|
On August 22 2014 02:55 GrandInquisitor wrote:
Dominance Factor, where kills are +2, deaths are -3, and assists are +1. A DF of 0 means going even. I think it is a better metric than KDA, and statistically it is more closely correlated with whether I win or lose.
but by far your worst champ with a 23% win rate has almost 0 dominance factor and your champ with 50/50 win rate has by far the highest dominance factor. in low amounts of games played that could be coincidence but since this is your top 5 champs surely that proves, at least for you, that its literally the worst metric ever.
|
GrandInquisitor
New York City13113 Posts
On August 22 2014 03:05 turdburgler wrote:Show nested quote +On August 22 2014 02:55 GrandInquisitor wrote:
Dominance Factor, where kills are +2, deaths are -3, and assists are +1. A DF of 0 means going even. I think it is a better metric than KDA, and statistically it is more closely correlated with whether I win or lose. but by far your worst champ with a 23% win rate has almost 0 dominance factor and your champ with 50/50 win rate has by far the highest dominance factor. in low amounts of games played that could be coincidence but since this is your top 5 champs surely that proves, at least for you, that its literally the worst metric ever.
...?
Obviously KDA or any KDA-related statistic will never be perfectly correlated with whether the team wins or loses. But how is it illogical that the champion I have the worst DF on is also the champion I am least likely to win the game with?
|
I have done econometrics that related how temperature variance affected the crime rates and looked like it was a good metric but then it really wasnt. xD
It is a homocedasticity problem, if you have good df / kda you are more likely to win, but also, if you are winning by other factors, you will get a good df / kda so not a good metric 
Im sure that earlier hours of the day relate to more wins to me... I should test that relationship xD
|
because on 40% of the champions listed, the DF was literally miles from anything remotely coherent. just because it happened to be lowest doesnt make it close to correct.
and if you were trying to improve your DF it would incentivize you to chase kills rather than take objectives, which is literally the opposite of what you should normally do. when dragons and barons and towers are all worth considerably more than kills, how does a system that only measures kill show value to your team?
seems like a silly system
|
On August 22 2014 03:05 turdburgler wrote:Show nested quote +On August 22 2014 02:55 GrandInquisitor wrote:
Dominance Factor, where kills are +2, deaths are -3, and assists are +1. A DF of 0 means going even. I think it is a better metric than KDA, and statistically it is more closely correlated with whether I win or lose. but by far your worst champ with a 23% win rate has almost 0 dominance factor and your champ with 50/50 win rate has by far the highest dominance factor. in low amounts of games played that could be coincidence but since this is your top 5 champs surely that proves, at least for you, that its literally the worst metric ever.
having a high DF on a top laner means you probably won your lane and participated in fights and stuff
junglers like jarvan only get lots of kills early on and then falls off sort of later so even with a high DF can lose
|
On August 22 2014 03:18 Usagi wrote:I have done econometrics that related how temperature variance affected the crime rates and looked like it was a good metric but then it really wasnt. xD It is a homocedasticity problem, if you have good df / kda you are more likely to win, but also, if you are winning by other factors, you will get a good df / kda so not a good metric Im sure that earlier hours of the day relate to more wins to me... I should test that relationship xD
I'm pretty sure that playing on the weekends/school holidays result in more losses for me xD
|
GrandInquisitor
New York City13113 Posts
On August 22 2014 03:19 turdburgler wrote:
because on 40% of the champions listed, the DF was literally miles from anything remotely coherent. just because it happened to be lowest doesnt make it close to correct.
and if you were trying to improve your DF it would incentivize you to chase kills rather than take objectives, which is literally the opposite of what you should normally do. when dragons and barons and towers are all worth considerably more than kills, how does a system that only measures kill show value to your team?
seems like a silly system
Where did you get the idea that DF is supposed to represent my value to my team? It's intended as a substitute for KDA. KDA thinks that going 1/2/3 is breaking even and equivalent to 1/1/1. DF thinks 1/2/3 is performing subpar and worse than 1/1/1. I have generally found that me going 1/2/3 is worse for my team than going 1/1/1, so I use DF alongside KDA. Any metric that considers only your K, D, and A is subject to limitations, but I think this is the best among those metrics.
|
If you want I can try to compute the best linear combination for k d a for predicting win/loss over millions of games.
