I’ve had an idea rolling around in my head for a few years now that I’ve kind of alluded to here from time to time but never taken the time to flesh out. It’s to do with assists. This kind of ties into my post in which I looked at how many of Erik Karlsson’s assists were assists on which he wasn’t trying to make any sort of an offensive play last year – in some ways this is kind of the second part of that, which I was bugged for by various people at various points in time.
I have a problem with assists. The problem is basically this: two assists get handed out for every goal that’s scored, regardless of whether the player getting the assist made any sort of an offensive play. If you watched last week’s Oiler game in Dallas, Justin Schultz got an assist on Ales Hemsky’s PP goal after he sort of rolled the back around the boards with his back to the play, Jordie Benn tried to clear it, it bounced off Ryan Smyth’s foot into the slot, Smyth missed with a swing at it and then Jordan Eberle grabbed control and flipped it to Hemsky for the goal. Assists: Jordan Eberle and Justin Schultz. (If you read back, you’ll see I mentioned Schultz about 100 words ago as having touched the puck.)
That was a PP goal but the same sort of thing holds true of ES goals, I think. There’s like a base level of offensive credit you will receive as a defenceman even if you do nothing more than roll the puck into the corners, simply by dint of the fact that hockey awards an assist if you were one of the last two guys to make a play on the puck.
This makes looking at assists as a measure of playmaking ability somewhat problematic. Now add this: Imagine that Jeff Petry played on an NHL team made up entirely of 19 copies of me. Now imagine that he played on an NHL team made up of 19 Wayne Gretzkys. Exact same Jeff Petry, but he’s going to have vastly different assist numbers, depending on which team he’s on. If you were to look at the assist numbers of Jeff Petry on Team MC79 and the assist numbers of Jeff Petry on Team Gretzky as a gauge of his playmaking effectiveness, you’d come to wildly different conclusions about his effectiveness as a playmaker.
Fortunately for Jeff Petry, he faces no risk of ending up on Team MC79 at the NHL level (sadly, no hope of Team Gretzky either). The range of talent from best team in the league to worst team in the league is much, much smaller. There’s still a problem though, with some guys playing in more goal rich environments than others – those guys are necessarily going to pile up more of the “had a touch on the puck and then some actual offensive players did something” assists.
Let’s take Caps’ defenceman Jeff Schultz and Devils’ defenceman Andy Greene to illustrate my point. Between 2007 and 2012, Schultz played 5128.97 at 5v5. He had 49 assists in that time, a rate of 0.57 5v5A/60. Greene played 5434.62 at ES and piled up 54 assists, a rate of 0.59 5v5A/60. They sound basically identical, right? Slightly more TOI for Greene, slightly more assists and an assist rate that’s a hair higher.
Except the Capitals scored scored 251 goals with Jeff Schultz on the ice and the Devils scored 181 with Andy Greene. The gap would be even bigger if Schultz had played as much as Greene. The conclusion is obvious: Jeff Schultz is an offensive wizard, whose contributions aren’t being properly captured by assists.
OR…I suppose, maybe it’s partly “The Caps were more fortunate with Schultz on the ice than the Devils were with Andy Greene” and “The Capitals were a better offensive team than the Devils during the period in question.” To the former point, the Caps shot 9.9% at 5v5 with Schultz on the ice to the Devils 7.1% with Greene. That is a massive difference (and unlikely to be one which either guy had anything to do with). To the latter, the Devils scored 2.27 5v5 G/60 when Greene wasn’t on the ice; the Caps scored 2.65 5v5 G/60 without Schultz. It seems reasonable to conclude that there’s an element of “Washington is better at scoring goals” going on here – the Team MC79 versus Team Gretzky problem that I referenced above, if on a somewhat lesser scale.
That information changes things a little bit, no? All of a sudden Greene’s assist rate looks a lot more impressive. It kind of suggests that there might be a better denominator to use when dealing with defencemen assists or, if not in place of A/60, at least in addition to it: A/GFON. Andy Greene got assists on 29.8% of the goals that New Jersey scored with him on the ice. Schultz got assists on 19.4% of the goals that Washington scored with him on the ice.
