SOCIETY FOR AMERICAN BASEBALL RESEARCH The Pythagorean Won-Loss Formula in Baseball: An Introduction to Statistics and Modeling

Transcription

1 SOCIETY FOR AMERICAN BASEBALL RESEARCH The Pythagorean Won-Loss Formula in Baseball: An Introduction to Statistics and Modeling Steven J. Miller Brown University Boston, MA, May 20 th, sjmiller

2 Ackowledgements For many useful discussions: Sal Baxamusa, Phil Birnbaum, Ray Ciccolella, Steve Johnston, Michelle Manes, Russ Mann. Dedicated to my great uncle Newt Bromberg: a lifetime Sox fan who promised me that I would live to see a World Series Championship in Boston. 1

3 Goals of the Talk Derive James Pythagorean Won-Loss formula from a reasonable model. Introduce some of the techniques of modeling. Discuss the mathematics behind the models and model testing. 2

4 Goals of the Talk Derive James Pythagorean Won-Loss formula from a reasonable model. Introduce some of the techniques of modeling. Discuss the mathematics behind the models and model testing. Discuss the 2004 World Champion Boston Red Sox! 3

5 Parameters: Numerical Observation: Pythagorean Won-Loss Formula RS obs : average number of runs scored per game; RA obs : average number of runs allowed per game; γ: some parameter, constant for a sport. Bill James Won-Loss Formula (NUMERICAL Observation): RS γ Won Loss Percentage = obs RS γ obs + RA γ obs γ originally taken as 2, numerical studies show best γ is about

6 Applications of the Pythagorean Won-Loss Formula Extrapolation: use half-way through season to predict a team s performance. Evaluation: see if consistently over-perform or under-perform. There are other statistics / formulas (example: run-differential per game); advantage is very easy to apply, depends only on two simple numbers for a team. 5

7 Probability density: p(x) 0; p(x)dx = 1; Probability Review X random variable with density p(x): Prob (X [a, b]) = b a p(x)dx. 6

8 Probability density: p(x) 0; p(x)dx = 1; Probability Review X random variable with density p(x): Prob (X [a, b]) = b a p(x)dx. Mean (average value) µ = xp(x)dx. Variance (how spread out) σ 2 = (x µ) 2 p(x)dx. 7

9 Probability density: p(x) 0; p(x)dx = 1; Probability Review X random variable with density p(x): Prob (X [a, b]) = b a p(x)dx. Mean (average value) µ = xp(x)dx. Variance (how spread out) σ 2 = (x µ) 2 p(x)dx. Independence: two random variables are independent if knowledge of one does not give knowledge of the other. 8

10 Guidelines for Modeling: Modeling the Real World Model should capture key features of the system; Model should be mathematically tractable (solvable). In general these are conflicting goals. How should we try and model baseball games? 9

11 Guidelines for Modeling: Modeling the Real World Model should capture key features of the system; Model should be mathematically tractable (solvable). In general these are conflicting goals. How should we try and model baseball games? Possible Model: Runs Scored and Runs Allowed independent random variables; f RS (x), g RA (y): probability density functions for runs scored (allowed). Reduced to calculating [ ] f RS (x)g RA (y)dy dx x y x or f RS (i)g RA (j). j<i i 10

12 Reduced to calculating [ ] f RS (x)g RA (y)dy dx x y x Problems with the Model or f RS (i)g RA (j). j<i i Problems with the model: Can the integral (or sum) be completed in closed form? Are the runs scored and allowed independent random variables? What are f RS and g RA? 11

13 Weibull distribution: f(x; α, β, γ) = Three Parameter Weibull ( ) γ x β γ 1 α α e ((x β)/α) γ if x β 0 otherwise. α: scale (variance: meters versus centimeters); β: origin (mean: translation, zero point); γ: shape (behavior near β and at infinity). Various values give different shapes, but can we find α, β, γ such that it fits observed data? Is the Weibull theoretically tractable? 12

