The State of NHL Microstats

By CJ Turtoro

To do list:

• Brief History of Microstats
• What We Know
• Right Now
• Player Evaluation
• Repeatability and Predictivity
• Future
• Research: Context
• Usage: Tactics

Who am I, And What are Microstats?

What is a CJ?

What Are Microstats?

• Formally: Records of all events that are involved in the progression from a defensive possession to an offensive shot attempt.�
• Informally: Any stat the the NHL doesn’t track�
• Examples
• Zone Exits
• Zone Entries
• Entries Defended
• Shot assists

*Note* All the above example are “individual” metrics

​

Entry Defense

Zone Exit

Zone Entry

Shot Assist

Shot Attempt

What We Know

State of Current Research

What Do We Know About Microstats In General?

• Carrying creates over twice as many shot attempts as dumping (Tulsky 2011, 2012) (Luszczyszyn 2015)�
• About a quarter of goals come after failed exits and over 40% of all goals come from “poor” exits (Novet 2018)�
• xG using pre-shot passing info is the most powerful expected goal model we have to-date (Stimson 2017)

What Do We Know About Microstats In General? (continued)

• Entry Defense and Exit vs Shot and Goal stats (O6 Analytics 2017)
• Weighted Entries (O’Connor 2017)
• No relation in entry generation vs supression (Novet 2016)
• Zone scores (Tulsky 2013)
• Ryan Stimson’s life, basically
• Browse his Hockey-Graphs author page for a start

Let’s look at all the playersssss

Right Now (part 1)

Player Evaluation

#SeemsLegit

Eller

RelT CF%: +0.18

RelT xGF%: -3.81

RelT GF%:-14.69

​

JvR

RelT CF%: +7.15

RelT xGF%: +7.99

RelT GF%: +1.79

​

#High5ebastion

5ebastion

RelT CF%: -0.31

RelT xGF%: -7.58

RelT GF%:-4.82

​

Zdeno

RelT CF%: -0.19

RelT xGF%: +0.17

RelT GF%: -0.06

We Figured out the Missing Link to Player Evaluation!

...Okay not really.

Right Now (part 2)

Repeatability & Predictivity

​

​

With data from Corey Sznajder and OffsideReview

DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA DATA

Statistics Used

• Stimson Stats
• Shot Assists
• Shot Contributions
• Entries
• Controlled Entries per 60
• Controlled Entry %
• Controlled Entry % (w/fails)
• Exits
• Controlled Exits per 60
• Controlled Exit %
• Exit Success %
• Entry Defense
• Controlled Entries Against per 60
• Controlled Entry % Against

• Traditional On-Ice Metrics
• CF%
• xGF%
• GF%

Questions #1

1. When do the A3Z “stabilize”?

​

• Are they “repeatable”?

When do they stabilize?

Forwards

Defencemen

Pearson Correlation (r)

When do they stabilize?

Forwards

Defencemen

Pearson Correlation (r)

When do they stabilize?

Forwards

Defencemen

Pearson Correlation (r)

When do they stabilize?

Forwards

Defencemen

Pearson Correlation (r)

When do they stabilize?

Forwards

Defencemen

Pearson Correlation (r)

Answers #1

• When do the A3Z stats “stabilize”?

Defencemen see stats stabilize around 20 games, Forwards might need 30+

• Are they “repeatable”?

Yes! Almost all of the A3Z stats are more repeatable than traditional on-ice metrics -- especially zone entries.

Questions #2

• Are A3Z stats predictive of CF%, xGF% or GF% for forwards?�
• Are A3Z stats predictive of CF%, xGF% or GF% for defencemen?�
• Which A3Z metrics are most helpful in predicting future events?�
• Are teams/scheme/usage important?

Maximum Predictivity by Team

Maximum Predictivity by Position

Defenders Forwards

Maximum Predictivity by Position (Rel, Adj.)

Defenders Forwards

Answers #2

• Are A3Z stats predictive of CF%, xGF% or GF% for forwards?

There is a signal there, but existing on-ice metrics are better.

​

• Are A3Z stats predictive of CF%, xGF% or GF% for defencemen?

A3Z metrics actually outperformed traditional on-ice metrics at predicting future on-ice results

​

• Which A3Z metrics are most helpful in predicting future events?

Exits>Entries for predicting future On-Ice results,

Entries>Exits for predicting future Rel results.�

• Are teams/scheme/usage important?

YES!!!

WE MADE IT!!

The Future of Microstats

More Research

• A3Z On-ice and Rel Stats
• Linemate Impact
• Special Teams
• Team-Specific Work
• Scheme/Tactics
• Tape-to-Tape Tracking Project

NHL Tracking

• NFL: Next Gen Stats
• NBA: Second Spectrum
• MLB: Statcast
• NHL: Derp

Concluding Thoughts ...

1. A3Z might be helpful for defender evaluation. It’s still early, but outperforming on-ice metrics convincingly is a good start.

​

• Exits/Entries and per60/% all have use

​

​

• There’s a lot of work to do

THE END