clearly every team already tries to have a gold advantage, why would you *not* try to do so? Gold is used as a proxy variable for "success" here; it's easy to measure and interpret, and it takes into account many different factors that would be bothersome to analyze independently (e.g. kills, stolen buffs, towers, etc.).
The "snowballing" is due to the simple fact that performing well (gaining gold) gives you an advantage later in the game (improved itemization). I don't think anybody who has played the game will dispute that having more gold increases your chance to win. If this were not the case, last hitting would be completely useless. The only question is how easy it is to overcome an early disadvantage, and that is not a question of cause and effect, it is a question of constant factors (i.e., cost and gold efficiency of items).
I also don't see how your proposition tests any meaningful hypothesis. The statistics Kronen gave are just that, statistics; they are observations he made. There is no new causal link being proposed here. if your team can gain a significant gold advantage early -- while aiming for a "standard" team composition, as all the teams in the data set did -- there's a good chance that your team is actually playing better than the other team. That was the conclusion drawn from the observations. The only surprising part was how high the correlation between early gold leads and the game outcome was; the idea that teams with a gold lead are more likely to win is not surprising.
If you are trying to suggest that gold itself is the cause for the teams winning the game, that is a completely different hypothesis from what you are proposing to test.
To test that, arrange random matches, ask both teams to play normally *except that one team has to forfeit a certain number of last hits* until the gold discrepancy becomes large enough. Remove this requirement after that point and see how the game plays out. (You could even repeat the games with the role of the gold-forfeiting team reversed, though that isn't statistically kosher given that games aren't independent. But it's a starting point.)
That would tell you whether gold was the *reason* the teams won that often -- or whether other factors are in play, e.g. better teams using their skill advantage to gain a gold advantage. (I'm simplifying here, but it'd go in the right direction towards answering that question.)
Also, there is absolutely nothing in the data that suggests that having a *10%* gold advantage gives you a *90%* chance to win. The data does not suggest that, I suggest you reread the original statement (and some of the replies) if you believe it does. There were no numbers provided for the win rate given a 10% gold advantage. The choice of 10% was an arbitrary cut-off.
|
On June 14 2012 04:55 bmn wrote:Show nested quote +On June 14 2012 01:57 hacpee wrote: Good job Kronen. We all appreciate your hard work. Now comes the hard part, analyzing what this means. Is the gold number the cause of snowballing, or is that gold statistic the result of snowballing.
Just as an example, lets say a team catches on to this number and starts doing whatever they can to get a 10% or higher gold lead early on. Will that increase their chances of winning to 90% or not? This is a very testable hypothesis and I surely hope that some team catches on and tests this out. I'm not sure what you're getting at here. Clearly every team already tries to have a gold advantage, why would you *not* try to do so? Gold is used as a proxy variable for "success" here; it's easy to measure and interpret, and it takes into account many different factors that would be bothersome to analyze independently (e.g. kills, stolen buffs, towers, etc.). The "snowballing" is due to the simple fact that performing well (gaining gold) gives you an advantage later in the game (improved itemization). I don't think anybody who has played the game will dispute that having more gold increases your chance to win. If this were not the case, last hitting would be completely useless. The only question is how easy it is to overcome an early disadvantage, and that is not a question of cause and effect, it is a question of constant factors (i.e., cost and gold efficiency of items). I also don't see how your proposition tests any meaningful hypothesis. The statistics Kronen gave are just that, statistics; they are observations he made. There is no new causal link being proposed here. if your team can gain a significant gold advantage early -- while aiming for a "standard" team composition, as all the teams in the data set did -- there's a good chance that your team is actually playing better than the other team. That was the conclusion drawn from the observations. The only surprising part was how high the correlation between early gold leads and the game outcome was; the idea that teams with a gold lead are more likely to win is not surprising. If you are trying to suggest that gold itself is the cause for the teams winning the game, that is a completely different hypothesis from what you are proposing to test. To test that, arrange random matches, ask both teams to play normally *except that one team has to forfeit a certain number of last hits* until the gold discrepancy becomes large enough. Remove this requirement after that point and see how the game plays out. (You could even repeat the games with the role of the gold-forfeiting team reversed, though that isn't statistically kosher given that games aren't independent. But it's a starting point.) That would tell you whether gold was the *reason* the teams won that often -- or whether other factors are in play, e.g. better teams using their skill advantage to gain a gold advantage. (I'm simplifying here, but it'd go in the right direction towards answering that question.) Also, there is absolutely nothing in the data that suggests that having a *10%* gold advantage gives you a *90%* chance to win. The data does not suggest that, I suggest you reread the original statement (and some of the replies) if you believe it does. There were no numbers provided for the win rate given a 10% gold advantage. The choice of 10% was an arbitrary cut-off.
