Fantasy Hockey Draft Kit
2023-2024
Prepared by:
Left Wing Lock, Inc.
P.O. Box 30131
Bethesda, MD 20824
Email: staff@leftwinglock.com
Web: https://leftwinglock.com
Most recent update: August 1, 2023
© Left Wing Lock, Inc. 2023. All Rights Reserved.
Contents
List of Figures
11
1 About This Document
15
1.1
Welcome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
1.2
Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
2 Draft Pick Value
16
2.1
Motivation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
2.2
How Much is the 2nd Overall Pick Worth? . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
2.3
The Draft Pick Value Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
2.4
Revisiting the Trade for 2nd Overall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
3 Auction Drafts
21
3.1
Auction Value Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
3.2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
3.3
Auction Drafts: Just Like Standard Drafts But With Money? . . . . . . . . . . . . . . . . . .
22
3.4
Strategy: Part I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
3.5
Strategy: Part II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
3.5.1
23
Generate Your Player Ranking List . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
CONTENTS
3.5.2
3.6
3
Generate Your Price Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
Tips and Tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
4 Important Trends in the NHL
25
4.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
4.2
Power Play Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
4.3
Power Play Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
4.4
Hitting is Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
4.5
Goal Scoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
4.6
Shots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
5 The Impact of 3-on-3 Overtime
29
5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
5.2
Number of Shootouts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
5.3
Goal Scoring Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
5.4
Impact on Fantasy Hockey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
5.4.1
30
Goals, Assists, and Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6 Mythbusters
32
6.1
The Sophomore Slump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
6.2
Playing for a Contract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
6.2.1
Unrestricted Free Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
6.2.2
Restricted Free Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
Goalies Are Good at the Penalty Kill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
6.3.1
Career Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
6.3.2
Can You Do It Again? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
6.3
CONTENTS
7 Shooting Percentage - Theory
4
40
7.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
7.2
A Quick Discussion on Coin Flips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
7.3
Applying Coin Flip Experiments to NHL Players . . . . . . . . . . . . . . . . . . . . . . . . .
42
7.4
Corey Perry’s 50 Goal Season . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
8 Individual Points Percentage - Theory
45
8.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
8.2
Case Study: Dougie Hamilton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
8.3
Case Study: Matt Duchene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
8.4
Case Study: Nicklas Backstrom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
8.5
Case Study: Alex Goligoski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
8.6
Case Study: Claude Giroux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
8.7
Case Study: Jiri Hudler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
8.8
Case Study: Eric Staal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
9 Assists: Theory
55
9.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
9.2
Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
9.3
Secondary Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
9.4
Primary Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
57
10 Projecting A Goalie’s Save Percentage
59
10.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
10.2 The Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
10.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
10.2.2 Looking at the League Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
60
CONTENTS
10.2.3 How to Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11 When Are We Sure of a Goalie’s Talent Level?
5
60
62
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
62
11.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
62
11.3 A Word of Caution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
12 The Repeatability of Fantasy Hockey Stats - Part I
67
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
12.2 Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
12.2.1 Conceptual Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
12.2.2 Mathematical Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
12.3 Putting It Together . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
13 The Repeatability of Fantasy Hockey Stats - Part II
70
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
13.2 Hits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
13.3 Blocked Shots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
71
13.4 Basic Scoring Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
72
13.4.1 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
72
13.4.2 Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.5 Power Play Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.5.1 Powerplay Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.5.2 Powerplay Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.5.3 Powerplay Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.6 Shorthanded Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
13.6.1 Shorthanded Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
CONTENTS
6
13.6.2 Shorthanded Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
13.7 Penalty Minutes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
13.8 Shots on Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
13.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
13.10 Important Note on Games Played . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
13.10.1 How Do Other Sites Do It? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
13.10.2 How Does Left Wing Lock Do It? . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
14 The Motivation for Enhanced Stats
79
14.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
14.2 A Simple Flip of a Coin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
14.2.1 10 Coin Flips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
14.2.2 100 Coin Flips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
14.2.3 1000 Coin Flips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
14.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
14.3 NHL Players as Coins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
14.3.1 Phil Kessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
14.4 Thoughts on Sample Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
83
14.5 The Motivation for Analyzing Shot Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
83
15 Getting to Know Enhanced Stats
85
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
15.2 Notation and Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
15.2.1 Simple Shot Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
15.2.2 Shot Attempts Differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
86
15.2.3 Unblocked Shot Attempts Differential . . . . . . . . . . . . . . . . . . . . . . . . . . .
86
CONTENTS
7
15.2.4 Fluctuations from League Average Performance . . . . . . . . . . . . . . . . . . . . . .
87
15.3 Assumptions of the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
88
15.3.1 EV, PP, and PK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
88
15.3.2 Score Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
88
16 The ± Statistic - Theory
90
16.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
16.2 Are Past ± Values Predictive? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
16.3 Projecting ± . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
91
17 Possession & Luck Charts
95
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
17.2 Four Types of Teams in the Pluck Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
17.3 Bubbles & Colors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
97
17.4 An Application: When to Trade a Hot Goalie . . . . . . . . . . . . . . . . . . . . . . . . . . .
97
18 Player Usage Charts
101
18.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
18.2 Description of the Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
18.3 Interpretation of the Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
19 The Relationship Between League Standings and Goal Differential
104
19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
19.2 The Standings vs. Goal Differential Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
19.3 How Many Goals Equal a Win? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
20 Anatomy of a Yahoo Pro League
107
20.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
CONTENTS
8
20.2 Description of Leagues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
20.3 Scoring Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
20.4 Traits of Winning Teams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
20.5 Traits of Average Teams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
20.6 Interpreting the Data
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
20.7 An FSI Draft Spreadsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
21 Fantasy Strength Index - FSI
113
21.1 What is it? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
21.2 How is it Created? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
21.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
21.2.2 How Do the Other Guys Rank Players? . . . . . . . . . . . . . . . . . . . . . . . . . . 113
21.2.3 Why is the FSI Better? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
21.3 How Do I Use it in My Fantasy Draft? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
21.3.1 Points Leagues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
21.3.2 Category Leagues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
21.3.3 A Real-World Fantasy Draft Example . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
21.3.4 FSI Spreadsheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
22 General Advice for Newcomers
118
22.1 Should I Draft Linemates? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
22.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
22.1.2 What Do We Advise? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
22.2 Is the Pre-season Important? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.2.1 Stats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.2.2 Injuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
CONTENTS
9
22.2.3 Line Combos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.2.4 Contract Holdouts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.3 What Happens After Age 27? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.4 Trading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.4.2 Don’t Send Insulting Trade Offers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
22.4.3 Trade Good Players to Teams That Don’t Need Them . . . . . . . . . . . . . . . . . . 120
22.4.4 Don’t Be Afraid to Overspend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
22.5 The Squeeze Play . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
22.6 How Do I Use My Bench Players? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
22.7 Types of Drafts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
22.7.1 Auto-Draft Advice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
22.7.2 Live-Draft Advice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
23 Using the Left Wing Lock Website
123
23.1 The Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
23.1.1 Starting Goalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
23.1.2 Line Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
23.1.3 Random Draft Order Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
23.1.4 Line Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.5 Line Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.6 News Feed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.7 Player Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.8 iPhone App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.9 Email Alerts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.10 Roster Maximizer
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
CONTENTS
10
23.1.11 Player Rankings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.12 Weekly Schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.1.13 Team Pages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
23.2 The Forum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
23.3 Site-wide Chat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
23.4 The Articles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
List of Figures
2.1
Draft Pick Value Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
4.1
Penalties Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
4.2
Power Plays Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
4.3
PPG Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
4.4
Hits Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
4.5
Goals Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
4.6
EVG Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
4.7
SOG Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
4.8
Blocked Shots Per Game (2013-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
5.1
OT & Shootouts (2011-present) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
6.1
Point Production in Sophomores vs. Rookies . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
6.2
Point Production in Non-rookies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
6.3
Goal Production by UFAs in Contract years . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
6.4
Point Production by UFAs in Contract years . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
6.5
Shot Production by UFAs in Contract years . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
6.6
Goal Production by Players Under the Age of 27 . . . . . . . . . . . . . . . . . . . . . . . . .
36
11
LIST OF FIGURES
12
6.7
Goal Production by RFAs in Contract years . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
6.8
Point Production by Players Under the Age of 27 . . . . . . . . . . . . . . . . . . . . . . . . .
36
6.9
Point Production by RFAs in Contract years . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
6.10 Shot Production by Players Under the Age of 27 . . . . . . . . . . . . . . . . . . . . . . . . .
37
6.11 Shot Production by RFAs in Contract years . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
6.12 Career Penalty Kill Save Percentage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
6.13 Change in PKSV% in Consecutive Seasons
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
39
7.1
The Normal Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
8.1
Dougie Hamilton (2012-2013 :: 2021-2022) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
8.2
Matt Duchene (2013-2014 :: 2020-2021) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
8.3
Nicklas Backstrom (2012-2013 :: 2019-2020) . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
8.4
Alex Goligoski (2011-2012 :: 2017-2018) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
8.5
Claude Giroux (2008-2009 :: 2016-2017) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
8.6
Jiri Hudler (2008-2009 :: 2014-2015) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
8.7
Eric Staal (2007-2008 :: 2015-2016) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
54
9.1
Year Over Year Secondary Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
56
9.2
Year Over Year Primary Assists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58
11.1 Simulation of 1000 Bad Goalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
11.2 Simulation of 1000 Average Goalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
64
11.3 Simulation of 1000 Good Goalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
11.4 Devan Dubnyk, Marc-Andre Fleury, and Michal Neuvirth . . . . . . . . . . . . . . . . . . . .
66
12.1 Scatter Plots for Stat X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
12.2 Scatter Plots for Stat Y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
LIST OF FIGURES
13
12.3 Stat X Data with Best Fit Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
12.4 Stat Y Data with Best Fit Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
13.1 Year-to-Year Hits Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
71
13.2 Year-to-Year Blocked Shots Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
72
13.3 Year-to-Year Goals Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.4 Year-to-Year Assists Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.5 Year-to-Year PPG Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
13.6 Year-to-Year PPA Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
13.7 Year-to-Year PPP Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
13.8 Year-to-Year SHG Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
13.9 Year-to-Year SHA Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
13.10Year-to-Year PIM Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
13.11Year-to-Year SOG Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
13.12Repeatability Data for Fantasy Hockey Stats . . . . . . . . . . . . . . . . . . . . . . . . . . .
77
14.1 Results of a computer simulation of 10 coin flips run one million times.
. . . . . . . . . . . .
80
. . . . . . . . . . .
80
. . . . . . . . . .
81
. . . . . . . . . . . . . .
82
14.5 Results of a computer simulation of 273 SOG run one million times. . . . . . . . . . . . . . .
83
14.6 Results of a computer simulation of 3,042 SOG run one million times. . . . . . . . . . . . . .
83
16.1 ± Data in Six Consecutive Seasons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
16.2 ± Data in Six Consecutive Seasons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
91
16.3 Relationship between future ± and past SHSV . . . . . . . . . . . . . . . . . . . . . . . . . .
92
16.4 Relationship between future ± and past SHSV . . . . . . . . . . . . . . . . . . . . . . . . . .
93
14.2 Results of a computer simulation of 100 coin flips run one million times.
14.3 Results of a computer simulation of 1,000 coin flips run one million times.
14.4 Results of a computer simulation of 60 SOG run one million times.
LIST OF FIGURES
14
17.1 Pluck Chart for the 2022-2023 NHL season . . . . . . . . . . . . . . . . . . . . . . . . . . . .
96
17.2 Pluck Chart: January 1, 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
99
17.3 Pluck Chart: April 9, 2017
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
18.1 Standard Player Usage Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
19.1 Relationship Between Standings Points and Goal Differential . . . . . . . . . . . . . . . . . . 105
20.1 Yahoo Fantasy Hockey - Rotisserie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
20.2 Yahoo Fantasy Hockey - H2H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
20.3 FSI Draft Spreadsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
21.1 2013-2014 Fantasy Hockey Draft Using FSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Chapter 1
About This Document
1.1
Welcome
the spreadsheets, please contact us so that we can
address the mistake quickly.
Thank you for purchasing the Left Wing Lock Fan- The easiest way to contact our team and receive a
tasy Hockey Draft Kit. You have placed a large response is to email us at: staff@leftwinglock.com.
amount of trust in us to help you prepare for your We are also available via phone at: 1-301-842-4370.
draft and we are committed to making you a fantasy
champion.
1.2
Two questions guide our team as we develop the annual draft kit:
Files
This draft kit is made up of three parts:
• will the information in this kit help you win?
• will the information in this kit make you a better
manager?
1. A pdf document (Theory) that lays out the theoretical underpinnings for our approach. Read
this if you want answers to questions that start
with how or why.1
Everything you find in this draft kit will address one
of the above two questions. Another guiding principle
of ours is that we believe in proof over guessing. We
theorize; we test; we test again. The information and
analysis that pass these tests makes it into the final
draft that you’ll spend your Summer reading.
2. A pdf document (Application) that applies all of
the theories to the 2023-2024 season. Read this
if you want answers to questions that start with
who or what.2
3. Spreadsheets filled with accurate statistical projections.
Your focus while reading this draft guide should be
on understanding and applying the conclusions of our
work. If, at any time, the details get in the way You are currently reading the Theory part of the Left
of that goal, please contact us. We love discussing Wing Lock Fantasy Hockey Guide.
fantasy hockey and we’d be happy to elaborate on
1 For example: why are goalies with great penalty kill numany of the ideas expressed in this document.
Finally, if you come across an error in this document or a number that doesn’t quite look right in
bers expected to regress the following season?
2 For example: who are the players expected to have rebound seasons in 2023-2024? What line combinations will the
Philadelphia Flyers use in 2023-2024?
15
Chapter 2
Draft Pick Value
2.1
Motivation
One of the goals of this annual draft kit is to tackle questions about fantasy hockey that seemingly have no
answer (or at least, no answers that have been published in the past). One of these questions that remains
unanswered involves determining the value of a draft pick.1
Draft pick value charts have been created in professional sports, with the most well-known example being
former Dallas Cowboys coach Jimmy Johnson’s NFL draft pick value chart.2 Eric Tulsky, now of the Carolina
Hurricanes, created a version for the NHL.3
These are interesting examples, but none of them are useful to us as fantasy hockey managers. Given that
all (or most) players are available to us in a fantasy draft, a 2nd round pick in fantasy hockey is remarkably
different from a 2nd round pick in real-life hockey. For this reason, we have set about to create the first ever
fantasy hockey draft pick value chart.
2.2
How Much is the 2nd Overall Pick Worth?
One of the reasons we were interested in creating this chart stemmed from the wide range of values that
fantasy hockey managers assign to draft picks. A simple question (on the surface) was posed to fantasy
managers and asked them what combination of draft picks would need to be exchanged in order to receive
the 2nd overall pick in return. Table 2.1 reveals a sample of just how differently fantasy managers value that
2nd overall pick.
1 We
are specifically talking about draft picks in fantasy hockey.
2 http://www.sbnation.com/nfl/2016/4/28/11494150/nfl-draft-trade-value-chart-explanation-history
3 http://www.broadstreethockey.com/2013/4/25/4262594/nhl-draft-pick-value-trading-up
16
CHAPTER 2. DRAFT PICK VALUE
17
Table 2.1: Suggested Trades for 2nd Overall Pick
Trade Scenario
A
B
C
D
E
F
G
H
Receive
2nd
2nd
2nd
2nd
2nd
2nd
2nd
2nd
overall
overall
overall
overall
overall
overall
overall
overall
Send
pick
pick
pick
pick
pick
pick
pick
pick
1st rounder, 2nd rounder
1st rounder, 3rd rounder
1st rounder, 4th rounder
1st rounder, 5th rounder
1st rounder, 7th rounder
1st rounder, 8th rounder
1st rounder, 10th rounder
2nd rounder, 4th rounder, 10th rounder
Most of the trade offers suggested begin with a swap of 1st round picks and then add in a second draft pick
on top of that. That extra pick is, of course, what this trade hinges on and it takes on many values ranging
from a 2nd round pick all the way up to a 10th round pick. In at least one trade suggestion, the swapping
of 1st round picks was not included.
With this wide range of subjective value offered up in hypothetical trades, just how is a fantasy manager to
know how much that 2nd overall draft pick worth?
2.3
The Draft Pick Value Chart
Inspired by the draft pick value charts of the NFL and the NHL, we set about to create the first ever fantasy
hockey draft pick value chart.
To accomplish this, we simulated several years of fantasy hockey drafts. We used historical average draft
position (ADP) values from each season to determine when a typical manager would take a player off the
board during the draft. We also made sure our simulation followed logical roster building rules (e.g., draft
starting lineup first, make sure each roster has the right number of players at each position, and so on).
We also used data from thousands of fantasy hockey leagues to determine the odds of each position being
selected in each draft spot (e.g., what are the odds that a fantasy manager selects a defenseman with the
7th overall pick).
We ran these simulations thousands of times. We then compiled the results of all of these drafts. We wanted
to know things like how many goals (on average) does the 17th overall pick score, what is a typical GAA of
a goalie taken with the 64th overall pick, and how many power play points are scored by the player taken
3rd overall. Since we knew exactly which player was chosen in every pick of every draft of every simulation,
we were able to compute these stats with relative ease.4
Once we determined the answers to these questions (and many, many more), we were able to determine the
impact (quantitatively) of each draft pick in a fantasy hockey draft. The impact of these draft picks was
wrapped up into a single number which was then scaled so that the first overall pick is always worth 1000.
4 Here,
of course, we used real-life statistics of these players.
CHAPTER 2. DRAFT PICK VALUE
18
Figure 2.1 is a plot of our results.
Figure 2.1: Draft Pick Value Chart
Using this chart is fairly straightforward. You find the value of each draft pick involved in the trade and add
up the values on each side of the trade. Going back to the original question of this chapter, we find that the
2nd overall pick has a value of 860. So, in order to make a fair offer to trade for the 2nd overall pick, you
need to make sure the picks you offer in return add up to something close to 860.
Reading numbers off of the plot is a bit difficult, so we’ve provided a table of data to assist you (see Table
2.2). Keep in mind that this is not a perfect model, but more of a guide to help you understand trade value
of draft picks.5 You might also wonder why 12th round picks have virtually no value in this model. The
answer is because 12th round pick players in fantasy hockey are easily replaceable on the waiver wire.
5 We used a standard 12-team default league in Yahoo as our guide in these simulations. A league with more scoring
categories might see a more severe slope to their plot. A league with more physical categories would see a less severe slope.
CHAPTER 2. DRAFT PICK VALUE
19
Table 2.2: Draft Pick Value Chart
2.4
Pick Number
Value
Pick Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
1000
860
779
721
676
639
608
582
558
537
518
500
484
469
455
442
430
418
408
397
387
378
369
360
352
344
337
330
322
316
309
303
296
290
285
279
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
Value
273
268
263
258
253
248
243
239
234
230
225
221
217
213
209
205
201
197
194
190
186
183
180
176
173
170
166
163
160
157
154
151
148
145
142
139
Pick Number
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
Value
136
134
131
129
126
123
121
118
116
113
111
108
106
104
101
99
97
95
92
90
88
86
84
82
80
77
75
73
71
69
67
65
64
62
60
58
Pick Number
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
Value
56
54
52
51
49
47
45
44
42
40
38
37
35
33
32
30
28
27
25
24
22
21
19
18
16
14
13
12
10
9
7
6
4
3
1
0
Revisiting the Trade for 2nd Overall
With our draft pick value chart in hand, we can now evaluate the sample trade offers for the 2nd overall
pick suggested to us by fantasy hockey managers. In Table 2.3, we sum the draft pick trade values in each
CHAPTER 2. DRAFT PICK VALUE
20
scenario.6 For the sake of discussion, we’ll assume we start out in draft position number 6 (you can repeat
the calculation for any other draft position you choose).
Table 2.3: Trade Evaluation
Trade Scenario
Value Received
Value Sent
A
B
C
D
E
F
G
H
860
860
860
860
860
860
860
860
1047
955
873
836
762
731
684
696
Table 2.3 gives us a rough guide that lets us know what a reasonable offer would look like for the 2nd overall
pick in the draft (if you held the 6th overall pick). You come out ahead in offers D, E, F, G, and H. Offer
C appears to be the most fair offer for both teams. Offers A and B would be considered a loss for you.
6 Again,
this is just simple addition of all draft pick trade values involved in the trade.
Chapter 3
Auction Drafts
3.1
Auction Value Tool
3.2
Update (August 16): We’ve added an
Auction sheet to the main spreadsheet.
If you’re in an auction-style league, consider using our
Auction Value Tool found in the “Auction” tab of the
main spreadsheet. This tool will complete all of the
calculations for you that are described in this chapter.
The entire process can be completed in seconds!
To use the tool, do the following: (1) enter the maximum budget allowed (e.g., 300) for your team in cell
H1 of the Auction tab; (2) enter the total number of
teams in your league (e.g., 12) in cell H2 of the Auction tab; (3) enter the total number of roster players allowed on a single team (e.g., 15) in cell H3 of
the Auction tab. After completing these steps, you
will notice that other cells in the Auction tab will
auto-fill with data. We recommend that you save the
file at this time. Then, go back to the “All” tab in
your spreadsheet. You will notice that the very last
column to the right labeled as LWLAUC will now
present you with the suggested auction values for every player. This column is typically found near column (DT) for Category Leagues and (CN) for Points
Leagues.
Introduction
Most people that are introduced to fantasy sports
spend the majority of their time playing in leagues
that employ the standard draft format. In the standard draft format, an ordered list of managers is
generated (either randomly or by way of rule) and
each manager chooses a player from a list of available
draftees until all rosters are completely full. Aside
from the draft order selection process, the only other
real tweak is whether your league chooses to use a
snake draft (the managers pick in reverse order in
even-numbered rounds) or a linear draft (the managers pick in exactly the same order every round of
the draft).
While many of you are veterans when it comes to
standard drafts in your fantasy leagues, one type of
draft that is less familiar is that of the auction draft.
If you’ve ever lamented being granted the 7th overall
pick (or worse) and realized you can’t own your favorite superstar player (Ovechkin, Crosby, Stamkos,
e.g.), an auction draft may hold particular value to
you.
In an auction draft, each team is assigned a budget
(predetermined according to your league’s settings).
You use this budget to acquire a full roster of players for your fantasy hockey team. An ordered list of
managers is generated (either randomly or by rule)
so that the auction process has structure. The first
team on the list will initiate the auction by nominat-
21
CHAPTER 3. AUCTION DRAFTS
22
ing a player (any player) to be auctioned (this action
is equivalent to this manager bidding on this particular player). The initial bid is typically entered as $1.
At this point, all managers in the league may now
enter bids on this nominated player. The bidding
continues until no manager is willing to outbid the
highest current bidding manager. At this time, the
highest bidding manager now owns that nominated
player. And this manager’s budget is now equal to
the original budget minus however much he spent to
acquire this first nominated player.
shocked to see Nazem Kadri or Elias Lindholm or
Martin Jones be the first player nominated in your
auction draft.
The process continues much like this until all managers have a complete roster. Want to own Alex
Ovechkin and Sidney Crosby this year? Go for it;
you’re only limited by the amount of money in your
budget. Tired of fretting over which round to draft
your first goalie? Tired of the manager in the 6th
slot always seeming to steal the player you wanted
to draft in odd-numbered rounds? Then an auction
draft might be just the thing you’re looking for.
Can I really own Ovechkin and Crosby? Sure,
why not. In one particular auction draft that the
Left Wing Lock staff participated, a manager outbid the league on John Tavares, Patrick Kane, Alex
Ovechkin, and Steven Stamkos.1 That particular
manager used $201 of his $300 budget to acquire
these players. He then had $99 to fill out the 16
remaining spots on his roster. That’s the beauty of
an auction draft: you control which players will be
on your roster.
3.3
Every manager will have two roster players after two
rounds, right? Nope, not very likely. Auction drafts
are strange beasts and you’ll see managers have five
roster players before you even have one; or maybe
you’ll be the manager with five players while others
have an empty sheet. Anything goes in auction drafts
- as long as you still have money in your wallet.
Auction Drafts: Just Like
Standard Drafts But With 3.4
Money?
Strategy: Part I
As you learn more about auction drafts, you’ll hear
about everyone’s favorite strategies. Some managers
don’t like to pay for goalies, so they wait until the
end of the draft and grab a couple at the $1 price
point. Some managers want to own several superstar
players, so they’ll pay heavily for 3-4 of these types
and then spread out the remaining budget on latedraft grabs. Some managers avoid superstar players
entirely and create a roster of players that are just
good enough.
While explaining how an auction draft works is fairly
straightforward, successfully drafting during an auction draft is another matter. Most of what you know
about fantasy drafting you probably learned in standard drafts. For example, you probably show up to
your standard drafts with a spreadsheet of projections and player rankings (go Left Wing Lock!) ready
to grab the best players in the early rounds. You’ve
determined what rounds you’ll target goalies and you
know just how long you can wait until the elite deAll of these approaches have their pros and cons, but
fensemen are off the board.
we’d like to provide a different angle in this draft kit.
Our auction draft strategy for you can be summed
Forget most of what you’ve learned.
up in two words: don’t overpay.
