Baseball for the Thinking Fan

Login | Register | Feedback

btf_logo
You are here > Home > Baseball Newsstand > Baseball Primer Newsblog > Discussion
Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand

Thursday, December 22, 2005

This Week’s Numbers at THT

Part 1:
This bit is pretty much what we used to call “Clutch Pitching”.  I think there’s plenty of research open in this area, if you can get past the sample sizes.

Part 2:
One problem I have with saying Furcal is worth $18 mil is that no one gets paid $18 mil, so there’s something wonky in the math=> interpretation “based on what MLB teams paid”.  There’s something off in that, but I can’t put my finger on it.

Part 3:
I forwarded that girlfriend link to my wife - it’s excellent.

Part 4:
That is sad, and I wish his family the best.

 

 

Other recent THT pieces I want to comment on, but don’t want to glut the links - Maybe I’ll do a THT blog, and critique the work weekly (and weakly):
Daily Graphing - I like this site, and had been otherwise, but I’m glad THT has highlighted it

Graphing Jacque Jones

I wonder if there is good research on a decline in LD% as one ages - I assume that is what occurs, but I don’t know for sure.

Graphing Heath Bell

Here’s where watching games matters - Bell leaves the ball up more in the majors - or did in ‘05.  Hitters adjust and wait on that mistake and drive it.  One of the problems with absorbing into DIPS is that you “forget” that second part - the hitter: if any pitchers throws balls in the middle of the plate, they’ll get ripped - Think of that 9-square chart of a hitter’s hot zones - Bell spent too much time in those zones.  If he doesn’t *adjust*, and that isn’t a luck factor, he’ll not get back.  In AAA, more players miss those pitches (as evident that the vast majority don’t make it in the majors).

Dan Fox (?)
Bunting

I have some work on this to publish here, but not with this conclusion (not an opposite conclusion - I was looking at something else).

But Dan hints that managers don’t bunt with “worse” bunters often enough.  I think the bunting skill isn’t as gaussian as one would assume - people who rarely bunt are particularly bad, and give up DPs at a damagingly higher rate.  I assert without proof.

More THT as the weeks come.

Chris Dial Posted: December 22, 2005 at 04:25 PM | 56 comment(s) Login to Bookmark
  Tags: sabermetrics

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.

   1. Der Komminsk-sar Posted: December 22, 2005 at 05:09 PM (#1790205)

Consider this a “bump”.

Chris, I agree with you fully on #2 - frankly, I’m not all that impressed with the salary estimators that I’ve seen, though I suppose I should whip out my own before complaining too publicly.

Bunting: One methodological question about Dan’s study - shouldn’t win expectancy, not run expectancy be used here?  Also, your hunch about bunting quality makes sense to me.

   2. Chris Dial Posted: December 22, 2005 at 05:16 PM (#1790217)

Stupidly, I changed the title of this - I need to mention those first few parts are about Studes top piece.  Guess I’ll edit this entry.

   3. studes Posted: December 22, 2005 at 05:19 PM (#1790223)

Chris, thanks for the THT link.  Much appreciated.

Regarding Furcal and $18 million, you’re right and I should try to refine my approach.  But my “system” (if you can call it that!) isn’t really a salary estimator, it’s a “fair market value” calculation.  Based on how much money was paid to free agents last year and what they contributed, Furcal is worth that much for one year, in my estimation.

I need to take this one step further by including a reliability factor (that factors in injury and aging) as well as a discount for multi-year contracts.  Then it will start to resemble more of a salary/contract estimator.

I still feel that, given his age, injury history and the fact that he only signed a three-year deal, the Dodgers didn’t overpay at all.

   4. Chris Dial Posted: December 22, 2005 at 05:24 PM (#1790231)

it’s a “fair market value” calculation.

I undersand it that way, and I disagree.  I htink you have something that makes it too high, because you may or may not be including “free players”.  I don’t know.

In your other article you make reference to someone being worth $23 mil.  Well, that’s bonkers because NO ONE pays that - except to ARod the MVP. 

I think you may need to change the equation of the line from linear.

   5. studes Posted: December 22, 2005 at 05:30 PM (#1790240)

I’m looking at just free agents, and not including other types.  If I made the equation logarithmic and followed the economics of baseball (where the 90th win is worth more than the 80th win) than excellent players would be worth even more.

The basic issue here is that baseball owners spend a lot of money on players like Jim Thome and Sammy Sosa (in 2005) every year and get virtually nothing for it.  When you include guys like that in your approach—and you should—guys like Furcal look that much more valuable.

I also agree that no one pays $23M.  Again, that’s not really what I’m saying.  There’s a distinction between worth calculation and salary/contract estimation.

   6. Daryn Posted: December 22, 2005 at 05:41 PM (#1790258)

I also agree that no one pays $23M. Again, that’s not really what I’m saying. There’s a distinction between worth calculation and salary/contract estimation.

That’s correct.  Every year there are a handful of players who, retrospectively based on the market, earned more than $20 Mil.  Due to the unpredictability of performance, that doesn’t mean anyone should pay $20 Mil.

   7. J. Cross Posted: December 22, 2005 at 05:52 PM (#1790283)

Players who made around $18 in 2005:

Jeter, Derek $ 19,600,000                
Mussina, Mike $ 19,000,000                  
Bagwell, Jeff $ 18,000,000                  
Clemens, Roger $ 18,000,000                  
Sosa, Sammy $ 17,000,000                  
Piazza, Mike $ 16,071,429                  
Jones, Chipper $ 16,061,802                
Johnson, Randy $ 16,000,000

   8. Der Komminsk-sar Posted: December 22, 2005 at 05:58 PM (#1790294)

it’s a “fair market value” calculation.
Sorry, I misspoke - I knew what you were getting at.

   9. Sam M. Posted: December 22, 2005 at 06:45 PM (#1790357)

In your other article you make reference to someone being worth $23 mil. Well, that’s bonkers because NO ONE pays that - except to ARod the MVP.

The fact that no one is paid that much doesn’t mean it’s bonkers to say someone is worth it, Chris.  It could represent a market failure or distortion.  For example, the very fact that there are only 30 teams when many more could be supported and profitable artificially limits the competition for the best players.  I would argue that the extent to which players are actually underpaid (e.g., Furcal by $5M in 2006) could represent cartel profits the owners are generating.

