More on Evaluating Defense by Rank on Team

The other day I wrote a post on evaluating forwards and defensemen based on their rank on their team. The purpose of that post is to show the value in breaking down performance beyond just Corsi but into Corsi For/Against and shooting and save percentages. I wanted to expand on that by looking at past seasons to see if there are trends that emerge.

In the previous post I ranked players on their team based on their total ice time because that is what Travis Yost did in his post on TSN.ca. I don’t believe this is the best methodology to use because if a first line player is injured and misses half the games he might get treated as a third line player even though when healthy he always played on the top line, and may in fact had the most ice time in those games. To fix this problem I use % of the teams ice time only in games that the player played in. To be more precise, here is my methodology:

  • I used 5v5close data to eliminate score effects
  • I used data from the first two thirds of the season (first 820 games of the season) because few trades occurred before that point in the season. Trades can mess up rankings of players and how do you rank a player that played on two different teams.
  • I only accounted for players who played in at least 20 games.
  • I then ranked players by their %ofTeam TOI stat which you can find on puckalytics.com.
  • Defensemen were ranked as D1, D2, D3, D4, D5, D6 and D7/Depth. D1-6 has one player from every team, D7/Depth defensemen are every other defenseman with 20 games played but not among the top 6.
  • I looked at 2011-12, 2013-14 and 2014-15 seasons individually as well as combining the individual seasons. The D1 group for the 3 seasons combined are D1 from 2011-12 plus D1 from 2013-14 plus D1 from 2014-15.

I did this for forwards as well but I will leave that to another post. There is just too much to look at to do it all in one post.

Let’s start off by looking at the GF% and CF% by defense rank using all three years of data.

DefenseRank_3yrCFPctGFPct

You will notice that there is relatively little variability in CF% from D1 through D5. A slight drop off but not significant. The drop off in GF% is much more significant with D1 and D2 above 50% but the rest of the defense below 50%. D6 is an interesting anomaly here as the CF% is quite low compared to any of the other defensemen, including the depth guys. Keep this in mind as we move through this analysis because D6 comes up as a bit of an anomaly a few times.

Let’s now have a look at shooting and save percentages.

DefenseRank_3yrShPctSvPct

From D1 through D5 there is a slow drop off in save percentage with a total drop off of 0.45 percentage points and then save percentage spiked up 0.55 percentage points from D5 before dropping off again for depth defensemen. Once again, D6 is a bit of an anomaly with elevated save percentages.

Shooting percentages are fairly stable across all defensemen rising a bit for D5 and dropping a bit for D6. The way D5 and D6 diverge from each other in both Sh% and Sv% is interesting. I should also point out that the correlation between Sh% and Sv% is -0.59 so they are fairly well inversely correlated.

Let’s have a look at CF/60 and CA/60.

DefenseRank_3yrCF60CA60

CA/60 is fairly stable across all defensemen except for D6 where CA/60 is elevated. CF/60 shows a mostly steady decline from D1 through to depth defensemen except for D6 which is abnormally low. Again, the the comparison between D5 and D6 is quite interesting here.

I should point out here that the correlation between CA/60 and Sv% is 0.56.

How does this all wash out in terms of goal rated for and against?

DefenseRank_3yrGF60GA60

D1 and D2 are on the ice for more goals for than against while D3 through depth defensemen are on the ice for more goals against than for. Once again though we see D6 being a bit of an anomaly with far fewer goals against and goals for. The far fewer goals against by D6 defensemen is driven by shooting percentage because D6 defensemen give up the most shots against.

That last finding is an important one. In general, D6 defensemen are poor at suppressing shots (worst among all defense types) but excellent at suppressing goals (best among all defense types). If this is true, it would mean when evaluating defensive ability of defensemen (which presumably D6 mostly are) we can’t evaluate solely on Corsi% or Corsi against rates. Save perentages seem to matter.

I want to explore this a bit further by looking at individual seasons to determine whether this is a quirk in the methodology or whether this actually happens each year. Here is the Sv% chart with the individual seasons.

DefenseRank_3yr_IndSeason_SvPct

Yes, in each of the 3 individual seasons there is a spike in Sv% for the D6 defensemen to the point that D6 defensemen are, on average, either the best save% defensemen on their team or the second best (with either D1 or D2 being better).

The takeaway so far is, defensemen do seem to have an ability to boost save percentages and it seems to be inversely correlated to their ability (or desire?) to suppress shots against. More shots against often leads to higher save percentages.

High, Medium and Low Corsi Teams

I got a bit curious now and I wondered if there were differences between high Corsi and low Corsi teams so I split teams into three groups:

  • High Corsi (HC) teams had a 5v5close CF% > 52%
  • Medium Corsi (MC) teams had a 5v5close CF% between 48% and 52%
  • Low Corsi (LC) teams had a 5v5close CF% below 48%

I chose these cut offs because it more or less divided the teams into thirds though there were slightly more teams in the Medium Corsi group than the other two.

