Tango on Baseball Archives

© Tangotiger

Archive List

OPS: Begone!
by Tangotiger

Team of same players

The following table presents various theoretical teams that are equivalent in expected Runs Per Game. The teams range from high-walks, low-power to low-walks, high-power. The equivalency is based on BaseRuns. If someone wants to run these teams through their favorite simulator, we should expect these teams to also be equivalent.

PA AB H 2B 3B HR BB Outs AVG OBA SLG BsR
620 600 160 46.6 6.2 24.9 20 440 0.267 0.29 0.489 75.6
640 600 160 37.7 5 20.1 40 440 0.267 0.313 0.447 75.6
660 600 160 30 4 16 60 440 0.267 0.333 0.41 75.6
680 600 160 23.4 3.1 12.5 80 440 0.267 0.353 0.378 75.6
700 600 160 17.6 2.3 9.4 100 440 0.267 0.371 0.351 75.6
720 600 160 12.4 1.7 6.6 120 440 0.267 0.389 0.326 75.6

Assuming that all these teams are equivalent, what metric can you construct using only OBA and SLG? The best-fit equation among the above samples is: 1.56 * OBA + SLG. Here is how OBA+SLG (OPS), the best-fit, and 3*OBA+SLG (3OPS) look:

AVG OBA SLG OPS Best-Fit 3OPS
0.267 0.29 0.489 0.78 0.942 1.36
0.267 0.313 0.447 0.759 0.934 1.384
0.267 0.333 0.41 0.743 0.93 1.41
0.267 0.353 0.378 0.731 0.929 1.437
0.267 0.371 0.351 0.722 0.93 1.465
0.267 0.389 0.326 0.715 0.933 1.493

This table shows that OPS undervalues teams with high OBA, and overvalues teams with high SLG. 3OPS does the reverse: it overvalues team's with high OBA, and undervalues teams with high SLG.

Team of real players

However, this is if you have a team of players that end up producing the above totals. The impact of a player with the above profile, inserted into a typical current team, will not be similar. Let's construct some new profiles of players, and insert them into a team of 8 typical profiles, so that these new teams of 9 players are all equivalent. (This will more closely resemble the real-world scenario.)

PA AB H 2B 3B HR BB Outs AVG OBA SLG BsR LWTS
620 600 160 44.4 5.9 23.7 20 440 0.267 0.29 0.479 74.1 -0.8
640 600 160 37.2 5 19.8 40 440 0.267 0.313 0.444 75.2 -0.4
660 600 160 30 4 16 60 440 0.267 0.333 0.41 75.6 0
680 600 160 23 3.1 12.2 80 440 0.267 0.353 0.376 75.2 0.5
700 600 160 16 2.1 8.5 100 440 0.267 0.371 0.343 74 1.2
720 600 160 9.2 1.2 4.9 120 440 0.267 0.389 0.311 71.9 1.9

You will notice that the BaseRuns of these players are not the same. This is because within the context of their teams, they do have the same impact, but if they were part of a team where all the players were like them, they would not have the same impact. I also present an extra column, Linear Weights, which shows that while they are still not equivalent, Linear Weights comes very close in capturing their equality.

Not presented here is the teams of each of these players with the 8 typical players. It's boring to look at. Just take it from me that these team BaseRuns are all the same.

Ok, so what we have here are 6 distinct types of players, all of which, if they were inserted into a typical current team of 8 typical players, would have each of their teams scoring the same number of runs. Now, let's find the best-fit equation. To make all these players equivalent using only OBA and SLG, the best-fit equation is: 1.75 * OBA + SLG. Here again is how OPS, the best-fit, and 3OPS sees these players.

AVG OBA SLG OPS Best-Fit 3OPS
0.267 0.29 0.479 0.769 0.987 1.35
0.267 0.313 0.444 0.757 0.991 1.382
0.267 0.333 0.41 0.743 0.993 1.41
0.267 0.353 0.376 0.729 0.994 1.435
0.267 0.371 0.343 0.715 0.993 1.458
0.267 0.389 0.311 0.7 0.991 1.477

Conclusion

The best-fit was based on these 6 samples of players. These are not representative of all players. If you really wanted to find the best-fit among actual players, you'll have to repeat what I did (inserting each player into a team of 8 typical players), but for a much larger sample. I would expect the best-fit equation to fall somewhere between 1.7 and 2.0. If you must rely only on OBA and SLG to establish a player's current run production, it would probably be easiest to do 2*OBA+SLG.

