The Wonderful World of Advanced Stats and Fantasy Hockey by @PITmounD - PensInitiative | Pittsburgh Penguins Blog | Rumors | News

The Latest

Post Top Ad

Tuesday, August 13, 2013

The Wonderful World of Advanced Stats and Fantasy Hockey by @PITmounD

by Dylan O'Brien (@PITmounD)

Advanced statistics in the NHL may seem like a daunting concept, but it's not.

This article is based on a Grantland piece entitled, "The Faker's Guide to Advanced Stats in the NHL," which I recommend everyone read as a companion to this one. Some ideas are adopted from the Grantland article, but what follows is intended to be an introduction to advanced statistics, and more specifically how to use them in the realm of fantasy hockey. If you can't spare a few bucks to buy Rob Vollman's "Hockey Abstract," hopefully you'll learn a few things here.

When you want to analyze, compare, and project player performance in the NHL, you can't just watch a ton of hockey games and use "the eye test" to accurately depict a player's on-ice contributions. This is why we have statistics. Most people are familiar with the traditionally tracked player statistics in the NHL: Goals, Assists, +/-, Takeaways, etc., but a lot of these traditional statistics aren't the most accurate predictors of a player's contributions to his team.

Take plus-minus for example. This particular "goal-based" statistic is one of the most common statistics (in fantasy hockey and in real life) used to support a player's defensive capabilities, and sometimes rightfully so; however, plus-minus is dependent on so many other factors -- linemates, defensive pairings, defensive scheme, quality of goaltending, etc. -- and when used as the be all end all of a player's defensive acumen, problems can emerge. An extreme example is Brian Leetch's 1997-98 campaign with the New York Rangers in which he was both contending for a Norris Trophy and a minus-36. Clearly, Brian Leetch wasn't an awful defensive player during that season -- or the four seasons prior in which he also posted  minus-ratings -- yet his plus-minus ratings were very poor on an underachieving Rangers club.

A lot of these statistics -- especially plus-minus -- tend to fluctuate wildly on a year-to-year basis, and are therefore not the most accurate method of predicting a player's future performance. As a general rule, using goal-based statistics -- like Goals/Assists, +/-, etc. -- is not the best way to predict future success. After all, goals are relatively infrequent events in the game of hockey, and basing all of our conclusions on infrequent events can be problematic. Simply put: not all statistics are created equal.

So where does this leave us? If we can't strictly use goals as our statistical basis for evaluating players, what do we use? The answer is puck possession. Since there isn't really a handy stat available that measures a player's puck possession, we use shots as a reflection of possession ... which brings us to the "advanced statistics."

On-Ice Shot Statistics:

Fenwick and Corsi are probably terms you've heard some nerd ranting about on Twitter when discussing something like "why Pascal Dupuis is so underrated," and you may not have known what they meant, or how they're applied, or even where to begin to try and understand them. But I'm here to tell you that not only are these "advanced" statistics anything but advanced, they are very useful in analyzing a team or player's future success and contributions, respectively.

Fenwick and Corsi are essentially shot differentials, and can be expressed as both a plus-minus or a percentage. There are a variety of nuances with each of these statistics, but the general idea of these stats is to measure a player's contribution to puck possession. A player with a high Fenwick/Corsi rating theoretically generates a lot more shots (possession) toward his opponents net -- all shot attempts are counted, including shots that miss the net, but Fenwick does not count blocked shots -- than are generated toward his own net. And since shot quantity, and not shot quality, is the best indicator of goalscoring, these players are more likely to be successful, i.e. produce goals at a higher rate than their opponent.

Fenwick and Corsi are wonderful statistics for predicting future performance, and in the world of fantasy hockey they can be utilized to mine those "diamonds in the rough" that simply haven't received the prime offensive opportunities, but will most likely produce when given the top-6 treatment and PP time.

There are so many different variations of the Fenwick/Corsi statistics (Relative Corsi, Corsi Quality of Competition, etc.), but these concepts can be learned with a bit more individual research on the subject. Voila!

