W HAT DOES IT MEAN TO FIND THE F ACE OF THE F RANCHISE ? P HYSICAL A TTRACTIVENESS AND THE E VALUATION OF A THLETIC P ERFORMANCE DAVE BERRI, ROB SIMMONS, JENNIFER VAN GILDER & LISLE O ’ NEILL WEAI Portland June Economics of the NFL
U NIVERSAL B EAUTY (F IRST D OWN ) “Beauty is in the eye of the beholder” Beauty affects our judgment from cradle to grave Sociological studies indicate proportion as a commonality Samuels (1994) says infants pay greater attention to symmetrical objects Honekopp (2006) finds human ratings of attractiveness confirm symmetry ratings
S YMMETRY : Q UANTITATIVE B EAUTY Measuring beauty in a quantitative manner Technological link between symmetry and human perception of attractiveness Gunes and Piccardi (2006) find high correlation between human ratings and digital ratings
B EAUTY IN THE L ABOR M ARKET Hamermesh and Biddle’s findings 1. Premium for beauty and penalty for ugliness 2. 3 reasons for premium or penalty Olson and Marshuetz (2005) suggest beauty has a hiring impact Our paper differs through use of symmetry analysis
D ATA : W HY Q UARTERBACKS ? (S ECOND D OWN ) Data Richness Acquired from NFL.com Subjects: 312 Quarterbacks from QBs seen as ‘the face of the franchise’, have a leadership role on team, role models for fans & young players, attract media publicity Contributing factors of Productivity measurement included in the “passer” rating Creation of 2 data sets: primary and secondary quarterbacks- which can be merged into one set
M ETHOD AND T HEORY (T HIRD D OWN ) Images provided by NFL homepage and Yahoo sports Theory: why would a GM hire a better-looking quarterback? Marginal revenue product Utility maximization Null Hypothesis, given that B 2 is defined as the coefficient on the beauty variable: H 0 : B 2 = 0 [no impact of beauty on pay] H A : B 2 > 0 [beauty has a positive effect on pay, given performance & experience]
S YMMETRY A NALYSIS Software: symmeter.com Three Examples of Analysis and Results Symmetry Value: % Symmetry Value: % Symmetry Value: %
D ESCRIPTIVE S TATISTICS VariableMeanStd DevMinimumMaximum Symmetry Cap Value Plays Attempts Pro Bowler Primary Quarterbacks VariableMeanStd DevMinimumMaximum Symmetry Cap Value Plays Attempts Pro Bowler Secondary Quarterbacks
F INAL M ODEL R ESULTS (F OURTH D OWN ) Model: lnSAL = b 0 + b 1 *PYARDS + b 2 *CPASSATT + b 3 *EXP + b 4 *EXPSQ + b 5 * DRAFT1 + b 6 *DRAFT2 + b 7 *VET + b 8 *NEWTM + b 9 *lnOFFSAL + b 10 *PB + b 11 *SYMMETRY + e t (1)
E STIMATION Dependent Variable: Log of Salary Years: 1995 to 2006 n = 480, all QBs Robust standard errors reported. Qualifying condition is at least 1 play in previous season; rookies excluded OLS then Huber Robust Regression
OLS RESULTS VariableCoefficient Standard Errort-stat PYARDS* CPASSATT* EXP* EXPSQ* DRAFT1* DRAFT2* VET* NEWTM* lnOFFSAL** PB* SYM** R-squared0.64
VariableCoefficient Standard Errort-stat PYARDS* CPASSATT* EXP* EXPSQ* DRAFT1* DRAFT2* VET* NEWTM* lnOFFSAL** PB* SYM**
N OTEWORTHY I MPLICATIONS Variables Primary Parameter Estimates Secondary Parameter Estimates Symmetry * Black * * Black*Symmetry Draft Draft2 * Pro Bowler Experience Experience QB Rating Change Team Year Attempts *Variable not statistically significant.
F UTURE R ESEARCH AND THANK YOU ( TOUCHDOWN ) Caveats Consider using one stat per QB (average, lifetime max?) Recent literature indicates CPI over-deflates: different deflators may give different results; earlier regressions had year summies Quantile Regression was used in JSE QB Race study QB & receiver performances interact-QBs and receivers are each credited in stats for yards gained- who was really responsible?