Coefficients for Models

[JVP3122]

So, I'm working on my first model, and I have a quick question. If I'm working with a runs scored model, and I'm adding up stats like singles, doubles, triples, etc, I just don't add them all together at equal weight. I'm wondering if there's a statistical test which would help me determine the weight that each term should hold or if I should just try different coefficients until the model makes sense. Any help's appreciated.

[a4u2fear]

Try finding the correlation between # of triples in a game and wins, then do the same with doubles etc.

[InTheRed]

My first instinct would be to weigh them as they are. singles =.25, doubles =.5, etc. That may be a start.

[evo34]

Rather than re-invent the wheel, you'd prob. want to look into existing formulas that translate baseball events (singles, doubles, etc.) into expected runs. E.g., Fangraphs' wOBA or Baseball Prospectus' TAv.

So, I'm working on my first model, and I have a quick question. If I'm working with a runs scored model, and I'm adding up stats like singles, doubles, triples, etc, I just don't add them all together at equal weight. I'm wondering if there's a statistical test which would help me determine the weight that each term should hold or if I should just try different coefficients until the model makes sense. Any help's appreciated.

[uva3021]

Run Expectancy

http://www.tangotiger.net/re24.html

download the retrosheet database and learn SQL language

[mebaran]

Contrary to intuition, doubles are not worth twice as much as singles, and triples aren't worth 3 times as much as singles. There is a drop off in value from doubles to triples (that extra base isn't worth as much, and most doubles will clear the bases anyways).

As far as regression analysis goes, I remember that OPS is a better indicator of runs scored than purely SLG. So take your sample (probably 3-6 years worth of hitting data) and do some basic linear regression with each hit and runs scored, etc.