Putting that aside, though, the broader issue I have is the idea that there is one universal number that represents what a player is “worth.”  I don’t care whether studes has it right, or whether he’s too high, or too low.  I have a problem with the concept, as I do with MGL’s fairly rigid idea of paying no more than X dollars for each marginal win.*  The thing is, not all marginal wins are created the same (or have the same value), and a player doesn’t have the same worth to every team that might be thinking about signing him.  Furcal doesn’t have one, single “value.”  He has a range of values, depending on whether the team needs a shortstop, whether it needs a lead-off hitter, whether it is a contender or rebuilding, etc.

I guess what I’m saying is, I don’t really see the point in putting one single dollar figure on a player’s value.  To me, it seems at least as misleading as it is revealing.


* - In fairness, he’s stressed that he sometimes expresses that in more absolute terms (“I’d NEVER do X . . . .”) than he really means.  So he’s not as rigid about it as he sometimes comes across, but still . . . he’s pretty adament about it.

   10. J. Cross Posted: December 22, 2005 at 06:54 PM (#1790375)

not all marginal wins are created the same (or have the same value),

Definitely.  A marginal win to a team projected to win 86 games is worth a lot more than to a team projected to win 76 games.

There are a probably just a few contending teams (or teams who view themselves as contenders) who have a need to fill at SS it could be an oversimplification to use a “fair market value” when there’s one unique item to be sold and only 2 or 3 serious buyers.

That said, I think calculations like these have some value in determining which players give the most bang for the buck.

   11. Backlasher Posted: December 22, 2005 at 06:56 PM (#1790379)

  * - In fairness, he’s stressed that he sometimes expresses that in more absolute terms (“I’d NEVER do X . . . .”) than he really means. So he’s not as rigid about it as he sometimes comes across, but still . . . he’s pretty adament about it.

I think he also recanted a little on the Damon issue basically saying that the Yankees were covered in all but CF ....

I absolutely agree with you; nevertheless, I think studs efforts can be worthwhile.  The problem, which I think you allude to, is when people get adamant about the result, even if they leave out key variables.

IMHO, I would think as a baseline negotiation move, you have three to four values in your mind when you go after players:

Value 1:  Offer price
Value 2:  The studs number
Value 3:  The likely win number
Value 4:  Your maximum output

These numbers do not have to be distinct or linearly escalating.  Its possible for 1=2=3=4, and for 2>4, or any other such combination.  If its not somewhere along the line of

1 <= 2 < 3 <= 4, you might determine its not worth bidding.  And if you win with 3>4, you suffer the bidders’ curse.

So if I’m a GM, a would like to have an assistant that is going to help me with determining 1 and 3; I’ll probably take sole responsibility for 4; and I’d have some spreadsheet jockey to help me with 2.

   12. GuyM Posted: December 22, 2005 at 06:58 PM (#1790382)

Thanks for posting this, Chris—a lot of good THT material has gone unposted in last few months.

In other words, the ability to keep baserunners from scoring is a repeatable skill, to a degree. For comparison, that’s about the same correlation as a pitcher’s home run rate (home runs per game).

Studes: a question and comment on this.  Question: to some extent, a high LOB% just reflects good pitching.  Is there any evidence that pitchers have an ability to pitch better with men on base?  (i.e. to consistently post a lower BA/RISP than BA/none on?) 

Comment:  Several THT analysts use y-t-y correlation as a measure of much a stat reflects true ability, as you do here.  That’s not quite right.  Correlation essentially measures the ratio btwn the true talent variance and the amount of noise in the statistic (which in turn is function of sample size).  Using correlation to tell us what is a true talent would only work if A) one SD in every stat was equivalent in RS or RA, and B) sample size (and thus noise) was same for every stat.  However, neither is true.

A low y-t-y correlation can mask important skill differences, if a small percentage change in the stat translates into a lot of RS (or RA), and/or if the stat has a higher measurement error than other stats.  A good example is BABIP, which of course has very low correlation, but where a “small”  difference in terms of % of the mean (say, .270 vs. .300) has big RA implications, and where the sample size (BIP) is smaller than for the stats it’s often compared to (K/9 or BB/9).

   13. J. Cross Posted: December 22, 2005 at 06:58 PM (#1790384)

Most post above (#7) was meant to show that the players actually making $18M a year are no great shakes.  Granted, some of them signed during that last “boom” but Furcal has to be better than 5 of those 8 guys (or “projects better than they projected for last year” to be specific).  Furcal could be “worth” $18M in a one year deal but I don’t think he’s worth $18M/yr. for 4 years.  He’s worth more than Damon though.

   14. Chris Dial Posted: December 22, 2005 at 06:59 PM (#1790387)

When you include guys like that in your approach—and you should—guys like Furcal look that much more valuable.

Every year there are a handful of players who, retrospectively based on the market, earned more than $20 Mil.

But here’s where I suggest the flaw is - your baseline isn’t correct in teh estimation - well, if you want to use this info for determining a team over/underpaid.

There is some issue with what it is - I think. 

As for “retrospective salaries”, I don’t that is a very good way to determine a good contract.  You can’t, IMO, say “what would you pay for Ted Williams today?”

I’ve claimed before, and written that you can assemble a very cheap FA team that will produce WS at a lower cost than this.  THis analysis - does it skew heavily by big FAs coming off huge seasons and getting big contracts?  Take Beltran or for heaven’s sake, Beltre.  Because those guys got a lot, when you could easily play someone else and match that doesn’t mean a team didn’t overpay.  $13 mil is too much for a SS of Furcal’s caliber. 

You can find that production for a lot less.

   15. Backlasher Posted: December 22, 2005 at 07:01 PM (#1790392)

Definitely. A marginal win to a team projected to win 86 games is worth a lot more than to a team projected to win 76 games.

I agree, we talked about this once in another thread.  Marginal wins likely have a multi step utility curve.  Its pretty flat up to say 60 but linearly increases, changes slope to a higher increase in the mid-80s, peaks somewhere in the low 90s, begins a gradual decline thereafter, and a somewhat steeper decline in the high 90s.

(That hypothesis is based on the weighted average of years in the wild card era, and the utility measure is either playoffs or championships.)

   16. Chris Dial Posted: December 22, 2005 at 07:04 PM (#1790400)

Is there any evidence that pitchers have an ability to pitch better with men on base?