In an auction draft, most of your typical strategies
1 Ok, so he didn’t own Ovechkin and Crosby, but you get
do not apply. Think Alex Ovechkin or Sidney Crosby
will be drafted first overall? Think again. Don’t be the idea.
CHAPTER 3. AUCTION DRAFTS
3.5
Strategy: Part II
23
• T - number of teams in your league
• R - number of players rostered by each team
Hidden within that simple, two-word approach is the
• A - available budget
fact that you must be willing to prepare for your auction draft. You cannot simply show up with a ranked
• Q - available bidding money for the league
list of players and think you’ll conquer the world. You
won’t; and the results might be painful. You need to
spend an afternoon, or an evening, preparing specif- Most of the variables above are self-explanatory, but
ically for the auction portion of the draft and the we’ll cover a few of the items that are not obvious.
payoff will be a playoff berth.
Since you must keep at least $1 available for every
roster spot not yet drafted on your team, your availThe following describes an approach to auction draft able budget (A) is always less than your maximum
strategy that should keep you from overpaying on any allowable budget (M ). Your available budget can be
one player during an auction.
thought of as the following:
3.5.1
Generate Your Player Ranking
List
The first step in preparing for your auction draft is
to create a list of ranked players. In a points style
league, this would mean your list is sorted by projected fantasy points (according to your specific scoring system). In a categories or rotisserie league, this
would mean sorting your list by how much each player
contributes overall to winning in your league (later in
this draft kit, we describe such an approach called the
Fantasy Strength Index, or FSI).
A = M − (R × $1)
(3.1)
The available bidding money for all teams (Q) can
be computed as:
Q = T × A = T × (M − R)
(3.2)
With those definitions out of the way, let’s set about
determining bid values. First, you need to determine
the baseline fantasy value (in fantasy points or FSI
value). There will be a total of T × R players drafted
in your auction league. You’ll want to go through
your list of ranked players (sorted by fantasy points)
and determine how many fantasy points the T × R
3.5.2 Generate Your Price Sheet
ranked player is projected to generate. Let’s label
that number as the baseline fantasy production, or
The next step is to formulate a price that you are
BFP.2
willing to play for each player in the draft.This step
is critical because how else would you know if you are With the baseline number (BFP) in hand, you’ll want
overpaying for a player. How you come up with these to create a new column in your spreadsheet. Label
prices is up to you, but we recommend the following this column Y . The value for Y is computed as folmethod.
lows:
First, let’s assign a bunch of labels to the numbers
we’ll be discussing in this method.
Y = F SI − BF P
2 This
(3.3)
number will be different for every fantasy league.
• M - maximum allowable budget (this is your This number might be 18 fantasy points or 212 fantasy points
depending on your point structure.
team’s budget, $200 for example)
CHAPTER 3. AUCTION DRAFTS
24
Essentially, this creates a metric in which the least
valuable player that gets drafted in your auction has
an Y value of zero.
Next, you will sum all of the Y values together into
one number. We’ll call that number F and it would
be computed as:
F =
N
X
Y
(3.4)
i=1
Again, this is something Excel does very easily; it
simply adds up all of the Y values for the players
within your specified range (that should include players ranked from 1 through T × R).3
Next, make a quick calculation using your calculator.
Determine the value of λ in this formula:
Q
λ=
F
(3.5)
Here, λ tells you how much money each fantasy point
is worth in your league. It is one of the most important pieces of your preparation for your auction draft.
Finally, make another column in your spreadsheet.
Label this column as BV (bid value) This will tell
you approximately how much to bid on every player
in your spreadsheet. To compute this value, use the
following formula:
BV = Y × λ
3.6
Tips and Tricks
Below you’ll find some advice to help you stay on
track while you’re in the middle of your auction draft.
1. Don’t Want - Don’t Bid - this one sounds
simple enough, but make sure you follow it. Do
not bid on any player that you do not want on
your roster. You might win!4
2. Track Opponent Rosters - to fully understand your ability to win a bid on a particular
player (especially later in the auction) it is critical that you track the remaining available budget
of every roster. if you know exactly how much
money your opponents have left (and how many
players they still need to draft), then you know
exactly how high you’ll need to bid on players
each step of the way.
3. Stay Out of Bidding Wars - you’ll learn early
that bids on some players will grow at almost
immeasurable rates. Before you finish blinking,
seven more bids have jacked the cost up on a
player by another $14. It can become a psychological game of out-bidding the other manager and before you know it, you’ve spent way
more on Player X than your spreadsheet allotted. When Player X reaches your BV value, it’s
getting near that time to walk away.
(3.6)
By now, your spreadsheet should have two new
columns in it: a Y column which serves as an adjusted FSI column and a BV column which serves as
a guide to limit how much you spend on each player
in the draft.
3 For
brevity, we called T × R just N in our formula.
4 And by win, we mean lose since you didn’t want this
player in the first place.
Chapter 4
Important Trends in the NHL
4.1
Introduction
obstruction-free type of hockey game.1 But, the end
result is that power play opportunities are down from
15 years ago and have remained flat for the past
After the 2004-2005 lockout, there was an upward decade. This has important consequences for you as
spike in the number of obstruction-related penalty a fantasy hockey manager.
calls made in NHL games (see Figure 4.1). The increase in penalties resulted in an increase in power The goal of this chapter, then, is to explore the implay opportunities, which in turn, led to an increase pact of this drop on how you draft as a fantasy hockey
in goal scoring.
manager and how you oversee your league as a fantasy hockey commissioner.
4.2
Power Play Opportunities
One of the obvious results that stems from a decrease
in penalty calls is a decrease in power play opportunities. But seeing that in printed words is not nearly
as powerful as seeing the data represented graphically. Figure 4.2 reveals the massive drop in power
play opportunities for NHL teams since the 2007-2008
season.
Figure 4.1: Penalties Per Game (2013-present)
While the trend has leveled off in recent seasons, the
overall drop amounts to nearly 50% over the past 15
years. This drop impacts fantasy hockey in ways you
may not have imagined.
With power play opportunities down, the fraction of
This increase in goal scoring (due to the increase points scored by NHL players that come from the
in penalties) would be short-lived. It’s debatable as power play is also down. This puts a premium on
to whether the officials slowed their rate of penalty
1 The reality is almost certainly a blend of these hypotheses.
calls or if the players adapted and are playing an
25
CHAPTER 4. IMPORTANT TRENDS IN THE NHL
Figure 4.2: Power Plays Per Game (2013-present)
26
Figure 4.3: PPG Per Game (2013-present)
players who are able to score while playing 5-on-5 4.4
Hitting is Up
hockey. Without a mandate from the NHL, there is
no expectation that power play scoring will rise in
A somewhat confounding result of less penalty calls
the near future.
in the NHL is the sharp increase in the number of
This drop in power play point generation is also im- hits that are laid out in NHL games (see Figure 4.4).
portant to fantasy hockey commissioners. Unless
your scoring settings were developed in the past few
years, then your original design has been impacted
significantly. We recommend that commissioners of
leagues older than 5-6 years, consider increasing the
weight of power play categories compared to where
you had them at the league’s onset.
4.3
Power Play Points
To be clear, a drop in power play opportunities almost certainly means that overall power play points
scored by NHL teams will drop. The only way for
this not to be true would be if NHL teams immediately and inexplicably became significantly better at
Figure 4.4: Hits Per Game (2013-present)
scoring on the power play. This has not happened,
of course, and the result is that power play goals are
trending in exactly the same manner as power play
opportunities; that is, they are down from 15 years The reason for the increase in hits is simple, but not
ago and have stayed relatively flat in recent years.
at all obvious. NHL teams on the power play largely
CHAPTER 4. IMPORTANT TRENDS IN THE NHL
27
play a game of “keep-away” from the team that is
shorthanded. The opportunities for physical contact
are few in number. With the decrease in power play
opportunities, teams are spending more time playing
5-on-5 hockey. It is this increase in even-strength
hockey time that results in larger hit totals in the
NHL.
If you’re in a league that assigns points to each hit
laid out by an NHL player, then be aware that players who produce lots of hits have become more valuable in recent years (and be aware that this trend
is intimately linked to the number of penalties being called in the NHL and not to a “more physical”
NHL). Commissioners of older fantasy hockey leagues
should reassess the weights assigned to hits as they’ve
become more valuable since you first instituted your
scoring system.
Figure 4.5: Goals Per Game (2013-present)
4.5
Goal Scoring
If you replace power play time with even-strength
time, the overall result should lead to less goal scoring
in the NHL. This is a clear result from the fact that
teams score at a higher rate when they have the manadvantage as opposed to even-strength. But, this also
means that the number of even-strength goals is increasing.
Figure 4.6 reinforces the advice we lent in Section
4.2; that is, you’ll be better off in fantasy hockey if
you place an emphasis on drafting players who have
a strong history of scoring at even-strength.
4.6
Shots
Finally, two important fantasy hockey statistical categories that remain unchanged in the face of large
drops in penalty calls are shots on goal (SOG) and
blocked shots (BS).
Figure 4.6: EVG Per Game (2013-present)
CHAPTER 4. IMPORTANT TRENDS IN THE NHL
Figure 4.7: SOG Per Game (2013-present)
Figure 4.8: Blocked Shots Per Game (2013-present)
28
Chapter 5
The Impact of 3-on-3 Overtime
5.1
Introduction
clare a winner. About 13.5% of NHL games required
the shootout to declare a winner. The end result here
is that 10.5% of NHL games were decided using the
Following the 2014-2015 season, the NHL voted to 4-on-4 overtime format that the NHL used up until
approve a number of rule changes. The most im- the 2015-2016 season.
portant of these changes (for both hockey fans and
fantasy hockey managers) is the switch from a 4-on4 overtime format (followed by a shootout if necessary) to a 3-on-3 overtime format (again, followed by
a shootout if necessary). The duration of overtime
periods remains unchanged at five minutes.
The intent of the rule change is to reduce the number
of NHL games that are decided by a shootout. This
implies that the NHL finds the shootout to be an unfavorable ending to a hockey game. And thus, we’re
left wondering why there is a shootout at all. Alas,
this column is not about NHL politics, but instead
how you can use these rule changes to your advantage as a fantasy hockey manager.
Figure 5.1: OT & Shootouts (2011-present)
5.2
Number of Shootouts
We’ll begin the analysis by looking at how this rule
change impacted the number of games decided in
overtime. Figure 5.1 reveals the breakdown of NHL
games that required overtime and shootouts since the
2011-2012 season.
Since the rule change, the number of NHL games requiring overtime did not change, nor was this number expected to change.1 But, you can clearly see the
impact of the new rule in Figure 5.1. Up until the
2015-2016 season, about half of all games that went
1 There
was a small dip in the number of games requiring an
Prior to the rule change, about 24% of NHL games overtime, but it was well within the typical annual deviations
required overtime (and possibly a shootout) to de- from the norm.
29
CHAPTER 5. THE IMPACT OF 3-ON-3 OVERTIME
30
to overtime ended up in a shootout.2 This number and divide them by the 1047.5 minutes to arrive at
dropped to 39% after the introduction of the new 0.177 goals per minute.6
NHL overtime format.3
Going forward, you should use this 39% number as
your guide as opposed to the historical 56% number.
But what does that mean for fantasy hockey managers? Let’s find out.
5.3
5.4
Impact on Fantasy Hockey
Now that we know the 3-on-3 goal scoring rate for
the NHL, we can use this information to determine
the impact on fantasy hockey leagues.
Goal Scoring Rates
5.4.1
Goals, Assists, and Points
We can use the number of overtime games requiring a
shootout to work backwards and determine how often
goals are scored during the 3-on-3 overtime periods.
This is significant because 3-on-3 goal scoring rates
haven’t been known with any accuracy prior to the
2015-2016 season.4
The most clear path to take here is to consider how
many extra goals will be scored in the NHL in future seasons as a result of the increase in overtime
goals. Using 11-year averages in the NHL, we know
that there will be approximately 289 overtimes per
In 2022-2023, there were 302 games requiring an over- season. The 2022-2023 data suggests that 30% of
time. Of these 302 games, 117 of them also required these overtimes will lead to a shootout. That means
a shootout to determine the game’s winner. We can that 70% of these overtimes will end with an overtime
use this information to approximate the number of goal.
overtime minutes during the 2022-2023 season.
Prior to the 2015-2016 rule change, only 44% of overWe know that 117 of these overtimes lasted the en- times would end with an overtime goal. So, now we
tire five minutes. The remaining 185 of these over- have a path for figuring out how many extra goals will
times were decided by a goal that happened some- be scored in the new overtime system. We compute
where between zero and five minutes. If we make the how many overtime goals were scored under the old
assumption that these overtimes, on average, lasted system and subtract those from how many are scored
2.5 minutes, then we arrive at the number of overtime under the new system. The difference ends up being
minutes in 2022-2023; that number is 1047.5 minutes. about 64 extra goals scored.
We also know how many overtime goals were scored
in 2022-2023. There were 185 overtime goals.5 Now
it’s a simple task to estimate the goal scoring rate in
the NHL for 3-on-3 hockey. We take the 185 goals
2 The
actual number was 56%, on average
can compute these numbers by taking the value of
the black bar and dividing by the value of the orange bar.
4 In the 2015 draft kit, we published a range of numbers
provided by various NHL sources that spanned from 0.10 to
0.27. Our own internal estimate for 3-on-3 goal scoring at that
time was 0.171 goals/minute.
5 This has to be true since 185 NHL games that went to
overtime did not need a shootout.
Over the past four seasons, the average number of
goals scored in the NHL was 6,734 goals. Thus, the
increase in overall goal scoring (as a result of the increase in overtime goals) amounts to about 0.95%.7
Let’s round this up to 1% to make our analysis a little
bit easier.
3 You
How can you use this for your draft? The number
6 For reference, 5-on-5 goal scoring rate is about 0.075 and
the 4-on-4 goal scoring rate is about 0.09.
7 Last year’s seasonal number was 0.76%. Our theoretical
estimates in last year’s kit was 0.72%, so we’re feeling pretty
good now!
CHAPTER 5. THE IMPACT OF 3-ON-3 OVERTIME
of goals scored by a team in the NHL is 218 on average. And assists are distributed at a rather consistent
rate of 1.71 assists per goal. This means that a typical NHL team is awarded about 591 points (goals +
assists) in a season. What does a 1% increase look
like for these numbers? That means a team will earn
an extra 7 points on the season. Breaking this down
into goals and assists (on a per-team basis), we’ll see
an extra 2.6 goals and 4.5 assists awarded due to the
new overtime rules (some teams will get more than
this and some teams will get less, of course).
If you know which three players a coach will use in the
new overtime format, then you can consider giving
them a bump in goals, assists, and points.8 How
big of a bump? Well, three players may be splitting
around seven points. That’s not a lot. It would be a
mistake to expect a huge jump in NHL scoring levels
to result from the change in overtime rules.
8 In the team chapters of this document, we post the most
frequently used overtime line combinations by each NHL team.
31
Chapter 6
Mythbusters
6.1
The Sophomore Slump
Season), we’ve plotted the point production (measured in points per game) of every rookie who played
in at least half of his team’s games that season. The
One of the most frequently cited laws of hockey is vertical axis (labeled as Sophomore Season) reprethat rookies are prone to slumping in their second sents the point production of those same players durseason in the NHL. Reasons cited for the sophomore ing their 2nd NHL season. Each dot, then, represents
slump include: increased workload, added pressure, a single NHL player with the x-coordinate as his point
scouting reports, and scheme changes. You’ll find production in his rookie season and the y-coordinate
this notion of the sophomore slump on nearly every as his point production in his sophomore season.
hockey website you visit: ESPN1 , NHL.com2 , Yahoo3 , Bleacher Report4 , Sportsnet5 , and The Hockey
Writers6 to name a handful. With all of these websites and hockey experts discussing the topic, it has
to be true, right?
If you already know how our team functions, you
know we’re not going to accept something as fact
just because a lot of people are talking about it. Instead, why don’t we explore this idea of the sophomore slump by analyzing the data to find out if it’s
true?
Figure 6.1 is a plot containing four seasons worth of
data. Along the horizontal axis (labeled as Rookie
1 http://insider.espn.go.com/nhl/insider/story/_/
id/9833178/nhl-does-sophomore-slump-exist
2 http://www.nhl.com/ice/news.htm?id=681241
3 http://sports.yahoo.com/blogs/nhl-puck-daddy/
Figure 6.1: Point Production in Sophomores vs.
tomas-hertl-and-the-sophomore-slump-212302302.html
4 http://bleacherreport.com/articles/
Rookies
2194179-nhl-players-most-likely-to-have-a-sophomore-slump-in-2014-15
5 http://www.sportsnet.ca/hockey/nhl/
monahan-bulks-up-wants-to-avoid-sophomore-slump/
6 http://thehockeywriters.com/
We’ve added a straight line to the graph to indiis-the-sophomore-slump-real/
cate where point production in the sophomore season
32
CHAPTER 6. MYTHBUSTERS
33
matches identically that of the rookie season. Therefore, all players to the upper-left of the black line
improved during their sophomore season, while all
players to the lower-right of the black line slumped
during their sophomore season.
The results are mostly random; that is, there are
roughly an equal number of players that performed
better and players that performed worse. But, there
is something really interesting going on with that
graph: of all the rookies who scored at least 0.5 points
per game in their rookie season, about 80% of them
slumped! In order to see this better, hold a piece of
paper up to your screen and block out the left-hand
side of the graph (players who scored less than 0.5
points per game in their rookie season). See it now?
This unblocked data that you see is what makes people believe in the concept of the sophomore slump.
Figure 6.2: Point Production in Non-rookies
We’re obviously not going to stop here. This chapter
is called mythbusters for a reason!
If the sophomore slump really exists, then the phenomenon should be unique to sophomores. This
statement provides us with an approach to test
whether or not the sophomore slump really exists.
We will now examine point production data (over the
course of four years) for all NHL players who were
not rookies. This list of players will include players
in their 2nd, 9th, 14th, etc. seasons. We’ll do exactly
what we did with the rookie/sophomore graph. We’ll
plot point production for these players in one season
and then plot the point production from these same
players in the very next season.
This data also looks mostly random. Again, players
in the follow-up season seem to be split between performing better and worse than in the preceding season. But, let’s focus on the players who performed
at a high level in the preceding season. Take a piece
of paper and block out all data to the left of the 1.0
points per game marker. The players we see (the
unblocked players) are those that scored about 80
or more points in this particular season. Now, look
at how they are arranged vertically. If you count
the dots (and we did!), you’ll find that 80% of these
players suffered a drop in performance in the followFigure 6.2 is a plot containing four seasons worth of up year. These players slumped; just like the sophodata. Along the horizontal axis (labeled as Season mores!
N ), we’ve plotted the point production (measured in
points per game) of every non-rookie who played in Where does that leave us? The sophomore slump, as
at least half of his team’s games that season. The promoted by many hockey writers, simply does not
vertical axis (labeled as Season N+1 ) represents the exist. If you believe in a sophomore slump, then the
point production of those same players during the junior slump and senior slump and “X” slump also
very next NHL season. Each dot, then, represents a exist, because apparently most (80%) of the players
single NHL player with the x-coordinate as his point in the NHL that perform well in one season tend to
production in one season and the y-coordinate as his perform worse in the following season. This is not a
slump at all. Instead, it is simple regression to the
point production in his very next season.
mean and it affects all NHL players in every season.
CHAPTER 6. MYTHBUSTERS
6.2
34
Playing for a Contract
A frequently published piece of advise in fantasy
hockey circles is that managers should target players in the final year of their contract.7891011 The assumption driving this advice is that players perform
at a higher level than normal during their contract
year. But is this assumption valid? It is true that
player performance is driven by a desire for money?
To test this idea, we’ve decided to split the analysis into two phases: unrestricted free agents and restricted free agents.
6.2.1
Unrestricted Free Agents
A simple approach to testing whether or not unrestricted free agents (UFAs) play better in contract
years is to compare their production (points, goals,
assists, shots) during the contract year with their career averages. If the “contract year” adage is true, we
should see a clear and obvious trend in the data that
reveals a performance bump for UFAs during their
contract years.
Figure 6.3: Goal Production by UFAs in Contract
years
duction rate as during his career. This line makes
the analysis on your end easy: players above the line
had UFA seasons where they performed above and
beyond their career norms. Players below the line
had UFA seasons where they performed below their
career norms.
To start with, let’s take a look at goal production by The overall pattern here is that UFAs perform worse
UFAs during their contract years (as compared to the in their contract year than they do over the course
of their careers. That’s not particularly surprising
career average).
given that most UFAs are of an age that puts them
Figure 6.3 is shown in goals per game, so a value outside of their prime production years. In fact, age
of 0.2 represents about 16 goals over the course of a alone should have been an indicator to you that UFA
season (which would be a lower-limit on fantasy rel- performance does not increase during contract years.
evant players in most leagues). The horizontal axis
represents the goal production (per game) for play- Yes, there will be the occasional UFA that performs
ers during their careers, while the vertical axis repre- at a higher level than normal. T.J. Oshie is one such
sents the goal production (per game) for those same example and you can find him near the top of Figure
12
But a few outliers should be expected. Preplayers during their UFA season. We’ve added a di- 6.3.
dicting
who these outliers will be in advance of your
agonal line to the graph that represents a player in
fantasy
hockey season is next to impossible. In more
his UFA season performing at exactly the same procases than not, a UFA will produce fewer goals in
7 https://goo.gl/rx7EHn
his contract year than he has, on average, during his
8 https://goo.gl/1Cf3QH
9 https://goo.gl/CymClL
10 https://goo.gl/o3JAak
11 https://goo.gl/VAippH
12 Oshie’s goal production in 2016-2017 was boosted by an
incredibly lucky shooting percentage. He shot with a 23.1%
success rate - nearly 75% higher than his career average.
CHAPTER 6. MYTHBUSTERS
35
career.
To be a bit more thorough, we’ll also look at points
per game and shots per game for these same UFAs
during their contract years.
Figure 6.5: Shot Production by UFAs in Contract
years
ages and this should be the driving principle for you
as you consider drafting them.
Figure 6.4: Point Production by UFAs in Contract
years
6.2.2
Restricted Free Agents
Point production follows goal production for UFAs
during their contract years. That is, the overall trend
is for UFAs to perform at a significantly lower level The biggest difference between unrestricted free
during contract years when compared to their career agents (UFAs) and restricted free agents (RFAs) is
averages.
age. RFAs, as a rule, are nearly always under the age
of 27, while UFAs are typically 27 and older (usually
Probably the biggest nail in the coffin for the UFA much older).
contract year myth is that shot production is noticeably weaker in contract years compared to the av- Since declines in player performance have been linked
erages during a player’s career. How often a player with age (in many studies), it makes sense to split
shoots the puck is one of the few stats a player has these two groups. It also makes sense that we should
control over. Figure 6.5 proves without a doubt that not take the results of our UFA analysis and try to
UFAs do not increase their shot output during con- apply them to RFAs; we’re dealing with a completely
tract years.
different subgroup of NHL players.
It is fairly safe to say that the myth of the “contract
year” for UFAs has been busted. It would be unwise
for you as a fantasy hockey manager to draft UFAs in
the hope that they see a contract-driven performance
boost. In fact, the opposite is true; most UFAs will
perform at rates significantly below their career aver-
To explore whether or not RFAs see a performance
increase during their contract years, we can’t simply
look at an RFA’s contract year performance and compare it to his career averages. The main reason for
this is that most players in their low-to-mid 20s see
increases in their performance. They shoot the puck
CHAPTER 6. MYTHBUSTERS
36
more which leads to more goals and more points. In- cally at goals per game production. The orange dots
stead, a better approach is to compare RFAs side-by- represent RFAs in their contract year, while the white
side with all NHL players under the age of 27.
dots represent all non-RFAs. The norm appears to
be that players under the age of 27 can generally expect a boost in goals per game production in a given
season compared to their career averages.
But, if you look closely at Figure 6.6, you should notice that the RFAs are almost all above the diagonal
line. Figure 6.7 plots the same data but removes all
of the non-RFA players. Most fantasy relevant RFAs
do perform better (as compared to their career averages) during their contract year. So, while many
non-RFAs under the age of 27 experience goal production boosts during a given season, most RFAs will
experience this boost. Thus, it appears that RFAs do
experience an increase in performance that might be
regarded as a contract year boost.
In the interest of symmetry, we’ll perform a similar
analysis on point production and shot production.
Figure 6.6: Goal Production by Players Under the
Age of 27
Figure 6.8: Point Production by Players Under the
Age of 27
Figure 6.7: Goal Production by RFAs in Contract
Generally speaking, the performance of RFAs is
years
boosted during contract years as compared to players
Figure 6.6 reveals how all NHL players (below the age in the same age group who are not RFAs. This sugof 27) performed in the most recent season compared gests a possible fantasy hockey draft strategy: given
to their career average. This chart is looking specifi- two players of similar talent level and age, choose the
CHAPTER 6. MYTHBUSTERS
Figure 6.9: Point Production by RFAs in Contract
years
37
Figure 6.11: Shot Production by RFAs in Contract
years
6.3
Goalies Are Good at the
Penalty Kill
Watch enough hockey over the years and you’re
bound to come across the old hockey adage stating
that your goalie has to be your best penalty killer.