None of this addresses the main problem, which is that the first 12 minutes of a game are the most inherently random minutes of a game because of the inability of players to have good ward coverage.
Thus the snowballyness causes 2 other things that (IMO) make the game less fun overall. 1. Forces players to play overly safe. 2. Eliminates several champions from the "viable" pool because of inability to "play it safe" (Rammus is a great example).
|
The crunch for winners bracket rounds 2 and 3 are up on the spreadsheet... time for a break.
|
On June 14 2012 06:43 cLutZ wrote:
None of this addresses the main problem, which is that the first 12 minutes of a game are the most inherently random minutes of a game because of the inability of players to have good ward coverage.
Thus the snowballyness causes 2 other things that (IMO) make the game less fun overall. 1. Forces players to play overly safe. 2. Eliminates several champions from the "viable" pool because of inability to "play it safe" (Rammus is a great example).
My post did intentionally not try to make any statements on how balance should be or what should be changed, I was just commenting on what I see as a misunderstanding of what the observations do and do not actually say. Trying to mix general opinion on how a game should or should not be with incorrect conclusions drawn from statistical observations only makes things worse.
But if you insist: - I don't really see why people consider the fundamental "snowball" property a problem. It is a core element of what DotA/LoL-style games are: Being ahead allows you to buy more/better stuff, giving you an advantage for future engagements. - How easy it is or is not to recover from an early disadvantage is something I can't comment on. I haven't played DotA, so I can't draw those comparisons. I'm also not at a level of play where relatively small gold disadvantages feel stifling, and the OP's statistics in isolation say very little about this question, which was my main point.
You say that the snowballing "forces players to play overly safe", but I don't see how that is the case at all. If the game tends to snowball, your incentive to take a lead early is larger, not smaller. It increases both the penalty for bad moves and the reward for good moves.
Perhaps early combat is too random in LoL, which would indeed act as an incentive for better players to play more passively. Perhaps wards are too cheap, perhaps the jungler's location is too predictable -- but those are separate, unrelated issues to whether the game snowballs or not.
|
Would be cool to have a more controlled data set. I would use only the seeded teams (since you'd expected games involving non-seeded teams to be one-sided) and I would only count games where one team was 10% or more ahead by the 12 minute mark (you use a sliding scale where you sometimes include games where there was less of a lead or the lead was established later in the game).
LoL definitely has a few problems that can lead to snowballing, such as oracle's elixir helping you stay ahead, but I feel it's nowhere near as bad as your initial reading of the data seems to suggest.
|
United States47024 Posts
On June 14 2012 06:43 cLutZ wrote: 1. Forces players to play overly safe
I disagree with this.
What makes players play overly safe in LoL is that the cost of playing safe is simply too low compared to the rewards for taking risks. It's relatively easy and cheap to ward up your lane against ganks, and there are comparatively few methods for bypassing wards when ganking (most of them are either invis heroes that aren't viable in competitive play, or long-cooldown global/semi-global ultimates that a professional player can easily keep track of the the cooldowns for). On top of this, when you feel you are at risk in lane, you can often just sit at your tower, and wait for the creep wave to get to you. If the creep pressure is even or pushing toward you, you lose minimal farm waiting out the danger.
This is different from DotA, where you cannot play "safe". The reason for this is that it's simply not possible to play safe. Your team is limited on wards, so you cannot cover all the entrances to 1 lane, without having a deficit of wards for the other 2 lanes. Furthermore even with wards, your lane is not safe. There are many more invisibility/global/semi-global/just-plain-long-range abilities that are capable of ganking you past wards, and particularly when reduced night-time vision is involved, there are just a lot of ways that you can be initiated on from further than you can see (e.g. night time vision for most heroes is 900, Blink/Blink-like abilities typically can be cast up to 1200 range). There's also Smoke of Deceit, which just lets any enemy ganker bypass wards with a consumable. At the same time, you can't just camp your tower in the face of these threats, due to the absolutely suffocating lane control allowing your opponent to freely deny creeps allows--he can aggressively pull the lane back to his side.
The best comparison I can make is--consider the level of threat you feel when Nocturne or TF is 6. Now consider that all 10 players feel that risk 95% of the time during the laning phase in DotA. It's simply not possible to "play it safe" then, because you would be sitting at your tower, not get a point of XP or gold, and the creep wave would never come back to you. In this scenario, you HAVE to take risks to get gold/XP. This is where the interesting decision-making happens--taking intelligent risks and getting away with it, vs. taking unintelligent risks and getting punished for it.
|
On June 14 2012 08:28 DanielZKlein wrote: Would be cool to have a more controlled data set. I would use only the seeded teams (since you'd expected games involving non-seeded teams to be one-sided) and I would only count games where one team was 10% or more ahead by the 12 minute mark (you use a sliding scale where you sometimes include games where there was less of a lead or the lead was established later in the game).