Guy,
this is what I was alluding to when Ireferenced “clutch pitching”.  In rsb a few years ago (7-10) it was regualrly observed that Cone and Ryan pitched exceedingly well w/ROB (I think, anyway).

I’ll have to be at home and check the back of TB IV.

   17. Backlasher Posted: December 22, 2005 at 07:07 PM (#1790409)

I’ve claimed before, and written that you can assemble a very cheap FA team that will produce WS at a lower cost than this.

Sure, if you maximize the free agent class, look backwards, and possibly play people out of position.

But the washouts in high dollars are likely a relatively fixed percentage.  I don’t know how much, but they are going to be offset by the reserve clause slaves giving good performances, and the free agent steals.  And both are available at a certain price point.

But the biggest thing, (and the one I want to go back and read) is the reliability in your projection.  That is the thing that kills most of these saberists in “prospect” deals.  They will project a Dan Meyer as an above average pitcher or a Drederick Barton as a HoF.  MLE’s are a hugely unreliable stat because they only select for players that actually make the ML and perform well enough to stay in the Major leagues.

So, I’m curious as to how reliable this Shandling number that studes cites.  I haven’t seen that before.  But that is the weighing criteria that would heavily push the numbers.  If your reliability is high, its worth more money.

   18. Der Komminsk-sar Posted: December 22, 2005 at 07:09 PM (#1790414)

#15 - Exactly. We use win maximization as a proxy for a team’s goals (presuming wins have a linear relationship with revenues), when it pretty clearly isn’t. [I’d argue that this, fundamentally, isn’t that different from the problems that a lot of studies on bunting suffer from.]

One question I have is: to what extent should we use cumulative stats (that measure value produced) versus rate stats (which proxy aspects of potential value)? Win Shares, with its ultra low replacement level, probably overrates players that spent a lot of time on the field.

   19. Backlasher Posted: December 22, 2005 at 07:14 PM (#1790423)

One question I have is: to what extent should we use cumulative stats (that measure value produced) versus rate stats (which proxy aspects of potential value)? Win Shares, with its ultra low replacement level, probably overrates players that spent a lot of time on the field.

That’s why I’m interested in that Shandling number.  As a baseline matter: rate will better predict performance; accumulation will better help with reliability.  I’ve never seen anyone really tackle the latter, other than murmur and sputter about sample size or noise.

You should be able to control for these things much better.  And you can certainly model that which the saberists call noise, metricize your reliance on your inference without using the clunky hypothesis tests, and even de-noise data.  You can also certainly determine better spaces to work with rather than throwing everything in one big sack (reference for Szym), but without cherry picking (Paging ACE FOS).

   20. Der Komminsk-sar Posted: December 22, 2005 at 07:19 PM (#1790438)

But the biggest thing, (and the one I want to go back and read) is the reliability in your projection. ... MLE’s are a hugely unreliable stat because they only select for players that actually make the ML and perform well enough to stay in the Major leagues.
This can be overcome - there’s a host of models that deal with self-selection/omitted variable bias issues - Heckit, for one. Of course, we’ll never isolate the scouting element from the process - a model which underestimates “worthy” players and overestimates “less worthy” ones isn’t the ideal either.

So, I’m curious as to how reliable this Shandling number that studes cites.
Without having done studies on his and competing MLEs, Shandler is a little more conservative than most, but still a skotch high, I think. He’s now considered the “industry leader”, fwiw.

If your reliability is high, its worth more money.
Sorry I’m confused - do you mean of the estimate or of the player (consistency of performance, ability to remain free from injury).

   21. Der Komminsk-sar Posted: December 22, 2005 at 07:23 PM (#1790445)

Sorry, I referred to the wrong Shandler numbers (you were referring to the Furcal comment, which is obvious in retrospect).  I’ve seen a few parties post reliability scores on their estimates, but never an attempt to measure how valid those scores were after the fact.  Anyone?

   22. Chris Dial Posted: December 22, 2005 at 07:32 PM (#1790462)

Sure, if you maximize the free agent class, look backwards, and possibly play people out of position

I disagree - it isn’t looking backwards - at least not what I am talking about doing/have done.

I’m not saying “teh Mets could have gotten so-and-so and he would have been better”, I am saying the Mets should sign so-and-so and he’ll outperform who they do play.

Take Josh Fogg.  He’s been non-tendered.  He’ll be better than Victor Zambrano for significantly less money. 

I suspect the Dodgers could have gotten Kaz Matsui to play SS for $7mil and he’ll perform about as well as Furcal.  A little worse, but not (how much is a marginal win? $2 mil?) 3 wins worth.

I suspect there are FAs that exist *now* BL, not looking backwards, nor necessarily projecting someone who’ll never get the chance.

   23. Sam M. Posted: December 22, 2005 at 07:39 PM (#1790480)

I suspect the Dodgers could have gotten Kaz Matsui to play SS for $7mil and he’ll perform about as well as Furcal. A little worse, but not (how much is a marginal win? $2 mil?) 3 wins worth.

Not to divert a general discussion of the concept into a specific quarrel over these players, but IMHO, Furcal is easily worth double Matsui’s salary.  But even if he weren’t, the assumption that he’s a better value assumes the Dodgers have something better to do with that $7M (i.e., they can produce more wins) than spending it on the difference between Matsui and Furcal.  I strongly doubt they could.  Moreover, it also assumes no value in the certainty the Dodgers get in knowing that position is filled for three years instead of having to do it all over again in 2006-07 when Matsui’s contract is up.  Your approach, Chris, would involve an awful lot of year-to-year roster reconstruction because you’d be signing a high number of one-year “bargains.”  I think there’s a cost associated with that.

   24. AROM Posted: December 22, 2005 at 07:39 PM (#1790483)

As far as I know its never been done.  I’ll have to get the baseball forcaster to see if they work for 2006.

Dan Meyer projected to be a slightly above average pitcher for 2005, and obviously was nothing of the sort, but that tells us nothing.

If we take a group of rookies with similar projections, and compare them to a group of veterans with the same projections, will the vets be more more reliable in the future?

I’d love to find out, but my gut tells me that if you go with the vets you’ll be as likely to get stuck with a Russ Ortiz as the guy going with rookies is to be stuck with Meyer.

