Jump to content


Yo! You're not logged in. Why am I seeing this ad?

Photo

Manny's defense, revisited


This topic has been archived. This means that you cannot reply to this topic.
89 replies to this topic

#51 OCD SS


  • SoSH Member


  • 6846 posts

Posted 20 March 2007 - 06:32 AM

As I am just 1 voice I will defer you to Tangos 2006 Scouting Report for who others rank below Manny.

I do not know exactly when Coco injured his shoulder, but his throwing was crap all year from what I could see.


To me, the value of Tango's fan polls are how they give a snapshot of conventional wisdom that's seperate from what the media just tells us. The problem with them I have is that they are a completely subjective accounting of defense; they don't measure defensive skill, they measure the opinion of it.

And to be clear, I'm not trying to defend Coco's throwing last year; I agree with you that it was crap. I was disputting your use of Coco's throwing in '06 as an illustration as to why Dewan's total defensive analysis of Manny in '03-'05 was not valid.

Edit: Val, I look forward to seeing your work on the subject. When looking at defensive systems that give a measure in runs I tend to treat it (rightly or wrongly) as a Pitching Independant Fielding Statistic; and use the relative comparison (between players) pretty closely, but treat the runs as much more variable. It would be nice to have a different measure that would serve to cross check those numbers.

It's also a bit frustrating, as ML teams like the Sox purportedly have it mostly figured out, but the fans are lagging in this information for lack of rescources.

Edited by OCD SS, 20 March 2007 - 06:42 AM.


#52 PedroKsBambino


  • SoSH Member


  • 14187 posts

Posted 20 March 2007 - 07:05 AM

The value or cost of an out will obviously change depending on the base out state. With 2 outs and no one on base, the value of the out is 0.116. With bases loaded and 2 out, the value of the out is 0.798 runs. The average cost of the out looking at the cumulative of all the various base outs states in all games played from 1999-2002 was 0.3 runs, so this is what is used.


I thought that UZR and other PBP metrics used a static run value, and it was the .8 value rather than adjusting by base/out state. Whether that overstates or understates the impact for any given player I don't know, but it overall likely works out across all players, I guess.

A simple question to ask, though, is why the value for a play not made in the PBP systems has a run value of .8 but an error in lwts has a run value of .49. While I don't pretend to have done the math, I find the lower run value a lot closer to observed reality on the issue...which doesn't make it right, of course.

#53 paulftodd


  • 133% banned


  • 1470 posts

Posted 20 March 2007 - 10:05 AM

I thought that UZR and other PBP metrics used a static run value, and it was the .8 value rather than adjusting by base/out state. Whether that overstates or understates the impact for any given player I don't know, but it overall likely works out across all players, I guess.

A simple question to ask, though, is why the value for a play not made in the PBP systems has a run value of .8 but an error in lwts has a run value of .49. While I don't pretend to have done the math, I find the lower run value a lot closer to observed reality on the issue...which doesn't make it right, of course.


Yes, they use the static value. I was just trying to answer VAL on why the out was valued at 0.3

Good question on error vs play not made. I guess because an error can be limited to a base being advanced, or bases, without costing you an out, and costing the defensive team 0.5 runs. When the error costs the team an out as well, add 0.3, hence 0.8 for error (additional ROB or baserunner advance) plus lost out. That said, I will defer to the experts.

#54 paulftodd


  • 133% banned


  • 1470 posts

Posted 20 March 2007 - 10:17 AM

And to be clear, I'm not trying to defend Coco's throwing last year; I agree with you that it was crap. I was disputting your use of Coco's throwing in '06 as an illustration as to why Dewan's total defensive analysis of Manny in '03-'05 was not valid.


Maybe I was not clear. I was saying Dewan in 2005 ranked Coco above Manny in thowing in 2005. Both were LFers in 2005. I assumed Coco 2006 throwing from CF was similar to his arm in 2005 in LF (not in results but quality), which may or may not have been a valid assumption if his shoulder was a valid excuse. If the assumption was valid, Dewan has to be wrong (Manny had 13 kills in 2005, Coco 1), at least IMHO.

#55 Tangotiger

  • 445 posts

Posted 20 March 2007 - 10:47 AM

Tango, I've read this example before and it makes sense at a certain level, but there are a few problems with it. Let's focus on my main one: why do you assign the .30 out value? I think I understand your using that value for your probabilistic work (using league average as the mid point), but in the cumulative model - which your example surely falls into, shouldn't the outval be .11 or thereabouts?


If you read the illustration, I don't use the .30 out value at all. The only thing I did was model a baseball game, and described how many runs a team will score with 41.6 batters faced, and what happens if a play that would have been an out was then allowed to become a hit (i.e., now 43.2 batters faced). And then figured how many runs scored.

And that's it. The answer was an extra .8 runs. No .30 or .50 talk whatsoever in order to get the .8 figure at all.

Now, to *explain* how one could derive the .8 figure without creating the illustration I did, then I tell you about the .3 and .5. But, that part is completely irrelevant to the illustration.

All I can tell you is for you to create your own model using whatever run modeler you want, and let you come up with the answer. If it's not .8, then we can talk.

#56 xjack


  • Futbol Crazed


  • 5162 posts

Posted 20 March 2007 - 12:10 PM

Joe Arthur speculates (and I agree with him) that this may be evidence that Manny was indeed injured.

How can that be? Everything I've read in the papers says he just quit. ;)

Seriously though, those pre-ASB/post-ASB zone rating splits are really interesting.

#57 elias

  • 253 posts

Posted 20 March 2007 - 12:34 PM

Here's my question about Manny and anything involving zone rating. He plays half his games in a starting position in a much different zone than any other LF'er. He starts each play in a signficatnyl different part of the field than most left fielders. That means the kind of balls he sees are also different than others. Think of how different you feel if you ever had to change seats in a classroom, or offices at work etc. Its a different position almost because of that wall. Very rarely, for 81 games a year, he plays a much different position than anyone else. Its somewhat close to the idea of playing the outfield version of SS between LF and whatever the extreme of what Manny has to do would be classified as.

To put Manny in the same class as those other mediocre guys is ludicrous. I find it impossibe to believe that Manny's defense resulted in that many runs for the opposition.

He is an MVP candidate year in and year out, and the attacks on him are getting pretty old and lame. He's a legendary player still dominating the sport. I don't get why people dont just shut up and appreciate that he's on a good team and involved in the game right now. It's too bad they can't, because it sure is amazing to watch.

