THE BOOK--Playing The Percentages In Baseball

We still don’t have perfect reconciliation with WARP - I don’t think positional values sum up to zero. (I don’t have my notes at work, so I’ll go over that and the other thread when I get home.) BRAA at the league level sums to about 880 runs, which is odd. I think it’s a .263 rep-level for 2008, probably a bit low. I may look at other years here - BP has upgraded the cards but not the sortable stats for WARP, which makes it harder to scrape the data.

What’s the issue with positional adjustments?

Previous thread on the topic is here:
http://www.insidethebook.com/ee/index.php/site/comments/the_new_warp/

Wow, ‘50s CF outhit the corner OF?  That would be one of the first things I would mention if someone at a bar told me that the league today isn’t as good as it used to be.

1) Richie Ashburn+
2) Duke Snider*
3) Larry Doby+
4) Jim Busby
5) Mickey Mantle*
6) Willie Mays*
7) Bill Bruton
8) Johnny Groth
9) Gus Bell
10) Bill Tuttle

* denotes a player voted into the Hall of Fame by the BBWAA, + a player voted into the Hall of Fame by the Veterans Committee.

That’s five Hall of Fame center fielders in one decade. Sure, it’s rare, but it happens. And that’s why we need to account for it in our evaluation of players.

You know, on the rare occasion that I do need to convert hits and walks and whatnot into runs, I do tend to use linear weights-style formulas. But ... deep down I still respect runs created for being able to derive an excellent estimate of runs scored without having to depend on league averages.

I have said this before--one of the biggest hurdles to acceptance of non-RC run estimators is that their creators (and those of us who are not creators ourselves but research run estimators and write about them) have emphasized the flexibility and ability to customize them.  But people seem to want one formula that stays concrete forever and ever, even if that makes for a poor model.

But the truth of the matter is that linear weights don’t need league averages to generate an “excellent estimate of runs scored”.  A simple ERP formula like (TB + .8H + W - .3AB)*.324 is just as accurate as Runs Created and does not use any league averages.  And of course there are many others.  Base Runs doesn’t need league average either. 

I don’t see how Runs Created can earn brownie points for this.

This question has probably been asked a million times.

I took a team.  Did my own projections.  Then used Base Runs on each player and then on a team level to determine runs scored.  When I did it for the team, it didn’t add up.  Is this because when you are doing it per player its calculated as if there are 8 average players in the lineup?

I guess thinking back on it I shouldnt have done Base Runs on a player level.  LWTS perhaps.

BaseRuns is not appropriate for individual hitters.  It is for teams and for pitchers.

You would need to use a team approach, and doing a with or without you.

Thanks.

And I guess if I wanted to look at individual hitters, linear weights are most appropriate?

Patriot, I agree with you—flexibility is a good thing, and I do prefer LWTS over runs created. But ... I’m speaking from the experience of the mid-1980s, when it was much easier to type the values for (H+BB)*TB/(AB+BB) into a handheld calculator. Linear weights always required an extra step—determining the out value. Not a big deal in this day and age, but I remember trying to calculate George Bell’s linear weights in the summer of 1987 (using the 1986 out value) and not knowing how much the higher league scoring level had skewed the results, and so it was difficult to compare the contributions of Bell vs Jimmy Key. Runs created was more straightforward and easier to calculate. As you probably remember, it wasn’t easy to find league averages during the season back then. Even now, league averages can be difficult to come by in a non-major league context (ie when you’re at a minor league game).

Of course, all of the above is anecdotal and irrelevant.

Just to yank your chain a little --

A simple ERP formula like (TB + .8H + W - .3AB)*.324 is just as accurate as Runs Created

Except runs created (in its basic form) doesn’t have any suspicious-looking “.324” floating around.

linear weights don’t need league averages

Tango’s wiki says “BR = .47S + .85D + 1.02T + 1.40HR + .33(W + HB) - ABF*(AB - H)
where ABF is the coefficient calculated so that the Batting Runs for a league is equal to zero”

Base Runs doesn’t need league average either.

The BsR entry on Wikipedia says “If you want to tailor a version to a particular dataset (such as 1993-2004, or the 1975 AL), all you have to do is determine the overall B multiplier.” The B multiplier, sometimes referred to as the variable X, is found by setting X = ((LgRuns - D) * C) / B / (A - (LgRuns - D))”

~~~~~~~~~~~~~~~

Come to think of it, even in 2009, why do you want to jump through all of those hoops just to knock a run or two off your RMSE?

All of your quotes illustrate exactly what I was talking about--the creators and advocates of those methods going out of their way to show you how to customize the method, rather than pretending as if the formula was somehow immutable.

The “unseemly” multiplier tacked on the end of ERP, for instance, is just a way to be able to write the formula more simply.  If you think of it as .5S + .8D + 1.05T + 1.4HR + .33W - .1(outs), or whatever, suddenly the unseemly multiplier is gone, and the coefficients are the average run expectancy deltas for each event.

Anyway, it’s true that the math just so happens to work out that basic RC needs no multiplier.  But that doesn’t make it any more fundamental of an estimator than any other.  It just so happens to require a scalar of 1.

Come to think of it, even in 2009, why do you want to jump through all of those hoops just to knock a run or two off your RMSE?

If you just want to estimate how many runs a normal team will score quickly, there’s nothing wrong with Runs Created.  But if you want to apply it to individual players or extreme teams, it breaks down.

For normal run of the mill situations, anything with any modicum of sense behind it will work.  OPS works.  TA works.  RC works.