The Association Elo Rating System

I wanted to write this up to try to stop the onslaught of “not sure what this is” comments around my emails. Figured the very end of the season is the perfect time to explain. Elo is a rating system measuring the relative strength of each owner’s team, taking into account every regular season and playoff game in The Association. It will take into account your win/loss as well as the margin of victory, so a close match at 80-80 is the same as 125-125. Here’s the formula when calculating the rating for Team 1:

Rn = Ro + K × (W1 / (W1 + W2) )

Ro is the prior rating. The K factor is the weight of the matchup. W1 is Team’s 1 score, W2 is Team’s 2 score.

The higher the K factor, the more sensitive the rating change. I have it set at 250 for regular season games, based on personal preference. For the playoffs, I lowered it to 100. This is similar to how they use k-factor in chess, a lower K-factor at higher levels of play will minimize the deflation of a team’s score if they lose. I wanted to avoid punishing someone by having a bad playoff run (either due to luck, injuries, or just playing harder teams), so I lowered the K-factor. For reference, fivethirtyeight has their factor set at 20.

I have adjusted this formula to 75% of the calculation. The other 25% is your score compared to the league average for that week. I decided to make this the average per week because a historical league-wide look does not factor in byes. Yes, byes affect everyone differently week to week, but on average you will have more scoring opportunities in week 1 compared to week 8

Average rating is 1500.

Each year your elo is reset. In fivethirtyeight’s elo model, they maintain 2/3 of the prior year elo, and 1/3 of the league average. I figured that was too high, there is more roster turnover in fantasy so I have half prior year and half league average.

Elo doesn’t do a great job in predicting future games. In fantasy football there is a lot of change week to week (and a wide variance of player scoring in the current week), so while someone rated at 1650 is likely to beat someone at 1400, it’s wouldn’t shock me if the underdog won. In my predictions I have a 250 point rating difference equal to a 75% chance of victory, really based solely on feel.