THE BOOK--Playing The Percentages In Baseball

So 32 percent of the major leaguers performed worse than the best free agent, had he performed at his true talent level?
Either the replacement level is set too high, or we have too many employed players who are really close to replacement level (unless perfomance variation is extremely wild)

Of course you also need to account for sample size, because PA and talent are negatively correlated.  It is the players who are closest to replacement level that will have the smallest sample size, and so of course a bunch of them will perform below replacement.  If you look only at hitters with 250+ PA, only 35 of 321 had negative WAR, and only 7 of those were -1 or worse.  Among pitchers with 70+ IP, only 15 of 215 had negative WAR (and only one who was worse than -1.)

I think this is the single biggest disconnect between saberists and JC/Berri:  understanding athlete performance as a sample of true talent.  JC just doesn’t get this, or perhaps I should say doesn’t agree.  But to me, it’s just a huge blind spot.

Is he still arguing against replacement level as a useful valuation concept?

It’s also not really 32% of a twenty-five man roster, as suggested in the passage. It’s 32% of players that appeared on a major league roster (I THINK- I didn’t check, but that makes the most sense) some of whom were career minor leaguers and rookies, and thus not expected to perform much better than replacement anyways. I’m not really surprised, and it doesn’t seem like an inefficiency.

So 32 percent of the major leaguers performed worse than the best free agent, had he performed at his true talent level?

That’s beautifully put.  You put all the guys who DID play MLB in one pool, and look at their OBSERVED performance.

You put all the guys that did NOT play MLB in a second pool, and look at their true talent level, and presume that that is also what their observed performance would have been (no random variation at all).

Compare the two pools, and you have 32% of players in one pool below the line set by the players in the other pool.

***

Guy: Substantial portions of his book is available at Amazon’s Look Inside.

In his player valuation, there are (at least) three problems I have found.  I am trying so very very hard to ignore commenting because of my potential bias, so I am hoping that you bright guys can spot the obvious mistakes and report on it.

re: comment #1

Of course there are many MLB players who are near the replacement level. For the most part there is very little difference in ability between the better minor league players and the typical MLB non-regular. And just about every MLB team has a few positions where the “regular” is essentially a replacement level talent. That is not due to lack of talent, but a natural result of talent distribution. Players with MLB abilities are rare, but as you move down the talent spectrum, from superstar to all star to above average (and so forth), there are more and more players who fit that description.

Tango:  I think I’ll be lazy and just wait for Phil Birnbaum to write a review!

Was JC making his point in an effort to claim that our replacement level is set too high? If so, wow. How could a PhD economist make that mistake?

Looking only at batters (defined as Fangraphs 2009 batters tab without a PA minimum and removing anyone with a “0” positional adjustment—any pitcher who didn’t play a non-pitcher position) I get 299 “full-time equivalents.  That’s defined as the total replacement runs divided by 20.  Makes sense: 10 per team.  Each team probably needs about 10 guys with 600 PAs each.  Or close enough.

Anyway, adding up all the “full time equivalents” below 0 WAR, I get 40 (just over one per team).  Above 0 WAR?  260, or just under 9 per team.  In other words, the 32% number is REALLY misleading.  It’s more like “13% of batters’ playing time was allocated to below-replacement-level performance.” 13 vs. 32, yikes.

"J.C. Bradbury is the preeminent analyst of baseball economics in the world.” Tyler Cowen, George Mason University and marginalrevolution.com—on book’s Amazon page.

I really enjoy Tyler Cowen’s blog, but now I’m worried he’s corrupting me.

What IS Bradbury doing with batting WAR for pitchers when he comes up with these numbers? It looks like he’s discarding them entirely, which is probably the kindest thing he could have done in making his point. (I still don’t think his point is correct, but still.)

"How could a PhD economist make that mistake?”

Dave, that’s sarcastic, right?  :>)

“J.C. Bradbury is the preeminent analyst of baseball economics in the world.”

