Fan vs. PECOTA Projections

[RShack]

That's the point I was trying to make many posts ago. For a shortstop with a .550-.600 OPS to be an average MLB player he has to be one of the 2-3 best defensive players in the world.

We have no reason to believe that LH is that. A lot of guys appear to be (and in fact are) really good, and then grade out at +5 or +10 plays on the +/- scale.

OK, so then, being +27 plays means you're one of the best SS's in the world.

Evidently, Everett suddenly got 25-plays worse in one measly year, right?

Or maybe he just played a lot less? Are these things adjusted based on innings?

How did John MacDonald all-the-sudden get great?

[Baltimoron]

OK, so then, being +27 plays means you're one of the best SS's in the world.

Evidently, Everett suddenly got 25-plays worse in one measly year, right?

Or maybe he just played a lot less? Are these things adjusted based on innings?

How did John MacDonald all-the-sudden get great?

Yes, being +27 means you're one of the best defensive SS in the world. Only 4 guys in mlb have reached +26 over the past two years, and mlb has some good baseball players.

As you have pointed out, +/- is a ranking of total plays made, so its not based on innings, but on total plays made.

Adam Everett broke his leg last year.

Yes, McDonald played very well last year, but he also played more than he ever had before as well. He was +7 in 2005 and 2006 over about 100 more innings at SS than the innings he played at SS in 07.

Think about how much variation there is in batting average over a full season, when a guy might get 700 or so plate appearances. McDonald had 450 chances last year, so keep in mind that sample size is still an issue. But its clear he is a very good defensive SS.

[RShack]

Yes, being +27 means you're one of the best defensive SS in the world. Only 4 guys in mlb have reached +26 over the past two years, and mlb has some good baseball players.

As you have pointed out, +/- is a ranking of total plays made, so its not based on innings, but on total plays made.

Adam Everett broke his leg last year.

Yes, McDonald played very well last year, but he also played more than he ever had before as well. He was +7 in 2005 and 2006 over about 100 more innings at SS than the innings he played at SS in 07.

Think about how much variation there is in batting average over a full season, when a guy might get 700 or so plate appearances. McDonald had 450 chances last year, so keep in mind that sample size is still an issue. But its clear he is a very good defensive SS.

OK, I agree with all that. I'm just thinking that it looks like there's more than just a couple SS's who are that good, but they just don't get to play enough. For example, while there are only 2 guys at or near the threshold for each of the last 2 years, those 2 guys are *different* guys each year. Over the last 2 years, we've got 4 guys, not 2. What happens if we go back a couple more years? Does the number go up even more? (I don't know.)

As I understand it, the fact that we've got different guys above the threshold for the last couple years is not so much about SS's being erratic as it is about who gets the opportunity to play. Therefore, it seems to me that saying that it requires Top-2-in-the-World talent is a bit of a reach. Rather, it's that only 1 out of 15 teams actually plays SS's who are that good in the field most every inning.

Is there any way to guestimate how many guys are that good, regardless of innings played?

[Baltimoron]

OK, I agree with all that. I'm just thinking that it looks like there's more than just a couple SS's who are that good, but they just don't get to play enough. For example, while there are only 2 guys at or near the threshold for each of the last 2 years, those 2 guys are *different* guys each year. Over the last 2 years, we've got 4 guys, not 2. What happens if we go back a couple more years? Does the number go up even more? (I don't know.)

As I understand it, the fact that we've got different guys above the threshold for the last couple years is not so much about SS's being erratic as it is about who gets the opportunity to play. Therefore, it seems to me that saying that it requires Top-2-in-the-World talent is a bit of a reach. Rather, it's that only 1 out of 15 teams actually plays SS's who are that good in the field most every inning.

Is there any way to guestimate how many guys are that good, regardless of innings played?

Regression to the mean tells us no one (sans Everett and perhaps Tulo) is really +28 good.

Its like Home Runs or any other stat. Go look at the leaders for the past few years. Arod's true home run power isn't 54 a year, but he did it last year. He is good, but not 54 a year good. RHow hit 58 in 06 but 47 in 07. He's clearly one of the best power hitters in the game, but he isn't likely a true 58 home run a year guy.

