THE BOOK--Playing The Percentages In Baseball

I’m looking at it as well - the writing style is a bit circular in my opinion, and I think things could be a lot clearer if he was more straightforward about getting to the point. I’m only a few pages in, though, so I’ll reserve judgment for later.

well, if I understand it correctly, one reason his measure comes out so well is that it “knows” the baserunner situation for each PA.  None of the other run estimators need to know this, they just look at total PA, BB, 1B, 2B, etc.

And if you’re going to go to that much trouble, why don’t you go ahead and include the out situation as well?  You’ve got a better chance of knocking in a runner on 3B with 1 out than with 2 outs.

Similarly, if you’re going to go to that trouble, how about giving a batter credit for moving the runner over?  I realize those values are probably included in some of the weights given to other outcomes--I probably understand less math than anyone else who reads this site--but if BBI% is calculated for specific batters in specific baserunner situations, I’d like to see something other than the guy gets credit for knocking in him or he doesn’t.  As far as I can tell, when calculating BBI%, a single moving a runner from 1st to 3d counts the same as a strikeout.  The value of that single comes in other places, but I don’t think the “timeliness” of that single comes in. 

And I don’t think there would be any distinction anywhere between a single that moves a runner from 1st to 2d and a single that moves a runner from 1st to 3d.  Is that important? Probably not.

Another factor that is probably not worth including would be the identity of baserunners.  I bet Longoria is going to have a high BBI% this year due in part to the sprinters getting on base in front of him.

The author gets such good results at the team-season level because he has much more detail than the other run estimators do, and he doesn’t even use it particularly well. (At the play-by-play level of data you can do all kinds of things, most of which are just a thinly-veiled form of the equation R=R.)

And, as he notes:

“If a player’s ability to create baserunners could somehow also be accurately determined, and
then joined through multiplication to this same player’s BBI Conversion Rating, we would
suddenly be very close to being able to estimate the number of runs scored by a team composed
of just that player and eight like him. All that would then be missing would be home runs.”

I suppose one could create a theoretical teams version of this estimator, but to what end?

As Hizouse notes, this system only gives credit to advances that score a run or reaching base safely. It does nothing to establish the value of advancing runners without scoring them - a walk with a runner on first will advance a runner to second, but that does not seem to get any credit in this system.

I don’t think it works as an individual-level run estimator. And it’s pointless as a team-level run estimator, due to the level of detail required.

The problem I had was with the introduction of RBI.  As I told him, I can give a great estimate of team runs scoring as something like:
R = (RBI-HR)/.92 + HR

(Or whatever that .92 would give me a best-fit.)

His problem is that he is not using independent variables, but actually using the outcome (RBI) to construct his equation.  You can do the same thing for pitchers, like those Cornell students did with their “strand rate”.

And, I’ll give you the perfect run estimator: RE24.  If you know the player’s stats in each of the 24 base/out states, you will get the run total to exactly match.  All I need is 24 separate equations.

We get a surprising number of papers like this sent to BTN ... I have to tell them that it doesn’t matter that they can beat Runs Created by using situational statistics, since the whole point of run estimators is to just use the basic stats.

And I agree with Colin on “R = R” estimators ... there was one guy on SABR-L a few years back who had an almost perfect estimator, but it used “bases left on” (runner left on third = 3, etc.), and when you do that, you can create an R=R that’s almost perfect.

Tom-- what exactly are you referring to with strand rate? I can’t find anything on the web.

"And, I’ll give you the perfect run estimator: RE24.  If you know the player’s stats in each of the 24 base/out states, you will get the run total to exactly match.”

Wouldn’t be exactly perfect.  Single with runner on 2B.  Sometimes it scores the run, sometimes it doesn’t.  But you’d certainly be very close.

One more thing. The document goes on and on about timing, and how some players hit better with men on base than otherwise.

Never, ever, ever, ever, EVER trust ANYONE - that includes me - who shows you a split and doesn’t show you the split for the average player. For the Retroera:

Empty: .261/.325/.419
Men On: .273/.342/.428

The two things that primarily effect this are pitchers going out of the stretch instead of the windup, and the defense playing differently to hold the runner. Neither of those are hitter skill.

