Filed under:

# Silver Nuggets: Advanced Statistics Explainers

A lot of Sens links, plus an explanation of some of the most common concepts in advanced stats.

Over the summer, we asked for some user input on things to introduce to the site. The response was overwhelming, and people seemed to be in favour of adding quite a few new features. One of the most popular was an advanced stats explanation series of articles. Now I don't consider myself an expert in these at all, but I know a bit, and I thought it might be good to talk about a few concepts every so often here in the Nuggets.

This time, I'll be looking at the most important ones to know: Corsi, Fenwick, and PDO, as well as the idea of score effects. If you're already familiar with these, feel free to skip. Or to read them and correct any mistakes I make.

Corsi/SAT:

Definition: Shot attempt differential, where shot attempts include shots on goal, blocked shots, and missed shots. Often expressed as a percentage (e.g. a team had 52% of the Corsi events during a game). Can be used to describe both teams and players.

History: This, like many so-called "advanced stats", was first named and tabulated by Vic Ferrari, creator of the Irreverent Oiler Fans website. He got the idea hearing then-Sabres GM Darcy Regier talking about the idea of tabulating all shots, not just those that hit the net. Not wanting to name the stat "Regier", he searched the Sabres' website for a cooler name, and settled on that of goalie coach Jim Corsi.

Why we care: The idea behind Corsi is that teams take far more shot attempts per year than just shots on goal, and larger sample sizes give better statistical results. We can prove that Corsi is far more repeatable from year to year than wins and losses for a team, and that Corsi correlates with points earned. Not every team with a good Corsi will do well, but it's a much better predictor than +/- or even shots on goal.

Fenwick/USAT:

Definition: Unblocked shot attempt differential, meaning only shots on goal and missed shots count. Can be used the same ways as Corsi.

History: This one was also named by Vic Ferrari, after a commenter on his site, Matt Fenwick, who proposed modifying Corsi to not include blocked shots.

Why we care: Fenwick's argument was that blocked shots aren't truly scoring chances, especially if they're blocks on things like points shots. He reasoned we'd get better results if they were removed from the equation. And to date, even-strength score-adjusted Fenwick (EVSAF) is the most repeatable stat we have for NHL teams. In other words, a team with a great EVSAF will probably be great next year, whereas a team with a great EV shooting percentage could be great or terrible the next year in the same category.

PDO/SHSV%:

Definition: The sum of a team's shooting percentage and save percentage (normally at even strength or 5v5). If calculated for a player, it's their team's shooting percentage and save percentage while they are on the ice.

History: Once again, this one came from a commenter on Vic Ferrari's site, whose username was PDO. The theory was that if every team were equal, every team would have a PDO of 100. A PDO higher than 100 suggests luck (either lots of shots go in while you're on the ice or your goalie is bailing you out or both), while lower than 100 suggests bad luck.

Why we care: #FancyStats pioneer Gabe Desjardins (of Behind the Net fame) described PDO as the most important stat to understand. The reason? Ferrari noted on his site that players with low PDOs often were afterthoughts in free agency, while players with high PDOs got long-term extensions they rarely lived up to. Since on-ice shooting and save percentages are mostly luck-driven for skaters, PDO is as close as we have to a proxy for luck.

Only extraordinary players can sustain a PDO above 100. For example, Sidney Crosby has a career 5v5 PDO of 102.0. Forwards have almost no impact on goalie save percentage, so this is likely due to Crosby being an excellent shooter and playmaker. This is useful to see in the context of this season, where Crosby is rocking a PDO of 99.7. It seems to be a lot of bad luck that's keeping him from being a dominant force this season. Expect him to rebound any game.

As a comparison, last season Mark Borowiecki had a PDO of 103.82 (at 5v5). We know Borocop isn't a better playmaker than Crosby, so he was probably being bailed out by his goalies. We can expect him to have a worse season than last year. On the other end of the spectrum, Colin Greening had the team's worst PDO at 94.38. It looks like horrendous luck got him stuck in the AHL. But in the 2012-13 playoffs before his criticized extension? A PDO of 101.81, meaning luck probably played a role in getting it.

Definition: A statistic, typically Corsi or Fenwick, is adjusted compared to the league average in the same score situation. Score situations are typically winning by 2+ goals, winning by 1, tied, losing by 1, and losing by 2+. A team's behaviour over those categories is compared to the league average, then re-weighted so that 50% is still the league average.

For example, if a team has a better Corsi than the average team when up by 2 goals by 1%, this is treated the same as having 51% of the shot attempts when the score is tied.

History: It's long been noted that as teams fall behind, they take more risks and generate more shots, while teams with the lead tend to get more defensive. This gets even more pronounced later in games. At first, people accounted for this by using something called "Close", which referred to the score being within one goal in the first two periods, or tied in the third. The problem with this is that often games aren't close, and so nearly 60% of available data was thrown out for not being close. Adjusting for score lets samples sizes be much bigger.

Why we care: As noted above, score-adjusted Fenwick is the most repeatable stat for teams that's been found. So clearly score-adjusting has its merits. But more than that, people always point out how important context is, and this helps to better understand context.

For all these stats, the results are only demonstrative. They don't tell you how to fix the problem, but they can point out problems and successes. So if a team is far better than the average when up by two goals, it would be important to note why. Do their D still pinch in the offensive zone? Do they plug up the neutral zone to make clean zone entries for the other team difficult? Are there specific players deployed? Looking at stats in the context of score effects opens up ideas that could otherwise be missed.