Did Fayne feign struggling this season?

The Oilers partially based their acquisition of Mark Fayne on information gained from analytics. James Mirtle of the Globe and Mail referred to Fayne as an “analytics darling.” (Not a term I would use, although somewhat common in the analytics lingo.) He was one of several players the Oilers acquired based on advanced statistics, which included David Perron, Teddy Purcell, Benoit Pouliot, and possibly Rob Klinkhammer. Because Fayne was considered a top-4 defenseman, a lot was expected from him. After all, he did play top-pairing minutes with Andy Greene in New Jersey.

At the beginning of the 2014/15 season, then coach Dallas Eakins put Fayne up against the toughest competition, along with Martin Marincin. Fayne (and Marincin) did not flourish. He, along with the Oilers, crashed and burned by December. Even after Eakins was fired, and Todd Nelson was brought on as interim coach, Fayne never did recover. He continued to struggle mightily. “But that’s because he faced the toughest competition,” one may interject.  Another could argue, “He’s the best defender on one of the worst defense corps in the league.” A few many even believe that Fayne played well, or as well as can be expected on a bottom-dwelling team. Finally, an embittered fan might explain Fayne’s performance with a more succinct, yet familiar refrain: “Because Oilers.” All these explanations could have kernels of truth to them, or not. Does the data support any of these explanations? Going the way of the scientist, I wanted to investigate the data and see what I could conclude.

My approach was scientific in that I began my data quest with questions (i.e., hypotheses) and then attempted to figure out how, or if, the data answered these questions (i.e., supported or didn’t support my hypotheses).  My questions included: Did Fayne feign struggling this season? (Did he actually struggle or am I seeing things that are not there?) If his struggles were real, what was the influence of the toughness of competition and quality of teammates? Assuming that his under-performance cannot be blamed on his competition or teammates, was their anything in his time with New Jersey that could have been a red flag? Was there some bit of data that should have given management (and analytics-inclined fans) a sign as to whether he could handle top-pairing, or even 2nd pairing minutes?

Scoring Chance +/-

I’ll begin with the most direct evidence by borrowing data from David Staples of the Edmonton Journal, who diligently tracks scoring chances from every game. Specifically, he records whether players contributed to offensive scoring chances or made errors in scoring chances against. He then computes a plus-minus of scoring chances. (Originally, the metric was named after Roger Neilson, who created it.). A positive value shows a player contributed to more offensive scoring chances than to errors when defending against scoring chances, and a negative value shows the opposite. The complete Oilers 2014/15 season can be viewed here. Fayne’s scoring chance differential was -2.52 per 60 minutes, which was 2nd worst among the Oiler’s regular defense; slightly worse than Ference, whose differential was -2.44. Only counting his errors on scoring chances against, Fayne averaged 7.76 errors per 60 minutes. This was actually better than most Oilers, except for (former Oilers) Petry and Marincin. Hence, his limited offensive contributions to scoring chances was the main culprit in his negative scoring chance differential. (Spoiler alert: His limitations on offense are a pattern throughout my analysis.) In any case, I submit that this is the first piece of evidence that Fayne under-achieved.

Passing Metrics

A second piece of direct evidence is passing data collected by Ryan Stimson and his colleagues for his Passing Project. These passing and shot attempt metrics are measures of a player’s offensive contributions that go beyond what is collected by the NHL. The metrics in the graph below include:

  • CC% and CC/60 Corsi Contribution (or Shot Attempt Contribution), which are individual shot attempts, primary passes leading to shot attempts, and secondary passes leading to shot attempts. These are given as a percentage (i.e., proportion of shot attempts a player is involved in when on the ice) and per sixty minutes. These metrics tell you how much offense goes through that player while on the ice and also how often they contribute.
  • Composite SAG and SG represent the total number of shot attempts and shots a player generated from both primary and secondary passes per sixty minutes. SAG/60 is solely for the player’s primary passing contributions.
  • Entry Assists represent the number of controlled entries a player assisted on. This is determined by the number of passes in transition (prior to entering the offensive zone) that was recorded for each player.
  • SC Contribution% and SCC/60 are identical CC% and CC/60, but represent only the scoring chances a player was involved in. Passing data for scoring chances was combined with War-on-Ice’s scoring chance (link to definition) data to arrive at a player’s total number of scoring chance contributions. SC SAG/60 represents the number of scoring chances set up from a player’s primary passes.

