THE BOOK--Playing The Percentages In Baseball

Good question!  Has anyone heard anything qualitative about his improvement this year?  Different arm slot?  Something unexpected (and good) during his 2007 injury recovery?  Simply a change in approach mandated from his coach?

His HR rate will absolutely shoot up next year, but the low walk rate this year tells me he’s definitely approaching things differently.  And given his success, he’ll certainly do the same again next year.

I voted $12/year, but as a GM I’d expect to be outbid by someone who goes up to $14 or $15.

When I ran this poll on May 19, 2008, here were the results:
http://www.insidethebook.com/ee/index.php/site/comments/is_cliff_lee_on_a_6_1_streak_or_a_0_1_slump/

The final tally was a: 4/34 :deal.

After 27 votes, it is now a: 4/43 :deal.

(Highlight the text between colons.)

This implied around a 2.25 WAR evaluation for Cliff Lee the first time, and is currently implying a 2.65 WAR evaluation now (if both contracts started in 2009).  I haven’t run the Marcels yet, but that second one seems pretty low.  I would guess that he must be at least a 3 WAR pitcher, if not a 3.5 WAR pitcher entering 2008.  Almost half the people have voted that he’s at most a: 10MM :a year pitcher, and I just can’t see how this is possible.

I personally wouldn’t go over 10M, but that is more of my philosophy on how to spend money.  Meaning I’d much rather put that money toward draft and international signings than take the gamble on him being more than a one year wonder or getting injured.

That being said, I would think at least a hand full of teams would give him at least 15M a year after the season he just had.

We heard at the beginning of the year that his velocity was up this year and that he was hurt last year.

However, I think this is the classic case of a pitcher who got extremely lucky in one year.  I have watched him pitch several times this year, and he is, to put it bluntly, nothing all that special.  A very good pitcher - probably.  A great one?  Not likely.  You can see great pitchers almost every time you watch them, from their “stuff,” their command, and the way they pitch.  Lee is not one of those pitchers.

I am afraid (for the team that spends beaucoup bucks on him) that he is likely a one year wonder.

That is not to say that this year, lucky or not, is not part of his “skill set” or that he does not have a good projection going forward - it is, and he does.

Let’s also not forget that pitchers usually get paid too much money in long-term deals because most teams do not properly consider the injury, decline, and attrition factor.  The basic “rule of ten” is that on the average pitchers throw 10% fewer IP every year going forward.  They also lose about a tenth (or more) of a run in ERA (I think) going forward.

Lee has thrown his fastball about 1.5 mph faster than he has in the past.  He uses the fastball a lot, 70% according to fangraphs.

I agree with MGL, he probably just had a career year.  It is possible for a slightly above average pitcher to put up a year like this with no other explanation than random variation.  I’ve seen it before.

After 53 votes, it’s 1MM more than what I said in post 2.

Basically, the first 27 votes said 4 years, x dollars, and the next 26 votes said x plus two dollars.

The main difference, I think would be in the change in K/BB differential.  He went from being +54 per 1000 batters in 2007 to +148. (Excludes IBB, includes HBP.) He was +75 in 2006.

I’ll just go out on a limb here and presume that of pitchers with at least 400 PA in back-to-back years, that puts him in the top 10 in the last 50 years.  And, of those 10 pitchers, in the following year, their KBB component numbers were far closer to the year T-1, than T-2.

Ah, what the heck… I’ll go check, and report back in a few minutes…

Going from +54 to +148 (a gain of +94) puts Cliff Lee in the top 40.

Included in there, ahead of Cliff Lee, is Pedro 98/99, when he went from +188 to +321!  That’s not really what we are talking about here.

Mark Langston went from -33 to +113 at age of 26.  Not really what we are after either.

Ok, let’s make it a little tighter.  Let’s look for all pitchers with a KBB differential of no more than +100, and who is between the ages of 28 and 32 (Lee is 30).  Under those conditions, what Lee did puts him 15th in the last 50 years.

