In just four games, the Boston Bruins handily dispatched of the Pittsburgh Penguins in the Eastern Conference Finals of the 2013 NHL Playoffs. The fact that Boston took the series shouldn’t come as a huge surprise to anyone; despite the star-studded lineup Pittsburgh can roll out including former MVPs Sidney Crosby and Evgeni Malkin and 2013 Norris Trophy nominee (as the NHL’s top defenseman) Kris Letang, Boston has a lot of depth both up front and on the point, as well as one of the top net-minders on the planet in Tuukka Rask. Boston was one of the best teams all season and there usually would not be much shame in losing to a team like that.
Usually.
But the fashion in which Boston dominated Pittsburgh for most of the series was not what I expected. The Penguins were the NHL’s highest scoring team in the regular season, averaging 3.38 goals for/game.* Not only that, Pittsburgh won all three regular season meetings with Boston, scoring eight goals in the process. These wins were despite the fact that Malkin didn’t play a game against Boston in the regular season and Crosby missed one of the three.
*note: even though Pittsburgh only scored two goals against Boston in four games, they were so prolific in scoring in the first two rounds of the playoffs that Pittsburgh is STILL the highest scoring team in goals/game in these 2013 playoffs.
So what caused Pittsburgh to only score two goals in four games against a Bruins team that allowed 2.67 goals for per game against the Penguins in the regular season? This article is Part 1 of 2. This part will cover the team philosophy of the Pittsburgh Penguins (one of them, anyway) while Part 2 (coming on Thursday) will discuss what went wrong specifically in these playoffs for this team of superstars. And yes, my contention is the two are related.
The Pittsburgh Penguins Take A Different Look At Hockey
It’s been written about before but it bears repeating: A lot of what happens in hockey is luck-driven. There are countless studies, blogs and columns that provide evidence that teams and players cannot drive shot quality and discuss the predictive nature of puck possession. Here are a couple:
– NHL Numbers looks at a couple of different common premises and why they’re incorrect (usually).
– Arctic Ice Hockey also looks at the falsity of a few different premises and discusses to what level individual talent (specifically, shooting prowess) can drive shot quality and account for team wins.
For those that just want to skip the reading (you shouldn’t) this is what we know:
– Fenwick – the rate of shot attempts for a team (or player) that doesn’t include blocked shots – has a high correlation to winning. Corsi (which includes blocked shots) also does, but not to the same level of Fenwick. This is because shot-blocking is a skill that can vary wildly between different teams and players.
– Teams and players do not drive shot quality to any significant degree. That is why Fenwick and Corsi are critical: over a large data set, teams will have roughly the same scoring chance/shot attempt ratio. It is the volume of shot attempts that is crucial (thus indicating high levels of puck possession), rather than trying to get a shot off “in a good spot”.
Puck possession is both team and player driven, shot quality is not. Puck possession is the best predictive tool we currently possess to figure out who’s going to win. Sure, there are problems – sample size is always an issue, luck is a significant factor in the outcome of most games – but it’s very, very useful.
Pittsburgh is trying to do something different. James Mirtle, one of the few “big name” hockey writers I have a great deal of respect for, wrote an article about two and a half months ago that discussed Pittsburgh’s attempt to acquire players (namely James Neal) that would be able to buck the evidence that I presented above. They were, in essence, trying to assemble a team that could drive shot-quality instead of shot-abundance.
This approach to hockey bears out in the evidence we have. Pittsburgh, despite all the talent, was 14th in the NHL in the lockout shortened season in Fenwick Close% (a team’s Fenwick is when games are within a goal in the first or second period or games that were tied in the third period). Even though they were just 14th in Fenwick Close%, and not even in the top 10 in shots on goal/game, they finished with the highest goal total in the NHL in the regular season.
Their approach – put highly-skilled shooters like Neal and Jarome Iginla (though not so much anymore) around highly-skilled playmakers– appeared to be working.
Predicted Goals For
Their approach was modeled by a team at The Sports Analytics Institute. Their model is called Predicted Goals For (PGS), which is the first in their four-step modelling process. Again, to save time:
– The model tries to remove the luck factor from the process. It should be noted that they do not discuss the luck factor in Fenwick or Corsi, but rather the luck factor in “traditional” statistics like goals, assists and plus/minus.
– PGS doesn’t necessarily try to predict who the best player could be in terms of shot quality but rather who the most effective player could be when put on a line with exceptional playmakers (e.g. Crosby and Malkin).
– PGS is a model that tries to define the true ability of a player to score a goal on a given shot using probability. The PGS model derives its information from shot totals rather than goal totals, like shooting percentage does. The advantage of using shot totals rather than goal totals is that shooting percentage can vary wildly over a given player’s career. Alex Ovechkin has shot as low as 8.7% (2010-2011) and as high as 14.6% (2007-2008) and over a full season that difference of 5.9% is an average Ovechkin season of 416 shots—meaning a variation of around 25 goals in 82 games.
This seems really exciting for stat-heads. It is a model that hopefully can quiet down the noise around goal-scoring (and eventually, goal-allowing) data, which are the biggest factors in win expectancy. However, by their own admission, there are limitations to this model in its early stages:
– “PGS does not account for situational play”. This is very problematic. What we call score effects, or how a situation in a game can determine puck possession numbers, can greatly skew overall totals. Teams with a lead late in a game usually go into a “defensive shell” which leads to lower possession numbers and thus lower shot totals. This valuation of lead protection can be an issue, as demonstrated here.
– “PGS is allocated to all the players that are on the ice at the time of the shot”. In other words, instead of just goals and assists being counted for the 1-3 players that contributed to these stat columns, all the players on the ice are credited with contributing to shot location, shot type and shot quality. I can understand there are some limitations to this – did a defenseman who just hopped on the ice really contribute anything to that 3 on 2? – however, this is at least the proper approach to take on a basic level.
– “… A Player’s PGS … for a game do not account for decreasing marginal returns to time on ice”. What this means is that PGS does not take into consideration how exceedingly high ice-time totals can decrease performance. All other things being equal, a 25 minute-per-game defenseman will be more fatigued in the late stages of the third period than a 15 minute-per-game defenseman. Again, this prevents obvious problems.
These problems are accounted for in the later stages of their four-step modelling process; Stage 2 (Predicted Wins) takes into account score effects while Stage 3 (Contribution to Winning) takes the players on the ice and decreasing returns on time on ice into account. These steps are critical to getting the whole picture into focus, but there is one more problem with this modelling process; while this undoubtedly improves upon traditional statistics like goals, assists and shooting percentage, how much does it improve on Fenwick and Corsi?
Part 2, what went wrong in this playoff run specifically for Pittsburgh, will be published on Thursday.
As always, thank you to Behind the Net, Hockey Analysis and Hockey Reference for their resources.
