Kerry Whisnant Creating a better formula than Pythagorean?

[mebaran]

I read Kerry Whisnant's paper on predicting baseball scores with the formula he devised. He claims that when compared to the Pythagorean Formula (runs scored^2/runs scored^2 + runs allowed^2), his formula cuts the error in half. By adding SLG. for and SLG. allowed into the equation, one can compensate for the distribution of runs, as well as the actual runs themselves.

His formula is:
W1/L1 = (RPG1/RPG2)^a(SLG1/SLG2)^b

where a = 0.723(RPG1+RPG2)^.373, and b = 0.977(RPG1+RPG2)^(-.947)

My question is directed at those who model baseball: Does his formula actually improve upon Bill James' decades old formula? What are it's limitations (ie, is it well suited to project single games or full seasons)?

In a few of my toy models, his formula in place of the pythag. formula almost always makes the favorite a larger favorite.

Thoughts? Experiences?

[Cookie Monster]

I am not into modelling, but I was an early reader of sabrmetrics.

The way I understand it, there are four depth levels. The deepest ones have more predictive value, the previous level is affected heavily by the random clumping of the deeper one.

1.- Wins
2.- Runs
3.- Batting results stats (OPS, etc).
4.- Batting components (lines, grounds, etc).

The Pythagorean formula was devised a long time ago as an improvement over raw W-L records. But there have been many advances since then. In the same way as runs are a better predictor of future wins than wins, advanced batting stats are better predictor of runs than actual runs. And batting components are better predictor of batting stats than actual batting stats.

So, of course Pythagorean can be improved. Baseball prospectus has been running "third level" expected wins for a few years. The formula you posted may be an small improvement over Pythagorean, but you can easily further improve it if you are so inclined.

[mebaran]

I am not into modelling, but I was an early reader of sabrmetrics.

The way I understand it, there are four depth levels. The deepest ones have more predictive value, the previous level is affected heavily by the random clumping of the deeper one.

1.- Wins
2.- Runs
3.- Batting results stats (OPS, etc).
4.- Batting components (lines, grounds, etc).

The Pythagorean formula was devised a long time ago as an improvement over raw W-L records. But there have been many advances since then. In the same way as runs are a better predictor of future wins than wins, advanced batting stats are better predictor of runs than actual runs. And batting components are better predictor of batting stats than actual batting stats.

So, of course Pythagorean can be improved. Baseball prospectus has been running "third level" expected wins for a few years. The formula you posted may be an small improvement over Pythagorean, but you can easily further improve it if you are so inclined.

Appreciate the input, CM. Where did you learn about this
, and what is random clumping?

[Cookie Monster]

I started getting interest in sabrmetrics, as many guys, reading Rob Neyer column at ESPN. Then I found "baseball primer" (now named baseball think factory) and spent many hours reading articles there everyday. So, I was not taught of the levels of baseball stats, but following the developments (as DIPS) anyone would infer the levels. Now there are many good reads, "The book" by Tango is a must, and you can find great info in the website. Another good sites: Hardball times, BTFactory, Baseball primer, etc.

Sorry about my poor English. By clumping I mean uneven distribution. For example, a team hitting 2 singles on 1st inning and 2 more in 2nd inning is expected to score 1 or 0 runs. The opponent hits 4 singles on 1st inning and 0 in second. It should score 2 or 3 runs. Both made the same number of hits and outs, but the 2nd team got lucky, as the hits were unevenly distributed.

[mebaran]

I started getting interest in sabrmetrics, as many guys, reading Rob Neyer column at ESPN. Then I found "baseball primer" (now named baseball think factory) and spent many hours reading articles there everyday. So, I was not taught of the levels of baseball stats, but following the developments (as DIPS) anyone would infer the levels. Now there are many good reads, "The book" by Tango is a must, and you can find great info in the website. Another good sites: Hardball times, BTFactory, Baseball primer, etc.

Sorry about my poor English. By clumping I mean uneven distribution. For example, a team hitting 2 singles on 1st inning and 2 more in 2nd inning is expected to score 1 or 0 runs. The opponent hits 4 singles on 1st inning and 0 in second. It should score 2 or 3 runs. Both made the same number of hits and outs, but the 2nd team got lucky, as the hits were unevenly distributed.

Ah ok, that's what I thought you meant. Thanks for the info.

[antifoil]

is there a way to predict clumping or is it compete random luck?

[Professor1215]

Apparently no one can answer your question.

[Pokerjoe]

Apparently no one can answer your question.

Why do you conclude that?

[mebaran]

Why do you conclude that?

He's a 99%er who thinks he just gets things without trying first?