Andy Murray,13 Terry Murray,14 Ron Wilson, and
Dave Allison.15 These are just a handful of (probably many) former NHL coaches that have used some
form of the phrase: your goalie has to be your best
penalty killer.16
On the surface, the statement seems rather benign
and perhaps based in truth. But the adage also presupposes that goalies can actually be good (or bad)
at the penalty kill. It is this presupposition that we
will explore here.
Figure 6.10: Shot Production by Players Under the
Age of 27
player who will be entering his RFA contract season.
13 https://goo.gl/h796sW
14 https://goo.gl/Vzqsbs
15 https://nordicquotes.com/author/Dave_Allison/4
16 You may not have heard of Dave Allison. He was brought
on to coach the Ottawa Senators 20 games into the 1995-1996
season after the team fired Rick Bowness. Allison would last
just 27 games (2-22-3). We’re pretty sure that his firing was
not related to his philosophy about the penalty kill; at least
we think so.
CHAPTER 6. MYTHBUSTERS
6.3.1
Career Evolution
One approach to determining whether or not goalies
are good (or bad) on the penalty kill is to look at the
career save percentage of all NHL goalies while on
the penalty kill (PKSV%). The idea here is that if
we plot the career save percentage for all NHL goalies,
we should be able to easily identify which goalies are
good at the penalty kill and which goalies are bad at
the penalty kill.
38
on the penalty kill, every NHL goalie fits within a
narrow band from 0.860 to 0.884.17
The recent NHL average for PKSV% is 0.872.18 After
2,500 shots, the difference between an average NHL
goalie and the “best penalty killing” goalie amounts
to 30 extra goals allowed - over the course of eight
NHL seasons.
To put this to you in a slightly different manner: the
difference between the best and average is less than
four goals over the course of an entire season.19 Taking it one step further, the difference between the best
and the worst amounts to 7.5 goals over the course of
a season.20 This result should make it very clear to
you that there is no separation in multi-season talent
between goalies on the penalty kill.21
6.3.2
Can You Do It Again?
If take a look at all NHL goalies who faced at least
200 shots on the penalty kill in the 2017-2018 season,
you’ll find that their PKSV% values are considerably
spread out.22
Figure 6.12: Career Penalty Kill Save Percentage
Sergei Bobrovsky of the Columbus Blue Jackets
stopped just 83.1% of the shots he faced on the
penalty kill. That was the lowest PKSV% in the
league for goalies facing at least 200 penalty kill shots.
Figure 6.12 reveals the cumulative PKSV% for all
17 It would take approximately eight seasons for a starting
NHL goalies who have played since the 1997-1998 goalie to reach 2,500 shots against on the penalty kill.
18 This is the weighted average of all NHL goalies over the
season. The vertical axis is the PKSV% and the horizontal axis is the number of shots a goalie has faced past five seasons.
19 The exact value is 3.75 goals.
while on the penalty kill during his career (PKSA).
20
For small values of PKSA (these data points represent goalies who are new to the NHL or played just a
few seasons during the time interval we’re analyzing),
you’ll see that PKSV% values vary considerably. For
these small sample sizes, PKSV% ranges from 0.820
to just North of 0.920. But, more importantly, observe what happens for goalies who face more and
more shots on the penalty kill during their careers:
their cumulative PKSV% falls in a much more narrow window of values. In fact, after 2,500 shots faced
This difference is equivalent to 1.38 extra wins (or 2.76
standings points) over the course of a season. It seems a bit
ridiculous to use labels such as best and worst when the difference in talent amounts to a standings improvement of less
than three points.
21 Figure 6.12 definitely supports the idea that goalies can
post PKSV% values in a single season that differ significantly
from their career average. But the fact that these goalies end
up with career PKSV% averages nearly inseparable from the
league average is proof that the one-season values are nonrepeatable.
22 The choice of 200 shots is not completely arbitrary. It is
equivalent to starting at least 42 games for your team, making
you the de facto starting goalie.
CHAPTER 6. MYTHBUSTERS
39
John Gibson stopped 91.6% of the penalty kill shots Figure 6.13 is the tool we need to answer the questhat he faced for the Anaheim Ducks - good for the tions posed above. It is rather clear from this graph
highest PKSV% in the league in 2017-2018.23
that the answers to the questions are: no, Bobrovsky
is not bad at the penalty kill; no, Gibson is not good
The difference between Bobrovsky and Gibson at the penalty kill; and finally, no, you cannot do it
amounted to 17 extra goals allowed by Columbus again.
in the 2017-2018 season. Equivalently, this could be
looked at as about six standings points.24 Columbus The vertical blue line in Figure 6.13 is a reference line
made the playoffs with 97 points (but nearly missed showing you the league average PKSV% over the past
as the next closest team had 96 points). Anaheim several seasons. If you look at all of the goalies to the
finished with 101 points. That four point swing can right of this line, you’ll find goalies who posted an
all be explained by the difference in the PKSV% of above-average single season PKSV%. Use your hand
the two netminders.
to cover all of the goalies to the left of this blue line.
Now, what do you notice about the vertical position
Is Bobrovsky bad at the penalty kill? Is Gibson good of most of the goalies to the right of the blue line?
at the penalty kill?
Yes, that’s right; the majority of these goalies saw
a negative change in their PKSV% in the following
What does it mean to be good (or bad) at anything? season. That is, the goalies who posted an aboveGenerally, what distinguishes talent (being good or average PKSV% in “Season N” followed that up with
bad) from luck is the ability of the individual to re- a below-average performance in the next season.
peat the performance over time. Essentially, what we
want to know is this: can you do it again?
Now, perform the same test on the goalies to the left
of the vertical blue line. Cover the goalies on the
right side with your hand. Determine the vertical
location of most of the goalies to the left of the blue
line. These goalies posted below-average PKSV% in
“Season N” and then followed that up by posting
above-average PKSV% in the next season.
If you post a high PKSV% in one season, the most
likely result for you in the following season is that
your PKSV% will drop. If you post a low PKSV%
in one season, the most likely result for you in the
following season is that your PKSV% will rise.
Figure 6.13: Change in PKSV% in Consecutive Seasons
23 Again, we’re limiting our dataset to goalies who faced at
least 200 shots on the penalty kill.
24 Near the end of this document, we explore how a team’s
goal differential is related to their overall position in the NHL
standings.
Goalies are incapable of repeating their performance
on the penalty kill from one season to the next. Since
penalty kill performance is not repeatable, it follows
that goalies are neither good nor bad at the penalty
kill.25 Their single-season PKSV% values are largely
random.
25 Performances that are non-repeatable are heavily influenced by luck. This does not mean that PKSV% is useless to
us. In fact, we can use this luck to our advantage. Be sure to
read about PKSV% in the Application part of the draft guide.
Chapter 7
Shooting Percentage - Theory
7.1
Introduction
If we were limited to passing along only one piece of
advice to fantasy hockey managers for their drafts,
we would choose the simple, but powerful idea that
shooting percentage can be used as a tool to predict
future goal scoring production. This one idea would
prevent managers from drafting players too high and
alert managers to grabbing later round steals.
This first half of this chapter will use coin flips as a
mathematical model for understanding just how unlikely it is for a player to shoot with a success rate
that deviates substantially from his career average. If
you’ve used our draft guide in the past, the coin flip
discussion will be familiar. In the Applications version of this chapter, we will use data from last season
to alert you to players who are the most likely to experience a drop in goal production in 2023-2024. Additionally, we provide several interesting case studies
about shooting percentage for you to examine.
7.2
A Quick Discussion on Coin
Flips
A great way to build the proper context for shooting
percentage is to consider the simple flipping of a fair
coin. By fair coin, we simply mean that the odds of
the coin landing on heads are 50%. What we’d like to
explore in this section is how likely or unlikely certain
coin flip results are in coin flip experiments of varying
number of tosses.
Coin flips are inherently random. As such, the results of a coin flip experiment begin to look more
and more like a normal distribution as you increase
the number of coin flips. Normal distributions are
convenient for analysis because we can quickly produce an expectation of results by looking at a simple curve. We’ll eventually use these normal distributions to understand how the shooting percentages
(and consequently, the goals) of NHL players behave.
Figure 7.1 shows a typical normal distribution. Any
random set of results from a coin flip experiment will
produce a curve that is normal.
What does the curve tell us? The curve tells us the
likelihood of seeing particular results from a coin flip
experiment (for example, what percentage of coin
flips were heads during our experiment). The very
middle of the curve would be the average value, which
for fair coin flips is 50%. As you move further to
the left or right away from 50%, the chance of seeing that particular percentage of heads becomes ever
smaller. But that’s just a qualitative understanding
of the curve which you probably already knew intuitively. Can’t we do better?
We can do much better. The graph below is broken
up into sections using vertical lines and each section
has a percentage assigned to it. If you look at the sections immediately to left and to the right of the cen-
40
CHAPTER 7. SHOOTING PERCENTAGE - THEORY
41
Figure 7.1: The Normal Distribution
ter line (50%), you’ll see that each of these sections
have been assigned 34.13%. What do these numbers
mean? They mean that 68.3% of all coin flip experiments will have results that fall within these two
sections. If you then choose to include the next section to the left and next section to the right (the sections labeled with 13.59%), you can then state that
95.4% of all coin flip experiment results will somewhere within these four sections in the curve. Taking
this one step further, if you include the 3rd sections to
the left and right of the centerline, you’ll find 99.7%
of all results.
Right now, this might seem pretty abstract. Where
did these sections come from? What does the SD
represent that is assigned to each section? The SD
refers to standard deviation. Standard deviation is
a measure of how spread out your results are in an
experiment. For example, imagine you perform two
experiments where you are measuring the height of
two groups of humans. In the first group/experiment,
you record the following results: 185 cm, 175 cm, 170
cm, 195 cm, 192 cm. In the second group/experiment, you record these results: 175 cm, 176 cm, 181
cm, 182 cm, 179 cm. Just by scanning the two sets of
data, you should be able to convince yourself qualitatively that the first group of data is more spread out
than the second group of data. If we computed the
standard deviations of the two data sets, we would
find that the first data set had a larger standard deviation.
We are now going to run some coin flip experiments.
The goal here is to understand how normal distributions completely explain the expected results of coin
flip experiments. We’ll start with a coin flip experiment where we flip the coin 10 times. Our goal is to
understand how likely (or unlikely) certain results are
in this type of experiment. To reach that goal, we’re
going to have to do a little math. The math is not
difficult and you won’t need to do any calculations
yourself (now, or later). But, by the time we get to
analyzing shooting percentages, I want you to know
where the numbers are coming from. The calculation
we’ll perform is computing the standard deviation.
The equation looks like this:
p
SD = N p(1 − p)
where N represents the number of coin flips and p is
the probability of seeings heads on a single coin flip.
For a 10 flip experiment, N would be 10 and p would
be 0.5. Putting these values into the equation above
yields a standard deviation (SD) of 1.6. Now that we
know what the standard deviation is for a 10 flip experiment, we can link the standard deviation to the
normal distribution above. In this experiment, five
CHAPTER 7. SHOOTING PERCENTAGE - THEORY
heads is the most likely result. If we subtract/add
one standard deviation from/to five, we get the following values: 3.4 and 6.6. We would say that if
our results from the coin flip experiment yield heads
somewhere between 3.4 and 6.6 times, then those results fall within one standard deviation of the mean.
Linking this to the normal distribution curve from
earlier, 68.3% of the time, you’ll get between 3.4 and
6.6 heads in a coin flip experiment of 10 flips.
42
sample sizes are highly unlikely to produce results far
from the mean.
If you were asked to bet $1000.00 on a coin flipping
contest as to what fraction of coin tosses will end up
as heads after 1000 flips, you’d be a fool to choose
any number that deviated far from 50%. Is it possible that the contest ends with 35% heads? Or 72%
heads? Sure; it is possible but it is extremely unlikely. How unlikely? In you ran a bunch of 1000
How about two standard deviations? Two standard coin flipping experiments, 99.7% of the time, you’ll
deviations for this experiment would be 3.2 (1.6 x 2). get between 452 heads and 548 heads (that is, 99.7%
Starting with the mean value of 5, we arrive at 1.8 of the time, your results will fall within three stanand 8.2 for our two new values. That is, any results dard deviations of the mean).
that yield between 1.8 and 8.2 heads are said to fall
within two standard deviations of the mean. The And this simple statement forms the basis of most
normal distribution curve tells us that 95.4% of the statistical arguments including those within this draft
time we run a 10 flip coin experiment, we will get kit. If you sit at your house and flip a coin 1000 times,
we cannot tell you how many heads you will get in
between 1.8 and 8.2 heads as the result.
your experiment. But, we can tell you (using the norSo far so good. We can perform the exact method mal distribution curves) what the most likely results
above on 100 flip coin experiments and 1000 flip coin are and which results are extremely unlikely. Simiexperiments. Rather than go through all the dirty larly, we can’t know what shooting percentage Phil
work, below you’ll find a table of results for the three Kessel will have in 2023-2024, but we can know the
experiments.
most likely shooting percentage he will attain (his
career shooting percentage). And betting on a significantly different number from his career value is
Table 7.1: Coin Flip Experiments
equivalent to expecting a coin flip contest to produce
heads 300 times on 1000 flips. It is a bet you will lose
Flips SD
±1 SD
±2 SD
almost every single time.
10
1.6
(3.4 - 6.6)
(1.8 - 8.2)
100
5
(45 - 55)
(40 - 60)
1000 15.8 (484 - 516)
(468 - 532)
7.3
These results above are hugely important to our discussions going forward. In a 10 flip coin experiment,
we could see between 34% heads and 66% heads in
68.3% of our experiments. If you increase the number of flips to 100, 68.3% of the experiments will yield
between 45% and 55% heads. Further increasing the
number of flips to 1000 yields between 48.4% and
51.6% heads. The takeaway here is that as you increase your sample size, the likelihood that you’ll end
up with a result close to your mean value is large. Put
another way, experiments with small sample sizes can
yield widely varying results. Experiments with large
Applying Coin Flip Experiments to NHL Players
Let’s take our new tool from the coin flip experiments
and apply it to NHL players.
If you take the top goal scorers (say, the top 200) in
the NHL, you’ll find that many of them have shooting
percentages at or around 12%. That is, these players
(on average) will score a goal on 12% of the shots
they take. Much like coin flips landing on heads,
a shot on goal has a certain likelihood of going in
CHAPTER 7. SHOOTING PERCENTAGE - THEORY
the net (becoming a goal). That likelihood for prolific goal scorers is about 12%, whereas for a coin it
is 50%. For small sample sizes (such as 10 shots,
or even 100 shots), there is a reasonable probability that a player’s shooting percentage will deviate
from 12% (much like the total number of heads in
a coin flip experiment can deviate from 50%). But,
the more and more shots a player takes, it becomes
more and more likely that his shooting percentage
will approach his talent level (12% in this particular
example). It will be your job in your fantasy hockey
draft to place your bets on players performing at or
near their career shooting percentages. These are the
only consistent, winning bets in fantasy hockey.
Let’s consider a 12% shooter in the NHL who takes
275 shots in a single season. These kind of numbers
correlate well to 1st/2nd line players in the NHL, so
it’s important that we understand how these players behave. We can go through the typical standard
deviation calculations we described in a previous section and we’ll find out that the standard deviation for
this type of player is right around 2%. Recalling that
68.3% of experiment results will yield values within
one standard deviation of the mean, we can say that
this type of NHL player will have a shooting percentage (for an entire season) between 10% and 14%
in 68.3% of the experiments. What I really mean
here by experiment is an NHL season (275 shots on
goal). Extending this, in 95.4% of his NHL seasons,
this type of player will have a shooting percentage
between 8% and 16% (two standard deviations from
the mean).
Since most NHL careers are about 10-15 years, it
would be very unlikely for a player with 12% scoring
talent to post a shooting percentage outside of the
8%-16% range more than once in his career. It is this
concept that forms the basis for why players who post
shooting percentages in one season that are wildly
different from their career averages are expected to
regress toward the mean. In our own studies (completed over a five-year period), players with high oneseason SH% typically have about a 90% chance of
experiencing a strong regression.
7.4
43
Corey Perry’s 50 Goal Season
In the 2010-2011 season, Corey Perry (who, up until
that time, had never scored more than 32 goals in a
season) erupted for a 50 goal outburst. Overnight, he
became every fantasy hockey manager’s dream pick
for the 2011-2012 draft. There was talk of 60 goals.
Seriously. Our draft guide that year claimed Perry
wouldn’t reach 40 goals in the 2011-2012 season. We
were roundly mocked.
Let’s apply our coin flip knowledge to 2011-2012
Corey Perry. Perry’s career shooting percentage (or
odds that one of his shots on goal becomes a goal) is
13.4%. He typically takes about 252 shots in a season.
If you run these numbers using our coin flip methods,
you end up with about 2.1% for the standard deviation of his shooting percentage. So, we can make
the following claims about Perry’s expected shooting
percentage during his career:
• 68% of the time, his shooting percentage should
be between 11.3% - 15.5%
• 95% of the time, his shooting percentage should
be between 9.2% - 17.6%
• 99% of the time, his shooting percentage should
be between 7.1% - 19.7%
When Perry scored 50 goals in 2010-2011, he was riding a 17.2% shooting percentage. Our normal distribution curve tell us that was an unlikely result. Perry
should produce shooting percentages outside of the
9.2% - 17.6% range in only 5% of the NHL seasons
that he plays (basically once in his career). The most
likely future performance for Perry would be in the
11.3% - 15.5% range. So when our staff was putting
together our projections for the 2011-2012 season,
we expected Perry to score between 35-40 goals. In
fact, Perry scored 37 goals in 2011-2012 and shot
with 13.4% success. This result caught many fantasy hockey managers by surprise - but it shouldn’t
CHAPTER 7. SHOOTING PERCENTAGE - THEORY
have! In 68% of the seasons played by NHL players,
they will shoot within ±2% of their career shooting
percentage. 95% of the time, these same players will
shoot within ±4% of their career shooting percentage. To base your fantasy hockey seasons on values
outside of these ranges can only be described as fantasy suicide.
44
Chapter 8
Individual Points Percentage - Theory
8.1
8.2
Introduction
Another metric that we can take advantage of in
fantasy hockey is that of Individual Points Percentage (IPP). IPP measures how often a player earns
a point while he is on the ice. The calculation for
IPP is rather simple; you take the number of points
awarded to Player X and divide it by the number of
goals scored by Player X’s team while he was on the
ice.1 Multiply that fraction by 100 and you have your
IPP value.
Case Study: Dougie Hamilton
In his first season (2021-2022) with the New Jersey
Devils, Dougie Hamilton disappointed fantasy managers by generating just 30 points in 62 games. At
0.48 points per game, Hamilton was on pace for just
40 points over the course of a full season—a significant dropoff from the 66 points he had been averaging
in recent seasons.
Fantasy managers, thinking that Hamilton might not
Typical values of IPP are 70% for forwards and 30%
produce as much on his new team, let Hamilton slip
for defensemen. Of course, elite players will exceed
into the 9th round (on average) during the Summer
2
these values.
drafts of 2022. Hamilton, in his previous two seaIf Player X scores a goal or assists on a goal, he earns sons had been drafted in the 4th round as one of the
a contribution toward his IPP. Since goal scoring and league’s top offensive defensemen.
playmaking are considered talents in the NHL, then
the IPP serves, on some level, as a metric for player
talent. But, since goal scoring and assist generation
are both processes that are subject to random fluctuations (luck), the overall IPP of a player is also subject to random fluctuations. And it is these random
fluctuations that we will use to our advantage when
drafting in fantasy hockey leagues. Similar to SH%,
we’re going to try to find players with IPP values way
beyond their career norms and use these numbers to
predict a regression in points in the 2023-2024 season.
1 The
calculation is performed at even-strength.
Malkin consistently maintains an IPP in the 75%
- 85% range.
2 Evgeni
Managers who let Hamilton slip until the 9th round
of their drafts were missing a key piece of the puzzle: Hamilton’s IPP. Figure 8.1 shows Hamilton’s
IPP over the course of his career (relative to his career average). It is clear from the graph that Hamilton’s production in the 2021-2022 season was muted
by bad luck; specifically, Hamilton’s IPP was nearly
30% below his career level. Managers who used the
Left Wing Lock draft kit immediately knew what this
meant; Hamilton was poised for a bounce-back season
in 2022-2023 as his IPP returned to normal levels.
Hamilton would not disappoint. He went on to generate 74 points in the 2022-2023 season for a new ca-
45
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
46
Figure 8.1: Dougie Hamilton (2012-2013 :: 2021-2022)
reer high. Hamilton was one of the steals of the 2022
Summer drafts as he finished the season as the fourthhighest scoring defensemen despite being available as
late as Round 9 in most drafts.
8.3
Case Study: Matt Duchene
Because of this poor season, Duchene was ignored in
fantasy drafts in the Summer of 2021. Despite ownership levels of 92% in his previous two seasons, Duchene’s ownership level sat at 0% on opening night.34
Instead of managers drafting Duchene in the 11th or
12th round (as had been the norm in previous seasons), Duchene went largely undrafted because managers expected another season of anemic offensive
output from the Nashville forward.
But users of the Left Wing Lock draft kit knew otherIn the 2020-2021 season, Matt Duchene generated
3 Left Wing Lock tracks this data on a daily basis during
just 13 points in 34 games for a 0.38 points/game
draft season each year.
pace. It was easily the worst season of his career
4 Duchene’s ownership level was not identically zero. We
as Duchene has consistently generated closer to 0.75 know this for sure because the Left Wing Lock team drafted
points per game across his career.
him in one of our leagues.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
47
Figure 8.2: Matt Duchene (2013-2014 :: 2020-2021)
wise. Figure 8.2 displays Duchene’s IPP values over
the past eight season. Pay particular attention to his
2020-2021 value, which is nearly 40% below his career
average. A wildly different IPP value in a single season is a strong indicator of future changes. Left Wing
Lock draft kit clients knew to expect a bounce-back
season from Duchene. In September of 2021, we publicly called for managers to stop ignoring Duchene in
fantasy drafts and mentioned that we expected him
to have a bounce-back season.5
5 https://twitter.com/Left_Wing_Lock/status/
1439249786163576842
The payoff for managers who trusted in the IPP data
was enormous. Duchene had the best season of his
career, scoring 86 points in 78 games. This performance was strong enough to leave Duchene ranked
20th overall in point production in the NHL for the
2021-2022 season.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
48
Figure 8.3: Nicklas Backstrom (2012-2013 :: 2019-2020)
8.4
Case Study: Nicklas Backstrom
As a result of this low offensive output, Backstrom
was hammered by most fantasy hockey ranking systems. Yahoo had him at 115.8 and ESPN had him at
111.6 - making him a mid-to-late 10th round pick in
12-team leagues. In the three previous seasons leading up until 2019-2020, Backstrom had typically been
drafted in Rounds 5-8 (with the more recent drafts
being the 8th round data as Backstrom had aged out
of his most productive years).
In the 2019-2020 season, Nicklas Backstrom produced
54 points in 61 games for a 0.88 points/game pace. It
was a disappointing season for the Washington center
and marked just the third time in his 14-year career
that he had a negative ± value for the season. You’d
When it came time to draft Backstrom in the 2020have to go back nine years to 2010-2011 to find the
2021 drafts, most fantasy hockey managers followed
last time Backstrom generated offense at such a low
the herd or so called “wisdom of crowds.” This would
level.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
prove to be a costly, yet avoidable, mistake. In our
2020-2021 draft kit, we noted that Backstrom’s IPP
value from the 2019-2020 season was suspiciously low.
Figure 8.3 is an eight-year view of Backstrom’s IPP
values. It’s normal for a player’s IPP value to fluctuate from his career average by 5-10% in a given
season. But when you see a fluctuation as great as
the one in 2019-2020, it’s your job as a fantasy hockey
manager to pounce.
We recommended our draft kit clients draft Backstrom earlier than his 10th round ranking at the major fantasy websites. We published notes about him
in both the IPP chapter and the Washington Capitals chapter of the Applications PDF. Just before the
season was to begin, we also posted our Backstrom
advice on Twitter so that a permanent record would
be established for people without access to last year’s
draft kit.6
49
This would turn out to be a mistake - and a predictable one at that. Figure 8.4 is a seven-year history
of Goligoski’s IPP. As is typical, Goligoski’s annual
IPP values bounced around slightly above and below
his career average (by about 10%). But in 2017-2018
(the year he set a career high in goals), Goligoski’s
IPP skyrockets to 36% above his career average. This
was your warning sign - and draft kit users from last
Summer were alerted to this potential problem in the
Arizona Coyotes chapter of the Application book.
