THE BOOK--Playing The Percentages In Baseball

Wednesday, September 19, 2012

Mail of the month: platooning players and leveraged value in WAR

An excellent question:

Hi Tom,

Long time reader, first time writer. It occurred to me that maybe the WAR calculations by Fangraphs (and most implementations) are overrating hitters that are heavily platooned. For example, at the time of writing, John Jaso has 47 PAs against LHP, and 282 against RHP. This is a very heavy platoon, as I only recall once or twice he’s made a start against a lefty.

So my question is this: if you have a RH batter and a LH batter that would both be replacement level over an entire season, and they have [replacement level] bats and league average platoon splits and they are perfectly platooned, they would combine for a WAR greater significantly greater than 0, despite being replacement-level players on an individual with no remarkable splits. How should this be accounted for? Why isn’t this brought up more often when describing players like Jaso, who, while he has been legitimately great this year against RH pitching, derives a lot of his “WAR” from the fact that he’s sat against lefties.

Thanks,
Eric

Excellent question! 

We see this also in fielding.  For example,  you can “hide” a poor fielder by putting him at 1B or DH, whereby their lack of opportunities prevents them from building up negative runs.  However, we account for this with the positional adjustment.

We see this also in pitching.  For example, you can “hide” a poor pitcher by making him a reliever and making sure he only pitches one inning.  We account for this with the role adjustment (relievers are compared to a much tougher baseline than starters).

Now, similar to the reader’s question, we can talk about the subset of relievers who are LOOGYs (left-handed one-out guys).  So, we can hide a particularly poor lefthanded pitcher by: a) making him a reliever and b) making him face alot of LHH.  A left-handed starting pitcher for example will face 20-25% of their hitters as LHH, but a LOOGY will face more than 50% of their hitters as LHH. (*)

(*) By the way, that’s another advantage for relievers over starters that we don’t talk much about, especially short-relievers.

However, we are not accounting for this with LOOGYs.  Should we?

Similarly and back to the reader’s question: should we account for starting players who play disproportionately against opposite-handed pitching?

I can see the arguments from both sides.

I’ll give you my argument that suggests we should NOT account for it: a LH-RH platoon combo at catcher will be more effective than a LH-LH platoon combo.  If the LHH and RHH of all four players in question have an overall .300 wOBA against a league average pitcher (70% RHP, 30% LHP), then aren’t we in fact getting more value out of the opposite-handed combo?  That the LH-RH platoon combo will let the LHH have a .310 wOBA, while the RHH will ALSO have a .310 wOBA?

In terms of true talent, they are all .300 wOBA hitters.  But in terms of impact, the LH-RH platoon combo has more impact.

But, do we really want to give that “value” to the hitters themselves?  Perhaps we should give the credit to the team for being able to construct their roster to maximize platoon effects?

Anyway, I’m still forming my thoughts here, so I’d love to hear what the Straight Arrow readers have to say about it.

Monday, April 30, 2012

Lefty-lefty matchup for the sake of lefty-lefty matchup?

According to this blog post, a healthy portion of fans still subscribe to this view.  Is the blogger properly relaying the view, or is it greatly exaggerated? (*)

(*) I think that great 1994 Expos didn’t have a lefty around, and it wasn’t an issue then.  Winning absolves issues.

The logical conclusion to this view is that if you have the worst lefty reliever in the league, that you would prefer to have him pitch to a lefty hitter, rather than have the best righty pitcher in the league (Verlander, Doc, Stras, Felix, Mo etc) face that very same lefty.  You know and I know that that’s bullsh!t.  That Verlander pitching against the advantage is better than the worst-lefty pitcher pitching with the platoon advantage.

The question on the table is how much that advantage for it to be in-play.  The basic platoon advantage, if you need a rule of thumb, as per The Book, is 20 wOBA points.  So, a .350 wOBA hitter will be .360 against opposite-handed pitching and .340 against same-handed pitching.  20 wOBA points is about 0.60 in ERA.  If you have a 4.00 righty ERA reliever and a 4.60 lefty ERA reliever, then it becomes a breakeven decision, as the 4.00 typical righty reliever will be about as effective as the 4.60 typical lefty reliever, against a typical lefty hitter.  Obviously, not everyone is typical, but that gives you the basis to make (or criticize) a decision.  This is why you can have a lefty Billy Wagner be a closer, even if he faces predominantly right-handed hitters in high-leverage situations: he was a GREAT pitcher, and it would be hard to find a righty pitcher in his bullpen that would be more effective against right-handed hitters(**).