"The two things that primarily effect this are pitchers going out of the stretch instead of the windup, and the defense playing differently to hold the runner. Neither of those are hitter skill.”

Colin, isn’t there also some selective sampling?  Bad pitchers have men on more often, hitters parks have relatively more PAs with men on, batting orders are optimized to have the best hitters up with more men on, etc.

I wonder how much of the difference is due to the stretch vs. windup and defensive positioning (was this covered in The Book?).

To be fair - we are given the average split in the linked document. But even then we’re shown a subset of the data, not everybody’s career MOB split. And we have no context for the discussion.

Looking at wOBA differential between bases empty and men on, the y-t-y correlation (min 100 PAs in the smaller split for both seasons) is only .01, which is insignificant. I don’t think it’s correct to say that “differences in performance can occur and do persist over consecutive seasons,” because there seems to be little to no persistance going on here.

And yes, Vinay, you’re absolutely correct about selective sampling here. I overlooked that. But again, that’s not a talent for the individual hitter.

Colin, regarding post #8, I think there might be one more factor that affects the runners on/bases empty split: the quality of the pitchers in those situations. You’re more likely to be batting with a runner on base when facing a crappy pitcher than a good pitcher. Again, that’s out of the hitter’s control, and he shouldn’t be rewarded for it

I believe that those of you who have made comments have, through no fault of your own, missed the most significant characteristics of Opportunity Runs.  I regret not having done a better job of communicating my ideas.  Please allow me to briefly try again, with several real-life examples. 

[Note: Although the following analysis does not cast as favorable of a light on Hanley Ramirez as some may prefer, please understand that I have the highest respect for his abilities and my evaluation technique had him rated the number one offensive shortstop in major league baseball in 2008.]

1968 Mets

In 1968, the New York Mets drove in 23.12% (353 out of the 1527) of the batters on their team who reached base safely (net of caught stealing and gidp).  This was the fourth-lowest “harvest” rate of any of the 1,316 teams playing in the major leagues during the 53-year period from 1956-2008, and it was very close to being the absolute lowest (1969 Padres at 22.51%).  Specifically:

Baserunners Batted In = BBI = RBI – home runs
‘68 NYM BBI = 434 – 81 = 353

Net times on base = hits – home runs + walks + hbp + roe – gidp – cs
‘68 NYM net times on base = 1252 – 81 + 379 + 43 + 83 – 104 - 45 = 1527

In the more technical and precise parlance of my report, the New York Mets had the absolute lowest BBI Conversion Rating of any of the 1,316 teams playing in the major leagues during the 53-year period from 1956-2008, with 353 BBIs, 458.693 standard BBI Opportunities, and a BBI Conversion Rating of .770 (=353/458.693).

2008 Hanley Ramirez

In 2008, Hanley Ramirez had the absolute lowest BBI Conversion Rating of any 2008 major league batter with 350 or more plate appearances, with 34 BBIs, 43.967 standard BBI Opportunities, and a BBI Conversion Rating of .773 (=34/43.967). 

Please note that Ramirez’s BBI Conversion Rating was very similar to that of the ’68 Mets.

Down and Dirty Run Production Computation for Ramirez

As stated above, the Mets drove in 23.12% of their baserunners with their BBI Conversion Rating of .770.  Let’s make the very reasonable assumption that Hanley Ramirez would drive in a slightly higher 23.21% of his own self-created baserunners with his (0.4% better) BBI Conversion Rating of .773.  Since Ramirez had a net times on base total of 235 (= 177 hits – 33 hr + 92 bb + 8 hbp + 8 roe – 5 gidp – 12 cs), that would yield a predicted 54.54 baserunners batted in (= .2321 x 235) for Hanley if he had eight teammates just like himself in the batting order. 

If we then give the Ramirez “team”: (1) a generous 20% boost in BBI count for his 10% higher “net total bases per net times on base” rate compared to the ‘68 Mets (1.33 vs. 1.21, counting extra bases on hits and stolen bases in the computation), and (2) a standard 1956-2008 scoring boost of 8% associated with uncredited BBIs (those due to errors, gidp, pb, wp, stealing home, and interference), we end up with an estimated total of 70.68 adjusted BBIs (54.54 x 1.20 x 1.08 = 70.68), excluding those which would accrue from driving himself home via home runs.  We complete the calculation by adding in Ramirez’s 33 home runs, and we end up with a “down and dirty” grand total run production figure of 103.68.

