Quote:
Originally Posted by Snowman
As far as how it could be calculated, there are several options. My preferred approach would be to build a hierarchical mixed-effects model (which controls for both fixed-effects and random-effects simultaneously). These models are extremely powerful. You could create time blocks for various periods where something of note happened (like 1942-1946 when the talent pool was heavily diluted due to players leaving for WW2), or pre-1950 for larger strike zones, or 1950-62 for larger strike zones, etc. You would hard code those into your data and treat them as fixed-effects. We could also control for offensive efficiencies of each era by measuring the delta between runners left in scoring position, among countless other ways (offenses were considerably less efficient when Grove was pitching)..........
|
All of this is just listing many variables involved, and I'm sure many more can be added. The obvious difficulties that remain are:
1. How do these variables play together? Are they additive, multiplicative, subtractive, and to what degree. How do you combine and weigh them?
2. How do you value them, with respect to specific players?
For example, let's say you are comparing 2 pitchers who both have a right fielder with a .985 fielding average. But one has a weak throwing arm and the other is Clemente. How much does having Clemente help, with his reputation discouraging runners taking an extra base?
First you'd need to give a weight to the variable - what impact does the right fielder's reputation play? Second, you have to value Clemente.
Suppose there are two catchers with equal fielding percentages, and throw out equal percentages of baserunners. But one is a very astute signal caller and the other is a dolt. Take Grove having Cochrane for example. First, how much can a smart, observant catcher help a pitcher? Second, what value do you assign to Cochrane (or Roseboro?)
All you have done is thrown out a bunch of factors to consider. The real trick would be to come up with an algorithm that can effectively combine and weigh the variables, and then, there's the (sometimes subjective - like the brains of a catcher) value you assign to each specific player involved.
In short, the above is not anywhere close to an actual predictive model.