(**) Though you could certainly have made a case for Dotel for a while there, the reality is that with a pairing like Wagner-Dotel, one was going to pitch in the 8th and the other in the 9th anyway, so it didn’t really matter how you lined them up.

Wednesday, March 02, 2011

Handedness platoon splits, 2002-2010

Good stuff.

Monday, October 11, 2010

Did Cox make the right decisions in game 3?

If you are a Braves fan, you are justifiably emotional about last night’s game.

Here are the 3 major criticisms I have read:

1) Not pulling Conrad for Diory Hernandez in the 9th, for defensive purposes.

2) Kimbrel throwing a slider to Sanchez on the 1-2 count.

3) Choosing to bring in Dunn, the lefty reliever, to pitch to Huff, who has a higher BA against lefties this year.

OK, number one is a legitimate beef I think.  I do no know how good a defender Hernandez is.  I have a feeling that Cox never even thought of replacing Conrad in the 9th.  I don’t think any of the commentators did either.  No one remarked about it on TV that I remember.

Despite Conrad’s recent woes in the field, is he really that bad of a fielder?  According to Chone’s TZ for the minor leagues, he was a slightly below average defender at 2B. Now maybe, with all his miscues recently, he was nervous and thus a terrible fielder at the time.  I don’t know.  Only someone like Cox might know that.  And of course if Conrad is much the worse fielder, how much does that actually cost in terms of WE?

Number two is somewhat of a legitimate beef I think, although it is very much a 20-20 hindsight thing.  If Sanchez had gotten out on that slider, no one would have said a thing.  Plus, it was a little bit of a hanger.  Had Kimbrel thrown that down and away, like he wanted to, again, no one likely would have said a thing.  Yes, he has a good fastball, and yes, Sanchez had just swung at some fastballs out of the zone (and missed them badly), but is he supposed to throw all fastballs?  He has a good slider. If he had thrown a fastball right down the middle and Sanchez had gotten a hit, fans could easily have criticized him for throwing too many fastballs.  I did not think that the slider was a terrible pitch to throw in that situation. He obviously missed his location which can happen with any pitch.  I don’t think there is much of a legitimate criticism here.  A little maybe, as I said, but not a lot.

#3 is way off the mark, I think.  There are several reasons why.  First of all, despite the fact that Huff is hitting better versus lefties, BA-wise this year, all lefties batters have positive platoon splits.  Virtually no lefty batter has a reverse platoon split. In fact, if you look at Huff’s last 3 years, he has a 1.15 platoon ratio for OPS, and even a 1.08 ratio for BA.  If we go back 5 years, it is even higher than that.  So, the idea that Huff’s higher BA against lefties in 2010 means anything (predictive) is nonsense.

Now, when you bring in a same-side pitcher, it is not just the batter’s true platoon ratio that you care about, but the pitcher’s as well.  Let’s say that Huff had a 1.00 ratio or even a reverse one.  If you brought in a pitcher with a large platoon ration, it still might be correct.  The platoon ratio for a particular batter/pitcher matchup, like anything else, is roughly a log-5 (odds ratio) combination of the pitcher and batter’s ratio.

And obviously the overall true talent of the pitcher has to be considered as well. If you bring in an overall (much) worse pitcher just to get a platoon advantage, it still might not be correct, although when you bring in lefty against lefty, in most cases you gain so much from the platoon advantage that even a mediocre overall lefty becomes a very good pitcher when facing a lefty batter.

So what about Dunn?  What is his true platoon ratio?  In a small sample, his MLB career platoon (OPS against) splits are large - 1.29.

For all pitchers, I compute estimated platoon splits from their actual numbers in the majors and minors, regressed the appropriate amount, given the types of pitches (and frequencies) they throw.  For Dunn, I have that number at 1.27.  For Kimbrel, based on his minor league stats, I have that at 1.31 (mostly because he throws the 2 pitches with the highest platoon ratios - fastball and slider).  So it looks like Dunn is WAY the better choice for Huff than Kimbrel.  What about Moylan or Farnsworth?  Moylan is a siderarmer. He has an estimated true split of 1.41. And Farnsworth’s is 1.21, also a large platoon split, probably because he throws mostly fastballs I think.

