Thursday, September 8, 2011

On Zone Starts

How valuable are offensive-zone faceoffs? How much should we adjust the Sedins' stats to take into account that they take three faceoffs in the offensive zone for every one at the other end? Are coaches using their guys effectively, or should they almost turn them into specialists like Vigneault does in Vancouver? Is it better for a coach to focus on zone starts, matchups or just roll lines to keep his guys fresh and the best playing the most?

These are just a few of the many, many questions that relate to zone starts that come up in hockey analysis. We'll put off dealing with most of those until later, this article will be more of a broad discussion and introduction to what I feel is a novel approach that we'll be using a lot.

The Data

I'm going to start with team stats. This is pretty strange because zone starts aren't a team issue. A team that has more offensive-zone starts than defensive has earned them; good zone starts aren't just handed down to the team by some suit. In contrast, a player only reaps what he sows, as it were, if he or a teammate ices the puck or possibly after a very short shift. Most faceoffs are taken with fresh players that were on the bench when the stoppage occurred. Despite that, using team stats is a good place to start because we can get a pretty good idea how the location of the most recent faceoff affects results.

The data come from a common source: NHL play-by-play and roster reports. In this case, I am using every game available for the 2010-2011 regular season. As usual, it's even strength with both goalies on the ice. Using the roster reports, which say when each player was on the ice during the game, I isolated these 5 situations:

Ozone, first shift - the most recent faceoff was in the team's offensive zone. All players that were on the ice for the faceoff remain on the ice.

Ozone, on-the-fly - the most recent faceoff was in the team's offensive zone. At least one player who was on the ice for that faceoff has left.

Neutral Zone - the most recent faceoff took place in the neutral zone.

Dzone, on-the-fly - the most recent faceoff was in the team's defensive zone. At least one player who was on the ice for that faceoff has left.

Dzone, first shift - the most recent faceoff was in the team's defensive zone. All players that were on the ice for the faceoff remain there.

To clarify, once any of the original faceoff guys have left the ice the rest of the time before the next stoppage is in the on-the-fly category. So the faceoff shift is only that first shift, even if later it so happens that the 10 skaters that were on the ice are out there together.

Note: I have separated out the team results to keep this post about the general concepts. If you want to see how good your team is in each situation, you can find that here.

Ice Time and Goals

Let's take a look at how much time was spent and how many goals were scored on average in each situation. This will give us a very rough idea how important faceoffs are in the offensive or defensive zone.

SituationIce Time%Goals%
Ozone, first shift452.511.4%21.213.8%
Ozone, on-the-fly765.119.3%32.321.0%
Neutral1538.938.7%5435.1%
Dzone, on-the-fly765.119.3%34.822.7%
Dzone, first shift452.511.4%11.47.4%
Total3974.1100%153.6100%

The first thing I noticed is that the first shift only accounts for about 37% of the ice time following a faceoff at either end. Part of that is that I'm being very strict defining the first shift; it would increase if I allowed for one guy to leave the ice, for example. In any case, you might wonder why we focus so much on who is on the ice for a faceoff when so much of a player's ice time after an ozone/dzone faceoff started with him jumping onto the ice.

The goals columns give you a pretty strong hint. Look at the second and fourth rows. Following a faceoff outside the neutral zone, once a change was made more goals were scored by the team that took the faceoff in their defensive zone! We'll later see that this is likely just due to random chance, but it seems clear that if you come on in an on-the-fly change it's more like a neutral-zone start than being on for a faceoff at either end.

Corsi

Let's look at the average team Corsi rate for each situation. I use the average team rate for each situation so we don't have the endogeneity effect I wrote poorly about a couple months ago. In other words, if we simply averaged out all the ice time we'd overestimate how important it is to be in the offensive zone because good teams tend to get more faceoffs in the offensive zone than bad teams so we'd be lumping in team quality with ozone-faceoff value.

SituationCorsi/60
Ozone, first shift39.957
Ozone, on-the-fly2.603
Neutral0.019
Dzone, on-the-fly-2.784
Dzone, first shift-40.103

Here you can see how much having a faceoff in the good zone helps your team out territorially. I'm sure that there is an effect didn't surprise you, but suspect how large it is might have. It's also interesting to note that this almost completely goes away once the first change happens. This shouldn't be a surprise either, teams don't change without the puck leaving the zone, but one might have expected more of a ripple effect. The most common way to change lines is to dump the puck and give the other team possession, albeit starting behind their own net. It does not appear that being able to breakout the Flying V gives you much of an edge.

This is a nice segue back to why we should care so much more about who is actually on for the faceoff and not just ice time afterward. Let's exaggerate and suppose that Henrik Sedin and Manny Malhotra spend the same amount of post-offensive-zone-faceoff time on the ice but Hank's is all first-shift time while Manny doesn't take a single offensive-zone faceoff. Sedin's ozone time will be extremely favorable, and Malhotra's only slightly so. In this extreme scenario we would need to adjust Sedin's stats a lot to take his type of ice time accurately into account. Being out there on the faceoff in front of their goalie is over 15 times as favorable as jumping on the ice after.

Conclusion: More To Come

When I thought of this method of separating out the ice time, I did a little fist pump. While I don't think this, or really any metric, will blow everything else away it seems like a good way to analyze zone starts and give us better insight into player value and coaching decisions. We will be using this and related methods a lot in the future, especially the coming weeks leading up to the season. Coming down the pipe is a new player metric to adjust for zone starts. We'll also do in-depth analysis on zone starts for the two teams that focus the most on them: the recently rivalrous Vancouver Canucks and Chicago Blackhawks. Are Vigneault and Quenneville outsmarting the rest of the league or is it a case of fancy coaching syndrome? If you ask nicely, we could probably do something similar for your favorite team, or even your favourite team.

Here is a link to the team data.

4 comments:

  1. In the second chart, how is the D-Zone Corsi higher than the O-Zone Corsi?

    ReplyDelete
  2. Yeah, somehow the minus signs got taken off. Both the on-the-fly and first-shift Corsis are negative for dzone faceoffs.

    Thanks for the correction.

    ReplyDelete
  3. Great post. Does this somewhat validate why we use zone starts as a situational stat for Corsi? Given that we're measuring ZS's using faceoffs, it still seems like quite an important adjustment.

    However, had you shown that on-the-fly shifts had a higher impact, our player evaluation guidelines would need to be seriously re-thought.

    ReplyDelete
  4. Thanks.

    You have it exactly right. The difference between the faceoff shift and neutral is quite large, which isn't the case for all shifts after that first one. This means it is very important to take into account that first-shift time for each end and not bad to ignore completely non-faceoff time as far as adjustments go. If it had been something like:
    ozone faceoff-shift corsi rate: +40
    ozone after change: +25
    neutral: 0
    dzone after change: -25
    dzone faceoff shift: -40

    then we'd really have to take a look at non-start time.

    ReplyDelete