May 20, 2003 - Kevin Harlow (www)

I had derived a 1.56 factor for OBP in a 4.5 RPG context. Derivation is at my homepage.

http://www.kevinharlow.com

May 20, 2003 - Ted Arrowsmith

From Moneyball, I got the feeling that the OBP*3 stuff had as much to do with the value of taking pitches and the effect this has on opposing pitchers. Is there any data regarding this? It would, I imagine, involve some serious programming to look to look at the effect of high and low pitch-per-plate-appearance guys who change teams.

Great work Tango!

May 20, 2003 - Patriot

If the effect of taking pitches and the like was so great, then a formula predicting runs that weighted OBA 3x as much as SLG would be a better predictor than one that doesn't. It isn't. For predicting team runs, something like 1.8OBA+SLG correlates best as Tango says.

May 20, 2003 - tangotiger (www) (e-mail)

The A's have an additional point that by being able to work the count longer, a team can "choose" their opposing pitchers to the point where the average opposing pitcher is worse than by random chance.

They "choose" the pitcher by forcing their opponent to bring in the 10th reliever, because they wore out the starter. While this is certainly conceivable, you would need a whole team of such batters for this to work. As well, there's no guarantee that your team will benefit from it, since your opposition's next opponent might reap the benefits.

In the end, we are talking about a max .20 run difference/GP (see a previous Clutch hit for calculation), if the whole team is like this, and they are the ones who get the benefit. I fail to see how jumping the OBA to 3x from 1.8x would capture this. The "extra pitches" is not a function of OBA, but of (BB+K)/PA. By jumping the number from 1.8 to 3, you are capturing only part of this effect (BB/PA), in a whole bunch of other noise (H,HR,outs). This extra 1.2 is sort of trying to rise above the noise to find the BB/PA. If this is what the A's are trying to do, I don't think they're doing it in the best way. It's hard to comment further, without having the specifics (like James / Todd Walker comment as the best #2 hitter). From what we think they are trying to do, they are wrong.

May 20, 2003 - Ted Arrowsmith

Maybe I'm mistaken about this, but if there's an interaction between OBP and slugging due to the value (if it really exists) of taking pitches, then it would not show up at the team level.

Let me use an impossibly extreme example to illustrate my point: a player is so good at taking and fouling off pitches that he averages 20 pitches per plate appearance. This allows his teammates to face pitchers who or more fatigued or, better yet, bad pitchers from the bullpen. The value of this player's ability to see 20 pitches per PA would not be primarily his OBP but the better OBP and SLG of his teammates when he's playing. The relationship between OBP and SLG at a team level would not change.

This is what I thought DePodesta was suggesting: that a player with a high OBP helps his team more than a high SLG guy becasue of this effect of taking pitches. Studying this effect (if it exists), especially since in reality the differences are just a couple pitches per PA, would be difficult from a computer programming perspective: do the other A's, all other things being equal (there's the rub), hit better with Hatteberg in the lineup than without him.

May 20, 2003 - Darren

As stated in Moneyball, it was "an additional point of OBP is worth 3 additional points of SLG."

I'm guess that this means from a baseline of about .300. That makes the following about equal:

.350 OBP == .450 SLG .400 OBP == .600 SLG .450 OBP == .750 SLG

Just a guess.

May 20, 2003 - tangotiger (www)

The "additional point" thing is what I'm capturing. It doesn't matter if you do: 3*(OBA-.3)+(SLG-.35) OR 3*OBA+SLG-1.25

It's the same thing.

========== As for the "wearing out the starter", Ted is correct in his approach. If you have a team of player's whose "true talent level" was .333/.400, this team would score about 4.5 runs per game. However, because these guys all work the count, they have a synergistic effect in tiring out the starter, and bringing in the 10th man. These guys, because they feed off each other in this manner, will end up with .343/.405 numbers (let's say). Now, all of a sudden, this team of talent of .333 with the synergy effect, acts just like a team of .343 with no synergy effect.