Passing Metrics - Mark Fayne 2014_15


Here is the way I read this graph. The top one-third percentile (67%+) is top-pairing defenseman range, 33%-66% is 2nd pairing, and below 33% is bottom-pairing. Reading from left to right, Fayne’s offensive contributions involving shot attempts (first 4 columns), he sits around the 2nd-pairing range. He is also 2nd-pairing for entry assists (i.e., controlled passes into the offensive zone). That’s one thing I remembered from watching games. That he had a solid first-pass on defensive zone breakouts. Where he drops off rapidly is his contribution to quality scoring chances, especially as a set-up guy. He was 10th percentile in primary passes leading to scoring chances. He then again sits around the 2nd-pairing range for scoring chance contributions, which includes scoring chance shots, as well as primary and secondary passes.

Is this evidence that Fayne under-performed? Depends on what we were expecting. As a top-pair, he under-achieved. As a 2nd-pair, he performed as expected, except when it came to setting up quality scoring chances. Surprisingly, Sekera is not that strong in setting up scoring chances either. Based on this, Sekera paired with Fayne might not be a good idea. Unfortunately, Schultz appears to be the only defender particularly good at setting up quality scoring chances. Both Fayne and Schultz are right-handed, so it’s unlikely, although not impossible, to pair them.

Quality Shots: Scoring Chances & High-Danger Scoring Chances

Next, I look at indirect evidence, namely, group measures such as a team’s shot attempt and scoring chance metrics. Based on these measures, especially the relative metrics, we make plausible inferences or guesses about a player’s contributions. I begin with Scoring Chances (SC) and high-danger scoring chances (i.e., shots from the slot area) provided by war-on-ice. These metrics are especially useful because they account for shot quality; locations and shot types that greatly increase shooting efficiency. (Unless otherwise stated, I present these scoring chance measures per 60 minutes.) We know that about 20% of shots from the slot (High-Danger) go in. Suppressing shots from the slot, then, has much more impact compared to shots from the blue-line, which only have a 4% efficiency rate.

Starting with suppression of quality shots, and relative to 170 defenders who played at least 750 minutes, Fayne’s High-Danger Scoring Chances Against  (17.16) and Scoring Chances Against (28.39) ranked 144th and 132nd, respectively. How do these metrics compare to other Oilers’ defensemen? Fayne’s High-Danger SC Against is 3rd worst, ahead of  Martin Marincin (19.13) and Keith Aulie (21.29).  With respect to general Scoring Chances Against, Fayne’s rank is better, but nothing special. When Fayne is on the ice, the Oilers are no worse, or better, when it comes to suppressing quality shots. In other words, the team’s defense is average with Fayne involved.

But when comes to the Oilers generating scoring chances with Fayne on the ice, the lack of quality shot production is glaring. Compared to other regular defenders, he ranked 169th in High-Danger Scoring Chances generated (11.12) and 167th in Scoring Chances generated (28.39). I’m hesitant to say this, but the numbers are pretty clear: Fayne’s scoring chance generation metrics are among the worst in the league. This wouldn’t be so bad if his defense compensated for this lack offense, but it doesn’t.

I could look at other numbers, like shot attempts for and against, to support the idea that Fayne struggled, but I think what I showed so far is sufficient. With the Fayne on the ice, the team’s defense in suppressing quality shots is par for the course , which isn’t saying much. What’s worse is that the team’s ability to generate quality scoring chances is severely hindered. So how does this all translate into team scoring? Based on shots from all areas, war-on-ice provides a Net Goals metric for each player. With Fayne on the ice, the team’s expected Net Goals was -0.41 (per 60 min.), which when multiplied by his time-on-ice, is -7.41 net goals. Only Aulie was worse (-0.54 per 60 min.).

Quality of Competition

The Oilers were expecting a solid top-4 defender, which his SAT% metrics with New Jersey suggested, but his Oilers’ numbers presented so far don’t support this notion. One explanation could be that Fayne faced the toughest competition on the team, which he did. However, as I have shown in previous posts (e.g., Gryba vs. Marincin & Sekera vs. Hamilton), a player’s quality of competition can be controlled. Specifically, using David Johnson’s WOWY (With-or-Without-you) tables, a player’s shot-attempt differential (SAT%) can be measured against different levels of competition. Moreover, the player’s SAT% can be compared to other players across similar levels of competition. If Fayne’s SAT% took a big hit only against tough competition, then we could conclude that quality of competition was major factor for his struggles. Let’s see if that is the case.