Among recent pitchers we have: Mike Scott (+161), Ron Villone (+126), Daryl Kile (+120), Matt Clement (+105), Mark Davis (+104), Juan Guzman (+104), Trachsel (+103).

The leader is Pedro Ramos (+161). The most recent was Shawn Chacon (+95).

I’ll look at the top 30, and let you know how they did the year after…

I’m also only interested in guys with at least 400 PA in the third year, so that’ll knock out some guys.

(By the way, I made a little boo boo.  I subtracted HBP instead of adding it.  So, disregard the previous post.)

The most recent example we have is Gil Meche, going +22 to +101.  The leader is Pedro Ramos at +161.

Anyway, I took all the pitchers with at least a +70 gain in KBB differential, aged between 28 and 32, with at least 400 PA in the two years in question, plus the following year.

Those pitchers averaged +46 differential in the first year, and +134 differential in the second year.  That gain was +88, a bit worse than Cliff Lee.  So, I think we’ve got a pretty good comparison group here.  In the third year, those 32 pitchers had a KBB differential of +94.  That is pretty much half-way between the two seasons in question (actually 55% the closest year, and 45% from two years prior).

This is pretty much what we should have expected with a Marcel forecast, which presumes a 3/2 weighting scheme.

So, unless we think Cliff Lee did something different, compared to Mike Scott, Mike Moore, Steve Carlton, Andy Pettite and the other players with a huge turnaround in their KBB differential, I think we’re going to have to presume that his Marcel forecast will be a decent forecast.

The K/BB change isn’t the only huge change - he’s also gone from being one of the most extreme flyball pitchers in baseball to an above average groundball guy.  His FB% by year since 2002:

44.8%, 43.6%, 44.7%, 43.8%, 48.4%, 49.7%, 35.1%

A 14% drop in flyball percentage is crazy.  Most of those became groundballs (a 10.6% increase) while some became line drives (a 4% increase).  If the FB% decrease is real, we’d have to adjust his HR rates down as well. 

The increased velocity, huge spike in groundballs, and drastic cut in walk rate would all seem to suggest that he reinvented his fastball this year.  I’m sure it is a career year, but there’s a dramatic in approach at the bottom of it.

Great point about GB/FB.  I need to get cracking on my PBP DB so that I can answer these questions better for the other guys.

The run value of a GB is about .10 runs better than a FB (including HR).  So, you should adjust Lee a good .25 runs per game, just on that basis.  That’s substantial.

Like I said, we’d have to look for his historical twins to see how persistent this change is (or is likely to be).

I voted $14 a year on this one and either 10 or 12 per year in the May poll (can’t remember which). But in looking at the numbers after voting, I’d change my vote to 10 per year at the most. His xFIP was 3.69 according to THT, and if that regresses to about 4.00 in 2009 (and progressively worse after), I don’t see how a 30-year old pitcher can be worth more than 10 a year.

The current implied WAR (based on the salary value you guys have) for Cliff Lee is 2.7 wins.

His WAR entering the season was around 0.5 to 1.0 wins.  His WAR this year (using WPA/LI) was around 7.0 to 7.5.  If we agree that his true value is half-way between those two figures, his WAR should be 4.0.  Apply a little aging, and 3.5 should be our expectations.  I cannot believe the call so far is for 2.7.  There seems to be a fantastic amount of reservation here by the readers of this blog.

I’d like to see Hardball Times, Primer, and Baseball Prospectus run a similar poll.  And LetsGoTribe.com .

I’m writing an article on Cliff Lee for the Hardball Times Annual.  I won’t go into all the details here because I’d like you to buy the annual to read it, but suffice it to say that he changed a lot of things between 2007 and 2008.  He changed his pitch repertoire, he added fastball velocity, his command improved, etc.  Dave is right when he says there was a dramatic change in approach for Lee this year.