In 2018-2019, Goligoski saw his goal production drop
by 75% and his point production drop by 23%. Managers who used that 15th round pick on Goligoski
were left holding the bag on a defenseman who generated very little offense, almost no penalty minutes,
and 1.2 shots per game. A 15th round pick isn’t going to kill your fantasy season by any stretch. But
imagine having used that 15th round pick on Erik
Gustafsson, TJ Brodie, or Vince Dunn.8
Managers who trusted in this idea of IPP were rewarded handsomely in the 2020-2021 season. Backstrom went on to achieve near point-per-game status
Case Study: Claude Giroux
with 53 points in 55 games making him the 20th best 8.6
point producer in the NHL. He also finished the season ranked 12th overall in power play production.
In the fantasy hockey drafts leading up to the 2017Not bad for a guy that all the major sites had ranked
2018 season, Claude Giroux was being selected late
as a late-round scrub.
in the sixth round. Managers had grown weary as
Giroux’s point production had dropped from 86 to
73 to 67 to 58 during a four-year span. His shot
8.5 Case Study: Alex Goligoski production in 2016-2017 reached levels not seen in
six years and his shooting percentage dipped to just
7%. Giroux was just 29 years old heading into fantasy
Alex Goligoski finished the 2017-2018 season with 14 drafts last Summer, but managers were behaving as
power play points and a career-high 12 goals. De- if he were in significant decline.
spite offering very little to fantasy hockey managers
in peripheral categories, Goligoski found himself be- There were signs, though, that Giroux was not in
ing drafted in most leagues entering the 2018-2019 a precipitous drop; instead, he may have been on
the receiving end of seriously bad luck in 2016-2017.9
season.7
8 All
three of these players would have been available to you
Evidently, managers saw his offensive production in at this point in your draft and none of them had IPP warning
2017–2018 as repeatable and worthy of a draft pick. signs.
6 https://twitter.com/Left_Wing_Lock/status/
1347964032800727041
7 Goligoski does not generate high levels of shots, hits, or
penalty minutes. He does block shots but not at an elite level.
9 There were several important factors that played a role in
Giroux’s downward turn over the past few seasons. For example, Giroux saw a revolving door of linemates in 2016-2017. He
also had two serious surgeries in the Summer preceding that
same season.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
50
Figure 8.4: Alex Goligoski (2011-2012 :: 2017-2018)
One of these signs was Giroux’s IPP value. Figure 8.5
reveals the career history of Giroux’s IPP values. His
career average IPP value heading into the 2017-2018
season was 72.5%.
Giroux’s IPP value in 2016-2017 was nearly 30%
lower than his career average. This was a clear sign
that his point production during that season was an
anomaly and that he would likely experience a significant bounce-back season in 2017-2018. As early
as August of 2017, we were promoting Giroux on social media (and in our draft kit) as a player who was
being selected way too late in drafts.11
Over the course of his career, Giroux had stayed
roughly within ± 10% of that average. But something tremendous happened in 2016-2017. Giroux’s
IPP value plummeted to 52.9%. Only two players on
that Flyers team had lower IPP values that season: Giroux’s massive drop in IPP gave us confidence
Dale Weise and Pierre-Edouard Bellamare.10
to continue our public push for managers to draft
10 Weise barely saw any playing time the following season
and Bellamare was left unprotected for the expansion draft.
11 https://twitter.com/Left_Wing_Lock/status/
901151015726088193
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
51
Figure 8.5: Claude Giroux (2008-2009 :: 2016-2017)
Giroux earlier than his sixth round position well into 8.7
Case Study: Jiri Hudler
September.12 His case was one of the most extreme
examples of an IPP shift that we had come across in a
decade and Giroux did not disappoint managers who
heeded the call. He finished second in NHL scoring During the 2014-2015 season, Jiri Hudler set personal
with 102 points.13
bests in goals (31), assists (45), and points (76). He
had a monster season and fantasy hockey managers
took notice. Hudler, who was typically drafted in
Round 14 prior to the 2014-2015 season, was now be12 https://twitter.com/Left_Wing_Lock/status/
ing selected by the end of the 7th round heading into
905811427344220160
the 2015-2016 season. This average draft position
13 Giroux was one of just three players to surpass 100
points in the 2017-2018 season. Connor McDavid and Nikita (ADP) meant that managers valued him as a top-50
Kucherov were the others.
forward for the 2015-2016 season.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
52
That same Summer, our team was sounding the 8.8
Case Study: Eric Staal
alarm about Jiri Hudler. In our 2015 draft kit, at
our website, and on Twitter, we strongly discouraged managers from drafting Hudler too early.1415 Now we’ll examine a player who posted an unusually
We held firm in our conviction even as Hudler’s ADP low IPP value in 2015-2016, indicating that he was
likely going to rebound in 2016-2017.
continued to move higher.
How did we know that Hudler was overvalued in
drafts in 2015? We used his historical IPP values
as our guide. In Figure 8.6, you’ll see Hudler’s IPP
values (relative to his career average of 74%) for seven
seasons leading into the 2015 draft.
What you see here is a steady ride at, and around,
70% IPP (what you would expect from a typical forward in the NHL). But in 2014-2015, Hudler’s IPP
jumps to 90% (about 16 points above his career average). It is clear that Hudler earned points at a rate
that exceeded his talent level. Some of these points
were the result of a high SH%. But the rest of the
points came from assists. And as you’ll find out in
the next chapter, too many points from assists can
be a warning sign for bad things to come.
Given that Hudler had an atypical IPP value in 20142015, we projected him to see a significant drop in
production for the 2015-2016 season. Hudler finished
the 2015-2016 season with 15 goals and 30 assists,
giving him only half the point totals he had in the
previous season. His IPP value for 2015-2016 was
71% (a value consistent with his career numbers and
on par with typical, non-elite NHL forwards).
During the 2015-2016 season, Eric Staal produced
just 13 goals and 39 points despite playing in every
game of the season.16 These were the lowest numbers
posted by Staal since his rookie season in 2003-2004.
As a result of these poor numbers, Staal’s pre-season
ranking at the major fantasy hockey websites took
a beating. His average draft position (ADP) heading into the 2016-2017 season was 164.9 at Yahoo
and 125.0 at ESPN. Clearly, fantasy hockey managers
were listening to the people in charge of rankings at
Yahoo and ESPN. But passing on Staal and letting
another manager take him in round 14 would prove
to be a significant mistake at your fantasy draft. And
this mistake could have been avoided.
In our 2016 fantasy hockey draft kit, Staal made
a short list of players for whom we expected a big
bounce back and his IPP data was the reason why.
Staal’s career average IPP sits at about 74% (the
same as Jiri Hudler in our earlier example). We’d expect to see Staal’s IPP values fluctuate around that
average in any given season. And we’d be especially
curious about any seasons in which Staal’s IPP value
deviated significantly from that average.
Jiri Hudler (in 2015) is the prototypical example of Figure 8.7 reveals Staal’s nine most recent seasons
how to use historical IPP values ahead of your fantasy of IPP data heading into the 2016 fantasy hockey
draft to guarantee that you avoid overvaluing players. draft. His 2015-2016 IPP value was just 62.8 - nearly
12 units below his career average. This suggested
(strongly) that Staal’s weak 2015-2016 was likely to
be followed up by a bounce back season in 2016-2017.
In 2016-2017, Staal rebounded to produce 28 goals
and 37 assists for a total of 65 points (his best season
since 2011-2012). With 211 shots on goal and significant power play production, drafting Staal proved to
14 https://leftwinglock.com/articles.php?id=2523&
title=Leery-of-Jiri
15 https://twitter.com/Left_Wing_Lock/status/
641323577829158913
16 Staal actually played in 83 games that season as a result
of being traded at the deadline from the Carolina Hurricanes
to the New York Rangers.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
53
Figure 8.6: Jiri Hudler (2008-2009 :: 2014-2015)
be a wise choice. If you were able to get him as late chapter in the Applications book where we’ll provide
as the 14th round, you likely ended up competing for you with lists of forwards and defensemen who are
the championship in your league.
likely to see big changes in production for the 20232024 season based on wild swings in their IPP values
Eric Staal’s wild IPP deviation of 12 points below his (relative to their career averages).
career average should serve as benchmark for you on
how to use IPP to find potential bounce back candidates for your fantasy hockey draft.17
Be sure to check out the Individual Points Percentage
17 Staal also benefitted from a lucky 13.3% shooting percentage in 2016-2017. But he would have produced a respectable
23 goals had he shot at his career level.
CHAPTER 8. INDIVIDUAL POINTS PERCENTAGE - THEORY
Figure 8.7: Eric Staal (2007-2008 :: 2015-2016)
54
Chapter 9
Assists: Theory
9.1
Introduction
1.68 assists awarded per goal in the NHL.4
In the 2013-2014 season, defenseman Duncan Keith
posted 55 assists in the 79 games he played for the
Chicago Blackhawks. The following season Keith’s
average draft position in fantasy leagues was 39,
meaning (on average) he was being taken very early
in the third round.1 Imagine the disappointment of
many fantasy hockey managers when 16 defensemen
finished ahead of Keith in assists in the 2014-2015
season.2
If a single assist is awarded on a goal, that assist is
considered a primary assist. If there are two assists
awarded on a goal, then the first of the two players
who touched the puck prior to the goal scorer will
be awarded a secondary assist. The player who most
recently touched the puck before the goal scorer will
earn a primary assist.
9.3
Secondary Assists
Should Duncan Keith have been selected so early in
fantasy drafts? Could the disappointment of fantasy
To better understand what happened to Duncan
hockey managers been avoided? Read on to find out.
Keith in 2014-2015, it is instructive to focus on his
secondary assist data from the 2013-2014 season. In
2013-2014, Keith earned 0.44 secondary assists per
game. That number probably sounds great to you
9.2 Definitions
if you owned Keith in 2013-2014, but it was a bad
omen for owners in 2014-2015. This huge amount of
secondary assists per game played put Keith on a list
Before we jump into the data, let’s all come to an of players most likely to see a significant drop in point
agreement on the types of assists in the NHL. NHL production in 2014-2015. Let’s learn why.
scoring judges are free to award between zero and two
assists on any ordinary goal.3 On average, there are To understand the problem, we’re going to generate
a plot for you that compares the secondary assists
1 Only three defensemen were being taken ahead of Duncan
earned by NHL players in one season to the change in
Keith in most drafts: P.K. Subban, Erik Karlsson, and Shea secondary assists in the following season.5 Figure 9.1
Weber.
2 The 2014-2015 season saw 15 defensemen outscore Keith
on a points-basis.
3 By ordinary goal, we mean those goals that do not include
penalty shot goals and shootout goals.
4 It’s worth pointing out here that, on average, there are
0.935 primary assists per goal and 0.745 secondary assists per
goal.
5 To be clear, the horizontal axis is the secondary assists
55
CHAPTER 9. ASSISTS: THEORY
56
Figure 9.1: Year Over Year Secondary Assists
contains three seasons worth of data for secondary
assists. One of the most striking results from the
graph (and the one most relevant to our discussion
of Duncan Keith) is that most players (about 90%)
who earn more than 0.3 secondary assists per game
(that’s about 24 secondary assists over the course of
a season), see a significant and predictable drop in
their secondary assist totals the following season.6
For most players, whether they see a small positive
or negative change from season to season in their secondary assist totals is mostly random. But the graph
clearly changes around the 0.3 indicator. This is the
part of the graph you want to use when preparing
for your fantasy hockey draft. The data reveals that
high secondary assist totals are not sustainable. That
is, players with very high secondary assist totals in
one season usually cannot maintain those totals in
7
per game in a season and the vertical axis is the amount (in the following season.
%) that this number changed in the following season.
6 Any player who lands below the black horizontal line at
zero experienced a drop in the rate at which they produce
secondary assists.
7 This is true for about 90% of players that fall into this
argument.
CHAPTER 9. ASSISTS: THEORY
If you’re wondering about the left-most side of Figure 9.1, it’s not nearly as interesting as it looks.8 The
reason this part of the graph looks suspicious is the
following: players who earn very few secondary assists in one season (think 1-2 secondary assists) will
often see large changes in their totals (positive or
negative). Why? Because if you only have one or
two of a quantity to begin with, then moving up or
down by one or two results in a very large percentage
change (two is 100% growth over one, zero is -100%
growth over one). It’s as simple as that. But wait,
you protest! Why doesn’t the graph ever go below
-100%? Because once you lose 100% of something,
you haven’t got anything else to lose.
Many of the players who post more than 0.3 secondary assists per game in one season will see drops
as large as 50% in the following season.9 This is exactly what happened to Duncan Keith in 2014-2015.
His secondary assists per game dropped from 0.44 in
2013-2014 (second best of all skaters in the NHL) to
only 0.20 in 2014-2015 (putting him outside the top
35 defensemen in the NHL).10
Figure 9.1 then suggests an exploit for fantasy hockey
managers in the coming season: find players with
large (greater than 0.3) secondary assists per game
totals from the 2022-2023 season and de-value them
in your upcoming fantasy hockey draft.
In the corresponding chapter, titled Assists: Application, we’ll reveal which players will experience the
biggest drops in 2023-2024 for secondary assists. For
now, let’s take a look at primary assists.
8 Actually,
it is kind of interesting, it’s just not very useful
for fantasy hockey managers.
9 Not all of the players will experience drops that big, but
Figure 9.1 clearly shows that many of the players see drops on
the order of 25% to 50%.
10 Before you get too clever, it is true that the Chicago
Blackhawks experienced a drop of about 15% in their goal
scoring from 2013-2014 to 2014-2015. But this drop in goals
was nowhere near large enough to explain the drop in Duncan
Keith’s secondary assists.
57
9.4
Primary Assists
A natural extension of this idea of analyzing secondary assists is to examine primary assists using
the same approach. Rather than working through
all of the same details, we’ll present the basics here:
the data reveals that players with high primary assist
rates in one season suffer from drops in the following
season.
The only real difference between the drops in primary
and secondary assists is the cutoff. For secondary
assists, we were looking for players with 0.3 secondary
assists per game played. With primary assists, that
cutoff rises to 0.4 primary assists per game played.
Figure 9.2 uses three years of primary assist data to
reveal a trend very similar to the one we noticed with
secondary assists: player who post high rates (greater
than 0.40) in one season frequently experience a drop
in their totals in the following season.
CHAPTER 9. ASSISTS: THEORY
Figure 9.2: Year Over Year Primary Assists
58
Chapter 10
Projecting A Goalie’s Save Percentage
10.1
Background
goalie who has faced more than 400 PKSA (shots
against while on the penalty kill) has been able to
maintain a PKSV% greater than .892.
Experienced fantasy hockey managers know that one
of the keys to consistently winning at fantasy hockey
is having access to accurate player projections. For
example, knowing that Mike Smith would post a SV% 10.2
The Method
of .910 in 2012-2013 (and not the .930 he posted the
year before) would have prevented you from drafting
him at too high a position. We actually projected 10.2.1 Introduction
Smith at a SV% of .914 that year and our projections
were based on methods that had been tested robustly
With this data in hand, we can now begin to develop
(not just simple hunches).
a method for projecting goalie save percentage. First,
Our aim with this section of the draft guide is to you need to understand that a goalie’s save percentgive you a starting point on how to create your own age is made up of three distinct units: even-strength
projections for the category of goalie save percentage. save percentage (EVSV%), save percentage while on
This is generally the category most attributable to a the penalty kill (PKSV%), and save percentage while
goalie’s skill level and can, in fact, be used to project on the power play (PPSV%). As a quick warning, be
many of the other fantasy hockey categories used for aware that some sites reverse the abbreviations for
goalies. We provide you with SV% projections in the PPSV% and PKSV%. If you’re grabbing data from
spreadsheets, but we feel it is useful to show you some a website, be sure you know which category you’re
actually looking at.
basic methods here as well.
In an important article1 here at Left Wing Lock, we
argued that there is very little (if any) difference in
skill level for goalies when their team is on the penalty
kill. The two most important conclusions from this
article are repeated here: the league average PKSV%
(save percentage while on the penalty kill) sits firmly
at (or around) .865 season after season and only one
With that said, how much does each unit contribute
to a goalie’s overall save percentage? To answer that,
you need access to data on all shots taken during the
three different types of shifts. It turns out, Left Wing
Lock has that data going back many seasons.2 For
this example, we’ll use data that is appropriate to the
2022-2023 season. From this, we know that 82.0% of
all shots are taken at even-strength, 15.0% are taken
1 https://leftwinglock.com/articles.php?id=2360&
title=Penalty-Kill-Save-Percentage-PKSV%
2 https://leftwinglock.com/articles.php?id=3329
59
CHAPTER 10. PROJECTING A GOALIE’S SAVE PERCENTAGE
60
while the goalie’s team is on the penalty kill, and the them specifically so that you can see the impact of
remaining 3.0% are taken while the goalie’s team is PKSV% on a goalie’s overall SV%. First, here is some
on the powerplay.
data for each goaltender:
10.2.2
Looking at the League Averages
As a quick exercise, we’ll examine how to compute
the overall save percentage of a goalie with knowledge of these three units. Our internal data tells us
that the league average for EVSV% last season was
0.9123. The PKSV% was 0.8607 and the PPSV%
was 0.9089. Thus, to compute the league average
for overall save percentage, you would perform a calculation using Equation 4.1. Using the league averages we noted above, you arrive at the following
figure: 0.9044. It turns out, that this figure agrees
completely with the simple method of adding up all
the league saves last season and dividing them by
the total number of shots faced. So, we have an internally consistent method here for computing save
percentages. If I asked you to make a best guess for
a goalie’s save percentage next season (but I didn’t
reveal the goalie’s name to you), you should respond
with 0.9044.
Table 10.1: Data from 2022-2023 Season
Goalie
Georgiev
Sorokin
Gustavsson
EVSV%
PKSV%
PPSV%
SV%
.929
.930
.931
.854
.894
.918
.966
.891
.956
.919
.924
.931
Looking at the data from Table 10.1, all three goalies
posted strong (and nearly identical) numbers at even
strength and yet their overall SV% values differ considerably. A fundamental issue in becoming a better
fantasy hockey manager is understanding that the
PKSV% can have a significant impact on the overall SV%. But that’s not enough; it is imperative that
you understand the lessons from the PKSV% chapter
and realize that PKSV% is not a repeatable talent it is largely driven by luck.
With that in mind, and knowing that the league average PKSV% is .8607, you can conclude the following:
Filip Gustavsson’s overall SV% was significantly inflated by good luck, while Alexander Georgiev’s overall SV% was significantly muted by bad luck. Ilya
Sorokin’s overall SV% was moestly influenced by luck
10.2.3 How to Project
on the penalty kill (but nowhere close to the degree
of the other two goalies in this comparison). So, if
How can we use the above to project a goalie’s save you were going to make decisions about your fantasy
percentage for the 2023-2024 season?
hockey roster for the upcoming season, the PKSV%
is one of the numbers you want to focus on. And, you
As a first approximation, we can assume that the
want to find a way to correct for this “luck” from the
number of shots faced on the different types of shifts
previous season. How do you do this?
at the league level is a pretty good estimate of what
each team will face. Yes, there will be differences, The first thing we can do here is make corrections
but your odds of guessing which teams will have more to the PKSV% and PPSV%. We’ll see what each
power plays than other teams next season is likely to goalie’s overall save percentage would have looked
introduce greater error than this approximation.
like had they performed at the league average (for
PKSV% and PPSV%). We arrive at the following
With that said, we’ll choose three goalies for our exnumbers:
ercise: Alexandar Georgiev, Ilya Sorokin, and Filip
Gustavsson. These three goalies have 2022-2023
EVSV% values that are nearly identical. We chose
CHAPTER 10. PROJECTING A GOALIE’S SAVE PERCENTAGE
= (.820) ∗ (EV SV %) + (.150) ∗ (P KSV %) + (.030) ∗ (P P SV %)
Table 10.2: Adjusted Save Percentage
Goalie
Georgiev
Sorokin
Gustavsson
EVSV%
PKSV%
PPSV%
ASV%
.929
.930
.931
.861
.861
.861
.909
.909
.909
.918
.919
.921
This is not a bad first step. We simply used the
EVSV% from last season, adjusted the PKSV% and
PPSV% to league average levels and recomputed the
overall expected save percentage. Immediately, one
thing becomes clear: had all three goalies performed
around the league average in PKSV% last season (i.e.
experienced similar levels of luck), their overall save
percentages would have been much closer (differing
only by one or two units in the third decimal place).
But we can do better than this.
Last year’s EVSV% is not the best place to start
when estimating a goalie’s EVSV%. An enhanced
approach would involve using the career EVSV% of
each goalie as the starting point. The projections in
the goalie spreadsheet that accompany this PDF file
use these career numbers as their starting point.
You can certainly make goalie save percentage projections more complicated than the method outlined
above. But, as a tangible and practical strategy
heading into your fantasy hockey draft, this method
should provide you with defensible and accurate projections.
61
(10.1)
Chapter 11
When Are We Sure of a Goalie’s
Talent Level?
11.1
Introduction
each goalie facing 10,000 shots. In a 30-team league,
you’d need about 250 seasons of data to match that
number of shots.
One of the key problems with goaltender evaluation
is the small sample size. And by small sample size
(at this point in the discussion) we’re not referring
to how many shots the goalie has faced at this point
in his career. By small sample size, we mean how
many goalies, for example, have followed the same
career trajectory as Sergei Bobrovsky? He posted the
following save percentages in his first three seasons:
.915, .899, and .932. If we were interested in digging
up all NHL goalies who have followed a similar trajectory, how many do you think we would find? No
matter the number, it wouldn’t be sufficient to build
an analysis upon.
To overcome this lack of data, we decided to approach
this problem using simulated data. We would programmatically create three different types of goalies
(a bad goalie with a .907 SV%, an average goalie with
a .914 SV%, and a good goalie with a .921 SV%) and
fire a lot of pucks at them. If you consider that a typical starting goalie plays about 60 games and faces
about 30 shots per game, then a 5 or 6 year span
of a goalie’s career would yield about 10,000 shots
against. So, we fired about 10,000 shots at our three
goalies. But if we analyzed only one goalie of each
type, we’d be opening ourselves up to small sample
size problems all over again. Instead, we simulated
1000 careers of each of the three types of goalies with
11.2
Results
Below, we’ve plotted the results of the 1000 simulated
goalie careers for each of the three types of goalies.
The SV% along the vertical axis is cumulative, i.e.,
it tells you the career SV% of a goalie based on how
many shots he has faced up to that point in his career.
For convenience, we’ve placed tick marks along the
horizontal axis to represent a single season (using the
assumption that a goalie plays 60 games and faces 30
shots in each game - neither of which are in any way
critical to the outcome of the simuations).
So, what is the answer to the original question that
prompted this article? We’d focus on the part of the
graph where the spread of the data becomes relatively
constant. The spread of the data starts to become
more narrow at about 1300 shots of data. Anything
before that number of shots against and we really
have no clue how good a goalie is. Without overcomplicating things, nobody would call you crazy for
suggesting that you probably want at least 3000 shots
of data before you start making bold claims about a
goalie’s talent.
62
CHAPTER 11. WHEN ARE WE SURE OF A GOALIE’S TALENT LEVEL?
63
Figure 11.1: Simulation of 1000 Bad Goalies
Consider the 300 shots against part of the graph. 11.3
A Word of Caution
Imagine a vertical line running through the data at
that mark. The simulation suggests that an average
goalie is capable of posting almost any imaginable
The best way to illustrate why you should show cauSV% over the course of 300 shots. An average goalie
tion in evaluating goalies (both in fantasy hockey and
(over a 10-game stretch) can look like a rock star or
in analysis of hockey in general) is to plot some reala complete dud.
life data along with the simulated data for average
goalies. Figure 11.4 shows a simulation of 1000 careers of the average NHL goalie (that is, a goalie with
a SV% of .914). As discussed earlier, it takes a large
amount of data for the noise to settle down in the
simulations and for us to know the talent level of a
goalie.
CHAPTER 11. WHEN ARE WE SURE OF A GOALIE’S TALENT LEVEL?
Figure 11.2: Simulation of 1000 Average Goalies
We’ve added real-life data for three NHL goalies: Devan Dubnyk, Marc-Andre Fleury, and Michal Neuvirth. Observe each goalie one at a time. In each
case, notice how all three goalies did not look like
strong goalies early in their careers. In fact, through
1000 shots, Fleury and Dubnyk looked rather awful.
If you were pressed to predict the future of these three
goalies after only 1000 shots, you probably would
have been very wrong. But it’s not just you, it’s
everyone.
64
CHAPTER 11. WHEN ARE WE SURE OF A GOALIE’S TALENT LEVEL?
Figure 11.3: Simulation of 1000 Good Goalies
65
CHAPTER 11. WHEN ARE WE SURE OF A GOALIE’S TALENT LEVEL?
Figure 11.4: Devan Dubnyk, Marc-Andre Fleury, and Michal Neuvirth
66
Chapter 12
The Repeatability of Fantasy Hockey
Stats - Part I
12.1
Introduction
One of the most common questions we field during
draft season is why don’t we include projections for
Stat X, where Stat X is usually ± or something like
shorthanded goals (SHG). We love getting questions
and this particular question is both interesting and
useful in fantasy hockey. We will address this specific
type of question here, but we’d like to use this chapter
to discuss a topic with a broader appeal: which stats
in fantasy hockey are repeatable?