   25. Dan Szymborski Posted: December 22, 2005 at 07:51 PM (#1790496)

MLE’s are a hugely unreliable stat because they only select for players that actually make the ML and perform well enough to stay in the Major leagues.

The problem is that the selection bias doesn’t completely work this way because MLEs miss high *and* low.  Players are selected from the minor leagues based on the expected performance, not on how well they match with their MLEs - we’d also run into a selection bias of teams picking players likely to exceed poor MLEs, making MLEs seem like they’re too low.

   26. Chris Dial Posted: December 22, 2005 at 07:58 PM (#1790511)

IMHO, Furcal is easily worth double Matsui’s salary.

Oh, that might be true, but he isn’t worth $13 mil.

Personally I find the argument like the HoF one - since X is in, then Y is in.

Since X *got* this much, then Y is worth this much.

Uh, no, X wasn’t wrth half that much - regardless of what teams paid - which is confounding to those that demand the market is what it gets - it isn’t - because there are free players laying around.  It’s stupid money.

Or I need to know the term for what I am refering to.

   27. Dag Nabbit is part of the zombie horde Posted: December 22, 2005 at 08:00 PM (#1790516)

I suppose I should whip out my own before complaining too publicly.

Huh huh. Heh.  He said, “whip out my own,” Beavis.  And “publicly.”

   28. Sam M. Posted: December 22, 2005 at 08:17 PM (#1790544)

IMHO, Furcal is easily worth double Matsui’s salary.

Oh, that might be true, but he isn’t worth $13 mil.

Personally I find the argument like the HoF one - since X is in, then Y is in.

Since X *got* this much, then Y is worth this much.

Wait a second, Chris.  You said in # 22 that they will “perform about the same,” and that Furcal isn’t 3 marginal wins better than Matsui (thus justifying a $6M difference in their 2006 salaries).  That suggests to me that you originally were saying Furcal is NOT worth “double Matsui’s salary.”

No matter.  My bigger issue with your point is that what what Y is worth MUST be measured in relation to what X (and Z and A and B and C) got.  All the contracts players are able to negotiate in a given FA year do inform/shape/constitute the market.  The fact is, there’s money out there, and when Burnett or Wagner gets what they get, it sends a big signal to teams interested in other FAs that the market is . . . well, what it is.  A marginal win on the FA market right now just costs more than $2M.

   29. Chris Dial Posted: December 22, 2005 at 08:49 PM (#1790606)

That suggests to me that you originally were saying Furcal is NOT worth “double Matsui’s salary.”

In that Matsui makes $7mil

If Matsui made $1mil, Furcal would be wo0rth double his salary.

   30. Chris Dial Posted: December 22, 2005 at 08:50 PM (#1790611)

Your approach, Chris, would involve an awful lot of year-to-year roster reconstruction because you’d be signing a high number of one-year “bargains.” I think there’s a cost associated with that.

No, it *may*, and that is done only for positions you can[‘t fill with your own farm system, and the very few FAs who *are* worth $13 mil.

   31. Chris Dial Posted: December 22, 2005 at 08:51 PM (#1790613)

The Braves do this very well.

   32. GuyM Posted: December 22, 2005 at 08:59 PM (#1790633)

I’ve claimed before, and written that you can assemble a very cheap FA team that will produce WS at a lower cost than this.

“because there are free players laying around.”

Even if we stipulate there are some above-replacement players available at “reasonable” cost, what this common complaint misses is that these are not 5-win or 6-win players.  They are 1- or 2-win players, maybe league average on occasion. You can accumulate some cheap WS this way, but you can’t build a championship—or even winning—team this way.  The market may let you buy 2 wins (6 WS) for $2M, but you can’t buy 6 wins for $6M.  “Well,” you say, “I’ll just buy 3 2-win players.”  No, you won’t—because roster limits and PT constraints don’t allow it.

This leads to an important larger point:  if win value is non-linear, then so is player value.  If win #90 is much more valuable than win #68 (and I agree), then a 6-win player is worth more than 3 times as much as a 2-win player.  A team can’t get to win #90 without several above-average players; conversely, the wins provided by elite players are more likely to be high-value wins than those provided by average and below-average players.  So it is entirely rational to pay more to elite players on a per-win basis. 

And I think this is true not only economically, but also in pure baseball terms.  If your metric is athletic competition—being better than other teams—win #90 matters more than win #68.  Or if your metric is fan satisfaction, the same is true.  Whatever you care about, some wins are more equal than others—and we need to account for that.

   33. studes Posted: December 22, 2005 at 10:44 PM (#1790849)

Guy, regarding y-t-y regression, I used a very simple regression analysis to project that LOB% is somewhat a repeatable skill.  You’re right that this is very simplistic, but I basically said that there wasn’t enough of a finding to reject the null hypothesis.  I think that’s appropriate, though I’m no regression expert.  Alas, it doesn’t run in the family.

Regarding salaries, my numbers are presented as a beginning baseline, not the be-all and end-all number.  In my mind, the market is king.  If the market paid $4 million per win above bench last year, then that’s what it’s worth, no matter what we may think.

We may be able to come up with better, more cost-effective strategies for building a team, but that’s not where I’m coming from.  I’m just interested in assessing the “market” worth of a player as best I can.

   34. GuyM Posted: December 22, 2005 at 10:55 PM (#1790864)

regarding y-t-y regression, I used a very simple regression analysis to project that LOB% is somewhat a repeatable skill.

I’m trying to distinguish between general pitching skill and the skill of pitching better with RISP.  Your high-LOB pitchers have a .246 BAA with nobody on, vs. .272 for the low-LOB pitchers—the first group is just better, period.  So I’m wondering if there’s a y-t-y correlation in the spread btwn BAA/RISP and BAA/nobody on.

   35. Mike Emeigh Posted: December 22, 2005 at 11:12 PM (#1790877)

Bunting:

When I looking at bunting by non-pitchers a couple of years ago, what I noted was that it was primarily a late-game strategy, usually executed either when the game was tied or when the team trailed by one or (occasionally) two runs. The stolen base (the other “obvious” one-run strategy), on the other hand, was generally used as an early-game strategy and was used more often by a team that was in the lead.

—MWE

   36. Backlasher Posted: December 22, 2005 at 11:26 PM (#1790890)

I’ve seen a few parties post reliability scores on their estimates, but never an attempt to measure how valid those scores were after the fact.