Please note that Opportunity Runs (considerably more precise in its techniques) rated Ramirez at 105.79.  Please also compare these figures with those of the eight well-known run estimators evaluated in my report, none of which includes baserunner context in the calculation:

Hanley Ramirez, 2008

Runs Created (James)—125.61
Runs Created t-1 (James)—135.79
OPS+ (Baseball-Reference.com version)—126.61
Linear Weights (Furtado version)—126.38 Linear Weights (Johnson version)—125.83
Total Average equation (Boswell)—133.47
DX equation (Cook)—125.73
Base Runs (Smyth)—126.76

Conclusion

Although not the norm, individual player discrepancies between evaluation systems of the size summarized above are not uncommon throughout the 53-year period studied.  Incidentally, Opportunity Runs predicted run output for the ’68 Mets at 466.99, or a deviation of 1.27% from the correct total was 473. 

One may argue, possibly at the risk of being incorrect, that clutch hitting and variations in BBI performance do not exist for individual players when given enough plate appearances (perhaps 20 years or more). But it seems quite unfair to ignore such evidence when granting MVP and other awards for past performance. 

My main point in writing this report was to bring attention to certain injustices in rating historical achievement.  Thank you for allowing me that opportunity.

Clint/12 was marked for moderation and is now open.

Clint: why not look at Linear Weights by the 24 base/out states (RE24), which is the only metric that can be fairly compared to yours:

http://www.fangraphs.com/statss.aspx?playerid=8001&position=SS

Hanley was +38 runs above average in 2008, which works out to 121 runs created. You’ve got Hanley at 15 runs below that, which is unsupportable.

I dispute that your method is more precise in its technique. 

***

My personal beef: no need to show things to two decimal places when saying someone created “125.63” runs.  That’s 126.  No knock on you, since (too) many people do what you do.

Clint: I find the fact that HanRam knocked in 10 fewer runners than the average player given the same PAs to be interesting, and that certainly is relevant to his value in 2008.

But HanRam was also, I suppose, better than average at not-making-outs with runners on base.  When figuring BBI, the outcome of PA in your calculation is binary; either there is success (RBI) or failure (no RBI).  RE24, which is new to me, would take into account the fact that some “failures” (i.e., a walk or hit that doesn’t score a runner, an out that moves up a runner) are better than others.

I also don’t understand why HanRam’s poor BBI conversion rating or the 1968 Mets BBI% is relevant to figuring out how much HanRam’s teammates could have or should have driven him in.

I would also recommend looking at Tom Ruane’s Value Added:

http://www.retrosheet.org/Research/RuaneT/valueadd_art.htm

It’s probably important to note that the BBI Conversion rate (B Factor) that Clint is talking about, is really only looking at the success rate of runner’s driven in (less HR).  This component really has nothing to do with Ramirez’s overall run creation, as it is ignoring the run scoring value of his own HR as well as what he does to set the table for others (captured in the BBI Creation -
A Factor).  For this component alone, he’s saying that Ramirez is missing 10 RBI’s (44 est. vs 34 actual).

RE24 is a fascinating technique, but its strength (giving precise fractional run credits and debits along the way, including for the intermediate steps between reaching base and actually driving home the runner) can also at times mislead readers lulled into acceptance by its appearance of highly-detailed precision.  Here is why:  in the end, each of the 24-state values are based on the average player’s ability to eventually “bring home the bacon” with a BBI.  Even if one uses multiple sets of 24-state values, customized for various team scoring strengths, the same basic assumption is still in place – each set of values will be based on the ability of the average player within each subdivided group to bring home the bacon. It is therefore NOT a perfect measuring system when dealing with players who have very high or very low BBI Conversion Ratings compared to their overall offensive numbers. 