Ya got to be careful about saying that his “approach” changed and therefore his Marcel is not accurate, because X, Y, and Z changed.

Depends on what X, Y, and Z is.  If X is velocity and it is sustainable, then I have no problem.  However, to say that his command improved, well, no duh. That is one way that a pitcher gets lucky!  It doesn’t just have to be balls in play.  How do you think that DIPS or FIP fluctuates randomly?  K rate increasing.  BB rate decreasing.  And HR rate decreasing.  Obviously.  All or part of that could be luck.

So could GB/FB ratio.  After all, it is basically a binomial subject to normal binomial variation.  Just about everything a pitcher does or the result of his pitch or BF is a binomial with a certain “p”.  Therefore anything can happen randomly.  About the only thing that does not apply to is velocity and spin.  It certainly applies to location, and the result of every pitch and every PA.

Now, if you tell me that his FB/GB ratio changed because he is throwing different pitches, well, maybe that is NOT just a random fluctuation.

But there are lots of ways to fluctuate randomly.  BABIP is not the only way.

Anyway, looking forward to the article as well as my free copy of THT!

A 14% drop in flyball percentage could be random variation, but the probability of that over 220 innings has to be very, very low.  It’s certainly the biggest drop I can remember. 

AJ Burnett had an 11% drop in 2005, and while his FB% regressed back towards previous levels in future years, it’s never gotten back to where it was in 2002-2004. 

Chris Carpenter had a 10.6% increase in his GB% in 2004, and he’s sustained his new GB% ever since.  In fact, Carpenter’s 2004 season is probably the closest thing we’ve seen to 2008 Lee, where a pitcher with some ability but a history of inconsistency becomes an ace out of nowhere.  For Carpenter, it was certainly the beginning of a new performance level. 

There are probably a few other examples of guys who had double digit swings in their GB/FB percentages, but those are the only two I remember off hand.  With Burnett, the Marcel would have been accurate.  With Carpenter, it would not have been.

MGL, I think I have a method to measure command.  It needs quite a bit of refining, but the basic premise is sound, I believe.  I intend to present it in a basic form in the article. 

There is a very clear reason that Lee’s command improved that has to do with his changed approach.  It’s not based on results like strikeouts, walks, and home runs.  It’s based on which pitches he threw to which locations in which situations.  The difference is so dramatic from 2007 to 2008, it’s clear that he intentionally changed his approach.  Obviously, I’m not inside his head to know why he changed what he did, but I can tell that he did make major changes.  Also, he has made some comments about what he has done with his delivery that line up with what I see.

After 111 votes, the average is 4/44.4.

If you knock out the top and bottom 5% of the votes, you get 4/46.  Knocking out the top/bottom 10% gives you a 4/46 deal.  This implies a 2.8 WAR.

On the other hand, I was using a 4.84MM per win, presuming 10% inflation.  If we use the old 4.4MM per win, this implies a 3.0 WAR.

So, it seems that the readers of this site might be considering his value a bit on the conservative side, but not that much on the conservative side.

Mike, no matter what, you cannot get true “anything” (except, as I said, velocity, and even then, not 100%) from the sample data.  That includes command.  And anything and everything else.  So while our estimate of a player’s true value with regard to anything always changes as his sample performance changes (the so-called running average), there is never 100% certainty in that.  So too of command!

There is nothing you can do about it!  When a pitcher throws a pitch he has a certain location in mind. While every pitcher has a unique true talent associated with how often he hits his spot, misses by an inch, 2 inches, etc., the actual result is random around that mean.