In the graph below, I present Fayne’s SAT% at different levels of competition (i.e., quality of competition [QualComp] SAT% ranging from 56% to 46%), I also compare him to Ference, who many would consider bottom-pairing at best, and Schultz, who some consider bottom-pairing (like myself), but on a defensively weak team like the Oilers, he is passable as a second-pairing defenseman. Before the analysis, my expectation was that all three were dominated by the strongest competition (i.e., QualComp SAT% of 54% or greater), but would improve considerably as QualComp softened to 50%. Then as competition dropped to players with a negative SAT differential (QualComp SAT% < 50%), Fayne’s SAT% would become positive. Finally, I was hoping that Fayne’s break-even point–the level of competition that players hold their own with a neutral SAT differential (50%)–would also be at least 50%, if not more. Here is what I found (click on graph to enlarge).



The first thing I noted was that Fayne was more similar to Ference than to Schultz, which is to say, not that good. In fact, Schultz looks amazing in comparison, but I don’t want to get carried away. Even against weaker competition, Fayne barely kept his head above water. This is most clearly seen in his break-even quality of competition SAT% of 46.9%. Notable players at this level include Devant Smith-Pelly, Nate Prosser, and Dalton Prout. Smith-Pelly I’ve know because he sometimes played with Anaheim’s top line, but in reality, he’s a bottom-6 forward (ranking 310 out of 410 forwards with 1.12 points/60 min). Prosser is in a bottom-pairing role for Minnesota and Prout is Columbus’ 4th more used defenseman. Overall, then, Fayne is holding his own against the bottom-half of opposing teams. This is what I would expect from a marginal second-pairing defenseman, or a solid third-pairing defenseman.

Quality of Teammates

I was still questioning whether Fayne was really that bad. Could it be because he played more minutes, compared to Schultz, with weaker possession players? One combination is the Oilers shutdown line, mainly Boyd Gordon and Matt Hendricks, and the other is any line with Yakupov. (Unfortunately, Yakupov is the weakest possession forward among the Oilers’ regulars). Fayne played 188 minutes with Hendricks and Gordon, whereas Schultz played 120 minutes; so that’s a significant difference. But then he played 136 minutes with Yakupov and Roy, and Schultz played 174 minutes with them. Moreover, when these players were on the ice with Schultz, they had a better SAT% (50%) then with Fayne (46%). Does this mean Schultz is a better defenseman than Schultz? I’m not suggesting that, but as these metrics indicate, Schultz appears to do a better job at generating offense than Fayne, which helps compensate a little for his deficits on defense.

Who Fayne had as a defensive partner also impacted his SAT differential. With Nitikin, it was almost respectable: 49.6%. With Marincin, it was only 46.3%. (Fayne’s adjusted SAT% was 46.7%.) Seems that being paired with a rookie, especially against the toughest competition, did not help Fayne. A few players did help Fayne gain a positive SAT differential. The main possession driver was Eberle, who helped Fayne exceed 50% when he was paired with either Marincin or Nikitin. I could go on to describe other combinations, but fortunately Micah McCurdy has done this work graphically with what he calls WOWY diagrams. This one below, if you follow it carefully, shows how Fayne and his teammates’ SAT differentials changed depending on whether they were with him or without him. I’ll use Marincin (#85) as example to help you understand the diagram (click on the diagram to enlarge).


wowy-1415-EDM-Defense - MARK FAYNE-shots

(You may want to increase your font size to for this part.) The bottom-right half, below the red line, is positive possession (good). Above the red line is negative possession (bad). Red-boxes indicate a teammate’s SAT differential without Fayne. If you look to the top-right and below the red line, you will see #85 boxed in red, which is Marincin (SAT% = 50.7%) without Fayne. Then if you follow the line from #85 to the left, you’ll cross the cut-off line (where SAT% = 50%), into negative possession (SAT% < 50%), you’ll see #85 in black. That’s Marincin and Fayne together with a 46.3% SAT differential. In short, Marincin appeared to be better without Fayne than with him, although that’s probably too simple a conclusion. If you look directly above this point (85 in black), you’ll see #85 in blue, which shows Fayne without Marincin. Fayne, then, whether he played with or without Marincin, his SAT differential changed very little. Still, there was small effect in defense with Marincin, namely, there were fewer shots against.