#58 Tangotiger

  • 445 posts

Posted 20 March 2007 - 01:09 PM

Manny is also a poor clutch hitter:

http://www.fangraphs...amp;position=OF

From 2002-2006, he's +24.6 wins, when not considering the men on base and leverage.

If you include the men on base, he's +26.4 wins, which is pretty good. He brings it on, with men on base. However, when you factor in the closeness of the game, he becomes +19.9 wins.

What this means is that:
a - he's still a monster (that +19.9 wins, even considering his chokiness, likely keeps him as one of the top 5 hitters in MLB in that time period)
b - he's not quite the monster his context-free stats make him out to be

You can also check out his career here:
http://www.baseball-...gi?n1=ramirma02

#59 xjack


  • Futbol Crazed


  • 5162 posts

Posted 20 March 2007 - 01:31 PM

Manny is also a poor clutch hitter

Well, he's a very disciplined hitter, and I think that's the key to understanding the clutch hitting numbers.

He walked 25% of the time last year in close and late situations, to the tune of a .459 OBP. With runners in scoring position and two out, he walked 30% of the time and his OBP was .545.

#60 Pumpsie


  • The Kilimanjaro of bullshit


  • 10608 posts

Posted 20 March 2007 - 03:16 PM

Well, he's a very disciplined hitter, and I think that's the key to understanding the clutch hitting numbers.

He walked 25% of the time last year in close and late situations, to the tune of a .459 OBP. With runners in scoring position and two out, he walked 30% of the time and his OBP was .545.


Bingo, xjack.

Bobby Abreu has that same tendency and so did Ted Williams, I believe. They're just not going to chase pitches out of the strike zone. To say that that translates to "Manny is not a clutch hitter" is just an out and out lie. Pitchers pitch to Manny as if he's a nuclear device on a hair-trigger. They'd much rather pitch to Trot Nixon or whoever else the Sox have had batting fifth, and who can blame them? If Manny chased, his production would slip further.

David Ortiz does not face this problem, of course. If Papi had to bat fourth with a Trot Nixon hitting behind him, his production would slide even more because he's not as disciplined as Manny is and can get frustrated if they're not throwing him anything at all to hit. That's why Tito has Papi third and Manny fourth.

This seems to be another case of viewing baseball through the lens of statistics without actually watching the games themselves.

Edited by Pumpsie, 20 March 2007 - 03:18 PM.


#61 Tangotiger

  • 445 posts

Posted 20 March 2007 - 04:15 PM

Bingo, xjack.

Bobby Abreu has that same tendency and so did Ted Williams, I believe. They're just not going to chase pitches out of the strike zone. To say that that translates to "Manny is not a clutch hitter" is just an out and out lie. Pitchers pitch to Manny as if he's a nuclear device on a hair-trigger. They'd much rather pitch to Trot Nixon or whoever else the Sox have had batting fifth, and who can blame them? If Manny chased, his production would slip further.

David Ortiz does not face this problem, of course. If Papi had to bat fourth with a Trot Nixon hitting behind him, his production would slide even more because he's not as disciplined as Manny is and can get frustrated if they're not throwing him anything at all to hit. That's why Tito has Papi third and Manny fourth.

This seems to be another case of viewing baseball through the lens of statistics without actually watching the games themselves.


It is not an out-and-out lie. And I wasn't referring to only the handful of PA in 2006. Just look at his career record. In late&close (1100 PA), his SLG is .497, while his career SLG is .600.

If you check out other heavies (Bonds, Thome, Junior, ARod), they lose 30-50 points from their overall SLG. Manny loses over 100. As for "pitching around", here's an excerpt from The Book, by Andy:

http://www.hardballt...around-batters/

Bobby Abreu is FANTASTIC with men on base, and not so good in late&close. Overall, he's still a net positive clutch hitter, enough that, while he's an overall worse hitter than Manny, he's better than Manny when you consider men on base and closeness of game.

#62 wade boggs chicken dinner


  • SoSH Member


  • 6710 posts

Posted 20 March 2007 - 04:42 PM

I posted a link to what Joe Arthur had found in the "What's Manny Really Worth" thread; the link appears to be broken. But assuming I was lucid at that point, here's how I summarized it:
I'll see if I can't find the original thread again.

Apparently Joe Arthur lurks here but is not a member (SHORT ASIDE: SHOULDN'T HE BE GIVEN A MEMBERSHIP HERE!) and he sent me a link that works: (you can change the page by adjusting the start & stop numbers)

I think it's already been linked, but in case it hasn't, here's a short piece that he did on CF and Coco.

Thanks Joe.

#63 Worst Trade Evah


  • SoSH Member


  • 10834 posts

Posted 20 March 2007 - 05:07 PM

It is not an out-and-out lie. And I wasn't referring to only the handful of PA in 2006. Just look at his career record. In late&close (1100 PA), his SLG is .497, while his career SLG is .600.

If you check out other heavies (Bonds, Thome, Junior, ARod), they lose 30-50 points from their overall SLG. Manny loses over 100. As for "pitching around", here's an excerpt from The Book, by Andy:


Well, late and close is kind of a data slice. With another data slice, say, 2 outs RISP (where Manny has just about the same number of appearances as late and close), Manny has a 1.073 OPS. Even late&close, his OBP is .416, putting his late&close OPS over 900. It's kind of hard to complain about that, especially in light of his numbers with 2 outs RISP, in Tie Games, etc.

I guess he's losing around a little ground in some "clutchiness" stat, but on the other hand, he destroyed the Yankees last year, and I'm not really worried about Manny up in key situations with the bat in his hand. You don't actually think Manny has any kind of a "clutch" problem, do you? As a retrospective assessment of Manny's results, yes, he's been down a little in some situations. Going forward, I don't think it means anything at all.

I like Abreu, but would you really rather have him at the plate in a big spot than Manny? I wouldn't.

Edited by Worst Trade Evah, 20 March 2007 - 05:08 PM.


#64 elias

  • 253 posts

Posted 20 March 2007 - 05:10 PM

I like Abreu, but would you really rather have him at the plate in a big spot than Manny? I wouldn't.


This says it all to me. Honestly, it is a no brainer for any manager in the league.

#65 dkappelman

  • 2 posts

Posted 20 March 2007 - 06:13 PM

I like Abreu, but would you really rather have him at the plate in a big spot than Manny? I wouldn't.


