By Matthias Kullowatz (@MattyAnselmo)
In so many words, Landon Donovan said that he thought the U.S. Men's National Team should have been more aggressive.
"[I]f I'm in that locker room before that game – before the Germany game, before the Belgium game – and the coach walked in and said we're playing a 4-5-1 and Clint [Dempsey] is up top by himself, I would have been disappointed. Because I would have said let's go for it. I want a chance to go for it and try to win the game."
The USMNT took a pounding against Belgium, facing 39 shots worth an average value of nearly five goals. The team's tactics mimicked those of many other teams in the tournament that were praying for a draw (and possibly the ensuing penalties, depending on the round). Iran against Argentina, Algeria against Germany, Mexico against Brazil.
These traditional tactics make some sense, as increasing the chances of a draw in regulation is beneficial for an underdog. But I'm not so sure it always maximizes that lesser team's chances of a positive result. I wrote a while back about how away teams in MLS did much better in the first half of games than in the second half when the game was tied. One possible cause might be that away teams fatigue faster, but I don't think that makes up the whole discrepancy. Tactics change, and though it might seem like a chicken-and-the-egg issue, I think it's the underdog that first decides to turtle up before the favorite becomes more aggressive.
The World Cup showed us similar results, with many teams stubbornly sticking to a conservative gameplan. Expected Goals during tied scores (even gamestates) serves as a reasonable barometer for how aggressive a team was able to be---or chose to be. Using ESPN's Soccer Power Index, I can estimate by how much a favored team should beat the underdog in terms of Expected Goals during even gamestates. SPI was able to predict these expected goal differentials with an R-squared value of 0.35. The scatter plot (for games in which at least 20 minutes were played in an even gamestate) is shown below with 95-percent prediction intervals.
Among the underdogs, only Japan in its game against Colombia performed worse than the United States against Belgium---that is, compared to how they were "supposed" to do. Much of the deviation from the regression line can be attributed to the SPI not being a perfect indication of team ability, as well as the variance of Expected Goals over a small sample size. But combined with the fact that the USMNT came out in a 4-5-1, the American's massive underperformance in getting quality shots off suggests that tactics were at least as much to blame as any discrepancy in ability.
Obviously, getting better shots---and more of them---is important to winning soccer games, and that's exactly what Expected Goals tries to measure. It's not surprising, then, that those underdogs that were able to outperform their predicted Expected Goals earned more favorable results collectively than those underdogs that were not able to do so. The only question that remains for any great deviation from expectation, then, is was it by choice or by force? The USA loss to Belgium reeks of choice.
I leave you with the list of results of those games in which at least 20 minutes were played in an even gamestate:
*The Algeria drubbing of Korea cannot be found on the graph, as it was such an extreme outlier.
**The SPIdiff can be seen as the number of goals by which a favorite was expected to beat an underdog, according Nate Silver's explanation.