I’ve only explained the effect of one teammate. So as you can see, this elegant diagram packs a lot information into one visual image, which is very cool.

The main take-home message of this graph is how many teammates’ SAT% improved with Fayne (numbers in black). You can barely see it, but one player’s SAT% was positive and that was Pouliot (#67 in black, which is behind #67 in red). That’s it. If Fayne was driving possession, if players gained an advantage offensively or defensively with Fayne, we would see a pattern in which “red” numbers move from left to right toward “black” numbers. We don’t see that pattern, even for Pouliot.  Pouliot’s SAT% is unchanged playing with or without Fayne. What we can conclude from this pattern is that Fayne is not a possession driver. Indeed, several player are worse with him than without him, which we can see in all the red numbers moving from right to left (e.g., Eberle #14 & Hall #4), from positive to negative SAT differentials.

What did Oilers management or analytics followers not know?

I believe I’ve shown that Fayne struggled as a top-pairing, and even 2nd-pairing defenseman this season. As per Staples’ scoring chance data, Fayne had the 2nd lowest plus-minus scoring chance differentials among the defense. Based on Stimson’s passing project data, Fayne’s passing effectiveness puts him in the range of an average 2nd-pairing defenseman. Notably, though, he rarely sets up quality scoring chances through primary assists. From the perspective of the Oilers’ generating and suppressing quality shots, the team (a) allows more (compared to the team’s average) High-Danger chances with Fayne on the ice, and (b) generates fewer high quality shots. Overall, the team’s expected Net Goals with Fayne on the ice is about -7. Is this due to his toughness of competition? I don’t see evidence for that . What I noted is that the team’s SAT differential with Fayne on the ice was not that strong even against weaker competition. As to the reciprocal influence of teammates, I noted that Fayne tended to reduce a teammate’s SAT differential. Some teammates improved Fayne’s SAT%, like Eberle, but given the one-way relationship, it supports the notion that Fayne is not a possession driver.

How did a strong possession player, a top-pairing defenseman on a strong defensive team like New Jersey, become what almost looks to be a possession liability? What did management miss (if anything)? What did the casual analytics follower miss? I say “casual” because Ryan Stimson, the Devils’ analytics blogger for SB Nation, was not a fan of Fayne and he had analytical reasons for that.

First, there is a hint that Fayne’s SAT% in 2013/14 was inflated by certain teammates, specifically, Greene and Jagr, who was the Devils’ best possession forward that season. When Fayne was paired with defensemen other than Greene (time-on-ice = 248 min), his SAT% fell significantly by over 10% to 46%. When Fayne wasn’t with Jagr, his SAT% fell by 5% to 52%.

Second, Stimson had collected passing data from the Devils 2013/14 season. (That was before he expanded the project league-wide.) One statistic that he focused on was a player’s passing efficiency, in particular, what percentage of shot attempt passes lead to shots on goal. He referred to this metric as Shot Attempt Generation Efficiency (SAGE). I think the idea is that if a player’s whose passes lead to more shots on goal, the player presumably has better passing skills, which, seems to me, involves awareness of the developing play, mobility, puck control, and accurate passing. In Stimson’s analysis of the Devils’ defense, Fayne had the lowest SAGE, which was about 30%. As Stimson concludes:

“Mark Fayne generated the third most shot attempts [compared to Devils’ defenders], but was the least efficient by a considerable margin. I think the Devils will be just fine in replacing him in the lineup. Certainly the volume [of shot attempts generated] was solid, but if there’s no efficiency, it’s just empty possession and inflating Corsi totals.”

I think Stimson’s findings provide a reasonable explanation of why Fayne did little to help the Oilers’ offense. We see confirmation of his passing inefficiency in the 2014/15 passing metric data, which shows a mediocre rate (45th percentile) of scoring chance contributions and an abysmal rate (10th percentile) of primary passes that lead to scoring chances. This 2013/14 data was not available to the public until after Fayne was acquired by the Oilers. Hard to say if Oilers management would have considered it seriously, given that it’s not official NHL data. Nonetheless, I think this is evidence that helps explain Fayne’s offensive weakness, which when combined with evidence of his SAT% being heavily dependent on Greene and Jagr, helps explain to some degree his struggles with the Oilers.