This extra effect the A's are capturing inside the OBA, by overweighting that metric. However, there's no reason to rely on such a noise-filled metric, when what you want is (BB+K)/PA or (pitches/PA). Because of the amount of noise, to try to capture the little extra pitches/PA in the OBA, you have to severely overvalue the OBA to find it.

May 20, 2003 - Ted Arrowsmith

Thanks for the analysis Tango.

There is a sense in which OBP may be 3 times more valuable then SLG that directly applies to the A's. An increase in runs scored through raising SLG may be about 3 times more expensive (in terms of salary or the quality of player needed in a trade) then the same increase in runs caused by OBP. Thus, it might make sense for the A's to focus on OBP.

May 20, 2003 - tangotiger (www) (e-mail)

The other reason for using "3" for OPS is if you are actively looking for those types of players. If you really really want guys with high OBA, then you would overweight OBA. You would do this because maybe you feel that it's a better predictor of future production. Or you feel that you need to get the players to toe the company line, or whatever. Guy like Vlad, Nomar, and Soriano would not be properly appreciated in such a system.

May 20, 2003 - McCoy

What SLOB how does it rate compared to 3OPS and 2OPS? I have often heard that it is better than OPS, is is also better than the others?

May 20, 2003 - Vinay Kumar

SLOB, which is OBP*SLG, works well at the team level (better than anything OPS-like). Note that RC=OBP*TB=OBP*SLG*AB=SLOB*AB.

It works well because it captures the interaction between OBP and SLG. However, for this reason, it doesn't work as well on players, because their OBP and SLG don't interact with each other (extreme example: when Barry Bonds is walked, he doesn't get a chance to drive himself in). This is what Tango was measuring in his second example, when adding different players to a team of 8 typical players; each player's OBP and SLG interact with those of his teammates.

May 20, 2003 - Art

Looking at Tango's 6 teams (the first table) even though they have the same expected number of runs scored the expected number of wins would only be the same if they had the same distribution as well... normally this would be the case (or close to it) but I wonder if high slg/low obp teams have a larger variance than the low slg/high obp teams... and that variance will affect the number of wins for that team. For a team like Oakland (good hitting & pitching) the lower variance would be beneficial (i.e. they'd prefer to score 5 runs every game than to score 2 runs half the time & 8 runs half the time). For a team like detroit (lousy hitting & pitching) they'd probably like the variance (getting shut out half the time & 6 runs half the time versus scoring 3 every game). Don't know if there's anything to this...

May 20, 2003 - Rob Fornoff

Why is AB the denominator here instead of PA? The top player in the list has to have 100 extra base hits turned into singles to become the bottom player in the list, but then he gets to add 100 walks, too. With all those extra plate appearances, it seems that bottom player will have more opportunities to produce runs, thus causing the OBP component to appear more valuable.

Or is outs the denominator in the comparison? (By denominator, I mean the number you have equalled out to all players; I don't know what else to call it.) I understand the purpose of what you're trying to do here, I just don't understand how you came up with your starting point.

May 20, 2003 - tangotiger (www) (e-mail)

This is how SLOB*k and SLOB*PA*k (where k is some constant to make things add up nicely) for 6 equivalent players from that last chart look: 

 81 	 76 


 81 	 78 


 79 	 79 


 77 	 79 


 74 	 78 


 70 	 76

SLOB by itself works ok, except at the real extreme. SLOB*PA works much better. SLOB*PA is essentially Runs Created, and we already know that BaseRuns is more logical/accurate than Runs Created.

The best one in this group remains static Linear Weights. The best one "on the market" right now is BaseRuns-generated custom Linear Weights.

May 20, 2003 - tangotiger (www) (e-mail)

Rob, you know what, you are right! I goofed.

While I was using outs as my baseline in the last chart, I should have used PA instead. Each player on the team should have the same number of PAs, not outs. Let me re-run the chart, and I'll publish the update on my site.

Good catch!

(Vinay, you are right about RC = SLOB*AB, and not PA as I mentioned in my last post.)

May 20, 2003 - tangotiger (www) (e-mail)

For the last example, I should have been more careful.

What happens is that I should fix the team outs to something. In my example, I actually fixed it to each player making the same number of outs (440) which is wrong.

