Sunday, January 11, 2009

The Bruins

The Bruins have been one of the surprise teams this year, what with them having the best record in the league at the halfway point (few would have predicted this to be so). They also have the best goal differential, so it's not as if they've been lucky in the conventional sense by winning a lot of close games. However, just because a team's record is proportional to its goal differential doesn't necessarily mean that it hasn't been lucky.

This is a chart showing how the Bruins have fared in various game situations so far this season. The numbers are as of 01/07/08. A couple things can be said about these numbers:

1. The Bruins success appears to largely be a product of the percentages. They have the best shooting percentage in the league, as well as the best save percentage. This also holds true at even strength.

2. For a team with such a good record and goal differential, the Bruins are anomalous in that they're pretty average with respect to shot differential. In fact, they get outshot on average.

The Percentages

The problem for Boston is that there isn't a great deal of repeatability in terms of the percentages, particularly at even strength. This post by Tyler at mc79hockey demonstrates how the sum of a team's even strength shooting percentage and its even strength save percentage tends to regress to 100 as the season progresses. The Bruins currently sit at ~105. If I was a betting man, I'd place money on that figure significantly decreasing by April.

Are Boston's percentages at all sustainable?

We know from past posts that, while fluctuations in the percentages do indeed have little sustain in the future, a team is able to reliably influence its shooting/save through shot quality. Shot quality is moderately correlated with the percentages and is substantially reliable. Thus, over a sufficiently large sample of games, there would still likely be team-to-team variation in the percentages, with this effect being mediated by shot quality.

In past seasons, the team that leads the league in shot quality for typically has a shot quality index of roughly 1.1. That is, that team takes shots that, on average, are 10% more likely to result in a goal than the average team.

Conversely, the team that leads in the league in shot quality against typically has a shot quality index of roughly 0.9. That is, that team allows shots that, on average, are 10% less likely to result in a goal against than the average team.

If we make the very conservative assumption that Boston currently leads the league in both shot quality for and shot quality against, then we can estimate the Bruins' expected shooting percentage based on these adjustments.

Expected shooting percentage = shot quality for index * league average shooting percentage
Expected save percentage = 1- ( shot quality against index * league average shooting percentage)

League average shooting%: 0.0917

Boston's expected shooting percentage: 1.1*0.0917 = 0.10
Boston's expected save percentage: 1-(0.9*0.0917)= 0.917

Boston's actual shooting percentage: 0.118
Boston's actual save percentage: 0.93

Therefore, even if we assume that Boston currently leads the league in both shot quality for and shot quality against, the Bruins' have still outperformed their expected shooting percentage and expected save percentage. While far from constituting definitive proof of good luck, it is suggestive of it.

In actuality, the Bruins have not been leading the league in either shot quality for or in shot quality against. Hockeynumbers tabulates data on shot quality that is periodically updated throughout the season. While the data is only available for specific game situations (EV, PP, SH), figures for overall shot quality can be obtained by dividing each team's expected goals for/goals against by their corresponding shots for/shots against total, and then expressing the resulting figure relative to the league average.

In addition to having a negative shot differential, the Bruins are below average in both shot quality for and shot quality against, thus making it even less likely that they'll replicate their impressive shooting/save percentage in second half. Indeed, Boston is in the red in terms of its expected goal differential, as is nicely illustrated here

Boston's 'true' even strength shooting/save percentage

As displayed in the table at the beginning of the post, the fact that Boston has managed to lead to the league in both shooting and save percentage is largely tied to even strength play -- that is, the overall percentages are largely being driven by the exceptional even strength percentages. Therefore, the sustainability of Boston's overall percentages is critically contingent upon sustaining its high percentages at even strength. Boston's shooting/save percentage at even strength will likely fall to something more reasonable by the time the season has ended. At the same time, however, it's unlikely that its EV shooting/save percentage is merely average.

To illustrate this, assume that Boston's true underlying even strength shooting percentage is exactly league average (~0.084), with the same holding true for its even strength save percentage (~0.916). The Bruins have taken 890 shots at even strength so far this season, while allowing 920. If a team with a true EV shooting percentage of 0.084 takes 890 shots, the probability of shooting 0.109 or better by chance alone is remote (about 4 times per thousand). Likewise, if a team with a true EV save percentage of 0.916 has 920 shots against, the probability of having a save percentage better than or equal to 0.939 by chance is equally minuscule (about 5-6 times per thousand). Thus, Boston's underlying EV shooting/save percentage -- while almost certainly lower than what they've attained thus far -- is probably above average. Therefore, a complete regression to the mean is unlikely.