In our earlier chapter on penalty kill save percentage,
we defined the term non-repeatable. Loosely put, if
past data for a particular stat can be used to accurately predict future data for a particular stat, then
we say that the stat is repeatable. If accurate predictions cannot be made using past data, the stat is
non-repeatable (and is therefore dominated by luck).
That’s the essence of repeatability in a nice, tidy
black-or-white description.
12.2
Correlation
12.2.1
Conceptual Arguments
In Figure 12.1 and Figure 12.2, we present some imaginary data for two fantasy hockey stats which we’ll
call Stat X and Stat Y. These are imaginary stats so
that we don’t impose any prejudice into the discussion.
For each stat, we’ve plotted data from one season
(Season N) on the horizontal axis and data from the
following season (Season N+1) on the vertical axis.
What we’d like to explore here is the following question: does having information about the data from
the earlier season help us predict the data we see in
the future season?
Let’s examine the plots from a qualitative perspective and make some general comments about what
It turns out that the universe is not binary. It is we observe. For Stat X, we observe that low values
uncommon to be able to use the terms repeatable in Season N+1 correlate with low values in Season
or non-repeatable in an absolute sense without some N. And high values in Season N+1 correlate to high
grey area. This chapter will set out to describe what values in Season N. Roughly speaking, the better a
we mean (mathematically) when we say repeatable or player was at Stat X in Season N, the better he was
non-repeatable. It will also allow us to assign a mea- at Stat X in Season N+1. Intuitively, there seems
sure to how much “grey” there is for certain fantasy to be some type of relationship here and we can say
hockey stats.
there is some level of repeatability for Stat X.
67
CHAPTER 12. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART I
68
to expect in the following year.
So, we have an intuitive way of seeing how data sets
relate to one another, but it would be much more
useful to us as fantasy hockey managers if we had
some sort of mathematical number or metric that we
could use instead.
12.2.2
Figure 12.1: Scatter Plots for Stat X
Mathematical Arguments
We just made some hand-waving arguments for how
much the Season N+1 data depended on the Season N data for two different stats: Stat X and Stat
Y. It turns out that there are rigorous mathematical
methods for measuring how much one variable depends upon another variable.1 But this is a fantasy
hockey draft kit, not a math book, so we’re going to
borrow the important results that we need instead of
going into a long discussion about them.
Imagine drawing a line of best fit through the data
for Stat X and another line of best fit through the
data in Stat Y. These lines of best fit are shown in
black in Figure 12.3 and Figure 12.4. Consider these
lines of best fit to be theoretical models that attempt
to predict the Season N+1 stats using the Season N
stats. While it’s obvious from the plots which model
works better at predictions, what isn’t clear is how
much better it is.
Figure 12.2: Scatter Plots for Stat Y
We can all agree that the points for Stat X are closer
to the model than the points in Stat Y. This is no
coincidence. In fact, this is actually a good way to
determine which model is more accurate (that is, better at predicting). We could send a Left Wing Lock
staffer into an office and have him measure the distance between each plotted point and the black line
for Stat X. He could then be forced to perform the
same assignment for Stat Y. Whichever model (black
line) ends up with the smallest combined distance
between all of the points and the model is the better
prediction.2
When we look at Stat Y, the data is not very revealing. Do smaller Season N values mean smaller Season
N+1 values? Nope, not really. What about larger
values in Season N? The data in Season N doesn’t
overwhelmingly give us an indication of what to expect in Season N+1. If you were to make a statement
1 You can read more about this topic by performing a web
about Stat Y, you would conclude that one year’s search for the terms coefficient of determination.
2 Math nerds of the world: don’t throw a fit here. I am
data doesn’t seem to be a reliable indicator of what
CHAPTER 12. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART I
69
And this number will tell you how good your model
fits your data. Put a more useful way, this number
will tell you how well your Season N+1 data is explained by your Season N data. This number is R2
(pronounced: r-squared) and is known as the coefficient of determination.
R2 can take on values from 0 to 1. If all of your data
points were found to be in complete alignment with
your line of best fit (model), then the R2 value for
that model would be 1. That would indicate that
your model is perfect and your predictions will always be exactly right. Instead, if your R2 value were
0, your model would be completely useless and have
no predictive power whatsoever. Real life situations
almost always fall somewhere in between.
Figure 12.3: Stat X Data with Best Fit Line
Going back to our two sets of data and models (the
best line fits in black), we can determine the R2 values. We’ve done this and found that the R2 value
for Stat X is 0.95 and for Stat Y is 0.06. Thus, we
can claim the following for Stat X : 95% of the Season
N+1 Stat X data can be explained by the Season Stat
X data. And for Stat Y : 6% of the Season N+1 Stat
Y data can be explained by the Season Stat Y data.
If you were using projections for a fantasy hockey
season based on these models, you would feel more
confident with your Stat X numbers than your Stat
Y numbers.
12.3
Figure 12.4: Stat Y Data with Best Fit Line
It turns out that you can assign a number to these distances measurements3 that we have been discussing.
really glossing over the ugly details about how one would actually perform this calculation. For starters, it would include
squaring the distances first before adding them together.
3 It’s not correct to call R2 a distance, but this is not the
place for a rigorous discussion of the topic. Instead, I want
most readers to come away with a conceptual feel for the topic.
Putting It Together
We can compute a single, numerical value to assess
how accurately a model fits the data of a particular fantasy hockey statistic. This value, R2 , will be
a powerful tool in understanding which stats are repeatable and which stats are not.
Chapter 13
The Repeatability of Fantasy Hockey
Stats - Part II
13.1
13.2
Introduction
In the previous chapter, we introduced the idea of
using R2 as a metric for understanding how accurate
we can expect a predictive model to be for a particular statistic. Generally speaking, we can make the
following statements about R2 values1 :
• 0.0 - 0.3: a weak, or even non-existent, relationship between the variables;
• 0.3 - 0.6: a moderate relationship which may or
not serve your purpose;
We’ll start with a stat that is growing in popularity
in the fantasy hockey world: hits. We have a good
reason for starting with hits and that’s because we
can use them to clearly demonstrate the principle of
repeatability. We’ve collected six years of NHL hits
data and we’re going to plot it in such a way that we
compare data from one season to the data from the
season following it.
Figure 13.1 reveals the relationship that past hits
data has with the projection of future hits data - and
it’s a strong one.
• 0.6 - 0.9: a strong relationship indicating that
the variables are connected;
• 0.9 - 1.0: a very strong relationship indicating
that the variables are measuring almost the same
thing.
Armed with this new statistical knowledge, let us
apply these concepts to the common stats used in
fantasy hockey leagues to determine which stats are
repeatable (and therefore, can be predicted with reasonable levels of accuracy).
1 This
Hits
is a good time to point out that large R2 values
don’t necessarily indicate causality. This is not something we’ll
come across in fantasy hockey stats, but it’s useful to know in
general.
What immediately stands out when looking at this
plot is that there is a clear relationship between one
season’s hits data and the hits data in the season following it. NHL players with low hits totals in one
season generally have low hits totals in the following season. NHL players with high hits totals in one
season generally have high hits totals in the following season. From a conceptual point of view, this is
exactly what we mean when we state that a stat is
repeatable.
But let’s take this a step further. How strong is the
relationship for hits between one season and the following season? We have a method for determining
that strength and we call it R2 . We’ve computed
70
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
71
Figure 13.1: Year-to-Year Hits Data
this value for the hits category to be 0.80. What this
means is that of all the variation in hits data for the
future, 80% of it can be explained by a linear model
using past data.2
portant factor in your drafting decision. What we’re
saying here is that our level of certainty in our hits
projection is high.
This is a great start. Remember, high R2 values mean
Blocked Shots
that the stat is repeatable. If the stat is repeatable, it 13.3
is dominated by a player’s skill (or behavior) and not
by luck. And therefore, the stat can be accurately
projected for future seasons. If your league uses hits Another example of a fantasy hockey stat with high
as a stat, then our hit projections should be an im- repeatability is blocked shots. In Figure 13.2, we plot
data for NHL players over a six season period. Again,
2 If R2 had been 0.32, then 32% of the variation could be
a clear relationship is established between data from
explained by past data.
one season and the season immediately following it.
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
72
Figure 13.2: Year-to-Year Blocked Shots Data
The R2 value for blocked shots has been determined
to be 0.86 and is one of the strongest correlations of
any stat in fantasy hockey. If you’re drafting players
based on their ability to blocked shots, you can do so
with great confidence.
Notice how the majority of the data points are
bunched up along a straight line flowing from the
lower-left hand side to the upper-right hand side.
This tight bunching of the data (almost forming a
straight line) will be seen in most data sets with a
high R2 value.
13.4
Basic Scoring Categories
13.4.1
Goals
Now we’ll get into some of the more common scoring
categories used in fantasy hockey. The results for
goals are shown in Figure 13.3. While there is a solid
relationship between past goals and future goals, our
draft kit does not use past goals to predict future
goals (more on this later!).
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
13.5
Power Play Categories
13.5.1
Powerplay Goals
73
How about power play goals? It turns out that these
aren’t very predictable. The results are plotted in
Figure 13.5.
Figure 13.3: Year-to-Year Goals Data
13.4.2
Assists
Figure 13.5: Year-to-Year PPG Data
13.5.2
Powerplay Assists
How about power play assists? We can do a little
better in this category than we can in power play
goals.. The results are plotted in Figure 13.6.
Figure 13.4: Year-to-Year Assists Data
13.5.3
Powerplay Points
You might notice that the data for assists is more Power play points as a whole are easier to project
spread out than the data for goals, indicating a than either of the individual power play categories
weaker relationship. And you’re be right; the R2 on their own. The results are plotted in Figure 13.7.
value for assists is smaller than that for goals.3
3 In
the summary of this chapter, R2 values are listed for
most fantasy hockey scoring categories.
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
74
egory is based on luck. Figure 13.8 shows us why.
Figure 13.6: Year-to-Year PPA Data
Figure 13.8: Year-to-Year SHG Data
13.6.2
Shorthanded Assists
Here is another example of a stat that is entirely
unpredictable. Of all stats used in fantasy hockey,
shorthanded assists are the least predictive (even
worse than ±). It would be unwise to base your
fantasy draft strategy on this category and for that
reason we do not provide projections for shorthanded
assists. Figure 13.9 shows us why.
Figure 13.7: Year-to-Year PPP Data
13.6
Shorthanded Categories
13.6.1
Shorthanded Goals
The R2 for the shorthanded assists category is 0.02.
That means that of all the variation in shorthanded
assists data coming in the 2023-2024 season, only 2%
of it can be explained by the data from the 2022-2023.
Put another way, 98% of the shorthanded assists data
in the coming season is unexplained by past data! Do
the four shorthanded assists by your favorite right
wing last season really mean anything?4
This is a good place to mention that we do not proHere is an example of a stat that is entirely unpre- vide projections for any category with a low R2 value.
dictable. There is nothing wrong with using this stat
4 They mean nothing.
in fantasy hockey, just understand that the entire cat-
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
Figure 13.9: Year-to-Year SHA Data
75
Figure 13.10: Year-to-Year PIM Data
Shots on Goals
You might ask: why not? The answer is now straight- 13.8
forward; a low R2 necessarily implies that the stat
cannot be projected using past data. Moreover, a low
R2 value implies that the stat is determined more by Figure 13.11 reveals the repeatability of the shots on
goal (SOG) category in fantasy hockey. The correluck than it is by player skill or player behavior.
lation is strong and one we believe to be extremely
important.
At the risk of stating the obvious, all goals begin as
shots on goal. If you can establish a strong theoretical basis for the repeatability of shots on goal, you
13.7 Penalty Minutes
can actually improve your ability to project goals.
Our team uses this method and this is why our goals
While some fantasy hockey leagues are moving away projections are the most accurate on the market year
from the penalty minutes (PIM) category, it is still after year.
widely used in the fantasy hockey world. It turns out
The average R2 value for shots on goal is 0.78, indithat penalty minutes are fairly repeatable. We plot
cating a strong relationship between past data and
six years of data in Figure 13.10.
future data. Unless you have a circumstantial reason
If you’re in a league that assigns negative points to to believe a player’s shot totals are moving signifi-5
penalty minutes, this data should still interest you cantly up or down, use past shot data as your guide.
greatly. It’s your job as fantasy manager to minimize
the impact of negative points from penalty minutes.
5 A good example of circumstances that might lead to
And the best way to do that is to have accurate changes
in shot totals would be a player moving from the 2nd
penalty minute projections in your hands on draft line to the 1st line or a player getting more time on the power
day.
play.
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
76
the goals category. Several years of data suggests
that about 60% of future goal data is explainable by
past goal data. If you were to build a model projecting future goals (using past goal data), your model
would be subject to significant variations (∼40%) unexplained by the data of past seasons. To do better than this, you would need to develop a model
(that used something other than past goal data as
the input) to improve the R2 relationship between
the model and the observed data.
Can it be done? Yes, in some cases, significant improvements can be made in models that use input
data different from simple past data. We’ve been
working on these models for a number of years to improve the projections that all draft kit clients receive
and some of our biggest gains in R2 have come within
Figure 13.11: Year-to-Year SOG Data
the Goals and Powerplay Goals categories. We’re
happy to report that our projections consistently outperform all other draft kits and our competitive ad13.9 Summary
vantage comes from developing models that improve
upon these R2 limits.6 We’re also proud to announce
Before you can set about making projections of future that we’ve developed a reasonably accurate approach
stats, you must test and understand the repeatability to projecting game-winning goals (GWG) that does
of these stats. If the repeatability (measured as R2 ) not rely on using past game-winning goals data.
is low, then past data is not a reliable indicator of
future data thereby making most projection schemes
highly inaccurate. In Figure 13.12, the R2 values
Important Note on
for most statistical categories of fantasy hockey are 13.10
presented in order of their repeatability.
Games Played
Looking at Figure 13.12 might have you wondering
what hope you have of selecting the right players in
the upcoming draft. But fear not, there are two important details we have not mentioned yet:
13.10.1
How Do Other Sites Do It?
Most major fantasy hockey websites include projections for the number of games played by each player
• This non-repeatability of several statistics is
in the upcoming season.7 Without much thought,
what keeps the game of hockey (and by extenfantasy hockey managers assume these projections
sion, fantasy hockey) interesting. It would be
are meaningful. As a consequence, the projections
boring if everything were predictable.
6
When making apples-to-apples comparisons, our projec• The data presented here only represents the limtions come out on top.
7 CBS, ESPN, Yahoo, and TSN are just a handful of fantasy
its on projections of future statistics if you’re
using past data of the exact same statistic.
hockey websites that build their projections based on a games
Take a minute to digest that second point. Consider
played number. All of the direct competitors of Left Wing
Lock, Inc. build their projections by starting with a games
played number.
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
77
Figure 13.12: Repeatability Data for Fantasy Hockey Stats
used by fantasy hockey managers are built upon the
projections of games played. To be clear, these sites
will first generate a projection for how many games
they think Player X will play in the upcoming season. The remaining projections (goals, assists, hits,
etc.) are then constructed using this number of games
played.
On the surface, this approach seems reasonable. If
Player X is going to play 66 games and Player Y
is going to play 80 games, then you should expect
the projections for these players to account for that
difference. If the two players are of similar talent
levels, then the total projections for Player Y should
be larger than those for Player X given the fact that
he’ll play 14 more games.
There is one major assumption being swept under the
rug by all of these websites that build their projections on a games played number: their games played
projections are accurate.
As it turns out, this assumption is dangerously
wrong. Using the same R2 analysis we developed
above, we tracked the major sites to see how well
their projected games played totals correlated to the
actual data in that season. The results, seen in Table
13.1, are terrible.
CHAPTER 13. THE REPEATABILITY OF FANTASY HOCKEY STATS - PART II
Table 13.1: Games Played Accuracy
R2
Site
Dobber
CBS
Yahoo
TSN
DailyFaceoff
78
look like an extremely weak defensemen in comparison. And as such, these comparisons (using different
games played models) are useless.9
0.17
0.14
0.13
0.03
0.015
Of course, many players will not play 82 games. Sickness, injury, and family emergencies all play a role in
absence. But we believe the data shows these events
are unpredictable. Simply put: past games played
data is a poor indicator of future games played data.
Don’t believe it? Table 13.2 measures the correlaThe values from Table 13.1, if they were added to tion for two different predictive models: (1) future
Figure 13.12, would all find themselves on the far left games played based on last season’s games played toside of the plot. That is, they would be considered tals and (2) future games played based on a player’s
useless for building a projection model.
three-year average of games played.
If you’re building a model to explain future games
played and 80% of the future games played data is
unexplained by your model... it’s time to delete your
model. In the worst case, 98.5% of future games
played data remained unexplained by the website’s
model.8
13.10.2
Table 13.2: Games Played Correlations
Model
Single Season
Three Year Average
R2
0.25
0.17
How Does Left Wing Lock Do With both models, we see weak results. These simple
models, however weak, still managed to outperform
It?
all of the major sites’ accuracy levels for games played
projections.
Our preferred approach to games played for skaters
is to simply evaluate every NHL player on an even
playing field. We project every player on a per-game
basis, or put alternatively, we evaluate every player
as if he were going to play 82 games.
It would be a mistake for you to rely on past games
played data as you try to forecast player production
in an upcoming season. All of the games played projections models that are published provide you with
accuracy levels that are on par with the shorthanded
A good example of this for 2018-2019 is Shea We- and +/- categories. That is, they are hardly distinber. Our projections are built on the 82-game as- guishable from flipping a coin. Since these websites
sumption. Weber is going to miss significant time all use games played projections to build their produe to injury. Some estimates place his return on jections for the other statistical categories (goals, asDecember 15; others have his return as December sists, hits, etc.), the inaccuracy of the games played
31 or even January 15. We all know Weber is go- projections are transferred to each and every one of
ing to miss multiple months but no one knows how these categories.
many games he will miss. Now, when you’re evaluating players in a spreadsheet, how in the world do
you quickly assess the difference between Weber and
9 We’re aware that some folks will blast us at the end of
another defensemen when you’re only looking at 50 the season for having projected Weber to score 14 goals and
games of projection data for Weber? He’s going to 25 assists in 2018-2019. But honestly, anyone with a 3rd grade
8 Congratulations
ping device.
to DailyFaceoff on inventing a coin flip-
reading level understands what we’re doing here. This is our
projection for an 82-game season. If Weber plays half the
season, he’d see half the production.
Chapter 14
The Motivation for Enhanced Stats
14.1
Background
is analyzed.
A primary goal of analysis in hockey and fantasy
hockey is the ability to use statistics to accurately
project the future performance of individual players
and teams. Traditional hockey statistics (goals, assists, +/-, etc.) are limited in their ability to achieve
this goal, due in large part to their non-repeatability.
One alternative approach to hockey analysis would
use puck possession as its fundamental metric. That
is, if a player or team is dominant, that dominance
should be reflected in the amount of time in which
they possess the puck. Unfortunately, the NHL does
not track nor publish data related to puck possession.
In spite of this lack of data, there are methods that
can be used to track puck possession.
The purpose of this document is to introduce hockey
fans (and fantasy hockey managers) to the topic of
Enhanced Stats. Briefly, Enhanced Stats involves the
use of NHL shot data to analyze individual players
and teams. The shot data is used as a proxy for puck
possession. Essentially, teams that are able to shoot
the puck more often are doing so because they are
more frequently in possession of the puck. It turns
out that teams that are able to consistently outshoot
their opponents typically end up winning games and
performing well in the playoffs.1 Thus, shot data can
play an integral role in the way the game of hockey
1 http://www.arcticicehockey.com/2010/4/13/1416623/
corsi-and-score-effects
Editor’s Note: The type of analysis outlined in this
chapter is sometimes referred to as advanced statistics. Readers should be aware that the math involved
in Enhanced Stats is limited to basic addition, subtraction, multiplication, and division. The term advanced statistics is misleading (since there is very
little math involved) and won’t be used in this document.
14.2
A Simple Flip of a Coin
A natural place to start a discussion of statistics is
with the simple idea of a coin flip. We’ll consider an
ideal coin in which there is a 50% chance of getting a
heads result if you flip the coin. What we would like
to explore is the following: if we have a set of coin
flip results, can we trust these results to be a good
predictor of future coin flip results.2
2 Another interesting question to ask here is at what point
can you determine whether or not the coin flipping is rigged. If
you get six heads after 10 flips, is the coin rigged? What about
60 heads after 100 flips? And 600 heads on 1,000 flips? Be sure
to check out the results of our three coin flipping simulations
to see the answer.
79
CHAPTER 14. THE MOTIVATION FOR ENHANCED STATS
14.2.1
10 Coin Flips
Let’s start with an experiment in which we flip a coin
10 times and record the number of times the coin
lands head side up. Since doing this experiment only
once doesn’t give us any worthwhile data, we’ll perform this experiment one million times. That is, we’ll
flip the coin 10 times and record how many heads we
see - and we’ll do this one million times. This type of
simulation can be done fairly quickly on a computer
and the results are shown below in Fig. 14.1. But
before you take a look at these results, try running
two or three experiments yourself (remember, each
experiment is only 10 flips).3
80
sult, what would you expect to happen in the next
10 flips?
What you should take away from these results is that
the results of one experiment involving 10 coin clips
does not have much predictive benefit on future experiments. Furthermore, using a single experiment
as the basis for future projections could lead you to
make rather erroneous statements regarding the coin.
14.2.2
100 Coin Flips
Next, we’ll change the simulation so that each single
experiment involves flipping a coin 100 times. And
the simulation will again run one million experiments.
The results are shown in Fig. 14.2. How often does an
Figure 14.1: Results of a computer simulation of 10
coin flips run one million times.
The results of this experiment match common sense:
the most likely result is to see five heads. What
should also be clear to you (both from personal experience and the results) is that it is not uncommon at
all to see four heads or six heads. In fact, even three
or seven heads isn’t unrealistic. If you performed this
experiment just once and got three heads as the re3 If you don’t have a coin handy, there is a virtual coin
flipper available at: http://www.random.org/coins/.
Figure 14.2: Results of a computer simulation of 100
coin flips run one million times.
experimenter see 40% or 60% heads? Not too often
at all. What about 30% or 70% heads? Something
has changed. We’ve created more events by flipping
the coin 100 times instead of 10 and the spread (or
variance) of our results has narrowed. It is becoming
less likely for us to see the results that were quite
common in the 10 flip experiment.
CHAPTER 14. THE MOTIVATION FOR ENHANCED STATS
14.2.3
1000 Coin Flips
Let’s run one more simulation. Each experiment this
time will involve 1,000 flips of a coin. And the simulation will run one million times. Fig. 14.3 reveals the
outcome of the simulation. Now we see a dramatic
change in the data: there is virtually no chance at
all of seeing a result in which 40% or 60% of the coin
flips are heads. And 30% and 70% don’t even show
up on the graph. We now have a fairly narrow range
of expected results from this simulation.4
14.2.4
in this chapter, we mentioned methods for detection of rigged coins. Fig. 14.3 provides you with that method.
If you encounter a coin that yields 420 heads on 1,000 flips
(for example), you can be certain that you are dealing with a
rigged coin. The odds of seeing so few heads are essentially
zero. But who has time to wait around for 1,000 coin flips?
According to Fig. 14.2, witnessing 30 heads or less on 100 flips
should also raise the alarm. Likewise, 70 heads or more on 100
flips would be a sure sign of a rigged coin.
NHL Players as Coins
In the previous section, we ran simulations on the
flipping of ideal coins - that is, coins with a 50% probability of landing with the head side up. What we’ll
do in this section is treat NHL players as coins. Instead of ideal coins though, we’ll be using weighted
coins (or biased coins). Here, the probability of landing on one particular side will not be 50%, but will
instead be determined by how often that particular
player scores goals.
14.3.1
4 Earlier
Summary
Our original goal was to explore the relationship
between sample sizes and their predictive abilities.
With only 10 coin flips in your sample size, using
the number of heads from a single experiment would
be practically useless in determining how that same
coin would behave in future experiments. As we increased the sample size to 100 coins, we gained additional predictive power in that our range of expected
results narrowed to a window of about 40% - 60%
heads. Finally, we boosted our sample size to 1,000
coin flips and the benefit was immediately recognized:
expected results narrowed to about 47% - 53%. Given
the three simulations above, it is clear which one provides your best chance at predicting future performance: it is the simulation in which you have more
data.
14.3
Figure 14.3: Results of a computer simulation of
1,000 coin flips run one million times.
81
Phil Kessel
Phil Kessel, a forward now playing for the Pittsburgh
Penguins, has a career shooting percentage (SH%) of
10.8%. What exactly does it mean to have a SH% of
10.8%? To explore this idea, we’ll model Phil Kessel
as a weighted coin. Instead of using heads and tails,
we’ll call the sides of the coin goals and non-goals.
Kessel’s coin model works in the following manner:
when we flip the coin, there is a 10.8% chance that it
CHAPTER 14. THE MOTIVATION FOR ENHANCED STATS
82
will land with the goals side up and an 89.2% chance
that it will land with the non-goals side up.
