THE BOOK--Playing The Percentages In Baseball

Does an influx of talent like this affect the positional scarcity adjustment? 

I guess this is a case in point for an ongoing question that I have.  The talent pool is fluid but the positions available is static.  In this instance it would seem like teams should have an easier time finding (trade or FA) a CF.  Am I misunderstanding the positional adjustment in the WAR calculaiton?

NO!  It does not affect the “positional” scarcity.  Because CF is not a position independent of LF and RF.

It could affect the overall OF/IF relationship, because we rarely see pure OF play the IF, though Schumaker might start being the rule.

I think Tango would wholeheartedly agree with this, but if you look at defense as being just defense and not defense as a CF’er, defense as a LF’er, defense as a 2B, which is THE CORRECT WAY of looking at defensive talent, there is no such thing as having a lot of CF’ers, few 3B, or any such thing.

It could be that there is better defense overall in the majors (or in the minors) right now for a variety of reasons, but that does not mean that there is a plethora of CF’ers, per se. Now, it just so happens that because there are many more minor league teams than major league teams, that if defense gets “better” in the majors, of course there will be more players in the majors who played CF (or SS) in the minors. That goes without saying.

And, if the style of hitting changes in the majors, the overall quality of defense will change in the majors, which is probably what is happening.

Saying that there is an “abundance of CF talent” in baseball right now, is like saying there is a an abundance of SS talent in the majors because most players in the majors played SS in Little League and High School! There are always going to be same number of CF’ers, LF’ers, SS’s, etc., in the major leagues, right?  More or less.  So, what Dave is saying is that there are more players in the majors who used to play CF in the minors, which is to be expected if there is a better quality of defense in the majors than in the past.

Or the whole thing (or part of it) could just be a random fluctuation, as there are 8 positions to choose from, and it is likely that one or more will be over or underrepresented in a small time frame just by chance alone.

Of course, there could be a change in the way major league teams view the defensive spectrum for whatever reasons, that causes one position or the other (again, from the minor leagues) to be under or over-represented in the majors.

I think Cameron was arguing that the current crop of CFers are unusually talented overall, not just defensively. 

That said, they appear to be the same hitters they’ve always been:
OPS(CF) / OPS (Lg)
2009:  752 / 752
2008:  753 / 749
2007:  758 / 758
So Cameron is only right if they are in fact better defensively than past cohorts of CFers.

And why couldn’t a certain group of CFers (defined as those who actually happen to play CF) be better defensively than has generally been the case?  I don’t see why that’s logically impossible.

"And why couldn’t a certain group of CFers (defined as those who actually happen to play CF) be better defensively than has generally been the case?  I don’t see why that’s logically impossible.”

That could be either through a fluke or because teams for some reason think that defense is more important in CF than they used to.

But, if defense in general is deemed more important in the major leagues, for whatever reasons, you will see more players in the major leagues from the right side of the defensive spectrum in the minor leagues.

Now, I have not RTFA, but if offense among CF this year is the same as its been relative to the whole league, and Cameron thinks that CF’ers are “better” this year, how does he know that they are better defenders?  In any given year, each position is “zeroed out” in UZR.

I just read the piece.  No numbers. Just an “observation” that there appears to be a “great crop of CF talent.”

Now, unless the current CF’ers in the major leagues are younger than before or their defense is above average as compared to past years (or their EQA or wOBA is higher than their OPS suggests), Dave appears to be suffering a bad case of “observational delusion,” at least according to Guy’s numbers!

While delusion can’t be ruled out (I was suffering from a pretty nasty migraine today, after all), I would be shocked if the current group of center fielders in baseball weren’t abnormally young. 

Of the 26 guys I listed, only Hunter, Cameron, Byrd, and Fukudome are 30+.  Most of the guys playing center field right now are 22-27, and while I know that guys move to the corners as they age, I can’t recall any years where there were so many good young players at the position.

I’m sure someone can give us the average age of either primary CFers or the average age of all CFers weighted by the number of innings or PA at the CF position both this year and in year or years past.

