THE BOOK--Playing The Percentages In Baseball

Minor detail: Jose Guillen was not traded but signed a free-agent deal. I was confused for a while which Guillen you’re refering to.

Great analysis, MGL—it’s a real joy to read sober-minded analysis.  I think that the problem lies in the fact that BP is a middlebrow site, that is, it’s somewhere between sabermetrics and old-time hack sportswriting.

#1, yes of course.  #2, thanks.

Funny, as Joe concludes:

League-average players have value, and Guillen is that

And that’s exactly what I called him!  A league-average player.  League-average players however make 9MM this year, and, if you are league-average at age 32, you are likely out of baseball at age 35.

Yeah, I wouldn’t have any problem calling him league average, although I have him as almost a win worse than that based on his poor UZR projection.  I HAVE heard that he can’t run anymore, which would not be surprising for an early thirties oft-injured player.  If he “can’t run anymore” it is unlikely he is even a league average defender in the OF, as my “speed versus UZR” study showed.

I don’t think that BP is quite at the (low) level of being “in between a sabermetric site and a hack mainstream site.” I think that they have too many writers (not all of them being even close to experts in sabermetrics), and they have a different agenda than just trying to “tell the truth” - which is to make money of course.  Plus, they are only a group of fallible human beings.  We make lots of mistakes here on this blog, but because of the format and our non-monetary agenda, we have a chance to vet many of them.

Guillen has a specific combination of tools:
http://www.tangotiger.net/scouting/sim2007_5753.html

All of his top comps are RF, 3B, and catchers (that is, heavy arms, heavy legs). 

Among the OF, his top comps according to Fans (UZR/150, from 03-mid07 are in parens): Vlad (-4), Jayson Werth (+5), Abreu (0), Hawpe (-4), Dye (+1), Nady (-2).  That doesn’t include their UZR arms, which likely would add 1 or 2 runs each.  Overall, Guillen’s top comps are basically average RF according to UZR.  The average Fan rating has them at just a sliver above zero.

If I had to guess on Guillen’s UZR for 2008, it will be the regressed about 80% toward the average of these 6 RF, be it UZR or Fans’ ratings.  I don’t think you can call him at anything worse than a -5 run RF.

In any case, you would need to cap his fielding in RF at -15 runs.  Anything worse (-15 runs in RF, plus the -5 positional adjustment) means that he’s a DH.  SS can go as low as being -25 runs in UZR, and 1B can be as low as -10 runs in UZR.  Anything worse than that means that they should only be a DH (or see if they can leverage their fielding at some other position).

Guillen suffered through some ankle problems in the first month of the season, serious enough to the point that he was being removed for Jason Ellison in late innings on a nightly basis.  For about the first 30-40 games of the year, he ran like a wounded solider. 

The ankles healed as the season went on, and his defense improved.  His range isn’t great, but I’m with Tango - he’s a -5 type fielder, not a -15 type fielder.

I think you’re misinterpreting the term “breakout year”.  I look at it to mean, “you should value the breakout year more than you normally would, and other years less than you normally would, in projecting forward”.

And truly, if someone had the skill to recognize Guillen’s 2003 as a breakout year, I’d bet their projection for him for 2004 would have been more accurate than something like Marcel.  In other words, if Marcel uses the 3, 2, 1 weighting, a breakout year might necessitate a 5, 2, 1 weighting.

Injuries (or fear of injuries) are by far the biggest determinant in fielding impact (UZR runs or whatnot).

One of the biggest UZR stars through 2004 was Mark Kotsay.  In 2005, with a bad back, he started his downward flight in UZR.  Rowand was a UZR star through 2005.  Then we know what happened in 2006.  Fielders have nothing to fall back on when they get hurt. 

Hitters on the other hand can compensate to some extent (Andres Galaragga is probably the best example).

When it comes to fielding, I’d need to know if he was injured, and if it’s a potentially recurring injury.

Not that it matters in this discussion, but Marcel uses 5/4/3.

As for “breakout year”, it is far more commonly accepted that it’s not simply a reweighting from 3/2/1 (50% for most recent season) to 5/2/1 (63% for most recent season).  That’s barely a change.