Well, he also ranks 2nd.  And 3rd…

Why is that Dave? Isn’t the performance necessary to reach replacement level performance, based on a formula with a lot of assumptions?
Tango says “For nonpitchers, that level is set at -2.25 wins per 162 games, below the average for that league.” Is that set in stone?
The positional adjustments can’t be changed?
I think it’s entirely possible that the performance needed to reach replacement level performance is high.

Well, he also ranks 2nd.  And 3rd…

Not really. I can come up with a few others, all of whom I’d rank higher - Fort, Krautmann and Zimbalist at least.

Argentinean, the ‘mistake’ I was referring to was not whether replacement level might be set too high (it is certainly possible), it’s that JC looked at only a limited sample performance, where many of the sample had few PAs…

OK, Colin—was just making the point that it’s a very small field. 

I don’t think of Fort as specialized in baseball especially.  I don’t know Krautmann—what has he written about baseball?

Krautmann has written this, for instance:

http://onlinelibrary.wiley.com/doi/10.1111/j.1465-7295.1999.tb01435.x/abstract

It’s a pretty good paper, about free agent value. There’s others I haven’t had time to read.

I think my larger point is that there are academic economists writing about baseball, and some of them really are worth reading.

To amplify Tango - I’ve skimmed some of the sample chapters of the book. I don’t think Bradbury has materially improved his approach to player evaluation. I’ve resisted getting the book and reading it, as I have other priorities and frankly I don’t see anyone else taking it seriously, so I don’t feel a strong need to talk about it. Anyone likely to listen to me on the topic probably already has strong doubts about Bradbury’s credibility on the matter.

I keep re-reading those pages 93-97.  Aside from the errors in understanding sampling, I can’t figure out what he’s trying to say - it seems like he’s disputing the notion that “replacement-level players should be valued at no more than the league minimum.”

How can you dispute a tautology?

Wow, that passage is horrible.  Just horrible.  I echo the, “How can he make such an elementary mistake?”

Unless our reading of that passage is somehow wrong due to it being taken out of context or some such thing, I will have to change my opinion of JC from a 3 to a solid 0 (on a scale of 1-10). Seriously.  That is just horrible.  Regardless of what his point is, he is clearly suggesting that those players are replacement level in talent.

I said there were at least 3 major problems.  Let me give you the first one: trying to not have a flat marginal win/dollar relationship.

His chart on page 76 captures it.  Personally, I think you can *easily* draw a straight line relationship in his own data.  I mean, how much different could the correlation be between his curved line to his data points and what my straight line to his data points?

Anyway, by forcing his line as he does, he guarantees that the total revenue will be no lower than around 160MM$, regardless of how bad a team you have.  And, since he extrapolates team to player, this means you can have a team filled with the worst players in baseball, and he still has to somehow get to 160MM$ in revenue.

To do that, he has to give alot of value just for appearing on the field, and limits the value for each marginal win.  Hence, the Jeff Francoeur = 12MM$ conundrum.  In order to fit a player who performed as the worst player in MLB, but who managed to play every game (what a fantastic test case), JC would value his performance far more than any of us, or MLB, does.  It’s the only way he can fit his model.

Anyway, because of this, JC boxes himself in. 

That’s issue #1.  I’ll talk about other issues later.

Tango, is that the image with five outlier points in the upper right corner?  That graph covers five years, so I’m guessing those are the five Yankee points.  Yankees, well, I’d certainly accept them as outliers.  Removing those points, it’s really hard NOT to draw a linear relationship between run differential and revenue.

[How does one view any pages between 8 and 237?  I could earlier today, but now I can’t.  Is there a trick?]

MGL: that quote is on page 96 of his book.  You can read that part of the book, in context (pp. 92-96), via Amazon’s Look Inside.

Sky: do a search for value or players or baseball.

What impact does JC’s non-linear model have?