Finally, I also, in my computations, estimate the average ERA that a reliever would have against RHB and LHB with average platoon splits:

Dunn

vs. RHB: 3.80
vs. LHB: 3.03

Kimbrel

vs. RHB: 2.91
vs. LHB: 3.75

Farnsworth

vs. RHB: 3.23
vs. LHB: 3.93

Moylan

vs. RHB: 3.51
vs. LHB: 4.86

So, it looks like Dunn was clearly the right choice, even if Huff had a slightly smaller platoon split than the average LHB (and there is not much evidence that that is the case - in fact, my estimated OPS true platoon split for him is quite high - 1.21 ).

Thursday, June 10, 2010

Regression equation for types of pitches versus platoon ratio

I created a data file containing the following variables and ran a multiple regression (using Excel) on it:

All pitchers in 2006-2009 who faced at least 300 RH batters (excluding IBB and SH).

All pitch type percentages were taken from FanGraphs.  Pitch type percentages do not add to 100% for each pitcher because I did not include in the data file cutters and knuckleballs - only fastballs (FB), sliders (SL), curve-balls (CB), change-ups (CH), and splitters (SF).

I computed each pitcher’s platoon OPS ratio for each year.  Platoon OPS ratio is simply opposite hand batters’ OPS divided by same-hand batters’ OPS, where OPS includes HPB, SF, but not IBB.

So each line in the data file was for one pitcher and one year (for example, Halliday, 2006), and included:

Variable X1: FB %
Variable X2: SL %
Variable X3: CB %
Variable X4: CH %
Variable X5: SF %
Variable Y: Platoon OPS ratio

N was 424 (424 lines of data, each line being a pitcher and year).

Dave Allen (I think - please correct me if that is wrong) in one of the THT annuals and on the THT site computed the following platoon differentials (runs per 100 pitches I imagine) based on the run values of each of the types of pitches. These are for RHP only and are “adjusted” (you’ll have to read the article to know what that means - I forgot).  I think he used pitch f/x data, but I am also not sure about that.

pitch     platoon
Slurve 1.12
Sinker 1.08
Heater 0.80
Slider 0.57
Rider 0.56
Cutter 0.41
Jumping fastball 0.27
Rising fastball 0.21
Power change 0.01
Tight curve -0.13
Roundhouse curve -0.69
Straight change -0.77

Anyway, just using the 5 pitch categories from FanGraphs that I mentioned above, here is the regression equation I came up with. Please tell me whether it looks right or not.

RHP (facing > 299 RHB in each pitcher season)

1.004 + .139 * FB + .224 * SL - .246 * CB - .008 * CH - .211 * SF

LHP (facing > 299 RHB in each pitcher season)

1.049 + .349 * FB + .119 * SL - .414 * CB - .582 * CH + .679 * SF

As a sanity check, if we plug in the average values for all RH and LH pitchers (who faced at least 100 RH batters in each year), we get the following predicted platoon ratios for all of these pitchers combined:

RHP 1.098
LHP 1.160

The average percentages from Fangraphs for 2006-2009, for all RH and LH pitchers who faced at least 100 RHB in a year was:

RHP

FB: 61.0% SL: 15.5 CB: 8.7 CH: 9.1 SF: 1.6

LHP

FB: 59.4% SL: 14.1 CB: 8.8 CH: 13.6 SF: .3

Sunday, February 14, 2010

How to do splits forecasts

I love it when the readers do stuff like this.  Good job.

Monday, February 08, 2010

How do you spot a lefty masher?

Pat asks:

However, lefty/right analysis has advanced since my adolescence, and I think this post from MGL is a must read for what I’m talking about. In it he says:

IOW, how a batter does against RH pitchers informs us on how he will likely do against LH pitchers and vice versa. Why? Because there is not much of a spread in true platoon splits among ML baseball players yet there is a large spread in overall true hitting talent among ML baseball players. So if we see a large platoon split, like for a player like [Ryan] Howard, it is likely a fluke. If a player does really well versus RH pitchers but terrible against LH pitchers, both the “really well� and the “terrible� numbers are likely fluky and the “truth� is somewhere in between.

Howard has a .719 OPS in the last 4 years versus LHP. How would we estimate his “true� OPS versus LHP? You might be tempted to just use the .719, which is not too good or you might be tempted to use the .719 and then regress that toward the league average for a LH batter of Howard’s physical characteristics, which might be around the same or a little higher – I don’t know. Both of these methods would be wrong. You cannot ignore the fact that he also hit 1.052 in OPS versus RHP over the same time period (last 4 years) and in many more PA. This suggests that he is a very good hitter overall (which he is) and that the .719 is somewhat of a fluke.

And then Pat asks the interesting question: do splits change by age?

