SlideShare uma empresa Scribd logo
1 de 25
DIGR: A Defense Independent Rating of NHL Goaltenders using Spatially Smoothed Save Percentage Maps Michael E. Schuckers* St. Lawrence UniversityStatistical Sports Consultingschuckers@stlawu.edu   *Thanks to Chris Wells, Ken Krzywicki, Dan Downs, Dennis Lock, Matt Generous
2009-10 Save Percentage Goalie			Gi			Team Pts Brodeur (NJD)		0.916		103* Luongo (VAN)		0.913		103* Turco (DAL)		0.913		88 Ward (CAR)		0.916		80 * Made Stanley Cup playoffs
Gi= Problem: Each goalie faces different distribution of shots  Goal of this paper Find statistical methodology to allow comparison Save Percentage
Rethinking Save Percentage s=shot type Pi(s) Ri(s) Xi(s) = Number of saves by goalie ion shots of type s Ti(s) = Total number of shots faced by goalie ion shots of type s Pi(s) = performance (save percentage) of goalie ion shot type s Ri(s) = percent/rate of all shots for goalie ithat were of type s
Rethinking Save Percentage Save Percentage Convert to`R(s) the league average distribution of shots faced
Data Downloaded from ESPN.com GameCast Every NHL regular season game 09-10 	Goalie  	(x,y) location of ( n= )74300 shots 	Opponents strength 	Shot Type 	Location* 	Home/Away Team *Madison Square Garden is a statistical nightmare in hockey
Shots s=(x-coord, y-coord, shot type, strength) All shots converted to single offensive zone Shot types  Backhand, Deflection, Slap, Snap, Tip-in, Wrap and Wrist Strength Even, Power Play, Shorthanded
Spatial Smoothing Use LOWESS* (locally weighted scatterplotsmoothing)  Nonparametric (no specific model) One map for each strength x shot type (21) Use all shots for given shot type (total weight 30) *Using loess in R
Why smooth? Luongo vs. Distance
Ryan Miller/ Slap Shots/ Even Strength
Ryan Miller/Slap Shot Even Strength Power Play Shorthanded
Tomas Vokoun/Slap Shot Even Strength Power Play Shorthanded
NiklasBackstrom/Slap Shot Power Play Even Strength Shorthanded
Rethinking Save Percentage Save Percentage Shot Quality Adjusted  Save Percentage  (E. g. Krzywicki (2010)) Defense Independent Goalie Rating (DIGR)
Application 49 goalies >600 shots faced in 2009-10 Regular Season Each shot (n=74300), each goalie 	predicted goal probability using smoothed maps Calculated DIGR
Results: Top 10 0.01 = 20 goals for a goalie facing 2000 shots
Results: Other Notables 0.01 = 20 goals for a goalie facing 2000 shots
Results Big* Winners(DIGR - Save Pct >> 0) 	Smith(TBL), Roloson (NYI), Huet (CHI),  Pavelec (ATL), Varlamov (WSH), Biron (NYI), Theodore (WSH), Leclaire (OTT),  Toskala (TOR, CGY)   Big* Losers (DIGR - Save Pct << 0) 	Rask (BOS), Howard (DET), Thomas (BOS) Big means > 0.0075 OR  15 goals on 2000 shots
Results (2000 shots using`R(s)) Rank	PlayerDIGR Goals 1		Miller(BUF) 		143 … 11Hedberg(ATL)		162 …. 21	Anderson (COL)	173 … 31	Ellis (NSH)		177 … 41	Huet(CHI)		191 … 49 	Toskala(TOR, CGY)	206 DPts=0.35*GoalDiff 6.7 3.9 1.4 4.9 5.0 19 11 4 14 15
Discussion Average season performance Standard Errors (Bootstrap) Shot target (holes 1 to 5) Injuries (e.g. Tim Thomas) Extension          G*ij=SsPi(s) Rj(s)
Turco takes Niemi’s shots June 2010 	Blackhawks win Stanley Cup 			Need Cap Space 			Fail to resign Niemi and sign Turco 			Saving $1.45 million			 GiGi*(DIGR) Niemi(CHI)		0.915		0.922 Turco (DAL) 		0.912		0.910 G*ij=SsPi(s)Rj(s) (i=Turco, j = Niemi) = 0.903
Turco takes Niemi’s shots Turco G*ij = 0.903 vsNiemi G*jj = 0.915 What’s the cost? Turco on pace to face about 1000 shots in 2010-11 1000 shots *(0.012) = 12 goals  12 goal *0.35 = 4.2 pts Turco Save Pct (2010-11) = 0.897
DIGR vs. ‘09-’10 Salary
Summary DIGR: Defense Independent Goalie Rating Three innovations - Spatial smoothing maps - Goalie ratings on comparable shot distribution - Mathematical framework
Thank You!schuckers@stlawu.edu  

