Crunchy Power Rankings: Week 4

If you followed the site last year, you'll know that I occasionally publish my own power rankings based on Opta statistics. Power ranking standards tend to be (intentionally) poorly defined, so the ranker has some creative freedom in deciding what exactly it means from one team to be ranked higher than another. The goal of the crunchy power rankings is twofold:

1. To make the ranking standards objective. I don't apply any subjective impression of a team's quality. I just calculate their underlying stats, weighted towards more recent results.
2. To dig beneath the results of luck. It'd be easy enough to rank teams by some kind of recency-weighted points per game. This is more or less a form table, which is popular on the continent. But if you read the stats work we do here and done elsewhere in the soccer statistics community, you'll know that the results of individual games are heavily impacted by tiny events (like whether a ball hits a post or goes a few inches left and becomes a goal), which for lack of a better term we'll call luck. It's possible for a team to play very well over a stretch of games but get fewer than an expected number of wins or draws due to tiny events with large impact. The fundamental assumption is that those results will regress. . that the team will, over the long term, get game results commensurate with its quality of play.

Of course, what counts as high quality of play is controversial. The question of whether possession stats are meaningful enough to even bother calculating is pretty well-worn, and if we built a power rankings based on time of possession, we wouldn't really be ranking anything important. That is to say, the correlation between possession and table results over the long term isn't very good.

So instead we look for other stats that correlate to results. Last season we settled on Duels Won, Passing Percentage in the offensive third, and Shots in the box. They cover three roughly separate phases of the game (defense, buildup, and attack) and all correlate reasonably well with PPG over a full season. Note that we could use shots on goal, which correlates even better, but in the quest to eliminate luck as much as possible I avoided it. The ratio of shots on goal to shots tends to regress to a mean for all teams, and a team with a short run of unusually high shot accuracy rate over a short period would be rewarded even though we'd expect that accuracy not to last. This happened last season when New York was putting up absurd goal totals early in the season through Thierry Henry and Kenny Cooper, but eventually their accuracy rate regressed closer to the mean (though they still stayed pretty high).

In the ongoing quest to improve the rankings, I've gone back and reviewed the stats and correlations based on last season's results. One significant change is that I've added the Recoveries stat, spurred by Matthew Doyle's article on mlssoccer.com. This innocuous-sounding stat represents winning loose balls, and it has an impressive correlation with long-term results.. over the last two seasons essentially equivalent to our previous gold standard: net shots in the box. In other words, your team's ability to win loose balls (generally in the midfield) is just as important to winning as your ability to take more shots in the 18-yard box.

A good argument for this addition is New England. Whenever a team's real table results diverge significantly from the cruncy power rankings, there are (at least) two possibilities. Either the team is getting unlucky and the CPR is providing a service in highlighting that fact, or the CPR model is missing something important. Last season, the Revolution consistently underperformed their CPR ranking. We had them, at times, somewhere in the top 5 teams, but they never really threatened for the playoffs. But when we add Recoveries to the model, they drop down to a mid-table team. Why? They were really bad at recoveries last season: one of the worst teams in the league. So there's an argument that, while I still think they were unlucky last season (look at the number of 1-goal losses they had), the model was missing the fact that the team was really bad at recovering loose balls.

I've also re-weighted the stats based on correlations in another full season of results (for the curious, duels: 0.31, off pass % 0.33, box shots 0.58, recoveries 0.58). Note that all stats are calculated as a net. . so the given team's results per 90 minutes minus the results they allow their opponent.

So with those changes, here's the inaugural edition of the 2013 Crunchy Power Rankings. As a reminder, the Total score is the sum of the individual stat results, weighted by that stat's correlation with PPG. The theoretical maximum is 100 and theoretical minimum is 0.

So what are the surprises in this edition? The biggest is the pair of Western Conference teams near the top. The Colorado Rapids and Portland Timbers are in the top 5 of the power rankings but near the bottom of the table, with 1 and 2 points respectively through three games played each. Portland has had an odd start to the season, going down early in all three of their games and then mounting a ferocious comeback at the end. There's some evidence that that skews a game's stats as the opponent tends to bunker in with a lead and provide more chances, so it might be overestimating Portland's quality as they've played teams largely content to back off the ball. Then again, Caleb Porter's Timbers look like a much better team than past editions, so I think there's something to the rankings.

As far as Colorado.. I haven't watched their season as closely. But their opening game loss came on Steward Ceus' rather memorable attempt to come out of the box and play a ball, which ended up bouncing over his head for an easy tap-in. That's the sort of tiny event that decides games, but which the Crunchy Power Rankings largely ignores. Their second match was a one-goal loss to the Union, and by all accounts the Rapids controlled most of that match. Then they drew Real Salt Lake away. So there's evidence that they're playing better than their results thus far. It'll be interesting to see if they and Portland can keep it up.

At the other end, the CPR isn't very impressed with the Union despite their 6 points. On opening week they lost heavily to Sporting Kansas City, then won the game over Colorado, which we've already discussed. Their third match was a sloppy 1-goal win over visiting New England on a corner kick goal. Their next game in two weeks at New York will be a nice test of whether they're really a playoff team or not.