Help on pitcher stuff...

[ScreaminPain]

I'm reworking a modeling program for MLB, and trying to upgrade to the use of sabermetrics as a tool. So far, things are working but I'd like some input about the use of DICE figures from those who may use them.

My question is this.
I'm using a fielding independent stat, DICE, for computation as a stat for Overall pitching, H/A pitching, and Last 3 gm pitching. These 3 stats are then averaged with a weighting method, that then gives me a DICE stat for that pitcher (Yes, I know I've omitted a Day/Night stat, but I'm fine with that.) Now, what weighting would work the most efficiently for the 3 stats I've got.

I'm currently using the following:
Season Overall: 3*
Home/Away 1.5*
Last 3 gms. 2.5*
...this divided by 7

I've never read anyones philosophy on weighting pitching stats, but I'm curious as to how others evaluate pitching when using different parameters.

[skrtelfan]

You are severely overrating the last 3 games by that method.

[ScreaminPain]

You are severely overrating the last 3 games by that method.

Maybe so....That's why I'm posing the question. Giving Last 3 gms a 2.5* value is an effort to catch a "hot" pitcher or one who is on the decline after a torrid start. What value would you suggest?

[Justin7]

Maybe so....That's why I'm posing the question. Giving Last 3 gms a 2.5* value is an effort to catch a "hot" pitcher or one who is on the decline after a torrid start. What value would you suggest?

If I had to guess, I'd use 0.

why don't you run a regression to see what weighting works best?

[ScreaminPain]

If I had to guess, I'd use 0.

why don't you run a regression to see what weighting works best?

Seems like it may be the most reliable way. ......thx

[skrtelfan]

If I were to give a guess I'd be pulling numbers out of my ass but I weight recent performance close to 0 except in terms of trying to predict line movement.

[unusialsusp5]

overanalyzing everything doesn't work. too much work results in human mind burnout. just bet and try to get lucky.

[ScreaminPain]

overanalyzing everything doesn't work. too much work results in human mind burnout. just bet and try to get lucky.

Gotta disagree with you here! There is a difference in "overanalyzing" and more accurate "computation". What if you could build a model with only a 5% added advantage? or even 3%?.....wouldn't it be worth the effort?

just bet and try to get lucky.

uh, NO!

[uva3021]

we build models to find value, then let the element of randomness take effect, and hopefully the expectation vs market model that you created creates a proposition that has an even probability but yet remains +EV because of the value placed by your model vs the market

so what I would do to weight, is find the best variable that produces the lowest RMSE, the highest correlations, and then use all the variables at its most significant correlated positions and run a regression analysis to find the precises weights

use retrosheet or the lahman database, then get the lines from statfox, and measure in accordance with wins

not only will all this work help you in the long run to create a successful base philosophy, it will distract your mind from inducing the parlay ritual due to boredom, because you will be otherwise preoccupied, therefore an indirect way of practicing solid money management