If there was a special 24-state matrix for the 2008 Hanley Ramirez Squadron, one with home runs pulled out of the numbers, leaving us basically with the scoring potential of baserunners, then it would have significantly lower net 24-state numbers than the norm.  In fact, they would much more closely resemble those of the 1968 Mets than they would the 2008 Shin-Shoo Choo Rippers (BBI Conversion Rating of 1.455), the 1963 Henry Aaron Bombers (Conversion Rating of 1.671, the highest I’ve come across so far) or the 1994 Cleveland Indians (highest team Conversion Rating of the past 53 years at 1.209).  And if such a customized Ramirez 24-state matrix existed, I think you would find that my numbers match up with it quite well.

Please remember also that, within the Opportunity Run system, over and above Hanley’s basic on-base event values, he receives an 8.7% BBI production boost for his high ratio of on-base events to outs, and an additional 8.8% BBI production boost for presenting “his” team with higher-than-expected BBI Opportunities due to the above-average number of baserunners he’s not driving in.  Even with these boosts, amounting to an overall 18.3% jump in BBI Opportunities Created over initial values, he would produce “only” 72.79 adjusted BBIs (runs minus home runs) and 105.73 runs total(excuse me, I meant 106 runs). 

The 73 baserunners who would score, subtracted from the net 235 that Ramirez put on base (= 177 hits – 33 hr + 92 bb + 8 hbp + 8 roe – 5 gidp – 12 cs), would leave the Squadron with 162 runners left on base per his 425 outs made.  This works out to an average of 1.14 LOB per inning (compared to the 1956-2008 MLB average of .83 and the ’56 Red Sox, at .99 the highest on record).  But that’s what happens when you have a combination of a very high on base outcome (1.477 BBI Opps Created Rating) and a very low conversion outcome (.773 BBI Conversion Rating, where 1.000 = the 1956-2008 MLB average for both categories).  This is not an entirely new situation for Ramirez, incidentally.  In 2007 he had a Created Rating of 1.331 and a significantly lower Conversion Rating of 1.055, and in 2006 he had a Created Rating of 1.242 and a significantly lower Conversion Rating of 1.027. 

Please understand, I am not being critical of Hanley Ramirez in an overall sense.  He’s the best offensive shortstop in baseball and I would absolutely want him on my team.  He just hasn’t been as valuable as people have give him credit for over the past several seasons.  One thing that perhaps everyone can agree with—he was a good choice for a leadoff hitter the past three seasons.  It will be interesting to see how he does this year in the number three spot in the order when we has gone a little farther into the season.

Clint/18: I disagree totally with your objections in paragraph 1.  Hanley Ramirez plays with 8 other players who are NOT Hanley Ramirez.  And so, we exactly want to know what happens when he plays with 8 average players.

Echoing Tango’s sentiments here… In using OR to evaluate individual players, Clint has repeated the flawed technique that James used to use with Runs Created.  Basically, he’s multiplying a players set-up component by his own drive-in component.  Which as we know over-values players who are exceptional at both aspects of the game.  This is the reason that James now uses his theoretical team approach within Win Shares. 

I’m wondering now, how the individual runs were calculated for the non-linear metrics such as RC and baseRuns.

In general terms, additive systems (linear weights and RC24) attempt to predict the value of a player in an average lineup, and multiplicative systems (Opportunity Runs and Runs Created) attempt to predict the value of a player in “his own” lineup.

I think it is safe to say that teams with personnel having overall average BBI conversion rates will be the ones most commonly having predicted team run outputs (via linear weights) that match actual run outputs.  This therefore also means that players having average BBI conversion rates will be the ones most commonly having predicted individual run outputs (via RC24) that match the player’s “true” run production. 

But not everyone is looking for the same thing in a run estimator, and so “true” is perhaps a bad choice of words.  That is, not everyone is looking for maximum run prediction accuracy at the team level. 

One school of thought is primarily looking for accuracy in the analyses of game strategies first and players second.  The other school of thought is looking primarily for accuracy in assessing past player performance first (with an eye toward the Hall of Fame and awards such as MVP), and strategies second.  Mr. Tango and others, of the first school, believes an average lineup is the proper way to go, while the second school, to which I belong, believes that the player’s self-created lineup is the proper way to go.  Both schools are correct, based on their goals. 

