Tuesday, March 27, 2007

Overtime incentives study

When an NHL team loses a game in overtime or in a shootout, it still gets a point in the standings. That means that, for overtime games, three points are divided between the two teams. But for non-overtime games, only two points are divided.

So teams have an incentive to allow more games to go into overtime.

I've done a little study to check that. It's still in draft form. You can find it here.

Comments are appreciated.

(Also, if anyone knows if there was a rule change or something responsible for the big jump from 2001-02 to 2002-03, please let me know.)

UPDATE, March 30: study is now updated (same link). I've added some tests suggested by commenters, and suggested some possible reasons for the observed effects.

Labels: , ,


At Wednesday, March 28, 2007 3:59:00 PM, Blogger Tangotiger said...

Very cool research.

The goals per game is the first major component to handle.

And, from what I did many years ago, it does follow a Poisson distribution. I don't remember what Alan did in his article that you pointed to, but I'm sure he found the same that I did. As for the not-even face-to-faces, it's very possible that teams play "closer" when faced with similar competition, and more wide open when not, so that in the end, you get the poisson.

Other rule changes would be penalties being called.

At Wednesday, March 28, 2007 8:47:00 PM, Blogger Phil Birnbaum said...

If goals do follow a Poisson distribution, then the proportion of ties should be significantly lower than that actual observed -- 18% in theory vs. 20-25% in practice.

I double-checked the theoretical numbers with a simulation, and they are accurate.

So perhaps goals are just *close* to Poisson?

At Thursday, March 29, 2007 11:33:00 AM, Blogger Don Coffin said...

One of the things that your argument suggests is that goals specifically in the third period should be lower. You might want to check that out as well.

At Thursday, March 29, 2007 2:01:00 PM, Anonymous Anonymous said...

very interesting study. a couple suggestions:

any ability to add seasons before 97-98 to beef up the data points in pre-ot extra points will help with statistical significance at least with the first rule change in 1999.

i think your best stab at normalizing tie % based on goals/game was taking the poisson prediction and increasing by some factor to get to about 2.5% adjustment factor (your 7% calculation imo is a result of running a regression on a small data set). (note that Ryder addresses the overabundance in ties on p. 12 of his poisson toolbox paper)

My last suggestion to help get stat. sig. results: feeding off of doc's post. while game strategies are designed to play for ties, perhaps teams don't play for ties in the first 2 periods because too much else can happen. but that the "play for tie" behavior shows up only in period 3 -- especially when the score is tied. If you have the data, you could rerun your #s using only games where the score was tied starting the 3rd period. I would bet your % of games going to OT would be starkly different from the seasons pre-1999 vs. post-99.

(you would have to rework the adjustment factor in goals/game -- but the poisson distribution can handle that)

At Thursday, March 29, 2007 2:26:00 PM, Blogger Phil Birnbaum said...

doc, nate,

great minds think alike, and all that ... I actually did all those things you guys suggested.

Third period goals roughly matched overall goals. But that might be because of empty-net goals ... I don't have data on EN goals, but can eliminate games where a team won by 2 or 3 and see what happens.

The proportion of OT games when the game was tied after two periods was flatter than overall -- in fact, the second lowest proportion was last season, when it should have been high, according to the hypothesis. The first two seasons in the study did show a lower proportion of third-period-tie OT games, just as it did of overall OT games.

The website I used had game data only back to 97-98 ... couldn't find any more.

Thanks, Nate, I did miss the discussion about overabundance of ties -- there's more in the paper about endgame strategies and such. Will re-read.

At Friday, March 30, 2007 1:36:00 PM, Blogger Tangotiger said...


You should join the following Yahoo groups:


(Something like that.)

Alan and I, among others, are members. There's some fantastic data available.

At Friday, March 30, 2007 1:47:00 PM, Blogger Phil Birnbaum said...

Geez, I'm already on that Yahoo group! I just never go there.

I just switched to digest mode, so I'll receive messages daily. Thanks.


Post a Comment

<< Home