Steve Fenn sets the MLS table.
The first month of 2013 MLS has just finished, but many are already trumpeting clubs’ achievements or bemoaning the lack thereof. Chivas USA and Montreal Impact gear is flying off the shelves apparently (that’s not true, but you get the drift.)
Yet, here is an important fact as sure as the stacks of yellow cards sitting in Oswaldo Minda’s locker stall.
Everyone in MLS has either 29 or 30 matches left in their season in which form will surely shift.
Is there a way to extrapolate from the four or five game observations seen to date?
First, let’s illustrate which clubs offensive & defensive achievements have to this point been so far outside the norm as to be unsustainable. The graph on the left is each club’s goals and shots on goal per game for each season from 2006 through 2013, sized based on number of games played. The graph on the right is the same format, but with the sum of goals allowed and shots faced by each club’s keepers. (Note: Goalkeeping stats don’t quite tie out to total defensive stats, but the assumption is they are close and as a note, were the only historical resource available. The defensive graph’s axes are reversed so that in both graphs clubs would be striving for the top right.)
Click [here] to see an interactive version where you can filter down by club or year.
The best fit lines illuminate the overall average strike rate for the league since 2006. Almost every one of the 2006-2012 club-seasons amass in the middle of the graph and around that line, displaying the general range one can expect almost every club to land in by the end of a season.
Thirteen clubs offensively and twelve defensively fall outside of that mass thus far in 2013, and a regression to the mean should be the expectation.
In other words, trust not the truly as good or bad as their performances over the first four or five matches. Chivas USA’s offense is an extreme example, scoring 2.0 goals per game off only 4.0 shots on goal (SOG). Meanwhile, Portland’s problem is very similar to Chivas’ advantage so far, allowing 2.0 goals per game even though Donovan Ricketts has only faced 4.25 per 90 (Perfectly possible that Ricketts is the reason for the need for regression here.)
(Editor’s note: Look away Chicago Fire fans)
Chicago’s shots totals on both sides of the ball are shockingly similar to the horrible squad Chivas USA ran out last year, but so far they are allowing even more goals and scoring less themselves per game. Surely they’ll improve, but it’s unclear whether they will do so enough to avoid “that laughingstock team” that fodders the jokes come September and October.
Virtually every club is outside of the norm on offense, defense, or both thus far.
Again, filtering within the interactive version [here] is very useful for seeing larger trends in the data.
When evaluating early-season results in European leagues, Simon Gleave of Infostrada Sports came up with 2 brilliantly simple solutions which he wrote about on his excellent blog, Scoreboard Journalism. First, he compares all results to the same fixture in the previous season. For example if a team drew at Old Trafford in 2012, but lost when visting Manchester United in 2013, they would be -1 for that fixture and ManU would be +1.”
With an unbalanced MLS schedule the cycles to this analysis for the domestic league take more care. If you can recall previous columns on MLS strength of schedule, the calculations there came in handy and did the footwork for the crunching here.
In that post, expected goal differential per game (xGDPG) for each month was calculated. So here, the difference between the March figure in that post and the actual goal difference per game each team has had through their first 4-5 matches is taken, ΔGDPG.
Second, Gleave simply uses the odds that casinos give for a win, loss, or draw in every match to calculate expected points per game (xPPG) for each team.
For example, if the casinos gave odds implying that they expect Team A will win a match 50% of the time, draw 30%, and lose 20%, then it follows that their expected points for that game are 3*0.50 + 1*0.30 = 1.80, and their opponents’ are 3*0.20 + 1*0.30 = 0.90.
Then, average the expected points for every match played so far to get xPPG. Subtract xPPG from actual PPG and another measure of each clubs’ over/under-performance is derived, ΔPPG. For further explanation, read Gleave’s post on the subject.
Click [here] to see an interactive version where you can filter down by conference.
The most interesting data points are certainly on the extremes, with Chicago clearly having fallen apart, and Seattle doing so to a lesser extent. Meanwhile Houston, Montreal, Chivas USA, Columbus, and Dallas all seem to have markedly improved since the 2012 regular season.
The Galaxy have also done well, but they had the easiest March fixtures in MLS, according to both the bookies and 2012 results. If you look back to the earlier graphs it showcases that their offense and defense have been far better than the norm. Part of their regression will likely come from simply facing better opponents.
Keep in mind that this study is descriptive, but with only the concept of regression to the mean holding much predictive value. The casinos will adjust their lines based on performance, so, in short, betting for/against clubs based solely on extreme ΔPPG scores might land you in Antoine Walker Bank Account Land.
Most MLS clubs will probably look quite different in the fall than they do right now. However, between strike rates on both sides of the ball indicating unsustainability, and our expectation adjustments scoring for improvements, we at least have more to go off than the standard MLS table right now.