First, let me tell you what his equation is, as per page 79:

1. Take a player’s runs above average, and multiply by 0.0641
2. Take his runs above average squared, and multiply by 0.000979
3. Take his runs above average to the power of 3, and multiply by 0.00000312
4. Add those three figures, and you get the dollars were above average.

Now, without a concrete example, we have no idea what all that does.  For you standard star, +40 runs above average, the figures from above come out to:
1. $2.6
2. $1.6
3. $0.2
4. $4.3

(All numbers rounded).

So, the average star makes 4.3MM$ above what the average player would make, if they both had the same playing time.

Here is the chart for +80 runs to +0 runs:

Runs Total
80 $13.0
70 $10.4
60 $8.0
50 $6.0
40 $4.3
30 $2.9
20 $1.7
10 $0.7
0 $0.0

Remember, those are dollars above average.

When you look at his chart on page 84, we see Pujols is 13.70MM$ above average, and Sizemore is +2.83MM$ above average and so on.  Basically, I think I did this right.

JC is suggesting the following:

A. If you have one player at +80 runs and seven at league average, the team will be +80, but dollar value will be +13MM$ above average.

B. If you have eight players at +10 runs each, the team will still be at +80, but the dollar value will be $7MM$ above average.

You see the danger about trying to make the wins non-linear?  EVEN if you can make the case at the team level that you should do it, you CANNOT make the case at the player level, without going against some obstacles.

Sky:  I had to sign of with Amazon account to search.

And oh my.  One reason JC rejects replacement value is because he believes baseball talent is normally distributed around the MLB mean.  JC concludes that there is not a large pool of replacement players.  Instead, there is a steep curve at the bottom of the talent distribution, and the alternative for a team to even a low-talent player is someone much worse. 

He shows a histogram of OPS to back this up.  And of course it’s true that there are more .750 OPS players in MLB than .690 players. But it seems not to occur to him that this is a function of defensive value and roster construction.  The reason there are few .690 hitters is that the only way to do that and play in MLB is you have to have tremendous defensive talent (probably at C or SS), which is itself scarce. 

If we knew every player’s true WAR/game, I have no doubt whatsoever we would see a nice pyramid, with more 0-1 players than 1-2, and more 1-2 than 2-3, and so on.  But JC really seems to think that .690 OPS is a more rare skill than .770.  Really astonishing.

In other words, Bradbury rejects the argument that there’s a sizable talent pool of those types of players in AAA, because he only looks at major league stats… incredible.

Wait, I’m looking at page 94.  JC’s really showing a histogram of MLB performance to argue that replacement-level is *scarce*?  Seriously?  Does ANYBODY think that it’s easier to find league average players than bad players?  Wow.  (Does JC make an attempt to measure the production of AAA players anywhere in the book?)

Nice aging study, too…

Colin/26: well-said.  Perfect in one sentence.

***

This is a good way to think of talent:
http://www.tangotiger.net/talent.html

Basically, JC is observing Chart 6, when we really are meaning to talk about Chart 2.

Tango/24—that’s all money relative to the average player.  Anybody see where Bradbury says how much the average player is worth?  (Trying to back out his “replacement level”.)

Yes, and this is another problem.  P.82.

He starts with 96MM$, and divides that between pitchers and nonpitchers, with pitchers getting 35MM$ (36.5%).  He uses a similar logic to Bill James, and you already know how wrong I think Bill is there.  My split for pitchers is about 43%.  Coincidentally, MLB pays its pitchers 43% of the payroll.  Again, economic theory has to at least either support reality, or show us why reality is wrong.

Anyway, with 61MM$ to nonpitchers, that means each player starts with about 6MM$ (if they have 600 PA).  There’s some positional adjustment as well.

Therefore, to get to the minimum salary, you’d have to be around -46 runs (!!).

Another reason for JC’s confusion on talent distribution is his not recognizing random variation. If teams set any minimum talent level, you will see a large cluster of players just above that level, and then a smaller number below.  The players below that level are a mix of 1) those near the cutoff who got unlucky that year, and 2) a few below the cutoff whose true talent was misdiagnosed. 

