Sunday, March 29, 2009

The Blackhawks

The Blackhawks have been an interesting team this year.

First of all, there's plenty of evidence which suggests that the Blackhawks are a pretty solid team. They've accrued one of the best goal differentials in the league up to this point in the season, despite playing in the league's toughest division in the better of the two conferences. They've also soundly outshot the opposition, both at even strength and in general. In these respects, it would be difficult to argue that the Blackhawks are not one of the league's best teams.

What I find unusual, then, is that the Blackhawks expected goal differential, as calculated at hockeynumbers, is only slightly positive. The Blackhawks have allowed some 20 fewer goals than what would be predicted on the basis of shot quality, while having scored some 20 more. Given that their shot ratio is more or less in line with their goal ratio, the implication is that Chicago has been below average in both shot quality for and shot quality against. Indeed, if the expected goals numbers are translated to yield a shot quality index for each team, the Blackhawks do in fact fare quite poorly.

The fact that the Blackhawks tend to allow high quality shots against is not surprising. Firstly, shot quality is repeatable on a year-to-year basis, with Chicago having ranked 27th in the league in that regard last year.

Secondly, teams coached by Joel Quenneville tend to allow high quality shots against on average. Outside of 1999-00, for every year that Quenneville has been a head coach in the NHL, his team has ranked in the bottom half of the league in terms of save percentage, which implies that his teams were surrendering high quality shots against. While it's perhaps true that Quenneville was burdened with poorer than average goaltending during his tenure in both St. Louis and Colorado, shot quality has been directly measured from the 2002-03 season onward, with the results tending to support the argument that Quenneville-coached teams are poor in terms of shot quality against.

The Contrarian Goaltender has also found evidence that the effect of Quenneville's coaching in St.Louis was to reduce the save percentages of his own goalies, who tended to have better save percentages prior to playing under Quenneville (see, for example, the comments section of this post).

With a team save percentage of 0.912, the Blackhawks goaltending has been nothing short of superb this season. However, the fact that the team apparently allows such high quality shots against, not to mention the fact that save percentage is at least partially a product of random statistical variation, inevitably leads to the question of sustainability. Of course, It's true that both Khabibulin and Huet have proven track records, and both goaltenders are certainly better than anything Quenneville had to contend with in St. Louis and Colorado. Nonetheless, I think that it's a potential cause for concern and, at the very least, something that one ought to be mindful of in evaluating the team's prospects for the postseason.

As for the team in general, I'm inclined to think that they're still pretty good, the shot quality numbers notwithstanding. As I intimated in the above paragraph, both Huet and Khabibulin are above average netminders and the Blackhawks team save percentage is in that sense somewhat sustainable. And while it is true that the Blackhawks have exceeded their expected goals for, I'm reluctant to ascribe the difference to luck considering their plethora of offensively talented and creative players. Furthermore, even if I happen to be wrong on these points, their tendency to convincingly outshoot the other team is, if nothing else, encouraging.

Friday, March 20, 2009

Home Recording Bias: Shots on Goal

In previous posts, it was shown how some of the statistics that are recorded by the NHL are subject to a home arena bias. Home arena bias seems to be most pronounced with respect to the RTSS data, which includes statistics like hits, takeaways, giveaways, blocked shots, and so forth.

However, this bias is also observed with less subjective statistics, such as shots on goal. Below is a chart showing how the recording of shots on goal has varied on a site-by-site basis over the last 13 NHL seasons, with the more interesting information highlighted. The values contained in each cell were derived as follows:

[ (Home shots for/60 minutes played + Home shots against/60 minutes played) - (Road shots for/60 minutes played + Road shots against/60 minutes played) ]

Basically, the formula boils down to this: the total shots on goal (by both teams) in games played by a particular team at home, minus the total shots on goal in games played by that particular team on the road, with ice time controlled for. Empty net situations were not included, both in terms of shots on goal and minutes played. I should also mention that the ice time data for 1994-95 to 1997-98 is approximate.