According to B-Ref, the average age this year of top 30 CFs (by playing time) is 27.0, the average age for all hitters is 28.9.  I would guess it’s pretty typical for CFs to be c. 2 years younger, but that’s just a guess.

Average age of center fielders:

yearID POS_Age
1973 26.7073
1974 26.4849
1975 26.4588
1976 26.5810
1977 26.0556
1978 26.4512
1979 27.2090
1980 27.3461
1981 28.0669
1982 27.3959
1983 27.3073
1984 27.2464
1985 27.3086
1986 27.2144
1987 27.0440
1988 27.7704
1989 27.5550
1990 27.8676
1991 27.4142
1992 27.7467
1993 28.0884
1994 28.4279
1995 28.2637
1996 28.2933
1997 28.4687
1998 28.4666
1999 28.1853
2000 27.7973
2001 27.9865
2002 27.9718
2003 28.3558
2004 28.4660
2005 28.6113
2006 28.5347
2007 28.2699
2008 27.4552
2009 27.2009

That’s seasonal age (age as of July 1), weighted by the number of games played as a center fielder. (Current season data pulled from Baseball Reference, everything else from the Baseball Databank.)

I’m sorry, but (especially given the OPS data noted above) nothing in Cameron’s article persuades me of his premise.

Well, crap.

It is not going to make much difference, but the age in 2009 is definitely younger than the average age over the last few years, by almost 1 year.  The only problem is that if we accept age 26-28 as the peak years for offense and defense combined, the fact that players are a little younger this year should have no bearing on the argument in question.

So, good catch Dave on the age thing. But, the basic premise, is, well…

That should be a lesson to all of us about drawing conclusions from observation or anecdote without “running the numbers.”

Now, there is still a chance that the current crop of CF’er is better defensively than, say, the last 3 years, in which case Dave’s premise would be correct…

Revised zone rating is up a bit, .932 to .922 last year.  Out of zone plays are up slightly too.

Can’t compare RZR y-t-y like that - the zones tend to shift around too much for that, especially in the outfield. (They use very small zones for RZR, and what’s defined as “in zone” changes depends on the number of plays made in that zone compared to BIP, IIRC.)

Let’s try offense, using Baseball Reference:

A look at offense:
Players, 250 games in CF, OPS 100+ or more
for 3 year periods, 10 years apart:

1967-1969 12
1977-1979 12
1987-1989 11
1997-1999 10
2007-2009 11 (using 240 games)

Players, 100 games OPS 100+

2009 9
average 1989-2008 11 per year (220 exactly)

No apparent proven offense gain

Dave,

Could’ve posted this on FanGraphs, but I do find it interesting that you completely omitted Aaron Rowand from the list. He’s right around 2 WAR (which makes him exactly worth the money he’s being paid, though not living up to a potentially bad contract shouldn’t disqualify him from the list you made), and his poor defensive year last year appears to be a big fluke. He’s certainly no worse than Fukudome, who really should be playing right anyway. Intentional omission or mere oversight?

What the young age shows is that there’s a move toward fielding talent in CF.  Going from an age of 28.6 for CF to 27.2 in 4 years is an enormous change.  That’s the biggest drop in average age in the period Colin covered above.

I don’t see anything Dave has to retract there.  I would not be surprised if it was accompanied with an average age increase in the corners (LF especially).

***

Colin: you made a math mistake, I think.  Did you do this:
1. Calculate seasonal age, as an integer.
2. Provided the average age, using the integer, and reported results as a decimal?

If so, read this:
http://tangotiger.net/wiki/index.php?title=Seasonal_Age

"Colin: you made a math mistake, I think.  Did you do this:
1. Calculate seasonal age, as an integer.
2. Provided the average age, using the integer, and reported results as a decimal?”

I do that all the time. Why can’t you do that? If half my players have a seasonal age of 25 (calculated in integers) and half as 26, my average age is not 25.5?  Of if 3/4 of them are 25 and 1/4 26, the average age is not 25.25?