14 Weibull Plots: Parameters (α, β, γ) Red:(1, 0, 1) (exponential); Green:(1, 0, 2); Cyan:(1, 2, 2); Blue:(1, 2, 4) 13

15 Weibull Integrations Let f(x; α, β, γ) be the probability density of a Weibull(α, β, γ): ( ) γ x β γ 1 f(x; α, β, γ) = α α e ((x β)/α) γ if x β 0 otherwise. For s C with the real part of s greater than 0, recall the Γ-function: Γ(s) = 0 e u u s 1 du = Γ(n) = (n 1)! (i.e., Γ(4) = 3 2 1). 0 e u u sdu u. Let µ α,β,γ denote the mean of f(x; α, β, γ). 14

16 Weibull Integrations (Continued) µ α,β,γ = = β β x γ ( ) x β γ 1 e ((x β)/α)γ dx α α α x β α γ ( ) x β γ 1 e ((x β)/α)γ dx + β. α α Change variables: u = µ α,β,γ = ( x β α = α ) γ. ( ) Then du = γ x β γ 1 α dx and 0 0 α αu γ 1 e u du + β e u u 1+γ 1 du u + β = αγ(1 + γ 1 ) + β. A similar calculation determines the variance. 15

17 Pythagorean Won-Loss Formula Theorem: Pythagorean Won-Loss Formula: Let the runs scored and allowed per game be two independent random variables drawn from Weibull distributions (α RS, β, γ) and (α RA, β, γ); α RS and α RA are chosen so that the means are RS and RA. If γ > 0 then (RS β) γ Won-Loss Percentage(RS, RA, β, γ) = (RS β) γ + (RA β) γ. In baseball take β =.5 (from runs must be integers). RS β estimates average runs scored, RA β estimates average runs allowed. 16

18 Best Fit Weibulls to Data: Method of Least Squares Minimized the sum of squares of the error from the runs scored data plus the sum of squares of the error from the runs allowed data. Bin(k) is the k th bin; RS obs (k) (resp. RA obs (k)) the observed number of games with the number of runs scored (allowed) in Bin(k); A(α, β, γ, k) the area under the Weibull with parameters (α, β, γ) in Bin(k). Find the values of (α RS, α RA, γ) that minimize #Bins k=1 + (RS obs (k) #Games A(α RS,.5, γ, k)) 2 #Bins k=1 (RA obs (k) #Games A(α RA,.5, γ, k)) 2. 17

19 Best Fit Weibulls to Data (Method of Maximum Likelihood) Plots of RS predicted vs observed and RA predicted vs observed for the Boston Red Sox Using as bins [.5,.5] [.5, 1.5] [7.5, 8.5] [8.5, 9.5] [9.5, 11.5] [11.5, ). 18

20 Plots of RS predicted vs observed and RA predicted vs observed for the New York Yankees

21 Plots of RS predicted vs observed and RA predicted vs observed for the Baltimore Orioles

22 Plots of RS predicted vs observed and RA predicted vs observed for the Tampa Bay Devil Rays

23 Plots of RS predicted vs observed and RA predicted vs observed for the Toronto Blue Jays

24 Bonferroni Adjustments Imagine a fair coin. If flip 1,000,000 times expect 500,000 heads. Approximately 95% of the time 499, 000 #Heads 501,

25 Bonferroni Adjustments Imagine a fair coin. If flip 1,000,000 times expect 500,000 heads. Approximately 95% of the time 499, 000 #Heads 501, 000. Consider N independent experiments of fliping a fair coin 1,000,000 times. What is the probability that at least one of set doesn t have 499, 000 #Heads 501, 000? 24

26 Bonferroni Adjustments Imagine a fair coin. If flip 1,000,000 times expect 500,000 heads. Approximately 95% of the time 499, 000 #Heads 501, 000. Consider N independent experiments of fliping a fair coin 1,000,000 times. What is the probability that at least one of set doesn t have 499, 000 #Heads 501, 000? See unlikely events happen as N increases! N Prob