Based on this analysis, I offer the following recommendations. (1) Fayne needs to be on the ice with a defenseman who is a puck-carrier and who is more efficient in their passing, especially in the offensive zone. Although I noted above that Sekera might not be a good idea, he does fit the bill of being a strong passer. And Fayne is better than Sekera in executing passes that allow controlled offensive zone entry. In other words, they do appear to compliment each other. As a 2nd pairing combination, Fayne with Klefbom could also work. Klefbom’s passing strength is in the offensive zone, which may help offset Fayne’s limited passing efficiency. (2) Against opposing top-6 forwards, Fayne should not be paired with weaker (i.e., bottom-pairing) defenders, especially rookies and prospects. (3) Reduced time-on-ice with weak possession players, otherwise these line combinations can be a recipe for puck possession disaster. Here I’m thinking of the Oilers’ shut-down line (Hendricks, Letestu, & Klinkhammer) and any line with Yakupov. I do realize Fayne was often put on the ice with the shutdown line, but their SAT% together–47.8%– is simply not that strong. Being realistic, though, the Oilers depth on defense is still lacking. There might not be a better choice than Fayne for this kind of deployment.

This analysis took me much longer than expected. Part of the reason is that I was surprised by some of my findings. I wanted to make sure I wasn’t missing something important, or concluding too much from too little. As always, I’m receptive to new evidence and lines of reasoning. If you believe that I have missed important information or reasoned wrongly, please comment below. Also, if you have any questions or confusion about what I presented, please ask away. My underlying approach in this blog is to make analytics as accessible as possible to hockey fans. Thanks for reading.

Walter Foddis Written by:
  • Daniel

    Didn’t he get pretty rough zone starts? Shouldn’t all of the comparisons here be corrected for this?

    Good stuff, though, makes you question projecting him to the top pairing next year.

    • Walter Foddis

      Thanks, Daniel. I didn’t include zone starts because analysis by David Johnson (http://hockeyanalysis.com/2015/03/16/zone-starts-corsi-and-the-percentages/) has shown that zone starts matter very little to SAT% (Corsi) and other indicators of player performance. There are exceptions, like Boyd Gordon, whose SAT% one season was affected by 2% based on his heavy d-zone starts. What’s telling for me is how poorly Fayne’s SAT% was even against weaker competition. Even adjusting for a generous 2%, Schultz outperformed Fayne by several %age points, or rather, the team did much better possession-wise with Schultz than with Fayne. I want to stress, though, that it’s not Fayne’s defense that is deficient (although it’s not that stellar either; just average relative to the team), but rather, that his lack of offensive contributions that drive down his SAT% and scoring chance differentials. I, too, worry about the top-pairing minutes with Fayne next season. I can see Fayne and Sekera get taken to task when the Oilers play the California teams. But then, what choice do the Oilers have? We still lack depth on D. I don’t expect to see significant improvements in our defense until 2016/17 when Nurse, Klefbom, and hopefully Reinhart have taken strides forward.

  • Daniel

    If I understand that article correctly, he gives an example of two players and shows that their zone starts don’t affect both CF/60 and CA/60, but only one of the two. In his example, heavy D-zone starts didn’t produce higher CA, but did seem to heavily influence CF, while heavy O-zone starts did the opposite.

    Is there a more thorough analysis like this that shows analysis across many players? Otherwise, it sounds like Fayne might be suffering from the same thing as Gaustad in the example: Punishing zone starts don’t force up his CA/60 (because he can defend), but they do push his CF/60 way down. Thus making his % numbers look horrible, particularly compared to Schultz, who got the opposite treatment. I don’t know what to say about the comparison to Ference… if they are similar, we are in deep trouble.

    • Walter Foddis

      First, thanks for following up and reading my source! Second, let me back up a bit. The Corsi (SAT%) analysis is just one piece. If we look at David Staples’ and Ryan Stimson’s data, which is directly observed and recorded, zone-starts become almost irrelevant. For me, these 2 sources of data explain a lot of why we don’t see much offense when Fayne is on the ice. His passing to generate offense ranges from average to below-average for a defenseman. What’s really telling is that he is extremely poor in directly assisting on quality scoring chances. Then we have Stimson’s data from 2013/14 that shows Fayne to be the worst Devil’s defenseman in setting up shots that lead to shots on goal (as opposed to blocked and missed shots). The difference here between Schultz and Fayne is night and day. A few percentage points in zone-start adjustment cannot explain these directly measured differences. Staples’ data also supports the notion that Fayne is not much of offensive contributor. His scoring chance contributions per 60 min of 5.2 was worst among the Oilers regular defenders, slightly worse than Ference (5.6) and much worse than Schultz (8.4).

