As the kickoff of any league’s season approaches, everyone wants to talk about predictions. Unfortunately, few in the media have any clear method behind their prognostications. A pundit’s expected league table (or power ranking, or whatever they call it) is usually little more than the previous year’s final ranking of teams, with some adjustments made based on how much the author (dis)liked clubs’ offseasons. There are absolutely media types who take a detailed approach, Matt Doyle for example, but they are the exceptions that prove the rule.
Instead of the 2013 MLS final standings, I will be using 2013 expected goal differentials (xGD) as my base for 2014 prognostication. From each club’s xGD, I’ve determined where the league-wide regression line suggests their natural point total lies. My MLS predictions aren’t entirely divergent from the pundits’ approach, though, as I will make some qualitative adjustments based on a variety of factors. Also, instead of simply ranking teams or slapping a single point target on each club, I will present my predictions as a range of likely outcomes.
Expected goals differentials are driven by shot locations earned and allowed. Basically, teams who regularly shoot from close to goal and straight on while their defense seldom allows such shots tend to excel. I wrote about it in more detail in The Shin Guardian last year, and showed that xGD was a better predictor from the first 17 matches of the season to the last 17 than were points or actual goal differential. All the tabs of the following data visualizations illustrate this in different ways:
The above is driven by data gathered by American Soccer Analysis, who allocated every shot for/against last season into one of these six bins. My first adjustments regard two clubs whose style of play can, I feel, most defy study based on shot locations alone. Real Salt Lake tend to create on the break and with throughballs, and such actions lead to shots of higher value than average in the same locations. Meanwhile, San Jose tend to create many shots off of headers, which are of lower value than getting the ball to strikers’ feet the same distance from goal. Thus, I’m shipping two points from Northern California to Utah. Oscar Pareja left the Colorado Rapids in order to return to FC Dallas, where his strengths could well fit the faux-hoops’ young roster like a glove, and the Rapids have yet to even name a replacement. That is worth at least one draw turning into a win and vice versa. Two points go from the Rockies to the Dallas suburbs. Obviously, 2013 stats are blind to Toronto FC’s massive offseason spending. Michael Bradley and Jermaine Defoe are enormous signings, and they are very likely to improve the Reds to some extent. Problem is, TFC has always been dismal before, and there is little reason to believe that they have enough pieces around their stars to contend for silverware. Let’s say Toronto is spending $7 million more on wages this year than they did last. Based on my analysis from a couple days ago, that on its own is generally good for about five points. But that’s fat from a certainty, so I’m also widening Toronto’s likely point range. The Philadelphia Union had a more modestly impressive offseason, with the additions of Maurice Edu, Austin Berry, Vincent Nougeira, Cristian Maidana, and others. Two more points for Philly. Last year the Montreal Impact were a good team, but they were old and lacked depth. Now the whole squad is a year older and a little shallower. The Vancouver Whitecaps lost their holder of the MLS golden boot, Camilo, and Lee Young-Pyo retired. The Chicago Fire unloaded Austin Berry, Jalil Anibaba, and others with no particularly impressive names replacing them.. The absences of Chad Marshall, Eddie Gaven, and Andy Gruenebaum should hurt the Columbus Crew. Plus all four of these clubs went through coaching changes. Maybe one or two of these coaches does a great job, but I think most of them will go through frustrating adjustment periods, so I’m taking two points away from each of them.
All of these adjustments are at their core generated by the things my 2013 expected goal difference data doesn’t know, like offseason changes and aspects of clubs’ regular shots taken/allowed that go beyond location. I decided on each of these shifts before looking at clubs’ specific point totals suggested by simply regressing xGD.
I plan to re-visit these projections after the season ends to see 1) how close I came to the actual results, 2) if my qualitatively adjusted predictions performed better or worse than purely quantitative xGD regressions would have, and 3) how consistent these clubs’ xGD were year over year.
Thiy is all quite experimental, but I hope that by having a clearcut method I am providing value now, and will improve upon it as a look on this model with a critical eye going forward.