Anyway, what I now did (see link) was started with the team outs (3960), and, making sure each player had the same number of PAs, found the 8 typical guys and the 1 variable guy that would produce 3960 outs.

Things actually change. The Best-Fit becomes 1.64 (and not 1.75). I suspect that the best-fit will fall somewhere between 1.5 and 2.0, and for ease, probably use 1.5.

(Static) Linear Weights now looks less good than originally. I like this change, as it shows that the component values should change if the underlying environment also changes.

Custom Linear Weights wouldn't have this issue. Though at this point, I don't want to pronounce that custom LWTS will see all these guys as the same. It would definitely see all the teams as the same (just like BaseRuns). I think there will be some differences among these players though through custom LWTS. I'm not sure how much difference though.

Great catch again, Rob!

May 20, 2003 - gd

Excellent work as always, tango.

It's just a trivial statistical nicety, but I noticed that the (OBP * x + SLG) equals about 1.0 for the range of x being discussed given the current league OBP/SLG figures. For example, in the 2002 AL, where the league OBP/SLG was .327/.424, 1.76 * OBP + SLG = 1.000. For Soriano his .332/.547 = 1.125. For Posada his .370/.468 = 1.119. From 98-03, the AL values are been 1.69, 1.63, 1.61, 1.74, 1.76, and 1.70. The 98-03 NL values 1.80, 1.68, 1.68, 1.76, 1.80, and 1.74.

I know adj-OPS+ is appreciated for the convenience of the scale, but since it provides inaccurate assessments of value, it's nice to know that OPS * (1.6 - 1.8) + SLG not only is more accurate but basically gives you a league (though not park) adjusted measure.

May 20, 2003 - MGL

Good work Tango!

I think Darren might be right about Beane's "3 times the value" comment (see his post). I also think that while Beane and company are parsecs ahead of their competition, that there is a large gap between his and Depodesta's knowledge of and efficient and correct use of sabermetric principles and that of many sabers on Primer, BP, and Fanhome (and wherever else they might lurk). IMO, the A's would be far better off hiring someone like James, Voros, Tango, etc., than trying to do sabermetric analysis themselves. It is kind of like when Brenley asked Matt Williams to sac bunt last night. I'm sure he is capable of doing so, but...

Also, I think that teams will quickly start catching up to the "player acquisition" principles being espoused and used by Beane and company, especially since they are now being publicized in a maninstream book (nice job Beane). I don't think it will be very long before it wil be very difficult to pick up high production (high OPS or whatever) but not traditionally highly regarded players cheaply. The next frontier for picking up undervalued players will be and should be DEFENSE, and other Super-lwts components. It will be a long time before teams start using things like UZR to evaluate the overall impact a player will have on their team. Right now, one of the best ways to pick up valuable players cheaply and "sell" players expensively who are not all that valuable (buy low and sell high), is to look for large gaps bewtween a player's traditional defensive rating (scouting, reputation, etc.) and their UZR (or other good defensive metric - are there any others?) rating. (I think the days are numbered as to being able to do that for offense.) This should provide for plenty of value for a while I think. In the book, Beane implies that they use some kind of defensive rating that sounds suspiciously like UZR, via some computer company or something. Anyone know more about that?

May 20, 2003 - David Smyth

One thing they didn't really explain in Moneyball is how DePodesta came up with the 3 to 1 ratio foe OBA and SLG. I think I figured it out...

The max OBA is 1.000, and the max SLG is 4.000. At either of these levels, the team will score an "infinite" number of runs, so the max OBA and max SLG are equivalent on the team level. The equalized ideal ratio is therefore 4.000/1.000, which comes to 4.0 (duh). The actual ratio of SLG to OBA in the mid-late 90s AL was about 1.3 to 1. So if you divide the 4.000 by the 1.300, you get 3.08, which they rounded off to 3.00. What this means is that there is 3 times as much proportional movement needed to get from the avg SLG to the max SLG, as there is to get from the avg OBA to the max OBA.