Let’s look at frequency of starting pitchers’ FIP (2009):
2.00-2.49 2
2.50-2.99 4
3.00-3.49 13
3.50-3.99 31
4.00-4.49 44
4.50-4.99 49
5.00-5.49 27
5.50-5.99 6

Obviously, it’s not harder to find a 5.25 ERA pitcher than a 4.75 pitcher.  Teams are basically drawing the line at around 5.00, but then are trying out a bunch of marginal pitchers and hoping they succeed.  Some don’t, and if they don’t improve soon will lose their jobs.  You would see the same pattern if MLB eliminated 12 teams and replacement level became 4.25, or wherever the cutoff is set.

This really clarifies for me why communication with JC fails so often.  He just doesn’t share a lot of our assumptions about how the world works.

Put another way, JC has both of these players as equals:

-20 runs, 600 PA = 4.3MM$
+15 runs, 300 PA = 4.3MM$

If that second player had 300 more PA with -35 runs of performance (that’s like, very very very bad… not sure if it’s ever been done), then he’d have no extra value over the other guy.  But, if he was -25 runs in those 300 extra PA?  Yup, then he’d be better.

Look, it’s so easy to find the examples where you are going to break a non-linear system of wins to dollars.  That’s why you should not do it.  It’s nothing but a headache, you are trying to solve for too many things.  And ultimately, the linear method is not only close to reality, it is incredibly simple to calculate in your head.

"Nice aging study, too…”

Sky, we had a long thread on JC’s aging study (at least one) a while back.  Probably 6-12 mos. ago.
IIRC, MGL presented some interesting aging analysis there.

"Therefore, to get to the minimum salary, you’d have to be around -46 runs (!!).”

Francouer could do it! 

Seriously, at least this model puts a replacement player at $4.3M instead of $12M.  That’s progress.  Still, it looks like JC has a 6 WAR (+40 runs) player valued at around $10M.  Is that right?

That’s what’s so weird about his model.  If your players are +/-40 runs around average, it is very close to a linear model: 1 win = 1MM$.  Since virtually all players are in this range, it seems… excessive… to do the model he presents.

JC’s model is for all players, not just free agents, right?  A rep level model might be something like $2M-$2.5M per win combining free agents and pre-arb players, so that’s $12M to $16M for a 6 WAR player.  Not awful.

Guy, my reference to the aging study was tongue-in-cheek.

I haven’t gotten to that part yet.

But, correct on your sentiment.  In my case, I have about 33 WAR for about 80MM$ marginal, or 2.4MM$ per win (FA, arb, slave).

So, right, we shouldn’t compare it to the 4MM$ to 5MM$ that we talk about for free agents, but the 2.4MM$ or so we talk about.

Actually, it looks like he used 96MM$ per team for total payroll, so that would mean about 86MM$ marginal dollars.

Anyway…

One obvious way to test if GMs are irrational, one that doesn’t make you retroactively adjust for luck, is to just look at the Marcels for that season.

I’ll try doing that tonight or tomorrow.  Can anyone estimate for me what batting replacement level is in terms of an easy-to-calculate stat, like OPS or (SLG + 1.7 * OBA), or something like that?

Phil: replacement level will be different by position.

You might as well stick to OPS+ so you can use b-r.com, and then take 75% of OPS+ of the positional average as the repl level. 

So, about 70 for SS/C, 75 for 2B/3B/CF, 85 for 1B/LF/RF.  Something like that.

Or, use the WAR in the Play Index to help you out.  You can’t ignore fielding.

OK, I’ll use 75% of the position average.  I don’t have WAR in my database and it’s probably too much work to incorporate it just for this.

Thanks.

Looks like Amazon removed most of the free pages.

Tango—my access to the free pages has waxed and waned today.  Try again tomorrow, maybe using a different Amazon account.

