Manny’ Defense
This started out as a study in which I questioned the wisdom of regressing Manny’s defense toward the mean, but after exchanging emails with MGL, I discovered that I had misunderstood his regression methods. I had thought that in projecting Manny’s future performance (or assessing his true talent level, as MGL calls it), he would include an element of league average OF. But that is not the case…
In fact, MGL considers age when determining the ‘mean’ to which he regresses. In Manny’s case, he would start with the following data (year and UZR from MGL, innings from BBRef):
Year, UZR, Innings
2000, -22, 799.3
2001, 9, 482
2002, -12, 581.3
2003, -20, 1073
2004, -20, 1087.7
2005, -50, 1225
From here, he weighted the years (5-4-3-2-1), and because he was using 5 years of data, he regressed Manny about 20 percent. But the surprising thing (to me at least) was that he regressed him toward -10 UZR--the average for a corner OF of Manny’s age during that period. He also tacked on about 5 runs for aging, and came up with this 2006 projection:
-30 UZR
In my attempts to outwit what I perceived to be shortcomings in MGL’s methods, I had gotten around -35 UZR--an inconsequential difference. Moral of the story--never doubt MGL.
Other moral of the story: Manny’s defense is likely to take a huge chunk out of his value in 06. Looking at his 05 BRAA from BPro (+44), if Manny can repeat his offense from 05, he’ll be worth something like 15 runs above average in 06. That’s something like what Jermaine Dye, Jay Gibbons, or Jason Lane did in 05 (assuming they were average fielders). Therefore, his contract is a tremendous albatross and trading him should be a priority for the Red Sox.
Thanks to MGL and MCoA for their help assembling the info for this.
Darren
Posted: December 30, 2005 at 11:18 AM |
116 comment(s)
Related News:
Sabermetrics,
Boston
Reader Comments and Retorts
Go to end of page
Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.
Well, sure. But I have seen nothing on this or any other thread that leads me to believe that -15 is anywhere close to an accurate representation of Manny's defense in 2005 or his real ability level. You have to take the most optimistic (IMO unrealistic) number, then assume you need to make an additional park adjustment, then also regress towards the mean. Too many assumptions, all in Manny's favor, that are just not realistic.
I would be quite confident advising Omar Minaya that Manny Ramirez would be at least 25 runs worse in left field than Cliff Floyd, and even that would be assuming Floyd is significantly worse than he was in 2005.
How on earth can you know this? These sorts of play-by-play estimates of defense have only just started being used to estimate defense. How do you know that -35 is Biblically bad? How do you know that Greg Luzinski wasn't -90?
You make tons of assertions, but it's basically coming down to your gut feeling that -35 or -40 or -50 are too large (on an absolute scale). If you have anything other than a gut feeling, then include it. These "it's not true because it just feels wrong" arguments are not terribly compelling.
I'm not saying I am convinced that Manny Ramirez is -50. I have no idea (if I had to guess, I'd guess he was a lot closer to -30), but I don't just dismiss the -50 because it feels wrong. That's not how the scientific method works. I certainly would never claim that a -50 has to be wrong without having something a little more substantial than "it feels wrong" fuelling my assertion.
In possibly the greatest season ever by a pitcher, Pedro prevented 64 runs more than a replacement pitcher in 217 innings. Manny played 1251 innings in LF (if I have my math right). So either Manny's performance was beyond historically bad, or the LF is no less than 1/6 as important as the pitcher in preventing runs (I realize that really we're talking about the spread since it's runs above average). Maybe defense is much much more important than I previously thought (I'd like that to be true since I used to be a good fielder and couldn't hit to save my life) or maybe Manny really did have a far worse season with the glove than Pedro had with his arm that year. But I'm going to need significantly more evidence to believe either of those conclusions. "Extraordinary claims require extraordinary proof."
I'm not sure where you got your 64 RAR anyway. According to BPro, Pedro had 119 PRAR in 2000. Barry Bonds was 149 BRAR in 2001. Manny's defense is nowhere near those levels. If you want a decent comp, Rafael Belliard was -22 BRAA in 315 PA in 1992 (roughly -45 BRAA if prorated to Manny's innings last year). I find it easy to believe that Manny is as bad a LF as Belliard was a hitter in their respective worst years.
In theory, every team puts their worst starting OF in LF. 50 runs worse than the average of those players is pretty extreme when you consider that SS appears to be bound somewhere around +/-30. Again, in theory, there should be the most variation at SS because it's at the thin end of the talent pyramid and SS's have the most opportunities. (So there should be the most variation in talent AND the largest sample to manifest that range). A LF at -50 in the same system where a -30 at SS induces guffaws should be suspect. Manny could be that bad. Defense could be a far bigger part of the game than conventional wisdom dictates (if it is normal for LF's to be +/-50 then team defense could have a maximum range of +/- 50 wins, roughly). There could also be something singularly unusual about last season, either in Manny's performance or in our analysis of it. If Manny really is asking for a trade (who knows?) he could be missing some plays on purpose.
3) I still haven't gotten clear as to whether the -50 is how many runs Manny cost relative to an average LF would given the same opportunity, or how many runs Manny would cost in an average park compared to an average LF.
I believe it's relative to a theoretical average LF subbed in for Manny, but I can't put my finger on MGL's uzr rundown at the moment. I found and old link to it, but now it just goes to the Primer newsblog.
I got a projection of 72 VORP+UZR for Tejada and 50 VORP+UZR for Manny.
