nderstanding the NHL Draft JUST a lLttle Better

UNDERSTANDING THE NHL DRAFT
JUST A LITTLE BETTER
By Corey Pronman
Coverage
Three topics:
1. Late birthdate effect
2. Drafting skill and role of chance
3. Scouts vs. Stats
Topic 1: Late Birthdate Effect
 Hockey has a unique rule regarding draft eligibility--- Players born after
September 15th are pushed back a year.
 Some recent big-name examples of players’ draft year pushed back:
 Leon Draisaitl
 Jack Eichel
 Taylor Hall
 Seth Jones
 Auston Matthews
 Nolan Patrick
 Sam Reinhart
 Matthew Tkachuk
Late Birthdate Effect
 Hypothesis: The late birthdate rule is impact player evaluations.
 Set-up:
 Draft data from 1990-2010, first year eligible, up to pick 210
 All leagues
 All skater positions
 Split data into late birthdate players (“LBD”) and non late birthdate
players (“NLDB)
 Omit pick slots where both groups aren’t represented
 Compare production by draft slot for both groups
Late Birthdate Effect
Non Late Birthdate vs. Late Birthdate Prospects Points/GP by Draft Slot 1990-2010
NLBD Pts/GP
5 per. Mov. Avg. (NLDB Pts/GP)
LBD Pts/GP
5 per. Mov. Avg. (LBD Pts/GP)
1.20
Points/ Game
1.00
0.80
0.60
0.40
0.20
0.00
1
21
41
62
82
103
Draft Slot
123
145
167
191
Late Birthdate Effect
 Breakdown by month
 Set-up
 Every CHL forward picked from 1990-2010
 First year eligible
 Had a CHL points/gm between 1.0-1.5 and > 20 GP
Late Birthdate Effect
NHL Points/Gm by Month
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
SEPT LATE
OCT
NOV
DEC
JAN
FEB
MAR
APR
MAY
JUN
JULY
AUG
SEPT EARLY
Guess These Players: All Current CHL Forwards
Player
Points/GP
Prim Pts/GP
Birth Date
Mystery Player A “Draft-1” 1.42
1.10
Late Sept
Mystery Player A Draft Yr
1.39
0.91
Late Sept
Mystery Player B Draft Yr
1.6
1.32
June
Mystery Player C Draft Yr
1.48
1.17
August
Guess These Players: Answers
Player
Points/GP
Prim Pts/GP
Birth Date
Nolan Patrick “Draft-1”
1.42
1.10
Late Sept
Nolan Patrick Draft Yr
1.39
0.91
Late Sept
Pierre-Luc Dubois Draft Yr
1.6
1.32
June
Nick Suzuki Draft Yr
1.48
1.17
August
Late Birthdate Effect
 Takeaways
 Always look at the month a player was born
 Draft -1, Draft +1 can be misleading terminology
 Late birthdate effect is not determinative, but it is important
 Simple adjustment, major implications
Topic 2: Drafting Skill & Role of Luck
 Drafting is very important to NHL teams, most contenders have a
high % of roster acquired through the draft
 Source of major investment by NHL teams, clubs spend up $2-3 MM
per year on scouting
 Given importance and investment, two important questions
1. How much consistency is this year to year in draft results?
2. How much does luck affect draft results?
Drafting Skill & Role of Luck
 Question 1: How much consistency is this year to year in draft
results?
 Set-up:
 Assign each draft slot from 1-210 an expected value point, based on games
played
 Assign each draft pick from 1990-2010 a surplus/deficit value based on
actual games played minus expected games played
 Example: Scott Gomez was top 10 from the 1998 class in games played,
picked 27th overall, but played twice as many games as expected as
someone picked 27th.
 Credit: Neil Paine at 538.
