Fenwick, the Hockey Stat Behind FF%

Fenwick is Corsi with one edit, and the editor was a Flames fan. In November 2007, Matt Fenwick posted his notes on Calgary's first 20 games at the Battle of Alberta blog, with two versions of the Corsi plus-minus at the bottom, one counting blocked shots and one leaving them out. "As a proxy for scoring chances for & against, I think it works best if blocked shots are excluded," he wrote. The version without blocks took his name.

His case came from the defending team's side of the puck. If Jarome Iginla winds up on a partial break and hits the post, that's a scoring chance. If Willie Mitchell gets a stick on the shot and deflects it into the netting, Fenwick argued, it isn't, and a stat that counts both the same fails to credit the defense. He granted that the version with blocks still had a use, as a proxy for zone time, or in his words, "Where The Puck Is."

The NHL adopted Fenwick's stat in February 2015 under the name USAT, for unblocked shot attempts, and its glossary credits USAT with taking shot-blocking performance into account where SAT doesn't. The league's numbers go back to 2009-10. Natural Stat Trick, MoneyPuck, Evolving-Hockey and Hockey-Reference all publish it under Fenwick's name, and it lives on inside expected goals, which prices the same unblocked attempts one by one.

Take Corsi and subtract the blocked attempts. The for and against columns, the five-on-five default and the share all work the way they do for Corsi, which is why they're almost always within a point of each other (31 of the 32 teams did last season). These are the inputs:

• Shots on goal, goals included
• Missed shots, including posts and crossbars
• Five-on-five ice time, for per-60 rates
• For a player, which attempts happened while he was on the ice

Here's a worked example, with invented round numbers. The Wings put 20 shots on goal, miss the net 10 times and get 15 attempts blocked, so their Corsi for is 45 and their Fenwick for is 30. The other team takes 40 attempts and Detroit blocks 10 of them, which leaves a Fenwick against of 30. On Corsi, Detroit won the night 45 to 40. On Fenwick it's dead even, because the Wings had more of their own shots blocked than they blocked.

The blocked column cuts both ways. A team that blocks a lot of shots shrinks its Fenwick against, and a team that gets a lot of its own attempts blocked shrinks its Fenwick for. Corsi counts every one of those attempts and doesn't care who did the blocking.

Fenwick measures the same thing Corsi does, territory, with a filter for the attempts that never reached the goalie. At the team level they're close to interchangeable. Across the 32 teams last season, five-on-five FF% and CF% correlated at 0.98, and 72% of the league's five-on-five attempts went unblocked.

Carolina led the league at 58.9% last season, Colorado was second at 55.9% and Toronto finished last at 45.1%. Detroit was 16th at 49.7%, five spots higher than its Corsi rank, a gap that's almost entirely about blocked shots (more on that below). Players spread about as wide as they do on Corsi. Among the 598 skaters who logged at least 500 five-on-five minutes, the median on-ice FF% was 49.6% and the top tenth cleared 55.0%, and a player's FF% and CF% correlated at 0.97.

Is Fenwick the better stat of the two? It's ahead only at the margins. In this site's split-season test (96 team-seasons, first half against second half, over the last three seasons), first-half FF% correlated at 0.44 with second-half goal share, a hair ahead of CF% at 0.42, while Corsi repeated a little better from half to half (0.81 to 0.77). There's one quieter advantage. In six seasons of league data ending in 2012-13, Michael Schuckers and Brian Macdonald found persistent rink effects in 12 arenas for Corsi events and five for Fenwick events, and blocked shots, the event Fenwick drops, showed them in 15.

Seeing what the blocks are doing

Detroit is the clean example. The Wings blocked 1,179 opponent attempts at five-on-five last season, tied with Seattle for the league lead and about 108 more than the median team, while getting 1,029 of their own attempts blocked, right at the median. That's why their FF% sat 0.88 points above their CF%, the second-largest such gap in the league behind Philadelphia's. Is blocking that many shots a skill or a sign that you're stuck in your own end? Fenwick can't settle that. It can show that Detroit's opponents out-attempted the Wings mostly with shots that never got through.

Building expected goals

MoneyPuck's and Evolving-Hockey's expected-goals models both start from Fenwick attempts. They price only unblocked attempts, since the NHL's feed logs a blocked shot at the spot of the block and keeps no honest record of where it was taken. Add those prices up and you get xG, so a team's xGF% is a Fenwick share with each attempt weighted by its danger. The expected goals page covers the model.

Reading close-game numbers

The NHL's glossary defines a close game as within one goal in the first two periods or tied in the third or overtime, and says limiting SAT and USAT to those minutes filters out the score effects that skew shot data, at some cost in sample size. The league still publishes the split. Detroit's five-on-five USAT% was 49.7% overall last season and 49.9% in close games, so the score didn't move the Wings much on this measure. Score-adjusted versions keep every minute and reweight it instead.

It throws away real information

A blocked attempt still tells you which team had the puck in the attacking zone, the point Fenwick himself conceded when he called the version with blocks a proxy for zone time. Over the last three seasons, Corsi repeated a little better from half to half.

Every unblocked attempt counts the same

A wrister from the point that sails wide counts as much as a rebound at the side of the net. That's the job expected goals does with the same set of shots.

Misses still depend on the scorer

Dropping blocks removes the least consistently recorded of the shot events, but missed shots stay in, and Schuckers and Macdonald found persistent rink effects for misses in nine arenas.

A player's number is mostly his team's

How much? Among the 598 skaters with 500 five-on-five minutes last season, on-ice FF% correlated at 0.73 with the team's FF%. Relative versions, on-ice minus off-ice, take most of the team's influence out.