I know you have spoken about reporting a number with more precision than the “inputs” but in this case, how do you suggest reporting the “average seasonal age” when I half my players are 25 (seasonal age) and half are 26?  25?  26?  That makes no sense.  Or maybe I am barking up the wrong tree.  I did not read your link yet!

“I don’t see anything Dave has to retract there.”

Are you assuming then that it is automatic that the defensive talent of a younger group is better than an older group?  If yes, nothing to retract, but I don’t think that is what he meant (that CF’ers today are younger so therefore better defensively).  If he meant that, he would have (and should have) said that.

"I did not read your link yet!”

The answer is there.  But, I’ll quote it for people with RTFA-aversion:

Age, unlike all other numbers used in computations, is truncated to the lower integer, rather than rounded to the closer integer. The result is that when you have to figure out a team’s seasonal age, the result will be biased by 6 months.

As an illustration, if you have one player that is 27.60 years old as of July 1, and another player is 28.40 years old as of the same date, the true average as of July 1 should be 28.00. However, the seasonal age format will first truncate 27.60 to 27, and then trucate 28.40 to 28. The average of 27 and 28 is 27.5, a number that is lower than both numbers it is based upon!

The correct action is to not truncate at all, and simply carry the decimal. Otherwise, if truncated is required, then the averaging of the ages should be done prior to truncation.

Because of this issue, some analysts determine the age as current year minus birth year, basically establishing the age as of Dec 31 midnight of the year in question. A player that is 27.60 years old as of July 1, will then be counted as being 28.10 years old, meaning truncated to age 28. A player that is 28.40 years old as of July 1 will then be counted as being 28.90 years old, meaning truncated to age 28. The average of the truncated numbers is 28.0. In short, in order to counteract the bias of 0.5 years from the truncation-then-average of the seasonal age process, you add 0.5 years first prior to truncation.

The benefit is that you only need to know the birth year of the player, rather than also know whether he was born before or after July 1.

That’s how the data is reported on B-Ref, and so I kept my numbers consistent with theirs. When the 2009 release of the BDB is out, it’s trivial for me to figure ages however anyone wants.

I don’t think it’s an issue, though, for this analysis - we don’t care about the actual age but the relative age. (Actually, I should probably compare it to the average age of the league as a whole. I’ll try to look at that tonight when I get home.)

OK, I was thinking that seasonal age was season year minus birth year.  If you use that, you can then average everyone’s age and use any number of decimal places to represent the average?

MGL: right, it will be closer:

Colin: well, Sean is wrong, and I told him as much on his blog.  Never would you ever truncate a number, and then find the average of a series of such numbers.  Ever.  And then, to report this to at least one decimal places?  You can’t report numbers more significant than they are.  Do you know how many math rules are broken?

Yes, all the numbers are relative, so it doesn’t matter in the aggregate.

But, if what you do is:
a) wrong
b) more complicated (you need to know what month he was born)
Why in the world would you do it?  Because b-r.com does it?

Because I was using b-r.com’s data - it’s the most convenient source I have of ages and defensive position played for players in the current season. For a quick little blog comment where it really comes out in the wash, I didn’t see the point in finding the birthdate of every center fielder in the majors this year.

Let’s say you have a group of 5 players, 4 of whom are 19 years, 355 days and the 5th is 20 years 40 days.

Let’s say you’ve done studies to determine how much someone improves by year, with the year truncated, so 19 years 0 days and 19 years 364 days are the same. Same holds true for each 1 year bracket.

Let’s assume that 19 year olds improve 5% a year, 20 year olds 3% a year, and there is no improvement the next 5 years.

So, your true math age of your sample is 20, but your “truncated average” is 19.2

How much will your group improve the next 2 years?

If your studies are right, then they really should improve 4.6% next year, not 3%.
And 2.4% in year 2, not 0.

You are really using age as a variable to determine the amount of future improvement, and not as the exact age.

How you calculate age is all relative of course.  That is a given.

However you are wrong with what you said, because you are treating age as discrete when it is continuous.