The accuracy of any run system, additive or multiplicative, is a function of the accuracy of its components and the manner in which they are assembled.  I believe that the accuracy “problem” with Runs Created at high and low-end levels of production is based on design features in two of its stated components and the fact that a third component is missing altogether.  Specifically:  (1) RC includes home runs in its on-base component, which, for maximum accuracy, it shouldn’t, (2) RC overvalues doubles, triples, and home runs in its slugging component, and (3) RC ignores differences in batter performance when the bases are empty versus when the bases are occupied.  That said, however, I believe that RC is also a brilliant, beautiful, elegant, and simple system that has given far more understanding to baseball fans than any deficiency can ever take away. 

The fact that RC multiplies its A and B components is not a flaw in design, but a feature which emulates real baseball.  A team is like a giant, amorphous “individual” who gets a crack at harvesting his own self-created baserunners.  Any system that incorporates this design feature has a built-in accuracy advantage – particularly at high and low ends of the production spectrum – compared with systems that don’t.  Multiplicative systems will be more accurate than additive systems in predicting run output over the full spectrum of teams – if their components are designed and assembled with equal skill. 

But prediction accuracy of systems at the team level is only the main priority for one group of analysts, and I suppose it would be best for all if we didn’t talk past each other so often.

I’ll also point you to this:

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

(Edit: link updated)

The purpose is to analyze what the players did in the past for his team and adjust for opportunity.  Context and the actual outcome of what a player did is important.  The article shows the limitation these non-contextual run estimators have in evaluating a players season.

RBI’s are a measure of what a players has actually done.  When adjusted for opportunity, it should be an important part of any comparison between players, such as for MVP consideration.  Ignoring RBI’s is one of the unfortunate consequences of the Bill James era, which correctly noted the importance of getting on base for the RBI guys.  Now that the means to adjust RBI for opportunity are available, there is no long any excuse to ignore it.

One minor weakness relates to getting on base and creating RBI opportunities. Getting on base does provide RBI opportunities for other players in the lineup.  But is David Ortiz earning a walk an equal opportunity for an RBI given his problems advancing on hits as an Ellsbury walk, or is a # 7 hitters walk with weak hitters behind him as valuable as a # 2 hitters walk.  I really would like to see some adjustment based on batting order position and base running ability, but maybe that’s asking for too much. 

His use of PF is unfortunate though.  Ted Williams gets penalized in a park that suppresses HR’s and RBI’s for LHB while providing great benefit to RHB (the basis for the high PF), while switch hitting Mickey Mantle benefited from Yankee stadium more as a LHB than he was penalized as a RHB, and gets a boost from the use of low PF due to it’s adverse effect on RHB’ers.  It would be best to not use PF for individual players unless it is clear the player got a boost or was harmed.  Tony Gwynn for example was hardly hurt playing in San Diego since he was a singles hitter, yet gets a PF boost because his park suppressed HR’s.

You are wrong, pft (and Mr. Meyer, to whom the rest of this is mainly addressed).

You are ignore the simple and fundamental way in which a baseball offense works. You are using simple cricket logic of one batter, one runner, with a binary logic of either a batter drives in a runner or he strands him.

This is not true, by which I mean that it is false, by which I mean that it is wrong. This is not a failure of anyone here to understand what you are trying to do, this is a simple extension of the fact that you are wrong and therefore your conclusions are wrong.

Imagine the start of an inning. The batter hits a single, and is on first base. The next batter strikes a single as well, sending the first batter to third base. Now there are runners on the corner. The next batter hits a single, scoring the first batter and sending the second batter to third, still with runners on the corners. The next three batters strike out, so we end up with one run scored in the inning.

Here is how it shakes out:

The first batter is credited with a run.
The second batter is credited with nothing, save a single.
The third batter is credited with an RBI.

But all of the batters did the same thing! Was the second batter’s single “untimely”? No; he hit it with men on base, didn’t he?

But the RBI ignores that. The RBI only cares about who was at bat when the run scored, not the entire sequence of events required to produce that one run. This is why the RBI can match up to team run scoring without having any real application at the player level - it gives no credit for advancing the runner. Anything above and beyond that is still wrong because you are starting out at wrong instead of starting out from something other than wrong.