While the home recording bias for shots on goal is not large, it is nonetheless clear that not all NHL arenas record shots equally. The recorders in Vancouver have over the years been very conservative in their shot counting, although the effect appears to have been moderated in the last couple seasons. There were more shots/60 in Colorado and Anaheim home games than there were in road games played by those two teams for every single season analyzed. Shot recording in Nashville has been generous ever since their inaugural season (although I'm not sure what happened in 2003-04), whereas the reverse has been true in Minnesota. In both New Jersey and Dallas, shots have been harder to come by since around the turn of the millennium. Finally, a bias towards overcounting seems to have materialized in Sunrise over the last couple years.

Of course, the above values are not necessarily demonstrative of a bias; they are merely suggestive of it. They ought to be supplemented with data on shooting percentage in order to allow for a more confident interpretation of the effect.

Why shooting percentage? Well, if an arena does in fact undercount or overcount shots on goal, the bias should concern saves rather than goals. The reasoning here is not difficult. Each shot on goal that results in a goal is necessarily a shot on goal -- there is no room for the exercise of discretion on the part of the shot recorder. However, in the case of a shot on goal that does not result in a goal, the shot recorder is permitted a modicum of discretion, and what constitutes a shot for some may not constitute a shot for others. Undercounting shots should have a positive effect on shooting percentage, whereas overcounting would be expected to have a positive effect on save percentage.

Therefore, it can be seen how accompanying data on shooting percentage will shed light on the extent to which a true bias is present. For the teams for which a bias is suspected -- Florida, Dallas, New Jersey, Nashville and Minnesota, I've included information below on the shooting percentage in games played by those teams, broken down into road and home situations. The data in the left column is the shooting percentage in road games played by the team indicated in the upper left hand corner. The data in the right column is the shooting percentage in that team's home games. It is important to stress that these figures do not merely represent the road and home shooting percentages of the team in question. Rather, the figures represent the overall shooting percentage (that is, both for the team in question as well as their opponents) in road or home games played by that team during the season indicated.

In terms of New Jersey, it seems that the tendency for undercounting shots at Continental Airlines arena began during the 2001-02 season. In every subsequent year, the shooting percentage in New Jersey home games has been higher than in New Jersey road games. Indeed, this bias has had some negative effect on the home save percentage of Devils goaltenders during the period in question (keeping in mind that the average home save percentage for NHL teams tends to be 0.003 to 0.007 higher than the average road save percentage).

The Devils led the league in shot quality against from 2002-03 to 2006-07. They also surrendered the fewest powerplays against during each of these seasons. While the Devils' road save percentage during the period in question reflects this fact, their home save percentage does not. At least part of the discrepancy can be accounted for by recording bias.

The shooting percentage data for Dallas games suggests that the bias emerged during the 1998-99 season. The effect appears to be large.

The data for Minnesota is less clear. The difference is in the predicted direction for four of the seasons (2001-02, 2002-03, 2003-04, 2005-06), the opposite direction for two of them (2006-07, 2007-08), with there being no difference in 2000-01. It's possible that:

a. The Wild simply play more conservatively at home.
b. The bias has lessened over time.
c. The results can be explained through some combination of the above factors.

The data on Nashville reveals that the bias is genuine, or at least was genuine prior to 2006-07 and 2007-08. The results for the last two seasons, taken together, indicate that the bias may no longer persist.

Finally, the data on Florida implies that there exists no recording bias at the BankAtlantic Center. The shooting percentage in Panther road games is largely indistinguishable from the shooting percentage in Panther home games. The fact that Florida goaltenders (Luongo, Anderson, Vokoun) have placed among the league leaders in save percentage in each of past several seasons (outside of 2006-07) have led some to conclude that their must be something amiss, given that the Panthers do not employ any type of defensive system and have not been an otherwise successful hockey team during that period. However, the most probable explanation is that the Florida has merely benefited from having a series of good goaltenders -- indeed, both Luongo and Vokoun have posted very impressive numbers elsewhere (granted, the data on Nashville suggests that the latter's save percentage may have been somewhat inflated by recording bias during his stay in the Music City).