THE BOOK--Playing The Percentages In Baseball

Wasn’t there a thread here on an updated study by Tango on RP a few years ago? My recollection is that at the end of the discussion, Rally (IIRC) posted a study which suggested that maybe HR shouldn’t be subtracted out, in support of B James. And Tango made a comment about confusing inputs and outputs. And then the thread died. I always felt that more discussion was warranted, but it never came…

The reason that I don’t like correlation is that you have so many players whose HR totals are not at the extreme that you learn nothing from it.

By pooling players as I have done, you are giving equal weight to groups of players, regardless as to the number of players in each group.

Otherwise, you can have a situation where say RC or TA correlates better with runs than BaseRuns.  What does that tell us?  It tells us that there’s not alot of extreme data points that would let BsR shine through.

I prefer a logical model rather than one that best-fits the most points, but that people will extrapolate to the extreme points anyway.

I sometimes tire of run estimators since we have no way of knowing how many runs a player actually contributed to HIS team (validation), and assuming a walk or HR has the same value regardless of team and batting order position is a poor assumption. In any event, it does not translate to actual runs produced in a game (the ones that change the scoreboard) . 

A BB that does not contribute to an ACTUAL run scored in the inning it occurs has a value close to 0 (unless someone can quantify the value of adding to the pitchers pitch count).

However, a 2 out BB with a man on second where the runner on 2nd scores after the BB on a base hit that follows has contributed to the run scored.

I like the idea of of giving assists like in hockey or basketball to be added to RP.  Total points is RP + A.  A player contribution to the teams run scored is simply total points/runs scored.

I also think a HR which then is followed by runs scored after 2 outs deserves an assist, or at least should not be fully deducted from RP.  When this was last discussed I said that while HR’s should be subtracted from RP, it should only be partially subtracted.

@3, It sounds like you really just want to talk about some leverage sensitive stat like WPA, and not run estimators at all.

Let me start with a small comment.  Tangotiger wrote:
“What is without question is that the HR needs to be subtracted.”

Can we at least agree that that is wrong?  I don’t mind an honest debate, but to claim that one’s position is correct “without question” overreaches by quite a bit.

It seems to me that the starting point for an argument about whether a statistic is useful is to ask “what are we trying to measure?” Since I have never used Runs Produced, that’s a bit hard for me to answer.  If the answer is “Runs Produced is intended to be a simple method of measuring overall offensive productivity” then Tangotiger’s argument makes a lot of sense.  (And I enjoyed reading it.) If the answer is “Runs Produced is intended to measure a player’s contributions to plays that score runs,” then subtracting out home runs doesn’t make sense.  I had thought Runs Produced was trying to do the latter, but I could be wrong.  If it is intended to measure overall offensive productivity, I have no idea why it’s advocates would start with numbers that are so dependent on line-up position and team offense.

Tangotiger urges us to not argue against RBIs, but it seems to me that not subtracting homers basically offsets some of the bias that RBIs build in.  Given two players with equal wOBAs but one has lots of homers, the homer hitter is going to bat in the middle of the line-up and he should be expected to have more RBIs (and possible more runs).

pft: you would probably like the Batter’s Blocks and Batter’s Assists that I introduced in this blog a year or two ago.

***

Detroit/5: it is not a “claim”.  It was the conclusion to my presentation.  If you want to debate my presentation, then debate that.  The conclusion follows the evidence.  I don’t start with the conclusion and then find the evidence.

OK, it’s a “conclusion,” not a “claim.” I didn’t mean to imply that you started with the conclusion and then found the evidence.  I expressly said I enjoyed reading the piece.  Your article was quite good.

However, the statement that “without question” the HR needs to be subtracted is flat out wrong.  Lots of folks have questioned subtracting the home run and there is at least a colorable argument for why one might want to do that.  If no one was questioning subtracting the home run, then you wouldn’t have bothered to write the article.  Reaching a sound conclusion is one thing.  Reaching a sound conclusion and asserting that it is without question is what I objected too.  It’s a nitpicky comment on my part to criticize a two-word phrase, but the “without question” phrase struck me as out of step with the rest of the piece.

I think your conclusion is correct once one asserts that the best Runs Produced formula is the one that most closely tracks wOBA.  I’m not sure that that’s the case.

Tango’s values for R+RBI, from BB thru HR, are .29, .47, .84, 1.21, and 2.60. Multiplying by .55 results in .16, .26, .46, .67, 1.43.

Remember the RPA run estimator of Mike Gimbel? He starts out with the following values: .165, .29, .41, .70, 1.44--almost a match for the above. Then he does his on-base adjustment, which I find kind of hard to understand because IMO he didn’t explain it very well, and winds up with a run estimate which is presumably about as accurate as the usual run estimators. So, in that on base adjustment, he must treat the HR so that it ends up with the correct value.

Anyone care to look at the RPA article (google it) and explain how that part works?

The point is that Gimbel’s approach may be the ‘proper’ way to get RC from R+RBI, instead of simply subtracting out the HR.

As I understand Gimbel’s “set-up” adjustment, it works like this (excerpted from longer, unpublished article on RPA):

SET begins with H + W + HB, adds .1 for doubles, steals, wild pitches, and balks, and adds .2 for triples.  Then CS and GDP are subtracted along with the run driving values for all-hit types.  This quantity is divided by plate appearances less home runs plus double plays. 

When you simplify the above procedure, what you get is:

SET = (.71S + .69D + .5T - .44HR + W + HB + .1(SB + WP +BK) - CS - DP)/(AB - HR + W + HB + DP)

A team’s SET is then compared to the league average SET, and is used to adjust half of iR (the weights David posted), with the other half of iR remaining unadjusted.  If a team’s SET is 5% higher than league average, then the final runs scored estimate will be iR, plus 2.5%.  To put it in a formula:

RP = iR*(SET/(2*LgSET) + .5)

When I find the intrinsic linear weights, though, they come out awfully squirrely:

.43S + .54D + .78T + 1.32HR + .38(W + HB) + .13SB - .44CS - .55DP - .057(non-DP out) + .13WPBK

Off of Patriot’s last post, Gimbel’s “set-up” RPA was pretty wacky.  I’m not sure if he always did it the same way, but in the annual I have, he:

1.  Figured out, using his own weights, a batter’s runs per plate appearance.

2.  Divided that in half, and set aside one half as the “drive-in” RPA.

3.  Figured out each player’s “effective on-base percentage,” which iIrc may well be the formula Patriot has in his second paragraph above.

4.  Divided the player’s EOB% by the league’s (which I think, at this time, was around .208). 

5.  Multiplied half of the RPA by the figure derived from step four, and called that the “set-up” RPA.

So if a guy started with a .120 RPA and a .230 EOB%, .060 of that would be “drive-in”, and .066 (.230/.208*.060) would be “set-up”, so the final RPA would be .126.

That’s for individuals; Patriot’s post appears to describe a method for estimating team runs, and he may have used different methods for the two.  I have that annual at home, so I can check it later.