Theoretically, a pitcher can hit his spot 100 times in a row and have the worst “true command” in baseball, just like it is theoretically possible that Bonds was actually the worst hitter in baseball (for his size, swing, etc.) who just got lucky (to the tune of 1 in a google of course).  For Lee, if the data suggests that he has changed something about his mechanics, approach, etc., such that his command was much better, then sure, our estimate of his command going forward changes, but there is still uncertainty in that.  His change in command is still a combination of a different talent level AND luck.  How much is one and how much is the other, we can estimate on the average, but we don’t and will never know for sure.  And yes, it is true that if pitcher A and pitcher B both improve command by 14% (whatever that means), but we find out that pitcher A actually lowered his leg kick (like, say Willis lately), and pitcher B did not, it suggests that pitcher A’s changes are more predictive than pitcher B, but we STILL don’t know how much of either pitcher’s changes are luck and how much are a change in skill, so we still have to apply some kind of Marcel.  A player’s previous performance never disappears and league means never disappear from the equation.  Two very important points for everyone to remember.

So while you can make inferences and assumptions about how a pitcher changed his approach and you can estimate what his “command index” or whatever you want to call it, is, you HAVE TO understand that you are working with sample data which has noise in it, and you are only estimating true talent.  And you have to understand that virtually all data has to be incorporated into a projection, although if you know what you are doing, you can often discount certain data (as long as you are very careful about your reasoning and methodology).

I think this is an important point lost on most people.

Again, I look forward to your article Mike, but please keep these points in mind when you draw conclusions about any player going forward.

What is amazing is that Lee is 4/44 and the Cards, a smart team, just gave Lohse 4/41.

MGL, I think we’re pretty much on the same page.  Having been trained as a physicist, I am keenly attuned to the fact that every measurement has an associated error.

I’m also not in the projections/predictions business.  A lot of people who read my Brian Bannister or Jack Cust analyses this spring assumed that what I said about how they were successful in 2007 meant that I thought they would carry that over to 2008.  In truth, I am agnostic on that point.  I don’t think we know much yet about how to regress or incorporate information garnered from PITCHf/x into projections about the future.

Jack Cust killed fastballs in 2007, and pitchers didn’t adjust and throw him more breaking balls throughout 2007.  We didn’t know whether or to what extent both of those trends would continue. (I haven’t looked to see how Cust did on those things since this May.) Right-handed hitters didn’t/couldn’t pull Bannister’s fastball in the air in 2007.  We didn’t know to what extent that would continue in 2008.  It turns out that RHH pulled a lot more of his fastballs in the air in 2008.

So I don’t feel competent to be in the projections business yet based on my knowledge of players gained through PITCHf/x.  Maybe someday I/we will get there.  In the mean time, I get great enjoyment from analyzing players and figuring out how they tick, even if I don’t know what that tells me about next year.

In the day or two since I wrote my previous comment, I’ve seen a couple references to other pitchers who “changed their approach” in 2008, one here on this site (Lohse) and one on another site (Davies).  “Changed their approach” can mean a lot of different things, and some of them might need to be regressed very heavily while others might have more carryover to 2009. 

The comment about Davies was that he had changed his approach to be much more aggressive in his last three starts and how his good results in those three starts are a good sign for his performance in 2009.  I’m pretty skeptical about that.  It could be true, of course, but I wouldn’t bet on it.  Three starts is an awfully small sample size.  I would guess that many young pitchers are “aggressive” and get good results for three straight starts.  Many of them probably don’t have that carry over well to the future.  The fact that these came at the end of the season doesn’t mean much.

The comment about Lohse was that he had changed his pitch mix to left-handed hitters.  Yes, he did.  That’s something more likely to carry over to 2009, in my guesstimation.  However, we don’t know (without analyzing more in depth) whether that change in pitch mix was responsible for an improved performance against LHH.  And we don’t know to what extent those kinds of changes tend to be permanent and to what extent the results tend to regress. (Or viewed from another angle, we don’t know how much of the change in results was related to skill and how much was related to chance.)

With Cliff Lee, his change in approach was so dramatic, I’m sure he’ll do the same thing next year, approach-wise, that he did this year, but I’m not claiming to know anything about what that says about the validity of his Marcel.  I’ll leave that for someone smarter than me to figure out.

