2023 SCF game 5 - pulling the goalie

[Hockey Outsider]

A lot of people have been talking about Florida's decision to pull Bobrovsky yesterday, down 4 goals with six minutes left in the game. Obviously, it was very likely they were going to lose either way. But did the decision make sense?

We can make a reasonable ballpark estimate of what would have happened, if the Panthers didn't pull Bobrovsky. Since they were down by four goals, regardless of what Las Vegas did, if the Panthers scored between 0 and 3 goals, they were going to lose.

In order to tie the game, they'd need to score 4 goals, while allowing none. Or they would have to score 5 goals, while allowing 1. And in order to win the game (in regulation), Florida would have had to score 5 goals while allowing none, or 6 goals while allowing no more than one, etc.

Using the Poisson probability distribution, we can make a reasonable ballpark estimate of how the game was likely to unfold. I'll spare everyone the mathematical details, but here's a table showing the outcomes:

(The rows show the number of goals scored by the Panthers, and the columns show the number of goals scored by the Knights. So, for example, the cell showing 2.1% is the probability of Florida scoring two and Las Vegas scoring zero).

Obviously there are a lot of simplifying assumptions that go into this. (The most significant point is I'm assuming the rest of the game unfolds at 5-on-5. A powerplay would change the analysis, and although I could incorporate that, it would take a lot of work, and it likely wouldn't change the outcome in a meaningful way). With that disclaimer in mind, if my inputs are reasonable, this suggests that the most likely outcome for the final six minutes of the game (59% probability) would have been no goals (in which case, Florida loses).

There was about a 31% chance of either team scoring a goal, and about an 8% chance of two goals being scored (either two for Vegas, two for Florida, or one each). Obviously, all of these outcomes would lead to Florida losing.

Adding everything up, I estimate that, had the Panthers not pulled Bobrovsky, they would have had about a 1-in-8,000 chance of tying the game (sum of the yellow cells), and about a 1-in-140,000 chance of winning the game (sum of the green cells).

Click to read more...

 

[Hockey Outsider]

Once the Panthers pulled Bobrovsky, two things happened. First, their expected goal differential plumetted (they had a virtually even expected goal differential before, and under 30% when the goalie was pulled). Second, the rate at which goals were expected to occur skyrocketed. In other words - when a team pulls its goalie, over the long run, it's going to do worse, and it's going to generate way more chances for and against. (This seems obvious).

Here's what the outcomes look like with Bobrovsky pulled:

As expected, there's way more activity. The probability of zero goals being scored dropped from 59% to just 9%. The probability of 3+ goals being scored soared from 2% to 43%. The goal differential was nearly even before, but now it's shifted clearly in Las Vegas's favour. Despite that, the probability of a Florida comeback is higher. Yes, there's a higher chance of them losing by 5, 6 or 7 goals - but who cares? They were going to lose either way. There's now a better chance - though still a very small one - that they score four goals (or score five and allow one, etc).

If they pulled Bobrovsky for the full six minutes, I estimate the Panthers would have had about a 1-in-800 chance of tying the game (and a 1-in-5,500 chance of winning the game in regulation). These are still very low percentages, but overall their chances of avoiding the loss increased about tenfold.

(I made a comment in another thread @tarheelhockey that maybe Maurice pulled the goalie in order to prove to fans and/or the owners that he tried his best. Maybe that was an unfair remark. Pulling the goalie absolutely gave the Panthers a better chance - so Maurice, as the coach, did what he was supposed to. Still, if I were the coach, I'm not sure if I would have bothered. Even with a tenfold increase, the probability was still just 1-in-800. A comeback was still highly unlikely. But with the entire season on the line, it's understandable).

 

[Hockey Outsider]

Also, I'll post this before someone else does:

 

[Bear of Bad News]

Was wondering the numbers on this when it happened - good timing and good post, as always.

 

[TheDawnOfANewTage]

These are the analytics everyone should get behind. Unfortunately there’s still a few know-it-alls who wanna just go by guesswork and feel.

Malcolm Gladwell interviewed some economist hockey fans a few years back, and their numbers were similar. If memory serves, they said that a team down by 2 should pull the goalie with like 9 minutes left, and if down by 1 it should be more like 4-5 minutes. Sounds insane, but sure enough we’re seeing goalies get pulled earlier. You’re already likely to lose, basically, so you wanna give your team as much chance to score as possible. Games should almost never end in a 1-goal difference. Either you tie it or the other team gets an EN goal, it’s just optics that have kept goalies in the net for so long.

 

[PK]

Does the model take into account that it was enough for Florida to score four goals first? What I mean is that a 4-1 with a pulled goalie wouldn’t work for Florida but in some instances they would score four goals and only then get scored on. Which wouldn’t happen in real life because after 4 goals, the goalie would be back on.

Not sure if I’m explaining it well.

 

[Hockey Outsider]

Does the model take into account that it was enough for Florida to score four goals first? What I mean is that a 4-1 with a pulled goalie wouldn’t work for Florida but in some instances they would score four goals and only then get scored on. Which wouldn’t happen in real life because after 4 goals, the goalie would be back on.

Not sure if I’m explaining it well.

No, it's a fair question. The model is essentially assuming that Florida keeps Bobrovsky on the bench for the full six minutes, regardless of what happens. So, according to the model, it's possible that they score 4 goals in three minutes, keep Bobrovsky out, and then allow an EN goal! Obviously that wouldn't happen in real life.

It's possible to refine the model, but it would take me several hours. My ballpark calculation suggests outcomes like that are extremely unlikely (because, if Florida were to tie the game, there would be very little time remaining). Qualitatively, this means that Florida's chances are probably slightly higher than what I posted, but maybe it's something like 1-in-750 (at best) instead of 1-in-800.

 

[BraveCanadian]

These are the analytics everyone should get behind. Unfortunately there’s still a few know-it-alls who wanna just go by guesswork and feel.

Malcolm Gladwell interviewed some economist hockey fans a few years back, and their numbers were similar. If memory serves, they said that a team down by 2 should pull the goalie with like 9 minutes left, and if down by 1 it should be more like 4-5 minutes. Sounds insane, but sure enough we’re seeing goalies get pulled earlier. You’re already likely to lose, basically, so you wanna give your team as much chance to score as possible. Games should almost never end in a 1-goal difference. Either you tie it or the other team gets an EN goal, it’s just optics that have kept goalies in the net for so long.

Has it ever worked?

 

[TheDawnOfANewTage]

Has it ever worked?

Has anyone tried it? We’ve seen a big increase in goalie pulls more towards the 3 minute mark, and a subsequent increase in games tied/won. You get more time to really wear down the other team, whereas before with a goalie pulled around 1:20 you just saw teams run outta time. That shouldn’t happen, is the big idea, either they bury you or you tie it up, and you’re unlikely to score two 5-on-5 goals with those 7+ minutes without a goalie pull. Statistics say it’s still worth going even earlier.

That said, we’ve also seen a change in strategy by the team in the lead, where they shoot from their own zone now. That’s also statistics-based, and if the economists’ numbers are from history maybe that new strategy changes things. The appeal was maybe that you could hem the leading team in their zone, and they’d struggle to get a breakout for the empty netter. Now they’ll just turn and shoot, perhaps negating some of the advantage from an earlier lull.

 

[Hockey Outsider]

Has it ever worked?

I'll post my 2nd meme in this thread (because it actually kind of fits):