No, this is *not* a controlled data set. You still don't control for the most important question: Whether the teams won because they were better, and having more gold being a side effect of being better, or whether the gold advantage was essentially unrelated to skill and that advantage caused them to win. And I'd guess (but that's just intuition) that the games where one team is much more skilled than the other are *more* useful for answering the question, not less useful (see explanation in parentheses below).
You want to control the data by creating a situation where either team is arbitrarily assigned a gold advantage and then plays it out from there. Since you can't just load save games and resume, there's no obvious easy way to do it, but you could fake it to some degree if you have teams willing to cooperate.
But as long as you just take games and observe without any way of controlling for skill, I don't see how you would be able to answer the important question mentioned earlier.
(You could probably try to answer the question if you had some giant repository of saved games played under controlled circumstances with many repeated team matchups without player/team changes; e.g. if teams spammed serious scrim games all day long and you had access to the outcomes. That way you could try to separately estimate "team skill" for a specific game a priori. But barring that, which doesn't seem realistic, I see no practical and meaningful way to do this: How well a team plays is a very unpredictable measure; every team's quality of play varies greatly from game to game, and even more so if you don't know the enemy's skill (different styles clashing; different team compositions also matter; etc.).
In fact, only looking at games where two roughly evenly matched teams play only makes things worse. If the teams are roughly evenly matched, you can't even make a guess on who should win a priori -- and if two teams are evenly matched, it's completely expected that the one with an early advantage wins; there's no way of getting useful information from the game outcome no matter which team had the initial advantage. If one team is clearly much worse than the other, and if the worse team has an early gold advantage, that is the kind of game where the predicted outcome between the two hypotheses (skill is paramount vs gold advantage is paramount) will differ, so those games will yield useful information. But even then, you'd need a lot of games to average out stuff like teams just having a bad day, etc.)
LoL definitely has a few problems that can lead to snowballing, such as oracle's elixir helping you stay ahead, but I feel it's nowhere near as bad as your initial reading of the data seems to suggest.
Snowballing is not a problem, it is a vital component of LoL and similar games!
Snowballing is the only reason there is any value in doing anything for gold, including last hitting, dragon, feeding kills to ad carry instead of support, and so on. It's not a problem, it's the whole basis of why the gameplay is as it is.
Removing snowballing completely trivial: Everyone starts at level 18 and gains an equal amount of gold over time, kills grant no experience or gold, and there are no jungle buffs. This would be boring as hell, but you'd solve the "problem".
Snowballing needs to be there to make it interesting (IMO): It greatly increases the stakes during the laning phase and it rewards high-level (and aggressive) play.
There may be a problem in the extent to which comebacks are difficult. That's a completely different game design question, a very important question, but a different one. As TheYango pointed out, there are many reasons DotA differs from LoL as to whether it pays off to play safe or not, but it's not because of the existence of snowballing.
|
New benchmarks are up! Wooo!
Chart of crunch can be found here and the subsequent quickie version with the percentile info and a clunky chart can be found here.
All of the VODs benchmarks are up except for the Finals. The most interesting statistic is that in all but 1 circumstance, the team that wins is never behind by more than 7%. I repeat, in the current subset of information, if a team falls behind by more than 7% (even less than my inital 10% supposition), chances are greatly increased that you're going lose. The only exception is a mystifying game Dignitas and CLG NA. Need to study that one more. But it's really really interesting.
On June 14 2012 10:50 bmn wrote: Snowballing is not a problem, it is a vital component of LoL and similar games!
Snowballing is the only reason there is any value in doing anything for gold, including last hitting, dragon, feeding kills to ad carry instead of support, and so on. It's not a problem, it's the whole basis of why the gameplay is as it is.
Removing snowballing completely trivial: Everyone starts at level 18 and gains an equal amount of gold over time, kills grant no experience or gold, and there are no jungle buffs. This would be boring as hell, but you'd solve the "problem".
There may be a problem in the extent to which comebacks are difficult. That's a completely different game design question, a very important question, but a different one. As TheYango pointed out, there are many reasons DotA differs from LoL as to whether it pays off to play safe or not, but it's not because of the existence of snowballing.
Perhaps we are misunderstanding each other. I'm treating snowballing as the mechanism that causes comebacks to be all but non-existent in LoL. Your saying "Snowballing is the only reason there is any value in doing anything for gold," makes me thinkg you're mistaking snowballing for just general good intelligent play. But let me clarify... I'm here because I find LoL too predictable and I wanted to see if there's math that can back up (or squash undeniably) my feelings. Having a game in which it is nigh on impossible to come back is a problem.
