Clairvoyance Blog: Ping, Winrate and Vayne Probs
Riot’s latest dev blog explores howÂ high ping affectsÂ winrate depending on a Champion’s skill cap:
ARTICLE BY CHRISTOKKIES, DESIGN BY NANCYMON
Nothing quite kills the joy of online gaming like latency. At one time or another weâ€™ve all experienced the pain of lag spikes during a pivotal, game-deciding moment.
One thing that Iâ€™ve been mulling over when it comes to latency (or lag) is how champions in League of Legends might be affected differently by slow communication between a playerâ€™s client and the game server.
My brother, who lives in Japan, plays on the NA server so we can play together. He intentionally avoids playing AD carries because, his logic goes, this mechanically intensive role is not conducive to being played in conditions of high latency.
But just how true is his supposition? Are certain roles (or champions) more susceptible to a drop in performance from increased latency?
To test this, I went all out. I created a series of statistical models that attempt to predict a gameâ€™s outcome (win or loss) based on oneâ€™s ping. As it turns out, the use of latency as a predictor of a gameâ€™s outcome is contingent on the champion being played. In other words, the outcome of a game with bad ping is easier to predict with certain champions more than with others.
The use of latency as a predictor of a gameâ€™s outcome is contingent on the champion being played.
Classical linear regression models make predictions about continuous variables, where numbers have a logical order (such as age or or number of wins on a champion). Because the outcome variable in this case is categorical and binary (win or loss), a type of regression known as logistic regression was used to determine the estimated probability of an outcome given latency. I analyzed over 95 million different occurrences of a champion appearing on Summonerâ€™s Rift. To emphasize relative differences in latency between players, I used difference from the average in-game ping (which weâ€™ll just call â€˜ping differenceâ€™ going forward) as the predictor variable rather than the absolute ping values. For example, if a playerâ€™s average ping during a game is 75, while the average ping of the game for everyone else is 70, that predictor value will be 5 (rather than 75).
It appears that more mechanically intensive champions are more affected by latency, while tankier champions or those with point-and-click abilities are less affected by latency.
The graph below shows the estimated probability of Vayne being on the winning team as a function of ping difference. As the graph demonstrates, the lower the average difference in ping, the better chance this champion that relies on extremely precise positioning will tumble into victory rather than the enemy team.
The estimated probability of Vayne being on the winning team as a function of her difference in ping from the average in game.
coeff. = -.0014, z-value = -41
This relationship is similar for Xerath. The higher a Xerathâ€™s latency is relative to other players in the same game, the harder time the Shurima demigod has landing skillshots on his opponents.
The estimated probability of Xerath being on a the winning team as a function of his difference in ping from the average in game.
coeff. = -.002, z-value = -20
I believe this intuitively makes sense. Landing a Xerath ultimate or a Vayne condemn when your opponentsâ€™ movements are more responsive than your own is difficult to put it mildly.
Certain champions, however, do not exhibit a strong relationship between latency and estimated probability of winning.
Looking at Warwick, for example, differences in the average latency between other players in the game is not a strong predictor.
The estimated probability of Warwick being on a the winning team as a function of his difference in ping from the average in game.
coeff. = -.0002, z-value = -4
Similarly, Singedâ€™s probability of landing on the winning team does not seem to be affected much by latency.
If a Singed has 30 less ping than the average champion in that game, he has a roughly 50 percent chance of being on the winning team. And if that Singed has 30 more ping than the average champion, the chance of him being on the winning team is…wellâ€¦ still around 50 percent.
My interpretation is that the skills necessary to excel at Singed are what I would consider more strategically focused. When and where do I ward? When do I split? When do I TP or group?
Latency should only be affecting the outcome of a match to the extent that it differs from that playerâ€™s normal ping.
Further, I am reluctant to rule out any effect of lag on Warwickâ€™s or a Singedâ€™s ability to win. Matchmaking is designed with the intent that players are matched against other players in such a way that they win 50 percent of their games on average. If high latency is systematic for a player, then the effect of lag on a championâ€™s ability to win should be muted. For example, if a player who normally has good ping suddenly has bad ping, then that player will probably be far more likely to lose that particular game than if that player consistently plays with bad ping from game to game. In other words, latency should only be affecting the outcome of a match to the extent that it differs from that playerâ€™s normal ping.
Finally, I would like to caveat these findings by noting that the interaction between latency, champion, and the estimated probability of winning may not be causal. The models shown here only have a single predictor variable, and itâ€™s possible that if we throw other variables into the mix that the effect of latency disappears. Having said that, I think that the data support the hypothesis that the effect of latency on the outcome of the game differs by champion. A future analysis using more complex models could perhaps provide more evidence of causality.
The estimated probability of Singed being on a the winning team as a function of his difference in ping from the average.
coeff. = -.0003, z-value = -2
Letâ€™s come back to the question of whether my brother is correct to avoid playing as an AD carry because of his high latency. To answer this question, I ranked the champions by the extent to which their estimated probability of winning was affected by latency (for the more statistically inclined, I ranked the champions by the normalized regression coefficient given by the Z value). I then took the median of the rankings for each champion by role. For example, Yasuo appeared to be the most affected by lag, so he was rank 1. Vayne was the second most affectedâ€”making long-distance duo-queuing with a Vayne main somewhat ill-advised.
We do see a difference in the effect of latency onÂ the estimated probability of winning based on role.
As it turns out, we do see a difference in the effect of latency on the estimated probability of winning based on role. Champions categorized as â€˜AD carriesâ€™ (median rank = 21) appear to suffer the most from latency, followed by the mid role (median rank = 50). Finally, support, top, and jungle all had relatively higher rankings (70, 79, and 93 respectively).
These findings are in line with the notion that roles are differentially affected by latency, and in fact, my brother is probably better off playing Top, Support, or Jungle over AD Carries or mid given his higher-than-average latency.
If you have any questions, feel free to ask me atÂ @NoL_ChefoÂ or e-mail me at firstname.lastname@example.org.