It's really a difference in what you consider "Clutch". Manny compared to most players did quite well in clutch situations. But compared to himself in non clutch situations, he did noticeably worse. It's probably fair to say he didn't "elevate his game". Some players, like Abreu and of course Oritz, did.

#66 Pumpsie


  • The Kilimanjaro of bullshit


  • 10608 posts

Posted 20 March 2007 - 06:17 PM

Some players, like Abreu and of course Oritz, did.


Don't tell this to Philly fans. They were all over Abreu for not coming up big in the clutch for the Phillies. This was a "known fact" in Philadelphia. Of course, they ARE Philly fans.

#67 mangotree101

  • 790 posts

Posted 20 March 2007 - 06:59 PM

It is not an out-and-out lie. And I wasn't referring to only the handful of PA in 2006. Just look at his career record. In late&close (1100 PA), his SLG is .497, while his career SLG is .600.

If you check out other heavies (Bonds, Thome, Junior, ARod), they lose 30-50 points from their overall SLG. Manny loses over 100. As for "pitching around", here's an excerpt from The Book, by Andy:

http://www.hardballt...around-batters/

Bobby Abreu is FANTASTIC with men on base, and not so good in late&close. Overall, he's still a net positive clutch hitter, enough that, while he's an overall worse hitter than Manny, he's better than Manny when you consider men on base and closeness of game.

I don't know. I like and trust stats as much as the next guy, but if the numbers tell me A-Rod is more clutch than Manny, I stop believing the numbers...

I know it doesn't mean much, if anything (SSS and all that), but in 2006 Manny's L&C SLG was .685, and his OBP was .545, as compared to his overall SLG/OBP of .619 and .439. He's turned things around!

#68 PedroKsBambino


  • SoSH Member


  • 14187 posts

Posted 20 March 2007 - 07:16 PM

Manny has had some terrific clutch hits, and I certainly don't worry about him being up in the clutch. That said, I think he's getting some halo effect here, too.

David Ortiz is an amazing clutch hitter. Perhaps the best in modern baseball. But Manny Ramirez, to my eyes, isn't anywhere near that. I am not stunned that he's below other top sluggers in the clutch, even. The Sox get a lot of wins because the middle of their order produces in the clutch, and Manny does well enough...but I think perhaps people are overstating how effective he's been there and perhaps some of that is his lineup neighbor.

Just my take; it's obviously a subjective one.

#69 Vermonter At Large


  • SoxFan


  • 3061 posts

Posted 20 March 2007 - 07:47 PM

If you read the illustration, I don't use the .30 out value at all. The only thing I did was model a baseball game, and described how many runs a team will score with 41.6 batters faced, and what happens if a play that would have been an out was then allowed to become a hit (i.e., now 43.2 batters faced). And then figured how many runs scored.

And that's it. The answer was an extra .8 runs. No .30 or .50 talk whatsoever in order to get the .8 figure at all.

Now, to *explain* how one could derive the .8 figure without creating the illustration I did, then I tell you about the .3 and .5. But, that part is completely irrelevant to the illustration.

All I can tell you is for you to create your own model using whatever run modeler you want, and let you come up with the answer. If it's not .8, then we can talk.

Tango,
I love you in a manly way, but that example is a little confusing on many different levels.

First of all, one hit more per team per game is a huge, huge margin, way beyond realistic expectations. In 2006, for example, the best hitting team in the A.L. averaged 8.10 NHRHs per game, while the worst averaged 7.71 NHRHs per game, for a difference of .39. Most teams operate within .20 NHRHs per game of each other. So using the extreme 8.1 and 7.7 NHRH split between your two teams and a league average BB/G value of 3.5, the lwts value of the batting portion of your two teams is actually 5.21 and 5.01, or a difference of only .20 runs. Subtracting for the 27 and 26 outs respectively, we come to values of 2.51 RPG for Ozzie's team, and 2.41 runs for the low team. Applying the same adjustment for 26 to 27 outs that you used, we get a run total for the low team of 2.12, for an overall difference of .39 runs, which is exactly half of the .78 run value you came up with. And this is the extreme example. In reality, the actual difference is probably quite a bit lower than .39 runs between average teams.

If those run values seem low, remember that we are not including HRs or other multipliers which add up to run totals.

Another logic test would be to look at how much an average PA is worth. The 2006 American League averaged .122 batting runs created per PA. If I multiply that times 41.6 PAs, I get 5.075 BRC, and by 43.2 I get 5.270 BRC for a difference of .195 runs. If I substitute actual runs (1.30 per PA) instead of BRC, I get 5.408 in 41.6 PAs and 5.616 in 43.2 PAs, for a difference of .208 runs per game. This is clearly about 25% of what you are estimating in your example, which is way too much for any possible excursion from the league average.

Having said all of that, I am not sure what your example actually has to do with defensive metrics. That it doesn't fit the .50 + .30 values doesn't mean those values aren't correct - just that the example doesn't prove that they are.

Since we are talking SS's, let's look at some other numbers.

First of all, we actually know how much run value a ball hit into a SS zone and not played is: about .47 runs. This is because there is almost no scenario in which a misfielded ball that a shortstop might be expected to field (or 2B for that matter) can be expected to be anything but a single (.47 runs). Actually, it's probably a few hundredths of a run lower than that, since there are probably a higher number of singles in a SS zone that don't advance runners at average expected rates, so they have less run-producing value.

Another interesting thing about SSs is that their fielding errors probably have lower run value than corner infielder and outfielders. Again, a misplayed ball is probably not going to exceed .47 runs, and throwing errors, which must surely comprise half or more of total SS errors, tend to be single base errors, so I would estimate that an average SS error is probably less than .40 runs. It occurs to me that breaking down errors like this could ultimately lead to a nice study reevaluating the defensive spectrum, since although middle of the defense players handle far more chances than the corner players do, the damage done through misplays is probably quite a bit less. There might be some interesting findings in such a study ...

Anyway I think we can use the .47 or even a .45 value of a playable hit as a good starting point for a SS. Next we have to decide what to do with outs. I would suggest using the .11 value in cumulative studies where we want to examine exactly how many runs a fielder yielded, or the .30 value if we are looking to produce something like a league average. I'll play with the cumulative model since that's my druther, and suggest a starting value of .56 runs for a SS fielding play.

Anyway, that's a start point. I'm running out of time and energy at this point, but we can surely build on this discussion from here.

Edited by Vermonter At Large, 20 March 2007 - 07:50 PM.