I'm not here to speculate on what causes the problem or how to fix it. I'm just trying to benchmark games and see if the problem exists and how it develops.
|
I'm curious about what happens when you throw out all the 2-0 series as "one team was TOO much better than the other team, and is obviously going to win before the game is even started"
The core of the issue is not that we're interested in how likely a comeback is when one team is significantly worse than another. We need to look at what happens when the teams are at least reasonably close in skill. TSM is always going to stomp a team that is far worse than they are, even within the first 12 minutes. It might happen that they botch early game and make a comeback, but that requires them to somehow lose the early game against a significantly worse team - it's not going to happen often.
That phenomenon isn't exclusive to LoL either - It's not very interesting when you have the #1 team in any sport stomping some of the lower teams in teh league, and leading by a huge margin halfway through the game (I love you Cubbies, but I'm looking at you here.) TSM vs CLG comebacks are what make watching LoL interesting, and those are not limited to 10% of games between top teams. Need to measure more against even teams. While we can't necessarily rate skill numerically, we CAN say where it's "roughly even", such as comparing numbers in matchups where the teams have relatively even win/loss rate against each other.
|
On June 14 2012 14:01 sylverfyre wrote: I'm curious about what happens when you throw out all the 2-0 series as "one team was TOO much better than the other team, and is obviously going to win before the game is even started"
The core of the issue is not that we're interested in how likely a comeback is when one team is significantly worse than another. We need to look at what happens when the teams are at least reasonably close in skill. TSM is always going to stomp a team that is far worse than they are, even within the first 12 minutes. It might happen that they botch early game and make a comeback, but that requires them to somehow lose the early game against a significantly worse team - it's not going to happen often.
That phenomenon isn't exclusive to LoL either - It's not very interesting when you have the #1 team in any sport stomping some of the lower teams in teh league, and leading by a huge margin halfway through the game (I love you Cubbies, but I'm looking at you here.) TSM vs CLG comebacks are what make watching LoL interesting, and those are not limited to 10% of games between top teams. Need to measure more against even teams. While we can't necessarily rate skill numerically, we CAN say where it's "roughly even", such as comparing numbers in matchups where the teams have relatively even win/loss rate against each other.
Now you're trying to figure out how to skew the numbers. =P
Take the top 5 for a specific tournament, only count games they played vs each other, do that for a time period of one year. That'd be as close as you could get to "even skill" while maintaining a somewhat healthy sample size.
|
On June 14 2012 10:50 bmn wrote: In fact, only looking at games where two roughly evenly matched teams play only makes things worse. If the teams are roughly evenly matched, you can't even make a guess on who should win a priori -- and if two teams are evenly matched, it's completely expected that the one with an early advantage wins; there's no way of getting useful information from the game outcome no matter which team had the initial advantage. I feel like you missed something here. Perhaps you assumed perfect play (or perfectly equally flawed play) from both evenly matched teams?
If two teams are evenly matched and one team manages to get a 10% gold lead, then that means there is a fluctuation of gold of at least 10% in a game between two evenly matched teams. Therefore, it should be reasonable to assume that the fluctuation of gold should fluctuate back to the losing team a certain percentage of the time and will sometimes result in a comeback. There is also the possibility of winning a game with a gold deficit. Either way, a team with a 10% gold deficit should sometimes be able to beat a team of equal skill. The % they should be able to do that is subjective to viewer opinions, but should be somewhere between 0 and 50% non-inclusive.
To many of the other responders, yes, the testing methodology is flawed. A more controlled environment or a formula adjusting for team skill would almost definitely result in less than the 90% win rate of this testing methodology. However, the final number is not the key component of this test. The testing methodology does exactly what it needs to do in a very simplistic and easily repeatable way: it gives us a number by which we can compare to future events.
What we have here is not a purely statistical endeavor. Instead, we have a mixture of statistics and viewer opinion. The viewer opinion is that there are too few comebacks. When combining the data and the viewer opinion, there is a feeling that a 90% win rate based on this test is too high. You can claim the data to be flawed in many ways, but this data is repeatable and will tell you something important when compared to future tests using the same methodology and similar circumstances. If Riot chooses to implement a new feature or make changes to the game to create less snowballing effect, it should show up by using the same test at a future MLG. If it doesn't, then it probably means that whatever they implemented did not lower the snowballing effect by a significant amount.
|
2 tournaments are coming up this weekend GESL and Dreamhack. Enough data to test this hypothesis. I generally agree, when a team is down early in gold it will snowball hard since there is a lack of hard cc like in dota.
|
|
|
|