That's an interesting insight, but what does it have to do with comparing the value of OBA and SLG for real-life ballplayers? The answer is...(drum roll, please)...nothing!! DePodesta screwed up, and so now Oakland is overvaluing OBA to the same degree as OPS undervalues it. This does not include the differences in the "price" of OBA and SLG, and I guess any pitch count benefit. My calculations showed a value of 1.7*SLG. Pumping that up a bit due to these extra considerations, and the Oaks should be weighting OBA as 2*SLG, not 3*

May 20, 2003 - David Smyth

Obviously, I meant that Oakland should use 2*OBA + SLG (not that OBA=2*SLG).

May 21, 2003 - McCoy

Question

Of the two, OBP and SLG, which is more important? By weighting OBP it appears that OBP is more important, but other studies show that SLG correlates better to run scoring. Am I correct about that?

May 21, 2003 - Sean Smith

What values for 1b, 2b, 3b, etc. are you using to calculate baseruns in the first chart?

May 21, 2003 - tangotiger (www) (e-mail)

See link for the values I used. For the categories I didn't use, I set them to "zero". It's not too important for what I am trying to do though.

As for the other question, you are asking if you can only know one thing, OBA or SLG, which one correlates to run scoring the best? I seem to remember Dan Werr doing a correlation study a month or 2 ago that showed the r to be pretty even between the two. That doesn't mean they are "equally important", especially if you have both.

As well, the coefficient itself (1.56, 1.64 or whatever) doesn't specificy the level of importance. If you made lilSLG = 1/4S + 2/4D + 3/4T + HR, all divided by AB, what do you think would happen? The best-fit would be 1.64*OBA + 4*lilSLG. That doesn't make lilSLG twice as important as OBA, now, does it?

May 21, 2003 - David Smyth

It's interesting that the best fit for the team was 1.56*, and for the individual added to the team it was higher, at 1.64*. I would have expected the opposite.

May 21, 2003 - Reality check

I also think that while Beane and company are parsecs ahead of their competition, that there is a large gap between his and Depodesta's knowledge of and efficient and correct use of sabermetric principles and that of many sabers on Primer, BP, and Fanhome (and wherever else they might lurk). IMO, the A's would be far better off hiring someone like James, Voros, Tango, etc., than trying to do sabermetric analysis themselves.

You're an idiot.

DePodesta is a Harvard graduate who did some pretty sophisticated econometric work to receive his degree. He's spent the past five years of his life dedicated, at least in large part, to performing sabermetric analysis on behalf of the A's. There's no reason to think that he's any *more* qualified to do sabermetric work than Voros or Tango etc, but there's also no reason to think that he's any less so, and if you're using "publication" as the standard of measurement, well, DePodesta has some very obvious reasons for keeping his work proprietary.

Most importantly, DePodesta has shown that he can get along with people, and influence the decision making process of a real, viable organization. Once again, I'm not saying that Voros or Tango couldn't, but it's far from a given. There are lots of very bright people - not just statheads - who don't have the interpersonal skills to get very far in business. See also Gimbel, Mike.

May 21, 2003 - tangotiger (www) (e-mail)

I agree that it would be a rush to judgement to make any conclusions without having all the information.

While you can conclude that using 3*OBA+SLG is a poor way to evaluate current run production, it's not so clear if you want to use that equation to try to evaluate future run production (or for other secondary reasons). And you certainly can't indict someone or some organization overall. Sample size! You need alot more evidence.

I also agree that being able to work with a group of people, respecting their views, regardless of what it is, as long as they respect your views as well, is very important. Respect, courtesy, professionalism. Isn't that the police motto?

However, I'll note that in the ESPN chat, Bill James said: Baldelli's a lot of fun. In my office we were making fun of some scout who compared him to Joe DiMaggio, but when you see him play you realize what people are reacting to. Of course, he doesn't have DiMaggio's entire package, but he does have more than half of it. I kinda didn't like the first part, which left me with the impression that the stat-heads and the scouts clash behind each other's backs. But, this was a throwaway sentence, so who knows what James meant.

Finally, as for anyone's ability to deal with people, I'm not sure that you can necessarily say that DePodesta is good or bad, nor could you say that with me, or Voros, or anyone else, unless you deal with these people on different issues in different settings (or you have some second-hand knowledge... definitely not third-hand or worse). I don't think that an executive is a better people-person, or can deal with people, than a non-exec.