Playing time with defense is more of an issue because a guy who can't hit but has great glove may not get into the lineup as much as he should (witness how little Everett signed for as a free agent), but a +28 is a great season for even the best defensive players in the world, because they have to play well, get some breaks, stay healthy and hit enough to ensure their team keeps playing them (but even the dullest team should realize how baller their defense is if they are really playing this well).

BTW, this is why the whole Tulowitzki postseason awards thing is a joke. Tulo was a legit MVP candidate, not just the roookie of the year and no brainer gold glove.

[Baltimoron]

Is there any way to guestimate how many guys are that good, regardless of innings played?

To get a rough feel you could prorate per innings (or per chance) the three years +/- stats for SS found here "under plus/minus" and then extrapolate to a full season's worth of innings/chances. There are sample size and other issues but it but it wouldn't be a bad place to start.

[Baltimoron]

For example, while there are only 2 guys at or near the threshold for each of the last 2 years, those 2 guys are *different* guys each year. Over the last 2 years, we've got 4 guys, not 2. What happens if we go back a couple more years? Does the number go up even more? (I don't know.)

No to belabor, but this is exactly why the concept of regression to the mean is so important.

Regression toward the mean refers to the fact that those with extreme scores on any measure at one point in time will, for purely statistical reasons, probably have less extreme scores the next time they are tested. Many extreme scores happen to fall with or against the initial score depending on whether your extreme score is extremely high or extremely low.

Consider an extreme example: a class of students takes a 100-item true/false test on a subject on which none of the students knows anything at all. Therefore, all students choose randomly on all questions leading to a mean score of about 50. Naturally, some students will score substantially above 50 and some substantially below 50 just by chance. If one takes only the top 10% of the students and gives them a parallel form of test on which they again guess on all items, the mean score would be expected to be close to 50. Thus the mean of these students would "regress" all the way back to the mean of all students who took the original test. No matter what a student scores on the original test, the best prediction of their score on the parallel form is 50.

If there were no luck on the test then all students would score the same on the parallel form as they scored on original test, and there would be no regression toward the mean.

Real situations fall between these two extremes: scores are a combination of skill and luck. If you choose a subset of people who score above the mean, they will be (on average) above the mean on skill and above the mean on luck. On a retest their previously above-average luck will revert to about average. They will therefore score above the mean due to their above-average skill, but not by as much as they did the first time because they will not be as lucky as they were the first time.

link

[RShack]

No to belabor, but this is exactly why the concept of regression to the mean is so important.

Well, I take your point, but I don't see why you're yelling about it ;-)

It's not that I don't understand the idea. To the contrary, I do understand the idea. It's not some obscure point that's hard to grok. The idea is that each guy has his own typical-level of performance, and you can't just take a single season's upward spike in performance (or downward dip in performance) and conclude that 1-season's worth of performance is the same thing as what you can expect of the guy in the subsequent year. Instead, you should expect that his subsequent year will be more like his typical year than like his spike or dip. (Of course, this idea is mitigated by another perfectly valid idea: that some guys can indeed get better, especially if they get the chance to play a lot.)

Based on that actual idea of "regression to the mean" (instead of assumptions about it), I don't see why anybody would say that Adam Everett's 43+ plays in 2006 is some bizarre 1-season freak performance. If instead of just concluding that, if we compare how he did in 2006 to 2007, it looks to me that he actually did the same in 2007. Now, I don't know exactly how many innings he played. In the general case, I don't think we can use AB's for that, simply because a lot of good-D guys are gonna get inserted late in games at precisely the moment when the team will get the benefit of their glove such that they don't get AB's. But, since the reason Everett didn't play much in 2007 was because of a broken leg, I'll guess that didn't apply to him in 2007. So, if we just go by AB's, then in 2006 he had 514 AB's and 43+ plus plays, which works out to a plus-play about every 12 AB's. In 2007, he had "only" 18 plus-plays to go along with his 220 AB's, which works out to a plus-play in about every 12 AB's. If you don't round it off and instead deal with stuff to the right of the decimal point, then his 2006 performance in 514 AB's would say that he should've had 18.4 plus-plays in his 2007 AB's, but he only had 18. So, the diff is less-than-half of a plus-play. AFAIK, nobody can make a 0.4 plus-plays, either you make one or you don't. His 18 plus-plays in 2007 is exactly what you'd expect from him, given his ABs, if you trust his 2006 performance as being real. So why would we assume that "regression to the mean" indicates that Adam Everett somehow failed to live up to his 2006 glove in 2007? He didn't fail to live up to that. Just going by AB's, it works out to exactly the same thing in 2007 as it did in 2006.