Mais conteúdo relacionado

Mais de Sloan Sports Conference

Flipping Coins in the War Room: Skill and Chance in the NFL Draft
Flipping Coins in the War Room: Skill and Chance in the NFL DraftFlipping Coins in the War Room: Skill and Chance in the NFL Draft
Flipping Coins in the War Room: Skill and Chance in the NFL DraftSloan Sports Conference
 
Dodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur Draft
Dodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur DraftDodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur Draft
Dodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur DraftSloan Sports Conference
 
Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...
Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...
Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...Sloan Sports Conference
 
Allocation and Dynamic Efficiency in NBA Decision Making
Allocation and Dynamic Efficiency in NBA Decision MakingAllocation and Dynamic Efficiency in NBA Decision Making
Allocation and Dynamic Efficiency in NBA Decision MakingSloan Sports Conference
 
Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...
Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...
Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...Sloan Sports Conference
 
Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...
Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...
Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...Sloan Sports Conference
 
The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...
The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...
The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...Sloan Sports Conference
 
The Real Reasons Behind the Home Field Advantage
The Real Reasons Behind the Home Field AdvantageThe Real Reasons Behind the Home Field Advantage
The Real Reasons Behind the Home Field AdvantageSloan Sports Conference
 

Mais de Sloan Sports Conference (9)

Flipping Coins in the War Room: Skill and Chance in the NFL Draft
Flipping Coins in the War Room: Skill and Chance in the NFL DraftFlipping Coins in the War Room: Skill and Chance in the NFL Draft
Flipping Coins in the War Room: Skill and Chance in the NFL Draft
 
Dodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur Draft
Dodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur DraftDodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur Draft
Dodging the Draft: Analyzing the Competitive Impact of Baseball’s Amateur Draft
 
Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...
Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...
Buying Tickets: Capturing the Dynamic Factors that Drive Consumer Purchase De...
 
How Much Trouble is Early Foul Trouble?
How Much Trouble is Early Foul Trouble?How Much Trouble is Early Foul Trouble?
How Much Trouble is Early Foul Trouble?
 
Allocation and Dynamic Efficiency in NBA Decision Making
Allocation and Dynamic Efficiency in NBA Decision MakingAllocation and Dynamic Efficiency in NBA Decision Making
Allocation and Dynamic Efficiency in NBA Decision Making
 
Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...
Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...
Moral Hazard in Long-Term Guaranteed Contracts – Theory and Evidence from the...
 
Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...
Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...
Scoring Strategies for the Underdog – Using Risk as an Ally in Determining Op...
 
The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...
The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...
The Importance of Being Open: What Player Tracking Data Can Say About NBA Fie...
 
The Real Reasons Behind the Home Field Advantage
The Real Reasons Behind the Home Field AdvantageThe Real Reasons Behind the Home Field Advantage
The Real Reasons Behind the Home Field Advantage
 

DIGR: A Defense Independent Rating of NHL Goaltenders using Spatially Smoothed Save Percentage Maps

