Presenting Expected Goals by Game
/By Matthias Kullowatz (@MattyAnselmo)
If you migrate over to the xGoals 3.0 tab, you can now find a "By game" option, complete with some team-level expected goals stats from each game this season.
Using our fun sorting capabilities, we might observe that three of the top six expected goal differential (xGD) advantages for home teams have been produced by the New York Red Bulls, and last week's drubbing of D.C. United was only New York's third-best showing of those three--at least, in terms of xGD. Of the top-20 xGD performances by home teams, the home team has won 19 of them and tied one.
Re-sorting that same column shows that Portland's performance in Seattle last Sunday was the 11th-best performance by an away team. Of the other top-20 xGD performances for away teams this season, the away team has come away with an average of 1.8 points (9W-7D-3L).
While xGD is a good indicator of a team's success, it does not guarantee it. For those with some intro stats under their respective belts, the simple correlation coefficient between xGD and points is 0.39, with a 95% confidence interval (0.28, 0.49). The epidemiology crowd may appreciate the contingency table to the right for which the sensitivity is 71% and the specificity is 64%. Modeling fanatics might want to know how the log-odds of winning change with respect to a one-unit improvement in xGD. The answer to that question is 1.13 (95%: 0.93, 1.33). Then those model fanatics would shoot me for using a correlation coefficient on data that is certainly not bivariate normal.
All of that is to say that teams that win the expected goals battle tend to win the game, but it is no certainty. Most importantly, while expected goals are not perfect, and soccer--with an outcome determined by a scarcity of special events--is not a sport easily explained by any single metric, expected goals are still meaningful. Statistics serve many purposes, and here at ASA we try to find those statistics that reveal sustainable skills and outcomes. Expected goal differential is one such stat, and hopefully, in this new table, it can help to describe the manner in which teams won, lost, and tied.