|
GrandInquisitor
New York City13113 Posts
I'd like to see that, stats from a larger data set. I did some experiments a while ago in another GD thread and came to the conclusion that it likely heavily depends on your role.
|
On August 22 2014 03:19 turdburgler wrote:
because on 40% of the champions listed, the DF was literally miles from anything remotely coherent. just because it happened to be lowest doesnt make it close to correct.
and if you were trying to improve your DF it would incentivize you to chase kills rather than take objectives, which is literally the opposite of what you should normally do. when dragons and barons and towers are all worth considerably more than kills, how does a system that only measures kill show value to your team?
seems like a silly system
Actually, no. I'm pretty sure most of us here know that chasing kills generally ends up in a negative result for your team most of the time. If a chased kill is so easy to secure, the player would do it normally anyway. If the chase even ends up in a trade, it'd have a net DF of -1. To pursue risky chases in effort to increase your own DF, it's far more likely that you end up in a negative situation.
If anything, the negative result of a kill trade actually adds more caution to foolish chases.
|
To make a KDA a reasonable metric, you would have to divide each assist up through the amount of people getting an assist. If you are the only one getting an assist for a kill, the kill probably wouldn't have happened if it weren't for you. If you just randomly get an assist on some guy everyone gets an assist on, that is worth a lot less.
|
On August 22 2014 04:00 Simberto wrote:
To make a KDA a reasonable metric, you would have to divide each assist up through the amount of people getting an assist. If you are the only one getting an assist for a kill, the kill probably wouldn't have happened if it weren't for you. If you just randomly get an assist on some guy everyone gets an assist on, that is worth a lot less.
I believe Spellsy made up a stat on an Ongamers article called a "Kill Share" that did that.
|
On August 22 2014 03:43 GrandInquisitor wrote:I'd like to see that, stats from a larger data set. I did some experiments a while ago in another GD thread and came to the conclusion that it likely heavily depends on your role.
I can tell you for sure it is champion dependent.
Champions like Katarina has high avg death per game on games she wins than a lot of champions have for the games they lose. Furthermore Anivia, for example, has very low deaths per game.
|
On August 22 2014 04:00 Simberto wrote:
To make a KDA a reasonable metric, you would have to divide each assist up through the amount of people getting an assist. If you are the only one getting an assist for a kill, the kill probably wouldn't have happened if it weren't for you. If you just randomly get an assist on some guy everyone gets an assist on, that is worth a lot less.
Yeah.
A system with % kill contribution, etc. Might make more sense. But there are some technical difficulties grabbing every player's KDA in a game.... so the data is a bit sparse on those.
|
On August 22 2014 05:01 Sufficiency wrote:I can tell you for sure it is champion dependent. Champions like Katarina has high avg death per game on games she wins than a lot of champions have for the games they lose. Furthermore Anivia, for example, has very low deaths per game.
And a successful Katarina will pile on the kills to go with her deaths.
On August 22 2014 05:08 Sufficiency wrote:Show nested quote +On August 22 2014 04:00 Simberto wrote:
To make a KDA a reasonable metric, you would have to divide each assist up through the amount of people getting an assist. If you are the only one getting an assist for a kill, the kill probably wouldn't have happened if it weren't for you. If you just randomly get an assist on some guy everyone gets an assist on, that is worth a lot less. Yeah. A system with % kill contribution, etc. Might make more sense. But there are some technical difficulties grabbing every player's KDA in a game.... so the data is a bit sparse on those.
The real issue is that you have to track every kill as it happens. There's 0 ability to reverse engineer it.
|
Kill contribution can be calculated post game by looking at everyone's kda.
|
On August 22 2014 05:10 Sufficiency wrote:
Kill contribution can be calculated post game by looking at everyone's kda.
From what I understood, he wasn't talking about kill contribution. He was talking about splitting up assist in a kill, where 1k1a results in the assister getting full credit for an assist, with 1k2a giving 1/2 credit for the assist to each assister. Unless a team finishes a game with 1 kill, or everybody takes part in every kill, it's impossible to tell where the assist clusters are by score screen.
|
|
|
|
|
Turn on LoL.
Ready to smack faces.
Ranked disabled.
WHAT THE FUUUUUUUUUUUUUCK.
|
|
|
|
|
|
EPT