Well, I haven’t.  I’ve seen people throw out confidence intervals, but that is not very reliable.  I’ve seen people throw out some weird hypothesis tests, but that also suffers from problems.  I don’t think PECOTA is disclosed, but its possible they may be using Bayes factors.

As for validity, Shandling does have an interesting article on that, with maybe a little shot at Tango.  (I wasn’t able to find the Shandling Number though).

I’d love to find out, but my gut tells me that if you go with the vets you’ll be as likely to get stuck with a Russ Ortiz as the guy going with rookies is to be stuck with Meyer.

Its not about going with rookies or going with vets, its about reliability.  Teams are projecting either expressly or implicitly with a lot more than the numerical models of the touts.  You can have more reliability in a rookie than a long term veteran b/c of that information.  What I said was limited to MLE’s, which are poor for forecasting.

Players are selected from the minor leagues based on the expected performance, not on how well they match with their MLEs - we’d also run into a selection bias of teams picking players likely to exceed poor MLEs, making MLEs seem like they’re too low.

I’m not sure what you are saying Szym.  If you want to assess the value of Drederick Barton in a trade, an MLE is poor.  If your trying to project major league output, any player that does not make the Major Leagues has a .000/.000/.000 output.  Therefore any tout method to assess reliabiilty should have shitty variance and be enormously high.

MLE’s may, or may not, provide value, iff you only apply them to the set of players that have made the major leagues and will stay in the major leagues.  They do not have value across the set of all minor leaguers.  Its how you end up with people overtouting the Jack Crusts of the world.

 

I suspect the Dodgers could have gotten Kaz Matsui to play SS for $7mil and he’ll perform about as well as Furcal. A little worse, but not (how much is a marginal win? $2 mil?) 3 wins worth.

I suspect there are FAs that exist *now* BL, not looking backwards

Guy has already discussed the increasing function to players.  I don’t need to readdress it, but I’ll adopt it by reference.

Moreover, if you have a system where you can show that Dodgers would achieve their goals with Kaz at a million dollar cost savings, more power to you.  I haven’t seen your system, so it would be improper for me to assess it.

I watch the Braves, and I think they do a good job of picking up value, but they also seem to be a little short in the final push department too.  The difference in a world championship can often be a Matt Diaz/Tommy Gregg versus a Wade Boggs on your bench.

I have seen other touts system, and good lord I’ve seen fanboys and their evaluations.  I have seen people go to town talking claiming the Braves were idiots for not signing Stairs,  then seen them be wrong every year on the Braves finish, and then right a column about how they couldn’t have expected X.

There are very few touts that will even put their system up for examination.  Those that do, Shandling does a good job of showing their incredularity in assessing value. 

If you study an organization long enough, you may find systematic errors in their decision processes.  That should be true of every organization under the sun.  Some of the systematic errors lead to really poor output.  But, I’ll be honest, I’m not finding that among critics at all.  Their analysis is that A is stupid or B is a genius, and then A is always chastised and B is always praised.

I don’t claim to be a tout.  I’m not trying to develop a forecasting system; I see little value in doing it if the only data is the current set of performance output.  At its best, the error is too high.

I’m much more interested in systems and processes, etc.  And with that being said, I don’t know if studs is right or wrong on his money assessment, I just think the studs number could be a good thing to have for negotiation.  Its merely one piece of evidence in trying to model the market, and nowing how you should bid, raise, and leave for certain transactions.

   37. Backlasher Posted: December 22, 2005 at 11:31 PM (#1790895)

When I looking at bunting by non-pitchers a couple of years ago, what I noted was that it was primarily a late-game strategy, usually executed either when the game was tied or when the team trailed by one or (occasionally) two runs. The stolen base (the other “obvious” one-run strategy), on the other hand, was generally used as an early-game strategy and was used more often by a team that was in the lead.

Interesting.  It does seem to confirm to general risk and utility applications.  When an outcome is close, you are willing to sacrifice resources for marginal advantages.  Leaders tend to take more risky strategies, e.g. stolen base has a larger risk/reward ratio for a marginal advantage, bunts has more certainty in the loss of a resource for a marginal advantage.

   38. base ball chick Posted: December 22, 2005 at 11:41 PM (#1790911)

i don’t know a y-t-y regression from a t-y-t progression, but i sure do know that some guys seem to pitch their way out of jams bettern others - kirk rueter used to do this. and i wonder how the ability to do this would change in different conditions - different outs, runner/s on different bases, and my interest - does the pitcher freak out if a runner gets on because of an error or a “hit” that shoulda been an out with a better fielder…

and i DID luuuvvvv that link. joel friesen got a great sense of humor. but you KNOW the poor boy still a dork cuz he didn’t know enough to put graph 7 first…

- and as for market value - well, ain’t that what people willing to pay? i mean, if i want, say, a copper-core all-clad sauteuse pan (in case husband is lurking) and he think it’s only worth 40 bucks, but a LOT more people think it’s worth more, then they’ll pay more and i’ll be out of luck. (and he hafta come up with something bettern graphs…) just like drayton mclane and baseball players…

   39. Backlasher Posted: December 22, 2005 at 11:55 PM (#1790936)

- and as for market value - well, ain’t that what people willing to pay? i mean, if i want, say, a copper-core all-clad sauteuse pan (in case husband is lurking) and he think it’s only worth 40 bucks, but a LOT more people think it’s worth more, then they’ll pay more and i’ll be out of luck. (and he hafta come up with something bettern graphs…) just like drayton mclane and baseball players…

Absolutely, but our argument is a little different.

You can buy that pan new at the store, or you can buy a used one of Ebay.  In the latter case, you might not trust buying that pan from a stranger.

Some people are saying, you don’t copper-core, you’ll just be fine with plain old aluminum because you can buy 3 aluminum pans for the price of 2 copper cores, and copper only lasts twice as long.  Others are saying, that’s not true if it affects the taste of the eggs.

I’m saying it doesn’t hurt to know what they are going for on ebay; it doesn’t hurt to know if you have aluminum alternatives, and, dammit, I don’t want shitty tasting eggs, unless I’m going to get some kickass pork chops.  So its nice to know all these things before you decide which pan, and which oven to buy, and how much you are going to pay for it.

