“All I know is that if I had Neilson chances data for the #NHL, I’d love to make trades with GM who goes by any other +/-”
If you follow the Edmonton Oilers on the internet, you’ve probably heard of Neilson Numbers. Basically, it’s scoring chance +/-, but with only those chances on which a player is involved being taken into account. There’s a fellow at the Edmonton Journal, David Staples, who writes about them a couple of times of week. He’s made a number of claims as to their superiority to scoring chances or Corsi as a measure of a player, because they capture only plays on which a player made a contribution. I’ve never seen him do any real sort of examination of them though and discuss the circumstances in which they deviate from the results produced by the team chances.
The gist of David’s argument, as I understand it, is that team based stats, like Corsi or scoring chances produce too many false positives/negatives. Guys are getting credited for things to which they did not make a contribution. I’m sort of uncomfortable with the idea that we can assign credit/blame like that because scoring chances happen within a certain context which is created by all of the players on the ice. If the forwards do a good job keeping the puck deep on a shift, there aren’t going to be any scoring chances again, which helps everyone’s numbers, even if the defencemen didn’t contribute to it. If the puck stays in the Oilers’ end for the entire game, defencemen and centres are going to be watching the numbers climb on their scoring chances against.
Let’s look at the scoring chances and Corsi for the Oilers’ eight most used wingers this season. I think there’s a consensus developing that scoring chance ratio and Corsi tend to track one another nicely. Vic Ferrari’s made the point and I’m pretty sure I’ve seen other people make it as well. Looking at the Oilers’ wingers, that’s broadly true this season. Hall (1st in Corsi, 1st in SC ratio), Hemsky (2nd, 2nd), Eberle (3rd, 4th), MPS (4th, 3rd), Eager (7th, 5th), Smyth (6th, 6th), Jones (5th, 7th) and Petrell (8th, 8th). Looking at this chart, you’ll notice that Hall, Hemsky, Eberle, Smyth and Jones are all pretty tightly bunched in SC against – 16.2/60, +/- 0.5. Eager, MPS and Petrell are all far lower – as I’ve written before, there tend to be fewer events when guys who bat down the lineup are on the ice, presumably because coaches instruct them to play more conservatively and they play against other lesser players who have also been instructed to play more conservatively, ensuring that time is filled with little consequence before the guys who are allowed to play interesting hockey come back on the ice.
Let’s look at David’s Neilson data. I’m particularly interested in the percentage of scoring chances on which they got tagged with a minus. Hall and Hemsky are reasonably close to one another, with Eberle some distance back, followed by another gap into which Smyth falls and then, way further back, Jones, who is tagged with a minus on only 23.9% of the scoring chances for which he has been on the ice this year. You can probably guess where I’m going with this.
These are the chances for which each player was on the ice and didn’t receive any credit/blame. Where I’m suspicious of Neilson numbers is this. When Hemsky’s on the ice, the Oilers have allowed 15.6 CA/60. Jones is at 15.7 CA/60. Oddly, Hemsky’s making mistakes that get him tagged with responsibility for chances against at about 156% the rate that Jones is. Logically, assuming all other things are equal (which they aren’t, but bear with me), shouldn’t Hemsky’s chance allowed number when he’s on the ice be higher than Jones? Why do the remaining Oilers make fewer chance causing mistakes when they’re out with Hemsky than they do when they’re out with Jones?
This doesn’t seem to make sense to me. Unfortunately, we can’t really look into why this might be with the data – it seems more than possible to me that guys like Hemsky/Hall handle the puck a lot more than Jones, which means more turnovers, which leads to more chances where they’re the ones getting tagged with errors. In effect, Jones is more peripheral to what happens on the ice than is Hemsky or Hall, and looks better because of it. Looking at David’s data, he has Jones with a slightly better Neilson differential than Hemsky or Hall. I’ve got serious doubts that that’s an accurate measure of their relative contributions (leaving strength of opponent/teammates aside).
David is adamant that this individual stat is better than the team based stat because of the possibility for false positives/negatives. I’m not convinced. There’s undoubtedly the possibility of false positives/negatives with team based statistics. The advantage that they have though is that they capture the things that players do that lead to scoring chances that aren’t captured in the scoring chances themselves. Why do the Oilers get outCorsi’d and outchanced so badly when Jones is on the ice if he contributes to scoring chance differential on par with a Hemsky or Hall? Is it not possible that Hemsky and Hall are doing something that tilts the ice that Jones doesn’t do, something that David isn’t capturing? Anyone got a better explanation?
Bill James had a line about how often a stat should surprise you, something like if it was surprising you more than 20% of the time, it was probably faulty. On the basis of this, I’ve got some doubts about applying Neilson numbers to wingers, specifically, and forwards more generally.