Also, because the groups do not contain players from all the different teams for this part of the analysis I am going to look at RelTM stats which will factor out team effects on the stats.

I am only going to show the defensive stats since they are most interesting for defensemen starting with CA/60.

DefenseRank_LC_MC_HC_CA60

Now this is quite interesting. The high Corsi teams seem to play the same style or have the same ability from D1 through depth defensemen as there is very little variability in CA/60. The low Corsi teams have their top defensemen give up far fewer shots against than their bottom defensemen while the middle Corsi teams are, well, in the middle of the two.

Let’s see how this translates to save percentages.

DefenseRank_LC_MC_HC_SvPct_2

Interestingly, it doesn’t translate as one might expect. One might have expected, based on earlier observations, that the higher CA/60 of D5-7 defensemen on low Corsi teams might have resulted in higher save percentages. This isn’t what happened. For low Corsi teams the top defensemen who gave up fewer shots against had higher save percentages and the defensemen lower down the depth charts that had higher shot against rates had lower save percentages.

The spike in save percentages that we saw earlier with D6 defensemen is due to the high Corsi teams. On high Corsi teams the D6 defenseman has a save percentage almost one full percentage point higher than the D5 defenseman. That is changing your goalie from a 92 save percentage goalie to a 93 save percentage goalie, or from good to elite.

So, how does this all translate into GA/60 stats?

DefenseRank_LC_MC_HC_GA60

This pretty much mimics the Sv% chart though inverted as better Sv% results in lower GA/60. The best GA/60 RelTM defensemen are the best defensemen on poor Corsi teams (don’t mistake this as being the best GA/60 defensemen, only best relative to their teammates) while the best Corsi teams are relatively even from top to bottom.

As you might expect, the worst Corsi teams have the highest CA/60 and the best Corsi teams have the lowest CA/60 team as shown in this chart.

DefenseRank_LC_MC_HC_CA60_NotRel

What is interesting is how these high/medium/low corsi teams rank in terms of save percentage.

DefenseRank_LC_MC_HC_SvPct_NotRel

There isn’t a big difference in D1 and D2 but after that the low corsi teams trail the high and medium corsi teams in save percentge by fairly good margins, particularly the second defense pairing.

Finally, here is the GA/60 chart.

DefenseRank_LC_MC_HC_GA60_NotRel

The low Corsi teams give up more goals when their top defense pairing are on the ice but the difference grows significantly larger as you move down the depth charts.

Conclusions

There is a lot to digest here and I am not certain I have fully grasped everything that is going on but here are a few of my theories and thoughts:

  • There do seem to be defensemen who can boost a goalies save percentage. In particular, D6 defensemen seem to have this ability, especially D6 defensemen on good Corsi teams.
  • Good Corsi teams also seem to have higher save percentages than poor Corsi teams. Does playing a sound puck possession game make life easier on your goalie? It seems it may. It may also explain why Stanley Cup winners are often good Corsi teams and also often good Save % teams. Of course, it doesn’t explain Cam Ward’s terrible save percentage on a good Corsi Carolina team. There is probably more to the story than that simple explanation but there seems to be some connection between Corsi and Save percentage.
  • Good Corsi teams have more balance throughout their defense, both in save percentage and Corsi Against rates. Is this a result of systems or a result of talent? I suspect a bit of both.
  • Although I didn’t show offensive stats above, generally, there wasn’t much to see in them. The one observation I will pass on is while D6 defensemen seem to be defensive specialists (low GA/60) D5 defensemen may be more offensive oriented defensemen (higher GF/60) in part due to elevated shooting percentages. I’ll look into offensive stats more when I look at forwards but it could be that D5 defensemen get a bit more ice time with offensive forwards.
  • Also not shown is any QoC stats but I did look at them. You can find some tiny differences in QoC among defensemen (and forwards) but they are so small it is difficult to imagine they have any measurable impact.
  • Finally, a caveat: Remember, these are broad generalizations. Not every team deploys their players in the same manner. Not every D6 defenseman is the same and not every D6 defenseman will boost their teams save percentage. These are just trends and generalizations and we must be careful when applying them to individual defensemen.

As I digest these observations more I may come up with more conclusions or theories and if you have any feel free to add them to the comments. When I look at forwards we will see some trends with offense as well.

