Some old-school headshots for ya.
This time of year, we tend to get one question more than any other: what do you think the lines will be? I haven’t the foggiest idea really, and I was awful at guessing how Oates would marshal his forwards last season. (The whole thing about mandating lefties play on the right wing still surprises me.) It seems everyone has his or her own guidelines for formulating lines, and ultimately they’re all subjective.
But if we did start with data, how would that work? What kind of line combinations could the Capitals have if they were purely evidence-based and is that ever a good idea?
I grabbed three years of data for all 2013-14 Capitals forwards and began sorting the players by each statistic. Some of these are going to be super insightful; others are going to be arbitrary and weird. In the end, I think we’ll have a better understanding of what objectively drives these coaching decisions– before qualitative judgments get involved.
The order of these line combinations go from most plausible at top to kinda crazy at bottom. All data are from even-strength performance during the last three seasons using Hockey Reference and Hockey Analysis as sources.
I started by assigning each forward a position. You may not agree with me putting Brooks Laich at left wing, and that’s okay.
I predicted this sort would be the most accurate because coaches pretty much always deploy their top lines more than their bottom lines, and at a glance that seems correct.
Martin Erat has jumped from the second line, his expected spot, up to the first. Erat, who spent most of the last three years in Nashville, is a reasonable substitute for Marcus Johansson, who surely has the most tenuous grip on his presumed top-line spot. Considerations of chemistry or defensive assignments might lead to overruling this sort and swapping Mojo for Erat, but overall I consider this a pretty strong lineup, and I wouldn’t feel bad at all about Washington taking their pregame rushes in this order.
I also performed similar sorts using assists and goals, but the results were all over the place (you can recreate them using my data if you want). Points seems to be superior as it doesn’t punish playmakers for not finishing (Joel Ward) or finishers for not getting assists (Jason Chimera).
We see Marcus Johansson regain his spot on the top line, but now as his pivot we’ve got Mathieu Perreault instead of Nick Backstrom. Perreault has had some prolific seasons despite limited ice time (hat trick vs Boston, anyone?), though there is some evidence that his success is driven more by fortuitous percentages than by driving play. That said, I was pretty keen on Perreault getting a shot at 2C earlier this summer, and I wouldn’t be too upset if he cracked the top six or even the top line at some point this season.
We should also note that Mikhail Grabovski, much praised by me, falls out of the top six under this system. I interpret that as a warning that the point-based logic– and all of them really– are imperfect. They all fail to take in account the varied contexts of the players (Grab0vski’s defensive deployment in Toronto last season for example), and as a result we should use them only as a way to kick off the conversation.
Here I’m using the percentage of all (Correction: unblocked) shot attempts that go towards the opponent’s net while the player is on the ice (the cool kids are calling it Fenwick). The easiest way to think about this is how the ice “tilts” while the player is on the ice.
Brooks Laich, who is not a strong possession player, wins the first-line left wing spot, but the competition was not tough. All Capitals LWs put up percentages under 50% (i.e. most pucks went the wrong way), though we should qualify some of these players– Erat, for example– took lots of defensive assignments. Tyler Dellow wrote a strong piece back in March about how and why players on the bottom lines tend to have negative possession scores. Doesn’t mean they’re bad players, just differently used.
The right wing is a different story. It’s an embarrassment of riches over there; every player is at or above 50%. The Capitals are stacked at right wing, though at this point it behooves me to point out that Ovechkin and Brouwer have seen their possession numbers degrade during the three-year window I’m using.
This table uses quality of competition– a statistic calculated by Hockey Analysis based on the average possession of the player’s on-ice opponents. My expectation was that this stat would align somewhat with the time on ice stat at top, but that does not seem to be accurate. Lately, QoC seems to have fallen out of favor as we acknowledge how savvy coaches put their best defensive players up (with not-awesome possession numbers) against the opponent’s best players.
Because of line matching, a team’s best players often face the toughest competition, yet this stat gives us a top line filled with players who are usually used defensively: Erat, Grabo, Ward. That’d be a terrific third line– perhaps deployed against the opponent’s best players, but it’s not the explosive scoring we expect in DC. So no, possession-based quality of competition is not much of use for choosing lines.
This is silly. PDO measures how far above or below average players veer regarding their shooting percentages — both their team’s and the opponent’s. Players above average are generally considered lucky, and players below average are considered unlucky. PDO is rarely used as a measure of skill (it was originally concocted to dismiss plus-minus’s fidelity as a measure of skill), but let’s see what happens anyway.
Welp, Chimera is out. Poor Jason Chimera and his lowly shooting percentages. Perreault, on the other hand, is sitting pretty with a curiously high PDO (1023) considering this is a three-year average. Shouldn’t we expect that number to trend towards 1000 in the future?
Just like QoC, there’s not much utility in using kinda-sorta blind luck to sort players into lines. That said, these lines don’t look so awful. Troy Brouwer and Nick Backstrom might disagree though.
Shooting percentage is an individual stat (unlike PDO, which includes all other players on the ice).
Once again, Matty Perreault has bobbed up to the 1C spot. I’m starting to get suspicious about that guy. Jason Chimera’s shooting is moribund, but high enough to bump Aaron Volpatti out of the lineup again. Troy Brouwer, whose high overall shooting percentage is buoyed by the power play, doesn’t fare so well at even strength and vacates the top six so Joel Ward can get a shot.
If a top line of Mojo, Perreault, and Ovechkin gives you pause, you’re not alone. This is good case study for why we don’t let the stats decide everything.
Even the alphabet conspires against Aaron Volpatti.
Here’s how I’d like to see things on October 1, though I won’t be heartbroken if it doesn’t happen.
I’m still wary of Marcus Johansson’s skills when he’s isolated from rock stars like Nick Backstrom, but the top-line Mojo experiment shall continue.
I’m also an avowed Eric Fehr fan, and I’d like to see him get some more time in the offensive zone. I think my second line would be a puck-domination machine.
That said, I’m not super sanguine about using net-crashy Troy Brouwer on the defensively-deployed third line or using Jason Chimera’s granite gloves on a purely checking line.
In conclusion, using previous time on ice is the best way I can think of to predict how Adam Oates will do things this season, but every method I used has a quirk or two. They’re all imperfect. It’s not easy to quantify chemistry between players (e.g. sometimes two players just want to play together despite no measurable synergy), and that’s why we have coaches.
Imagine a team coached by a robot that used even-strength shot-attempt differential to make his lines. That would be a lame team. Coach-a-tron 5000 would get canned before Thanksgiving.
What lines do you wanna see on October 1st?
Russian Machine Never Breaks is not associated with the Washington Capitals; Monumental Sports, the NHL, or its properties. Not even a little bit.
All original content on russianmachineneverbreaks.com is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported (CC BY-NC-SA 3.0)– unless otherwise stated or superseded by another license. You are free to share, copy, and remix this content so long as it is attributed, done for noncommercial purposes, and done so under a license similar to this one.