|
On June 10 2012 13:05 ocelotter wrote:
This would be even more interesting if data was gathered for each individual minute, not just the 12th minute (why was 12 chosen? seems arbitrary).
12 min was a completely arbitrary choice on my part. The rationale being that by 12min the early game is usually over, and the game usually isn't halfway done as the average game is over 25min.
To answer your question data is gathered at 12min and every single minute after. I note all lead changes after that point in time up until the end of the game.
Something I'm considering if really get serious about data crunching (if I get some help and guidance to do it properly) would be to take flat percentile gold disparity checks every 3minutes on the dot and chart the overall progression. That could prove interesting. But for right now, given that I'm away from a computer traveling for the next 3 days, it's going to be a struggle to get the last day of benchmarks up regardless.... Ugh.
|
On June 10 2012 08:55 TheYango wrote:Show nested quote +On June 10 2012 08:54 Takkara wrote:
It's weird though to say things like there is an "issue if 2/3 of the average game time is spent only determining 10% of the final outcome". I don't think these numbers come close to really saying that. LoL has been criticized more in the past for allowing too many comebacks rather than too few. Certainly snowballs happen. But there are plenty of 10% gold games that are not snowballed.
It has? I don't remember that ever being the case.
Well I dunno about criticism, but there have been a lot of balance changes both increasing and decreasing the amount of snowballing. LoL was originally designed with ~30 minute games in mind (for 5v5), and the amount of snowballing is one way to affect the average game time.
Increase - introduction of snowball items, spell/buff/elixir scaling (incl. Nasus/Veigar/Sion's infinite scaling), scaling towers/creep stats, lower death penalties on stacks, lowering maximum kill streak bounties,
Decrease - kill streak bounties, diminishing returns on repeat kills, delayed spawns on epic monsters
etc
Basically when games were lasting too long, adding more snowballing could shorten them -- there are probably other ways to augment game length than snowballing (the only ones I can think of revolve around map objectives) to be explored.
edit (misunderstood Takkara's post): I guess the thesis should be tweaked, since the correlation is still just between a lead and a victory. One way to try factoring out skill differences would be to chart the extended matchups between two teams (such as only analyzing CLG vs TSM over multiple tournaments) to get a "skill ratio" between the two, and apply that to the number of games decided at 12 minutes, however recent player swaps would still augment that data.
|
you would have to filter out the skill differences of the teams playing, because those corrupt your data, they lead it in a way that your theory is supported!
If a team is more skilled, it will lead at minute 12 because of skillful play and then continue to get advantages and win becuse of skillful play. This will corrupt all your data in a way that your thesis is supported.
Since there is high chance that 2 teams out of all those invited into MLG possess significantly different skill, data ís probably not useable to prove your thesis.
those statisics, imo, are only useable to try to prove non-snowballing.
For the opposite you cant really make a statement.
you would need a lot of statistical magics to repair the data, the endresult would have a lot of cases and a lot of % statements
|
15-2-1
That is the score (from your Reddit post as best I could work it out - it was a bit hard to follow? Maybe I'm just really tired but it seemed odd to me...) consider just games with Bo3 fully played out (ie a 2-1 score).
That is an even bigger disparity than before? (Ignoring ties). Definitely makes a strong case for snowballing in LoL in my opinion. And sort of debunks the 'more skilled' team winning issue people are having.
|
On June 12 2012 01:28 LaNague wrote:
you would have to filter out the skill differences of the teams playing, because those corrupt your data, they lead it in a way that your theory is supported!
If a team is more skilled, it will lead at minute 12 because of skillful play and then continue to get advantages and win becuse of skillful play. This will corrupt all your data in a way that your thesis is supported.
Since there is high chance that 2 teams out of all those invited into MLG possess significantly different skill, data ís probably not useable to prove your thesis.
those statisics, imo, are only useable to try to prove non-snowballing.
For the opposite you cant really make a statement.
you would need a lot of statistical magics to repair the data, the endresult would have a lot of cases and a lot of % statements
??? You just need a good skill proxy and control for it.
|
On June 12 2012 02:16 rackdude wrote:Show nested quote +On June 12 2012 01:28 LaNague wrote:
you would have to filter out the skill differences of the teams playing, because those corrupt your data, they lead it in a way that your theory is supported!
If a team is more skilled, it will lead at minute 12 because of skillful play and then continue to get advantages and win becuse of skillful play. This will corrupt all your data in a way that your thesis is supported.
Since there is high chance that 2 teams out of all those invited into MLG possess significantly different skill, data ís probably not useable to prove your thesis.
those statisics, imo, are only useable to try to prove non-snowballing.