Sunny Mehta said...

Good post. I too have been wondering about Boston, while simultaneously betting against them. :)

Tyler's post about PDO numbers did seem to show that, on average, teams' S% + SV% tends to regress to the mean when it's out of whack. It doesn't prove that one particular team can't sustain a better-than-average S% or SV% throughout a given season, though. You alluded to that fact, and I'm sure Tyler himself wouldn't disagree either.

It'd be nice to see how teams fare over sample sizes larger than one season. But the problem is that rosters change. Would that great offensive Buffalo team a few years ago have been able to sustain a high S% the following year, and the following year? What about a team's SV% with Roberto Luongo in net? Who knows.

From just glancing at numbers on Behind The Net from last season and this season, I noticed a couple peculiar things. Ottawa had an excellent ES S% last season. This year it's dead last. They didn't make a lot of personnel changes, right?

Pittsburgh had a great S% last season. This year it's still excellent. Ditto for Dallas. As you mentioned in a previous blog comment, Minnesota is a team that seems to always outperform its ratios.

As for why some of it shows up in shot quality models (e.g. Minnesota having an excellent SQA) and some of it doesn't (e.g. Pittsburgh has an SQF of average), who knows.

Possible explanations:

1) Data recording error.
2) Skilled offensive players being able to outperform SQF models.
3) Randomness.
4) ???

I'm really curious to see what happens to Boston in the second half.

overpass said...

From just glancing at numbers on Behind The Net from last season and this season, I noticed a couple peculiar things. Ottawa had an excellent ES S% last season. This year it's dead last. They didn't make a lot of personnel changes, right?

I looked into Ottawa's complete collapse after the All-Star break last year. ES S% was the big culprit, dropping from from 9.5% before the break to 7.5% after the break, so it didn't start this year for Ottawa. Their forwards are essentially the same group, but the depth forwards after the big line basically lost any finish they ever had halfway through last year. I think a lot of it has to do the fact that the offensive support from the blueline is way down, with Corvo, Redden, and Meszaros all having left.

Vic Ferrari said...

Matt linked to the stats for when the score was tied, it's relevant here I think.

I mean usually teams that have the better of the territorial advantage at evens end up with the lead a lot late, and they tend to sit on it. So not so much in the way of odd man rushes. So that dimishes the shot proxies measures a smidge.

The Bruins look to have had far the better of the play at evens as well, but also the luck then (and I don't dispute that Thomas and Fernandez are better at stopping pucks than most, and that the Bruins have more than their fair share of guys that can create plays and finish them ... still, those numbers are off the hook).

So the super percentages working for them when the score is tied diinishes their overall shots-stuff results, because being the absolute bomb when the score is tied through luck means that you'll be sitting on the lead more than you otherwise would. Their strong special teams factor in their as well, of course.

And the opposite applies to CBJ. Anyone who has watched them play knows that they are the real deal at even strength, but the tragic special teams have left them trailing more than you would expect ... so their underlying numbers at evens, at least by shot proxies, flatters them a bit.

And of course focussing on the wackiest looking tree in the forest without offering up the fores itself as context ... dangerous business at the best of times imho.

Sunny Mehta said...

Excellent points, Vic. I always love how you seem to make a point to keep context, or "the forest", in mind.

I meant to say this in my previous post, but I happened to check out your TOI scripts for the ES Tied situations this season. Boston's percentages are crazy high there as well. But their Corsi, while certainly good, is nowhere near the teams like San Jose or Detroit.

Any thoughts on that? I hear what you're saying about the fact that when you run sick good with the percentages and build leads a lot, it's gonna affect the overall shot ratio numbers at ES.

But WHEN THE SCORE IS TIED, shouldn't a team's Corsi and percentages be pretty much independent? I.e. - shouldn't Boston (a "very good" team) be dominating Corsi at ES w/score tied regardless of how high their percentages are?

