The good, the bad, and the unlucky: What Expected Points tell us about the 2018 MLS season

/

By Eliot McKinley (@etmckinley)

Expected goals (xG) has finally made it, the Times of London are including an alternate table for the English Premier League based upon per game xG for this season. While using only which team had the highest xG in a game for determining a winner is problematic, it is still a step in the right analytical direction.

Trying to determine a team’s luck is a time honored tradition here at American Soccer Analysis. In 2015, both Jared Young and Sean Steffen wrote about using xG to see if a team is good or lucky. In case you haven’t read What to Expect When You’re Expecting Goals or need a refresher, each shot in a game is associated with an xG value based upon where and how the shot was taken. xG gives you a probability that a goal will be scored from that shot. The higher the xG, the more likely that a goal will be scored. Summing up the xG in a game tells you generally which team had the better scoring chances, but as xG is probabilistic it is not as simple as saying the team with higher xG should win. If you simulate a game 100,000 times based upon the xG of shots taken you can derive the probabilities that the game will end up with a home win, draw, or away win. Doing this for every game played allows you to tell which teams are over- or under-performing their expected points (xPoints), giving you a sense of which teams are lucky or unlucky over the course of a season. There are obvious limitations to this type of method, as xG doesn’t encompass all factors that go into a game (e.g. it is blind to own goals), but it generally will get you pretty close.

By measure of points minus xPoints, The luckiest team in MLS play so far this season is FC Dallas. FC Dallas has 8.9 more points than expected based upon the shots taken in each of their games so far. If the table was based on xPoints rather than actual points, FC Dallas would be 5th instead of 1st in the Western Conference. The unluckiest team is San Jose, earning 12.6 fewer points than expected. While they are favorites to take home the Anthony Precourt Memorial Wooden Spoon, perhaps San Jose isn’t bad, but rather mediocre and unlucky. I have no doubt this will warm the hearts of the Quakes faithful. Toronto FC is following closely behind San Jose, underperforming xPoints by 10.8. If xPoints were points, Toronto FC would be in playoff position rather than near the bottom of the Eastern Conference table.

Zooming out to the entire league we can start to classify teams by their “luck” and how “good” they are. I’ve defined a team as “lucky” if they have more points per game than xPoints per game, and “unlucky” if the opposite (N.B. overperforming xG is not just chance, or luck, but that’s the term I’m going with here). A team is “good” if they are above the league median in xPoints per game, and “not good” if below.

Atlanta United is both really good this season and also a bit lucky. Atlanta leads the league in xPoints/g and is in the middle of the league in Points-xPoints/g. Remember, that xPoints is based upon both xG for and xG against, so even though a team may be perceived as not finishing well (unlucky), they could make up for it if their opponents are not finishing chances either. New York City FC and New York Red Bulls join Atlanta in the group of good teams that are a bit lucky.

Staying in the Eastern Conference, the Columbus Crew have the distinction of being the best unlucky team in the league this season. On a per game basis, Columbus trails only Atlanta in xPoints, largely on the back of a stellar defense, but are the 6th unluckiest team in the league. While the Crew have been creating plenty of chances, they just haven’t been converting them into goals.

Moving to the “not good” teams, the Chicago Fire are the worst team in the league by xPoints at only 1.01 per game and only slightly unlucky. So unlike San Jose, Chicago may just not be a very good team this season and haven’t had a sting of results against the statistical run of play. Conversely, Real Salt Lake has the 5th lowest xPoints per game, however, RSL is also over performing their xPoints by the second highest amount. This luck has them in a playoff position rather than competing for the bottom spot in the Western Conference.

xPoints, being a xG derivative, provides a method of ascertaining how a team is performing beyond points earned. While ultimately finishing at the top of the table is the goal, the process of getting there matters. While being lucky can propel a team up the table, reversion to the mean is likely. As noted on a recent American Soccer Analysis podcast, this should give FC Dallas fans some concern, but also Columbus Crew fans a reason for hope on the field (#SaveTheCrew).

Note: After finishing writing this article, I found that James Tippet is already doing an expected points-based table for the English Premier League.

Team Season Games Points xPoints Points - xPoints GD xGD Points/g xPoints/g Points-xPoints/g GD/g xGD/g
FC Dallas 2018 23 42 33.1 8.9 7 1.4 1.83 1.44 0.39 0.30 0.06
Salt Lake 2018 24 35 27.2 7.8 -7 -10.1 1.46 1.13 0.32 -0.29 -0.42
New York 2018 23 47 40.6 6.4 22 13.9 2.04 1.76 0.28 0.96 0.60
New York City FC 2018 24 47 42.8 4.2 17 15.5 1.96 1.78 0.18 0.71 0.65
Seattle 2018 23 32 28.0 4.0 0 -5.7 1.39 1.22 0.17 0.00 -0.25
Portland 2018 21 37 33.1 3.9 7 7.0 1.76 1.58 0.19 0.33 0.33
Kansas City 2018 23 39 35.6 3.4 12 8.8 1.70 1.55 0.15 0.52 0.38
Montreal 2018 25 30 27.3 2.7 -10 -13.4 1.20 1.09 0.11 -0.40 -0.53
Minnesota United 2018 24 29 26.4 2.6 -10 -13.2 1.21 1.10 0.11 -0.42 -0.55
Atlanta United 2018 24 48 45.5 2.5 22 23.7 2.00 1.90 0.10 0.92 0.99
L.A. Galaxy 2018 24 36 33.5 2.5 6 -0.8 1.50 1.39 0.11 0.25 -0.03
Vancouver 2018 24 33 30.7 2.3 -9 -6.6 1.38 1.28 0.09 -0.38 -0.27
Los Angeles FC 2018 23 36 34.3 1.7 6 3.6 1.57 1.49 0.07 0.26 0.16
New England 2018 23 29 27.5 1.5 0 -7.3 1.26 1.20 0.06 0.00 -0.32
Chicago 2018 25 23 25.3 -2.3 -14 -15.4 0.92 1.01 -0.09 -0.56 -0.61
DC United 2018 20 21 24.1 -3.1 -5 -5.6 1.05 1.20 -0.15 -0.25 -0.28
Orlando City 2018 24 23 26.3 -3.3 -20 -12.0 0.96 1.10 -0.14 -0.83 -0.50
Columbus 2018 24 39 43.2 -4.2 2 13.3 1.62 1.80 -0.18 0.08 0.55
Philadelphia 2018 23 30 34.9 -4.9 -7 3.0 1.30 1.52 -0.21 -0.30 0.13
Colorado 2018 23 23 29.1 -6.1 -9 -5.8 1.00 1.26 -0.26 -0.39 -0.25
Houston 2018 23 27 35.7 -8.7 5 7.3 1.17 1.55 -0.38 0.22 0.32
Toronto 2018 23 23 33.8 -10.8 -5 4.0 1.00 1.47 -0.47 -0.22 0.18
San Jose 2018 23 16 28.6 -12.6 -10 -5.6 0.70 1.24 -0.55 -0.43 -0.25