I agree that arrogance is a turn-off to most people, and that's something that a speaker should be conscious of. Mike Gimbel, who I've had occasion to e-mail from time-to-time, seems like a pleasant enough fellow. But I've heard from many many people that he is insufferable. That by itself, truth or perception, will keep Gimbel out of MLB, in my view.

May 21, 2003 - Reality check

Finally, as for anyone's ability to deal with people, I'm not sure that you can necessarily say that DePodesta is good or bad, nor could you say that with me, or Voros, or anyone else, unless you deal with these people on different issues in different settings (or you have some second-hand knowledge... definitely not third-hand or worse). I don't think that an executive is a better people-person, or can deal with people, than a non-exec.

Well, working in the corporate sector, I can confirm that. Although it's a minority, there are plenty of a**holes who do plenty well for themselves in business, and plenty of really nice people who fall short.

But I'm not talking about being friendly, and I'm not *exactly* talking about managerial skills. Rather, I'm talking about the ability to be persuasive when dealing with people who have dissenting or at least ambivalent viewpoints, which at the very least involves some combination of:

1. The ability to listen to people (or "read people", if you will), in order to understand what their priorities are and why they believe what they do. 2. Rhetorical ability ... which in a non-virtual setting becomes a subset of charisma. 3. Prioritizing one's research and aligning it with the goals of the organization ... i.e. knowing which battles to fight.

Now, there are a lot of different ways to go about accomplishing those things - Beane has a certain personalitiy, DePodesta's is virtually the opposite, and they're both highly effective. But to really do all of those things well is fairly unusual, and I would guess that among the pool of the 15 or 50 or 500 or whatever leading analysts, there's a lot more differentiation in terms of interpersonal ability than technical ability.

May 21, 2003 - tangotiger (www) (e-mail)

I agree with your comment on the corporate world (as I've been here for...geez, almost 13 years... my "corporate world" anniversary will be in 1 month).

Rather, I'm talking about the ability to be persuasive when dealing with people who have dissenting or at least ambivalent viewpoints, which at the very least involves some combination of:

That sentence alone is interesting to read!

But to really do all of those things well is fairly unusual, and I would guess that among the pool of the 15 or 50 or 500 or whatever leading analysts, there's a lot more differentiation in terms of interpersonal ability than technical ability.

That's an interesting thought too. I'm not sure if there is more differentiation in one or the other, or how you would qualify/quantify all that. And even if the differentiation is more in one category, the impact of that differentiation might not be as much as the other category.

Did it just feel like we had an OBA v SLG discussion? (More differentiation in SLG, but more impact with OBA differences.)

As with everything, there's degrees of impact to everything, and it's rather pointless to label them black/white (not that that's what I think anyone is doing here). Even if you have a terribly insufferable analyst, his work might be of such quality that it tips the scales towards good. Even if you would be able to classify DePodesta as a mediocre sabermetrician (and I'm not doing that), the rest of his skills might be so strong, that he can make an impact with his research, while others might not (even with better "stuff").

The fact that a successful organization has him employed, and he is highly regarded by other successful people, even though his experience is not as vast as other baseball execs, must show that his total package is something to respect highly. He's a mover and a shaker, and he gets things moving and shaking in generally the right direction.

May 21, 2003 - David Smyth

I think some people are being to kind to DePodesta, in some sort of PC fear. This is a saber site and discussion. How easy Paul D is to get along with, and what his college was and degree was in, has next to nothing to do with anything (how do you like *that* sentence, Tango). Let me quote the relevant passage from moneyball about the 3 to 1 ratio: "He proceeded to tinker with his own version of Bill James' runs created formula. When he was finished, he had a model for predicting run production that was more accurate than any he knew of. In his model an extra point of OBA was worth 3 times an extra point of SLG."

Now, that is a definitive statement. Whether it's from DePodesta, James, Gimbel, Einstein, or me, it is either true or false. All the evidence suggests that it is false, and that even if you give a bonus for taking pitches and the lower salary cost of walks, you'll be hard-pressed to even get the ratio up to 2 to 1.

