The last several years have seen the influx and normalization of a new way of studying hockey. A new set of statistical measures, designed to capture different or more complete elements of the game, has come to the forefront of hockey analysis. Measures like Corsi, Fenwick, and a host of more complicated statistics have joined the conversation in terms of analyzing hockey performance. However, he analysis of one position has remained fairly unaffected by this shift towards "fancy stats." How goaltenders are assessed and evaluated has remained fairly constant throughout the growing popularity of these new statistical measures.
Corsi and Fenwick Save Percentages
If the new-wave of statistical measures can be used to enhance the hockey audience's understanding of the game in relation to puck possession and the effect of special teams, surely these measures can also be used to paint a more complete picture of the performance of a goalie in the same way they enhance understanding of the skaters on the ice.
There are two obvious choices for creating a more advanced statistical measure for goalies. The first is to divide the Corsi attempts by the number of attempts a goal was prevented. This would be a similar metric to save percentage as it is currently calculated.
Another possibility is to take Fenwick and divide by the number of attempts where a goal was prevented, exactly the same as with Corsi. As this would create a save percentage based on the shot attempts the goalie in question would have faced during the game, it seems, at first glance, to be a more accurate view of what a goalie does and more in line with how goalies may see the game.
Although a few alternative measures and techniques have emerged (such as Michael Schuckers' Defense Independent Goaltending Rating, War-On-Ice's Adjusted Save Percentage, Hextally Analysis, and Save Percentage Plus), to a large extent most analyses and discussion in the new frontier of advanced stats have had a tendency to stick with plain-old save percentage.
A modification as to how save percentage is calculated may produce a more accurate assessment of the game in regards to how goaltenders view the game. When a goaltender reacts to the play on ice, what they are actually reacting to is a scoring opportunity--a shot attempt.
The argument for using a Corsi-based measure is derived from the concept that goaltenders don't know if one of the skaters is going to block a shot for them. They must react to every scoring chance as viable and give it their full attention.
Alternately, it can be argued that blocked shots should not count towards any analysis of a goaltender's performance as no action the goaltender took was responsible for preventing the opposition from scoring. In this light, Fenwick becomes the ideal measure to evaluate how a goaltender responds to the opposition's attempts to score.
Average percentages by Corsi-Against Groups
Traditional Sv% numbers are typically published by season. As a result, the numbers of Corsis, Fenwicks, and shots goaltenders face varies widely. In order to properly evaluate goalies, the number of trials goaltenders have faced should be included, meaning one needs to consider the number of shots, Fenwick, and Corsis thrown towards each goaltender.
For this analysis, Corsi-Against (CA) categories per season were used, dividing goaltenders into four groups, using 5v5 data for completed NHL seasons starting in 2007. Those four groups divided goaltenders into the following categories: those who faced less-than 500 CA, those who faced between 500 and 1000 CA, those who faced between 1000 and 2000 CA, and those who faced more than 2000 CA. All data was obtained from stats.hockeyanalysis.com
Average CSv% values are highest, followed by FSv% and Sv%. Associated standard deviations (SD) are lowest for CSv% and highest for Sv%. As the number of events increases, generally speaking, the associated save-percentage increases while the SD decreases, regardless of save-percentage type. One notable exception is for Sv%, for which percentage is more irregular.
Generally speaking, there are a lot of positives to the use of FSv%. It has several benefits. Firstly, it appears to provide results that are more consistent across Corsi Against groups than Sv%. The lower standard deviation also allows for a more confident estimate of a goaltender's possible future performance. Irregularities in deviation from the standard make it harder to estimate what may happen in the future. In short, if the purpose of using statistical measures is to use past performance to gauge the most likely future possibilities, it makes sense to use the measure which conforms most closely with predictive calculations.
Furthermore, as mentioned above, goaltenders have to react to more than shots on goal (which have a fairly narrow definition), and because of this, a simple save percentage doesn't accurately record a goaltender's perspective on the game in front of him. While save percentage only shows a part of what goaltenders do, it doesn't necessarily do justice to all the goaltender has to do to help his team win. A Fenwick-derived save percentage may not include all aspects of what the goaltender does, but its more inclusive parameters allow for a better overall picture. The fact that goaltenders often feel the shots on the clock are not an accurate description of the game adds weight to the idea that other measures may be useful for evaluation.
Finally, there are other metrics that are used to evaluate a goaltender's performance. Measures such as games won, individual awards and recognitions, and shutouts are often used during the discussion of a goaltender's quality.
Take, for example, the case of Martin Brodeur, a goaltender generally assumed to be one of the best goaltenders. Brodeur had a fairly unremarkable save percentage throughout his career. If goaltenders are evaluated primarily on their save percentage, which is the first level of analysis for many, Brodeur should not be held in the esteem he is. However, with 125 regular season shutouts, another 24 playoff shutouts, and over 600 wins over the course of his career, Brodeur's quality is often evaluated using different metrics. If Brodeur can be evaluated using the metrics that highlight his strengths, there is no reason why other metrics for evaluation cannot be used to develop a more complete picture of what makes a quality goaltender. There is no reason, especially in this age of statistical dominance, why FSv% can't be one more tool used by those evaluating the performance of a goaltender.
As the number of different measurement tools for the game of hockey grow, more information about what makes a good hockey player surfaces. There's a memoir titled The Crazy Game by Clint Malarchuk, in which he speaks to the different style of play, lifestyle, and hockey that was part of his career in the mid-1980s and the early 1990s. Despite a fantastic number of changes and advancements to the way hockey is played and evaluated, goaltenders are still judged by the same standard. The goalposts for what makes an average or good goaltender simply shift based on what the baseline is at any point in time. As Malarchuk mentions, Grant Fuhr was an excellent goaltender, but during the time he played, it wasn't unusual to see a goaltender allow more goals than they currently do. If he was judged by the standards in place today, he might not fare as well in the assessment.
Deriving a save percentage based off of Fenwick instead of simply shots may seem like moving the goalposts; it's actually more of an effort to obtain a clear view of where the goalposts are. If it's possible to use the statistical measures currently in place to evaluate more clearly what a goaltender faces, it's possible to more concretely understand when a goaltender is good, average or bad. This evaluation is the point of any sort of a statistical measure. It's about being able to apply the most accurate label possible, and FSv% should allow for a more advanced understanding of a goaltender's quality.