- NEW! 2017 Mar 10: 2017 model now uses data from both 2017 and 2016 seasons. It uses data from 2015-2016 to determine how heavily to weight 2016 data in 2017.
- NEW! 2017 Mar 4: Modified 2017 model to use only 2016 data until the algorithm can be modified to combine 2016 and 2017 (otherwise the early season model would be very unstable).
- NEW! 2017 Feb 26: Created simulation of the upcoming 2017 season using 2016 data (and 2015 expansion data to estimate the strength of the expansion teams).
- 2016 Sep 6: Added CONCACAF Champions League qualification odds.
- 2016 Sep 1: Added Supporters' Shield odds.
- 2016 Aug 26: MLS Cup and playoff odds.
- 2016 Aug 24: Supporters' Shield odds on team pages.
- 2016 Aug 19: New mobile-friendly pages; added page for each team.
- 2016 Aug 17: Simulations now take 2016 team performance into account.
- 2016 Aug 16: Simulations now take home-field advantage into account.
This system is an attempt to model the Major League Soccer season, and provides useful insight into the probability of various occurrences in the remainder of the season.
The algorithm projects the results of future matches by simulating them, using data about which teams are playing and where the match will be played. It uses the league-wide season-to-date scoring average to project the average number of goals scored per match, and contains an estimate of the league-wide home-field advantage calculated from previous results. Team-level data (see Power Rankings) are added to this model.
The algorithm results in a reasonable model for wins, losses, and draws. The actual schedule is used, and each match is simulated individually. The season is simulated a number of times (currently 100,000), and the results of the simulations are tallied and presented.
The algorithm uses the correct tiebreakers to break ties in the standings (wins, goal differential, goals scored) until reaching the "disciplinary points" tiebreaker. The system is not aware of various teams' disciplinary points, nor does it contain a model for estimating future disciplinary points, so remaining ties after the first three tiebreakers are broken randomly.
The power rankings algorithm uses 2016-2017 regular season results (including context, like home-field advantage) in order to estimate the performance of each team offensively and defensively. The offensive and defensive ratings are combined to produce a single power ranking.
The power rankings correspond to the relative frequency with which each team should score a goal against each other team. As such, they represent the probability that a team would win against other teams in a golden goal situation. High-scoring teams will tend to have more wins (and losses), and fewer ties. For this reason, given two teams with the same power ranking, the higher-scoring team should perform better in the standings.
- (Done 2016 Aug 16)
Model home advantage
- (Done 2016 Aug 17)
Model team quality, (Done 2016 Aug 18) and publish the rankings produced
- (Done 2016 Aug 26)
Simulate the playoffs and generate odds that include playoff outcomes
- (Done 2017 Feb 26)
Use results from previous seasons to help estimate team quality
- (Done 2017 Mar 10)
Use both 2016 and 2017 data to calculate 2017 model