Keyens: Why use Elo system in warfronts

I'm going to elaborate on Ocho's points separately (specifically with how they apply in Rift).

Originally Posted by Ocho

So ELO isn't reliant on balanced matchmaking, imbalanced matchmaking still provides the right data with a large enough data set. As long as matches are being played, it can generate an accurate score over time.

To elaborate on this, ELO operates on expected outcomes. Matchmaking is the process of trying to get those outcomes as close to 50/50 as possible. However in cases that isn't possible, ELO shifts are adjusted by your expected outcome. Lets use a chess example for a moment. Presume for the moment you are playing against a Grandmaster. When you lose a game in this case your ELO says it should basically be not affected negatively because the odds of you winning the game are so small. The same is true for their ELO going up. It should barely move because they are expected to win that game in the vast majority of cases. If you were to win however, your 'certainty' (which is a statistic that is tracked that roughly equates to "is this player as skilled as ELO implies they are") would be shaken and the system would move your ELO more drastically to attempt to correct for what is generally perceived as a change in skill. This is also true in even matchups. If you start winning more often than 50/50 (lets say 60/40) when the matching says you should be 50/50, your ELO will start to move more quickly as the system adjusts to your new skill level. Remember that the goal of ELO *on its own* is not to make matches. It is to project the expected outcome of any given match. Using this data, the matchmaker attempts to make balanced matches. There are various dials we can turn here and many different variables that impact match quality, so beyond that it would be a little more complicated than I want to go into here. Suffice it to say ELO is a pretty solid indicator of expected match outcome for any given matchup. This does not mean however that all of your matches will be even, because that is gated by lots of things beyond the matchmakers control such as available queue population and microvariance in matchup (such as you playing something that counters someone of higher skill, giving you an edge you wouldn't normally have).

Originally Posted by Ocho

WN8 is an interesting one. If you note games where that data is available, people with high WN8 ratings usually have pretty similarly high win percentages. Even a small deviation (more than say 1%) away from 50% is pretty significant over time. The rare outlier is a person with great mechanical skills at the game, but poor situational awareness (or just a bad team player). These people tend not to win matches, which makes them effectively bad at winning the game, so ELO is more accurate in that sense.Or to put it more simply, it doesn't matter if you're topping the charts if you're killing the wrong people.

WN8 is interesting as Ocho noted, but it has a major flaw. The less the individual game types favor tracked stats the less accurate it is, and the more variant gametypes any one game serves, the less accurate it is. This is a double whammy in Rift because not all of our gametypes can be well represented in tracked stats (for example a player who drives off a couple attackers in Domination maps but doesn't score kills is doing good work, but would score poorly), and WN8 struggles to accurately portray skill in various gametypes (ELO somewhat suffers from this as well because you can be better at one over another, but as a whole is far less susceptible to it). There are ways to combat this (for example you could get a rating for every unique game type that exists, but this has drawbacks too).Long and short of all this for those who'd like to skip the lecture is that ELO is a really effective system for predicting match outcome, which is all it is designed to do. There are other ways to improve match quality, but many of them have risks or might prove to worsen the matches for outlier players. As a general rule, improvements to systems that influence player behavior and perception will likely have a bigger impact that any change we could make to ELO and matchmaking.

Let's block ads! (Why?)

Keyens: Why use Elo system in warfronts