No, it’s more like if you consider Beltre’s 2004 season, that if this was his breakout year, you’d weight that season far far more than his prior seasons.  You’d do at least 80% of the weight, if not 90%, with a tiny bit of allowance that 2004 was not a breakout year.

http://www.baseball-reference.com/p/putzjj01.shtml

Probably follows your 80-90% definition of breakout year.  Of course, this is a sample size of 1, but if I were a betting man I’d have bet a fairly sizable chunk of money that Putz would have had a 2007 that was very similar to his 2006, and not 2005-earlier.  Why?  Because he learned to throw the splitter before 2006.

I don’t believe research that says breakout years have no predictive impact.  The problem is, we aren’t looking at enough data.  Put a column in the model for “learned a new pitch”, and see what happens.  Same with “lost 8 mph on his fastball”, “got laser eye surgery”, “started hanging out with Greg Anderson”, “hat size grew by half an inch”, and “gained 30 lbs”.

I guess your point is that, at the time, we’d consider Beltre’s 2004 and Putz 2006 as two instances of the same “breakout” thing.  And I can see why you’d conclude from a bunch of those types of seasons that there’s no true new level of talent established.  But I guess the baseball romantic in me wants to believe that there IS a way to identify true breakout seasons - true seasons that establish a new talent level - and I think the data is there to do so if we look for it.

Certainly there is a way to identify it.  You CANNOT identify it by looking at the seasonal aggregate data, which is what virtually everyone does.

Mike, pitchers are much more likely to have seasons where their true talent changes considerably.  That is the main reason why we weight recent seasons more aggresively with pitchers.  Certainly, if you KNOW that something changes, like a new pitch or an injury, or recent great health (maybe), then you can do your own “seat of the pants” weghting, or tweak the projection, or what have you.

With batters it is MUCH harder to identify something fundamental that has changed, BEFORE (or concurrent with) the fact.  After the fact, everyone has a “reason” for the breakout, just like all successful teams had great chemistry (as Tango says, let’s identify the good or bad chemistry before the season starts, just one time), etc.  Players are always coming into camp in the best shape of their lives, injury-free (finally), yada, yada, yada.

We are pretty much just talking about looking at the numbers Mike.  No one (with any sense) is going to argue that if you KNOW (or even suspect) something has changed fundamentally with a player, that you want to use the same old Marcel to project his performance.  But without knowing anything but the numbers, the research has shown that ANYTHING but the usual Marcel weightings do NOT work as well even after a banner year, either poor or good.  There are some qualifications to that.  For certain component stats, at certain ages, and certain types of banner years, there are better weightings than Marcel, but they are subtle.

But my point still stands (and much of my initial post was a little hyperbolic and sarcastic with a feeble attempt at humor, so don’t take it too seriously or literally) that the way most people look at “breakout” seasons (a heavy weighting to that year) is simply not correct.  And I don’t like to even use that term.  A season is a season.  And sure, after a breakout season, a player’s projection, by definition goes up a lot - just not as much as most people think.  As well as the fact that prior years are still very important.  A lot of people will think that a player, say at 27 or 28, when players are supposed to have breakout seasons, who posts a +20 at age 28 after a +3 and +5 is about the same player going forward as a player who posted that same +20 after a -12 and a -3 or even a -12 and a +7.  They are not.  All years count.

So much in that post, MGL!  One comment that comes to mind for me: I disagree with you regarding consistency.  Consistency is not just a statistical phenomenon.  Some players have a skillset that tends to yield more consistent results over time.  A strikeout pitcher, say, than a pitcher who relies on his fielders.  Or power vs. contact hitters.

Secondly, I believe some players are inherently streakier than other players and that can have an impact on seasonal results.  Tony Gwynn was a remarkably consistent offensive player.  Cal Ripken tended to have a few great among many average seasons (offensively) (caveat—I named those two players off the top of my head due to their recent Hall of Fame induction.  May not be a valid comparison).  I think consistency is a player attribute, often related to injuries, true, but sometimes related to something about the player himself.  We can argue about whether it’s good or bad, or how to measure it, but I do think it exists.

Just a comment on Guillen’s arm. For 2005-2007 in right field I have him at +2.1 runs total in 335 adjusted games with exactly average performances in 2005 and 2006 and a better 2007. This metric looks at all advancement opportunities and uses run expectancy. His performance in 2007 ranks him eighth among regular right fielders with the likes of Markakis and Abreu.