This tells me that we need to be thoughtful before we go assuming that a stellar year somehow is misleading just because of idea of "regression to the mean". Sometimes that's relevant, but sometimes it's not. Just because somebody has a great D year, that doesn't mean it's a fluke. To the contrary, we've got 2 years in a row of Adam Everett doing 30-some-% *better* at plus-plays than he was in 2005 when evidently he was more valuable than Jeter. In 2005, he made a plus-play for every 16 AB's. Since then, he's made one for every 12 AB's. So, it seems to me that one possibility is that he was getting better even though he was 29, maybe just from playing more. I'm not concluding that, because I don't know how he did in 2003 and 2004, but however he played over the last 2 years, it doesn't seem that "regression to the mean" has much to do with it. I agree that the idea of "regression to the mean" is important. However, I don't buy that it's *SO* important that we should just *assume* that's what's happening when a guy puts up good-D performance. To the contrary, if we really trust stats, then we should not assume that. Instead, we should actually look and see.

[TinCup]

Well, I take your point, but I don't see why you're yelling about it ;-)

It's not that I don't understand the idea. To the contrary, I do understand the idea. It's not some obscure point that's hard to grok. The idea is that each guy has his own typical-level of performance, and you can't just take a single season's upward spike in performance (or dip in performance) and conclude that 1-season's worth of performance is the same thing as what you can expect of the guy in the subsequent year. Instead, you should expect that his subsequent year will be more like his typical year than like his spike or dip. (Of course, this idea is mitigated by another perfectly valid idea: that some guys can indeed get better, especially if they get the chance to play a lot.)

Based on that actual idea of "regression to the mean" (instead of assumptions about it), I don't see why anybody would say that Adam Everett's 43+ plays in 2006 is some bizarre 1-season freak performance. If instead of just concluding that, if we compare how he did in 2006 to 2007, it looks to me that he actually did the same in 2007. Now, I don't know exactly how many innings he played. In the general case, I don't think we can use AB's for that, simply because a lot of good-D guys are gonna get inserted late in games at precisely the moment when the team will get the benefit of their glove such that they don't get AB's. But, since the reason Everett didn't play much in 2007 was because of a broken leg, I'll guess that didn't apply to him in 2007. So, if we just go by AB's, then in 2006 he had 514 AB's and 43+ plus plays, which works out to a plus-play about every 12 AB's. In 2007, he had "only" 18 plus-plays to go along with his 220 AB's, which works out to a plus-play in about every 12 AB's. If you don't round it off and instead deal with stuff to the right of the decimal point, then his 2006 performance in 514 AB's would say that he should've had 18.4 plus-plays in his 2007 AB's, but he only had 18. So, the diff is less-than-half of a plus-play. AFAIK, nobody can make a 0.4 plus-plays, either you make one or you don't. His 18 plus-plays in 2007 is exactly what you'd expect from him, given his ABs, if you trust his 2006 performance as being real. So why would we assume that "regression to the mean" indicates that Adam Everett somehow failed to live up to his 2006 glove in 2007? He didn't fail to live up to that. It works out to the same thing in 2007 as it did in 2006.

This tells me that we need to be thoughtful before we go assuming that a stellar year somehow is misleading just because of idea of "regression to the mean". Sometimes that's relevant, but sometimes it's not. Just because somebody has a great D year, that doesn't mean it's a fluke. To the contrary, we've got 2 years in a row of Adam Everett doing 30-some-% *better* at plus-plays than he was in 2005 when evidently he was more valuable than Jeter. In 2005, he made a plus-play for every 16 AB's. Since then, he's made one for every 12 AB's. So, it seems to me that one possibility is that he was getting better even though he was 29, maybe just from playing more. I'm not concluding that, because I don't know how he did in 2003 and 2004, but however he played over the last 2 years, it doesn't seem that "regression to the mean" has much to do with it.