27 Team RS RAΧ2: 20 d.f. IndepΧ2: 109 d.f Boston Red Sox New York Yankees Baltimore Orioles Tampa Bay Devil Rays Toronto Blue Jays Minnesota Twins Chicago White Sox Cleveland Indians Detroit Tigers Kansas City Royals Los Angeles Angels Oakland Athletics Texas Rangers Seattle Mariners d.f.: (at the 95% level) and (at the 99% level). 109 d.f.: (at the 95% level) and (at the 99% level). Bonferroni Adjustment: 20 d.f.: (at the 95% level) and (at the 99% level). 109 d.f.: (at the 95% level) and (at the 99% level). 26

28 Testing the Model: Data from Method of Maximum Likelihood Team Obs Wins Pred Wins ObsPerc PredPerc GamesDiff Γ Boston Red Sox New York Yankees Baltimore Orioles Tampa Bay Devil Rays Toronto Blue Jays Minnesota Twins Chicago White Sox Cleveland Indians Detroit Tigers Kansas City Royals Los Angeles Angels Oakland Athletics Texas Rangers Seattle Mariners γ: mean = 1.74, standard deviation =.06, median = 1.76; close to numerically observed value of The mean number of the difference between observed and predicted wins was.13 with a standard deviation of 7.11 (and a median of 0.19). If we consider just the absolute value of the difference then we have a mean of 5.77 with a standard deviation of 3.85 (and a median of 6.04). 27

29 Conclusions Can find parameters such that the Weibulls are good fits to the data; The runs scored and allowed per game are statistically independent; The Pythagorean Won-Loss Formula is a consequence of our model; Method of Least Squares: best value of γ of about 1.79; Method of Maximum Likelihood: best value of γ of about 1.74; both are close to the observed best

30 Future Work Micro-analysis: runs scored and allowed are not entirely independent (big lead, close game), run production smaller for inter-league games in NL parks, et cetera. What about other sports? Does the same model work? How does γ depend on the sport? Are there other probability distributions that give integrals which can be determined in closed form? 29

31 1. Baxamusa, Sal: Weibull worksheet: Bibliography Run distribution plots for various teams: 2. Miller, Steven J.: A Derivation of James Pythagorean projection, By The Numbers The Newsletter of the SABR Statistical Analysis Committee, vol. 16 (February 2006), no. 1, A derivation of the Pythagorean Won-Loss Formula in baseball (preprint, 13 pgs). sjmiller/math/papers/pythagwonloss Paper.pdf 30

32 Appendix: Proof of the Pythagorean Won-Loss Formula Proof. Let X and Y be independent random variables with Weibull distributions (α RS, β, γ) and (α RA, β, γ) respectively. α RS = RS β Γ(1 + γ 1 ), α RA = RA β Γ(1 + γ 1 ). We need only calculate the probability that X exceeds Y. We use the integral of a probability density is 1. 31

33 Prob(X > Y ) = = = = x=β x y=β γ x=0 α RS = 1 γ x=0 α RS x=0 γ α RS ( x α RS ( x γ α RS where we have set α RS x=β x ( x β α RS y=β f(x; α RS, β, γ)f(y; α RA, β, γ)dy dx ) γ 1 e ( x β α RS ) γ ) γ 1 e ( x α RS ) γ [ x y=0 γ α RA γ α RA ( ) ( ) y β γ 1 e y β γ α RA dy α RA ( ) ( ) y γ 1 e y γ ] α RA dy α RA ) γ 1 e (x/α RS) γ [ 1 e (x/α RA) γ] dx ( x α RS 1 α γ = 1 α γ RS ) γ 1 e (x/α)γ dx, + 1 α γ RA = αγ RS + αγ RA α γ. RS αγ RA 32

34 Prob(X > Y ) = 1 αγ α γ RS = 1 αγ α γ RS = 1 1 α γ RS α γ RS 0 γ ( x ) γ 1 e (x/α) γ dx α α α γ RS αγ RA α γ RS + αγ RA = α γ RS +. αγ RA We substitute the relations for α RS and α RA and find that (RS β) γ Prob(X > Y ) = (RS β) γ + (RA β) γ. Note RS β estimates RS obs, RA β estimates RA obs. 33