Overtime

by Hawerchuk on Nov 24, 2009 9:00 AM EST in Analysis

I've mentioned a few times that overtime and shootout outcomes are essentially "luck." I wanted to put some numbers on it so you can see what I mean.

In the long-run, overtime winning percentage will converge to the same value as a team's overall winning percentage. But there's no such thing as "the long-run," with overtime: teams rarely play even 120 minutes of OT in a single season. The most likely outcome of a five-minute overtime period is a tie, which drags overall winning percentage towards the .500 mark, no matter how good a team is. Here's how teams have performed in OT (and the SO) relative to their record in regulation time:

	R^2	Regression to the Mean
SO 2005-09	0%	97%
OT 2005-09	0.7%	93%
OT 1999-04	11.2%	69%
OT 1983-99	6.7%	79%

The NHL played overtime 5-on-5 for the first 16 seasons it was in effect, then went to 4-on-4 for five years, and we're now in the fifth season of the OT/SO regime. While there was some relationship in the past between a team's overall record and its OT record, the introduction of the shootout has completely severed that link. Why play for 2 points in OT against a better team when you can go to the shootout and have a better chance of winning? As I've mentioned before, the incentives in overtime favor bad teams slowing down the pace against better teams.

Now those sample sizes aren't huge, so I thought I'd simulate overtime performance using a constant goal-scoring rate equal to the league-wide scoring levels at 4-on-4. Here's the relationship I get between a team's overall winning percentage and its performance in overtime if I assume that a team plays 20 overtimes per season:

That's 80% regression to the mean - a spread in winning percentage from .250 to .750 gets squeezed down to the .450 to .550 range. And even worse, with the advent of the extra point for overtime losses and shootouts, the number of points a team can expect to get in games that go to overtime regresses 90% from their true winning percentage. A team with a .250 true winning percentage - an expansion team, perhaps - still has a 40% shot at having a winning record or better between the shootout and overtime.

Overtime might seem like it gives a better team another chance to win the game, but five minutes is too little time to have much of an impact. In addition, the completely fluky nature of the shootout gives weaker teams an incentive to play for a tie in OT, so it's likely that overtime winning percentage is regressed even more than 90%. Overtime is a nice gimmick, but it's like Final Jeopardy! - you can be winning all game and one wrong answer leads to your downfall. And to drag the analogy even further, the shootout is like replacing Final Jeopardy! with a quick dart-throwing contest - it emphasizes a skill that's irrelevant to the rest of the game and gives you random outcomes.

Tweet Comment 14 comments | 0 recs |

Do you like this story?

Comments

I would like to see a return to the days when there wasn’t a shootout at all. And instead, increase OT to 10 minutes or just do away with it.

I think the shootout makes things worse, because the shootout gives you a chance to get 2 points.

by tsukemonoki on Nov 24, 2009 11:22 AM EST reply actions

This made me think of something and now I’m curious — has there been any kind of look at how much luck is involved in winning a single hockey game?

by MathMan on Nov 24, 2009 1:33 PM EST reply actions

It’s a recurring theme on Vic Ferrari’s site (see links on sidebar). But you’d have to be able to define every team’s “true talent,” in order to determine how much luck is in their record. Nobody has been able to figure that out yet.

by Hawerchuk on Nov 24, 2009 1:45 PM EST up reply actions

That’s 80% regression to the mean – a spread in winning percentage from .250 to .750 gets squeezed down to the .450 to .550 range. And even worse, with the advent of the extra point for overtime losses and shootouts, the number of points a team can expect to get in games that go to overtime regresses 90% from their true winning percentage.

Put those two factors together, and read it like this: a spread in Points percentage from .500 to 1.000 gets squeezed down to the .700 to .800 range.

And there’s the rub. Over any decent chunk of games, you can’t lose. The percentages are in every team’s favour. Get to enough overtimes and you’re stealing points, plain and simple. Don’t get to enough, and you’re giving points away.

One disturbing trend I have noticed this year is that fewer games are being decided in overtime, just 33 of 92 if I’m adding my morning standings correctly.

Percent of 3-point games decided in shootout:

2005-06: 145/281 = 51.6%
2006-07: 164/281 = 58.4%
2007-08: 156/272 = 57.4%
2008-09 159/282 = 56.4%
2009-10 33 / 92 = 64.1%

It’s preliminary just a quarter of the way into the season, so standard “sample size” caveats apply, but it’s a disturbing trend. So is this:

3-point games as a percent of all games:

2005-06: 281 / 1230 = 22.8%
2006-07: 281 / 1230 = 22.8%
2007-08: 272 / 1230 = 22.1%
2008-09: 282 / 1230 = 22.9%
2009-10: 92 / 337 = 27.3%

Put it all together, and early results have yielded a season-over-season increase in regulation ties of about 19%, AND an increase in overtime ties of around 14%. All of which strongly suggests to me two things about tied games:

1) More than ever, teams are either playing to tie the score, or playing to the tie score (you might have to read that carefully), in the last 10, 20, 30 or more minutes of regulation. OT is a desired outcome, especially but by no means entirely for the underdog. The time to play for the lead is relatively early in regulation, and then not again until overtime (if then).

2) Teams are playing to the tie score in OT for the very reason you suggest : they like their chances better in the shootout than in open play. In this case it would be the underdog more likely to be playing Hang On Harvey hockey while the superior team has no disincentive to open it up in overtime, unlike the strong one that exists in regulation (risk losing the Bettman Point).