If the improvement is 5% among “true” 19yr olds (meaning those from 19.000 to 19.364 years.days), and 3% for those 20.000 to 20.364 years.days, then someone who is 19.355 years.days old is pretty much at bit over 4% expectation in the aging curve.

It is not like you have a staggered development cycle where if you were born one day after the arbitrary cutoff point that you are either in the 5% or 2% group.

***

My point is that once you have decided to truncate at some arbitrary point (which is fine), and then you decide to find the “average”, you CANNOT now report it to one (or four!) decimal points.  We literally don’t even know what that means.

Just as in my example where you have someone who is 27.183 years.days old and someone who is 28.182 years.days old should give you an average of 28.000 years.days, what is instead being reported is that the average age is 27.5 years old.  How can the average of two numbers be less than BOTH numbers?

The correct thing to do is to report the average as either 27 or 28 (whichever, I don’t care), because that is all that the significant digits allows you to report.  By doing that, you are expressing an uncertainty at the first decimal place, which is true.

***

In any case, it drives me bananas that the average ages being reported are biased by 6 months.  And we seem to be doing this because Bill James decided what a seasonal age is, and no one wants to challenge him.  I’m not playing ball.

***

Colin: I didn’t know you were handling someone else’s dataset.  In that case, you use what you’ve got, I agree.

Tango, I’m not sure why you conclude “I don’t see anything Dave has to retract there.” I agree that the age change over last 4 years appears to be non-trivial.  But the average age is not enough below the historical average to just assume this is an outstanding group of defenders.  We’d need some evidence, like CFs making a larger than usual share of BIP outs. 

And the bottom line is these guys are at peak age, 27, yet are still just league-average hitters (at a position we expect league-average hitting).  There’s no reason to expect them to improve, given their age.  So what makes them outstanding?

I’m not saying they are or are not outstanding.  I didn’t study the issue.  Prima facie, it looks like a reasonable thesis.

Let’s see… I’ve got my WAR forecasts entering 2009.  We can look at those, and see if CF stands out.

I had 56 nonpitchers with a WAR of at least 3.  That would mean, we’d expect around 7 for each position, if no position had “extra” talent.

9 CF
9 3B
8 RF (includes Markakis, Rios, Span, Ichiro)
8 1B
8 2B
6 SS
4 C
3 LF
1 DH

Those RF would easily be CF on other teams.

The real story is how little talent you can find in LF. 

Clearly, teams are trying to put alot of their hit-no-field players in LF, and they are failing (overall).  Just as the overflow of quality CF has seeped into RF (Rios in CF with Sox, Ichiro would be CF, Markakis and Span blocked, like Gutierrez and Victorino were blocked), teams SHOULD be putting their overflow into LF as well (but they aren’t).  So, you are left with abundance of talent in CF.

Normally, the fielding spectrum is CF to corner OF.  But, we’ve seen plenty of players go from RF to CF, precisely because there’s so much talent in CF, that those guys are moving around to make way for others.  Franklin Gutierrez is another example of a guy who was blocked at RF because of Sizemore. 

So, I think Dave’s basic assumption is correct.  There is an overflow of CF talent, and teams need to be able to move that talent around somehow, as there’s a not-coincidental dearth in LF talent.

To tie that back to your (and MGL’s) earlier point that CF and LF/RF are not independent positions:

It appears to me that the right conclusion to draw here is that there is a pool of 20 “talented” outfielders, and teams are choosing to put almost all of them in Center and Right, and very few of them in Left, for some reason.

Well, the “some reason” is inefficiency in valuing offense and defense.  Some teams see LF as somewhere to put your big hitter, without consequence to their fielding.

Ryan Langerhans (and Endy Chavez to a lesser extent) epitomizes this.

Langerhans is a below average hitter, but “good enough” for a CF.  But, he’s a good-not-great fielding outfielder, but certainly is above average for the corners, but his hitting is “not” good-enough for the corners (whatever that means).  What do teams do with this guy?  Plug him in a hole, move him team to team, let him languish in the minors, etc.

