Friday, November 1, 2013

Inside Game: The Effects of Net Drive 5v5

Gallagher, Eller and Galchenyuk lead the Habs in Net Drive 5v5
Like moths to a flame, stats-folk are drawn to numbers and even when there are seemingly obvious reasons to be cautious, the temptation can often be too difficult to resist. So when it comes to drawing nectar from the NHL’s play-by-play game data, just how close can you get to shot distance and location without getting burned?

Until the NHL takes itself more seriously and stops relying on unverified play-by-play data, collected by part-time NHL employees with day jobs who are making what you probably think they’re making for a few hours’ work, a few times a week; trusting shot data carries a greater risk of combustion than sweet reward.

Simply put, the Elias Sports Bureau can process and track all the data they want for the NHL but when the raw data collected is based on live, screen-touching guesstimates of part-time individuals tracking a game in near-constant motion with no built-in redundancy or review process, delving too deep into shot data isn’t so much advanced as it is folly.

Trust Issues

So, how far do I trust play-by-play shot data? Just enough to paint in broad strokes to highlight patterns within teams and the league. Unfortunately, with passes and pass location completely untracked, there just isn’t data to break down what exactly happens in the final moments before a goal.

Short of analysing video of what preceded each and every shot attempt, all we have to go on is what type of shot was supposedly taken and if it went in, caused a rebound, missed the net, was blocked, tipped or deflected, etc., and more importantly of late; the on-ice coordinates of where the shot attempt was supposedly taken from, excluding blocked shots.

The thing is, trust in play-by-play data is entirely dependent on the NHL’s give-a-damn about verifying accuracy and once you get past goals, assists and TOI, their give-a-damn quotient towards reliable stats is easily the lowest among the four major North American sports. That lack of caring by the NHL is also the achilles heel of hockey analytics; stat-folk caring a heck of a lot more about the numbers than the NHL's part-time employees entering them no validation or oversight.

Get hit location wrong and nobody really cares so long as the count is somewhat accurate and before shot coordinates became publicly available, nobody really cared where they came from so long as the right players were credited with the goals and assists. pulled on-ice coordinates from their Ice Tracker prior to last season but other sites still have access to them, offering them up for analysis and with that, the topic of “shot quality” swung into high gear.

The problem I had studying the Habs shot quality for the past three seasons was finding a sweet spot where the shot data could be trusted enough to actually be useful. I wanted to know how much of an impact net drive was having on goal production, or in the case of the Montreal Canadiens over the past 15 years, their lack of it.

During that time, I compared 5v5 goal distance and location from video highlights spanning 500+ games across the NHL to the stated goal distance recorded by the NHL’s part-timers and location data from sites that kept offering it. The NHL’s play-by-play distance data was consistently wrong by an average of 20% and with that, so was the location data, obviously. My trust and interest in on-ice coordinates ended there but play-by-play shot distance appeared to still have value.

The Sweet Spot

What the video highlights showed was that more than 70% of 5v5 goals are scored within 30 feet of the net, with the greatest concentration towards the middle of the ice, from the goal crease out to just beyond the hash marks and fading on either side towards the faceoff dots. The percentage might be higher than some may have expected but where most of the goals are scored from shouldn’t be.

From there, I analysed play-by-play shot data from every NHL game last season and the average 5v5 goal distance registered within 25 feet of the net, excluding empty net goals was a comfortable 67%. Knowing that recorded shot distance is typically off 20%, 25 feet became my trust sweet spot between what the NHL part-timers are recording and what’s actually happening on the ice. For this story on the effects of net drive 5v5, settling on 25 feet wasn’t an arbitrary decision; it’s as close as I dare get to the flame. As for location, I’ll let others risk getting burned on shot coordinates or distance averages.

Now, before I go into more detail, here’s a small sample of the NHL's shot location accuracy stat-folk must deal with; their most blatant play-by-play mistakes involving the Habs last season. Mistakes that were never corrected:

March 05, 2013 - Montreal Canadiens @ NY Islanders

1 EV 5:21 GOAL MTL #14 PLEKANEC(10), Snap, Neu. Zone, 107 ft.
Plekanec actually scored from about eight feet away, not 107.

2 PP 12:10 GOAL NYI #17 MARTIN(2), Snap, Def. Zone, 153 ft.
Good example of an NHL part-time employee marking the shot at the wrong end of the rink. Martin actually scored from about 21 feet away, not 153.

2 PP 1:23 GOAL NYI #26 MOULSON(11), Snap, Neu. Zone, 111 ft.
Moulson actually scored from about 20 feet away, not 111.

March 19, 2013 - Buffalo Sabres @ Montreal Canadiens

1 EV 18:59 GOAL BUF #63 ENNIS(8), Wrist, Def. Zone, 140 ft.
Ennis actually scored off a tip from about 13 feet away, not a wrist shot from 140.

