The Sabermetric Revolution: Assessing the Growth of Analytics in Baseball by Benjamin Baumer, Andrew Zimbalist Page B

by Sabermetrics
    We saw previously that Oakland had the largest positive team effect in our payroll model. In this context we want to understand the deviations from the general relationship between winning percentage and payroll, so we compute the model without team fixed effects, and later investigate the relationship of those residuals for patterns with respect to our composite saber-intensity (SI) metric. The equation tested below is WPCT = f(PAY, PAY 2 ).
    Oakland, St. Louis, Atlanta, the Chicago White Sox, Minnesota, and Florida have the largest average residuals.
    Table 24. The Relationship Between Win Percentage and Payroll, Without Team Fixed Effects

Building a Composite Sabermetric Index
    Our goal here is to build a single index that will measure sabermetric intensity. To do this, we need weights for each of the elements of saber-intensity. A reasonable choice for those weights is the normalized coefficients from a double-log model for WPCT as a function of the sabermetric variables that we are tracking. That model is presented below. Since the baserunning statistic can be negative, we transform it so that it is always positive. The natural log of WPCT is the dependent variable.
    The value of these coefficients gives us a sense of the relative impact that each has on WPCT. Note that while a team’s fielding metric (DER) appears to have the largest impact upon WPCT, each of the six terms has a statistically significant effect at the 10 percent level or better.
    In order to use these coefficients as weights, we drop the coefficient for the intercept term, take their absolute values, and then normalize them.
    Table 25. Double Log Model to Weight Components of SI Index

    Table 26. Final Component Weights in SI Index
log(OBP)
0.2367
log(ISO)
0.0374
log(FIP)
0.2903
log(DER)
0.4328
log(brun)
0.0008
log(sacbunt)
0.0020
    This suggests that roughly 43 percent of our sabermetric index will consist of a team’s fielding metric (DER). These weights are then applied to our six saber-intensity elements to generate our composite index of saber-intensity (SI). The results for the top thirty teams in SI between 1985 and 2011 are presented in Table 17 in Chapter 7 .
    When we regress the residuals from the WPCT and payroll regression on our saber-intensity index, the results are presented in Table 27 .
    Note that since the average team has an SI of 1, the predicted WPCT residual (the expected impact of all factors other than payroll) is zero for the average team. 6 It is worth reiterating that there is no reason to believe, independent of sabermetrics, that a team with a high sabermetric index would be more successful than their payroll would indicate. Yet the regression model above shows that our sabermetric index explains nearly 37 percent of the variation in team winning percentage that is not explained by payroll.
    Table 27. Relationship of Win Percentage and Team Saber-Intensity (SI)

MODELING THE SHIFTING INEFFICIENCIES
IN MLB LABOR MARKETS

    In this section we describe how we modeled our test for the morphing inefficiencies in baseball’s labor market. Our approach is an extension of the one employed by Hakes and Sauer in 2006. 7
    There are two main components to this procedure. First, we construct a model for team performance in terms of simple performance metrics. In this manner, we gain an understanding of what skills translate into team success. Second, we construct a model for how those skills are compensated on the labor market. Market inefficiencies are reflected in the differences in the estimates between these two models.
    Hakes and Sauer identified three largely orthogonal qualities that reflect on-field performance for both batters and pitchers: Eye (walks plus hit by pitch per plate appearance), Bat (batting average), and Power (slugging percentage divided by batting average). Since there is a natural equality to the way in which a team’s offense and defense contribute to their success, the Hakes and Sauer model