I remember JC making a comment in the book about teams not settling for what others consider replacement level talent because it’s so much worse than what you can get for as little as a million or two.  Well yeah, when you think rep level is -46 runs, no way you settle for that.

JC has Jayson Werth being worth $16-17M per year over 5 years: http://www.sabernomics.com/sabernomics/index.php/2010/10/what-is-jayson-werth-worth/.  I don’t know if that’s “right,” but it’s not far off.  So he must have a free agent model that generates plausible results in at least some ranges. 

At the same time, his age analysis is all screwed up, as you’d expect given his aging curve work.  He estimates that Werth will only lose about 6% of his value over a 5-year contract, starting at age 32.  I’d guess that 32-year-old players lose at least 25% of their value over those ages, and maybe closer to 50%.

I’ll tell you why it “works”, or gives normal numbers.

He starts off with a low win value, but then applies almost no aging!  And on top of that, will give him a pretty stable number of PA.

While on my side, I apply a more reasonable single-year win value and heavily age him (in PA) and moderately in terms of talent level.  The sum total, over a 4-6 year period, makes us both similar.

Basically, his two wrongs makes my one right.

He says replacement level players are scarce...so, if the league doubles in size, then we should expect the average talent level to DRAMATICALLY fall.  Or if the league size cuts in half, we would have an abundance of replacement players.  The league just happens to be exactly the right size that we’re just to the left of this spike in the talent distribution.  Yikes…

If you doubted what JC meant:

http://www.sabernomics.com/sabernomics/index.php/2010/11/hot-stove-myths-again/

Replacement Players are Cheap and Abundant — Why do you think that a third of the league is composed of “below replacement” players?

Aren’t just by the laws of statistics 50% of “true” replacement level players going to end the year with a negative WAR?  Shouldn’t their spread in possible observed value be something like a bell shaped curve, with a portion of that curve being under replacement level (for almost all players)?

So how would we know if a players “true” value was really below replacement level or not, just because his “observed” value was?
vr, Xei

Tango @46,

If JC here is saying that replacement level players are abundant, doesn’t that contradict what we’ve seen him indicate in his salary modeling in a more recent thread here?

re 48 -

I believe there is sarcasm in JC’s sentence, let me requote

Other myths in Hot Stove Economics:
Replacement Players are Cheap and Abundant — Why do you think that a third of the league is composed of “below replacement” players?

It’s not the number of players, but how many PA/IP do they get (On the Pirates, more than other teams)

Most of the replacement players reside in AAA, floating just below the surface. Each MLB has 15 players on the 40 man roster reassigned to the minors, and another 180-200 players on minor league contracts, all of these in the minors subject to recall at any time, to replace someone on the 25 man roster. In the next week or so 500-600 players (6 year pros left off 40 man rosters) will became minor league free agents, to be signed to minor league contracts and settling in to new AAA and AA teams next year. The definition of freely available talent. Occasionly a guy like Garrett Jones will catch on as a MLB regular, but after 1000 PAs prove himself to be the 28th to 30th best 1b in MLB. Replacement level.

It also depends on where you set replacement level. If you measure actual payroll, then marginal salary is fixed - it’s the sum of all team’s payroll minus (30 teams times 25 players per team times minimum salary). Setting replacement level clsoer to mean will decrease the number of marginal runs, and thus increase the cost of each run - a fixed number of dollars divided by a smaller numbers of runs, and then also increase the theoretial maximum salaries in the model. Setting replacement lower, further from the mean will decrease the cost of each run, and lower the top salaries. Replacement could then be estimated by finding a best fit of projected vs actual salaries.

Brian,

Thanks, my sarcasm detector needed sharpening.

Sky is trying to engage him:

http://www.sabernomics.com/sabernomics/index.php/2010/11/hot-stove-myths-again/comment-page-1/#comment-110540

I don’t think this quote is a fair assessment:

“Replacement Players are Cheap and Abundant — Why do you think that a third of the league is composed of “below replacement” players?”

