• # Safe is Death

## by Tyler Dellow • November 28, 2013 • Hockey • 6 Comments

This is a little bit of a vegetables post but it’s sort of fun so maybe like vegetables with some icing on them. We’re getting to the point that stats are big enough that people are starting to ask questions like “Lots of times the team with more 5v5 Corsii loses the game and what’s the deal with that?”

I’ve been mulling this over. I thought it might be helpful to break games down a little bit and look at what the data tells. In order to do this, I coded every Corsi event between 2007-12 based on the score at the time. Obviously, each event has two different score codes – a one goal lead for one team is a one goal deficit for the opposition. I then took every game and calculated a Corsi% for each team when the game was tied, when there was a one goal lead, a one goal deficit – whatever situation arose during the game.

This permitted me to sort score-tied games based on what the Corsi% was and look at how things changed depending on what the score tied, +1, +2 etc. Corsi% was in a given game. I’ve got three tables that I’m going to present. We’ll start with score-tied Corsi%.

So, for example, when a team posted a score-tied 5v5 Corsi% of between 60.1% and 65% in a game during the 2007-12 seasons, it got 60.3% of the 5v5 goals. Pretty straightforward.

I have 12,276 games in my sample (the NHL didn’t record data for 12 games). 12,230 of those games had at least one 5v5 Corsi event with the score tied. The biggest, most obvious thing that jumps out at me is this: as a team’s Corsi% increases with the score tied in a game, the likelihood that it will score more of the 5v5 goals with the score tied increases too.

There’s a little bit of a PDO effect for teams in games in which they post a sub-50% Corsi% but it’s not very big. It’s not big enough to overcome the effect of being outshot. In general terms, there’s not much difference in that 45% to 55% range but once you get outside that, there’s an effect. There’s a pretty massive difference once you get above 55% or below 45% in a game in terms of the score-tied Corsi%.

Some nerdy math notes – I wouldn’t read anything into the shooting/save percentages being higher away from the centre of the sample. Most games with, say, a 70.1% to 75% Corsi% for one team didn’t have many shots taken with the score tied. It’s more likely that a team will have seven shot attempts out of ten than 14 out of 20. I’ve included a column that notes this effect – S/G is how many shots per game with a Corsi% of X there were, on average.

You’ll note that this increases as we get closer to 50%, when it falls to 11.8 from 22.2. Again, this is a math driven thing as opposed to something real – you can’t have a score-tied Corsi% of 50.1% to 55% without at least 13 shot attempts with the score tied. You can have a 50% score tied Corsi% with only two attempts. As my rules for calculating this are implicitly considering goals, the percentages are skewed by that. You should not conclude that the finishing and goaltending is worse when the Corsi% is exactly 50%.

The most common Corsi% in a game being between 55.1%-60% and 40%-45.9% is, I believe, a function of the probability of a goal being scored on a given 5v5 shot. If you need at least 13 shots to get below that (which implies both teams shooting 7.7% or worse at 5v5, which is below the league average), there are going to be more games where the Corsi% is outside that 45% to 55% range. Again, this is a function of the likelihood of a goal on a given shot and the volume of shots needed to get into that 45-55 range.

Now let’s look at what happens when a team is leading by a goal. How does the game change?

Keep in mind that the ranges here aren’t complementary, like with score tied Corsi%, where one team being at 45% with the score tied means that the other team in the game was at 55%. In this table, a 55% team when leading by one means that there’s another team which was at 45% when trailing by one, which isn’t shown in this table.

Two things leap out at me: first, the same rule applies: as your Corsi% increases when you’re leading a game by one, so too does the percentage of the goals that you score. (There’s a bit of a wrinkle right on 50% for some reason.) Second, there’s a bit more of a PDO effect here. The leading team does see a bit of a save percentage/shooting percentage advantage, to the point that they actually outscore on average if they get out-shot attempted by somewhere between 45.0% to 49.9%. You’d still rather get 50.1%+ of the shot attempts.

You’ll note that the most frequent Corsi% during portions of games where a team is leading by a goal is between 40.0% and 44.9%. I think that there’s a pretty big hint here as to why we frequently see teams that have better records when they’re out Corsid or outshot. Hockey’s the kind of game where, if you’re up one nothing, odds are that you’re going to win because it’s a low goal sport. Even if you’re getting whipped in terms of shot attempts, you’ve still got a shot at a goal and there’s no guarantee there will be any more goals anyway. My suspicion, which I think that the data supports, is that many of the games in which a team is out shot attempted and wins anyway, the shot attempts when they’ve taken the lead flip the needle the other way.

That doesn’t make it necessarily a good strategy. The table’s pretty stark about that.

Two goals up?

The rule still pretty much applies: more shot attempts = more of the goals. There’s a more severe PDO effect here but the game doesn’t flip over to the point that you’re better off allowing more shot attempts than you generate here if you evaluate things on the basis of goals scored.

Hockey’s a funny game because it’s difficult to separate the process from the results. As there are relatively few goals scored in a game, you can have lousy process and, with the right breaks, good results. The more angles I look at this from, the more I come to the conclusion that NHL teams as a whole play too conservatively with a lead and, as a result, are actually less effective at converting leads into wins than they should be.

Email Tyler Dellow at mc79hockey@gmail.com

### 6 Responses to Safe is Death

1. November 28, 2013 at

So is Tyler’s vision starting to go, or did some old folks campaign to get text size enhanced? I keep thinking the cat must’ve stepped on CTRL while I was using my scroll wheel or something because the text is so big.

2. Tyler Dellow
November 28, 2013 at

Just took it down a notch. I was getting complaints.

• November 29, 2013 at

I appreciate it. It was large enough that it was weirding me out to read it that way. This is pretty decent.

3. Peter Lynn
November 28, 2013 at

I’m not kidding when I say that the simple act of increasing the font side on your site may marginally hasten the adoption of advanced stats in the NHL by making it that much easier for people (particularly older fuddy-duddy types) to read your posts..

4. Tom Benjamin
November 29, 2013 at

That doesn’t make it necessarily a good strategy. The table’s pretty stark about that.

It isn’t a strategy, I don’t think. Teams don’t set out to get outshot. They try to get shots and they try to prevent them. While teams do try to get shots, they also try to get good shots because good shots are much more likely to be a goal. For better or worse players frequently pass up low percentage shots to try for a better chance.

NHL teams probably do play too conservatively when they get a lead, but I don’t think there is anything anybody can do about it. It is a human trait, a flaw in our ability to assess risk. Losses loom larger than gains. Players can’t help themselves and they assess risks differently when the game is not tied. While I think they carry it too far, they are not exactly wrong either.

Suppose a player could take a chance that 55% of the time produced a goal and 45% of the time it produced a goals against. Look at the big picture and we would always want the player to go for it no matter what. Get 55% of the goals in your games over the course of a season and you are going to be successful.

Frame it differently – bring it from the big picture down to the actual game level – and it isn’t get quite so simple. Once we are up a goal, the equation becomes a 55% chance of extending the lead and a 45% chance the opponents tie the game. The math is the same and a goal is still a goal but the answer for most of us changes. The opponent’s possible next goal is more valuable than our possible next goal. The goal that extends the lead becomes less and less valuable relative to the goal the opponents could get to tie the game as the clock winds down.

The players should evaluate the risks differently and play somewhat more carefully. The mistake that gets made is when a team suddenly stops taking any risk at all.