Nice system in a “New NHL” that’s supposedly all about offence, eh?

PS: Gabe, I know most of my stuff is in a different realm than yours, but you may find these two articles about the Bettman point to be of interest to you:
Another Day, Another Bettman Point last March

What is the value of a game-tying goal? last week

This has been a pet peeve of mine for years. With the daily puzzle that is the NHL standings, who needs Sudoku?

Writer for The Copper & Blue and primary shareholder of Zorg Industries

by Bruce McCurdy on Nov 24, 2009 5:48 PM EST reply actions

Bruce,

Thanks for the lengthy reply! It’ll be interesting to see if those early-season percentages play out. If teams in one division focused on tying each other, they’d all make the playoffs…Lots of opportunities here for collusion that didn’t necessarily exist before.

I’ll check out your columns when I get a chance!

Gabe

by Hawerchuk on Nov 24, 2009 7:56 PM EST up reply actions

One thing about 4 v 4 OT is that it might actually limit the benefits of overall team depth. The majority of NHL teams can scratch up 6F and 4D that can play, and that’s roughly how many teams roll in OT. I wonder if the talent that takes the ice in OT isn’t a lot more bunched together than you might see otherwise.

Matchsticks and Gasoline

by Robert Cleave on Nov 24, 2009 9:23 PM EST reply actions

I’d like to see the NHL either a) make all games 3 points in the standings or b) go with 10 minutes of 4 on 4 OT and scrap the bonus point.

I don’t like the fact that some games are worth 2 points and others worth 3, but when I subtracted out the OT/SO bonus point from the overall standings, I was rather surprised to see that it rarely makes a difference in terms of which teams qualify for the post-season.

All things Thrashers + stats: www.birdwatchersanonymous.com

by The Falconer on Nov 24, 2009 10:29 PM EST reply actions

Tampa did sneak in to the playoffs as the 21st seed in 06-07, but otherwise it hasn’t upended the standings too much. I wonder if seeding by conference and unbalanced divisions makes as big a difference?

by Hawerchuk on Nov 25, 2009 12:58 AM EST up reply actions

it rarely makes a difference in terms of which teams qualify for the post-season.

2005-06
Edmonton Oilers 28-28 (13-13) = 95 points
Vancouver Canucks 34-32 (8-8) = 92 points

Oilers were an exactly .500 team in both regulation and overtime, Vancouver slightly better in regulation and also .500 in overtime. But they made the mistake of not playing enough overtime games.

By the Three-Point Must rule, it would have been 126 points for Vancouver, 123 (= .500!) for Edmonton. Of course, as an Oiler fan I’m happy my team took advantage and almost won the Stanley Cup as a result, but i still don’t think it’s right.

To your more significant point, Falconer, it rarely makes a difference because all bubble teams are forced to play politics within games to maximize their return, so they all rack up 8 or 10 or 12 Bettman points, however you want to define them (to me BPs are overtime wins rather than losses, but you can make a case either way). Thus the bubble for 8th out of 15 has swollen from 82 points (.500) to ~92 ( ~ .560). Not only does that make for standings from Bizarroworld, far more seriously it compromises the integrity of the game on the ice.

Writer for The Copper & Blue and primary shareholder of Zorg Industries

by Bruce McCurdy on Nov 24, 2009 10:55 PM EST reply actions

That Oilers team is my typical example for the pitfalls of handing out bonus points for OTLs and SOWs. The Oilers were not one of the top eight teams in the West that year, but got a playoff spot on the strength of a random number generator. From a lottery pick to game 7 of the finals.

by Resolute on Nov 25, 2009 10:56 AM EST up reply actions

Anaheim should have been 9th last year, but snuck in at 8th…The Sharks should protest!

by Hawerchuk on Nov 25, 2009 11:02 AM EST up reply actions

to me BPs are overtime wins rather than losses, but you can make a case either way

That Oilers team is my typical example for the pitfalls of handing out bonus points for OTLs and SOWs.

Hey Resolute, that’s an interesting take on Bettman Points. Awarded to the loser in overtime (2 and 1 instead of 2 and 0) but awarded to the winner in a shootout (2 and 1 instead of 1 and 1). I’ve been struggling with that assignation — who really gets the Bettman Point? — and my take has been it’s a sequential thing: that once the “traditional” two points are split in regulation, whoever wins either OT or the SO gets the Bettman Point. But your (?) spin on it certainly has some merit.

All I know for sure is it’s one hell of a mess.

Writer for The Copper & Blue and primary shareholder of Zorg Industries

by Bruce McCurdy on Nov 25, 2009 1:38 PM EST up reply actions

I could live with the “bonus” point being awarded via the “tie prevention mechanism” as long as teams that win in regulation are rewarded for that achievement. Make all games worth 3 points. Pretty simple.

All things Thrashers + stats: www.birdwatchersanonymous.com

by The Falconer on Nov 25, 2009 9:10 PM EST up reply actions

Yup

I’ve been on that horse since about 15 minutes after the NHL first announced the Bettman Point Ver 1.0 in 1999. It balances what you lose with what you win; the two teams split the three points without “creating” a third one out of thin air. It isn’t exactly rocket science.

Writer for The Copper & Blue and primary shareholder of Zorg Industries

by Bruce McCurdy on Nov 26, 2009 2:34 AM EST up reply actions