For the opposite you cant really make a statement.
you would need a lot of statistical magics to repair the data, the endresult would have a lot of cases and a lot of % statements ??? You just need a good skill proxy and control for it.
But analyzing a single tournament (or even tournaments at all) is misleading because some teams will appear more often in the data set. A team that plays in such a way that supports the hypothesis (i.e. taking a lead and keeping it until victory) will skew the data, because winning ensures both staying in the sample and more supportive data : \
|
How did you come up with the 12 minute control? Just to trying to chime in some constructive criticism.
Because it's so arbitrary, there can be any point in the game where this happens; and actually, I think if you analyze and cherry-pick certain points of time, you might find information that contradicts this.
For this to be truly valuable, I think you might need to collect gold information throughtout the entire game, perhaps in 1-minute increments. Then analyze for how long the lead was kept, and cross-analyze what points of time intersect.
Great initiative though 
|
On June 10 2012 09:37 TheYango wrote:Show nested quote +On June 10 2012 09:22 Zato-1 wrote:
OP: Statistically speaking, the only fault I find with your data is that you can't just assume equal skill levels. These lopsided stats could just as well (theoretically) be explained by vast differences in skill, where the more skilled team pretty much always wins the game and 90% of the time has a 10% (or more) lead in gold by minute 12. If you could somehow get past this obstacle and show that you're comparing teams which are more or less even in skill, then I'd have no objections.
e.g. if you had stats for a best-of-49 series of LoL games between two teams, where one team wins 25 games and the other team wins 22 games, and in 90% of all of these games the team that was 10% ahead in gold by minute 12 wins, then I'd agree that snowballing is a problem at the skill level of these teams. Even accounting for that, the difference between predicted/actual result should be more than 10%. Consider this: in a "healthy" game, you want to see diverse and innovative play. Particularly in lopsided matchups, you should expect to see weaker teams prepare innovative strategies and drafts against the stronger teams. Inherently this "cheese factor" should create a divergence of greater than 10%, even if EVERY game is lopsided, because you would expect cheesy/innovative strategies to have significantly better than a 10% winrate. The fact that there's such a narrow divergence means one of two things: - "Cheese" games result in lopsided results anyway (either they're too successful and the cheesing team wins too often, or they're not successful enough, and the stronger team wins anyway) - Not enough "cheese' games are being played--people are, for whatever reason, always opting for the safe strategies, even in super-lopsided games where cheesing should give them a better shot at winning Neither of these necessarily signify a problem with the game's design, but they are issues that should be investigated--if people are opting not to play cheesy/innovative strategies, or they're not being successful enough, we should take care to try and understand why this is the case.
You're missing something.
A very narrow divergence could also mean that it's comparatively easy to not make big mistakes. Basically (yes, sorry, I'm bringing the whole skill-ceiling thing into this again to some degree) what those numbers can mean is that once a team is ahead it's very hard to give up that lead.
Whoever is - even slightly - ahead has a very easy time exploiting that lead while the enemy team has a very hard time coming back. The small advantages that slight lead creates are "too big" from a design perspective to allow comebacks.
I don't consider the cheese explanation valid because that should be something that evens out in the long run (not neccessarily over one tournament, but certainly over one year) due to the fact that cheese > greedy play > safe play > cheese. If these don't even out (or aren't used) it's again either a design problem or people are plain and simply too dumb to do it properly. I'd bet on design though when it comes to League. =P
|
The main problem with LoL in comparison to sports, is that getting an advantage in lol translates to getting an advantage later on.
in basketball/soccer/hockey etc etc. getting a point doesn't make it so you're more likely to get a point later on.
major fixes to snowballing is going to be making EXTREMELY strong midgame items for cheap. For example a haunting guise that costs half of what it costs now, or ionic spark getting its price reduced, and buffing 5gp10 item stats to make them cost effective.
|
we don't need statistics for that. everyone knows snowball exists. I think the root of the problem is items>>>>experience. Once one team gets core items its impossible for other team stop them unless they have same items.
I think riot needs to rework some of the item-gold stuff. Make items very expensive or reward with less gold champion kills and more gold for jungle mobs. Or they should outright nerf all items by 30% I guess.
|
I don't know the game all that well in terms of the competitive setting, but does this also affect DotA play as well?
Or is there something inherent to DotA that prevents this kind of snowball-y play? If not, then it's a problem of the genre, not design.
|
United States47024 Posts
On June 12 2012 04:25 WaveofShadow wrote:
I don't know the game all that well in terms of the competitive setting, but does this also affect DotA play as well?