 

This article has 3 Comments

  1. It was asked on twitter so here are the D6 identified.

    2011-12
    BRAYDEN MCNABB
    ROSTISLAV KLESLA
    MARCO SCANDELLA
    ADAM PARDY
    RYAN O_BYRNE
    JAMIE MCBAIN
    ANDY SUTTON
    LUKE SCHENN
    KRIS RUSSELL
    MARK STUART
    MATT NISKANEN
    MATT TAORMINA
    BRETT CLARK
    ADAM MCQUAID
    ROMAN JOSI
    AARON JOHNSON
    MATT GREENE
    AARON ROME
    JEFF SCHULTZ
    TJ BRODIE
    JUSTIN BRAUN
    ANDREAS LILJA
    TOMAS KABERLE
    SHELDON BROOKBANK
    JAKUB KINDL
    MARK EATON
    KEATON ELLERBY
    BRIAN LEE
    STEVE MONTADOR
    JEFF WOYWITKA

    2013-14
    LADISLAV SMID
    JON MERRILL
    NIKITA NIKITIN
    JAKUB KINDL
    PAUL RANGER
    SCOTT HANNAN
    MICHAEL STONE
    MARK PYSYK
    SERGEI GONCHAR
    MARK STUART
    SAMI VATANEN
    OLLI MAATTA
    BRIAN STRAIT
    NATE GUENIN
    LUKE SCHENN
    RYAN MURPHY
    KEVAN MILLER
    ERIC GRYBA
    JORDAN LEOPOLD
    FRANCIS BOUILLON
    JOHN ERSKINE
    VICTOR BARTLEY
    SAMI SALO
    DYLAN OLSEN
    JOHN MOORE
    RYAN STANTON
    NICK LEDDY
    MATT GREENE
    CLAYTON STONER

    2014-15
    BEN CHIAROT
    RADKO GUDAS
    NICK LEDDY
    BEN LOVEJOY
    KEVAN MILLER
    DAVID SCHLEMKO
    NICKLAS GROSSMANN
    JORDAN LEOPOLD
    MARK FAYNE
    MATT IRWIN
    STEPHANE ROBIDAS
    MIKE WEBER
    XAVIER OUELLET
    ROBERT BORTUZZO
    MATT HUNWICK
    BARRET JACKMAN
    TIM GLEASON
    MARK BOROWIECKI
    RYAN STANTON
    NATHAN BEAULIEU
    NATE GUENIN
    VICTOR BARTLEY
    PETER HARROLD
    MATT GREENE
    DYLAN OLSEN
    NATE SCHMIDT
    TIM ERIXON
    DERYK ENGELLAND
    MATT DUMBA

    There were also questions about QoC. Here are the group averages.

    Rank OppGF60 OppGA60 OppGF%
    D1 2.294 2.262 50.349
    D2 2.303 2.258 50.492
    D3 2.276 2.261 50.166
    D4 2.251 2.256 49.947
    D5 2.230 2.260 49.664
    D6 2.210 2.267 49.374
    Depth 2.192 2.247 49.391

    QoC drops as you go down the list but the differences are relatively minor. ~0.1 GF/60 spread.

    Here are D6 QoC by HC, MC and LC teams.

    OppGF60 OppGA60 OppGF%
    HC D6 2.212 2.287 49.19
    MC D6 2.205 2.269 49.29
    LC D6 2.218 2.240 49.76

    almost no difference.

    Here are some zone start stats by group.

    Rank DZFO% OZFO%
    D1 31.4 29.7
    D2 31.1 29.5
    D3 31.0 30.9
    D4 31.4 31.0
    D5 31.0 31.4
    D6 30.7 32.6
    Depth 29.6 34.3

  2. These are the D6 players from High Corsi teams. This is the best group at boosting Sv%.

    2011-12 ADAM MCQUAID
    2011-12 STEVE MONTADOR
    2011-12 JAKUB KINDL
    2011-12 MATT GREENE
    2011-12 MATT NISKANEN
    2011-12 KRIS RUSSELL
    2011-12 AARON ROME
    2013-14 KEVAN MILLER
    2013-14 NICK LEDDY
    2013-14 MATT GREENE
    2013-14 JON MERRILL
    2013-14 JOHN MOORE
    2013-14 ERIC GRYBA
    2013-14 SCOTT HANNAN
    2013-14 JORDAN LEOPOLD
    2013-14 RYAN STANTON
    2014-15 KEVAN MILLER
    2014-15 TIM ERIXON
    2014-15 DAVID SCHLEMKO
    2014-15 XAVIER OUELLET
    2014-15 MATT GREENE
    2014-15 VICTOR BARTLEY
    2014-15 NICK LEDDY
    2014-15 ROBERT BORTUZZO
    2014-15 RADKO GUDAS
    2014-15 BEN CHIAROT

  3. Hi David, great work. I was wondering, did you look at scoring chance data or blocked shots in your analysis? Your conclusion that 6th defenseman show a knack for suppressing goals despite high corsi against totals got me thinking about shot quality. Do these players play a more conservative style — allowing more attempts from outside scoring areas — in favor of locking down dangerous areas? The data also reflects these players are largely poor at creating offense for their teams, which could support this theory. Curious to hear your thoughts and see the forward data. I wonder if any of those findings will inform these better.

Comments are closed.