You want to look at credit for actual runs scored? Look at Tango’s batting assists:

http://www.insidethebook.com/ee/index.php/site/article/introducing_batting_assists_and_batting_blocks/

Until that point you are taking wrong and doing some fancy math to it so that it can be wrong to two decimal places.

Saying again, just for clarity’s sake: I understand what you’re doing here, seemingly better than you do because I understand that it’s wrong. Please do not assume that my disagreement here is based upon my not understanding the concepts here.

There are two common questions a run estimator can ask when applied to individual players:

1. How many runs did he add to his team?
2. How many runs would a team made up of 9 clones of this player score?

Hanley Ramirez hit worse than normal with runners on base.  He also hit fewer times with runners on base than most other players. (about 33% last year, average player gets 45%).

To answer question #1, seems like a good approach would be to take the typical run estimate, 126 or so, and subtract 10 given that he drove in 10 fewer runs than you’d expect (though make sure you aren’t penalizing him for drawing walks with runners on base).  You absolutely CANNOT answer this question by pretending Ramirez’s RBI ability is interacting with his on base ability.  His teammates drive him in, and his teammates are driven in (or stranded) by him.

Question #2 is theoretical but still valid.  This approach can’t work though.  Ramirez will not have 66% of his PA without runners on base when he’s batting in an order with 8 other .390 OBP Hanleys.  He’d probably have runners on for 55-60% of his PA in such a lineup. Something has to give:

Either more of his good PA happen with runners on base, and his BBI% goes up and a lot more runs score than your estimate, or:

He continues to hit poorly with runners on base.  Since there are so many more runners on base, his overall stats end up much worse than what we currently see in his stat line.

Batting Assists and Blocks are an interesting concept, but they are themselves missing situations where runners are moved forward on the basepaths but who do not score on that play, or score ever. 

But even if that problem was corrected, what then do you do with these numbers?  How, specifically, do they relate to the scoring of runs, and what sort of run value would you assign to them?  RC24, which measures every movement and non-movement around the bases one small step at a time, assigns a run value to each, and adds them up in simple and tidy fashion, seems to address all issues addressed by Assists and Blocks, but stated in terms of runs.  RC24 is a very interesting and nearly perfect evaluation system.  Nearly.  Its only significant flaw is that its analysis is based on average MLB run outcomes whether we are dealing with Shin-Soo Choo (BBI Conversion Rating of 1.455), Hanley Ramirez (BBI Conversion rating of .773), or everybody in between.  That’s a lot of territory to cover accurately by one matrix which cares nothing for BBI Conversion Rates. 

I believe that utilizing a custom 24-state matrix for every player – based on all of a player’s offensive credentials, specifically including his ability to bring home the bacon with a BBI—would yield basically identical results to Opportunity Runs.  It would take another large study to demonstrate this conclusively, but perhaps some of those who have used their very valuable energy and intelligence to discuss Opportunity Runs could begin to directly assist in that process.

Let me be clear, I fully understand the concerns expressed about Opportunity Runs because I have had the same concerns myself with earlier versions of the equation.  I believe I have effectively dealt with those concerns in the current design of the OR system, and have addressed them in some detail in both my report and in the paragraphs below.

First, to summarize the concerns:  some readers believe that “intermediate events” are missing from the OR calculation because only the first step in run production (getting on base), the last step (driving in the runner), and, of course, the home run step, seem to be counted.  This is not true.  But if it was true, such a design flaw would then undervalue three categories of hitters: (1) those who have little-to-no power and whose array of hits therefore features a disproportionally high number of singles and walks with runners on first, (2) hitters like Hanley Ramirez, who have plenty of power but who suffer from unfortunate timing with runners in scoring position, and (3) high-quality hitters who draw a significant number of walks with runners on base.

These are all excellent concerns, and in the remainder of my response I will describe three specific ways that I have dealt with them.  Thank you for caring enough about the subject to plow through a few more paragraphs of discussion.