Or is there something inherent to DotA that prevents this kind of snowball-y play? If not, then it's a problem of the genre, not design.
DotA has certain common issues, but there are also distinct mechanics that contribute to make the game much less snowball-y than LoL, such as:
- TP scrolls, buybacks, and SC/WC3 high-ground mechanics creating a stronger defender's advantage, allowing the losing team more possibilities to come back
- significant tradeoff between gold-efficiency and slot-efficiency on items--small midgame items are often much more cost-effective than their lategame counterparts, allowing room for a team that's behind on gold to make timing-based plays, exploiting the buildup on larger items
- limited wards and Smoke of Deceit, meaning that if a team has an advantage, they can't ward up the entire map and have absolute map dominance
|
On June 12 2012 05:15 TheYango wrote:Show nested quote +On June 12 2012 04:25 WaveofShadow wrote:
I don't know the game all that well in terms of the competitive setting, but does this also affect DotA play as well?
Or is there something inherent to DotA that prevents this kind of snowball-y play? If not, then it's a problem of the genre, not design. DotA has certain common issues, but there are also distinct mechanics that contribute to make the game much less snowball-y than LoL, such as: - TP scrolls, buybacks, and SC/WC3 high-ground mechanics creating a stronger defender's advantage, allowing the losing team more possibilities to come back - significant tradeoff between gold-efficiency and slot-efficiency on items--small midgame items are often much more cost-effective than their lategame counterparts, allowing room for a team that's behind on gold to make timing-based plays, exploiting the buildup on larger items - limited wards and Smoke of Deceit, meaning that if a team has an advantage, they can't ward up the entire map and have absolute map dominance
You missed the most important thing in my eyes, which is the ability to prevent the carry from reaching their next item by killing them and taking their gold away.
If you kill the AD carry in LoL, ya that's great, but he still has the same amount of money, and is barely behind in when he will get that next big item. Where in DotA you can bring a carry saving for radiance (3800 takes a while to get) down a LOT of gold by taking them out repeatedly preventing them from snowballing even if they got some advantage early.
|
but aren't carries in dota like 10 times more powerful than carries in lol anyway?
|
On June 12 2012 02:42 r.Evo wrote:Show nested quote +On June 10 2012 09:37 TheYango wrote:On June 10 2012 09:22 Zato-1 wrote:
OP: Statistically speaking, the only fault I find with your data is that you can't just assume equal skill levels. These lopsided stats could just as well (theoretically) be explained by vast differences in skill, where the more skilled team pretty much always wins the game and 90% of the time has a 10% (or more) lead in gold by minute 12. If you could somehow get past this obstacle and show that you're comparing teams which are more or less even in skill, then I'd have no objections.
e.g. if you had stats for a best-of-49 series of LoL games between two teams, where one team wins 25 games and the other team wins 22 games, and in 90% of all of these games the team that was 10% ahead in gold by minute 12 wins, then I'd agree that snowballing is a problem at the skill level of these teams. Even accounting for that, the difference between predicted/actual result should be more than 10%. Consider this: in a "healthy" game, you want to see diverse and innovative play. Particularly in lopsided matchups, you should expect to see weaker teams prepare innovative strategies and drafts against the stronger teams. Inherently this "cheese factor" should create a divergence of greater than 10%, even if EVERY game is lopsided, because you would expect cheesy/innovative strategies to have significantly better than a 10% winrate. The fact that there's such a narrow divergence means one of two things: - "Cheese" games result in lopsided results anyway (either they're too successful and the cheesing team wins too often, or they're not successful enough, and the stronger team wins anyway) - Not enough "cheese' games are being played--people are, for whatever reason, always opting for the safe strategies, even in super-lopsided games where cheesing should give them a better shot at winning Neither of these necessarily signify a problem with the game's design, but they are issues that should be investigated--if people are opting not to play cheesy/innovative strategies, or they're not being successful enough, we should take care to try and understand why this is the case. You're missing something. A very narrow divergence could also mean that it's comparatively easy to not make big mistakes. Basically (yes, sorry, I'm bringing the whole skill-ceiling thing into this again to some degree) what those numbers can mean is that once a team is ahead it's very hard to give up that lead. Whoever is - even slightly - ahead has a very easy time exploiting that lead while the enemy team has a very hard time coming back. The small advantages that slight lead creates are "too big" from a design perspective to allow comebacks. I don't consider the cheese explanation valid because that should be something that evens out in the long run (not neccessarily over one tournament, but certainly over one year) due to the fact that cheese > greedy play > safe play > cheese. If these don't even out (or aren't used) it's again either a design problem or people are plain and simply too dumb to do it properly. I'd bet on design though when it comes to League. =P
I don't know what your argument is?