I absolutely LOVE that script of yours, the two problems I have though are that you don't have a column for actual time on ice. So the raw corsi can be misleading if one team spent a significantly higher amount of time at ES w/score tied than another team. (Is there anyway you could easily add a column for raw ice time totals?)

Also, of course, there's the sample size issues. One FULL SEASON isn't even a big enough sample, let alone half a season, let alone half a season's worth of minutes in which the game was tied at ES.

JLikens said...


Yeah, I agree in that it would be very interesting to see the spread in percentages over an adequately large (i.e. hundreds and hundreds of games) sample. It's unfortunate that a single season fails to provide that.

RE: Ottawa, Pittsburgh, and Dallas

Not sure what Ottawa's problem is. They outperformed their expected GF in 03-04, 05-06, 06-07 and, while I haven't checked, likely in 02-03 and last year as well. Perhaps the decline in finishing ability is related to the exodus of skilled and/or offensively gifted players over the last few seasons -- namely, as overpass alluded to, puck moving defensemen. That said, I think that they've also been terribly unlucky this year in terms of scoring goals.

While Pittsburgh has been average in SQF this year, they were one of the best in the league in 06-07 (by Ryder's numbers) as well as last year (by Javageek's numbers). I suspect that their high percentages are a combination of having a plethora of talented offensive players (as you mentioned) and shot quality (part of this is shot selectivity, I think).

Dallas was one of the worst teams in the league last year in terms of shot quality against, which suggests that they might have done away with the defense-first philosophy in favor of a more open system. This would certainly (partly) explain their shooting percentage numbers (both for this year and last), as well as the decline in team save percentages (likely not all Turco's fault).

RE: Vic Ferrari

Yeah, I completely overlooked the leading/trailing effect and its influence on shooting/save percentage. You're right in that the effect of this has been to make the underlying numbers prejudicial to Boston's actual level of EV play, which has been above average (as ascertained through the fact that they outshoot the other team when the score is tied).

As you said, context is critical here.

Vic Ferrari said...

I didn't mean to seem short, JLikens. For what it's worth, I think your stuff is terrifically honest.

Sunny Mehta said...

JLikens and overpass,

re: Ottawa and the possibility of their demise being somewhat related to some bad shooting luck

I have two words: Antoine Vermette

He is on pace for more shots this season than last season. His average distance of shot this season is IDENTICAL to what it was last season.

Last season his ES S% was 9.4. This season it's 1.3 percent.


Tom said...

I'm not even remotely qualified to engage in the nitty-gritty data analysis, but I see the Bruins often enough to know what's happening with their shots for/against ratio:

The Bruins have some of the biggest, nastiest defensive players in the league (Chara, Lucic, Stuart, Hnidy, Thornton, etc.) and run a defensive system that usually has at least 3 players in front of the net. Sometimes they stack all 5 players in the slot, especially when defending a lead.

So when opponents take a shot, more often than not it's a pretty low-percentage attempt. Shots from the point, from sharp angles, etc. When you've got a good goalie tandem and a bunch of large defensemen who stay at home, low-percentage shots are fairly harmless. On a typical play the goalie will make the save and the defense will sweep away the rebound. Perhaps the opponent gathers the loose puck and takes another low-quality shot -- but that's not improving their chances much. Game after game, the Bruins allow upwards of 10 shots in a period but the goalie only makes 2 or 3 challenging saves.

At the same time, the Bruins (when healthy) feature forwards like Kessel, Ryder and Sturm who score most of their goals on quick wrist shots. I would go so far as to say that the Bruins have the best quick-strike offense in the East, in large part because they can absorb 3 or 4 shots against, break quickly down the ice, and score a goal with 1 shot.

At the end of the day, it's nothing unusual for the Bruins to dominate a game yet still take fewer shots than their opponent. For example, they beat Atlanta (a clearly inferior team) by 4 goals despite taking 8 fewer shots. In fact, they've been outshot in 5 of their last 6 wins. During their 10-game win streak they were outshot 7 times.

I guess you can call that "luck" from a statistical standpoint, but only if you make the assumption that a team must be lucky to win if they're outshot. The Bruins are a textbook counterpoint to that assumption, since their gameplan doesn't involve a shot counter.