Tuesday, March 24, 2009

I love it when someone else writes my posts

Bill Simmons, this time, on ESPN.com talking about the statistical revolution in sports, and specifically in basketball.

If I wrote that, there would be two major differences:

1. It would be about hockey, not basketball.
2. 7,335,204 fewer people would read it.

I don't begrudge the man his audience of course, but I feel a kinship with the premise, which is that certain sports (baseball, football) are much easier to break down statistically. I think the reason is that they are much less fluid than basketball and hockey. Baseball is easiest because it is the most static - there are innings, and further there are at-bats and individual pitches. Go to the Baseball Analysts and you can find posts describing the various run-creation formulae... and even insanity such as charting certain pitches placed in certain parts of the strike zone with various permutations of outs and baserunners, and how many runs tend to be scored off those pitches.

(Note - not all that insanity is bad, actually; and lest Bill Conlin and Murray Chass start pounding their 1915 manual typewriters in frustration over all this newfangledry, there are also posts breaking down pitchers' mechanics and old-school "I know good baseball when I see it, consarnit" scouting reports.)

Football is an interesting case. It breaks down similarly by play, but there is a wrinkle that baseball lacks: there are people in football whose sole job is to make it possible for other people do their jobs: primarily the offensive lineman and the humble blocking fullback, who are thus tough to quantify. The venerable Paul Zimmerman, SI's Dr. Z, keeps charts to help him rate the overall performance of such players, but it involves watching truly staggering amounts of game film every week, and thus it's very difficult to duplicate and make generally accessible. (Aaron Schatz and the Football Outsiders folks have taken Dr. Z's work to the next level.)

This wrinkle sometimes hurts cold objective analysis. Why are some defensive units extremely effective with few or even no Pro-Bowlers, while a unit with four can be inferior? Or to use Simmons' basketball example - Carmelo Anthony is clearly superior to Bruce Bowen, but the Spurs played better with Bowen than they ever would have with Anthony. They wouldn't have been able to use a guy like Carmelo in Bowen's role as top-flight defender and grit/hustle player. And basketball runs into far more of those difficulties than football - it is much more free-form. However, it is also high-scoring and has a shot clock; it breaks down into possessions of a sort, resulting either in points or a stop, and that gives the nascent statistician an edge.

What then of hockey, the most fluid of the four major sports?

Well, it's so unstructured that play can conceivably continue without a whistle for an entire 20-minute period, so long as nobody takes a penalty or knocks the puck out of play. Baseball is so structured that players can only play on offense one at a time, have rigidly-defined jobs on defense, and cannot return when substituted. Hockey is so loose that both offense and defense require complete team efforts, in which anyone may find themselves crashing a crease or working in the corner, and in which players make substitutions during the play.

The most structured position is that of goaltender. Not surprisingly, the goalie has the most complete set of statistics: percentage of shots saved, goals per game, wins, losses, and shutouts. This doesn't account for the quality of the shots but over the course of a full season such things tend to even out, or are partly reflected in other numbers. Then there's offense. Hockey features the lowest scoring of the four majors, so again, that's a relatively simple matter of who scored and who set up the goal. (Sometimes not even that. The NHL didn't officially record assists for a while.)

But as a low-scoring game, it's the defense that suffers. The plus/minus is a start, but the stat has a lot of "noise" to it. It sounds simple: plus/minus measures the difference between goals a team scores and those it permits while a particular player is on the ice. If my +/- is +5 it means that, while I've played, my team has scored five more goals than it has allowed at even strength. If yours is -10, well, that means the team is fifteen goals "worse" with you out there than it is with me. Right?

Well, not really. A poor player on a good team can float on the results of stronger teammates. A great player on a bad team will tend to have bad numbers. Some players play almost 30 minutes per game while others play only 7 or 8. A team's best defenseman and checking forwards are expected to shut down the opponents' most dangerous scorers while a third-pair defender benefits from playing against lower-skilled foes. You could be expected to handle future Hall-of-Famers for half of every game while I can only be trusted against scrubs for ten minutes here and there.

And it gets more complicated - if I wasn't around for those ten minutes, would you do worse at your job because you wouldn't be as rested during the game? If I sink to even or minus, should I be benched? What if that sinks your rating to -20? What if your rating is so low because you play with checking forwards who aren't expected to score? What if it's because we have a terrible goalie, so you are more likely to be on the ice when he flubs an easy shot?

With most statistics, outliers are a problem - they skew the overall results and aren't reliable. With a stat like +/-, it almost forces the reverse, where an outlier is not an anomaly but a true measure. That indicates something badly gone wrong with the original stat. Sure, the only player with a positive rating on a bad team full of high minuses is probably that team's best defender, but if takes that kind of a result, the stat is not all that useful. We could tell such a disparity without any stat at all; but if a guy is +8 and another is +10 over the course of 82 games, is that a statistical flub or a true indication that one has had a marginally better season on defense?

Attempts to improve hockey statistics are also underway, but much less known, and one big reason is that they're rather like Dr. Z's Pro-Bowl charts - they aren't that accessible and they do some very strange things with the little raw data available. Measuring strength of opponent, for example - sticking with you and I, the two defenseman - you face better opponents, measured in part by their goal and assist totals, in part by their minutes played (thus reflecting what their coaches think of their skill)... But are not my opponents' lower scoring totals due in part to their own lesser playing time? In part, isn't it also because I am knocking them down with a stick? In spreadsheet terms, it's something of a circular formula: the results of the calcuation are part of the calculation itself.

The only way to be really sure of our relative value would be to play me half the game against All-Stars and you ten minutes against shlubs, and see if I can duplicate my +5 while you remain at -10. Good luck finding the team that is willing to do that merely in the name of statistical analysis. Much the reverse - they would want to see some cold facts to back up the idea that I am suited for a tougher role. In fact, that's why the Montreal Canadiens developed +/- in the first place, back in the '50s (the league at large didn't adopt it until 1967, I believe). Absent that they must go with their eyes and their gut. Well and good for the team, but harder for the hockey fan who wants to argue that their favorite guy deserves a larger role, and who wants to appeal to something more than "I really like that guy." (I am your favorite guy, of course... What? My hit counter proves that I'm a bench jockey? Whyioughta...)

The worst thing would be to drag into hockey the ridiculous conflict between stat geeks and harrumphy "because I say so" sorts. Both call themselves the clear-eyed lovers of fact, and the other side is regarded rather as the emissaries of the Devil - either wanting to empty the sport of its soul, or of its brains. I'd rather not be either a vampire or a zombie, thanks all the same. I don't object to good stats, nor to personal observation. They're both simply means to describe and measure performance, and when they are at odds, something is up. What I object to are bad stats and prejudiced observations, which confuse and mislead. We get this a lot with Martin Brodeur, who recently became the winningest regular-season goalie in NHL history. Some people think he's too brilliant for words, others insist that he's only decent and not excellent, both based on their lyin' eyes. Others point only to the numbers to say he's the greatest, while others point to entirely different numbers to insist that he's adequate at best. (Really, I've heard it.) It's almost like he's the Derek Jeter of goalies...

...except, of course, that Jeter is clearly overrated (though still darn good), while Brodeur is in fact a superb keeper. So say my own lyin' eyes, at least until the stats catch up.

PS - speaking of lyin' eyes, your eyes are correct about the different look to the blog. I swapped templates and tinkered with some of the colors during a dull moment.

No comments: