Having 3 corners on the field is now the norm in the NFL. Since many corners get significant playing time, I wanted to see if I could group together corners by their performance during the 2020 season. In this article I use a K-Means Clustering to group together corners with similar statistics. This was inspired by Arjun Menon’s articles clustering NFL RBs and WRs which can be found here Clustering 2020 NFL Running Backs, Clustering 2020 NFL Wide Receivers

The K-Means Clustering process includes taking provided inputs, scaling them, and then creating clusters that minimize distance between points within clusters, while maximizing the distance between cluster centers. To learn more about K-Means , I highly recommend new Chargers Analyst Alex Stern’s tutorial https://alexcstern.github.io/hoopDown.html

Using data from Pro Football Focus, I clustered corners that had at least 50 snaps in both man and zone coverage in the 2020-21 season using the following features.

  • Man Coverage % (Man %) – Percentage of coverage snaps in man
  • Man Coverage Snaps per Target (Man S/Tgt)
  • Forced Incompletion Rate in Man Coverage (Man FINC%)
  • Passer Rating Against in Man Coverage (Man QBRating)
  • Zone Coverage Snaps per Target (Zone S/Tgt)
  • Forced Incompletion Rate in Zone Coverage (Zone FINC%)
  • Passer Rating Against in Zone Coverage (Zone QBRating)
  • Slot % – Percentage of coverage snaps played from the slot
  • Missed Tackle Rate (Miss TKL)
  • PFF Run Defense Grade (Run DEF)
  • PFF Penalty Grade (PEN)

Man % and slot % give us an idea of a player’s role. Coverage Snaps per Target tell us how often a defender is targeted. Generally if a defender is doing his job and playing tight coverage they will not be targeted often. FINC% gives us an idea of how a defender performs when they are targeted. FINC% measures how often a defender causes an incompletion due to an interception, pass breakup, or tight coverage when targeted. One of the most important jobs for a corner is to limit big plays, and Passer Rating Against is a good indicator of whether corners are allowing big plays when targeted.

From this data I was able to create 11 clusters.

*Note that Man and Zone QBRating have inverted scales – so higher in the graph represents lower/better Passer Rating allowed.
*Note that Cluster 5 had a Man S/Tgt center of 7.5. I zoomed so that it would be easier to compare the rest of the data.

Cluster 1 – Low Tier Slots

  • Worst zone coverage snaps per target
  • Worst zone FINC%
  • High Slot%
  • Highest Missed Tackle Rate
  • Low Run DEF grade

On average corners in this cluster have solid stats in man coverage but struggle in zone coverage and in run defense. Most of the players in this cluster play in the slot above 65% of plays in coverage. Cameron Sutton and Kevin Johnson are slight exceptions, playing just over 50% of their snaps in the slot.

Closest fit (Least distance to center/average of cluster) – Darnay Holmes

Worst fit – Blidi Wreh-Wilson and Justin Coleman

Cluster 2 – Physical Man Corners

  • High Man%
  • Good Man Passer Rating Against
  • Low Zone Coverage Snaps per Target
  • Low Zone FINC%
  • Good Zone Passer Rating Against
  • High Run DEF grade
  • High Penalty grade

Cluster 2 contains some high end corners like Marlon Humphrey, Bryce Callahan, J.C. Jackson, and William Jackson. On average, corners in this cluster play a high amount of man coverage and are great at limiting big plays when targeted. The cluster averages for passer rating allowed when targeted in man and zone respectively are 71.8 and 78.8. This cluster is also fundamentally sound with high run defense and penalty grades.

Closest fit – Darqueze Denard

Worst in cluster fit – Jeffrey Okudah

Cluster 3 – All Around Outside Corners

  • Above average Man Coverage Snaps per Target
  • Above average Man Passer Rating Against
  • Above average Zone Coverage Snaps per Target
  • Above average Zone Passer Rating Against
  • Low Slot %
  • Worst Penalty Grade

Cluster 3 has probably the best mix of good coverage stats in man and zone. The average zone cover snaps per target for the cluster is 10.2, good for 2nd best. However the tradeoff may be that this cluster also has the worst penalty grade by a wide margin with an average grade of 35. The players in this cluster mostly check out, Marcus Peters, Tre’Davious White, and Xavien Howard are among the best at their position. Marshon Lattimore has shown some high end play, and also is known to be prone to penalties, committing the 4th most out of corners with 11 penalties last season.

Closest fit – Steven Nelson

Worst fit – Marcus Peters

Cluster 4 – Playmakers

  • High Zone%
  • Highest Man FINC%
  • High Zone FINC%
  • Good Zone Passer Rating Against
  • Low Slot%
  • Good Penalty Grade

Cluster 4 contains a lot of solid corners who particularly excel making plays on the ball. This cluster has the highest FINC% in man coverage (19%) and impressively also excel at playing clean, with the best Penalty Grade out of any cluster. These corners on average excel in zone coverage and are put in situations to succeed often, playing zone for about 75% of their snaps. 

Best fit – Akhello Witherspoon and Jaire Alexander

Worst fit – James Bradberry

Cluster 5 – Jalen Ramsey

  • Lowest Man%
  • Off the charts Man Coverage Snaps per Target
  • Best Man Passer Rating Against
  • High Zone FINC%
  • Best Zone Passer Rating Against
  • Lowest Missed Tackle Rate

Jalen Ramsey’s greatness combined with a small sample of plays in man coverage causes him to break the scales in most man coverage stats. Ramsey posts a ridiculous man cover snaps per target of 19 and a passer rating allowed of 39.6 (3rd/129 included corners). The Ramsey cluster also has the best Passer Rating Against in zone coverage at 57.2. Ramsey truly is in a league, and cluster of his own.

Cluster 6 – Outside Corner Zone Specialists

  • Lowest Man Coverage Snaps per Target
  • Bad Man Passer Rating Against
  • Best Zone FINC%
  • Above average Zone Passer Rating Against
  • Low Slot %

Cluster 6 contains corners who play mostly from the outside and excel in zone coverage. This has been Josh Norman’s calling card his entire career playing in heavy zone schemes in Carolina, Washington, and Buffalo so it makes sense to see him in this cluster. 

Best fit – Breon Borders, Pierre Desir

Worst fit – Sydney Jones

Cluster 7 – Belichick Tree

  • Highest Man %
  • Low Man FINC%
  • Low Man Passer Rating Against
  • Highest Zone Coverage Snaps per Target
  • Low Zone FINC%
  • High Missed Tackle Rate

The Belichick tree cluster contains 5 of the 6 players from the Patriots, Dolphins, and Lions. All teams that were coached by Belichick or a former Patriots defensive coordinator. Belichick, Flores, and Patricia all target similar corners and utilize them similarly with league high amounts of man coverage. The system puts a lot of pressure on its corners and can expose bad corners (Desmond Trufant) while allowing others to excel (Stephon Gilmore). This cluster is grouped together mostly by the extremely high man coverage percent rather than the ability of the corners. Stephon Gilmore and Byron Jones may have had slight down years in 2020 with PFF grades of 58 and 61 respectively, but both have shown that they are capable of taking on the toughest assignments consistently. McCourty, Trufant, and Igbinoghene however all struggled greatly. Trufant and Igbinoghene were both bottom 10 at the position in PFF grade.

Best fit – Stephon Gilmore

Worst fit – Jason McCourty

Cluster 8 – Slot Prototypes

  • Highest Slot%
  • High Man Coverage Snaps per Target
  • High Run Defense Grade
  • Low Zone Coverage Snaps per Target
  • Low Man and Zone FINC%

These players all play the majority of their snaps from the slot and on average have the key skills teams look for in slot players. High man coverage snaps per target is particularly impressive from guys who play most of their snaps in the slot, since it is easier for slot receivers to get more separation due to more space and an unlimited route tree compared to outside receivers. Also slot corners often have more responsibility against the run, so this cluster’s high run defense grade is important to their success.

Best fit – Taron Johnson

Worst fit – Tramon Williams

Cluster 9 – Big Ten Cluster

  • Bad Zone Passer Rating Against
  • Bad Penalty Grade
  • Best Run Defense Grade
  • Above average Man FINC%
  • Low Man%

I named this group the Big Ten Cluster because their stats were boring, the corners are mostly bad to mediocre, but the cluster had the best run defense grade. Denzel Ward seems like the misfit because he is a pretty good corner. Also, Ward is the only player in the cluster who actually played in the Big Ten.

Best fit – Kevin King, AJ Terrell

Worst fit – Daryl Worley

Cluster 10 – Big Play Prone

  • Worst Zone Passer Rating Against
  • Worst Run Defense Grade
  • High Man %
  • Average to below average in almost all man/zone stats
  • High Penalty Grade

The Big Play Prone Cluster is about average in S/Tgt and FINC% in man and zone, but has the worst zone passer rating against and is below average in man passer rating against as well. These players can keep up with their man and make some plays but can be prone to gamble too much allowing big plays or committing penalties.

Best fit – Avonte Maddox

Worst fit – Lamar Jackson

Cluster 11 – Man Coverage Liabilities

  • Worst Man Passer Rating Against
  • Low Man%
  • Low Man FINC%
  • Low Man Coverage Snaps per Target
  • Below average Zone Coverage Snaps per Target
  • Below average Zone FINC%

The name of this cluster really sums it up. On average these players really struggled in 2020, especially in man coverage. Even Chris Harris Jr. who has been a top tier corner since entering the league in 2011 had struggles in man coverage last season, allowing a Passer Rating of 149.3 in man coverage, and compiling a 38.4 PFF grade in man coverage.

Best fit – Sean Murphy-Bunting

Worst fit – Damon Arnette and Chris Harris Jr.

Cornerback is a very volatile position and coverage statistics certainly are a step behind passing, receiving, and pass rushing metrics in terms of predictability and overall ability in capturing a players value. However I think the clusters are good general groupings of corners last season based on style and ability.

Thanks for reading! Feel free to send any critiques or follow me on twitter @joey_dicresce.