We’re going to run three simulations for the Phil
Kessel coin:
• a 60 flip experiment
• a 273 flip experiment
• a 3,042 flip experiment.
You might be wondering why we have chosen such
odd numbers for our experiments. The 60 flip experiment will represent Kessel’s shot total in his first 15
games played in the 2012-2013 season; the 273 flip experiment represents the average number of shots that Figure 14.4: Results of a computer simulation of 60
Kessel takes in an 82-game season; the 3,042 flip ex- SOG run one million times.
periment represents the total number of shots Kessel
has taken in his NHL career.
The first experiment is a fascinating one; Kessel, a
seven-time 30+ goal scorer5 , had managed only two
goals in his first 15 games of the 2012-2013 season (he
had taken 60 shots on goal during this time frame).
We’ll use one million identical Kessels and have them
take 60 shots on goal each. The results are shown in
Fig. 14.4. Notice that the likelihood of Kessel scoring only two goals in a 60 shot span (3.3% SH%) is
not that high, but it’s certainly possible. Had you
tried using this 60 shot span to make assumptions
about Kessel’s future goal scoring, you would have
been sadly disappointed. Kessel went on to score 20
goals on 161 shots, giving him a SH% of 12.4% for
the season. Much like the 10 flips coin experiment,
you can’t use small sample sizes to make accurate
projections of future performance. Fantasy hockey
managers who traded Phil Kessel early in the 20122013 season became acutely aware of this notion.
What happens if we use a shot sample size of 273
shots on goal (recall this is the average number of
shots on goal by Kessel over an 82-game span)?
Fig. 14.5 provides us with a glimpse of the range of
5 Kessel’s 20 goal shortened season in 2012-2013 projects to
34 goals over 82 games.
expected goals scored by Phil Kessel. Taking a look
at these results should convince you of one thing: the
more shot data you have on a player, the better your
ability to project his future shooting percentage. Notice in the 60 shot experiment, that shooting percentages ranging from 0%-20% were in the realm of
possibility. In the 273 shot experiment, that range
narrows to about 5% - 15%. As it turns out, all of
Phil Kessel’s 12 NHL seasons have produced shooting
percentages within this exact range.
Finally, we look at Phil Kessel’s career numbers. He
has taken 3,042 shots over the course of his career.
With that data in mind, what happens to the range
of expected SH%? It is now only about 5% points
wide, ranging from about 8% to 13%. Incidentally,
this gives us some insight into the talent level of Phil
Kessel: given the 3,042 recorded shots on goal, it
would be very unlikely to see him record a SH% outside of the 8% - 13% range for an extended period of
time.
CHAPTER 14. THE MOTIVATION FOR ENHANCED STATS
Figure 14.5: Results of a computer simulation of 273
SOG run one million times.
Figure 14.6: Results of a computer simulation of
3,042 SOG run one million times.
14.5
14.4
Thoughts on Sample Sizes
It should now be quite clear to you that using small
sample sizes gives you virtually no predictive power
when projecting the future number of heads in a
coin flip experiment or the number of goals scored
by an NHL player. In a typical NHL season, approximately five players will score more than 40 goals.
That means, for the overwhelming majority of NHL
players, the sample size for goals in a season is less
than 40. Given what you’ve just learned from the
coin experiments, using goal data to project future
goal scoring is going to produce inaccurate results.
83
The Motivation for Analyzing Shot Data
If the number of goals or shots on goal isn’t enough
to form a useful data set, then how can this data set
be constructed? The solution is to include all shot
data in your data set. By all shot data, we mean
goals, shots on goal that were saved, missed shots,
and shots that were attempted but ended up being
blocked by the opposing team.
You might be asking yourself how much you can really gain by switching from a goals analysis to a shots
analysis. The question is fair and the answer may surprise you: for Phil Kessel’s 2012-2013 season, instead
of having only 20 events (goals) in your sample size,
Using shots on goal data should provide a significant you end up with 313 events (all shots combined). And
boost. But, it would take an entire season to see how about that 15 game stretch where Kessel scored
sample sizes in the 200-300 event range (and that’s only two goals? You’d have 114 events in your sample
for the high-end NHL shooters). The problem with if you included all of his shot attempts.
projecting the performance of NHL players is now
front and center: sample sizes of in-season goal and The method of using all shot data to analyze NHL
shot totals are too small to yield useful predictive teams and players will be called Enhanced Stats. An
benefits.
immediate benefit of using all shot data is that your
CHAPTER 14. THE MOTIVATION FOR ENHANCED STATS
sample sizes can be grown significantly, thereby allowing you to make more accurate player projections
using data over shorter time intervals. In the next
chapter, we’ll introduce Enhanced Stats to you and
reveal other benefits of the system to assist you in
your pursuit of hockey and fantasy hockey analysis.
84
Chapter 15
Getting to Know Enhanced Stats
15.1
Introduction
the statistic they are reading about. It is worthwhile
to discuss the new notation and terminology here.
The idea of using all shots (and not just goals or shots
on goals) to analyze NHL teams has been around
for decades. Rather than analyze skaters (or teams
as a whole), Jim Corsi first used shots to measure
the workload of his goalies.1 The first references to
this type of analysis being used on skaters was the
work done by Vic Ferrari2 in a 2007 article about the
Edmonton Oilers.3 This analysis has since evolved
into a greater understanding of which players and
teams are driving play in the NHL.
15.2.1
Simple Shot Counting
We’ll start with the most basic statistic of Enhanced
Stats: shot attempts for. Shot attempts for (SAT F )
is simply the total number of shots taken by a player
or team. The total number of shots taken by a team
(shots for) is found as follows:
SAT F = SOG + M S + BS,
(15.1)
where SOG represents shots on goal (both those that
are saved and those that result in goals), M S repre15.2 Notation and Terminology sents shots taken that missed the net, and BS represents shots taken that were blocked by the opposing
5
Currently, modern hockey analysis uses meaningless team. An equivalent metric can be defined for shots
names4 to represent the rather meaningful statistics taken by the opposing team. This sum is known as
that form the foundation of the field. In an effort to shot attempts against (SAT A) and is computed in
avoid this trap, we’ve made the decision to use the the same manner:
NHL’s nomenclature. These are names that should
SAT A = SOG + M S + BS.
(15.2)
give readers an immediate clue as to the definition of
Together, Eqs. (15.1) and (15.2), form the basis for
1 http://o.canada.com/sports/
most of the statistics used in Enhanced Stats. All
count-this-nhl-guru-as-more-than-just-a-number
2 This is the blogger’s pseudonym.
future terminology in this document will depend on
3 This site is now private: http://vhockey.blogspot.ca/
these simple definitions.
2007/10/corsi-numbers.html
4 In some instances, they have renamed basic hockey concepts that have existed since the beginning of the game. For
example, the term corsi is being used to represent the total
number of shots a team has taken. We already have a word
for that: it’s called shots.
5 This is a good place to point out that these individual
shot definitions are mutually exclusive. That is, for example,
a shot cannot be defined as both a missed shot and a blocked
shot. Each shot taken falls into no more than one of the three
types: shots on goal, missed shots, and blocked shots.
85
CHAPTER 15. GETTING TO KNOW ENHANCED STATS
15.2.2
Shot Attempts Differential
A more useful look at shot counting involves looking
at a team’s shot production compared to their opponent’s shot production. The shot attempts differential SAT is defined as shot attempts for minus shot
attempts against. It is a simple measure of whether
or not a team is outshooting their opponent. The
shot attempts differential is computed as follows:
SAT = SAT F − SAT A.
(15.3)
Alternatively, one can look at a team’s shot production relative to the total number of shot attempts in
a game (or season). By expressing the shot attempts
for as a percentage of the total shot attempts for and
against, it becomes immediately clear if a team is
outshooting their opponents. The shot attempts differential percentage SAT % is defined as:
SAT % = 100 ×
SAT F
.
SAT F + SAT A
(15.4)
86
team calculations. You simply compute the SAT F
and SAT A for events while that particular player is
on the ice. There is one useful modification you can
make when analyzing individual players: correcting
for the differences in time-on-ice. Since some players
see 25 minutes of ice time and others only see 8 minutes, their shot totals would be wildly different. To
account for this, we simply express the shot differential as a rate statistic. We’ve chosen 60 minutes as
our time frame. This stat is written as SAT /60 and
represents the expected shot attempts differential for
a player over a 60 minute interval of time. For reference, the top five skaters (as measured by SAT /60
in the 2022-2023 season) were:
• Derek Stepan (CAR)
• Stefan Noesen (CAR)
• Matthew Tkachuk (FLA)
• Paul Stastny (CAR)
• Brent Bruns (CAR)
A value of 50% for the SAT % would indicate that
the team played in a balanced game (or season) as
measured from a shot differential perspective. For Remarkably, six of the top seven skaters in the SAT%
the 2022-2023 season, the top five teams in SAT % metric come from the Carolina Hurricanes roster.6
were:
• Carolina Hurricanes (60%)
• Calgary Flames (57%)
• Florida Panthers (54%)
• New Jersey Devils (54%)
• Seattle Kraken (53%)
15.2.3
Unblocked Shot Attempts Differential
As mentioned earlier, the core reason for using Enhanced Stats is that they serve as a reasonable proxy
for a team’s puck possession. If a team is consistently outshooting an opponent, then that team has
possession of the puck more often. An argument has
been made 7 that these Enhanced Stats can be used
as a proxy for scoring chances and therefore the inclusion of blocked shots is really not necessary at all.8
Given that four of the teams listed above were playoff teams (with two earning over 110 points each in
the standings), it would seem that SAT % might be
6 We imposed a requirement that the players appeared in
a reasonable approach to discovering teams destined
at least 41 games during the season.
for playoff success.
7
You can also compute SAT and SAT % for individual
players. The formulas are identical to those for the
This site is now private: http://vhockey.blogspot.com/
2007/11/driving-possession.html
8 Furthermore, blocked shot totals suffer from rink bias - an
apparent inflation of certain stats by home arena statisticians.
CHAPTER 15. GETTING TO KNOW ENHANCED STATS
Thus, a simple adjustment can be made to the Enhanced Stats to account for this change. We’ll redefine the basic SAT F and SAT A metrics by removing the blocked shots component. Thus, we have the
unblocked shot attempts for and unblocked shot attempts against defined as:
U SAT F = SOG + M S
(15.5)
U SAT A = SOG + M S.
(15.6)
and
With these adjusted definitions for U SAT F and
U SAT A, we can easily compute the remaining modified metrics. For completeness, these metrics are:
U SAT = U SAT F − U SAT A
and
U SAT % = 100 ×
15.2.4
U SAT F
.
U SAT F + U SAT A
87
Table 15.1: Data for EVSV% and EVSH%
Season
2022-2023
2021-2022
2020-2021
2019-2020
2018-2019
2017-2018
2016-2017
2015-2016
2014-2015
2013-2014
2012-2013
2011-2012
2010-2011
2009-2010
2008-2009
2007-2008
EVSV%
EVSH%
.912
.914
.915
.917
.917
.920
.921
.923
.922
.921
.920
.921
.921
.919
.919
.920
.088
.086
.085
.083
.083
.080
.079
.077
.078
.079
.080
.079
.079
.081
.081
.080
Imagine if you were to take the EVSV% and EVSH%
(15.7) values above and add them together (and, for convenience only, multiply that sum by 1000). It turns
out, that the result of this quick calculation is 1000.
That is, if you sum the league averages for EVSV%
and EVSH% you’ll get the number 1000.10
(15.8) Next, imagine calculating this same sum for each of
the 32 NHL teams individually. We’ll post a few of
these here for you from the 2022-2023 season:
Fluctuations from League Average Performance
• Boston Bruins - 1036
• New York Islanders - 1015
• Calgary Flames - 980
During the 2022-2023 season, NHL goalies posted an
• Columbus Blue Jackets - 978.
average save percentage of .912, while NHL skaters
posted a .088 shooting percentage. These numbers
These are just a few extreme examples for NHL
reflect even-strength hockey only. 9 League averteams. What is particularly interesting about these
ages for both save percentage and shooting percentsums is that teams are unable to sustain (for very
age over the past 16 years are reported in Table 15.1.
9 Numbers
computed using Left Wing Lock internal data.
10 Intuitively, this should make sense to you since every shot
on goal must result in either a save or a goal. So, the sum of
EVSV% and EVSH% should be 1 (which becomes 1000 after
we perform our multiplication of convenience).
CHAPTER 15. GETTING TO KNOW ENHANCED STATS
long times) sums that stray too far from 1000. In
2008, Tyler Dellow performed a study on these sums
that revealed that teams generally regress back to a
sum of 1000 given enough time.1112 In Dellow’s simple study he looked at five years of NHL data and
broke each season into four quarters. He computed
the sums (EVSV% + EVSH%) for each NHL team at
the end of the first quarter of each season and then
computed the sums again for the remaining portion
of each season. The results of the study are stunning:
• the 20 best teams from the first quarter had average sums of 1031;
• the 20 best teams dropped to 1005 in the remaining three quarters of the year;
88
15.3
Assumptions of the Model
15.3.1
EV, PP, and PK
You have probably noticed that both goal scoring
rates and shot production rates increase dramatically
when a team is on the power play (P P ) as compared
to when a team is at even-strength (EV ).13 Likewise, when a team is on the penalty kill (P K), the
number of shots they face rises. Rather than allow
these lopsided shooting situations to affect the overall
shot data, it is common practice to remove this data
entirely from the set. Therefore, when we speak of
Enhanced Stats, we will always be referring to game
situations involving even-strength hockey (more precisely, five skaters vs. five skaters).14
• the 20 worst teams from the first quarter had
average sums of 970;
• the 20 worst teams jumped to 998 in the remaining three quarters of the year.
These sums (EVSV% + EVSH%) will be labeled
SHSV. Briefly, these SHSV values provide a measure for how much a team is straying from their expected performance. The take-away is this: a sample
of teams with a high average SHSV is likely to see
their future performance decline; on the other hand,
a sample of teams with a low average SHSV will likely
see a rise in their future performance. This idea of
changing SHSV values is an example of a statistical
phenomenon known as regression toward the mean.
Like all other Enhanced Stats, SHSV can be computed for individual players as well. To perform the
calculation for an individual player, you want to find
the team’s SV% and SH% only for the times when
that particular player was on the ice. These two percentages are known as onSV % and onSH% respectively. To be clear, you are not simply using the SH%
of the individual player.
11 This
site is offline: http://www.mc79hockey.com/?p=2996
original reference to these sums is generally credited
to Brian King.
12 The
15.3.2
Score Effects
Another aspect of a hockey game that can affect shot
differentials is the score. Consider a hockey game
that is close in score (perhaps tied) in the third period. Both teams are motivated to score and continue to throw shots at the opposing net in an effort
to win. Contrast that game situation with one where
a team has a two or three goal lead in the third period. The team with the large lead is likely to employ
a defense-first strategy, thereby reducing their shot
output. Not only does a defense-first strategy lead
to less shots by the team with the lead, it also yields
more shots by the team without the lead. The shot
reduction by the leading team combined with the shot
increase by the lagging team skews any measurement
of shot differentials.
13 Teams can generally stop 92% of SOG at even-strength.
That number drops to 87% when a team is on the penalty kill.
14 Note that the NHL considers the end moments of a game
with a goalie pulled to be a 5-on-5 EV situation (even though
one team has six skaters on the ice). The data at Left Wing
Lock removes these extra-attacker situations (which occur any
time a goalie is pulled) because we believe they don’t accurately reflect what we’re trying to capture when we say “evenstrength” situation. In summary, Enhanced Stats data is computed using 5-on-5 EV situations with all extra-attacker situations removed.
CHAPTER 15. GETTING TO KNOW ENHANCED STATS
There are a number of ways to account for these effects. The most rigid approach is to remove them
from the data set entirely. In this instance, one would
compute shot differentials only when the game is tied.
But, this approach seems contrary to one of the major advantages of using Enhanced Stats: the benefit
of larger sample sizes over small time intervals. An
alternative to throwing out all of the non-tied game
situation data is to use game data when the score
is tied or close.15 A slightly more complicated approach has been suggested recently 16 which adjusts
the shot differentials for situations when a game is not
tied. This proposal has the benefit that it allows you
to keep all of the 5-on-5 data from a game no matter
the score. It should be pointed out that the last of
these three suggestions appears to have the greatest
predictive value (based on testing). This should not
surprise you since it is the method with the largest
sample size. In the end, whichever score effect adjustment you prefer, know that they ultimately will
agree with one another the deeper you get into the
season.
15 The term close here refers to game situations when the
lead is no more than one goal in the first two periods or the
score is tied in the third and overtime periods.
16 http://www.broadstreethockey.com/2012/1/23/
2722089/score-adjusted-fenwick
89
Chapter 16
The ± Statistic - Theory
16.1
Background
± data for an individual NHL player and we’ll plot
more than 1,200 data points!
Very few statistical categories in hockey and fantasy
hockey elicit more groans than the ± category. No
matter what side of the fence you sit on regarding
±, the fact is that it is still used as a standard category by some of the major fantasy hockey providers.
As such, it is our duty to help you understand the
category in as much detail as possible.
16.2
Are Past ± Values Predictive?
The most pressing question one can ask about ± is
whether or not the use of past ± values allow you to
successfully predict future ± values.
Figure 16.1: ± Data in Six Consecutive Seasons
The short and simple answer to that question In Figure 16.1, the relationship between the ± valis no.
ues of consecutive seasons is revealed. Notice how
the data appears to be randomly distributed in this
Let’s back that up with some data. We’re going to plot. If the datasets were highly correlated (implying
look at ± data from consecutive seasons in the NHL. some type of linear relationship), it would be readily
We’ll do this for the most recent six consecutive sea- apparent to us by looking at the plot. Instead, our
sons.
instincts tell us that future ± values don’t seem to
be related to past ± values.
Essentially, we’re creating a plot where the x-axis represents ± data from a previous season and the y-axis But, let’s use a little bit more rigor.
represents ± data from the very next season. Thus,
a single data point on our plot reveals two seasons of Figure 16.1 is staggeringly blunt in its conclusion:
90
CHAPTER 16. THE ± STATISTIC - THEORY
91
future ± values are completely independent of past these same players performed the very next season
± values.1
after posting a ± value of 0 in the previous season.
For those of you who read the chapter on the repeatability of fantasy stats, the R-squared value for this ±
relationship is 0.08. The key takeaway here is
that using past ± values to predict future ±
values will be a complete failure. Surprisingly,
there are fantasy hockey websites that still pretend
that their projected future ± values are meaningful.
If you remain unconvinced by these arguments, we
ask that you look at one more plot of data. We’re
going to be looking at exactly the same data set: the
1,200+ points over the past six consecutive seasons.
But this time, we’re going to look at a specific slice of
the data. We’re going to focus on every player during
this six season interval that posted a ± value of 0 in
one of the seasons.
What do you notice? You should notice that in the
following season, the ± values of these players range
anywhere from -30 to +40; that’s a span of 70 units!
So even looking at one specific ± value from a season
doesn’t give us any hope of predicting what is going
to happen in the following season.
You can perform your own slices of this data at home.
Just use some paper to block out certain regions of
the plot and focus on specific ± values on the x-axis.
In every situation, you’ll see that the ± values in
the vertical direction span a continuous and unpredictable range of values.
16.3
Projecting ±
All hope is not lost. After some extensive testing,
we have developed a method to assist fantasy hockey
managers in their quest to pin down future ± values
for NHL players. You should not be surprised after
reading the earlier sections in this chapter that this
new method does not rely on past ± data at all.
Figure 16.2: ± Data in Six Consecutive Seasons
Our method for making statements about future ±
values involves the use of an enhanced stat known as
SHSV.2 The SHSV stat measures (to some degree)
how lucky or unlucky a player was at even-strength
hockey during the season. As a reminder, to compute
SHSV, you simply add the team shooting percentage
to the team save percentage while that specific player
was on the ice.3
Figure 16.2 reveals that specific slice of data involving
players who posted a 0 value in ± during one of the
six seasons. Obviously, all of the points along the
x-axis are 0, but the y-axis is the interesting part of
the plot. Remember, the y-axis reveals to us how
To be clear, a player with a very high SHSV value has
been on the ice while his team had a high shooting
percentage (lots of even-strength goals) and while his
team has stopped a lot of pucks from going into the
net (few even-strength goals against). A player with a
1 We’re being a little sloppy here. Strictly speaking, we
mean there is no linear relationship between future ± values
and past ± values. We have not excluded a non-linear relationship between the two - however unlikely that may be.
2 This is the official NHL name for the stat. In previous
universes, this has also been known as PDO.
3 Again, these calculations are done at even-strength and,
for convenience, the SHSV is multiplied by 1000.
CHAPTER 16. THE ± STATISTIC - THEORY
92
Figure 16.3: Relationship between future ± and past SHSV
very low SHSV is just the opposite (few even-strength
goals for and lots of even-strength goals against). As
we noted in our two chapters on enhanced stats, most
players will not have consecutive seasons with extreme values of SHSV. That is, since SHSV regresses
to the mean, it would be rare for an individual player
to have back-to-back seasons with a very high (or
very low) SHSV value.
another even-strength stat was correlated.4 Let’s see
how this worked out.
In Figure 16.3, we plot the SHSV values of a player
in one season on the x-axis and the change in ± value
of that same player from the previous season to the
current season. We’re looking for a relationship that
suggests that future ± values have some dependence
on past SHSV values.
The motivation for us to consider SHSV as a possible
indicator of ± values was simple. All ± events are just You can see a reasonably strong correlation (Ra measure of goals for and goals against. Nearly every
4 The ± calculation is also influenced by shorthanded
± event occurs at even-strength, so why not test if
events, but those make up a small fraction of the overall total.
CHAPTER 16. THE ± STATISTIC - THEORY
93
Figure 16.4: Relationship between future ± and past SHSV
squared is 0.39) between past SHSV values and the
change in a player’s ± value from that season to the
next. Looking at Figure 16.3, it is clear that players with very low SHSV values in one season will see
double-digit increased in their ± values the following season. Players with very high SHSV values in
one season will see double-digit decreases in their ±
values the following season.
Things remain a little messy in the middle of the
chart. That is, players with near-average SHSV values, undergo unpredictable changes in their ± values.
That’s ok. The projection of ± values has been elu-
sive for years in the fantasy community. But now we
have a way of projecting big changes for some NHL
players. That’s progress!
In Figure 16.4, we take a second look at the data.
We’ve removed all the data for players that posted
SHSV values near the league average.5 The updated
plot uses only the extreme SHSV values from the previous season.6
5 The
league average SHSV value is 1000.
the purpose of this experiment, we consider extreme
SHSV values to be < 980 or > 1020.
6 For
CHAPTER 16. THE ± STATISTIC - THEORY
Now, this is really interesting. Here we have data
with an R-squared value of 0.60. Nearly 90% of the
players with extreme SHSV in the previous season
saw predictable changes in their future ± values (that
is, players with high SHSV experienced big drops in
their ± and players with low SHSV experienced big
gains). Furthermore, about 70% of the players saw
double-digit changes in their ± from one season to
the next.
Figure 16.4 suggests a method for predicting big
changes in ± from one season to the next: find the
players with extreme SHSV values in the current season and you’ll know which players will see doubledigit changes in their ± values in the next season.
In the Applications version of this chapter, we apply
this theory to the 2023-2024 season.
94
Chapter 17
Possession & Luck Charts
17.1
Introduction
In a broad sense, two of the most important statistical ideas that you can use to analyze a team are
possession and luck. Possession stats tell us which
teams are consistently outshooting their opponents.1
Measurements of luck tell us which teams are playing
beyond their means.
the chart plots possession on the horizontal axis (xaxis) and luck on the vertical axis (y-axis). Specifically, the x-axis is USAT% and the y-axis is SHSV.4
Let’s ignore the bubbles and colors for now and imagine each team is just a dot. If your dot lives on the
right hand side of the chart, then your team consistently outshoots their opponents during games. If
your dot lives on the left hand side of the chart, then
your team is consistently being outshot by their opWith that in mind, we developed a chart a few years ponents during games.
ago to analyze hockey teams. We call this chart, the
Pluck Chart.2
If your dot lives in the top half of the graph, then
your team’s results this season have been boosted by
good luck. If your dot lives in the bottom half of the
graph, then your team’s results this season have been
17.2 Four Types of Teams in muted by bad luck.
the Pluck Chart
These charts contain a significant amount of data to
process. But we are confident that once you practice
with them, you’ll find them extremely useful.3
Easy enough. Now, let’s put it all together. Teams in
the upper-left quadrant of the chart are weak possession teams (they are consistently outshot by their opponents) and their results have been boosted by luck.
We can call these teams weak & lucky. Examples
from this quadrant include the Arizona Coyotes and
St. Louis Blues.
Figure 17.1 is the 2022-2023 version of the Pluck
Chart. We’ll start with a very broad overview and
then work ourselves into the fine details. Essentially, Teams in the upper-right quadrant of the chart are
strong possession teams (they are consistently out1 Teams that consistently outshoot their opponents, more
shooting their opponents) and their results have been
often than not, end up in the playoffs.
boosted by luck. We can call these teams strong
2
You might notice that this name is a not-so-clever confluence of the words possession and luck.