      There is more analysis on zone-starts, I found several more by David Johnson and Eric Tulsky, but fortunately we may not have to dig too deeply into that. Johnson provides both unadjusted and and zone-start adjusted metrics for Fenwick (unblocked shot attempts). Fayne’s unadjusted Fenwick: FF60=33.4, FA60=39.9. His zone-start adjusted Fenwick: FF60=35.4 & FA60=41.6. Fayne’s offense improves in the adjusted figure, but his defense is similarly worse, which leaves a for/against ratio relatively unchanged.

      So where does this all leave us? We have to hope to hell that Sekera can carry Fayne, that McLellan implements systems that improve our overall defense (Johnson has an article that suggests coaching systems have a much bigger impact on a player’s Corsi than other factors), that our players learn quickly to execute these systems, and continue to improve given that several are still not in their prime, or just got there (e.g., Eberle, Hall).

  • Daniel

    I think this is great, as picking these things apart are pretty key to judging what we’ve got going into next year.

    I’m afraid I struggle to see how zone starts could not have any influence on Corsi results, so I just dug into it. Pulling stuff from Hockey abstracts, I looked at the spearman correlations between different statistics and player’s Corsi Rel. Here’s what comes out:

    For all players, QoC correlates 0.173, QoT correlates 0.615 and Off zone starts correlates 0.468

    When I looked specifically at D, I was surprised to see that QoC correlates -0.047, QoT correlates 0.425 and Off zone starts correlates 0.387

    This suggests to me that zone starts really are affecting Fayne’s results, but that the difference in QoC they faced cannot explain why Fayne did not out perform Ference by more (though Ference did have slightly better QoT). However, I really do feel that the separation between Fayne and Schultz can be explained by their zone starts and their QoT, which both heavily favored Schultz.

    I just hope that if you give Fayne a better partner, like Klefbom or Sekera that things might just turn out a little better.

    • Walter Foddis

      Daniel, if you’re curious, the Oilers’ defense passing metrics are summarized by a blogger I follow, Sunil Agnihotri (http://bit.ly/1LLwwjf). Although zone starts may have some affect (but see my links below), I think the data clearly shows that Schultz is a superior passer compared to Fayne. This, in my view, helps explain why there is more offense generated (whether shot attempts or scoring chances) with Schultz on the ice. For instance, the rate at which Schultz produces primary assists on scoring chances ranks above the 95th percentile; 8th best among a sample of 224 defensemen. Fayne ranks 197th on this same metric.

      Here is more analysis from David Johnson on the nature of the weak relationship of zone starts to Corsi and other statistics. I cite David frequently because he’s done the most recent analysis on the topic. Plus, his is the only hockey statistics site that provides zone-start adjusted metrics.

      Zone Starts, Corsi, and the Percentages – David Johnson

      The above article was met with strong criticism from Sam Ventura (formerly of war-on-ice.com, but now an analytics consultant for the Penguins). Here is Johnson’s response to Ventura:

      Zone starts and impact on players statistics

      One of David’s key points is:

      “For me, the main evidence that zone starts don’t have a significant effect on a player’s overall statistics is if I remove the 45 seconds after all offensive/defensive zone face offs (which basically ignores the entire shift) the majority of players have the same CF% +/- 1% and only a handful with heavy offensive or defensive zone starts have an effect in the +/- 1-2%. If removing all shifts that start with an offensive or defensive zone start does not dramatically impact a players overall statistics you simply cannot conclude that zone start bias plays a prominent role in driving a players overall statistics.”

      It’s going to be interesting to see defensemen step up their game this season. I like our prospects, but don’t think any of them are NHL-ready. Let’s hope that Fayne and Nikitin at least perform little better compared to the defensive gong show fans had to endure last season.

      Thanks again for sharing your thoughts.



    • Walter Foddis

      Also of possible interest:

      How much do zone starts matter part II: A lot on their own, not that much in aggregate, Matt Cane


      Cane summarizes his finding as follows:

      “After all, even with our simplistic but blunt approach here, we see that very few players require significant adjustments to their numbers due to their zone starts. While it’s true that more accurate numbers are always better, it’s certainly becoming clearer that outside of a handful of players the effect of zone starts on possession aren’t as great as we may have thought, and may actually be negligible for a large portion of the league.”