   40. studes Posted: December 23, 2005 at 12:19 AM (#1790971)

So I’m wondering if there’s a y-t-y correlation in the spread btwn BAA/RISP and BAA/nobody on.

Well, I don’t know.  But there is a correlation in LOB%, which is an indirect measure of that spread.  One pitcher that immediately comes to mind is Tom Glavine, who was known for having much better stats with RISP, at least from the late 90’s to the early 00’s.  Looking at Clemens’ stats at Retrosheet, batters have batted .229 with no one on vs. .215 with RISP for his career.  Jim Palmer’s splits were .231/.213.  Eerily similar.

I did find that there is a positive correlation between overall pitching skill (I used xFIP) and LOB%.  Shandler found that pitchers with higher K rates, lower BB rates and a higher groundball rate will have higher strand rates.

   41. Chris Dial Posted: December 23, 2005 at 12:40 AM (#1791000)

Even if we stipulate there are some above-replacement players available at “reasonable” cost, what this common complaint misses is that these are not 5-win or 6-win players. They are 1- or 2-win players, maybe league average on occasion. You can accumulate some cheap WS this way, but you can’t build a championship—or even winning—team this way. The market may let you buy 2 wins (6 WS) for $2M, but you can’t buy 6 wins for $6M. “Well,” you say, “I’ll just buy 3 2-win players.” No, you won’t—because roster limits and PT constraints don’t allow it.

This leads to an important larger point: if win value is non-linear, then so is player value. If win #90 is much more valuable than win #68 (and I agree), then a 6-win player is worth more than 3 times as much as a 2-win player. A team can’t get to win #90 without several above-average players; conversely, the wins provided by elite players are more likely to be high-value wins than those provided by average and below-average players. So it is entirely rational to pay more to elite players on a per-win basis.

Well, don’t I suck.

Part 2 first: Me in my first post “I think you may need to change the equation of the line from linear. “

Didn’t you just say that?  And I agree with this - but I don’t think Furcal *is an elite player*.

Which I said in some other post (that there are some elite players.

The Braves are a great example of what I am talking about:
3 homegrown elite players - Jones, Jones, Smoltz
3 homegrown decent players - Furcal, Giles, Estrada, HRamirez
1 store-bought eliteplayer - Hudson (okay a trade, but whatever)
LF, RF, 1B, SPs all cheap FAs (IIRC).  Juan Cruz, Thomson, Jordan, Sanders, Franco

Just half their roster off the scrap heap - for about two years, and then they completely obverhaul those players.

You are telling me it can’t be done, but it *is* done by the Braves for 15 years. 

BL, I think, sees this, but then he hints that not winning the WS but once means this system fails even if it wins 95 games every year for 15 years (which is of course bonkers).

   42. Backlasher Posted: December 23, 2005 at 12:49 AM (#1791009)

BL, I think, sees this, but then he hints that not winning the WS but once means this system fails even if it wins 95 games every year for 15 years (which is of course bonkers).

It isn’t bonkers if its true.  The Braves should have more WC, but that is based on the mid-90s braves that featured all home grown and some expensive store bought purchases, e.g. Grissom, McGriff.

The Braves are damn good, but when they just merely fill holes, they don’t have push.  And this is reflected in their actual results, and in the impressions: everyone, including touts, oddsmakers, and even you saberists made them longshots.

The only way this strategy is viable is if you completely buy into that facially invalid crapshoot model.

But yes, I know exactly what you are getting at.  But more important, the Braves are expressly not doing this with MLE or numeric projection systems based on STATS collected data.  They are doing it on talent evaluation from observation.  Did Cather have great MLE’s.  How about Lightenberg?  What was Hammond’s projection?  Julio Franco?

This scrap heap that the Braves are raiding isn’t the same scrap heap that has Kaz Matsui and Jack Crust.  And the touts only can look at Matsuit and Crust.  They have little or now information on Hammond or Franco.

   43. greenback calls it soccer Posted: December 23, 2005 at 01:32 AM (#1791045)

What I said was limited to MLE’s, which are poor for forecasting.

Is there real evidence of this?

   44. Dan Szymborski Posted: December 23, 2005 at 01:58 AM (#1791062)


I’m not sure what you are saying Szym.

What I’m saying is that if major league executives have the ability to pick players that will fail to meet good MLEs, then they should also be able to pick players that will exceed bad MLEs and be useful, which is a selection bias that would make MLEs look like they’re too *low*.

MLE’s may, or may not, provide value, iff you only apply them to the set of players that have made the major leagues and will stay in the major leagues. They do not have value across the set of all minor leaguers. Its how you end up with people overtouting the Jack Crusts of the world.

The thing is, “Jack Crust” was *not* an MLE God - he was a walk guy who was playing on a horrific team (Baltimore) that should’ve given him a chance because they had nothing.  Cust wasn’t going to play over anyone good.  The Orioles then played Jose Morban, a Rule 5 middle infielder at DH more than a dozen times.

Here are my translations for Cust, adjusted to a neutral park:

1999:  267/347/434
2000:  251/374/400
2001:  240/367/417
2002:  231/348/428
2003:  260/381/394
2004:  213/317/379
2005:  219/340/374

   45.  nddc Posted: December 23, 2005 at 02:29 AM (#1791095)

What I’m saying is that if major league executives have the ability to pick players that will fail to meet good MLEs, then they should also be able to pick players that will exceed bad MLEs and be useful, which is a selection bias that would make MLEs look like they’re too *low*.

In principle, this is what makes the Heckit model that Der Komminsk-ar mentions in #20 so promising.  There are two equations - one predicting whether the prospect makes the majors, and another predicting performance for those that do make it. 

The model works best when there are variables that are included in one equation, but not the other.  Past performance certainly belongs in both equations, but the structure and practice of the GM office is more likely (IMO) to affect the decision to offer a ML contract than it is to affect performance.  [And the structure and practice of the ML field staff, the manager and coaches, probably belongs in the performance equation but not in the contract equation.]

Actually, this would be great from an econometric standpoint because certain variables belong in exactly one equation.  I keep qualifying my enthusiasm because it’s so hard to measure the variables I’m talking about in the paragraph above.

   46. greenback calls it soccer Posted: December 23, 2005 at 02:44 AM (#1791112)

Here are my translations for Cust, adjusted to a neutral park:

Should a 20-22 year old with a 250/360/420 line project better than Cust has done in his mid-20s?