Shackie...I swear the mods are going to have to invoke a new "Shack rule"...no more than one post a day that exceeds 50 words!! You are the king of verbosity.

I AM kidding really. Just playin'

It is enlightening material.

[Baltimoron]

My regression to the mean comments had nothing to do with Everett. We have lots of data on Everett from which to assess his real mean or true talent. Regression to the mean becomes less meaningful as our individual's sample size grows.

To get back on track, I referenced regression to the mean in response to this:

For example, while there are only 2 guys at or near the threshold for each of the last 2 years, those 2 guys are *different* guys each year. Over the last 2 years, we've got 4 guys, not 2. What happens if we go back a couple more years? Does the number go up even more? (I don't know.)

DUCY regression to the mean is relevant to your comments about 4 guys and their one year performances over 2006 and 2007?

Its one thing to say that a player could play at +28 (or +21), its another thing to assert that such a level is reflective of their true talent.

[RShack]

OK. I have a couple questions:

1. Do you have a quick-and-dirty RC calc using that 1.8 factor that produces better results than the one that treats OBP and SLG equally?

He did one here:

So, the real equation is this:

Runs: (1.73*OBP + SLG) * 0.27 * PA

Am I being a dope about this?

When I try that, I get a RC number that's more than double what I get with AB * OBP * SLG.

What am I doing wrong?

[Baltimoron]

Am I being a dope about this?

When I try that, I get a RC number that's more than double what I get with AB * OBP * SLG.

What am I doing wrong?

It’s a marginal formula, meaning that it measures the change in OBP and change in SLG against the change in RE (i.e., LWTS). My bad.

So, it’s not Runs equals all that. Use SLG*OBP (*.87*PA)

But baseruns would be better.

[Baltimoron]

Sean Smith had Luis + 4 runs over 139 innings at SS in 2007.

Justin Inaz had LH +4.1 runs as a SS in 2007.

The problem, of course, is one of sample size, but each of these guys had him 4 runs above average in like 1/10th of a season at SS.

Given a fulltime SS plays 1300-1400 innings, Hernandez was on a pace in 2007, which if sustained, would make him the best defensive SS in baseball by a fair bit. That's sorta like saying the guy with 5 homes runs over his first 70 plate appearances will hit 50 - a lot can happen and the sample size is inherently unreliable.

So if Luis plays a little worse than he did last year in his short time at SS in 2007 over a full season in 2008, Adam Loewen will be the happiest guy on the planet and the Os would have a value at SS, even if he is really, really bad with the bat. Was his short stint reflective of how good he is with the glove?

[Baltimoron]

Did you all see this article on THT?

It is a look at yet another defensive metric. The great thing about this metric is that it gives us an ability to measure players back to 1956. The article inlcudes the top defensive players at each position by his system from 1956 to 1986. This is anothe +/- system. Keep in mind that since it ends at 1986, it doesn't include the last 10 years of Ozzie Smith's career. Orioles fans will like his rankings at 3B (Brooks by a wide margin), SS (Belanger, Aparicio top 2 spots), and CF (Paul Blair by a pretty comfortable margin).

Brooks was like 110 and the Blade like 70 runs better than the next closest fielder from 1956 to 1986:

OMG

That is absurd- the two best fielders over this three decade period (yes the math is goofy because careers get cut off but..), by a huge margin, playing on the same side of the infield.

[Baltimoron]

What kind of scouting and minor league defensive numbers are there for LH? I never realized he was that good last year - albeit a very small sample size. Was he always regarded as a silly good (like best around) SS?

[Hallas]

Brooks was like 110 and the Blade like 70 runs better than the next closest fielder from 1956 to 1986:

OMG

That is absurd- the two best fielders over this three decade period (yes the math is goofy because careers get cut off but..), by a huge margin, playing on the same side of the infield.

I'm surprised at how well Belanger comes out on these metrics. I always thought he was a close second to Ozzie but a lot of this new data says that Blade out did the Wizard.

Granted I've only had the opportunity to look at Ozzie play so I might be a little biased.