  • Blaine

    I kinda hate the “zone starts don’t matter” theory. I know as it is mentioned in aggregate they balance out, but if you look at hockeyprospectus and where they break up Corsi for all three sones it tells you something. Now they only did this until 2014 but here are some numbers –
    Hall OFZCF% 78.69, DFZCF% 32.28, NZCF% 46.76
    Gordon OFZCF% 69.28, DFZCF% 32.23, NZCF% 36.56
    Fayne NJ OFZCF% 63.88, DFZCF% 36.50, NZCF% 51.65

    Now if you look at Fayne’s DFZCF% of 36.50 he is higher than all the Oilers that year except for Perron.
    1.Perron DFZCF% 43.35 2. Smyth DFZCF% 36.33 3. Marincin DFZCF% 36.04

    So this does suggest he is an excellent DFZ deployment. Plus if you look at the variance in all three zones, despite the aggregate, it is telling that zone starts do have a significant role in your overall stats, I’ll break out some numbers on this later which still do support the aggregate but are worth noting.

    Most over though in disregard for zonestarts I prefer “deployment”. When your deployed in a defensive role and it is your job to weather the storm, to swing the momentum you’ll see some unfair bias to your stats. When your put out there during your opponents highest swing and peak of momentum it is likely your going to give up more High danger scoring chances, your going to give up shots and your going to get less shots and quality chances as your teams momentum is at a low. You are gassed, and plus, putting you out for the least funnest part of the game gasses you even more. While your offensive predators get all the glory you get all the strain and drain! Plus your sent out there to turn the tides and if you successful they send out the predators to capitalize – not usually you. So they get the the “freebies” that you in essence earned!

    Now to the zonestart numbers. I’ve got different ones than the article and see a slightly different range based on them. But lets say –
    DFZ 30% – kinda low for DFZ but adjusted for OFZ which can peak much higher
    OFZ 70% – really good players can post above 80 easily and even over 90!

    So DFZ at a CF% of 30% with a 40% OZ% gives you – 30% for 60% of the time and 70% for 40%.
    so 0.3 * 0.6 = .018
    0.4 * 0.7 = 0.28
    0.28 + 0.18 = 0.46 or a CF% of 46%
    but if you have a 30% OZ% your looking at another drop
    0.3 * 0.7 = 0.21
    0.7 * 0.3 = 0.21
    0.21 +0.21 = 0.42 or a CF% of 0.42
    So does 10% zonestart difference = a -4% in CORSI? NO, you have to count Neutral zone as well and even on the fly. So if you say your combined OZ and DZ starts equal about 50% of your total and neutral zone and on the fly are another 50% than you look at a 2% difference per 10% zone start. Which is negligible almost – as suggested in aggregate. But its the deployment, the strain and drain that is suggested in a zonestart that implies a more significant impact on your stats!

    *Extra Math* (you can ignore)
    Some of the offensive powers post a 90% OFZCF%. So if you have a 30% DFZCF% but your up against a 90% your numbers are going to be worse. 90 + 30 = 120 now a generous balance would be just to keep the ratio of 3:1 so you’d get at 75% vs 25% = 100. But why this is generous is because your defensive numbers only dropped 5% and we assumed your opponents dropped by 15% But realistically someone so far above average is likely going to drop less than his opponents almost to the point where I’d flip this 5% and 15% around so you’d have a 15% DFZCF% and your opponent would have a 85% OFZCF% against you, maybe a 20% to 80% (or a 10 and 10). But that is still equal and while you sit at an average 30% your opponents has a 20% positive deviance from the average so it’s unlikely it would be an equal (10 and 10). You might want to take the full 20 point adjustment and give the guy with a 20% higher than average his fair share which is 20% * 20% or 0.04 so then take that 10 and 10 and create a difference of 4 so it would be 12 and 8 that would make his corsi about 78% and yours about 22% which sounds as close as you can get ( I might actually prefer to flip them and take 12 off your 30 and 8 off his 90 giving you a 82% vs 18% – as it seems it would be difficult to ever get a 90% if your only posting a 78% vs the average this how 85% makes sense to me too!)<- this is also the math that I've noticed to be lacking in Quality of Competition the balancing of the 2 numbers to equal 100 and the bias in the adjustment when doing so – whatever it is deemed to be.

