Beware of What Zone Starts are Telling You, Part II: Shot Metrics

In Part I of this series on zone starts, we discussed how coaches’ are limited in their ability to completely control line deployment, particularly when we talking about only offensive and defensive zone shift starts (which make up a very small portion of all shift starts), as well as why the difference between zone starts and shift starts further muddies the relation between metrics like ZSR or OZS% and the actual intention of the coaches.

I hinted at the end that while starting in a specific zone has a large effect on the shot metrics for that particular shift, the overall effect of zone starts on measures like Corsi and Fenwick is actually quite small. In Part II of this series, we’ll explore why that is the case, and discuss how metrics that adjust for zone starts function.

The connection between zone starts and shot shares (Corsi, Fenwick, etc) is easy to imagine: if you take a faceoff in your zone, you’re more likely to surrender shots against; if you take a faceoff in the opposing zone, you’re more like to generate shots for.

The influence of shift start locations on Corsi-for percentages is actually quite large:

Micah Blake McCurdy of hockeyviz.com took this analysis a step further and mapped out shot rates based on time from the beginning of shifts. This is an eyeful at first, but it contains some fascinating revelations.

The solid lines in this graph are shots* for rates, while the dotted lines are shots* against rates. Notice the huge spike in shots for among offensive zone faceoffs and the corresponding spike in shots against among defensive zone faceoffs – they quickly reach nearly triple the 5-on-5 average:

However, that spike lasts a very brief period of time. By far the biggest effect occurs within the first 10 seconds. By 15 seconds after the drop of the puck, we’re looking at a difference of about 20 shots* per hour (one shot every five seconds), and by 20 seconds, the effect is almost nonexistent. All lines converge and eventually begin to separate after about 40 seconds. At this point, all the for rates are grouped together, and all the against rates are grouped together. The longer the player stays on the ice, the more their for rate decreases and their against rate increases.

The takeaways here are as follows:

With the most common shift length being 53 seconds (as found by McCurdy), players are constantly staying out long enough to begin erasing the benefits they saw by starting in the offensive zone, and the longer they stay out, the more likely it is they get trapped – and then things can really get messy.

I’ll offer my apologies if all of this is just adding to the level of confusion. This is simply the groundwork while the real evidence of the effect of zone starts lies in the aftermath of formulas that adjust for them.

The solution to taking into account zone deployment without letting it provide more influence than it deserves is to use zone adjusted adjusted shot metrics.

There are several different adjustments out there, though many of them rely on the same methods: determine the amount of shots that occur in shifts starting in various zones, and use the varying totals to apply coefficients to individual shots. Matt Cane did this in his article on the subject, as did Micah McCurdy.

When Micah adjusted the Corsi-for percentages of all players from the 2014-15 season, he found that for the vast majority, there was almost no difference:

This would indicate that variations in zone starts have a rather unimpressive effect on shot metrics as a whole. The explanation for the discrepancy between this and individual effects noted above is that

The formulas that we’ll focus on are those of Emmanuel (Manny) Perry – as the webmaster of Corsica, his numbers are widely available and thus they make the most sense to use – and they will show a slightly result (though similar) result. Perry’s are derived based on Micah’s methods and discoveries, but his formulas include a significant addition: they factor in Micah’s assertion that the effects of zone starts are largely contained in the first 20 seconds following a faceoff. As such, events that occur within 20 seconds of a faceoff are weighted differently than events that occur more than 20 seconds after a faceoff.

Perry’s site adjust not only for zone starts, but also for score state (leading, trailing) and venue (home/away), leading a massive amount of various coefficients that are applied to shot events:

Unfortunately, Corsica doesn’t provide stats with zone adjustments only – just score adjustments and score + zone adjustments. So to compare, we’ll look at the difference between score and score+zone adjustments for each player from the 2007-08 to 2015-16 seasons, sorted into the following histogram:

As you can see, the vast majority of adjustments are within two percent, with some outliers in either direction. Like I said, the results are a little bit different – Perry’s formula causes adjustments that are somewhat higher than Micah’s simpler formula – but the overarching conclusion is still the same: zoen starts do not have a great effect on shot metrics.

Here Perry described the greatest adjustment, belonging to Nashville Predators centre Paul Gaustad:

Let that sink in: Paul Gaustad, who took 387 more faceoffs in the defensive end than the offensive zone had the Corsi-For percentage of his entire 2015-16 adjusted up by just seven and a half percent. That is why claims like “Sure, Horvat’s CF% is only 44 percent, but look at his zone starts!’ don’t hold much water. The difference between Horvat’s score adjusted CF% and score and zone adjusted CF% in October was just 0.78. Horvat, who took 17 more defensive faceoffs than offensive faceoffs in October, has a long way to go to reach Gaustad levels.

You might be wondering why the adjustment is so small, given that the chart at the beginning of this article showed such large benefits for individual zone starts. The explanation for the discrepancy is the sample size problem that I noted in Part I: offensive and defensive zone starts make up only a small portion of all players’ shifts – the rest of them, neutral zone faceoffs and on-the-fly changes, regress the personal metrics back towards the individual player’s natural capabilities.

The bottom line, once again, as Micah Blake McCurdy put it:

Now that we’ve seen what the data says about the average NHL player, in Part III, we’ll analyze how individual ability helps move the needle when it comes to zone starts, even if the overall effect is small, and why determining which players should frequently start in which zones is not as obvious as it seems on the surface.