Opportunity Run Component A1 gives relative weights of 1.00, 1.15, and 1.30 to singles, doubles, and triples, as opposed to, for example, weights based on average eventual scoring probabilities (MLB 2003-07) of 1.00, 1.71, and 2.33, or weights based on the very simple and obvious base count of 1, 2, 3.  The weights eventually chosen were the most accurate ones I could discover in attempting to design Component A as a device to accurately “track” the actual baserunner situations confronting batters on all 1,316 teams from 1956-2008. And they happen to work very well in this regard, as we will see in a moment. [The reasons why these weights work are discussed in some detail in the full report.] In any event, such weights greatly benefit players in the low-BBI, high-intermediate advance categories mentioned in the first paragraph above. For example, Hanley Ramirez was rated 29% above average in Component A1 (this, for example, in comparison to his traditional on base average of .400, which was not only just 22% above average, but which includes home runs in the computation).  Suffice it to say that Component A1 values Mr. Ramirez and other hitters in the disadvantaged group a great deal. 

Opportunity Run Component A2 then gives a BBI Opps Created boost to players with an above-average ratio of weighted on-base events to outs.  This was done to model reality in that higher ratios, on average, mean extra runners getting into scoring position with less than two outs.  This component particularly benefits above-average hitters with a high rate of singles (and sometimes doubles, both of which “invisibly” move runners into much better position to score).  Hanley Ramirez, for example, received an 8.7% boost in BBI Opps Created via this component.

Opportunity Run Component A3 then gives a final BBI Opps Created boost to all players with a below-average BBI Conversion Rating. (This component also boosts hitters with good conversion ratings, but who are walked excessively and have BBI Conversion ratings below the level that they would have if they were not pitched around so often). Again, all this was done to model reality.  Such hitters and their teams actually “produce” more baserunner situations than their basic event count would indicate because they are leaving an above-average number of runners on the basepaths from one batter to the next.  Component A3 benefits every hitter in the disadvantaged group.  Hanley Ramirez specifically received an 8.8% BBI Opps Created boost via this component.

All things considered (A1 x A2 x A3), Hanley Ramirez received the second-highest BBI Opps Created Ratings (1.477) in all of Major League Baseball in 2008, behind only Chipper Jones.  This means he was nearly 50% better (!) at setting the table for other hitters than the average player from 1956-2008. If one may say such a thing, BBI Opps Created “likes” the disadvantaged group a great deal.  I do not believe that the intermediate run production contributions of Hanley Ramirez, or any other player, are being shortchanged.  My “belief” aside, there is considerable statistical proof of this. 

Namely:  BBI Opps Created successfully tracks BBI Opps Faced (weighted baserunners available for hitters to harvest) at an average deviation rate of only 1.99% for all 1,316 teams during the entire period 1956-2008 (with each team calculated individually).  This is considerably more accurate than the best estimator equation is at predicting team-by-team run outputs over the same period of time (average deviation rate of 2.44%; please see analysis in my report).  After folding home runs into the mix, Opportunity Runs achieves an average deviation rate in predicting total run production of only 1.49% on a team-by-team basis for all 1,316 teams during the entire period 1956-2008.

This is 61% of the deviations of the best competitor, and approximately half the deviations of, for instance, OPS+.  Opportunity Runs are also within 1% of actual team run production 40% of the time and have deviations of more than 5% less than 2% of the time. And since we are dealing with teams that have plate appearance totals equivalent to ten years in the career of a full-time player, these are significant accuracy issues, and impossible ones for non-context run estimators to overcome no matter how they weight and combine their events.

On an individual player basis, non-context equations are all capable of greatly overvaluing or undervaluing at seemingly random moments without user knowledge.  The largest discrepancies between Opportunity Runs and other run estimators are almost always in the 30-40% range for players with 60 or more runs produced in any given season.

In conclusion, Opportunity Runs has worked extremely hard to be fair and accurate with all hitter categories, and I believe that it has been quite successful in this attempt.  Thanks very much for the opportunity to present this information in the past few days.

Just as important as math in this discussion is the fact that, at its heart, the whole idea of Opportunity Runs is both logical and reasonable.  To use an analogy:

“Food placed on the table (BBI Opps Created) times percent of food consumed (BBI Conversion Rating) equals the amount of food consumed (team BBIs, or individual player BBIs on a one-man team).  Add dessert (home runs) and you’ve got the full meal calculated (total run production).”

As long as BBI Opps Created can reliably predict BBI Opps Faced, everything is great.  That it most certainly can do, and, as some of you may have guessed, this was not a trivial accomplishment. 