I’d also like to see how players do split wise over the course of a career. Obviously the skill of hitters diminishes over time, but it’d be interesting to see if the splits are larger. Ryan Howard certainly is not “old” at 30 years old, but his skill set and physical size certainly have shifted over the years. While he was never slender, Howard is certainly a “bigger” guy than he used to be, and probably a good amount slower. Besides that, as players get older they tend to lose some hand-eye skill, an effect that may be magnified when facing a pitcher of the same handedness. Here are Howard’s wRC+ from 2006-2009, going from overall to versus lefties and then versus righties.

2006: 166, 133, 182
2007: 140, 110, 159
2008: 123, 91, 143
2009: 141, 71, 178

Fangraphs now has some Splits data

Finally. 

Personally, I’d like to see each split data have its own subtab under the Career/Year tab.  If I’m interested in L/R splits, I’d like to see the STandard / Advanced / Batted Ball all together, not navigate through the other splits I don’t care about (at that moment).  Also, as more splits are added in (like men on base, role, park, etc), it just makes this more cloudy.  Basically, splits should extend horizontally (add more tabs) rather than vertically.

David’s taking suggestions, so post it here or there.

***

I linked to Granderson’s splits, who has a .270 wOBA against LHP (685 PA) and .380 against RHP (2211 PA).  (If David wants to wow us, when he does his leaderboards, give us the “differential”, and also shows us the batter handedness on that table.)  Andy said that for LHH, that you would regress the observed split (110 points in this case) 50% toward the mean if you had 1000 PA against LHP.  In this case, with 685 PA, you would regress 1000/(1000+685) or 60% toward the league observed splits for LHH, which I think is like 27 points. 

So, you regress 110 60% toward 27, or 110*.4 + 27*.6 = 60.  So, our estimate of Granderson’s handedness split is .060 in wOBA.  He’s a career .358 wOBA, with 24% PA against LHP.  We can reconfigure his observed split to be:
.358 = .24(x) + .76 (x+.060)

So, his LHP adjusted-observed split is .312 and his RHP split is 60 points higher or .372.  This is his observed baseline.  If he’s a true .350, as opposed to his observed career .358, you bring him down by 8 points on both sides.

In any case, with his excellent fielding, even his poor hitting against LHP means he’s an average player.  When you are an average player against the platoon, this is not really the kind of guy you need to be platooning.  (Unless of course you already have a good platoon partner, so you might as well take advantage when you can.)

***

Here is Andre Ethier in high-leverage situations.  Dave in the THT Annual noted how Ethier had back-to-back years of great high-leverage performance.  We see that this can extend even to 2007.  Here are his wOBA in hi-lev, 2007-09: .403, .451, .454.  (Hi-Lev must be based on LI of at least 2.0, since we have about 10% of Ethier’s PA.  This is different from b-r.com’s hi-lev that uses LI of 1.5 or higher, which is about 20% of PAs.)  Anyway, compare that to his overall career wOBA of .363.  His incredibly sucky hi-lev in his rookie year really drags down his career split in hi-lev.

Anyway, love this stuff.

Wednesday, October 28, 2009

Is Ryan Howard really that bad versus LHP?

Many people, including some analysts, have suggested that Howard should be platooned.  And that the Phillies might be in trouble in the WS because the Yankees are going to throw 4 or 5 lefties in 7 games.  I think that the conventional wisdom is simply that Howard is not a good hitter versus a LHP.

I am here to challenge that wisdom of course.  But, many of you will say, look at his stats versus LHP.  An OPS of .719 in 890 PA over the last 4 years!  Not too good.  Not good at all.

I say, stats, schmats!

Try this one on for size:

Say we have two players. Both are left-handed. Player A has an actual wOBA versus RHP of .450. Against LHP it is .300. Again, those are actual numbers in say 3 full seasons.

Say player B is .350 versus RHP and also .300 versus LHP in the same number of PA (versus both RH and LH pitchers) in 3 seasons as well.

Who has the higher projection versus LHP and by around how much?

The answer is that Player A has a much higher projection versus LHP than player B, even though they both had the exact same historical performance in the same number of PA versus LHP.

How can that be?

Because how a batter does overall (versus lefties and righties) tells us more than just how he does against LH or RH pitchers alone, even if we want to estimate how they will do in the future against either lefties or righties alone.

IOW, how a batter does against RH pitchers informs us on how he will likely do against LH pitchers and vice versa. Why? Because there is not much of a spread in true platoon splits among ML baseball players yet there is a large spread in overall true hitting talent among ML baseball players. So if we see a large platoon split, like for a player like Howard, it is likely a fluke. If a player does really well versus RH pitchers but terrible against LH pitchers, both the “really well� and the “terrible� numbers are likely fluky and the “truth� is somewhere in between.