The OP is correct, there is something strange about these results. A lead at 12 mins (basically the 1/3 or 1/4 mark in the game) should not be translating to much into victories. This is because the longer the game goes on the more randomness should be smoothed out. 12 Minutes is basically just reflecting the laning phase. A stronger team should win even if it modestly lost in the laning phase.
Look at this win probability chart for the super bowl.
http://live.advancednflstats.com/
What the OP is saying is, if the NFL was LOL, if at halftime the pats are winning 10-9 they have a 90% chance of winning. That shouldn't make sense.
|
On June 10 2012 08:44 TheYango wrote:Show nested quote +On June 10 2012 08:41 zulu_nation8 wrote:
This doesn't apply at all to solo queue though. Who gives a shit about solo queue?
Ranked Solo Q is the farthest the majority of players will ever see in LoL, including myself. Therefore Solo Q is very important to us.
|
In my opinion the best way to deal with this would be to introduce many more Doran-like items that are really really strong for their cost but are slot-inefficient.
It might not be enough but it'd be a step in the right direction
|
Wow; for TL I would've expected a better understanding of stats.
"The team that takes as little as a 10% gold lead by 12 minutes wins over 90% of the time"
The key word is in bold. What that means is the stats take into account HUGE advantages at 12 minutes too.
That means you cannot infer that if a team is up 10% they will win 90% of the time!!!
You would have to take stats for teams up exactly 10% and no more to find the expected win percentage.
|
On June 10 2012 07:34 Kronen wrote:Yea, it is impossible for me to account for "skill" of a team given that that metric is entirely subjective. You could make the argument seeds are important, but that's odd too. Ultimately, the numbers should work out closer as the brackets progress. But, I'm speculating if they won't. Even split series have shown the snowball effect. It's just a matter of who gets the lead first. In series which go to 3 games, you'd think that the snowball factor wouldn't play as large of a factor though too because the "skill" factor would be more evenly matched. Show nested quote +this, the better team is gonna be the better team from minute 1 to minute 35 to minute x... probably about 90% of the time  No, not necessarily. That just means that the sport you're watching is designed in such a way that any lead is a foregone conclusion. That is a problem for your sport because it is inherently predictable (and therefore boring). In any other sport it's possible to come back from an early deficit through skill and game elements. I'm currently hunting for soccer and basketball statistics for some comparison, but even DOTA functions differently. Gold and xp changes in dota happen very frequently. Watching any other sport or esport there is give and take. Leads are lost and gained. Like the basketball example above, if a sport becomes that predictable there's something very very wrong.
Yeah but in basketball you can't buy items that make you faster and stronger with the baskets you make. The answer is in the title- snowballing. The better team gains a lead at the beginning and will always be ahead in items
So I agree with your sentiment of "not necessarily" and think your reasoning is solid, but I still believe that the better team is just going to outfarm, outbuy, and ultimately outplay their opponents most of the time
I could also, be wrong. There are champions that can't farm against certain others (a mid kassadin getting denied a lot) or bad matchups or one player underperforming which represent a number of cases where a lead can either be blown or a comeback made, or a win despite having less gold.
There are so many champs being played by so many gamers it's hard to map it out (imo) from just raw data but i think by and large the better team just knows how to farm more.
Sorry I know I sort of carry on and don't make clear points all of the time but thanks for reading
|
On June 12 2012 05:45 Klive5ive wrote:
Wow; for TL I would've expected a better understanding of stats.
"The team that takes as little as a 10% gold lead by 12 minutes wins over 90% of the time"
The key word is in bold. What that means is the stats take into account HUGE advantages at 12 minutes too.
That means you cannot infer that if a team is up 10% they will win 90% of the time!!!
You would have to take stats for teams up exactly 10% and no more to find the expected win percentage.
No, you're putting words in the OP's mouth(as did others).
If any team takes a lead by 12min, they are much much more likely to win said game. The game is setup in such a way that any early lead is basically insurmountable except by human error and throwing a game with sub-standard play.
That is his original statement; the statistics support that going down to 'as little as a 10% lead at 12 minutes.'
Obviously some of that 90% win rate is easily predictable(a huge early game lead), but that ANY lead of 10% or more leads to a win 90% of the time, including games which are part of a competitive series(as shown by someone else), is staggering, even if the 'only 10%' wins are a fairly small number of said victories. Any way you slice it, it means that comebacks, large and small, are not happening.
|
|
|
|
|
|
EPT