Drafting Skill & Role of Luck: Minnesota/Dallas Franchise
Dallas Drafting Value Year 1 vs. Year 2
10
8
Year X + 1
6
4
R² = 0.2294
2
0
-4
-2
0
2
4
-2
-4
Year X
6
8
10
Drafting Skill & Role of Luck: Detroit
Detroit Drafting Value Year 1 vs. Year 2
12
10
8
Year X + 1
6
4
R² = 0.0052
2
0
-6
-4
-2
0
2
4
-2
-4
-6
Year X
6
8
10
12
Drafting Skill & Role of Luck
NHL Drafting Value Year 1 vs. Year 2
25
20
15
R² = 0.0004
Year X + 1
10
-25
5
0
-20
-15
-10
-5
0
-5
-10
-15
-20
-25
Year X
5
10
15
20
25
Drafting Skill & Role of Luck
 Question 2: How much does luck affect draft results?
 Why is this important given answer to prior question? Skill can still be shown
over a long period of time.
 Set-up:
 Use previous data and calculated values
 Calculate difference from sum of value of all picks for each team’s actual outcome
subtracted from expected outcome
 Calculate standard deviation for each team’s outcome (every team has different
picks)
 Finally, take difference and divide by standard deviation to get an index of
improved outcome over expected, herein called Draft Score
 Credit: Andrew Thomas, now with Minnesota Wild
Drafting Skill & Role of Luck
Draft Score
2
1
0
-1
-2
-3
Drafting Skill & Role of Luck
 We expect a certain amount of variance in Draft Score due to chance
 If drafting was 100% luck, the variance of all the normalized Draft
Scores would be 1, however here it is about 1.33.
 This suggests that 25% of draft results are due to skill when adjusted
for standings, 75% are due to randomness.
Drafting Skill & Role of Luck
 Takeaways:
 The effect of luck at the NHL Draft is fairly large
 Teams show talent over a long period of time, but over a short period of
time skill isn’t identifiable
 Overreacting to individual picks or drafts is likely bad analysis
Topic 3: Scouts vs. Stats
 Common debate about which form of evaluation is superior
 Hard to completely separate, because scouts are biased by statistical
performance.
 Set-up:
 Compare scout and stats rankings of first year eligible drafted CHL forwards





from 1990-2010.
Basic definitions of ‘scout’ and ‘stats’
Scouts: Slot in NHL Draft where player is picked
Stats: Most CHL points (Very simplistic approach)
See relationship between the two rankings and NHL production
Example: Brandon Dubinsky was the 12th CHL forward (60th overall) picked in
2004, but his 78 points was 2nd highest among eligible players. (He’s now #2 in
NHL scoring among eligible players, 9th overall).
Scouts vs. Stats
Table of correlation coefficients of
Scout/Stat rank vs. NHL points
Group
Scout Rank
Stat Rank
All forwards
0.54
0.49
Forwards with 300+ NHL points
0.31
0.36
Forwards with 1+ NHL point
0.40
0.48
Forwards with no NHL points
0.32
0.29
Scouts vs. Stats
Table of correlation coefficients of
Scout/Stat rank vs. NHL points
Group
Scout Rank
Stat Rank
Average of Scouts/Stats
All forwards
0.54
0.49
.58
Forwards with 300+ NHL
points
0.31
0.36
.38
Forwards with 1+ NHL
point
0.40
0.48
.50
Forwards with no NHL
points
0.32
0.29
.35
Scouts vs. Stats
 Takeaways:
 Correlation between ‘stat’ rank and actual NHL draft order/’scout’ rank
was 0.49 yet, it achieved nearly identical results.
 Scouts and stats are identifying different players while achieving nearly
identical results.
 A case for synergy of the scouting and statistical analysts. A simply average
of opinions could produce significantly better results.
Summary
 Look carefully at a player’s birth month, be extra skeptical of players
born after September 15th.
 NHL teams are not consistent in their year over year drafting.
 Luck is a very large part of NHL Draft results, but there is long-term
skill.
 Scouts and statisticians perform comparably working alone, and can
improve each other’s results with cooperation.
QUESTIONS?