#70 paulftodd


  • 133% banned


  • 1470 posts

Posted 20 March 2007 - 09:15 PM

Well, late and close is kind of a data slice. With another data slice, say, 2 outs RISP (where Manny has just about the same number of appearances as late and close), Manny has a 1.073 OPS. Even late&close, his OBP is .416, putting his late&close OPS over 900. It's kind of hard to complain about that, especially in light of his numbers with 2 outs RISP, in Tie Games, etc.

I guess he's losing around a little ground in some "clutchiness" stat, but on the other hand, he destroyed the Yankees last year, and I'm not really worried about Manny up in key situations with the bat in his hand. You don't actually think Manny has any kind of a "clutch" problem, do you? As a retrospective assessment of Manny's results, yes, he's been down a little in some situations. Going forward, I don't think it means anything at all.

I like Abreu, but would you really rather have him at the plate in a big spot than Manny? I wouldn't.


I do not believe that Manny is "not" clutch, I think Manny is Manny regardless of the situation but in the clutches sitiations, guys like Manny see the other teams best RP and are pitched around more Papi on the other hand defines clutch. I had put this on another thread, slightly off topic though but relevent to the dicussion in recent posts.

Fan Graphs Clutch = WPA-(OPSWins*pLI)


No question Manny is a great hitter but based on Fan Graphs Clutch data Manny be an even worse clutch hitter than A-Rod. Five year Clutch totals for some selected players

Papi 3.57
JD Drew 2.56
Bernie Williams 2.22
Andrew Jones 2.04
Jermaine Dye 1.92
Tek 0.63
Garry Sheffield 0.31
M. Loretta 0.23
Kevin Millar 0.09
Jeter -0.08
Coco Crisp -0.19
Nomar -0.67
Carlos Beltran -1.22
Mark Bellhorn -1.42
Vernon Wells -1.49
Damon -1.52
Lugo -1.56
Pujols -1.87
Lowell -1.87
Alex Gonzalez -2.06
Jason Giambi -2.33
Edgar Renteria -2.39
Paul Konerko -2.50
A-Rod -2.71
Barry Bonds -3.69
Jim Thome -4.75
Helton -5.76
Manny -6.55

I could not find anyone with a lower 5 yr total than Manny but these were the only players I checked. For 2006 A-Rod was dead last, Papi # 1, and Manny was 136 out of 162 qualified hitters

IMO Manny hits the same regardless of the situation, not "clutch" like Papi , but not a choker, and his poor results here may be related to his lack of protection and not seeing anything to hit in tougher situations and facing the other teams best RP.


There are some surprising names at the bottom of the list so you wonder if this metric is accurate.

Another measure is RBI Pct from BR. If I could separate it out for close and late situations it would of course be a better measure but I can't. But since 2003, Manny is ranked 20th in MLB in RBI Pct for all players with a minimum of 1000 ROB in the period at 17.6%. Papi is ranked 2nd at 19.02 (Mike Sweeney is first), for reference Pujols is 18.71% and A-Rod is ranked 34th at 15.82%). Jim Rice was 17.8% in his peak years from 1975-1980 (his data calculated from BP).

From 2000-2002, Manny was ranked # 2 in RBI pct at 19.71% (Sweeney again was #1). Has Manny regressed, or is it just that Papi is cleaning the bases ahead of him and taking the easier pickings. Probably more of the latter than the former.

#71 bowiac


  • I've been living a lie.


  • 9509 posts

Posted 20 March 2007 - 09:56 PM

I like Abreu, but would you really rather have him at the plate in a big spot than Manny? I wouldn't.


I don't think Tango would either - while Abreu has been more clutch than Manny in the past, I don't think he was saying that he necessarily expects that to continue. You need to regress those numbers pretty heavily.

#72 Tangotiger

  • 445 posts

Posted 20 March 2007 - 11:19 PM

Tango,
I love you in a manly way, but that example is a little confusing on many different levels.


I'm trying to stay away from mathematical gymnastics. Let's just stick to reality and to models, rather than doing what you are doing.

Ok, so instead of removing 1 out and making that a hit, remove 0.1 outs and add 0.1 hits. Work it out. Use a realistic model for how runs are created. I suggest this:

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

But, you choose absolutely anything you want. I'll guarantee you that you'll get .08 more runs scored by turning 0.1 sure outs into 0.1 sure singles.

***

Update: I just ran it with Markov, and I get an extra .083 runs.

Edited by Tangotiger, 20 March 2007 - 11:23 PM.


#73 Tangotiger

  • 445 posts

Posted 20 March 2007 - 11:31 PM

I don't think Tango would either - while Abreu has been more clutch than Manny in the past, I don't think he was saying that he necessarily expects that to continue. You need to regress those numbers pretty heavily.


I never said anything about the future. I only explained the past.

Obviously, you'd regress the numbers heavily as bowiac is saying.

As for the other comment about "a slice" for late & close: that's the point of Leverage Index. You don't have to "slice" it. Every single PA is given a level of crucialness from 0.01 to 10.00 (average of 1.00). And, Manny simply didn't cut it (relatively speaking for him). Even granting that he didn't (relatively) cut it, he's still OUTSTANDING in the clutch relative to almost every MLB player since 2002 (oustide of Bonds, Pujols, Ortiz).

And ARod was one of the worst clutch hitters in 2006. Ortiz, Pujols and Jeter were at the top. Ortiz 2005/2006 may have had the greatest clutch performance of all time. WPA works and delivers.

Finally, please don't quote how Manny did in 80 or 100 PA as evidence of anything. 2 standard deviations = .100 in OBP. A guy can go .350 OBP in 100 PA, and you are 95% sure his real talent level is between .250 and .450. That encompasses just about every player in MLB.

#74 xjack


  • Futbol Crazed


  • 5162 posts

Posted 21 March 2007 - 02:19 PM

Finally, please don't quote how Manny did in 80 or 100 PA as evidence of anything. 2 standard deviations = .100 in OBP. A guy can go .350 OBP in 100 PA, and you are 95% sure his real talent level is between .250 and .450. That encompasses just about every player in MLB.

Fine.

Over 439 at bats with runners in scoring position from 2004-2006, Manny's OBP was .462. That's 85 points higher than his OBP over 803 at bats with nobody on. There's little question pitchers are more careful with him when the stakes are higher.

#75 Tangotiger

  • 445 posts

Posted 21 March 2007 - 03:41 PM

I'm more concerned about the late & close.