   47.  nddc Posted: December 23, 2005 at 02:50 AM (#1791115)

About reliability (which is the topic of the post where BL called me out - is this what you are asking about?) - here’s what occurs to me:

Generally speaking, you’d have more reliable forecasts with more years of prior data.  But there’s another process working in the opposite direction, decreasing reliability:  age and workload.  Fluke injuries can happen to anyone at any age, but the chances of falling off a cliff performance-wise increase with age and wear and tear.  Big drops in performance will kill your forecast accuracy when you evaluate a prediction system after the fact.  Big jumps in performance will too, but those are rarer (assuming the needle never meets the tush - for the moment, <u>please</u>).

Definitely scouting could account for some of the factors that are associated with big changes in performance.  If you could build the factors that scouts notice into the model(there I go again;  a big if) then you could somewhat neutralize the trend toward unreliability that’s caused by age and work.

   48.  nddc Posted: December 23, 2005 at 03:03 AM (#1791133)

I think that’s appropriate, though I’m no regression expert. Alas, it doesn’t run in the family.

I always wondered about that, Studes.  I used to recommend a book called Using Econometrics by someone named A.H. Studenmund.  Nice book, but the damn thing kept floating in and out of print, though.  Any relation?

   49. Buzzards Bay Posted: December 23, 2005 at 03:27 AM (#1791168)

What is the “People component”.....this is fascinating stuff…..the attempt to quantify…but I think the Braves model is telling because it doesn’t rest solely on the gymnastics of mathematics…..likewise is there a reason Palmer and Clemens mirror each other with RISP….. all things being equal…..Rafael Furcal has the rare ability to change a game ...the way he plays in addition to his talents….very few guys can do it…...

   50. PhillyBooster Posted: December 23, 2005 at 03:32 AM (#1791171)

What I’m saying is that if major league executives have the ability to pick players that will fail to meet good MLEs, then they should also be able to pick players that will exceed bad MLEs and be useful, which is a selection bias that would make MLEs look like they’re too *low*.

Is this true?  Are there examples of a significant players with sub-replacement MLEs who turned out to have decent major league careers?

My assumptions from looking at the ‘real’ prospects vs. the Custs/Pickerings of the world is that a quality MLE may be a “necessary” but not a “sufficient” factor.

   51. Backlasher Posted: December 23, 2005 at 04:04 AM (#1791224)

n principle, this is what makes the Heckit model that Der Komminsk-ar mentions in #20 so promising. There are two equations - one predicting whether the prospect makes the majors, and another predicting performance for those that do make it.

If accurate, this would be an excellent improvement, I’m not familiar with Hackett.

What I’m saying is that if major league executives have the ability to pick players that will fail to meet good MLEs, then they should also be able to pick players that will exceed bad MLEs and be useful, which is a selection bias that would make MLEs look like they’re too *low*.

Yes, they could be too high (where high is an overestimation of the performance) or too low (where low is an underestimation of performance).  But that doesn’t make the MLE have more value, or less value for forecasting, because its still dependent on the distribution of the selection function. 

As to value, its possible that would make MLE accurate for all players that have been selected, but it still doesn’t help you make the selection.

And moreover, the “make it” distribution is not enough.  Unless “make it” also encompasses the ability to persist.  The only way that MLE can independently have value is if MLE, or a large number of its strongly influential components, also influence the selection and persistence function.

Moreover, the value for a forecast is still lean until you can predict opportunity, or the rate projection is constant for all number of chances, which I don’t believe will be true.

Also, remember that if you are trying to get expected value from a set of choices, its going to be a probablity curve on teh aggregation of outputs.  There will be at least three distinct curves (we could search for more couldn’t we ACE FOS), one the districution on the set of all former major league players, one the distribution on the set of all former minor league players, and one the distribution on the set of all players who never played in either.  I trust I don’t need to prove these comprise sets with different statistical characteristics.

The last set is currently pretty neglible, but may be to be subdivided if we get more foreign players that go straight to the majors.  Things like Zips addresses set one.  Things like MLEs address set two.

So once you come up with your zips and MLE’s, if you want to forecast, to find the contribution its going to be Aggregation(ZIPS * p(persist)) for curve one and Aggregation (MLE * p1(persist)).  Now in either case, those functions are obviously very dependent on the persist function. 

Even if we have absolutely no other information, its nontrivial to show that p > p1 in most any space.  That is, if you take the set of all major league players, any randomly selected member will have greater probability of being in the majors the next year, then a randomly selected member from the set of all minor league players.

So until you can deconstruct the set of minor leaguers into subsets based on the likelihood or making the majors that MLE space is going to have little value.  Now there may be some ways to do that, but acey doesn’t like them too much from prior conversations.

And both have limitations until/unless you can scale your rate, give it a reasonably narrow interval or high inference score.

And they all have limitations if you can’t aggregate via simple summations, which I don’t think you can.

Is there real evidence of this?

There is a real proof that could be constructed.  Evidence should not matter.  I’ve given you the verbaige to construct it.  It would take too much time for me to do it.  The only problematic element is if there is interdependency between your rate projection and your persistance distribution.  In which case, the value would increase for forecasting purposes proportionate to the influence between the two distributions.

   52. Backlasher Posted: December 23, 2005 at 04:05 AM (#1791226)

And Ace, the above is just one reason that I think pure discriminant analysis will have you chasing your tail for years and years and years.

   53. studes Posted: December 23, 2005 at 05:16 AM (#1791310)

Any relation?

Yup, that’s my big brother, who I talked about in the above article.

   54.  nddc Posted: December 23, 2005 at 12:11 PM (#1791584)

Yup, that’s my big brother, who I talked about in the above article.

Aha.  I found his book very useful (especially the exercises) when I was teaching.  Very sad story about his APBA colleague.

   55.  nddc Posted: December 23, 2005 at 12:33 PM (#1791589)

So until you can deconstruct the set of minor leaguers into subsets based on the likelihood or making the majors that MLE space is going to have little value. Now there may be some ways to do that, but acey doesn’t like them too much from prior conversations.