Endy Chavez has the far superior glove, so it’s easy to see how he can last longer.

So, there’s plenty of good-fielding, average-hitting guys in CF who won’t see much action in LF because their hitting would be “exposed” (from management’s thinking).  It takes someone of Chavez’ glove to be able to live with that kind of guy in LF.

But, a run is a run is a run.  To think in terms of needing minimum threshholds is ridiculous.  But, that’s what those teams are doing.

Look at the dearth of fielding talent in LF among regulars:
http://tangotiger.net/scout/index6.php?prim_fld_cd=7

(Those games are as of mid-August).

And compare it to RF:
http://tangotiger.net/scout/index6.php?prim_fld_cd=9

LF is the new DH.

PO/G:  LF/CF/RF
09: 1.94 2.59 2.02
08: 1.90 2.59 2.05
07: 1.98 2.64 2.05
06: 1.93 2.54 2.07
05: 1.93 2.51 2.03
04: 1.87 2.51 2.07

Maybe a small increase in CF talent. Maybe.  But I don’t see the rest of the story.

Note that LF is filled with backup players. So, at the very least, you should break down your numbers between starters and backups.

Actually, I already did that, with UZR, a year or two ago (I’ll look for the thread in a sec).  And LF was the ONLY position where the regulars had a worse UZR than their bench players.

Here’s the thread I was remembering, in plays per 162 GP:

http://www.insidethebook.com/ee/index.php/site/comments/replacement_level_fielding/

Here are how the regulars fare relative to the bench player:

1B: -0.7
2B: -2.1
SS: +4.1
3B: +11.1
LF: -3.4
CF: 0.0
RF: +3.8

LF has a disproportionate number of regulars who are hitters-who-can’t field, and their backups who are the opposite.

RF backups are worse fielders than the RF regulars.  Considering that there is severe overlap between backup RF and backup LF, we can see how we might think there’s a huge gap in fielding between regular RF and regular LF.

It doesnt matter whether you use 7/1 or 1/1 as seasonal age as long as you’re consistent. I would prefer 1/1 since it’s easier to compute.

I am not treating age as discrete; all of the studies on improvement are doing it.

So, all of the information you are going to use to calculate improvement are.

Does anyone compute future player projections based on whether someone is an old 23 year-old versus a young 23 year-old?

I “know” there is not a stagnant development curve, it’s very apparent in youth sports where there is an age cutoff.

You are using the age to determine the amount of improvement, which you/everyone have/has in 1 block increments.

"It doesnt matter whether you use 7/1 or 1/1 as seasonal age as long as you’re consistent. I would prefer 1/1 since it’s easier to compute. “

I already said:
“Yes, all the numbers are relative, so it doesn’t matter in the aggregate. “

***

“Does anyone compute future player projections based on whether someone is an old 23 year-old versus a young 23 year-old? “

Well, they should.

***

Regardless, my point stands that averaging 27.6 with 28.4 and showing the results as 27.5000 is wrong.  How do you say that in English?

“The average seasonal age is 27.5000, which is really 6 months lower than the actual average.”

***

I will keep stipulating that all the numbers are relative.

***

OPS is also equally indescribable as is “average seasonal age”.

Averaging 27.6 with 28.4 and showing the results as 28.00 is even wrong. And I hate it when I see it.

We are not using age as numbers, they are brackets.

It’s not the numbers, it’s what you are using them for.

You really don’t care about the age of a player, you care about how much improvement there is.

Maybe I should take a crack at relative age.

Say you get 1% rebate if you buy up to $10,000 and 2% rebate if you buy over $10,000

Year 1 you buy $9,000 you get $90
Year 2 you buy $11,000 you get $220

You buy $10,000 a year on average, but you don’t get $200 on the average.

You are correct that you need to weight the rebate rates (as you would the development rates).

If, for example, the development rate was 5% for those aged 22.000 and it was 2% for those aged 23.000, and -6% for those aged 24.000, and you had an overall average age of 23.000, the average development rate is not 2%.

