May 22 2012 05:05PM
(--NOTE-- This piece was written by Rob Pettapiece of The CIS Blog , known also for his work on players who claim to be in the the best shape of their lives and he is also the man behind the significant discovery that Canadian teenagers age . Visit his more general sports blog here . Primarily a baseball guy, I often joke with Rob that he knew about Corsi and advanced hockey stats before he knew anything about hockey.)
We know that players with more offensive zone starts tend to have better offensive numbers. But are those players given these OZone starts because they are better offensively, or do they merely appear better offensively because they are given these zone starts? In other words, we know there's a correlation, but where is the causation, if any?
The Canucks' justification for trading Cody Hodgson suggests that you can inflate a player's statistics if you give him more offensive zone time, so at least one team (and many of its fans) believe that it's the zone starts that make the player, to some extent. But how can we figure out just how much a zone start is worth to a player's offensive numbers?
The Study - Ozone and Dzone groups
To start, we will simply take out the offensive and defensive zone starts completely.
We went through the play-by-play for both the 2010-11 and 2011-12 regular seasons and extracted, for each player, the number of goals and shots on goal recorded by his team while he was on the ice -- but only in a specific situation: following a neutral zone faceoff, with the score tied, at 5on5.
This removes the advantage of an offensive zone start (or the disadvantage of a defensive zone start) from that shift for that player. It also removes score effects. We didn't count up blocks or missed shots due to arena discrepancies, and we counted the total shots, rather than just his own, to get a bigger sample. (Note: we only looked at players with at least 500 minutes played across all 5on5 situations in a season.)
As a result have, for every player: time on ice, shots for, and goals for in that specific situation - in other words, a zone-start-independent measure of this player's offensive output.
With this method to evaluate players independent of zone starts, we can then match two skaters who are similar by that measure (and thus assumed to be generally equal offensively), but otherwise different in terms of their distribution of offensive and defensive zone starts. In particular, players with an 82-game zone-start differential of +50 or greater (offensive zone starts minus defensive zone starts, expressed per 82 games) were put in the "OZone group", and those with a differential less than -50 were put in the "DZone group".
For example, in 2011-12 Manny Malhotra played about 104 minutes following a neutral zone faceoff (again, score tied, 5on5), with the Canucks recording 22 SOG and 2 goals in that time. He is safely in the DZone group, with the most extreme negative differential of any skater, at -516 zone starts per 82 games. We want to match him as closely as possible to someone in the OZone group, someone who (ideally) has 104 minutes, 22 SF, and 2 GF.
As it turns out, his best match among the OZone players is Yannick Weber (+68 zone-start differential), who played just under 100 such minutes, with the Canadiens recording 30 SOG and 2 goals. Malhotra's total time on ice at 5-on-5 was about 762 minutes, 104 of which we've already covered here, leaving 658 minutes following an offensive or defensive zone start (or a neutral zone start where the score wasn't tied). In those 658 minutes, Vancouver scored 23 goals, or 2.1 goals per 60 minutes. Weber, on the other hand, played 584 other minutes, in which Montreal scored 20 goals, or 2.05 GF/60. So in this case we don't see a zone-start advantage.
Of course we're not just basing this on one pair of players. There are actually 245 matched pairs of players across the two seasons ('10-11 and '11-12), and we followed the above procedure for all of them. The average results are below:
It's the last two columns we're most interested in and they suggest the OZone players were on the ice for seven more goals, or an increase of 0.44 goals per 60 minutes. This increase is not a surprise given that they averaged 202 more net offensive zone starts, which we’ll round off and call +100 vs. average.
The OZone players already had a very small (+0.02) advantage in the zone-start-independent situations, which reflects the fact that we matched on three things (time on ice, shots, goals) and not just goals. That difference (0.44 minus 0.02) means our zone-start effect is estimated to be about 0.42 extra goals per 60 minutes of this kind of ice time, given 100 extra offensive zone starts vs. the average.
The average player in our OZone group played 824 such minutes, so that’s about 5.7 goals in a season. Of course, the player in question doesn't score all of those goals himself, but we can estimate how many more points he gets: among all players in our OZone group, they picked up a goal or assist on 58% of their team's goals. 58% of 5.7 is 3.3 points per season.
In other words, the average skater, when given 100 extra offensive zone starts over the course of a season, can be expected to score three more points than he otherwise would. This result is more extreme, of course, for a Sedin: in 2011-12, Daniel Sedin had a differential of +423 per 82 games, in about 876 zone-start-affected minutes. For every extra 100 zone starts, the above says he would gain 3.3 points, so our results suggest he gained about 14 extra points over the whole season from zone starts alone - or 20% of his even strength point total. For Henrik, it’s about the same.
Those numbers seem incredibly high - earlier this year Gabriel Desjardins estimated that Daniel Sedin “would have 7-9 fewer EV points this season if he got 50% o-zone starts”, not 14. It’s possible that our estimate is off - we’re only using, on average, 160 minutes per player to establish his zone-start-independent scoring rate, or as little as 8% or 10% of his overall ice time. Unfortunately, that’s about as much as we can use given how few shifts begin in the neutral zone, not to mention the score differential and even-strength requirements as well. Including ice time after zone starts would defeat the purpose of trying to isolate them.
We could also look directly at points scored by the player in question, rather than goals scored by his entire team, but that introduces more sample size problems of its own. Besides, even if a Sedin doesn’t record a goal or assist himself, he still benefits from the offensive performance of his linemates, no matter what version of Corsi/Fenwick/scoring chances we’re using.
Going back to the questions we asked at the beginning:
1) It certainly appears that you can pump up a player’s offensive stats with more offensive zone time. That is, it’s not simply that better offensive players get more OZone starts, but that the favourable zone starts do indeed affect a player’s point total.
2) This effect is probably at least a couple of points per season for the average player, and while maybe not 14, certainly quite a bit higher for those who benefit from the "Sedin treatment".