My plane leaves soon, so I won’t have time for a long post…

Organized Baseball gets you started on the process of separating major leaguers into groups based on their probability of being in the majors next year.  Members of the group with the highest probability - let’s call it Group AAA - are more likely to be in the majors than members of the next group, which we’ll call Group AA, and so on.

MLEs, as I understand them, are an attempt to standardize performance measures across all these different contexts so we can compare apples to apples.  We could compute AAEs if we wanted to, but the major league context is more substantively interesting, so we convert everyone’s stats to that numeraire.

Once that’s done, we can get a Marcel-like prediction based on the principle that the best predictor of next year’s stats is last year’s stats, or the last few years’ stats.  But until you standardize your performance measures, you can’t do this.

If accurate, this would be an excellent improvement, I’m not familiar with Hackett.

If accurate, the method would deal with your interdependency problem between the contract and performance functions.  If there are unmeasured factors that are common to both equations, you need to account for that by seeing if the errors are correlated.  If they are, you have evidence of selection bias - assuming that there is bias, and that we’re able to measure all the factors that explain why a player makes the bigs.  If we can’t do that, there will be omitted variable bias in both equations.

And Ace, the above is just one reason that I think pure discriminant analysis will have you chasing your tail for years and years and years.

Yeah, ain’t it cool?

I’m just saying that I’d rather start on this route than start on a route where you immediately start chasing your tail.  In my experience, factor analysis and cluster analysis gets you to chase mode faster than this would.  Will either pan out ultimately?  Never know until somebody tries.

I see the immediate goal as the development of an accurate, continuous measure of performance, not as a way to split people into groups based on performance levels.  Once we’re there, we can argue about the splitting criterion if we want to, but we are still working toward goal #1.  I shouldn’t say “we” because I’m not doing anything to help, but hopefully you understand.

   56. GuyM Posted: December 23, 2005 at 06:37 PM (#1791907)

Me in my first post “I think you may need to change the equation of the line from linear.”  Didn’t you just say that? And I agree with this - but I don’t think Furcal *is an elite player*.

Yes, Chris, you said that, but it didn’t sound like you thought high-value players should be valued even more highly than a linear model would indicate.  As Studes said in a post above, if he had used a non-linear model then Furcal would be valued even higher, a view you certainly don’t seem to agree with.  Sorry if I misinterpreted.

Agreed that Furcal isn’t an “elite” player. My comment wasn’t meant to be about Furcal, but about all elite/good players (say, 2+ wins-above-avg). As for valuing Furcal, a lot depends on how you value his defense: the THT Win Shares say 30% of his 2005 value was fielding, but BPro says it’s over 50% (41 FRAR).  How does your (Chris’) system rate him?

The Braves are a great example of what I am talking about:
3 homegrown elite players - Jones, Jones, Smoltz
3 homegrown decent players - Furcal, Giles, Estrada, HRamirez
1 store-bought eliteplayer - Hudson (okay a trade, but whatever)
LF, RF, 1B, SPs all cheap FAs (IIRC). Juan Cruz, Thomson, Jordan, Sanders, Franco….
You are telling me it can’t be done, but it *is* done by the Braves for 15 years.

So many different issues are involved here, it’s hard to sort out.  If your point is that it’s better to have players who can’t negotiate a market-value contract, well, sure it is.  I think we can all agree that Furcal at $6M is better than Furcal at $13M.

And if your point is a team can sometimes get cheap 1-win and 2-win players off the scrap heap, I think we all agree there, too.

But the Braves’ success has also depended on having several good/great players, who were paid market value at that time.  The Jones boys, Smoltz, Maddux, Glavine—none of them came cheap. 

Well, don’t I suck.

It’s the holidays—don’t be so hard on yourself.

You must be Registered and Logged In to post comments.

 

 

<< Back to main

BBTF Partner

Support BBTF

donate

Thanks to
1k5v3L
for his generous support.

Bookmarks

You must be logged in to view your Bookmarks.

Hot Topics

NewsblogPrimer Dugout (and link of the day) 7-22-2014
(16 - 7:16pm, Jul 22)
Last: Eric J can SABER all he wants to

NewsblogOTP - July 2014: Republicans Lose To Democrats For Sixth Straight Year In Congressional Baseball Game
(2711 - 7:16pm, Jul 22)
Last: The Id of SugarBear Blanks

NewsblogAs shifts suppress offense, time has come to consider a rule change
(6 - 7:14pm, Jul 22)
Last: Pat Rapper's Delight

NewsblogRangers' Yu Darvish Pushes for a Six-Man Pitching Rotation - NYTimes.com
(11 - 7:07pm, Jul 22)
Last: Srul Itza

NewsblogMLB: Astros telecasts catching on to advanced metrics
(8 - 7:04pm, Jul 22)
Last: Walt Davis

NewsblogOT: Monthly NBA Thread- July 2014
(815 - 7:04pm, Jul 22)
Last: andrewberg

NewsblogChase Headley traded to New York Yankees from San Diego Padres - ESPN New York
(75 - 7:00pm, Jul 22)
Last: Baldrick

NewsblogFSAZ: D-backs cut off McCarthy’s cutter controversy
(25 - 6:59pm, Jul 22)
Last: Walt Davis

NewsblogMisremembering Mantle's Final Season
(46 - 6:53pm, Jul 22)
Last: Jolly Old St. Nick Still Gags in October

NewsblogTony Oliva turns 76; Gardenhire: 'He should be in hall of fame'
(41 - 6:44pm, Jul 22)
Last: Tubbs & Minnie Miñoso don't fear Sid Monge

NewsblogOT: The Soccer Thread July, 2014
(317 - 6:41pm, Jul 22)
Last: Darkness and the howling fantods

NewsblogThree Moves The Red Sox Should Make - Tony Massarotti - Boston.com
(26 - 6:18pm, Jul 22)
Last: Joe Bivens, Minor Genius

NewsblogTrading for Price would be right move for Cubs | FOX Sports
(69 - 5:57pm, Jul 22)
Last: SteveM.

NewsblogGeorge "The Animal" Steele Mangles A Baseball
(123 - 5:40pm, Jul 22)
Last: Canker Soriano

NewsblogOT: NBC.news: Valve isn’t making one gaming console, but multiple ‘Steam machines’
(673 - 5:32pm, Jul 22)
Last: TFTIO can't talk like this -- he's so sorry.

Page rendered in 0.9110 seconds
52 querie(s) executed