Good stuff.  Here’s what the Fans say for RF (any RF not listed means he was +/- 1 run).  These are runs per 162 G.

Arm Player
5 Young, Delmon
5 Victorino, Shane
5 Francoeur, Jeff
4 Cuddyer, Michael
4 Rios, Alex
4 Hawpe, Brad
4 Markakis, Nick
3 Kearns, Austin
2 Guerrero, Vladimir
2 Guillen, Jose
...
(2) Encarnacion, Juan
(2) Lewis, Fred
(3) Hermida, Jeremy
(3) Wilson, Preston
(3) Murton, Matt
(3) Pena, Wily Mo
(3) Nixon, Trot
(3) Green, Shawn
(4) Floyd, Cliff
(5) Cust, Jack
(5) Gomes, Jonny
(5) Sosa, Sammy

I have to believe that Arms is the easiest thing for Fans to judge.  John Walsh did a series a year or two ago that also compared to the Fans.

If Dan or MGL want to compare their 3-yr arms runs to the above, it would be most interesting.

Related links from John Walsh:
http://www.hardballtimes.com/main/article/cannons-and-popguns-rating-outfield-arms/

http://www.hardballtimes.com/main/article/more-guns-in-the-outfield-center-and-left-field/

Dan, I have him at zero in 05 and 07 and -2 in 06, using exactly (more or less) the same methodology.  Let’s split the difference!

Studes, I believe that the Earth is flat and although I am a pretty smart guy overall, that does not make it so.

If you have some evidence that consistency among players is a skill (other than the K pitchers, etc., example you give, which is correct, although I am not sure how much difference that makes - but we can surely check), give me a ring.  It is easily checked of course.  And why would you give examples of players who were more or less consistent throughout their careers as if that were evidence of your belief?  We KNOW that ABSENT a “consistency skill” there will be plenty of players who, by chance, were anomolously consistent or inconsistent for their entire careers, right?

O.K., it is the holliday season.  I’ll give you a pass on using examples for “illustrative purposes.” Let’s just not mistake that for evidence!

If I have a chance later on, I’ll actually look at the data.  I have wondered that for a while - whether “consistency”, over and above a player’s “profile” (K and BB rate, which are obviouslt more “stable"), is indedd a skill, and if yes, to what extent.  Of course, most of things (consistency, clutch, etc.) are probably a skill to one degree or another.  The real question is whether it is a “significant” skill, however you want to define “significant” of course.  For example, Andy found that clutch hitting is probably a skill, but so small that is makes no practical difference, except perhaps as a tie-breaker or as a little “extra” value (positive or negative) over the course of a long career.

Yes, you can probably watch a player throw once or twice and accurately rate his arm (like speed, just watch a player run hard once), but…

Arm value is different.  It includes accuracy, which is NOT as easy for fans to judge, although if they see a player enough, I think they can.  It also includes speed to the ball and setup.  Fans can probably judge setup (although they are more interested in arm strength I think), and I think they largely ignore speed to the ball.  Maybe you should include that if you don’t already.

Finally, the reputation of the OF’er (what the runners and coach think) is part of his arm value.  The fans are probably not including that at all.  An OF whose rep is less than that of his arm is going to have more value and vice versa.

The nice thing about the fan ratings though is they can “adjust” for where the ball was hit, whether a runner made a mistake, whether no arm would have gotten a runner, etc.  In fact, the fans may be better than the numbers for a year or perhaps even two.  I am not sure of the ratio I would use to combine them, but for arm, I might use 75/25 or 60/40 fans for one year and 50/50 for two.

Of course it depends on how the fan data is collected.  There are a lot of potential biases among the fans as we have seen.  It would be a lot better if the fans would “rate” a player who they have never heard of and have to put how how many times they have seen him, etc.  I am sure we have some fans who see a player a few times and then pretty much just put down what they “have heard” or their observations are influenced by what they have heard (or read) or what they have seen in years past, etc.

BTW, anyone know why I cannot select certain text with my cursor and/or keys using IE 6 on this blog?  I can only select text from my cursor to the top of the page.