3 We mean this in two ways. You can use the charts to
analyze hockey teams and you can use the charts to gather
intel for use in fantasy hockey.
4 USAT% is a measure of whether or not you’re outshooting
your opponents at even-strength. SHSV is the sum of your
shooting percentage and save percentage at even-strength.
95
CHAPTER 17. POSSESSION & LUCK CHARTS
96
Figure 17.1: Pluck Chart for the 2022-2023 NHL season
& lucky. Examples from this quadrant include the Anaheim Ducks and Chicago Blackhawks.
Boston Bruins and Toronto Maple Leafs.
94% of all Stanley Cup finalist teams over the past
Teams in the lower-right quadrant of the chart are decade came from either the upper-right quadrant or
strong possession teams that had their results muted the lower right quadrant of the Pluck Chart.5
by luck. We can call these teams strong & unlucky. Examples from this quadrant include the Cal5 The 2017-2018 Washington Capitals won the Stanley Cup
gary Flames and the Los Angeles Kings.
Finally, teams in the lower-left quadrant of the chart
are weak possession teams that had their results
muted by luck. We can call these teams weak &
unlucky. Examples from this quadrant include the
with a bubble that was technically in the upper-left quadrant
(when averaged over the entire regular season). But, a deeper
analysis reveals that the Capitals made significant structural
changes to their team that resulted in their bubble living in
the upper-right quadrant from March 1 through the end of
the regular season. This bubble location in the upper-right
quadrant continued throughout the 2018 playoffs.
CHAPTER 17. POSSESSION & LUCK CHARTS
17.3
Bubbles & Colors
97
17.4
An Application: When to
Trade a Hot Goalie
So far, we’ve used the Pluck Charts to organize NHL
teams into four quadrants based on their possession The team chapters will reveal how to use these charts
and luck metrics. But these charts contain much to understand player deployment by modern NHL
more information than that.
coaches and the impact this deployment has on fantasy hockey production.
Each team is labeled as a colored bubble on the chart
(not a dot as we noted earlier for simplification). But, we’d like to take a minute here to describe one
Both the color and size of the bubble tell us details of the many ways that you can use the league-wide
about each team. The color of each bubble, ranging possession and luck chart to your advantage during
from dark blue to dark orange, tells us the team’s your fantasy hockey season.
even-strength save percentage (EVSV%).6 A color
scale to the right of the chart helps you understand This seems to be a good place to discuss what to do
the numbers. The idea here is that if your color is with goalies who are dominant in the first half of a
far from average, then you’re team has experienced fantasy hockey season. In the 2016-2017 season, Deatypical save percentage numbers at even-strength.7 van Dubnyk posted a SV% of 0.941 and GAA of 1.75
The league average EVSV% has been around .915 the through December 31, 2016. He also had a winning
percentage of 66%. Had you owned him from Octopast few seasons.
ber through December, you were likely dominating
The size of each bubble tells you the team’s even- the goalie categories in your fantasy league.
strength shooting percentage (EVSH%). The larger
the bubble, the larger the value. League aver- But, if you held onto Dubnyk (instead of trading
age shooting percentage (for a team and at even- him), you likely got crushed in the goalie categories
strength) is 8.5%. It is uncommon for teams to stray for the remainder of the season. Dubnyk posted a terrible save percentage of 0.908 and GAA of 2.69 from
very far from this number for long periods of time.
January 1, 2017 through the remainder of the season.
Essentially, what is happening here with the bubbles His winning percentage during that time period had
is that we’re taking the y-axis value for each team and dropped to 58%.
splitting it into two individual components (EVSV%
is bubble color and EVSH% is bubble size). And this Why did Dubnyk’s stats drop so catastrophically in
the second half of the season? Could this have been
simple extension creates a very powerful tool.
predicted?910 What action should you have taken as
Finally, we’ve drawn a white dotted line from the a Dubnyk owner?
upper-left to the lower-right of the chart. Generally
speaking, the teams that end up in the playoffs are Let’s take a look at the NHL Pluck Chart (Figure
17.2) for the morning of January 1, 2017. On that
the teams that land above the white dotted line.8
morning, the Wild sat in 2nd place in the Western
Conference with 50 standings points and just nine
losses. Note: in the 2016-2017 season, we used Or6 This is the combined EVSV% of every goalie on the team.
ange & White for the Pluck Chart instead of Orange
7
This is a point of contention among hockey followers.
Some teams actually do have a great goalie (or a bad goalie).
So, this bubble color is catching some of that talent in its metric as well.
8 The white dotted line method correctly predicted 28 of
the 32 (88%) playoff teams over the past two seasons.
9 It was predicted in Episode 61 of the Left Wing Lock
Fantasy Hockey Podcast.
10 https://leftwinglock.com/articles.php?id=2867&
title=Fantasy-Hockey-Podcast-Episode-61
CHAPTER 17. POSSESSION & LUCK CHARTS
& Blue.
98
ners. Most managers would have lined up for a
chance to get Dubnyk in January; and winning that
Minnesota’s bubble can be found way up at the top trade wouldn’t have required getting a star in return.
of the chart. It’s a brilliant dark orange indicating
that the team is getting very lucky with their evenstrength save percentage. It’s one of the largest bubbles indicating that the team is getting very lucky
with the even-strength shooting percentage. And finally, the bubble is on the left hand side of the vertical
indicator of 50% puck possession meaning that the
Wild were consistently outshot by their opponents.
Every piece of information in that January 1, 2017
Pluck Chart is screaming at Dubnyk owners: sell,
sell, sell! It would have taken courage, but the only
smart decision for Dubnyk owners on January 1 was
to trade Dubnyk. Curiously, you could have traded
Dubnyk for anything and come out on top since his
second half of the season statistics were so terrible.
But, given that Dubnyk was posting some of the best
goaltending numbers in the NHL at the time, you
could have easily traded Dubnyk for an average to
above-average goalie and a forward or defenseman
upgrade.
Do you protest? Why on Earth would you trade the
best goalie in the NHL (at the time) for an average or slightly above-average goalie? Why? Because
that goalie you traded for would have outperformed
Dubnyk in the second half of the season. Additionally, you would have been upgraded at forward or
defense in the process. This is how fantasy leagues
are won; making the tough decisions that most managers aren’t willing to make.
Now, let’s take a look at what happened to the Minnesota bubble by the end of the season. Figure
17.3 reveals that the vertical location of Minnesota
dropped from about 1038 to 1018. And since the
size of the bubble was hardly changed, the drop was
almost entirely related to a drop in save percentage
(take notice of how much the color of the bubble has
changed).
This is a great example of how to use a Pluck Chart
mid-season to execute a trade on unsuspecting part-
CHAPTER 17. POSSESSION & LUCK CHARTS
Figure 17.2: Pluck Chart: January 1, 2017
99
CHAPTER 17. POSSESSION & LUCK CHARTS
Figure 17.3: Pluck Chart: April 9, 2017
100
Chapter 18
Player Usage Charts
18.1
Background
The vertical axis is labeled as Quality of Competition. In short, this is a measure of how strong of
an opponent was on the ice when you were on the
The idea for player usage charts was first proposed ice (e.g., were you facing Sidney Crosby or were you
by Rob Vollman in 2011. The concept resulted from facing Ryan Reaves). The actual measurement for
a desire to visualize (with a single graphic) how a this axis is computed using a metric developed by
team’s players were being used on the ice. Consider the Left Wing Lock staff. If you’d like to learn more
these charts to be something of a situational aware- about the metric, please contact us.
ness at the team and player level.
The color of the bubble is a measure of the player’s
The charts do not tell you if a player is objectively Relative SAT. SAT is simply a measure (at evengood or bad at hockey. Instead, the charts tell you strength) of how much you outshot your opponent
whether or not a player has been effective in the spe- by (it’s negative if you were being outshot). Relative
cific role assigned to him by the coaching staff.
SAT is computed relative to how your team performs
when you aren’t on the ice. A darker orange bubble
is used for players with strong Relative SAT num18.2 Description of the Charts bers (these players were effective at driving play; they
are considered puck possession players). Blue bubbles (ranging through lighter shades to darker shades)
There are several pieces of data that go into a player represent players who struggle when it comes to puck
usage chart and it’s easy to get overwhelmed. So, let’s possession (these are players who are on the ice when
take this a step at a time and build up our knowledge their teams are being outshot). It is important to
realize that the color scale is adjusted for each NHL
base before jumping in.
team; two players on two different teams that have
The horizontal axis is labeled as Offensive Zone the same color bubble may have different Relative
Start %. Imagine if you take all the starts of all SAT values.
the shifts for a player. Next, imagine removing any
shift starts that happen in the neutral zone. You The size of the bubble is a measure of your average
are now left with a player’s shifts which start either even-strength time on ice (AEVTOI). Larger bubbles
in the offensive zone or the defensive zone (from his indicate more ice time.
perspective). The Offensive Zone Start % tells you
what fraction of these non-neutral zone shift starts
happen in the offensive zone.
101
CHAPTER 18. PLAYER USAGE CHARTS
18.3
Interpretation
Charts
of
102
the
The charts for each team are described in detail in
each team chapter. In the first few chapters, we spell
out exactly how we’re arriving at each conclusion so
that you can get a feel for how these charts work.
Figure 18.1 represents a typical player usage chart.
The chart is divided into four quadrants (labeled I,
II, III, and IV). Each quadrant represents a distinct
type of player that is commonly found in the NHL.1
When you look at each chart, you’ll notice a black
vertical line and a black horizontal line that represent
the center lines of the chart. Players (represented by
bubbles) that reside on the right hand side of the
chart are more often starting their shifts in the offensive zone, while players on the left hand side of
the chart are more often starting their shirts in the
defensive zone. The location of your shift start is important for many reasons, with an obvious one being
that you are either close to, or far away from, the net
you’re trying to score into.
For players that reside on the top half of the graph,
they are typically seeing ice time against the tougher
players on the opposing teams (these minutes are
sometimes referred to as tough minutes). Players on
the bottom half of the graph typically see ice time
against the 3rd or 4th line opponents (these minutes
are sometimes referred to as sheltered minutes).
Quadrant I (the upper right part of the chart) typically contains players you might refer to as twoway players. They are strong at both ends of the
ice. Quadrant II (the lower right part of the chart)
contains players that are seeing so called easy minutes since they generally start their shifts in the offensive zone and the competition they face is not
strong. Quadrant III (the lower left part of the chart)
contains players that generally start their shifts in
the defensive zone but face competition that is not
1 This should be considered an idealized representation of
what the NHL actually behaves like. The real world is not so
black or white.
strong. Finally, Quadrant IV (the upper left part
of the chart) contains shutdown players who generally start their shifts in the defensive zone and face
tougher competitors.
CHAPTER 18. PLAYER USAGE CHARTS
Figure 18.1: Standard Player Usage Chart
103
Chapter 19
The Relationship Between League
Standings and Goal Differential
19.1
Introduction
Perhaps the strongest indicator of whether or not a team will make the playoffs is their goal differential. To
find the goal differential, you simply take a team’s total goals scored and subtract out the team’s total goals
allowed.
Generally speaking, teams with a positive goal differential will qualify for the playoffs and teams with a
negative goal differential become draft lottery teams.1 There are always exceptions. For example, the
Calgary Flames posted a +8 goal differential this past season but missed the playoffs by two standings
points. In the 2021-2022 season, both the Vegas Golden Knights (+18) and Vancouver Canucks (+13)
missed the playoffs (by 3 and 5 points, respectively) despite having positive goal differentials. In the 20182019 season, the Montreal Canadiens missed the playoffs (by just two points) despite posting a +13 goal
differential. In the 2016-2017 season, the Ottawa Senators qualified for the playoffs with a -2 goal differential
while the Tampa Bay Lightning did not qualify for the playoffs with a +7 goal differential. In the 2015-2016
season, the Boston Bruins missed the playoffs with a +10 goal differential while the Philadelphia Flyers
qualified for the playoffs with a -4 goal differential. The Detroit Red Wings also qualified that season with
a -13 goal differential.
The relationship is not perfect, but the fact that we can very briefly list the teams that violate the rule is
promising. Just six teams (in total) from the six most recent 82-game seasons have missed the playoffs with
positive goal differentials.
1 We’re omitting details from the 2019-2020 season here since the regular season was cut short by 15% and the playoff field
was expanded under unique circumstances. We’ve also omitted the details from the shortened 2020-2021 season for similar
reasons.
104
CHAPTER 19. THE RELATIONSHIP BETWEEN LEAGUE STANDINGS AND GOAL DIFFERENTIAL105
19.2
The Standings vs. Goal Differential Chart
Figure 19.1 plots several years of standings points data versus the goal differential data for NHL teams. Now
we can easily see a clear relationship between the two:
Figure 19.1: Relationship Between Standings Points and Goal Differential
19.3
How Many Goals Equal a Win?
One question that arises naturally when you’re projecting how teams will finish in the standings is: how
many goals of differential is equivalent to a win? Figure 19.1 can help us answer that question.
CHAPTER 19. THE RELATIONSHIP BETWEEN LEAGUE STANDINGS AND GOAL DIFFERENTIAL106
If you find the line of best fit for this dataset, you’ll find that the equation of that line is:
y = 0.3684x + 91.867
(19.1)
Here, x represents a team’s goal differential and y represents the number of standings points that team will
earn. For those of you who remember our discussion of R2 from the Repeatability of Fantasy Hockey Stats
chapter, the relationship above has an R2 value of 0.924.2
This equation is then quite useful. For example, if you want to know approximately how many standings
points a team will earn if they have a goal differential of zero, simply put a zero into the equation for x. The
result is that a team with 0 for a goal differential can be expected to earn approximately 92 points.34
Thus, another way of thinking about things here is that a team that scores at least 92 points in the standings
has good odds of qualifying for the playoffs.5
But, let’s get back to that original question: how many goals of differential is equivalent to a win? A win
gets you two points in the standings. So, we then just need to figure out how what goal differential creates
two extra points in the standings. We simply divide 2 by 0.3684 to produce 5.429. So, if you want your
team to get an extra two points in the standings, they are going to need to generate an additional 5.429 goal
differential. Want to jump four points in the standings? You’re going to need a +11 goal differential.6
2 R2
values can range from 0 to 1. The higher the value, the stronger the relationship between the two variables.
rounding up the 91.867 for convenience.
4 The 91.867 is also known as the y-intercept for those of you wanting to impress your 8th grade math teacher.
5 In the 2022-2023 season, 92 points was the cutoff for qualifying for the playoffs in the Eastern Conference while Western
Conference teams needed 95 points.
6 Again, this is just an approximation. If you double 5.429, you end up with 10.86.
3 We’re
Chapter 20
Anatomy of a Yahoo Pro League
20.1
Background
PRO250 League, $500 to compete in a PRO500
League, $1000 to compete in a PRO1000 League.
They’re back! Yahoo Fantasy Hockey offers users the
chance to compete in Yahoo Pro Leagues for a chance
to win prize money. There are two levels of Yahoo
Pro Leagues distinguished only by the cost of the
buy-in (and subsequent payouts). This section of the
draft kit provides you with a behind the scenes look
at how competitors win money in these leagues. Our
emphasis will be on Rotisserie style leagues, but many
of the tips in this chapter can be applied to Headto-Head (H2H) style Pro Leagues (with the caveat
that Head-to-Head leagues are influenced to a greater
extent by luck than Rotisserie leagues).
You may pay the entry fee using a credit card or other
payment method or, if you have sufficient funds available in your Yahoo Sports Fantasy account, once you
join a league, your Yahoo Sports Fantasy account balance may be debited for the respective entry amount.
You are then entered into a league with 11 other
random competitors. The only choice you have over
which league you join is the draft date and time (e.g.,
you can choose to participate in the 8:00 PM draft
on September 29).
Yahoo Pro Leagues pay out money to the top three
finishers in the standings at the end of the season (or
The data presented here have been compiled over a the top three playoff teams for H2H leagues). Below,
several year period and is found exclusively in this you can find the payout data for the two league types:
draft kit. We hope our users will be able to apply this
knowledge to dominate the Yahoo Fantasy Hockey
Pro League landscape.
20.2
Description of Leagues
Table 20.1: Pro League Payouts
Yahoo Pro Leagues are available in several flavors:
for the 2022-2023 season, payment of an entry fee
was required to register a team. The fee to register
a team depends on the type of league for which you
are registering: $20 to compete in a PRO20 League,
$50 to compete in a PRO50 League, $100 to compete in a PRO100 League, $250 to compete in a
107
League
PRO20
PRO50
PRO100
PRO250
PRO500
PRO1000
1st
2nd
3rd
$120
$300
$600
$1,500
$3,000
$6,000
$70
$180
$360
$900
$1,800
$3,600
$30
$70
$140
$350
$700
$1,400
CHAPTER 20. ANATOMY OF A YAHOO PRO LEAGUE
20.3
Scoring Settings
And now for the goalies:
The Yahoo Pro Leagues use the default scoring settings on Yahoo. For starters, this means that every league will have 12 managers (any less, and the
league is folded before the draft and your money is
returned).
Rotisserie leagues use the settings shown in Figure
20.1 (note: trade deadline date changes with each
season), while H2H leagues use the settings shown in
Figure 20.2. The scoring categories are identical, but
there are subtle differences in how roster moves can
be made.
20.4
Traits of Winning Teams
108
Table 20.2: Goalie Categories
20.5
Category
Raw
Standings
Wins
GAA
SV%
SHO
75.65
2.28
0.921
14.14
9.69
9.77
9.59
10.34
Traits of Average Teams
For comparison, we post the raw scoring data for
average fantasy hockey managers in Yahoo Pro
Leagues.
We’ve analyzed thousands of Yahoo Pro Leagues in Let’s start by looking at the scoring categories for
an effort to determine how winning managers behave. skaters.
Below, we’ll present both the raw data (e.g, how
many goals do winning teams score) and the standings data (e.g., how many standings points for the
Table 20.4: Skater Categories
goals category did the winning team earn) for Yahoo
Category
Raw
Pro Leagues.
Goals
187.84
Note that the team with the most goals receives a
Assists
323.87
score of 12 in the standings for that category. While
+/35.80
the team with the least goals receives 1 point in the
Hits
980.67
standings for that category.
PPP
161.50
SOG
1922.26
Let’s start by looking at the scoring categories for
skaters.
And now for the goalies:
Table 20.2: Skater Categories
Category
Goals
Assists
+/Hits
PPP
SOG
Raw
Standings
209.06
358.29
58.99
1089.23
182.13
2112.56
9.68
9.77
8.53
8.82
9.68
10.19
Table 20.5: Goalie Categories
Category
Raw
Wins
GAA
SV%
SHO
63.37
2.47
0.915
9.87
CHAPTER 20. ANATOMY OF A YAHOO PRO LEAGUE
109
Figure 20.1: Yahoo Fantasy Hockey - Rotisserie
20.6
Interpreting the Data
difference in standings points is only about 2 (8.53 6.5). The main point here is that large differences in
the raw data of categories sometimes result in very
A first glance at the data might lead you to believe minor differences in standings points. The +/- catethat there is a huge gulf between winners and av- gory happens to be one of those categories - and that
erage managers in the +/- category in Yahoo Pro is great news for all of us since +/- is one of the least
Leagues. Winners typically end up at +58.99 and projectable statistics in all of fantasy hockey.
average managers usually end up at +35.80. But if
One of the biggest differences you can make in adoptyou focus on that large difference, you’ll make an iming a strategy for Yahoo Pro Leagues appears to be
portant mistake; the raw data is not the important
the selection of your goalies. We have multiple reafactor here. How many standings points does an avsons for this conclusion. First, some of the biggest
erage fantasy hockey manager get in the +/- category
differences between winners and average managers
(or any category for that matter)? 6.5. So, despite
happen in the goalie categories (remember, the standbeing outperformed in the +/- category by over 23
ings differences are more important than the raw data
units (an actual difference of almost 65%), the actual
CHAPTER 20. ANATOMY OF A YAHOO PRO LEAGUE
110
Figure 20.2: Yahoo Fantasy Hockey - H2H
differences). Beyond just the data we present here,
we’ve also examined the drafting habits of winners of
Yahoo Pro Leagues. All of the winning managers in
these types of leagues, the most commonly drafted
position in the 1st round turned out to be the goalie
position. This is no coincidence.
A common mistake of managers in fantasy hockey
leagues is trying too hard to dominate a particular
category. Consider the following scenarios. Which
team gets more standings points in a Rotisserie
league: Team A with 252 goals (leading their league)
or Team B with 221 goals (leading their league)? The
answer, of course, is that both teams earn the same
number of standings points. One could make the argument that Team A may have “over drafted” the
CHAPTER 20. ANATOMY OF A YAHOO PRO LEAGUE
goals category at the expense of some other offensive category. Instead of trying to dominate a particular category, we strongly recommend a balanced
approach to drafting in Yahoo Pro Leagues. If you
notice the average standings points for winners, most
of the values are approximately 10 (not 12!). This
suggests that a balanced approach where you aim for
3rd place in every category would be sufficient for
winning many Yahoo Pro Leagues.
This balanced approach is the basis for the FSI (Fantasy Strength Index) ranking system pioneered by
Left Wing Lock. Instead of attempting to be best
at a few categories and risking progress in others, we
suggest being “good enough” in all categories.
To this end, we recommend using the FSI system as
your drafting system in all Yahoo Pro Leagues. You
might not have the sexiest roster at the end of the
draft. But if you’ve been diligent in making sure that
your team hits the F SI = 100 goal in each category,
you should be in a strong position to win money in
any Yahoo Pro League.
20.7
An FSI Draft Spreadsheet
One of the most useful items you can have at your
fantasy hockey draft is a spreadsheet that autocomputes your FSI totals in each category as you
draft in real-time. We’ve created a spreadsheet that
will perform all of these calculations for you during
your draft. The only required input from you is a
simple copy/paste of the data for a particular player
you’re drafting from the Left Wing Lock draft kit
spreadsheet. You’ll also want to input the pick number for your team for the 1st round of the draft.
This FSI Draft spreadsheet will compute (on-the-fly)
the FSI totals in each scoring category as each round
passes by. You’ll be able to keep an eye on each
category’s FSI value as you draft and you’ll always
know which categories you are ahead/behind in for
each round.
111
Remember, in these Yahoo Pro Leagues, an FSI score
of 100 in each category is your ultimate goal. Don’t
forget that we do not make projections for the +/category in the draft kit and, thus, we do not track
your +/- FSI score during your draft. For a reallife demonstration of how such a spreadsheet works
during a live fantasy hockey draft, be sure to check
out Figure 21.1 in the next chapter.
The FSI Draft spreadsheet is available to you (free
of charge) in the Draft Kit Portal at the Left Wing
Lock website. A preview of this spreadsheet is shown
below. Note that all orange boxes require input of
some kind. Please contact us if you have any questions on how to use it during your draft. We believe it
will provide you with a powerful tracking advantage
during your draft.
CHAPTER 20. ANATOMY OF A YAHOO PRO LEAGUE
Figure 20.3: FSI Draft Spreadsheet
112
Chapter 21
Fantasy Strength Index - FSI
21.1
What is it?
• We know how many PIMs you’ll need to beat
out 80% of your competitors
The Fantasy Strength Index (FSI) is a proprietary
player ranking system developed by Left Wing Lock,
Inc. The ranking system is strongly data-driven
and is customized to your specific scoring system.
Leagues that include Hits and SHG would have different rankings from those that include Blocked Shots
and TOI. These differences are reflected in the FSI
created for your league.
• We know the average final standings position of
all teams that drafted Alex Ovechkin
You might be alarmed by the statements above, but
you shouldn’t be; we’re on your team. The above list
is a just a small fraction of the data at our disposal
as we help you prepare for your fantasy hockey draft.
21.2.2
21.2
How is it Created?
How Do the Other Guys Rank
Players?
Poorly. Believe it or not, the other guys never ask
you for your scoring system. No, I’m serious. They
believe in the one size fits all draft kit. Your league
Every year, our staff compiles data on tens of thou- uses Blocked Shots? Too bad, the other guys don’t
sands of fantasy hockey leagues. To give you an idea include those stats in their player rankings. So, how
of some of the data we have access to, consider the do they rank players then?
following:
The standard approach to ranking players used
by other sites is to simply compare Player X’s goal
• We know how many goals every fantasy hockey totals to the goal totals of the player who scored
team scored
the most goals last year. For example, Steven
Stamkos scored 60 goals in 2011-2012. His ranking
• We know the average number of SOG recorded in that category would be 100 (or 1, depending on
by every 3rd place team
whatever scale you choose to use). Alex Ovechkin
scored 38 goals last season and his ranking would be
• We know the best, worst, and average GAA for 100*(38/60) which turns out to be 63. This process
is repeated for various scoring categories and all of
every manager that won a fantasy league
21.2.1
Introduction
113
CHAPTER 21. FANTASY STRENGTH INDEX - FSI
the individual categories are added up in the end to
determine the rankings of the players. One of the
many problems with this approach occurs when the
league leader is a statistical outlier. Last season,
only two other players scored more than 40 goals.
