Where the home-road split revealed a discrepancy in recorded shots, I then looked at the shooting percentage data to determine whether there was, in fact, a recording bias. I reasoned that if a discrepancy was due to bias, rather than randomness or other factors, there ought to a concomitant increase (in the case of undercounting) or decrease (in the case of overcounting) in the shooting percentage of both teams in games played in the arena in question.
Since that time, others, such as Sunny Mehta, Vic Ferrari, Chris Boersma and Tom Awad, have also examined the issue through work of their own, all of which I would recommend reading.
Two of the arenas that I suspected might be overcounting shots were Bank Atlantic Center in Sunrise, the home of the Panthers, as well as Continental Airlines Arena (and, perhaps, the Prudential Center as well) in New Jersey. While the data on shooting percentage suggested that shots were likely undercounted in New Jersey, the same was not true of Florida.
The purpose of this post is to take a somewhat more refined look at the topic so as to properly determine the existence of bias. While my initial post looked at overall shot totals and overall shooting percentage, it failed to consider the influence of specific game states, such as special teams play and the playing to the score effect. As both of these factors are known to influence shots on goal as well as shooting and save percentage, merely examining the data in aggregate can be more misleading than illuminating. In order to mitigate these concerns, the data below has been broken down according to game situation.
Firstly, the data for Florida. Presented below is a table comparing the number of shots taken by both teams in Florida road games versus Florida home games, broken down by game state and season. Shots that resulted in an empty net goal have been excluded. This essentially confirms what was already known -- that more shots are recorded in Sunrise than elsewhere. Interestingly, the putative bias appears to be confined to even strength, with no effect on special teams.
I've also prepared a similar table that compares the shooting percentage (again, for both teams) in Florida road and home games. As with the previous table, the percentages do not include empty net goals.
Looking at the data, it's difficult to argue for any sort of shot recording bias. The aggregate shooting percentage in Florida home games is identical to the aggregate shooting percentage in Florida road games. The same is essentially true at even strength with the score tied. If shots were, in fact, being overcounted, then one would expect to to observe a lower shooting percentage in Florida home games. But such is not the case.
In the comment thread of this post made by the Contrarian Goaltender, Vic Ferrari surmised that some of the apparent shot recording biases are not biases at all, and that some arenas really do consistently feature more or fewer shots than average, perhaps due to team style, strategy or some other like factor. I think that's probably the best explanation in this case. The Panthers, for whatever reason, seem to play a more exciting style of hockey at home, which serves to increase the shots on goal numbers while leaving the shooting percentage data unaffected. This is consistent with the shot discrepancy being restricted to even strength.
The data for New Jersey tells a different story. Unlike in Florida, New Jersey home games have featured a deficit of shots, rather than an excess.
More significantly, however, this deficit in shots has been accompanied by an increase in the shooting percentage in Devil home games. This implies that the deficit is due to recording bias, rather than some other factor.
Looking solely at even strength play, the shooting percentage in Devil road games from 2003-04 to 2008-09 has been nearly an entire percentage point lower than the shooting percentage in Devils home games during the same period. While the difference may not seem large, it is greater than what one would expect to observe through chance alone. I've included a separate table below that shows a range of expected shooting percentage values, expressed in the form of confidence intervals, for both EV play with the score tied as well as for EV play in general.
This table shows the range in values where one would expect the overall shooting percentage for New Jersey home games to be found, during the period under review (2003-04 to 2008-09), if it is assumed that:
- There is no shot recording bias
- The 'true' shooting percentage in Devils home games is equivalent to that observed in Devils road games.
One final comment: some will have noticed that more shots were recorded in New Jersey home games than road games for both 2007-08 and 2008-09. I take this to mean that the shot recording bias is likely no longer in existence. While it is true that the shooting percentage in Devils home games continues to be higher than in Devils road games, the difference is probably meaningless in the absence of an actual difference in recorded shots. Perhaps the switch to a different arena was accompanied by a change in shot recorders.