I already said that he was better with men on base than his overall numbers suggest. What he sucked in (relative to his fantastic overall self) was his performance late & close.

This is what I said:

From 2002-2006, he's +24.6 wins, when not considering the men on base and leverage.

If you include the men on base, he's +26.4 wins, which is pretty good. He brings it on, with men on base. However, when you factor in the closeness of the game, he becomes +19.9 wins.

What this means is that:
a - he's still a monster (that +19.9 wins, even considering his chokiness, likely keeps him as one of the top 5 hitters in MLB in that time period)
b - he's not quite the monster his context-free stats make him out to be


Edited by Tangotiger, 21 March 2007 - 03:43 PM.


#76 elias

  • 253 posts

Posted 21 March 2007 - 03:59 PM

I'm more concerned about the late & close.


The biggest problem is that the stats don't show that in close and late situations Manny's mindframe changes significantly with Ortiz directly behind him. If we had samples of three or 4 guys batting in front of David Ortiz and what would happen to their numbers I bet they would resemble Manny's. To weigh Manny's clutch ability without incorporating this, and to suggest anything about his tendencies by the numbers is very shortsighted, and one sidedly simple.

#77 Tangotiger

  • 445 posts

Posted 21 March 2007 - 05:09 PM

The biggest problem is that the stats don't show that in close and late situations Manny's mindframe changes significantly with Ortiz directly behind him. If we had samples of three or 4 guys batting in front of David Ortiz and what would happen to their numbers I bet they would resemble Manny's. To weigh Manny's clutch ability without incorporating this, and to suggest anything about his tendencies by the numbers is very shortsighted, and one sidedly simple.


Ortiz is the #3 hitter, and Manny is #4.

Are you suggesting that Manny's SLG drops 100 points, compared to the other big sluggers who drop 30-50 points in late and close situations because he has no one to "protect" him in late & close (but that this doesn't manifest itself in similar men on base situations, but early innings)? That Manny needs protection only in late&close situations. And that this only applies to Manny, and no one else?

#78 PedroKsBambino


  • SoSH Member


  • 14187 posts

Posted 21 March 2007 - 05:49 PM

Ortiz is the #3 hitter, and Manny is #4.


Not that if affects the larger points you are making, but Ortiz hit behind Manny for 2004, which is in the dataset you cited (2002-6). Otherwise, though, generally Ortiz has hit 3 and Manny 4.

Edited by PedroKsBambino, 21 March 2007 - 05:50 PM.


#79 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4657 posts

Posted 24 March 2007 - 12:41 PM

Late to the discussion, per usual. Skiing was great, though.

First, concerning Manny's defense, 32 runs below average, which would be more like - 36 fielding when considering his "arm" runs saved, makes little sense. I am guessing that number is per 150 games, although that was never stated in the article. In 150 games an average major league left fielder will catch around 290 balls and, in order for Manny to be -36 runs, using the correct (sorry, Val - more on this later) .8 run value for a missed ball, he would have to miss 45 balls others would have caught. This means that Manny would have to miss just over one in every six balls an average left fielder would catch, or one every 3 1/2 games. This does not pass the smell test, imo. FWIW, MGL used -15 runs in his "What is Manny Really Worth" article MGL article, while combining multiple systems over three years and including fielding and arm, and I concur with that number.

Not to put words in Vals mouth, but I understand his skepticism with absolute defensive run values and, although I disagree with his comments on said values (which I refuse to discuss in any depth again), I think I may have an idea where his issues arise. Putting the entire onus on a fielder for missing a play in a certain zone, which most systems do, is perhaps too rigid. Pinto's PMR determines the percentage of plays successfully made in each zone and applies credit/blame based on that number. For example, if an average shortstop successfully turned 75% of all balls hit at a certain speed in a certain zone into outs, then a player missing a play in that zone would be responsible for 75% of the value of that "hit," or .6 runs, not the full .8. Again, not to speak for Val, but perhaps this is what he is sensing is wrong with many defensive metrics, the all or nothing aspect of making the play that can overstate a defensive liability considerably.

Lastly, while I find WPA an interesting and useful stat, I certainly do not think it is nearly as accurate an assessment tool as linear weights, which is what OPS Wins is effectively estimating. When there is a 9 plus run a season difference in the two numbers over the past five seasons, as there is for Manny, there is certainly something going on there, but I am not confident enough in the accuracy and effectiveness of the WPA number to definitively say that Manny is costing his team a win every season because he is a poor clutch hitter.

If anyone has any questions about the metrics I discussed, please feel free to e-mail me and I'll do my best to explain. I usually attempt to do this in my posts, but time constraints will not permit that today. Better yet, buy The Book and read all about it yourself.

Yo la tengo,
Fris

Edited by Frisbetarian, 24 March 2007 - 12:45 PM.


#80 Vermonter At Large


  • SoxFan


  • 3061 posts

Posted 25 March 2007 - 09:55 AM

Thanks for tying this discussion up a little, Fris, and FWIW, I never mind playing Charlie McCarthy to your Edgar Bergen ...

I've been wicked busy and haven't done anything more in this thread, but frankly, I kind of lost myself in this discussion. The main problem is that I really don't know all there is to the calculation of the new UZR so it's hard to get a good reference point for this discussion.

I understand the .8 run calculation as an approximation of run values of misplays over the average fielder, but there are two essential points I tried to make:

1. That value does vary widely according to zone. I do not know if they use varying run values according to which zone the misplay was in, but I hope they do.

2. That .8 run value is valid only when comparing players to "league average," using the .3 out value to zeroize the numbers. I understand this, but would also welcome some calculations and studies at the cumulative level (using the .11 out value) to get a feel for exactly how many runs players give up over the course of a year. This to me would be sort of a grand sanity check, because we know very well how nearly all other runs against are calculated, and if these numbers fit nicely into the overall model of cumulative team and league runscoring, then I think we would surely know how accurate they are.

Here is what that looks like, btw, using the 1992 Cincinnati Reds as an example:

Batting runs allowed:			832.6 runs  As the lwts value of all hits given up
Residual runs allowed:		   158.0 runs  As the lwts value of all BB, HBP, SB, CS, WP, BK given up
Defensive out value:			-426.0 runs  As the lwts value of all outs made (using -.11 per out)
Defensive runs allowed:		   46.2 runs  As the lwts value of all errors made.
Theoretical runs allowed:		610.7 runs  Total runs allocated by lwts measurements of the above
Actual runs allowed:			 609.0 runs  Actual runs allowed.