It's true that we don't have a detailed explanation from DePodesta, and that Lewis might have gotten the details wrong. In an earlier post, I gave a theory as to how Paul D got his result, and that it was bogus. Maybe my theory is wrong (I doubt it), but the bottom line is that 3 to 1 is way off, based on what info we have to evaluate what DePodesta meant. Let us not kiss his ass because he has a "job". There is no doubt that the people named such as Voros and Tango could do DePodesta's essential job, and probably better.

May 21, 2003 - Reality check

How easy Paul D is to get along with, and what his college was and degree was in, has next to nothing to do with anything.

It has to do with the following comment by MGL,

"IMO, the A's would be far better off hiring someone like James, Voros, Tango, etc., than trying to do sabermetric analysis themselves"

which neglects the fact that the technical side of DePodesta's job is just one part of his responsibilities. Understanding the priorities of his organization, being able to pitch his ideas successfully; these things are also very important. The tragic part of the Mike Gimbel experiment, and the experiences that Craig Wright, Eddie Epstein etc. had before him, is that they were lone voices in the dark and had no discernable influence on the direction of their organizations. Yes, DePodesta's in the right place at the right time. But he's also the right man for the job, as attested to by the fact that he's received offers to be a GM, a position in which his responsibilities would extend far beyond being a Stat Boy.

Let me quote the relevant passage from moneyball about the 3 to 1 ratio: "He proceeded to tinker with his own version of Bill James' runs created formula. When he was finished, he had a model for predicting run production that was more accurate than any he knew of. In his model an extra point of OBA was worth 3 times an extra point of SLG."

I happen to believe - and no, I can't offer proof - that Lewis in some way or another didn't convey DePodesta's findings properly. The A's certainly haven't *behaved* like an organization that believes that OBP is three times more important than slugging average.

May 21, 2003 - tangotiger

I think that you should give the benefit of the doubt when you can. I've heard nothing but good (in fact great) things about DePodesta, so, without him actually saying anything, I give him that benefit.

Now, I can interpret the 3 thing as being "you know, I've got this great formula, and you know what, this correlates highly to 3*OBA+SLG. I don't use 3OPS, I have my own, but as it turns out, it's close to 3OPS. BaseRuns, which I don't use, is close to 1.6OPS. I'm sure Tango/David don't use 1.6OPS, but their equation is close to that".

I don't think that explanation is unreasonable, is it?

May 22, 2003 - terry (e-mail)

It's not the A's that are the best example. It's the M's. The A's are a team in constant flux, and the fact that they are still so good is a tribute to Billy Beane. However, he is not a one trick pony. He is smart, period. He adjusts to his available talent. He lost Giambi (hell, both Giambi's), and he is living and dying with unproven players and his big three pitchers. The A's are not particularly patient any more.

The M's are the poster boys. Olerud, Boone, Cirillo, Cameron, both catchers, Edgar, even Ichiro who almost never hits the ball foul (see Tony Gwynn, who never walked and never struck out because he never hit foul balls), Wynn, only Guillen is a hacker. They may not walk 700 times this year because pitchers just won't nibble when there are men on base. However, they work pitchers so hard and put so many men on base that those fatter pitches add up to a lot more runs than they deserve to score.

They are 29-15 right now, and it is mostly because of their ensemble offense. A 40 year old DH that couldn't beat my mother in a foot race is hitting .333 with 10 homers, and the rest of the cast feeds off of him. He might go down, but the rest of them will continue the quest. No atbat is wasted, and no inning is given up. The M's score more with 2 out and nothing going than anyone. They just don't give away any atbats. 3 years of this has sunk in, and they just don't give up.

I know that this sounds like some announcer's crap. I don't know how to say it right. I have been watching this game for over 30 years. I have never seen a team like this. They just don't waste atbats. Maybe the 1998 Yankees.

Terry

May 22, 2003 - David Smyth

Another way to test for the best OBA factor is to use the "plus 1" method, and find out which factor produces the most correct ratio between avg event values. The critical one in this case is the HR/BB. Doing that, and keeping in mind that OBA includes the IBBs as regular walks, I find the best factor to be 1.7. Tango got 1.56 and 1.64, I think, but he only included a handful of possible teams. Until someone shows otherwise, I would use 1.7*OBA + SLG.