I’ll just say that he added two pitches--two-seam fastball and slider (he had those pitches in his arsenal in 2007 but he used them extremely rarely in 2007), and he made a huge change in how he locates his fastball.  He also improved his command, which I’m using to mean the ability to consistently put the pitch where he wanted it, and added about 1 mph to his fastball speed.  Some of those things might not be that big a deal in and of themselves, but added together, they make Cliff Lee look like a very different pitcher this year than last.

I recognize that it is easy to conflate a change in approach and a change in results.  The latter has some element of chance, perhaps a large element of chance, involved.  It’s easy to go looking for what changed in the approach when you know that something changed in the results.  I have tried in all my articles to be forthright about this challenge.

On top of everything else, a pitcher (or batter) may change his approach, be very successful, and then his opponents make adjustments and all of a sudden that success is reduced or even eliminated (if it is not optimal).

So yes, it is difficult to tell to what extent any change is, one, intentional, and likely permanent, and two, likely to have continued success.

All that being said, while I am skeptical, like you, of many players changing their approach and being more successful (everyone who has a couple of games of success has “changed their approach, right?), there is little doubt that Lee is one of those pitchers who did in fact experience (whether intentional or not) some major changes which has led to dramatic success.

My principal point with Lee is that you don’t have to watch him for very long to see that he just doesn’t have the “stuff” of a truly great pitcher and than this year’s numbers are most likely a fluke to some degree.  It’s not like we have a guy who throws 95-96 that we expected great things from and he finally had a great year. I think that mainly Lee just had a once in a lifetime year as far as command goes and as soon as he has a normal year of command for him (whatever that is - he is no Maddux though), his numbers will be eminently reasonable - an above average starter by maybe half a run at most and that’s it.

Sounds like on Mar 1, 2009, MGL will be issuing a challenge to all-comers.

***

MGL is right that he may have made real changes, but that doesn’t mean that the other side won’t adapt (or that he can keep it up).  As someone else noted on another topic, my favorite analogy is a weather pattern.  You can have a hurricane, and it’s real, and it does real damage, and it comes along again, and it’s real, and then it doesn’t do the same damage. 

We need to establish the level of persistence of whatever changes he has made, in terms of outcomes.

After 127 votes, the mean is 4/45.  Taking out the top/bottom 10%, and we have a 4/46 deal.

I think we’ve pretty much reached consensus.

After 127 votes, the mean is 4/45.  Taking out the top/bottom 10%, and we have a 4/46 deal.

So, Cliff Lee = Kyle Lohse.  Huh, didn’t realize that.

So, Cliff Lee = Kyle Lohse.  Huh, didn’t realize that.

Amazing, isn’t it.  Yet, not too many people are criticizing the Lohse deal.  Obviously, Lee is going to get paid a lot more than that.  Is he a FA now?

I think Lee has 2 years left… what a sweet sweet deal for the Indians.

Would you be able to briefly explain the methodology behind the 4.4M per WAR on the open market?

Thanks.

You can get all the particulars in last year’s THT Annual (or was it the 2007 one?), but the short of it is that if you take all the dollars spent on free agents and divide it by the wins above replacement those players generated, you get some 4 or 5 million$ per win.

Tango #29: I thought $4.4M was last offseason’s number for a WAR? Shouldn’t this year’s be $4.48M or whatever, or am I wrong to assume a 10% inflation this year as (I think) has been the case when you did this in the past?

Oops. I meant $4.84M as this year’s approximate number for a WAR… I can do basic math. I just can’t type.

Right, I’m presuming 10% inflation.  I don’t want to say yet for sure that that’s what it is, but that should be our operating philosophy.

Thanks. So am I right in assuming this year’s number for a WAR $4.8M until we hear otherwise?

Yes, I am using 4.84, until someone presents me evidence to the contrary.