So, just to put this in context, even if we take the most favorable of the pbp-based defensive evaluations, we still have Miguel Tejada as more than 20 runs better than Manny. Tejada should get another few runs for his baserunning, as well.
So, that -47, in real terms, is actually about -38. That's still very, very bad, but I think it helps a little. Manny's defense, by UZR, cost the Red Sox about 40 runs below average in 2005, not 50.
(This also means that the -30 projection from Darren and MGL in the introduction, can be understood as a -22 projection based on Manny's typical playing time.)
Unless I'm wrong about what these numbers mean - and I'm only running off memory here, so please correct me if I'm all wrong - I think this significantly shortens the gap between the different measures. I substituted Manny's 2005 UZR into my quick projection spreadsheet, and it only cut his projection by ~5 runs. That's still worth noting, and if the trade happens and a lot of money changes hands, it might make a difference, but it's not a sum worth going to theoretical war over.
At -50, the difference in Manny's projection for the simple 5/4/3 calculation is ~10 runs.
So, to return to my point from post #7/107, the question can be re-framed as asking whether Manny Ramirez projects as 20 or 30 runs worse than Miguel Tejada.
One could make a more far-reaching case that we should still not trust the pbp metrics for Manny's defense, and it's even better than -15 to -25 per season, but I find that argument unconvincing. We've established what kind of defender Manny is, now we're just haggling over price.
MGL posted in the gold glove thread a while back that Manny's 2005 UZR was -47. Apparently, since then, he has changed UZR enough that Manny's 2005 UZR is now -50.
But if Manny costs the Sox 50 runs, he was adding about half a run to the ERA of each of their rotation starters, so really the team got fine pitching, bad as it may have looked. Meanwhile, the other guys, Damon and Trot & Co. and the apparently undistinguished bunch of infielders, were perhaps really very good fielders.
I guess I can accept that.
I still wonder about Fenway -- not that it skews the raw numbers, but that the numbers may have different values there. A ball is hit to the track in Boston, Manny runs into the wall, Damon picks up the ball, and the batter stops at first. A ball is hit to the track in Detroit, Manny falls down chasing it, the batter ends up on third.
I don't know if that means anything. It may mean that it's easier to "hide" a truly bad fielder in Fenway's LF than in almost any other park and place: a missed chance has less impact there than in other left fields. A lot of defense takes place after a chance has been missed or made.
Of course, Manny has to play in Detroit, too. And the mitigating effect of all those singles-instead-of-triples might be pretty minor in any event.
One question I have is, is it legit to regress an apparent outlier even more than the standard amount when doing a projection? I believe there is a technique in statistics which involves discarding the outliers. So, the idea apparently shouldn't be dissmissed out of hand.
Red Sox 2005: 4.31 dERA, 4.74 ERA, 5.07 RA
They pitched 1429 innings in 2005 and allowed 752 earned runs, 805 runs total. Taking away 50 runs from those (with the same number of innings, for lack of something better to do) would give them an ERA of 4.42 and a RA of 4.76.
It seems to me, given the differences between dERA and ERA that there is ample room for Ramirez to be about -50, Renteria to be bad (though considerably better than Ramirez), and the rest of the team to, on average, be average.
This is all back-of-the-envelope stuff and shouldn't be taken too seriously, but it seems to me that if -50 were WAY TOO BIG for Ramirez, subtracting his 50 runs from the Red Sox's numbers would give them an ERA vastly better than their dERA (which would not jibe with the general impression of the rest of the team's defense as pretty much mediocre, on average).
Those numbers in the intro are /150 games. I don't know why, but for some reason that got edited out when I changed the intro. Sorry. So you're right to lower the number, but you've lowered it too much. (I also think that in the games where you put Manny at DH, you have to adjust for forcing Ortiz to play 1B somehow, and forcing a good bat like Youks to the bench).
As for Rally Monkey's #s, Dial's spreadsheet comparing various systems has Rally's #s at -34 (or did you adjust this for games already). One other number you didn't include was DSG's which was -35. (Dial was -34)
One other factor: MGL emailed me to say that the numbers he gave me above did not include arm. Here are those numbers:
01: 0
02: -2
03: +6
04: -6
05: +6
Overall, I think that it'd be very close to a wash for future projections (maybe 1-2 runs better).
I specifically said that this just contradicted my intuition - I might be wrong, but I believe that to say a LF can have that much impact on runs allowed relative to a pitcher's impact is an extraordinary assertion and demands extraordinary proof.
I'm not sure where you got your 64 RAR anyway. According to BPro, Pedro had 119 PRAR in 2000.
Sorry, my mistake, I meant to say 64 runs above average, which I think is the important comparison here, since Manny is comared to average. If Manny were as much worse than a replacement fielder as Pedro were better than a replacement pitcher, well, he'd be in Little League. (I know he looks like he his sometimes.)
Before we spend too much time (or have we already) looking for flaws in the metric, can anyone give any sort of indication as to the variability of uzr? I'm not just talking about projection - what is the inherent error in the measurement. Mgl changes the method in some way during the year and Manny's number changes 6%. There are other parameters I imagine he's not changing which also have some error associated with them. Do any of us have any idea what sorts of magnitudes we're talking about? Even if there are no systematic errors to produce the -50 number, if the random error is +/- 30 runs, that's a big deal. You'd have to discount the defensive projection relative to the more defined aspects.
To repeat a couple things I've said earlier - even if the error is 50%, uzr is by no means useless, and if we can move Manny in such a way that even if he were -10 or something on defense it would be fruitful, that would be spectacular.
You must be Registered and Logged In to post comments.
<< Back to main