Resources and Databases for Fenwick/Corsi statistics: (database) (Hockey Prospectus Glossary $) (shot quality v. quantity) (Why Possession matters)


In addition to Corsi and Fenwick, there are tons of useful statistics that fall under the "advanced stats" umbrella, notably PDO -- which is essentially a measurement of a player's luck. While most in the statistics community would prefer the term "random chance," this statistic essentially measures whether a player or team had good or bad luck.

PDO can also be expressed as an individual or team statistic and is a team's on-ice shooting% (while a player is on the ice) + a team's on-ice save%. Anything above 1.000 is on the high-end, and generally a team/individual regression is expected. On the other hand, anything below 1.000 and we can assume that the team/individual was a victim of bad "puck luck" and is due for a bounceback season.

A great breakdown of PDO from Steve Burtsch of Pension Plan Puppets:

For fantasy hockey purposes, PDO is an incredibly valuable tool when looking for "sleeper" or "bust" candidates for the upcoming season. Let's take a look at an example from the Penguins roster: Chris Kunitz. In the 2012-13 season Kunitz posted career numbers, scoring over a point-per-game and shooting at a 19.5% clip. His PDO -- which is the sum of On-ice shooting% and On-ice save% while that player was on the ice -- was 1074, well above the statistical mean. What we should take from this is that not only was Kunitz's individual shooting% well above his career average (12.8%), but his PDO was highly unsustainable. This is not to say that Kunitz won't have a good season playing top-line minutes with Sid, just that we should expect a regression from Kunitz's 35+ goal / 80-point pace of 2012-13.


At this point you're probably asking yourself, "where is the goaltender's place in the world of advanced statistics?" Well, when it comes to goaltenders there is really nothing major that you probably haven't  been exposed to in one way or another. We all know about the traditional goalie statistics: wins, goals against average, save%, and shutouts. I'm here to tell you that only one of those statistics (more like 1/2 of a statistic) is really statistically relevant when evaluating goaltenders.
Various statistical analyses have shown that statistics like Wins, GAA, and even Shutouts are more attributable to the overall team (and random chance) than to the goaltender himself, so we are basically left with Save% -- more specifically Even-strength Save%. Generally speaking, since a goaltender can't control special-teams situations and Sv% tends to fluctuate wildly during 5v4 and 5v3 opportunities, 5v5 Sv% is the best indicator and predictor of a goalie's performance. Again, generally speaking, the most successful goaltenders will be those with a high 5v5 Sv% who play on teams that aren't often shorthanded and conceding goals with a man in the box. Michael Clifford has an excellent article, "Fantasy Hockey: Evaluating and Projecting Goaltenders," which goes into much more detail on this (somewhat new) argument.

If you only know one thing about evaluating goaltenders, make sure that you pay attention to 5v5 Sv% and remember that evaluating the team is nearly as important as evaluating the man between the pipes.

Resources on 5v5 Sv% / Shot quality: (database) (shot quality) (Reimer / Bernier)

As always, a little bit of individual research into the topic is never a bad thing if you want to understand these statistics and how they can give you an edge against your fantasy hockey competition.

In closing, I wanted to bring up one of the points from the Grantland piece which I find is very important, and that is context. All statistics should be viewed in context, not as the answer to the question, but more like a piece of the larger puzzle that is our understanding of a player or team. These stats -- Corsi, Fenwick, PDO, 5v5 Sv% -- don't paint the entire picture of a player, and are not intended to do so. There are many mitigating factors that can explain a poor Corsi/Fenwick rating -- linemates / quality of teammates / quality of competition / zone-starts, etc. -- and as always, these statistics should be viewed alongside a variety of other statistics.

Nobody wants to get suckered into placing too much importance on a particular statistic, or worse on a small sample size, so these are pitfalls you want to avoid when using these types of statistics. Similarly, when discussing sample size the general rule is: the more statistical evidence available, the more accurate the results. Falling into the trap of evaluating players based on a relatively small amount of sample data can be very troublesome and lead to inaccurate predictions.

Just as you'd want a variety of perspectives on, say, which dog is the best breed to have as a pet, you'll want to keep in mind a variety of different statistics when evaluating players and teams in the NHL.

Check back every couple weeks for my latest piece on fantasy hockey developments around the NHL with an advanced stats flavor!

No comments:

Post a Comment

Post Top Ad