Home

The Problem: Try Bonus Points Most rugby competitions are scored using competition points rather than just wins and losses to try and make matches more interesting. The usual system awards 4 points for a win, 2 for a draw, 0 for losing, and then two ways to get a bonus point: 1 point is awarded for losing by seven or fewer points, and one is awa...

Week Three Projections With the week off, model review was done in the last post. They’ve been pretty spot on so far. Club ratings have been updated from last week’s outcomes, so we re-ran simulations and have new projections. For a bit more depth, check out the match predictions in the Current Projections page. Or check out the Six Nations St...

Offseason Changes There’s been a few big changes to this league, most notably some changes in teams. The LA and Austin teams were removed for violation of league rules, and the Chicago Hounds have been added as an expansion club. This puts the league at twelve teams: Atlanta, Toronto New England, New Orleans, New York, and Washington in the Eas...

Model Review After 6, our club level model is 5 of 6, and the player models are 6 of 6. Interestingly, the player model was a bit less accurate than the club model for the Ireland v France match, which may point to a bit of an overvaluation of Irish players, or an undervaluation of French players. If such a misevaluation exists, it may be cause...

Current Table Team Wins Point Differential Competition Points Ireland 1 24 5 Scotland 1 6 5 France 1 5 5 Italy 0 -5 1 England 0 -6 1 ...

Method Now that I can generate expected performances for any given team, I figured I could test out what I can do by trying to simulate an entire competition, and the Six Nations came around at the perfect time. I simulated each match in the competition and updated team rankings using the simulated result before simulating the next week’s matche...

Now that I’ve finally filled the holes in my dataset and have some decent predictions, I was able to start on version two - a glicko model. The ability to generate predictions with uncertainty and work with distributions should allow for more nuanced and useful predictions. But its a big change to make, so I’m implementing it in parts. To be Agi...

Model improvement was something I wanted to tackle much earlier, but some issues with data coverage became very apparent over the summer and became a higher priority. Now that I have most of the leagues I want, I can focus a bit more in the math. Before tinkering with the computation of a team’s elo (because intuitively, a player should contribu...