One, it’s more representative to say that about 15% of plate appearances come from players who performed worse than replacement level (as you note in your book.)

Two, because there are so many players with talent levels near replacement level, there are bound to be MANY who underperform their talent level. In other words, while their production (in very limited playing time) is below replacement level, our best estimate of their actual talent level is NOT below replacement level.

Any thoughts on those, especially the second?

The question is worded fantastically well.

Yes, great wording on the question.  Clear and concise. I eagerly await JC’s response.  The first part is easily answered:  15% of the PA is true, as is 30% of the players.  He (JC) is NOT going to admit that he was being misleading when he said “30% of the players”, even though, while true, it is clearly misleading, especially when your point hinges on the size of the number (and 30% is a lot higher than 15%).

Another specific question that Sky or someone else can add is, “Given that 15% percent of all PA from around 30% of the players (however many PA per player that comes out to) are below replacement level, and given what we know about random fluctuation in finite numbers of PA around a player’s true talent mean, how may of those players would you estimate are actually true talent below replacement players?”

On the other hand, if we are going to go that route (that observation does not equal true talent), how many of the above replacement players are actually below replacement and got lucky?

Maybe the total IS 30% of all players. I don’t know.

So the real question is, “Given the entire gamut of observed performance, below AND above replacement level, how many players do we estimate are true talent below replacement level players?”

Well, he did reply and said this:

“I don’t recall writing that replacement players take 15% of all plate appearances. And if I had, I’m not sure why it would be more representative or fairer than what I wrote above. I did write that 16% of players supposedly cost their teams $1 million or more, which was in anticipation of the argument that the negative values could be explained by expected natural deviations in performance from replacement players (p. 96).”

I’m not sure JC understood the question. Or wanted to answer it. Sounds like he confused the “15%” with his own “16%”, which comes from somewhere else. Probably Sky should have clarified where he took 15% from (JC had 32% instead, correct?).

It was my fault on the 15% thing—he’s right about what he wrote.  Working on a response.

Main point is that $1M really isn’t that much in Fangraphs terms.  About 3 runs below replacement level, which is easy to hit in 25 or 100 PAs given a true talent level at or a little above replacement level.

Sky: you have to ask a very specific and limited question, like a lawyer.  That’s the only way to get a straight answer from him.

Am I doing this right?

SD for wOBA = SQRT(wOBA*(1.1-wOBA)/PA)
and
runs difference = (obs_wOBA - talent_wOBA)/1.15*PA

For a .300 wOBA in 100 PAs, the SD is .049, meaning 3 runs is a .266 wOBA, and .7 SDs below the true talent level.  Being less than .7 SDs below the mean happens about 26% of the time.

The 1.15 is not a hard-and-fast rule.  It’s probably closer to 1.20 these days, not that it’ll change much the end result.

I don’t know what this means: “meaning 3 runs is a .266 wOBA”.  It would have to depend on the baseline wOBA and number of PA.

Ok, I think I understand.  You start with a true talent level of .300 because that’s where replacement level would be.  Got it.  That comes out to 2.83 runs using what I said.  So, instead of .266, you would have a wOBA of .264, which would be 0.73 SD below true talent level.

"I’m not sure JC understood the question. Or wanted to answer it.”

That’s always the question with JC, isn’t it?  (I suspect the answer is usually “both.")

Sky, it’s your time of course, but I think you’re on a quixotic mission here:  JC’s not going to change his mind on this.  He has derided the whole idea of replacement level too many times.  Look at the sarcastic tone of his comment:  “Why do you think that a third of the league is composed of “below replacement” players?” It’s not just that he thinks he has a somewhat better analytic framework, he has publicly committed himself to the idea that replacement level is an idiotic idea.  (Which I guess it could be, in a world in which 4.30 ERA pitchers were more abundant than 5.50 pitchers.) There’s no imaginable face-saving way for him to move off that position now.  (In contrast, changing Francouer’s value from $12M to $3M can is relatively easy, especially if you don’t acknowledge your critics). 