Lest you think I cherry-picked a team whose theoretical and actual runs are nearly equal, I should say that this model normally comes within +/- 10 runs.

Now where do the UZR runs fit into this model? Well, they don't on their own, because the values of everything else add up to within 10 runs of zero. So fielding runs would modify batting runs against and defensive out values to one extent or another. I would certainly like to know how much, but at the team level a single is a single, an out is an out, and it doesn't matter who the "blame" is assigned to.

I can, of course, break these numbers down to individual pitchers, using the top three categories to measure pitching performance, and calculate their performance on a per BF (or per 27 outs if you prefer) level, which is very useful.

Perhaps the most wonderful thing we could gain from being able to isolate the effect of team defense in terms of cumulative runs would be to modify pitching performances to be truly defensive-independent.

Anyway, as I develop a larger set of data, it may be possible to perform correlations of this data against things like RF, or DERA or other non-quite related measurements to see if there is a relationship, especially with the outlier teams.

So anyway, back to the numbers listed in the model, I would like to know how many plus and minus plays UZR assigns per team, per season. If anyone knows where I can get that data, I can do a quick check against the above data for a sanity check.

Edited by Vermonter At Large, 25 March 2007 - 09:57 AM.


#81 Vermonter At Large


  • SoxFan


  • 3061 posts

Posted 25 March 2007 - 10:35 AM

I left the most important point out of the model above - the holistic relationship between a pitcher and his defense.

Since the hits and outs and everything add up to zero on the team level, any runs assessed by defensive play have to be subtracted from pitching. This is the main basis for my skepticism of the absolute .8 assignment of runs, as you interpreted, every single contact event that stays in the park falls into that holistic relationship. No non-home run contact event is going to be strictly a function of the pitcher, nor strictly the fault of the defense.

There has to be a more/less universal base ratio for an event - say 50-50 as a start point. Depending on how hard the contact is, and the vector and all of that, that ratio goes up or down the moment the ball comes off the bat. Say it's a weak chopper to the left side - a weak contact event, suddenly that becomes an 80% defense contact event. The third baseman fields the ball cleanly and makes a good throw. That's 90% of the complexity of the fielding event and the first baseman completes his 10% part of the fielding play by catching the throw without pulling his foot off the bag. It's an out, -.11 runs. The pitcher gets credit for 80% of that out, or -.089 runs. The third baseman gets 18% of that out for fielding the easy chance and making a good throw, or -.020 runs. The first baseman gets the other -.001 runs for catching the ball.

Something like that, anyway. I think the monolithic value for the plays creates a huge warp in the run-scoring universe at any rate, and a more holistic viewpoint is needed to bring these numbers into something resembling real-world values.

#82 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4657 posts

Posted 25 March 2007 - 11:26 AM

Val -

I think you are confusing things a bit with regards to your metrics. A linear weight assessment of a player (or team in your case) usually shows the number of runs above or below an average player (team), with the total linear value of hits/walks/errors/outs/etc for the entire league adding up to zero. For any interested, there is a Wiki article explaining this. In this type of analysis, the value of an out is between .25 and .30 runs, depending on the league run environment. What you have done to get the total runs to equal the actual runs allowed is to effectively take the out value of approximately .28, then subtract league runs divided by league outs (which equals about .17) to get your plugged in .11 number. The actual cost/value of an out is your .11 plus the league average number of runs per out, which will be .25 - .30 runs depending on the season.

Make sense?



Thanks for tying this discussion up a little, Fris, and FWIW, I never mind playing Charlie McCarthy to your Edgar Bergen ...

I've been wicked busy and haven't done anything more in this thread, but frankly, I kind of lost myself in this discussion. The main problem is that I really don't know all there is to the calculation of the new UZR so it's hard to get a good reference point for this discussion.

I understand the .8 run calculation as an approximation of run values of misplays over the average fielder, but there are two essential points I tried to make:

1. That value does vary widely according to zone. I do not know if they use varying run values according to which zone the misplay was in, but I hope they do.

2. That .8 run value is valid only when comparing players to "league average," using the .3 out value to zeroize the numbers. I understand this, but would also welcome some calculations and studies at the cumulative level (using the .11 out value) to get a feel for exactly how many runs players give up over the course of a year. This to me would be sort of a grand sanity check, because we know very well how nearly all other runs against are calculated, and if these numbers fit nicely into the overall model of cumulative team and league runscoring, then I think we would surely know how accurate they are.

Here is what that looks like, btw, using the 1992 Cincinnati Reds as an example:

Batting runs allowed:			832.6 runs  As the lwts value of all hits given up
Residual runs allowed:		   158.0 runs  As the lwts value of all BB, HBP, SB, CS, WP, BK given up
Defensive out value:			-426.0 runs  As the lwts value of all outs made (using -.11 per out)
Defensive runs allowed:		   46.2 runs  As the lwts value of all errors made.
Theoretical runs allowed:		610.7 runs  Total runs allocated by lwts measurements of the above
Actual runs allowed:			 609.0 runs  Actual runs allowed.

Lest you think I cherry-picked a team whose theoretical and actual runs are nearly equal, I should say that this model normally comes within +/- 10 runs.

Now where do the UZR runs fit into this model? Well, they don't on their own, because the values of everything else add up to within 10 runs of zero. So fielding runs would modify batting runs against and defensive out values to one extent or another. I would certainly like to know how much, but at the team level a single is a single, an out is an out, and it doesn't matter who the "blame" is assigned to.

I can, of course, break these numbers down to individual pitchers, using the top three categories to measure pitching performance, and calculate their performance on a per BF (or per 27 outs if you prefer) level, which is very useful.

Perhaps the most wonderful thing we could gain from being able to isolate the effect of team defense in terms of cumulative runs would be to modify pitching performances to be truly defensive-independent.

Anyway, as I develop a larger set of data, it may be possible to perform correlations of this data against things like RF, or DERA or other non-quite related measurements to see if there is a relationship, especially with the outlier teams.

So anyway, back to the numbers listed in the model, I would like to know how many plus and minus plays UZR assigns per team, per season. If anyone knows where I can get that data, I can do a quick check against the above data for a sanity check.