That’s four goals with a combined distance of about 62 feet (Avg 15.5) registered as a combined distance of 511 feet (Avg 127.8). Think a scoring change would trigger a play-by-play data correction? Nope. Let’s look at a recent Habs game:

October 29, 2013 – Dallas Stars @ Montreal Canadiens

1 EV 12:02 GOAL MTL #49 BOURNIVAL(3), Slap, Off. Zone, 63 ft.
The goal was initially credited to Diaz but changed shortly after to Bournival, who had the puck deflect in off his body from about six feet away. Beyond the goal and assist credits to Diaz and Plekanec, nothing else was changed in the play-by-play data.

Net Drive Effects 5v5

Prior to Thursday night’s games, there has been a total of 12,015 goals, shots and misses recorded 5v5 and 33.9% of them came within 25 feet of the net. That translates to a game average of 10.9 shot attempts in-close per team. Unfortunately, blocked shots don’t have a recorded distance so I have to exclude them from this story. What’s especially interesting is that 33.9% has accounted for 68.7% of all 5v5 goals scored, within 1% of last year’s average. To put that into context, you’d have to shoot the puck 4.3 times as much beyond 25 feet to score as much within 25 feet 5v5.

The league average save percentage this season on all shots beyond 25 feet is .963, which is within .010 of last year’s average. Within 25 feet of the net, the average SA% drops to just .852, once again within .010 of last season. It certainly doesn’t take long for the league averages at distance to settle within a tight band of fluctuation.

Looking for evidence that the closer you get to the net, greater the odds of scoring, no matter what kind of shot you take? The league as a whole is currently shooting 14.8% with 25 feet 5v5. That's a 30+ goal scorer for anyone playing top 6 minutes that's willing to consistently drive the net. In addition, 18.7% of all 5v5 goals scored came off rebounds, where an initial save was made within 3 seconds. As for shot selection, 54% of 5v5 goals within 25 feet were scored with a wrist shot.

So what's the lesson for teams looking to improve 5v5? Generate more shot attempts in-close while limiting the opposition’s chances. Driving the net for a shot, pass, rebound, tip, deflection or screen clearly has its rewards, as does a quick release. And given the average league SH%, missing the net in-close should carry some form of team punishment program to drive home the fact that pucks on net do matter.

As for the Canadiens last season, their 5v5 in-close chances represented 33.5% of their total, slightly below the league average but much better than previous years. With a Fenwick of 54.3% within 25 feet, combined with a 17.1 shooting percentage vs. 15.8% against, the Habs were +13 on 5v5 goals scored and they won more games because of it. For teams like the Leafs who were getting outshot in-close last season by more than 2 shots a game, the only salvation for being out chanced in-close 5v5 is rebound control and excellent goaltending. I'm just not sure how long you want to hang onto that carton of milk before things go sour. Habs fans know it shouldn't be a long-term lifestyle choice on how to play the game.

Early Trends for the Habs

The 2013-14 season is just 13 games young so let’s look at how the Habs are doing so far. Their 5v5 in-close chances currently represent 39.6% of their total, well above the 33.9% league average while their Fenwick within 25 feet sits at 51.1%. Overall, the Habs are off to a decent start, generating 12.6 shot attempts in-close per game but they're also missing the net a full shot a game over the league average and that's costing them goals.

One area of concern for Michel Therrien is the Habs shooting just 11.2% in-close 5v5. Fortunately, Price and Budaj are limiting shooters to just 9.0%, thanks in large part to fewer rebounds. As a net result, the Habs are +3 on 5v5 goals scored in-close. If it wasn't for quality goaltending to start the year, their lack of in-close efficiently could've been a big problem. Bullet dodged... for now.

Carey Price is certainly playing much better than last year 5v5. Opponents are shooting just 9.8% in-close on him so far with the league average currently at 14.8%. Last year, Price was beaten in-close 5v5 14.5% of the time. Some regression to the mean should be expected in Price’s in-close numbers but hopefully they’ll be timed with the Habs improving their own SH% while he settles into an even strength SA% range that will no longer be described as “average”.

If the Habs are to go anywhere this season, that's a term that needs to be erased from memory and only Price can do it. While the regular season would be a good start, the jury is out until he performs in the playoffs and that's where I'm hoping Stephane Waite can help the most.

League Highs and Lows

Meanwhile in San Jose, folks wondering why the Sharks are off to a great start can look to their inside game. They're averaging 13.1 5v5 shot attempts per game within 25 feet and shooting 16.9%. At the other end of the scale we have the New York Rangers, averaging 11.2 shot attempts and shooting an embarrassing 5.4%. In Philadelphia, the Flyers are averaging just 10.0 chances and shooting only 8.7%. How about the defending Stanley Cup Champions? The Blackhawks are cruising along just fine, averaging 12.8 in-close chances per game and shooting an impressive 18.2%.

And then there's the Toronto Maple Leafs, the bane of stat geeks everywhere. You know the hockey gods have a sense of humour when they average a miniscule 8.6 5v5 hot attempts in-close and shoot a staggering 21.9%. That said, there's something to be said about pucks on net because the Leafs in-close are only missing the net 1.8 times a night, almost a full shot less than the league average and missing half as much as the Habs do.

Given the league average SH% is currently 14.8% within 25 feet 5v5, there's strong evidence the Leafs sky-high SH% is partly related to the team being among the league's best at getting pucks on net in-close for rebounds.

Next up, a look at the effects of net drive on special teams.