Meghan Hall

This data set includes all regular season NHL games from the 2015 season through the 2019 season, until the pause. “Shots” encompasses all unblocked shot attempts: goals, shots on goal, and missed shots. Blocked shots are excluded because those are recorded at the location of the block, not the location of the shot.

In the play-by-play game data made public by the NHL, most events include a location: an X coordinate (measuring from the center line, 0 to 100) and a Y coordinate (measuring from mid-rink, -43 to 43) to represent where on the ice the event took place. These locations are manually recorded, and since humans make mistakes, we would expect to see some random error that would even itself out over time. But by looking at all the shot locations per arena (i.e., all the shots recorded in an arena for a given season, for both the home team and all the away teams), we can investigate to see whether there is systematic error in certain arenas. That is, do certain arenas record shots less accurately?

We can look at this by comparing the distributions of shot locations in the entire league to each home arena. For example, when looking at the X coordinates in the graph below, the most popular recorded location for a shot is at 79 feet (for context, the goal line is at 89 feet). League-wide, over this time frame, 2.7% of all shots were recorded at that coordinate. However, the individual arena values range from 2.1% (Toronto) to 3.7% (Pittsburgh).

You can see the full distribution below for the X coordinate. The dark line is the league average, while the gray lines represent each individual arena.

Two initial reactions might be: those differences don’t look that large or maybe there are true characteristics that drive those discrepancies. Perhaps Pittsburgh, for example, takes so many more shots closer to the net that it skews the average shot location for all the shots recorded at that arena.

However, we can test that by looking at the exact same distribution, but this time focused on away teams rather than by home arenas. So instead of being driven by all the shots recorded by the home arena, the individual lines below encompass all the shots recorded while a certain team was the away team.

There is much less variation than in the first graph, so it’s clear that there are some differences in recorded shot locations among arenas.

This question of possible systematic error is not a new one—Gabe Desjardins wrote an article for Arctic Ice Hockey back in 2010 that focused on the shot locations in Madison Square Garden in particular, and the graphs in that piece inspired my approach here. Alan Ryder in 2007 also found systematic bias in shot locations, particularly at MSG, and many others, including Michael Schuckers in 2014, have written about rink effects in the recording of events generally. I hadn’t seen shot location data like this recently and wanted to update it for the last handful of seasons.

Why does this matter? Expected goals models that use public data rely heavily on shot location, as it’s an important feature in those models. Being aware of possible systematic bias among arenas can add context to the results.

In order to summarize the differences per team over this time period, we can create a measure of comparison by finding the absolute value of the difference between the league average and the team’s value at each coordinate and summing those values. The top and bottom 10 teams (i.e., the teams that had the largest and smallest differences from the average, respectively) are shown below for the X coordinate, which has more variation.

Team	Measure
T.B	0.341
CAR	0.219
NYR	0.213
ARI	0.209
BOS	0.208
ANA	0.205
PHI	0.201
STL	0.185
MTL	0.184
TOR	0.165

Team	Measure
EDM	0.079
L.A	0.092
NYI	0.093
NSH	0.098
CGY	0.102
WSH	0.103
VGK	0.103
COL	0.107
DET	0.112
S.J	0.119

You can explore this data by team and by year in the dashboard below, which is also available here. (If you’re on reading this on mobile, clicking the link will likely give you a better viewing experience.)