Is there an equivalent to bball's points per possession in Hockey?

**VegasDave** · 03-09-09, 02:36 PM

Originally posted by curious

Hi,
I am trying to find predictive stats for hockey. Points per possession and the pythag theorem are pretty predictive for college bball (for some reason don't work at all for NBA), so I am wondering if there is an equivalent stat for hockey.

Would it be points per shot on goal?

Shot on goal per possession?

I don't know much about hockey.

Interesting question.

Unlike in basketball where a team takes the ball up the court and then sets up a "possession", plenty of times in the NHL possession is not established, and as such you can't really determine what a "failed possession" is on paper.

The most obvious correlation is success (or lack thereof) on the powerplay, but this is only bits and pieces of the game, not the whole game.

I guess even strength goals per time even strength is the closest, but I don't really know if that is helpful at all.

**Justin7** · 03-09-09, 03:38 PM

It seems likely, though it might be a different exponent. If you gave me a few thousand games, I could do a regression.

**Justin7** · 03-09-09, 03:42 PM

Another thing you might look at, in addition to shots per possession: scores per shot. This gives you two different dimensions of normalization, which might work better than a simple scores per possession.

**curious** · 03-10-09, 08:17 PM

Originally posted by Justin7

Another thing you might look at, in addition to shots per possession: scores per shot. This gives you two different dimensions of normalization, which might work better than a simple scores per possession.

Now the trick is to find the right exponent for the pythag formula.

**Justin7** · 03-10-09, 08:27 PM

Originally posted by curious

Now the trick is to find the right exponent for the pythag formula.

If you have 2 or 3 years of data, it should be easy.

**curious** · 03-10-09, 08:38 PM

Bill James take on this

I found this quote.

...bill james and his pythagorean formula used in baseball and applies it to hockey games. the most accurate method he found for predictive purposes involves poisson probabilities.

I need to try to track down where James said this.

**Data** · 03-10-09, 08:59 PM

Search this forum for "Poisson". At some point Ganchrow posted a link to a paper in pdf format that, as I recall, is a decent read on hockey, nothing advanced but very solid basics.

**MonkeyF0cker** · 03-11-09, 06:42 AM

I would search the web as well. There are quite a few papers and articles discussing hockey and poisson distributions.

**MonkeyF0cker** · 03-11-09, 06:59 AM

Another process you may wish to explore is inter-arrival times of goals scored...