Look, a couple months ago Phil presented unambiguous evidence that JC’s entire pitch count analysis was mistaken, and was based on an embarrassingly-obvious methodological error.  JC’s response was to double-down and continue promoting the paper (he has linked to it a few times since then).  If he couldn’t bring himself to admit error then, there’s no way he can do it here.

I’m actually sympathetic to JC’s underlying idea, which is that teams generally act rationally.  I agree with him that some saberists are too quick to conclude MLB decision makers are wasting vast sums of money on stupid decisions—many of these claims involve bad analysis, not a mistake by decision-makers.  The larger the inefficiency an analyst thinks he’s found, the more carefully he should review his work.  But here JC is just trying to score debating points:  “Teams waste at least $1M on 16% of their players.  Ha ha. Like that could happen.” And it does sound implausible.  UNTIL you realize this means that a bunch of players were just 3 runs below replacement! And that this includes both unlucky replacement-level players PLUS a few mis-evaluated sub-replacement guys.  So of course there will be a bunch of -$1M guys every year, even if there are a ton of replacement level players. 

(Interestingly, JC is a huge fan and supporter of Dave Berri, who makes much greater claims than any baseball saberist about incredibly massive inefficiencies in basketball.  If you believe Berri, NBA coaches and GMs could improve just by throwing darts at the board.  I don’t know how JC squares his admiration for Berri with his own belief in mostly rational sports outcomes.)

There’s no imaginable face-saving way for him to move off that position now. 

I’m reminded of your other post:
http://www.insidethebook.com/ee/index.php/site/comments/warp/#4

I had said this, in criticizing Clay’s replacement level:

The really strange part is not that he sticks to his guns, but does so in the face of everyone else who studies the issue.  And without a very convincing argument.

Replacement-level has been defined generally speaking as a team with a .300 win level.  If Clay insists on using a .150-.160 level, then he’s got to use a different name.  Something like “WAA” or “WADA” for “Wins above AA”.

And you (Guy) said this:

I’m not surprised:  BPro has gone too far down the WARP road to now just tell the world, like Emily Litella, “Never mind.”

Except this is exactly what Clay did in the BP annual (2009 I think).  He did say he was wrong.

And, BPro went further in hiring Colin, who is pretty sympathetic to the views espoused in this blog (a blog that has taken a strong stance against several BPro stats), and is in the process of cleaning up all the holes in the BPro metrics. 

That was pretty big of them.

***

JC should learn that even if he is smarter than each one of us, there’s no way he’s smarter than the collective of us.  And if he’s not smarter than the collective, then he better have figured out something that the rest of us missed (possible if improbable).  Until he does that, he will always lose his arguments, and he will remain wrong.

For example, the ten thousand of us are not smarter than 10 million baseball fans who believe in RBI, but we did figure something out that they didn’t consider appropriately (the impact of number of runners on base having a highly biased effect). 

What’s JC’s argument on replacement level?  Faulty logic.

Tango:  Man, how did you remember that?  But point taken.  I don’t know Clay or other senior BP folks at all, so I shouldn’t have made an assumption about how stubborn he would be.  My bad.  And BP does deserve a lot of credit for the changes they’ve made. 

That said, I do feel pretty confident in this prediction, at least in the short-term.  JC has demonstrated a strong animosity toward saberists on numerous occasions, and is committed to trying to do things differently (even when saberists already have reached a close approximation of the right answer).  He and Berri have both written a lot about how inferior the work of us “amateurs” is to what they do, and Berri too never admits error in public.  So I think it would be extremely painful for JC to reverse course on the replacement issue.

Maybe in a couple years JC will develop still another valuation model that implicitly accepts replacement level (as he changed his French valuation).  But I’d guess he will never admit he was wrong, and for sure will never acknowledge that we were right.

"Look Inside” isn’t working for me right now—anyone remember how JC adjusts values for relievers?  Is it in his talent estimate or the conversion to dollars?