As far as increasing the OBA weight because of the "wearing down the pitcher" theory, it should be kept in mind that another way to make the pitcher exert himself more is by the threat of power. As you all have read, the lack of worrying about a batter's power is frequently cited as a primary reason why dead-ball pitchers were able to pace themselves and save their best stuff for crucial moments. I don't see why the same argument shouldn't apply to power hitters and non-power hitters in today's game. So that would suggest an increase in the weight of SLG, canceling out the added OBA weight for taking pitches. All in all, just use the 1.7 if you want to use WOPS at all (Weighted OPS).

May 22, 2003 - tangotiger (www) (e-mail)

David, I agree that the 1.64 value is a little suspect since it is based only on those 6 players that I happen to construct. I mentioned that 1.50 to 2.00 would be the correct value, if you were to look for it.

I've used the plus-1 method in the past, and I find I can minimize the runs error by using 1.83 as the coefficient for OBA. That is, 1.83*OBA+SLG. I think that as long as you use something between 1.5 and 2.0, you'll be ok, or at least better than not. I suppose if you really wanted to find the best-fit via the "plus 1" method, you'd look at 200 regular hitters, and figure it out that way.

(For the uninitiated, the "plus 1" method was described in the "Runs Really Created" series last year. Check out the archives.)

May 22, 2003 - David Smyth

When you got the 1.83, Tango, did you include an adjustment for the IBBs? Without it, I got 1.8. With it, I got 1.7

May 22, 2003 - tangotiger (www) (e-mail)

Interesting. You know, I'm pretty sure I never include the IBB, but it was several months ago when I did that 1.8 thing. Interesting results though. I suppose we should compare it to the full-blown BsR version in that case.

May 30, 2003 - Kurt

-----Quote----- Let me quote the relevant passage from moneyball about the 3 to 1 ratio: "He proceeded to tinker with his own version of Bill James' runs created formula. When he was finished, he had a model for predicting run production that was more accurate than any he knew of. In his model an extra point of OBA was worth 3 times an extra point of SLG."

I happen to believe - and no, I can't offer proof - that Lewis in some way or another didn't convey DePodesta's findings properly. The A's certainly haven't *behaved* like an organization that believes that OBP is three times more important than slugging average. ---End Quote---

From the verbiage of the Moneyball quote it seems quite obvious that they are, in some way, obfuscating the actual formula. If they are hiding it, we have no reason to believe that any of that section is usable. But looking at the exact words used... an EXTRA point of OBA, an EXTRA point of SLG... it sounds like they are using basline values.

3*(OBA-x)+(SLG-y)

Knowing what those baseline values are would seem to be the trick. .200 and .250? .300 and .100? .167 and .300? Where you set those greatly varies the effect of trippling the OBA.

June 2, 2003 - tangotiger

3*(OBA-x)+(SLG-y)

This works out to 3OBA+SLG-(3x+y) which works out to 3OBA+SLG-k

Therefore, it is irrelevant what "k", "x", or "y" is. Whatever numbers you choose won't affect the ranking of the players, or the degree of their rankings, relative to each other, than if you simply used 3OBA+SLG

June 15, 2003 - Dominic Rivers (www) (e-mail)

I have published a study of the shortcomings of OPS at my site: http://mysite.verizon.net/vze1yy5u/

I was rather surprised to read that your study determined the exact same multiplier - 1.64 - as I found for the 2001-2002 time period. I did find that the late 1990's multiplier was close to 3, as DePodesta suggests in Moneyball.

December 25, 2003 - Julio Rojas

Hi guys. I´m just doing a little analysis on the relation between OBA and SLG. I took a nonparametrical approach to estimate the density of the ratio between SLG and OBA. Because we would like to have some sort of coefficient to use as a multiplier for OBA, I tried using SLG/OBA (this way xOBA=SLG). I took the 165 players that qualified for this year "batting title", calculated this ratio and, using kernel smoothing, estimated the probability density. One should expect to have a unimodal density (almost a normal density, one value being notoriously bigger than the rest), but I found a bimodal density. The modes were around 1.2 and 1.45, which are far from the estimates you have, and from the value you expect for a bigger sample. I don't know how well this result fix the problem, but it generates a bigger problem, because the bimodal density only says there are two very definite groups which should be treated and analysed with different coefficients. Well, my two cents. Sorry for my poor english.