Defensive Metrics thread continued

[GotowarMissAgnes]

This basically sums up what I've tried to express in this thread pretty well, and probably more succinctly than I have.

Again, the problem with this is that it starts with the assumption hitting and fielding come from the same distribution of outcomes. They don't. Hitting is a draw from a low mean, high variance distribution. Fielding is a draw from a high mean, low variance distribution. A random draw from the first is likely to be characterized by lots of variability. A random draw from the second is not.

Again, they're quite comparable when measured in runs.

Again, they're not. Runs are simply a linear transformation from the underlying rate data.

[Big Pimpin]

Again, the problem with this is that it starts with the assumption hitting and fielding come from the same distribution of outcomes. They don't. Hitting is a draw from a low mean, high variance distribution. Fielding is a draw from a high mean, low variance distribution. A random draw from the first is likely to be characterized by lots of variability. A random draw from the second is not.

Again, they're quite comparable when measured in runs.

Again, they're not. Runs are simply a linear transformation from the underlying rate data.

Ok, a linear transformation from the underlying plays number, which is derived from an entire matrix of probabilities, some of which are high and some of which are low.

I love this stuff.