Minnesota Wild’s Possession Metrics This Year: A Goat or a New Car!

I’ve only been writing for Hockey Wilderness for three months, so I think it’s important to let you know that I’m willing to admit when I just don’t understand something. For example, Garret Hohl posted an article at hockey-graphs.com about “Scoring First and Conditional probability.” It wasn’t so much Hohl’s article that tripped me up, but a link in the article to the Wikipedia entry for the Monty Hall problem.

For those who aren’t old enough to remember, Hall was the original host of a game show called “Let’s Make a Deal.” The show offered audience members the opportunity to be contestants on the show in different games. But the Monty Hall problem is specifically related to the game in which the contestant was given the choice of three doors. Behind one of the doors was the big prize—a new car!—and behind each of the other two doors was a goat. After the contestant chose one of the doors, Hall would reveal what was behind one of the non-chosen doors and Hall always revealed a goat. Then, Hall would give the contestant the option to stay with the contestant’s chosen door or switch to the remaining closed door.

The paradox of what a contestant should do was originally submitted as a letter to the American Statistician in 1975 and the solution resulted in a bit of controversy, because the response was that the contestant should always switch doors when given the option. If the contestant stays with the original door choice, he has a 1/3 chance of winning the car. But if the contestant switches doors, he has a 2/3 chance of winning the car and that’s based mostly on the host’s behavior.

Here’s the Wikipedia page. I’ve probably read it through three times and I find the solution to be completely counter-intuitive. But the best explanation is probably illustrated best by this box on the Wiki page. I guess your best chance is to assume that you picked a door with a goat behind it with your first guess and Monty has just revealed the second goat. But the notion that those are not 50/50 odds just can’t seep into my small brain.

That’s a long introduction, but I thought it might give you a window into things I think about, and it leads me to discussing the Minnesota Wild as we round the first turn and head into turn two of the season. One of the reasons that some of your HW writers and editors have been looking forward to the 20th game of the Wild’s season is because some of the team-level metrics we like to look at start to be predictive after a quarter of the season is concluded. Basically, after 20 games, you can see how teams are likely to shake out and you can tell which teams are strong and which might be soon looking at a dive.

The first stat we all look at is 5v5 CF%, which is simply the percentage of unblocked shot attempts for and against at 5v5. Lots of sites now have this information. I got mine this time from puckalytics.com and we can see that the Wild has played 1,063 minutes 5v5 in 22 games and the team’s CF% is 48.71, ranking 21st in the league, the bottom third between Colorado and Winnipeg. Well, that’s not good. The Wild are 18th in the league at shot attempts for at 53.45 CF/60 and 22nd in the league at shot attempts against with 56.27 CA/60.

I wrote a couple weeks ago that I think that the Wild’s Corsi number is depressed, because the Wild’s high shooting percentage and high save percentage resulted in the team playing with the lead a lot. That hasn’t been the case nearly as much recently, but I think score effects have likely played a role in the Wild’s poor underlying metrics. One of the sites one get CF% adjusted for score effects is puckon.net. And, what do you know?, the Wild’s Score Adjusted (SA) CF% is 50.7%, which ranks 16th in the league, or the middle of pack. And we’ve known for a few years now that adjusting for score effects makes for a better prediction at the team-level of how a team is going to perform.

But then things get strange. Puck On Net also allows for the user to adjust for venue, meaning home or away. And that adjusting for score effects and venue provides even better predictive measures. If you adjust the possession metrics for both score and venue, the Wild comes in at 51.1%, which ranks 14th in the NHL. Rising the ladder?

And, finally, Puck On Net added another measurement last fall to adjust for events, meaning weighting the four different events types used to calculate Corsi: shots, missed shots, blocked shots and goals. Adjusting our shot attempt metrics for event, score and venue increases the predictive value of the measurement. If we adjust for ESVA Corsi, the Wild rank 9th in the league at 52.8%. That’s pretty remarkable. The Wild have never finished in the top 10 in possession in the modern stats era, topping out at 16th two seasons ago.

Now, I feel like I’ve read articles that have referred to CF% and adjusted CF% and I’ve always assumed that a team would finish in about the same place regardless of the variation of the measurement used. In fact, if we go back to last year, the Wild finished 23rd in SA Corsi, 23rd in SVA Corsi and 21st in ESVA Corsi. Two years ago, the Wild finished 16th, 16th and 16^th, respectively.

Corsi as a tool is designed to measure quality at the team level. Generally, the teams that out shot-attempt their opponents tend to win more often. So, are the Wild around the 9th best team (top third) or around the 21st best team (bottom third) or somewhere in between? Much like the Monty Hall paradox, I don’t think I’ve gotten my head around that yet. Maybe it’s still just too early and these numbers will shake out, but it’s something to keep an eye on as we eye the remaining three quarters of the season.