But, what is that an issue here?

The ONLY thing I’m saying is that showing a group of players as averaging “27.5000” years is wrong because:
1. the significant digits
2. the value itself is not “age” as we understand it, but the average of the floor of the age

There is no such thing as an average age of 27.x, if you’ve already decided that “age” can only be expressed as an integer (and at that, the floor of the integer).

In what other discipline would they allow you to take an average of a series of already-truncated numbers?  Not only that, but allow you to report that number as a non-integer?  Other than “age”, this doesn’t happen anywhere.(*)

The only thing we are arguing over is the things we agree on!  Yes, it’s all relative anyway.  No, you can’t apply the development rate to the average of the group, since the development rate is not linear.

(*) I even argued with an actuary about this, when he set the rates for people at 69 1/2 years old.  Why? I said.  Why don’t you set the rate a bit lower at age 69, or a bit higher at 70?  Why do you want us to calculate the person’s age at 69 1/2 to set the new rate?  But, I had to program it the way he wanted it.  And of course, now customer service has to spend time talking to the customer to explain how their “seasonal” age triggers the new rate on Sep 4th, even though they were born Mar 4th.

Oh, and don’t ask me about leap years, and people born on a leap day.

My new all-time favorite stat was when I looked up to see if any Canadien in hockey had matched or exceeded Bill Russell’s 11 team championships.

Henri Richard tied Russell with 11 team championships. He was born on Feb 29, so he had won more championships than he had birthdays.

Going back to the CF issue....

I don’t see what the starter/backup distinction is telling us.  If LF starters are worse fielders than past LF’s, and RF features a stellar group of starters, then that should show up in the overall position stats (since their backups are often the same players).  But there has been no change in the distribution of plays made per game:
1999/2009
LF:  1.95 / 1.94
CF:  2.58 / 2.59
RF:  2.04 / 2.02
How is it these LF slugs are making the same number of outs as LFs did a decade ago?  And why aren’t all these “CFs” playing in right making any more outs than past RFs? 

On offense, the RFers ARE slightly better than LF this year:
LF: 107 OPS+
CF: 100
RF: 111

Overall, I don’t see much evidence that we have an unusually good cohort in CF or RF, or an unusually bad group of LFs.  I think this must be a case of a few standout players (both good and bad) shaping our perception of larger groups.

What we DO have now is much better hitting middle infielders than in the past.  2B now provides league average hitting (OPS+ of 101 and 100, this year and last), while it averaged about 95 a decade ago.  SS is also up but less dramatically:  an OPS+ of 92 now, vs. about 89 a decade ago (and that was the heyday of ARod/Nomar/Jeter).  Unless there has been a loss of defensive talent at 2B, it may be that MLB teams recognized the inefficiency Tango has talked about before—that 2Bmen were no better fielders than 3bmen, but worse hitters—and upgraded offense at 2B.

Perhaps we’re also seeing some general reduction in offensive variance?  As SS and 2B have gained ground, both LF and RF provide less offense than ten years ago.

I don’t know when the LF fielding talent started to go away.  But, looking at total outs made won’t necessarily help us, unless we also know how many balls are being hit there.  Not to mention that if you hit balls the opposite field more, those might be better candidates for hits.

All to say that, you can’t presume that the spray is the same.

going back to the age issue…

I read the link and understand what was said

However, it’s not just that Sean at br does it that way.

As I have detailed elsewhere, I got started in stats in sandlot baseball. Little League is as far as I know the largest sandlot organization, but they all assign players by age. LL uses Aug 1 as their cutoff date, their Senior Division goes up to 15. I’ve did a couple seasons of Pony Baseball, and as far as I recall they were also Aug 1, and then the AAABA and all it’s member leagues (up to age 20) use Aug 1. If you are 21 on July 31, you can not play in AAABA. If you are 21 on Aug 1, you are good.

So, for continuity, I use Aug 1, as it seems to be the consensus among amateur organizations where age does matter.