Fun With Stats Episode 2: Dancing on the Grave

For the first ten weeks of the 2011-12 NHL season, hockey fans were forced to endure the success of your Minnesota Wild.  The Statgeeks told us about the inevitable Great Regression, yet we did not heed their warnings.

As the articles stacked up against the Wild’s success, like so many missed shots against, the Fenwick and Corsi champs vowed we would see our fragile tower crumble.

And it crumbled.  Oh, did it crumble.  Not just a freefall from 1st to 22nd.  The rumblings of a Winter Classic coming to our little town surely are gone.  Re-alignment which would restore the Wild to playing with teams that are actually in the same time zone, nixed indefinitely.  The careers of skilled forwards Pierre-Marc Bouchard and Guillaume Latendresse are in danger.  Captain Mikko Koivu has now been tagged with the dreaded “injury-prone” label.  Any and all good karma, mojo, and vibes which seemed so long overdue cruelly vanished, like so many Minnesota professional sports teams before.

Mikko Koivu raising the late-season white flag

Perhaps it is pity, then, that so few of those Statgeeks have come out to claim victory.  Maybe it’s modesty or humility.  But maybe they checked the numbers, and saw that it did not add up.

In my previous entry, I analyzed the then-seven-game losing streak and concluded that the Wild’s shot-based statistics were not consistent with the on-ice success (or lack thereof).

Small sample size?  Luckily, the Wild have treated us to a 29-game-streak of incompetence.  So let’s dig right in and take a look at the stats.

Here are the old-fashioned goal-based statistics, before the collapse, which occurred right around the time that Pierre-Marc Bouchard was violently boarded and concussed with no supplemental discipline by Sherriff Shanahan.  They should be fairly self-explanatory.

Games Played: 30
Goals per Game: 2.63
Goals Against per Game: 2.13
Power Play Percentage: 17.0%
Penalty Kill Percentage: 85.5%
Save Percentage: 93.4%
Shooting Percentage: 10.2%

And here are the shots-based statistics.  Note that I re-calculated the Fenwick and Corsi as ratios, which is the sum of scoring chances for divided by the sum of scoring chances for and against.  Fenwick includes both shots and missed shots, while Corsi includes shots, missed shots, and blocked shots.

Shots per Game: 25.8
Shots Against per Game: 32.1
Fenwick Ratio: 0.44
Corsi Ratio: 0.43

Now my stats are incomplete.  I only have total shots, missed shots, and blocked shots from the end of the game.  Statsgeeks love them some Fenwick Tied, and they have some great reasons why.  When teams are down, they tend to be more aggressive and take more shots.  When they are up, they tend to sit back and be more defensive.  So they look at shots, missed shots, and blocked shots only when the score is tied.  They also love road stats, because every scorekeeper gives his team better statistics (never mind the fact that almost every team plays better at home).  

So here’s the data from November 30, two weeks before the collapse:

Fenwick Tied: 0.41 (30th in the league)
Fenwick Tied Home: 0.45 (28th in the league)
Fenwick Tied Road: 0.35 (30th in the league)

Yikes.  Any way you slice it, the Wild were routinely out-shot and out-chanced by their opponent.  Yet the bottom line is winning games.

Win Percentage: 71.7%

Somehow they managed to win more games and earn more points than any other team in the league, even though the advanced stats said otherwise.  Clearly this was the Prophecy of the Great Regression, right?

Now let’s look at the losing streak:

Games Played: 29
Goals per Game: 1.79
Goals Against per Game: 3.10
Power Play Percentage: 13.2%
Penalty Kill Percentage: 80.6%
Save Percentage: 90.2%
Shooting Percentage: 6.5%

Goal scoring went down drastically.  Goals against went up drastically.  Special teams sputtered out.  Save percentage (foreshadowing!) went down drastically.  Shooting percentage went down drastically.  In every area the Wild are dramatically worse.  The bottom line bears this out:

Win Percentage: 31.0%

The Wild have taken less than half the points they did early in the season.  That’s not just coming down to Earth, that’s racing Brendan Fraser to it’s delicious, chocolate-y core.

Clearly their already dismal shot-based statistics have plummeted as well?  Obviously you cannot score goals if you don’t take shots, and if you give up more shots, you’re going to see more goals against.  It all comes down to shots, right?

Shots per Game: 27.4
Shots Against per Game: 31.5
Fenwick Ratio: 0.47
Corsi Ratio: 0.46

Hmm...shots for went up.  Shots against went down.  Fenwick ratio went up.  Corsi ratio went up.  This doesn’t add up at all.  Maybe the Fenwick Tied can clear some things up:

Fenwick Tied: 0.46 (29th in the league)
Fenwick Tied Home: 0.46 (28th in the league)
Fenwick Tied Road: 0.46 (23rd in the league)

So...the Wild were going to regress to the mean because they were being out-shot so often, yet they improved in every shot-based category both in absolute and relative terms, yet they are losing at an incredible pace?  Or maybe they’re overdue for a Great PROgression?

Or maybe it’s just a bunch of bologna, like I said last time around.

Obligatory snarky picture to illustrate my point.

Look, the Wild clearly overachieved early on.  And they’ve clearly been hit by injuries in the second half.  They shouldn’t have been as good as they were, and shouldn’t be as bad as they are.

But we were asked to believe that shot-based statistics were predicting the collapse, and what I’ve just pointed out states that is pure coincidence.

Just like save percentage is a systems-based statistic, Fenwick, or Corsi, or shots, tied, at home, wherever, they are all systems-based statistics.  You can compare WITHIN a team, but not AMONG teams.  We’ll look at that someday, I promise.

1 comment:

  1. Just coming to this very late as a result of a Puck Daddy link to one of your articles that had a link to this article. You get the idea. You know about the internet.

    Anyway, I thought I should point out that while they did improve during the losing streak, they remained awful relative to the rest of the league. In other words, the regression (in this case an elastic effect from one extreme to another that produced a more realistic mean winning %) is entirely appropriate and well predicted by their continued struggles to possess the puck and direct it at the other team's end.