Here are my 07 and then 04-07 numbers:

Arm Player
5 Young, Delmon +13! Of course he was -17 in UZR.
5 Victorino, Shane 15/9!
5 Francoeur, Jeff 18/9
4 Cuddyer, Michael
4 Rios, Alex 19/11
4 Hawpe, Brad -5/3 (+13 in 06)
4 Markakis, Nick -8/-3
3 Kearns, Austin -6/-1
2 Guerrero, Vladimir -3/-1
2 Guillen, Jose 0/0
...
(2) Encarnacion, Juan -3/-2
(2) Lewis, Fred -1/-1 (Fred Lewis?)
(3) Hermida, Jeremy 1/0
(3) Wilson, Preston -2/-5 (He is bad as a CF!)
(3) Murton, Matt 2/2
(3) Pena, Wily Mo 0/1
(3) Nixon, Trot -5/-4
(3) Green, Shawn -5/-5 (along with -4/-8 in UZR.  He is a replacement player.)
(4) Floyd, Cliff -4/-3
(5) Cust, Jack -4/-4
(5) Gomes, Jonny 2/0 (he is mostly a DH)
(5) Sosa, Sammy -2/-3

Hawpe, Markakis, Kearns, and Guerrero seem to be way off.  They seemed to have gotten most of the bad ones correct.

Tango, do they rate players independent of position?  And then do you adjust for position?  Obviously an average arm in RF would be tremendous in LF.  Etc.

Yes, independent of position.

That’s the cool thing doing it that way.  I know exactly how much the translation should be in arm, between LF and RF.  I’ll tell you that tomorrow when I get in.

Looks like all the very good and very bad, we have great agreement on. 

For Vlad, it’s possible that my weighting of “release/footwork”, “accuracy”, and “strength” is not correct.  I gave them each equal weight.

Hawpe seems to be ok.  Just this year, the UZR seemed off.  Markakis is probably the biggest difference.

Finally, the reputation of the OF’er (what the runners and coach think) is part of his arm value.  The fans are probably not including that at all.

Should we be?  I thought the project was meant to assess actual physical tools.

Re: consistency/aberrations.  What you are describing is an outlier datapoint.  (e.g., Beltre 2004).  It can mess a lot of things up when you put too much faith into an outlier.

That said, we do have a few metrics to give us some insight concerning whether a breakout/breakdown season is the product of a new skill or luck (i.e., an outlier).  For example, a pitcher goes from being a 2.50 ERA guy to a 4.50 ERA in the same season that his BABIP goes way up.  We know BABIP is largely a function of chance.  I suppose a few investigations could be undertaken to see what factors might distinguish players who have an abberation season vs. developing a new skill.

Sorry, I meant “reputation” in terms of what the baserunners do.  If an OFer has a reputation for a powerful arm, justified or not, runners will not attempt the extra base and vice versa.  So an OFers arm value is his arm PLUS what the baseruners do.  The fans are only rating the first part of the equation, of course.

Here’s a Banner Years study of mine from a few years ago:
http://www.tangotiger.net/archives/artBanner.shtml

It looked at overall performance levels.  It’s possible that focusing on component-based will give you different answers.

Studes, I believe that the Earth is flat and although I am a pretty smart guy overall, that does not make it so.

Obviously, you’ve been reading Thomas Friedman lately!

Seriously, I don’t have the skill or mathematical chops to figure out whether consistency is an actual skill or not, or whether it is “significant.”

But count me among those who believe that people won’t come to a firm “no” conclusion based on any statistical analysis.  There will always be room for our opinions, thank goodness.

There is NEVER a firm answer when analyzing sample data. I have been preaching that for years.  I have also been preaching that there is ALWAYS a significant chance of the existence of a small effect when we cannot find any effect, also given less than an enormous data set (less than an infinite set actually).  All we can do is say that we think there is an X percent chance (or we are Z confident) that there is (or is not) a Y effect.  Which is usually good enough for government (or baseball) work.

Actually, in reality we are usually looking for evidence that a significant (however you want to define that of course) effect exists.  If we are lucky enough to have enough data, we can usually answer the question of whether, “There exists evidence that a significant effect exists or doesn’t.” Again, in practice, at least in baseball research, occasionally the results are equivocable.  Then, if we can increase the sample size being studied, or change the methodology in some other way, we can often come up with a “less equivocable” (I guess technically ALL answers are equivocable when dealing with sample data, as I already said) answer.  Sometimes, we are simply stuck with an equivocable answer, or no answer at all, if you will.

As for opinions about factual things without any factual data to back them up…

We get enough of that from politicians.