by wozearly » Mon Apr 08, 2019 2:35 pm
For information: resurrecting a post I made about the ELO calculation before, which explains the way it works and some of the oddities it creates at Master Tier
==========================================================================================================================================
Daft and unintuitive though it seems at a glance, the ELO calculation is working exactly as designed - though not perhaps in an ideal way for the reality of high ELO TOS.
Assuming the people you're playing against are not all Master ELO, and some are quite some way below you...
When you play as a solo faction, your ELO average is high enough above the others that between that and the Master K factor, you only gain +1. In effect, the calculation is already predicting you will very likely win based on your ELO, so doesn't want to move your ELO score much in response to a win. The K factor should similarly be shielding you against severe losses vs low ELO opponents as a neutral; but you can expect to get -2 or more for a neutral loss.
When you play as a group faction, the calculation takes the average ELO of the group. Your effective ELO drops substantially in the calculation's eyes when you're grouped with lower level players (a lone Master in the Town is heavily mitigated, a lone Master in the Mafia less so) so it doesn't see that faction's win as being so certain. Therefore you gain more ELO from Town/Mafia win because the calculation believes the match as a whole is more even. In effect, it expects your allies to bring you down to their level.
And yes, the net effect of this can be a bit silly. To a player, it seems to be madness that at Master Tier a win with a Werewolf would gain +1 where a win with Retributionist would gain +7. But, as it stands, that's the calculation working as designed - albeit not necessarily working as intended.
One fix that has been suggested already by Flake is to massively tone down the impact of other players' ELO in the calculation, and gear it more heavily towards outperforming your faction's winrate - or, preferably, your faction's winrate for the ELO tier in which you're currently playing.
==========================================================================================================================================
Added to this, although Ranked does adjust ELO to an extent to balance between the expected faction winrates, these winrates (as far as we know) are taken from Ranked as a whole. In lower tiers, Mafia and NK wins are notably more likely - in higher tiers, Town is typically using one or more meta-game strategies which, statistically speaking, raise their chances of winning more often than not. This imbalance means that Mafia and NK games are slightly 'punished' in terms of ELO gain/loss at higher winrates, where Town games are slightly 'rewarded' in terms of ELO gain/loss.
Once you move into Gold/Diamond Tier or above, assuming you're a decent player, the combination of Town's higher win-rate, the 'reward' for Town games and the increasingly greater chance you'll be placed with Town members of a lower ELO all contribute towards a tendency for ELO to rise over time.
In terms of fixing this, it isn't practical to base ELO on actions as Yemac floated (e.g. WW gets ELO for kills, LO gets ELO for seeing visits). Although its attractive in principle, it would require an incredibly complex calculation system, introduces inherent unfairness between roles (e.g. a more passive role, such as Mayor or Spy, has fewer options to gain ELO) and may also encourage counter-productive behaviour (e.g. people acting to try to maximise their potential ELO gain over actions that would be more likely to contribute to a win).