One of the oldest saws in the sports-narrative world is the notion that defense wins championships, and hockey is no exception. Ever since the mid-1990s, when the advent of successful defensive schemes like New Jersey’s neutral-zone trap and Detroit’s left-wing lock brought an end to the NHL’s “firewagon era”, many observers have taken for granted that, while fast-moving, exciting hockey might be what the fans like to see, stout defending and a slower pace are what leads to the Stanley Cup. Since 2009, every postseason loss by the Pittsburgh Penguins has been accompanied by the narrative that they lack a true “shutdown defenseman” (strangely, this hasn’t prevented them from winning 203 games over that time period, or from being one of the league’s better shot-prevention teams since their Cup win). It was the same sort of thinking that led the Washington Capitals, in 2011, to abandon an approach that had won them the President’s Trophy just a season before, without the hoped-for results. More recently, this piece at the Shutdown Line blog asked whether the Carolina Hurricanes can contend with the high-event style they played in 2013. Insofar as hockey analysis has identified measures of team play that are strongly correlated with winning, it seems appropriate to ask whether fast-paced, high-event teams are, in fact, unsound. How much does the pace of a team’s game matter?
In order to study this empirically, I first needed a measure of the event rate or pace at which teams play. For this, I decided on the total rate of shot attempts per 60 minutes, or (Corsi events for + Corsi events against)/60. A team that consistently plays games in which tons of pucks are being thrown at the net can be characterized as a firewagon squad, while a low rate of attempts suggests a team spending a lot of time in the neutral zone. These data are available for each team, by season (going back to 2007-08, for a total database of 180 team-seasons), at stats.hockeyanalysis.com. Initially, I wanted to look at how consistent teams’ event rates were over time, to get a sense of how reliably a team can be characterized as “firewagon” or “trapping”. As such, I graphed each team’s event rate over time; in order to make the data somewhat readable, I broke the teams out by division.
Right off the bat, there are some interesting relationships in the data. Specifically:
- It does look like high-event teams can still win things: in the 2010-11 season, the team playing at the second-highest event rate (116.8 total Corsi/60) was none other than the Boston Bruins.
- The 2007-08 Red Wings played at a relatively slow-paced 96.6 rate, which I guess can happen when you almost never allow shot attempts against.
- The slowest-paced team of the past three seasons? You can probably guess: New Jersey (94.3, 91.3, and 90.5 Corsi/60 in 2011, 2012, and 2013 respectively).
- The teams in the Metropolitan division vary widely in the style they play, in contrast to the other three divisions.
- When it comes to specific teams, reputation and data don’t always agree. Regardless of their all-offense reputation, the Penguins don’t play an exceptionally high-event game. Nor do “skill” teams like Detroit, Chicago, or Vancouver. In contrast, teams often known for their defense (e.g., Boston, Ottawa, San Jose, Phoenix) actually play pretty fast hockey.
I also compiled these numbers over the past 6 seasons by game score, to look at whether the pace of the game changes by score. These data are depicted in the table below. As you can see, hockey games slow down when one team is up relative to tied games, though the effect isn’t particularly dramatic. It does suggest, though, that leading teams slow the game down more than trailing teams speed it up.
I also wanted to explore the relationship of event rate to other variables of interest. In addition to the event rate data described above, I also gathered teams’ Fenwick Close data from Behind the Net, and team win percentages from the standings records at ESPN. I imported these data into R, and used the pglm package to estimate logistic regressions modeling these variables as a function of the event rate (pglm* allows you to take serial correlation into account when estimating the variance-covariance matrix, which is important for avoiding biased standard errors). The goal of these analyses was to estimate odds ratios characterizing the effect of event rate; a ratio of 1.00 indicates no relationship. The results? Event rate had a statistically significant negative relationship to both Fenwick Close % (OR = 0.591) and win percentage (OR = 0.896).**
So what does all of this mean? First of all, it means that the pace of a team’s game is meaningful in determining whether or not they’ll be successful. It’s not absolute by any means (for example, low-event hockey didn’t help the Devils much last season), but teams playing firewagon hockey are consistently less likely to win than more conservative teams. If I’m offering an explanation, I’d guess it comes back to goaltenders: unless you can trot out a goalie playing at the level of circa-2011 Tim Thomas, a high-event style means your defense is going to leak goals in a big way. Of course, there’s that relationship to Fenwick Close to consider as well; over the past six seasons, high-event teams have tended to be poor possession clubs as well, and few will argue that a weak puck control game is a recipe for success. My thinking (again, I’m speculating) is that teams playing a consistently low-event style are those that play effectively in the neutral zone, which is likely to show up in their shooting differential as well. It could also be related to the common tendency to play a dump-and-chase style; good possession clubs (think Chicago, Los Angeles, or St. Louis) are often very adept at moving pucks out of their end, making it very hard to generate shot attempts against them off of dump-ins. At the end of the day, all this means that the conventional wisdom on firewagon hockey holds at least some water. If you want to win in the NHL, you need to be able to slow things down.
* pglm = generalized linear models for panel data
** There were a number of other regression analyses I tried to pursue. I wanted to look at the correlation between event rate and shooting and save percentages; problem is, at the team-season level, these measures don’t vary enough for the regression models to be stable, and I wasn’t comfortable interpreting the results. I also wanted to explore multivariate regressions estimating the effect of event rate on winning probability with puck possession and/or PDO held constant. The strong correlation between event rate and Fenwick Close confounded the former analysis, and the latter analysis was very difficult to interpret given the limited variability mentioned earlier.