I ask because of this: http://www.sabernomics.com/sabernomics/index.php/2010/11/2010-most-valuable-players-pitchers/

Matt Belisle is at $15M, the fourth most valuable NLer, just behind Ubaldo Jimenez.  7 out of the top 20 NLers are relievers.  I wonder if it’s again a lack of replacement-level issue—valuing a 3.50 ERA from a starter the same as a 3.50 ERA from a reliever or something similar.  But still, the lack of innings would make it tough to catch starters.  (Not that JC is using ERA—Joba makes the top AL pitcher list.)

Tango:  what is JC’s theory for why top salaries are higher than the MRP delivered by the best players even in their best seasons?  One would think that a player’s salary should match the average MRP delivered, which will obviously be quite a bit less than these leader tables show.  Given his numbers, it seems that no player other than perhaps Pujols could possibly have an expected MRP of over $15M/yr, and probably less than that.

Sky:

Oh. My. God.

He has Tim Lincecum’s 2011 value the same as his reliever (Brian Wilson).

How the heck are we supposed to take JC seriously?

I don’t think he has a reliever adjustment, so he likely treats a 3.00 ERA from a reliever identical to a starter.

I presume that he doesn’t include chaining, and so he gives Wilson a leverage of 2 and Lincecum obviously 1.

So, he’ll forecast say Wilson with a 2.25 ERA or something and Lincecum with a 3.25 ERA (let’s say).  Lincecum would be 1 run better than league average and Wilson would be 2 runs better per 9IP.  If we give Lincecum 216 innings and Wilson 72, that makes Wilson +16 runs above average and Linceum is +24 runs.  But Wilson has a leverage of 2x, so the effective runs is +32 for Wilson and +24 for Lincecum, or +3.2 wins for Wilson and +2.4 wins for Lincecum.

That’s just for wins above average.  You still need the base wins per IP, whatever that happens to be.

Anyway, Lincecum gets the advantage on IP, and Wilson gets the advantage on wins above average, and in JC’s math, the two work out to the same thing.

Hopefully people will see the ridiculousness of the results and treat his process as wrong.  Hopefully, his latest thread will be his wakeup call.

Sorry, that was 2010 value, not 2011 forecast.  That makes it slightly better.

He’s still wrong about not having the reliever adjustment.

But now it becomes a philosophical issue regarding the leverage (chain or no chain).

To be fair, $15M for 92 IP of a 2.68 FIP in Coors Field isn’t too crazy.  On the FG scale, that’s about 3.5 wins.

Of course, it’s crazy relative to the other pitchers’ (and hitters’ price tags.  Given that the top dogs barely surpass the $20M mark, I’m guessing these are done using league-wide rates (like $2.5M per win), not free-agent rates (like $4M per win).  Makes the top earners more “on”, but makes the relievers and mediocre guys “off"/overpaid.

Wilson had a FIP of 2.19 on 75IP, and Lincecum was 3.15 on 212 IP.  So, my illustration of their 2011 forecast was actually pretty much a match for their performance in 2010.

So, the issue is the relief adjustment, and chaining.

I think chaining needs to be done, but, as I said, this is a philosophical issue and if JC prefers to not chain, that’s ok.  The problem is that you end up with some absurd results at times. 

Given that you can make a good case either way (chaining, no chaining), I think seeing the results would pretty much sway someone to the chaining side.

Off topic, but JC has new paper out evaluating managers:  http://www.sabernomics.com/sabernomics/index.php/2010/11/how-important-are-managers/.

I don’t have a book on me, but can anyone summarize how JC goes from the team-level wins vs. revenue curve to individual player values?

He’s got a post up explaining his issues with the Fangraphs valuation methodology and has been quite engaging in the comments: http://www.sabernomics.com/sabernomics/index.php/2010/11/on-per-win-estimates-of-baseball-players-worth/

Sky, I just started a new thread on this.