There are 174 defencemen who’ve played at least 3000 minutes at 5v5 between 2007-12. I’ve assembled the data on them from Behind The Net and then sorted the list of them by the percentage of goals scored at 5v5 with them on the ice on which they got assists. I’ve then created ten groups of players, alternating between groups of 17 and 18, with two groups of 17 at the end.
There’s something going on there that’s interesting. First of all, I’ve got serious doubts that defencemen have much of an impact on shooting percentage with their playmaking – you’ll notice that the group that gets the most assists per goal actually has a pretty shoddy shooting percentage (eighth of the ten groups).
The thing that I find really interesting (and, sort of surprising) is that the defencemen who accrue assists at a higher rate of goals tend to be on the ice for more goals for than defencemen who don’t accrue assists on such a rate of goals. On one level, this seems sort of unsurprising. Defencemen who are more likely to get assists on goals that are scored are probably more able to do offensive things, which should, all other things being equal, lead to more goals.
On the other hand, I’ve never seen anything that sort of suggests something about the size of the gap between offensive defencemen and non-offensive defencemen. This hints that there might be an answer we can suss out. From that perspective, this is very interesting stuff. I’ve re-organized the information presented in that table above, splitting my group into two groups of 87, based on the rate at which they are credited with assists on 5v5 goals.
You’ll note that there’s a pretty significant goal rate difference between the two groups. Even if you allow for the lower on-ice S% of the second group (I’ve got my doubts that it means anything), you’re still left with about a 0.15 G/60 difference between the group that gets assists on goals and the group that doesn’t. Obviously, this comes from the shot rate difference, which is kind of intriguing – it seems reasonable to assume that defencemen who do good things with the puck lead to more shots on goal and, by extension, more goals. If you’re talking about a fifteen 5v5 minute a night defenceman, it’s a difference of three goals for between each group. That doesn’t seem like a lot but it’s not nothing. It is, of course, more significant around the edges. There may also be a defensive benefit to the puck not being in the defensive zone as much, although that remains to be seen.
I’m not sure that I’d be willing to draw conclusions from this yet – there are some possible explanations for this that I’d be interested in checking. I’ve included the ZoneStart data, although I don’t think it’s enough a difference to explain the discrepancy. In the case of top four defencemen, who tend to play with top six forwards but not with one exclusive line, I doubt that quality of teammates would be the explanation. Quality of competition is another potential explanation although, again, with top four defencemen it’s hard to attribute this to that.
The irritating thing about points as a measure of offensive production is that they don’t really tell us anything about the contribution that the individual player made to the goal. As in the case of the Justin Schultz assist that I mentioned at the opening of this piece, players are going to get credited with a certain number of assists simply by virtue of being on an NHL hockey rink. What makes this interesting is that it hints at a measure of offensive skill of a defenceman that strips away credit for that which the player probably doesn’t deserve credit.
I’m going to write some more posts about this – there’s a lot of stuff to cover but I thought it’s worth mentioning that this does seem to pass the old Bill James smell test about stats that surprise you too often being worthless. The top ten names in terms of A/GFON: Rob Blake, Mark Streit, James Wisniewski, Brian Campbell, Brian Rafalski, Marek Zidlicky, Keith Yandle, Erik Karlsson, Dmitry Kulikov and Andrei Markov. At the other end: Francis Bouillon, Greg Zanon, Mark Fistric, Aaron Ward, Mark Eaton, Eric Brewer (if you got tired of hearing him referred to as a puck mover by ignorant national media while he was an OIler, you can smile at that), Mark Stuart, Rostislav Klesla and Shaone Morrisonn.
That being said, there are some surprises too: Dion Phaneuf (21.7%) and Jay Bouwmeester (21.6%). Both slot in right below Marc Staal (21.8%) who doesn’t really have any sort of a reputation for shot blocking. Oilers’ fans will be unsurprised by Ladislav Smid’s abysmal rating – in that five year span, he got an assist on just 20.4% of the 191 5v5 goals scored while he was on the ice. Tom Gilbert checks in at a hair under 27.9% of the 279 5v5 goals scored while he was on the ice. There are no prizes for guessing that the Oilers generated shots at a higher rate with Gilbert on the ice.