#83 Vermonter At Large


  • SoxFan


  • 3061 posts

Posted 25 March 2007 - 12:48 PM

Val -

I think you are confusing things a bit with regards to your metrics. A linear weight assessment of a player (or team in your case) usually shows the number of runs above or below an average player (team), with the total linear value of hits/walks/errors/outs/etc for the entire league adding up to zero. For any interested, there is a Wiki article explaining this. In this type of analysis, the value of an out is between .25 and .30 runs, depending on the league run environment. What you have done to get the total runs to equal the actual runs allowed is to effectively take the out value of approximately .28, then subtract league runs divided by league outs (which equals about .17) to get your plugged in .11 number. The actual cost/value of an out is your .11 plus the league average number of runs per out, which will be .25 - .30 runs depending on the season.

Make sense?

Sure, but linear weights are more universal than that. The weights themselves were first actually developed without the out values, then Pete Palmer, apparently with input from David Smyth, added them mainly to zeroize the run values to the league average, which I would term the relativistic model of linear weights. Tango's work added situational variation to the weights themselves, which I would categorize as the probalistic model of linear weights. My work, which is mainly for historical analysis, is what I would term as a cumulative model of linear weights, which assigns absolute values to everything regardless of era.

The values for most of the events is consistent across all three models, plus or minus a few hundredths of a run. It's the out values (both base outs and residual things like caught stealing) that vary from model to model (actually, in the probalistic model, everything varies according to situation, but averages out to essentially the same values as the relativistic model.)

I think all three models have application in baseball analysis. My work with the cumulative model is holistic in nature, in that I look for consistency between individual, team and league run-scoring. My Batting Runs Created/Batting Runs Allowed calculations are kind of a hybrid between Palmer's old Batting Runs, and James' Runs Created, although I didn't realize that when I first worked them out (thus the clever hybrid name was accidental, lol).

Interestingly, Tango wrote an analysis a few years back advocating his version of David Smyth's "Base Runs." In that analysis, he actually worked out the linear weights for James' Runs Created, and those numbers are very similar to the ones I use. Here is the link to that chart on his web page: LINK

I hope this clarifies a bit where I am coming from in my analysis.

#84 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4657 posts

Posted 25 March 2007 - 05:59 PM

But you need to have apples with apples, so to speak, when you are analyzing this type of data, and when you are determining the cost of a missed play you have to use the full cost/value of an out (which will bring the league net to zero - your "relativistic" or "probalistic" model) or you are understating the out value by the league runs per out. There certainly is a place for analysis using the lowered out value, but not when you are attempting to determine the actual value or cost of a out in a game situation.

Look at it this way - last season AL teams scored an average of about 5 runs a game. Using linear values of : singles +0.48, doubles +0.77, triples +1.07, home runs +1.40, walks/hpb +0.32, intentional walks +0.185, strikeouts 0.292, ground outs 0.313, fly outs 0.278, stolen bases +0.193, caught stealing 0.437. (average out .3) and totaling all the players in the league leaves you with close to zero runs. With these values you are easily able to determine then value of a hitter above or below an average hitter. Using these values, you can also go through every play in every game and will once again come up with a net of zero runs, give or take a few. These values represent what actually occurs in the games. Your .11 value for an out does not accurately represent an out because you are removing the run value per out from the actual out value. It is an incorrect number to use as an absolute run cost in this situation as Tango's model clearly shows above.

My head hurts.

Yo la tengo,
Fris

#85 Vermonter At Large


  • SoxFan


  • 3061 posts

Posted 25 March 2007 - 06:11 PM

But you need to have apples with apples, so to speak, when you are analyzing this type of data, and when you are determining the cost of a missed play you have to use the full cost/value of an out (which will bring the league net to zero - your "relativistic" or "probalistic" model) or you are understating the out value by the league runs per out. There certainly is a place for analysis using the lowered out value, but not when you are attempting to determine the actual value or cost of a out in a game situation.

Look at it this way - last season AL teams scored an average of about 5 runs a game. Using linear values of : singles +0.48, doubles +0.77, triples +1.07, home runs +1.40, walks/hpb +0.32, intentional walks +0.185, strikeouts 0.292, ground outs 0.313, fly outs 0.278, stolen bases +0.193, caught stealing 0.437. (average out .3) and totaling all the players in the league leaves you with close to zero runs. With these values you are easily able to determine then value of a hitter above or below an average hitter. Using these values, you can also go through every play in every game and will once again come up with a net of zero runs, give or take a few. These values represent what actually occurs in the games. Your .11 value for an out does not accurately represent an out because you are removing the run value per out from the actual out value. It is an incorrect number to use as an absolute run cost in this situation as Tango's model clearly shows above.

My head hurts.

Yo la tengo,
Fris


Hah. Sorry, my head hurts too. All I am saying is that if you use -.11 instead of -.30, using otherwise the same exact formula, you get a total of 11240 runs, which is pretty close to the 11262 runs the league actually scored. It's just a different aspect. I agree that Tango's model is better for game situations and that type of analysis, but it's also useful to know how many runs were actually scored through different means without the zeroization for higher level analysis.

Sometimes I think we worry too much about normalizing stats. Looking, for instance, at the 1960's we tend to normalize everyone's numbers based on lower league run-scoring averages. Well, if you look at raw production, you can see that guys like Aaron, Mays, Clemente, Robinson, Mantle and Kaline didn't take a hit at all in their actual run production as league run-scoring dipped - their raw run production stayed pretty much the same. So why on earth do we normalize their stats to league averages that they transcended?

So there are uses for the -.11 run value in analyzing the historical game. It's more difficult to work with on some levels, but for serious analysis beyond who is better than whom, or in searching for root causes of certain trends, I think it's preferable.

As I alluded to in my second post above, I think that the overall effect of defense on run-scoring is a more granular thing than the UZR formula allows for in what I believe is it's current form. I think that it is possible to evaluate the numbers so that we can better understand the interaction between pitching and fielding in overall defensive runs allowed, but I think that understanding is sorely lacking at this juncture. Until someone builds a model that takes all of this granularity into account, and looks at pitching and fielding together in a very detailed way and holistic manner, we're just throwing out numbers for the sake of numbers. I don't think that understanding is going to come from zeroizing to the league average, either.

Edited by Vermonter At Large, 25 March 2007 - 06:37 PM.


#86 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4657 posts

Posted 25 March 2007 - 07:01 PM

Hah. Sorry, my head hurts too. All I am saying is that if you use -.11 instead of -.30, using otherwise the same exact formula, you get a total of 11240 runs, which is pretty close to the 11262 runs the league actually scored. It's just a different aspect. I agree that Tango's model is better for game situations and that type of analysis, but it's also useful to know how many runs were actually scored through different means without the zeroization for higher level analysis.