    • Walter Foddis

      Thanks for the thoughtful analysis, Blaine. I’ll definitely have to look at the HockeyProspectus analysis. Important to remember, though, that zone starts determines about 50% of where a player is on the ice. The other 50% is transitional play and line changes on the fly. After a first read, I’m a little confused by some of your analysis, but perhaps because it’s 5 AM and I haven’t had my coffee yet! When I have time, I’ll revisit your comments with a clear mind. In the meantime, if you haven’t already read them, I would highly recommend the David Johnson articles I linked to in the comments above. In one article, Johnson concludes:

      “For me, the main evidence that zone starts don’t have a significant effect on a player’s overall statistics is if I remove the 45 seconds after all offensive/defensive zone face offs (which basically ignores the entire shift) the majority of players have the same CF% +/- 1% and only a handful with heavy offensive or defensive zone starts have an effect in the +/- 1-2%. If removing all shifts that start with an offensive or defensive zone start does not dramatically impact a players overall statistics you simply cannot conclude that zone start bias plays a prominent role in driving a players overall statistics.”

  • Walter Foddis

    I think I’m following what you’re arguing, and it makes some sense. I see now that you do mention “on the fly” needs to be accounted, but it does look like you’re putting too much weight on zone starts overall. Seems like you’re saying offensive and defensive zone starts account for 50% of zone time, which is incorrect. (Or maybe you do know that zone starts only account for 50% of zone time and I’m confused about what you’re saying.)

    As to balancing Corsi to equal 100% in accounting for quality of competition, I’m not quite sure what you mean. If a player’s Corsi is 40% against opposition player X, then obviously player X’s Corsi is 60%. What am I missing?

    If you’re not familiar, Johnson’s opposition WOWY tables show 3 sets of Corsi: (1) player vs. opposition, (2) player without opposition, (3) opposition without player. That way you can compute how much a player’s Corsi varies from his average against a certain player i.e., comparing (1) and (2). I use (3) to compute the oppositions’s Quality of Competition levels.

    As to Fayne’s performance, more generally, I used 5 sources of data, including David Staples’ scoring chance contributions, to form my conclusions about Fayne. Just focusing on Corsi will give a narrow view. Important to consider is a player’s individual Corsi contributions, which we can now do with (at least to some degree with the Oilers) by using Ryan Stimson’s Passing Project data, (see my blog entry for more details). This shows how players directly contribute to shot attempts and scoring chances.

    This data is perhaps the most telling on how mediocre Fayne is as a passer. As I note in my blog entry, almost all of his poor Corsi differential is due to a lack of contribution to offense. His defensive side of Corsi (Corsi Against), relative to the team, is average. His poor passing metrics help explain this offensive weakness. If I just looked at Corsi, or defensive zone starts (let’s say), I could say that his poor Corsi can be “explained” by zone starts, but when we see that his passing is not that great, we know that he is not helping his cause because of his own skills.

    Notably, his rate of passing and shooting (per 60 min) relative to the league for Corsi contributions is 50th percentile and scoring chance contributions is 40th percentile. In other words, he’s performing at a level of a marginal 3/4 D-man (i.e., 34th to 67th percentile).. What’s even worse is his rate of primary passes leading to scoring chances, which is around 10th percentile. A play-maker he is not. He seems to be OK in passes leading to controlled o-zone entries, although 60th percentile rank is still 3/4 D-man level.

    Finally, I forgot to mention this, but Steven Burtch has a measure he calls delta or dCorsi. Check out the labs menu at war-on-ice.com. He uses multiple regression analysis to arrive at a player’s expected Corsi-For and Corsi-Against, which accounts for coaching systems, quality of teammates, quality of competition, and zone starts. This expected value gives us the Corsi numbers of an average defensemen

    For Fayne, Burtch’s model predicted an expected Corsi-For of 48.7 and Corsi-Against of 57.8. Fayne’s actual Corsi-For and Against were very close to those estimates, and only slightly negative: Fayne’s actual CF=48 & CA=56.3. So using this statistically-derived estimation, Fayne still performed below the average defenseman.

    To me, this provides further confirmation that he is a middling 3/4 D-man. I think to put him up against top lines of good teams (i.e., top 16) is too much for him to handle. Mind you, with a better coaching system, and the additions of key players like Sekera and McDavid, along with improvements expected from developing players like Nugent-Hopkins, Yakupov, and Klefbom, perhaps Fayne’s performance (and resultant numbers) will benefit from this overall team improvement.