Yeah, I would say a K of 50 is far too high personally (especially if you're starting at 1200) and is definitely part of the reason you see massive fluctuations. I don't think I've ever implemented Elo with a K above 32 and base rating of 1450. I had an implementation that adjusted K in bands like FIDE do as well, and increased K for playoff games, which worked okay.
There was an attempt to adjust rating change based on margin of victory but it led to some wild changes and I couldn't balance it... also had one where I tried to tweak based on normalising pt differential based on strength of schedule, but that didnt work because id either have to create a moving range of games to calculate SOS from, or the formula would be basically broken for half of every season.
Elo is a good framework for an objective rating system though. I think it would probably work best for calculating power rankings here, because it would give you an easy way to gauge form as well (just sum rating change over a fixed period), meaning you wouldn't necessarily have to rank based entirely on elo rating, but you could incorporate delta into it and rate a team at 1200 who just jumped 35 pts in their last game higher than a team at 1210 who dropped 25 to end up there on the grounds that the first team is "hotter" right now.
All-time Elo would be very interesting, but a ton of work to put together.
There was an attempt to adjust rating change based on margin of victory but it led to some wild changes and I couldn't balance it... also had one where I tried to tweak based on normalising pt differential based on strength of schedule, but that didnt work because id either have to create a moving range of games to calculate SOS from, or the formula would be basically broken for half of every season.
Elo is a good framework for an objective rating system though. I think it would probably work best for calculating power rankings here, because it would give you an easy way to gauge form as well (just sum rating change over a fixed period), meaning you wouldn't necessarily have to rank based entirely on elo rating, but you could incorporate delta into it and rate a team at 1200 who just jumped 35 pts in their last game higher than a team at 1210 who dropped 25 to end up there on the grounds that the first team is "hotter" right now.
All-time Elo would be very interesting, but a ton of work to put together.
I impersonate a programmer for a living
Father of the League Wiki • Friendly Neighbourhood Angry Black Guy™ • NOT British
Originator of the Sim League Cinematic Universe (SLCU)
Super capitalists are parasites. Fite me.
Alternatively, if you agree, you can support a grassroots movement dedicated to educating and organising the working class by buying a digital newspaper subscription. Your support would be greatly appreciated.