And multiplying the number of outs by .19 (difference between .30 and .11) will also, obviously, give the total number of runs scored. Because you are subtracting runs per out from the actual out value it naturally follows that if you merely mutiply the runs/out rate by the number of outs you will get the total runs scored.

Sometimes I think we worry too much about normalizing stats. Looking, for instance, at the 1960's we tend to normalize everyone's numbers based on lower league run-scoring averages. Well, if you look at raw production, you can see that guys like Aaron, Mays, Clemente, Robinson, Mantle and Kaline didn't take a hit at all in their actual run production as league run-scoring dipped - their raw run production stayed pretty much the same. So why on earth do we normalize their stats to league averages that they transcended?

So there are uses for the -.11 run value in analyzing the historical game. It's more difficult to work with on some levels, but for serious analysis beyond who is better than whom, or in searching for root causes of certain trends, I think it's preferable.

As I alluded to in my second post above, I think that the overall effect of defense on run-scoring is a more granular thing than the UZR formula allows for in what I believe is it's current form. I think that it is possible to evaluate the numbers so that we can better understand the interaction between pitching and fielding in overall defensive runs allowed, but I think that understanding is sorely lacking at this juncture. Until someone builds a model that takes all of this granularity into account, and looks at pitching and fielding together in a very detailed way and holistic manner, we're just throwing out numbers for the sake of numbers. I don't think that understanding is going to come from zeroizing to the league average, either.


Agreed, and the reason for my previous post on absolute blame for missed balls vs partial responsibility based on league averages for the positions in that zone at that ball speed, etc.

I'm having a beer now. Maybe three.

#87 paulftodd


  • 133% banned


  • 1470 posts

Posted 25 March 2007 - 07:45 PM

I just want to quote from tangos/mgl/ad's The Book on the value of the out

The run values presented above are the run values relative to the average player. As we see in the out line, in this typical situation the average team will score 0.538 runs to the end of the inning. However, in situations when an out does occur, the average team wil score 0.240 runs......Before the out, we expected the team to score 0.54 runs. The player who got the out dropped the run expectancy for the inning by 0.30 runs....This is the cost of the out.



Obviously, above average teams and below average teams will have different run values for the out, and it will depend on situations, and vary by game (eg 2-1 game vs a 9-2 game), but this is the average.

100% of the cost is being charged to the hitter here. But we also want to balance the ledger, what is a debit on one side of the ledger (offense) is a credit on the other side (defense).

Defense as we know is some part pitching, some part fielding. Fielding comes into play only on BIP. As VAL indicated, the fielder has a better chance of fielding certain types of BIP for outs than others, and the pitcher has some control over the type of batted ball they allow, for example in an article in THT 2007 Annual, it is said the pitcher has 35% control over a GB and the hitter 65%. While a fielders skill in fielding is important to making an out from a ball hit to them, it is also dependent on his positioning. The positioning of a fielder is dependent on scouting of opposition hitters, coaching to communicate to players where they should position themselves for certain hitters for certain pitches, and communication to the fielders as to what pitch is going to be thrown next.

So when all is said and done, to make sure the ledger is balanced, how much credit for a BIP that is not fielded for an out is due to the hitter, pitcher, fielder. It would be interesting if someone could make a rational argument as to what the various percentages are for each for the various BIP. Charging the fielder the entire 0.30 runs seems unfair and not logical, unless it is an error. It may be useful to do so as a way of comparing fielders, but to say that Manny costs the teams 20-30 runs due to his fielding is misleading, especially when they ignore the contributions of his throwing. How an OFer approaches GB that go into the OF are also factors in a BR taking the extra base.

That said, Manny has regressed defensively, he did so last year due to his knee (?), and his age means he will regress further. Since DH is not an option for Manny, unless Papi spends more time at 1B, some plan needs to be put in place for Manny as to where he will play in coming years unless the Red Sox are going to decline his option after next year or trade him to a team that could use him as DH.

#88 Pumpsie


  • The Kilimanjaro of bullshit


  • 10608 posts

Posted 26 March 2007 - 06:16 AM

That said, Manny has regressed defensively, he did so last year due to his knee (?), and his age means he will regress further. Since DH is not an option for Manny, unless Papi spends more time at 1B, some plan needs to be put in place for Manny as to where he will play in coming years unless the Red Sox are going to decline his option after next year or trade him to a team that could use him as DH.


Unless he recovers from his injury and his defense improves this year over last year. We'll have to see what happens.

#89 twothousandone

  • 2867 posts

Posted 26 March 2007 - 08:30 AM

In the bottom of the ninth, with a lead at home, should the Sox pull Ramirez? For Pena? For Hinske? If by some stroke of luck, Carlos Lee were on the Red Sox, would you put him instead of Ramirez just for defensive purposes? How about Scott Hatteberg (just to pick a randon suggestion)? Kevin Youkilis?

After we've colectively put a number on Ramirez' defense, there are two questions --
1) Is he so bad he should be pulled when defense is much more crucial than offense (protecting a lead)?
2) Do the Red Sox really have a better option than Ramirez?

#90 Tangotiger

  • 445 posts

Posted 26 March 2007 - 02:46 PM

In the bottom of the ninth, with a lead at home, should the Sox pull Ramirez? For Pena? For Hinske? If by some stroke of luck, Carlos Lee were on the Red Sox, would you put him instead of Ramirez just for defensive purposes? How about Scott Hatteberg (just to pick a randon suggestion)? Kevin Youkilis?


Ramirez will cost you say 0.09 runs per game, or 0.01 runs per inning compared to an average fielder.
His bat is worth 0.07 runs per PA.

If he already batted in the 8th, he has very limited value in the 9th as a batter. Of course, if his team mounts a comeback, not only could you use his .07 runs per PA, but they'd be of much higher leverage by then.

The key is leverage, and the replacement. Assuming it's a 1-run game, the leverage is now double or triple. If you have a great fielder on the bench, say someone who is +.01 runs per inning, now all of a sudden you have a .02 runs per inning swing, and when leveraged, is equivalent to .04 to .06 runs per inning. That's a huge gain.

Of course, if the team blows the lead, now his .07 runs per PA would become valuable (if he were to come up).

It's a series of tradeoffs that also has a human element: how will the player respond to this move? Fortunately, Manny is a well-adjusted person, that he'll trust the manager to make the right move, and he'd respond in a professional manner.