Or, one could try to look at the whole concept separated into little pieces. If someone asked a fan if it would be of any interest for him or her to see a statistic like the BBI Conversion Rating (BBI/BBI Opps Faced), either for a team or a player, he or she might say something like “Of course! That sounds like a great stat. Where can I get my hands on it?” After pondering the idea a moment, the fan might add “Of course, I would hope that it could somehow be taken into account that certain players are pitched around more often than others.  Assuming that occurs, yeah, it sounds like a great stat.” [Please see Opp Runs factors A1, A2, and especially A3 regarding this walk issue.]

If someone separately asked a different fan if it would be of any interest to him to see an equation that could very accurately predict the baserunner situations faced by every major league team from 1956-2008, he might say something like “As an intellectual exercise I would very much like to know how someone would do go about doing such a thing, but I’m not sure what I would do with it once I had it.” When asked if it would make any difference if all the players on the team in question had absolutely identical stats, the fan would look at the questioner blankly and say “Well, I don’t see why that would make a difference. In all the teams of the past 53 years, there must have been some of them with player performance that was tightly bunched.  You said this equation works accurately for all the teams during the era, right?  Then I guess what difference would it make if all the players on a team had identical stats?”

And finally, if someone asked a third fan if he could think of anything that he could do with the baserunner predictor equation [component A] and the BBI Conversion Rating [component B] if he had them both in his hands at the same time, the fan might say, “Heck, yes. If I was really bored, I could multiply the two, get baserunners batted in, add home runs, and, presto, I would have figured out run output for a team—or, better yet, run production for an individual player.  You know, that actually sounds like a really good way to rate a ballplayer’s offensive contributions.”

After thinking a bit longer, he might add “But the baserunner prediction equation better be pretty darn good, and very versatile, or total run predictions would be all over the map and not worth a bucket of warm spit.”

Right at this point is where my confusion with the comments of certain readers begins. 

It is implied by some comments that it is fairly trivial to accurately predict baserunner situations for 1,316 teams over 53 years, one team at a time. For this reason alone, Opportunity Runs has apparently been discounted by certain readers. The reality is that A2 and A3 took a great deal of thought to conceptualize, months to fine tune, and are unique in the world of run estimator equations. Together they not only greatly enhance the accuracy of Opportunity Runs, but are versatile enough to deal effectively with conundrums like the Hanley Ramirez situation. 

In any event, once component A has rather amazingly predicted the denominator of component B, the goalposts have been mysteriously hauled away in a pickup truck right after the snap and just before the kick. It suddenly somehow became invalid to use Created and Faced in the same final run estimator equation because .  .  . they match too well! Of course they match too well. That is exactly the idea. 

I mean, if they don’t match—not just in the aggregate, but on a team-by-team basis—the final Opps Run equation won’t be worth a darn.  And if they do match, then the equation has been validated for players as well as teams.

Put simply, because of the interlocking nature of its components, the accuracy of Opportunity Runs is an effective demonstration of the validity of the concept as a whole.

And if they do match, then the equation has been validated for players as well as teams.

This statement is absolutely not true as been discussed several times in the past.  Teams have a very narrow variation in their statistics compared to individuals and you absolutely cannot extrapolate from that narrow range to the wider range of players.  Evaluating a metric intended to be used on players by testing its resulats on teams is meaningless.

The reality is that A2 and A3 took a great deal of thought to conceptualize, months to fine tune, and are unique in the world of run estimator equations.

I guess in an ideal world the effort put into a project would always be rewarded by the ultimate value of the end product.  Unfortunately, in the real world that is not always the case, and it certainly isn’t the case here.  I don’t expect you to believe me any more than you have the others that have been critical of your metric, but what you have ended up with, with all the weightings and manipulations, has neither the theoretical basis or the elegance of the run value added metric that nearly perfectly measures the past accomplishments of a batter by including the context of the entire baseout situation faced by the batter for every plate appearance.  Measuring past accomplishments is much less important than predicting future performance, so there is going to be much less interest in a metric that admittedly is not designed to predict future performance, but there should be no interest in a metric that does it less well than an already existing metric.