So, this entire scale is based upon the goal scoring
of one player in one season. Furthermore, this type
of arrangement ignores all of the data from real-life
fantasy hockey leagues. Instead of ranking fantasy
players based on how they help your fantasy team,
this approach ranks fantasy hockey players based on
scoring output of one real-life player. Finally, the
sites that use this approach never ask you for your
scoring system. If they don’t include Hits in their
ranking system (but your league use that category),
then the rankings are going to be worthless to you.
You’ve just wasted $20.00.
114
contribution in the goals category. We get it by
taking 100*(60/215) to arrive at 27.9. Note: we
multiply by 100 so that each FSI rating is out of
a possible total of 100. Therefore, Stamkos’ FSI
rating (in the goal category) would be 27.9, and this
was the highest FSI rating (in the goal category) of
all NHL players last season. Better yet, you have
a quick way of knowing what fraction of the total
goals you’ll need to win (in this example, you’re at
27.9%).
Comparing the 27.9 rating to the 100 rating
used by the other guys, you can instantly tell which
system is more useful. The 100 score tells you
nothing outside of the fact that Stamkos scored the
most goals last season (but you already knew this!).
The FSI, on the other hand, not only tells you that
Stamkos scored the most goals, it tells you how much
he contributes to your overall team goal of 215 goals.
Our staff uses this process on every NHL player and
in every possible scoring category (for skaters and
21.2.3 Why is the FSI Better?
goalies alike). There are some subtleties introduced
(particularly with some of the goalie categories),
The FSI, first and foremost, requires your exact but we won’t hammer you with the details here.
scoring system as the input to the system. Any It is enough to understand that we’ve developed a
ranking system that does not take your scoring ranking system that uses tens of thousands of fantasy
setting into account is worthless to you.
league data and, more importantly, is tailored to
your specific scoring settings.
Because our staff has access to the data noted
in Section 26.2.1, we know how many goals your
team will need to win the category or finish in 4th
place in the category. Likewise, we know many 21.3
How Do I Use it in My
assists you’ll need, how strong of a GAA is required,
Fantasy Draft?
and how many SOG are required for you to win
your league. Let’s focus on goals for a moment.
Instead of basing our rankings on one player in
21.3.1 Points Leagues
the NHL, we choose to use the data on goals from
tens of thousands of fantasy hockey leagues. Let’s
assume that the average winner of a fantasy hockey A points league is one in which assignment of points
league needs 215 goals to win his/her league (that’s a are made to various categories and the winner of the
made-up number). That means you need to assemble league is determined by which manager has the most
a team that can achieve that number. Stamkos, points at the end of the regular season. These types
with his 60 goals, contributes tremendously to this of leagues generally do not have playoff systems and
total. But, we’re not interested in superlatives, we the commissioner is responsible for setting up the
want to know a mathematical number for Stamkos’ scoring system in such a way that goals count for 2
CHAPTER 21. FANTASY STRENGTH INDEX - FSI
115
points, assists count for 1 point, saves count for 0.2 have an FSI ranking for every scoring category that
points, etc.
your league chooses. Given the league settings noted
above, a skater on your spreadsheet would have an
The FSI ranking system for a points league is FSI ranking for G, A, SOG, and PPP. A goalie on
fairly straightforward. In your spreadsheets, you’ll your spreadsheet would have an FSI ranking for W,
see a column marked as TFSI (total FSI). This is the GAA, and SV%.
ranking column and players with the highest TFSI
values are the best players to pick in your league. You might be tempted to add these FSI rankAs you draft, you’ll want to fill up your roster (LW, ings together and form a total FSI ranking. But
RW, G, etc.) with the players who provide you with we caution you against relying too heavily upon
the highest TFSI values.
this strategy for the following reason: players with
really high FSI rankings in some categories might
Be aware that we have removed the +/- cate- have a fairly low ranking in another category. While
gory from consideration when developing this TFSI you’re forming your team during the draft, you
score. The +/- category is one of the least repeatable don’t care about total FSI rankings, you care about
statistics in hockey; that is, knowing a player’s +/- making sure your team is strong in many categories.
value in one season provides you with very little So, it is more useful for you to use the individual
predictive power in projecting his +/- for the next FSI rankings for the particular categories than it
season.
is for you to use a total FSI ranking. If you want
alternatives, consider using the newly developed
PR column in our spreadsheets (it stands for
Performance Rating). This performance rating does
21.3.2 Category Leagues
the work for you and removes any doubt as to which
players are best overall contributors to your league’s
Category leagues use scoring systems in which the scoring system. Best of all, it minimizes the impact
goal is to do as well as possible in several different that one-category wonders have on most ranking
scoring categories. Typically, these types of league systems.
are referred to as Rotisserie and/or Head-to-Head
leagues. Consider the following example involving a Finally, because of how the FSI is formulated,
12 team league using the categories of G, A, SOG, the ideal scenario is to draft a team with an FSI
PPP, W, GAA, and SV%. In a Rotisserie league, score (in each category) that is as close to 100 as
the manager accumulating the most G that season possible (in a standard 12-team, 6F, 4D, 2G starting
would be given a score of 12 in that category, while position setup). For example, if you draft a team
the manager with least number of G would earn a with a starting roster that has a total Goals FSI (or
1. This process is repeated for all of the scoring G FSI) close to 100, your team is likely to finish in
categories. The individual scores of each category the top three of that category. If your league has
are added up at the end of the season and the less or more than 12 teams, your aim will be to keep
manager with the highest overall score is deemed your FSI scores as close to each other as possible.
the winner. Head-to-Head leagues work in a similar
manner, except that you’ll be competing against one
manager each week and the category winner will 21.3.3 A Real-World Fantasy Draft
receive 1 point (for each category they win).
Example
The FSI ranking system for these leagues is far
more complicated. Instead of one single FSI ranking Mike, from the staff of Left Wing Lock, joined a
(as given in the Points leagues), each player will public league in the Yahoo system this season so
CHAPTER 21. FANTASY STRENGTH INDEX - FSI
that he could track his draft while employing the
FSI ranking system. This particular league was a
category league that used the following categories:
the skater categories were G, A, +/-, PIM, PPP,
SOG and the goalie categories were W, GAA, SV%,
and SHO.
A draft spreadsheet was created in advance so
that Mike could track the FSI contributions of
every player during each round. As each player was
added to the spreadsheet, the FSI contributions were
automatically updated in every category. Thus, if
Round 8 came along, Mike would immediately know
his strengths/weaknesses in every league category
and this would help him make crucial drafting
decisions based on logic and data instead of emotion
and hunches. We strongly encourage all category
league participants to create a similar spreadsheet
for their drafts.
and PPP might prove problematic. You might notice
that Halak’s FSI contributions don’t match those in
the spreadsheet. There is a very good reason for this.
Mike is expecting to get 134 starts from Luongo and
Lehtonen. Thus, he can’t count all of Halak’s totals;
he can only count a fraction of them since Halak will
only provide 30 starts toward his team totals (164
possible starts - 134 starts leaves 30 starts for Halak).
Because he had this spreadsheet doing the math for
him, Mike was able to keep his team balanced across
many categories in real-time. This is just one of the
ways that the FSI ranking system (combined with
an FSI spreadsheet at the ready) can help you make
on-the-fly decisions during your high energy draft.
21.3.4
If your league imposes caps on the number of
games you can use at each position (and that’s
probably a smart way to set up your league), then
you should only enter the anticipated starting roster
into your spreadsheet. In Mike’s league, the starting
roster consisted of: 2 C, 2 LW, 2 RW, 4 D, and 2G.
Each position was limited by the following caps: 164
C games, 164 LW games, 164 RW games, 328 D
games, and 164 G games. Ignoring injury, you should
easily reach these caps at the skater positions while
using your starting roster. In goal, on the other
hand, it would be rare for two goalies to combine
for more than 140 starts - therefore, a third goalie
needs to be drafted and his stats should be used in
determining your total FSI values on draft day.
As Mike made each selection, he copied the
FSI data from his draft kit spreadsheet into this
pre-formatted FSI spreadsheet. Here is how Mike’s
draft went (note: only starting roster players are
included in the team FSI calculations): Looking at
the scoring categories for Mike’s league, it is clear
which categories he should be strong in and which
categories might give him some trouble. G, A, SOG,
W, GAA, SV%, and SHO are projected to be strong
categories for Mike based on his draft. While PIM
116
FSI Spreadsheets
Starting in 2018-2019, FSI spreadsheets will be
included in all Skater Spreadsheet downloads for
non-points leagues. The FSI calculator is linked to
your projections/rankings spreadsheets so that you
can quickly see how you are performing in real-time
during your draft.
To use the FSI spreadsheet, enter the appropriate values into the orange-highlighted cells. You
can simply type in a player’s LWLRANK (from
the Projections Sheet) as you draft them and the
calculator will auto-populate all of the necessary FSI
values for each category, the player’s name, and the
player’s position. As you enter more drafted players,
the calculator will keep running totals for each of
these categories.
CHAPTER 21. FANTASY STRENGTH INDEX - FSI
Figure 21.1: 2013-2014 Fantasy Hockey Draft Using FSI
117
Chapter 22
General Advice for Newcomers
This chapter will serve as a general strategy guide for
your draft and some in-season tips. What you read Conventional wisdom pushes you to form the
here is certainly not the only approach to drafting duo and pick Player A. But does this duo hold any
and playing fantasy hockey, but it is a successful one. advantage? No. Player B, statistically, produces on
the same level as Player A. Thus, from a fantasy
hockey perspective, there is no benefit to choosing
Player A over Player B. But, it gets worse for
22.1 Should I Draft Linemates? conventional wisdom followers. Imagine if Ovechkin
suffers an injury. As Ovechkin’s linemate, Player
A will typically see a drop in production while
22.1.1 Background
Ovechkin is on the IR. An injury to one half of a
duo on your fantasy team affects two players. What
Drafting line mates, or duos as they are sometimes would have happened if you had drafted Player B
called, is oft-cited advice in fantasy hockey circles. A instead? Nothing. Drafting duos provides no statisduo is a pair of forwards that play on the same line in tical advantage over drafting non-duos; moreover, it
the NHL. Some examples of well-known duos would actually increases the damage to your team should
be Sedin-Sedin and Getzlaf-Perry. The idea here is one of the linemates suffer an injury because those
that if the line is dominant, you will reap the rewards two players are intimately tied together.
of having both players on your fantasy hockey team.
Another interesting reason to avoid drafting
linemates (when you don’t have to) is that the
ownership of linemates inherently produces more
22.1.2 What Do We Advise?
schedule conflicts for your fantasy roster. If you’re in
a league that doesn’t place a limit on games played,
Drafting duos is not a part of the strategy recom- then drafting linemates may actually work against
1
mended by the Left Wing Lock staff. Imagine that you.
you have drafted Alexander Ovechkin in round 1 of
your draft and it is now turn for your second pick.
You have your eye on two players: Player A and
Player B. Both players have nearly identical numbers
over the past two seasons, have comparable histories,
1 Thanks to Left Wing Lock user Jeremy G. for writing to
and are of the same age. Player A is Ovechkin’s
us about this schedule conflict idea.
linemate; Player B is not.
118
CHAPTER 22. GENERAL ADVICE FOR NEWCOMERS
22.2
Is the Pre-season Impor- 22.3
tant?
119
What Happens After Age
27?
As mentioned in Section ??, skaters over the age of
27 generally begin to show a decline in their shooting percentage and therefore are subject to declines
This one is quick and easy: don’t be swayed by
in their overall goal scoring output. Note that we
solid or poor individual performances during the prewent through every player projection in our spreadseason. They mean virtually nothing when it comes
sheet and applied a “tax” on the point production by
to regular season performance. For starters, teams
players over the age of 27.
are not facing tough opponents, line combos are fluid,
and some of the team strategies are experimental. Finally, the sample size is simply way too small to be
useful.
22.4 Trading
22.2.1
Stats
22.2.2
Injuries
One caveat: you SHOULD pay close attention to injuries during the pre-season. And what better place
to keep an eye on those than the Left Wing Lock
Fantasy Hockey Newsfeed?2
22.2.3
Line Combos
Generally speaking, line combinations in the preseason are fluid. You have younger guys battling for
positions, but most of the vets don’t see much action. It is difficult to determine where players will
end up based on a few pre-season games. Your best
option is to pay attention to the Left Wing Lock forum as active discussion of all teams will be at your
fingertips.3
22.2.4
Contract Holdouts
Be wary of drafting players with extended contract
holdouts, especially if they’ve missed camp and preseason.
2 http://leftwinglock.com/news/
3 http://leftwinglock.com/forum.php
22.4.1
Introduction
If you’ve visited our forum lately, you’ll notice that
the number one sub-forum at the website is the
Trade Advice forum. This is no surprise. Knowing
who and when to trade for what is probably the
most important, yet least understood, aspect to
succeeding at fantasy hockey. No one has all the
answers and those who claim to are full of crap.
The line you’ll hear over and over again is: buy
low and sell high. But this strategy is more easily
discussed than it is implemented. This off-season,
the staff at Left Wing Lock set out to change that.
In an effort to assist you in your trading endeavors,
we’ve created two new tools at the website.
22.4.2
Don’t Send Insulting Trade Offers
You run the risk of losing the chance of ever trading
with that manager for the rest of the season. You
might even turn off other managers from trading with
you (there is lots of chatter behind the scenes in the
leagues I’ve been a part of). Instead, make your first
offer (presumably favoring your side) and attach a
note with the offer letting the manager know that
CHAPTER 22. GENERAL ADVICE FOR NEWCOMERS
120
this is a suggested offer and that other players and 22.5
The Squeeze Play
deals are on the table. You don’t want your offer to
close off the line of communication; you may need
This section covers a trick that can be used right
this manager later.
at the end of the season by diligent managers. At
Left Wing Lock, we call this the squeeze play. This
particular trick only works in leagues which put a
22.4.3 Trade Good Players to Teams cap on the number of games played a manager can
employ at each roster position. If this applies to
That Don’t Need Them
you, read on to learn how this trick can help you win
your league.
If you want good players to come your way in a trade,
you’re going to have to give up good players. One Note that some leagues (Yahoo, e.g.) go out of
particular strategy that can be effective for a number their way to explain this scenario in their rules
of reasons is to trade your good players to teams that section - so it is by no means cheating.4 Other
don’t really need them. Now, what teams wouldn’t managers are going to take advantage of this, so you
need good players? Well, they all need good players, need to make sure you do too.
so what do I really mean here? Imagine you are in a
position to trade a very good LW and you’re looking The squeeze play, in a few words, allows a manager
for a defenseman in return. If you trade this LW to to play more games at every roster position than
a team with bad LWs, then you are going to make the Yahoo caps suggest. In one of the leagues I’m
one of your competitors even stronger. The better involved with this season, the league settings allow
approach (if you can pull it off) is to trade the LW for the following number of games played:
to a team that has solid LWs (but those LWs are not
as good as the one you are trading to them). Their
• C: 164
team improves - but not as much as the team with
weak LWs would have improved. This allows you to
• LW: 164
get your defenseman (thereby improving your squad)
without improving your competitor’s squad by very
• RW: 164
much.
• D: 328
• G: 164
22.4.4
Don’t Be Afraid to Overspend
Also in this particular league, we have starting roster
If you’re trading with a team that has no chance slots for 2 centers, 2 left wings, 2 right wings, 4
of catching you in the standings, you should not be defensemen, and 2 goalies.
afraid to overspend to get the player you want. As
long as the trade makes your team stronger over- One example of how to make the squeeze play
all, overspending can actually be an effective trading happened to me recently at the position of right
strategy, particularly when used later in the season. wing. I had already used 163 games at the right wing
Trades are not about getting fair value; trades are not position and therefore had one game remaining. I
about how good the players are in real life. Trades are then waited for a night when I had two of my right
about winning your fantasy league. Make the moves wings playing on the same night. I played both of
4 http://help.yahoo.com/kb/index?locale=en_US&y=
that make your team stronger overall and ignore the
PROD_ACCT&page=content&id=SLN6836
reactions of all other managers.
CHAPTER 22. GENERAL ADVICE FOR NEWCOMERS
121
the right wings that night and I earned points from
both, despite only have one game left according to
Yahoo. This happened for me on April 2, using
Dany Heatley and Corey Perry. I was able to earn
12 total fantasy points instead of the 4 (or 8) I would
have earned by only starting one of those players.
suffers an injury. Always make sure that your
starting players play as many games as possible.
Your starting players are your best 6 forwards and
best 4 defensemen (on a 10 man roster). It makes
no sense to let a lesser quality player use up a game
that was intended for a higher quality player.
This works for all of the positions. In my particular league, I should be able to pick up 3 extra
forward games (one at C, LW, RW), as many as
3 extra defense- men games (to max this out, I’d
wait until I had only 1 game remaining and then
pick a night when all 4 of my defensemen were
starting), and 1 extra goalie game. Thus, it is
possible for my team to earn points for 7 extra
games played - a total that can make the difference between a 1st and 2nd place finish. Think
about that - 7 extra games played is the equivalent
of extending your fantasy hockey season by a full day.
BUT...in the case of goalies, the situation is
slightly different. While again, you should always
start your top two goalies in ALL situations, goalies
in real hockey never play 82 games. Thus, you will
always have leftover, unused games unless you let a
bench goalie or two play some games. The amount
that you have them play depends on how often your
starting goalies play. You may find that you need
to allow for anywhere between 10-50 games for your
bench goalies during the season. Waiting until the
end of the season would be a mistake!
Don’t expect the ideal scenario I described above
Types of Drafts
to just fall into your lap. You’re going to have to 22.7
plan and prepare to have any shot of making this
method succeed. But, if you’re willing to put in
a 20 minute effort into analyzing your roster with 22.7.1 Auto-Draft Advice
some 3rd grade level mathematics, you can find the
squeeze points that most fantasy hockey managers If you play in a league which features an Auto-Draft,
don’t know about. Good luck!
then a computer system will automate the draft
selections for all of the teams. While this option is
convenient and a time-saver, we recommend that you
choose to play in a league with a Live-Draft. The
22.6 How Do I Use My Bench Live-Draft option affords you much more control over
the choice of your players and it is usually a lot of fun.
Players?
In most leagues, you will be required to draft some
players who fill up the bench spots on your roster. A
common mistake is for managers to use bench players
as substitutes on nights for when their starters are
not playing. If you are using the standard settings
(which caps the number of games you may start
a player at 82 games), then you should never use
your bench players as substitutes for your starting
forwards and defensemen. The only time you should
substitute a bench player for a starting forward or
defenseman, is when one of your starting players
In the event that an Auto-Draft is your only
choice, we highly recommend that you pre-rank the
players ahead of time. If you choose not to pre-rank
your players, Yahoo (or your other FH system) will
use their own player rankings to determine your
selections. Later in this guide, we’ll detail the pitfalls
of this yearÕs Yahoo player rankings. But, in the
past, Yahoo has made severe mistakes including
allowing retired players to be on the rankings list
and confusing Mike Richards with Brad Richards (in
2006-2007, ouch!), to name a few.
CHAPTER 22. GENERAL ADVICE FOR NEWCOMERS
We’ll provide detailed player rankings at all
positions for you later in this guide to help you
create an Auto-Draft rankings list.
22.7.2
Live-Draft Advice
The single most important piece of advice for a
manager participating in a Live-Draft is to show up!
It always amazes me how many managers fail to
show up to a Live-Draft and leave their team to the
mercy of pre-rankings. With that said, read on to
understand more about how Live-Drafts work.
When you show up to your Live-Draft, you
will have an open window on your computer screen
which lists the other managers, the players available
(players already taken will be greyed-out), and a
running list of roster positions you have already
filled. Each manager gets two (2) minutes for their
selection. The time goes by quickly, especially if
a number of managers do not show up. Since the
draft is always a snake-draft (reverse order for each
round), it is possible for you to have to make two (2)
selections in a very short span of time. Be prepared.
The Live-Draft format is all click-and- drag, so it is
very easy to understand in realtime.
We recommend having your pre-ranked players
list (either our’s or your own private list) printed out
on paper and at your desk while you participate in
the draft. It is often helpful to have a cold beer on
hand too. A highlighter or red pen is a great tool
for crossing out players that have been taken from
your wish list. Also, keep a few blank pieces of paper
nearby to make notes about what other teams are
up to (e.g. Team A has already taken two goalies
early, they likely wonÕt pick another until late in
the draft).
122
Chapter 23
Using the Left Wing Lock Website
Don’t let the purchase of your draft kit be the last 23.1.2 Line Combinations
step in your preparation for the draft. There are
many resources at the Left Wing Lock website that
will assist you in the draft and throughout the season. We publish daily line combinations for all NHL
teams. Our line combination data is not a simple
depth chart of four lines that you find on most fantasy hockey sites. We start with raw, official NHL
data and compute the frequency with which every
23.1 The Tools
line combination is used in every game. What does
this mean for you? It means you have free access to
One of the strongest aspects of our website that sets how every player is used in every situation in every
us apart from other fantasy hockey websites is the game.
fantasy hockey tools section. Below, we’ll describe
But that’s not all. You can now search for line combihow to use each tool to maximum advantage.
nations for any game played during the NHL season.
And you can search the line combinations for a specific player.
23.1.1
Starting Goalies
Since 2006, we have published the starting goalies
for all NHL teams on a daily basis. We are the
longest-running and most accurate starting goalies
website. The goalies are updated throughout the day 23.1.3 Random Draft Order Generator
and have an accuracy level of 99.9%. What does this
mean? It means we’re not perfect. We will get some
goalies wrong (usually due to very late, unannounced
changes because of illness). Last season, we missed Need to set a random draft order for your fantasy
draft? Use our random draft generator tool. We
one goalie out of 2,542 entries.
randomize your draft order in a fair and transparent
Since the start of 2014-2015, users have been able to manner. The results are immediately emailed to all
scroll through to any future (or past) date they want managers in your league to avoid any potential funny
to see which goalies are expected to start.
business.
123
CHAPTER 23. USING THE LEFT WING LOCK WEBSITE
23.1.4
Line Production
Use this tool to find out (on a daily basis) which lines
in hockey are producing the most goals.
23.1.5
Line Matching
This tool allows you to figure out which opposing
players are on the ice when Sidney Crosby is on the
ice, for example. Basically, you can find out which
lines teams use to face opposing lines.
124
App has had the capacity to send you push notifications for any NHL players you wish to sign up for.
23.1.9
Email Alerts
Don’t have an iPhone? Sign up for our email alerts.
You’ll receive an email whenever the status of a starting goalie has changed at our website.
23.1.10
Roster Maximizer
The roster maximizer tool reveals to you how many
schedule conflicts you will introduce to your fantasy
23.1.6 News Feed
hockey team by adding another roster player given a
set number of starting positions and a known number
We publish a near-real-time news feed of all notes of rostered players. The tool has use both during your
relevant to fantasy hockey. This includes injury up- fantasy hockey draft and during your fantasy hockey
dates, contract updates, trades, illness, roster up- season (waiver wire & trades).
dates, and more.
23.1.11
23.1.7
Player Rankings
Player Profiles
Our player profiles are quickly becoming a hot commodity in the fantasy hockey community. Consider
it a one-stop location to find out everything about a
particular player: recent stats, season-long stats, line
mates, time-on-ice charts, fantasy trade value, and
more.
We publish fantasy hockey player rankings - customized to your scoring settings - and updated on
a daily basis.
23.1.12
Weekly Schedule
Want a quick look at the entire weekly NHL schedule
in a grid format? How about next week too? This
is the tool for you. Find out which teams play the
23.1.8 iPhone App
most games in any given week. Additionally, this tool
allows you to quickly assess the strength of schedule
Our iPhone App, the only starting goalies App on for each team and determine which games are most
the mobile market, allows you to receive push no- likely to produce a lot of PIMs.
tifications when goalies of your choosing have been
confirmed as the starting goalie for today’s games.
It also includes a look at the line combinations (EV,
23.1.13 Team Pages
PP, & PK) for all NHL teams.
Since the start of the 2014-2015 season, the iPhone
Similar to the player profiles but for teams instead.
CHAPTER 23. USING THE LEFT WING LOCK WEBSITE
23.2
The Forum
The forum now has over 10,000 members and over
100,000 posts. This is a great resource for those of
you who want/need a community to discuss your fantasy hockey trades, conquests, and frustrations.
23.3
Site-wide Chat
Along with your forum account, you’ll have access
to a site-wide chat room to discuss anything from
whether David Backes will play center or right wing
to what you should get your significant other for the
holidays. Come join us, we have a great group to
interact with.
23.4
The Articles
Of course, for the 2023-2024 season, the staff at Left
Wing Lock will be providing you with advice articles
on fantasy hockey on a daily basis. Whether you’re
looking for advice on which players will bring you the
most PIMs this week, which goalies have the best
match ups, or which star player will start to see a
decline in his goal scoring - it will be covered in our
Articles section of the website. Feel free to chime
in with your comments (your forum login credentials
will work for our Articles section too!).
125