  • 1. DIGR: A Defense Independent Rating of NHL Goaltenders using Spatially Smoothed Save Percentage Maps Michael E. Schuckers* St. Lawrence UniversityStatistical Sports Consultingschuckers@stlawu.edu   *Thanks to Chris Wells, Ken Krzywicki, Dan Downs, Dennis Lock, Matt Generous
  • 2. 2009-10 Save Percentage Goalie Gi Team Pts Brodeur (NJD) 0.916 103* Luongo (VAN) 0.913 103* Turco (DAL) 0.913 88 Ward (CAR) 0.916 80 * Made Stanley Cup playoffs
  • 3. Gi= Problem: Each goalie faces different distribution of shots Goal of this paper Find statistical methodology to allow comparison Save Percentage
  • 4. Rethinking Save Percentage s=shot type Pi(s) Ri(s) Xi(s) = Number of saves by goalie ion shots of type s Ti(s) = Total number of shots faced by goalie ion shots of type s Pi(s) = performance (save percentage) of goalie ion shot type s Ri(s) = percent/rate of all shots for goalie ithat were of type s
  • 5. Rethinking Save Percentage Save Percentage Convert to`R(s) the league average distribution of shots faced
  • 6. Data Downloaded from ESPN.com GameCast Every NHL regular season game 09-10 Goalie (x,y) location of ( n= )74300 shots Opponents strength Shot Type Location* Home/Away Team *Madison Square Garden is a statistical nightmare in hockey
  • 7. Shots s=(x-coord, y-coord, shot type, strength) All shots converted to single offensive zone Shot types Backhand, Deflection, Slap, Snap, Tip-in, Wrap and Wrist Strength Even, Power Play, Shorthanded
  • 8. Spatial Smoothing Use LOWESS* (locally weighted scatterplotsmoothing) Nonparametric (no specific model) One map for each strength x shot type (21) Use all shots for given shot type (total weight 30) *Using loess in R
  • 9. Why smooth? Luongo vs. Distance
  • 10. Ryan Miller/ Slap Shots/ Even Strength
  • 11. Ryan Miller/Slap Shot Even Strength Power Play Shorthanded
  • 12. Tomas Vokoun/Slap Shot Even Strength Power Play Shorthanded
  • 13. NiklasBackstrom/Slap Shot Power Play Even Strength Shorthanded
  • 14. Rethinking Save Percentage Save Percentage Shot Quality Adjusted Save Percentage (E. g. Krzywicki (2010)) Defense Independent Goalie Rating (DIGR)
  • 15. Application 49 goalies >600 shots faced in 2009-10 Regular Season Each shot (n=74300), each goalie predicted goal probability using smoothed maps Calculated DIGR
  • 16. Results: Top 10 0.01 = 20 goals for a goalie facing 2000 shots
  • 17. Results: Other Notables 0.01 = 20 goals for a goalie facing 2000 shots
  • 18. Results Big* Winners(DIGR - Save Pct >> 0) Smith(TBL), Roloson (NYI), Huet (CHI), Pavelec (ATL), Varlamov (WSH), Biron (NYI), Theodore (WSH), Leclaire (OTT), Toskala (TOR, CGY) Big* Losers (DIGR - Save Pct << 0) Rask (BOS), Howard (DET), Thomas (BOS) Big means > 0.0075 OR 15 goals on 2000 shots
  • 19. Results (2000 shots using`R(s)) Rank PlayerDIGR Goals 1 Miller(BUF) 143 … 11Hedberg(ATL) 162 …. 21 Anderson (COL) 173 … 31 Ellis (NSH) 177 … 41 Huet(CHI) 191 … 49 Toskala(TOR, CGY) 206 DPts=0.35*GoalDiff 6.7 3.9 1.4 4.9 5.0 19 11 4 14 15
  • 20. Discussion Average season performance Standard Errors (Bootstrap) Shot target (holes 1 to 5) Injuries (e.g. Tim Thomas) Extension G*ij=SsPi(s) Rj(s)
  • 21. Turco takes Niemi’s shots June 2010 Blackhawks win Stanley Cup Need Cap Space Fail to resign Niemi and sign Turco Saving $1.45 million GiGi*(DIGR) Niemi(CHI) 0.915 0.922 Turco (DAL) 0.912 0.910 G*ij=SsPi(s)Rj(s) (i=Turco, j = Niemi) = 0.903
  • 22. Turco takes Niemi’s shots Turco G*ij = 0.903 vsNiemi G*jj = 0.915 What’s the cost? Turco on pace to face about 1000 shots in 2010-11 1000 shots *(0.012) = 12 goals 12 goal *0.35 = 4.2 pts Turco Save Pct (2010-11) = 0.897
  • 24. Summary DIGR: Defense Independent Goalie Rating Three innovations - Spatial smoothing maps - Goalie ratings on comparable shot distribution - Mathematical framework