Howard has a .719 OPS in the last 4 years versus LHP. How would we estimate his “true� OPS versus LHP? You might be tempted to just use the .719, which is not too good or you might be tempted to use the .719 and then regress that toward the league average for a LH batter of Howard’s physical characteristics, which might be around the same or a little higher – I don’t know. Both of these methods would be wrong. You cannot ignore the fact that he also hit 1.052 in OPS versus RHP over the same time period (last 4 years) and in many more PA. This suggests that he is a very good hitter overall (which he is) and that the .719 is somewhat of a fluke.

Anyway, what would be the correct method for estimating Howard’s true OPS versus LH pitchers using his last 4 years of stats?

First we’ll estimate his overall true OPS. In his career, it is .966. We’ll regress that and call it .930, which is a typical projection for him. Now we have to take his observed platoon ratio (I like to use ratio – some people use a differential) and regress that. His observed platoon ratio for those 4 years is 1.052/.719, or 1.46. For the average lefty, it is 1.20 and we just don’t see that much variation among players in their true platoon splits. IOW, that 1.46 is likely very (but not completely) flukey. We might regress that 1.46 80% toward the league average of 1.20, to get 1.25. That is Howard’s “true� estimated platoon split.

Now we simply apply that to his overall estimated true OPS of .930 and the fact that he faced 62% RHP. That gives us an OPS of .805 versus LHP and 1.006 versus RHP.

.805 is a far cry from .719

So if we are going to talk about how Howard is likely to do versus LHP, let’s at least get our numbers right. .805 is higher than Werth and Jeter versus RHP. We don’t hear anyone talking about how terrible Jeter and Werth are versus RHP, which they face most of the time, do we?

Tuesday, April 22, 2008

The Greatest Platoon of all-time

Rob Neyer is proposing Frank Thomas (RHH) and Barry Bonds (LHH) be signed on some team as the DH.  That would be fantastic!

Thursday, April 10, 2008

Is there a hangover effect for hitters following a knuckler?

Editor’s Note: I’ve changed the name of this thread to better capture the exciting research that is appearing, starting at post 14.  MGL’s original blog post appears after the jump, and the URL remains as it was in its original form.

Read More

Friday, November 09, 2007

Personal Platoon Splits

Nate discusses platoon splits.

Secondly, these platoon splits will allow us to adjust for changes to player context. What I mean by that is, if a left-handed hitter moves from a division (like the NL Central) without very much left-handed pitching, and to a division (like the AL Central) with plenty of good left-handed pitching, we’ll able able to account for that in his projection.

Sounds both reasonable and unreasonable.  So, let’s go to the data and see how many PA LHH faced:

Read More

Wednesday, January 24, 2007

How Did Ted Williams Do Against Lefties?

There is a thread going on at Baseball Fever that discusses Ted Williams.  In the very first post, there is some fascinating data.  Below is my analysis of that data, and I encourage others to take on a more exhaustive study, especially focusing in on lefty hitters at Fenway.

Read More

Friday, January 12, 2007

Switch Hitters

I recently learned that Google lets you easily publish spreadsheets online.  It’s very cool and nice.  Check out this handedness data, historically.  This is how to read the first line:

Read More

Thursday, August 24, 2006

Lefties and Righties

Righties throw faster than lefties, in MLB.  In The Book, I discussed the spread in platoon advantage for RHH and LHH is likely simply an artifact of the selection bias of the pitching talent in MLB.  Dewan’s data is great data to have.

Hat tip: studes

Friday, August 11, 2006

Edmonds, and the decline against lefties

This article discusses an old aging ballplayer whining about not being paid out his lottery ticket next year.

A 10 million$ contact against a 3 million$ buyout.  So, do you pay him the extra 7 million$?  He just has to be about 3 wins above replacement, meaning 1 win above average in full-time play.  Seems like a lock, unless the injury has changed his talent level.  Or…

Read More

Batter Pitcher Matchups

As readers of The Book know, this has little meaning.  Even so, lots of fun:
http://www.baseball-reference.com/friv/bvp.cgi?n1=raineti01&n2=hershor01

Wednesday, August 09, 2006

Pinch Hitting: No Bang Off The Bench?

Andy penned an article on pinch hitting at Baseball Prospectus:
http://www.baseballprospectus.com/article.php?articleid=5404