THE BOOK--Playing The Percentages In Baseball

LOB% is a lousy proxy for situational pitching. Which makes it completely useless, because we don’t need a proxy; all we need is component ERA.

Why does no one ever talk about component ERA, anyway?  Why doesn’t FanGraphs list it?  It’s a logical progression: ERA, ERC, FIP, xFIP, where with each step you remove one element of luck.  Leaving ERC out just makes people ascribe far more ERA variation to the BABIP fairy than is justified.

B-R.com shows OPS in the situational pitching.

I’d like for Fangraphs to show wOBA for pitchers in their splits in the advanced (and really, all their pages… wOBA for pitchers… that’s what I use).

Now that I made this suggestion, it’ll take about 0-3 days for David to consider and implement.  Dude is amazing fast.

ERC tells the story much more clearly than wOBA splits do, though.  Showing the situational splits but not showing ERC is akin to showing BABIP but not FIP, or showing HR/FB but not xFIP.

What’s important is being able to see how everything translates into runs, to be able to say things like “Pitcher X’s ERA was 0.75 higher than it would have been with normal sequencing.” Just listing the splits doesn’t do that.

"If you take the best pitchers according to say FIP and the worst according to FIP, you will find the LOB% will be much higher for the good pitchers than the bad ones.”

How much is “much higher”? In my mind, it ranges from 69% to like 76%. Beyond that is luck, but maybe I’m wrong.

LOB% is influened by skill.  For a given pitcher in a given year/month/day, it can be lucky or unlucky, but in general LOB% is strongly influenced by K rates, walk rates, and ground ball rates.  Relievers generally have a wider “normal” range of LOB% outcomes and a higher average LOB% overall than starters.

The range that you (BWoodrum/4) noted is accurate, and the normal range of 7% doesn’t seem like much on that scale but if you flip it around and think of it as “score rate” with a normal range of 24% to 31%, then you see that it can make a huge difference in relative run prevention.

Thanks for the feedback Ed D. If I could do the video over again, I’d like to change it so that it more clearly reflects that a 50% LOB is unlucky, but 70% may be normal. I think I rushed that part too much.

Here’s the article in which I showed the positive relationship between xFIP (just plain pitching skill) and LOB%.  I wrote this the same year I rolled out LOB%.

Studes, are we missing a link?  “Here’s the article ...”?

Oops!  Click my name.

http://www.hardballtimes.com/main/article/ten-things-i-didnt-know-a-couple-of-weeks-ago/

Or… not my name.  Just the link.

"69% to like 76%”

That is an ENORMOUS difference. 

A guy with 10 runners on base per 9IP makes the difference of .07 strand rate as .70 runs per game.  It’s huge.  And that’s based on their “true” strand rate.

I love the series you have, but calling strand rate luck is wrong.  I suggest you don’t push it as anything close to what BABIP is.

***

The point is this: when you have a runner on base, the reason he will score or not is based on how the next batter(s) will do.  And that is going to be based on how good the pitcher actually is.

The reason Mo and Pedro had ridiculously low strand rate is because they don’t allow players to get a hit or walk off them.  Players make an out with runners on base (and with bases empty).

"high strand rate” not low.

To try to hammer it home: the correlation of BABIP to FIP is close to 0.  The correlation of strand rate to FIP is going to be fairly strong.  The strand rate is a byproduct of a great FIP and a great BABIP with men on base.

There is luck involved, obviously, but the pitcher’s overall FIP is going to be a strong indicator, and over a period of a few seasons, it’ll be the leading indicator.

Let’s not talk about luck here…

All this talk has got me thinking an amendment is in order.

Would I be correct to say “pitchers with seasons far from their career LOB% are likely to return to their career norms” instead of trumping 72%? I realize this is basically just touching on regression to the mean (something deserving its own series), but I’d rather not mislead viewers.

That would be better, yes.

Follow-up questions: Can I use your likeness Tango? And by likeness, I mean: Can I portray you as an android cat?

If you do a search for
tango tiger
on images.google.com

I like the first two that pop up (plush toy, and dancing tiger).

This one is also funny:
http://www.celesteyarnall.com/graphics/DSC07014%20Tiger%20Tango%201sm.jpg

"To try to hammer it home: the correlation of BABIP to FIP is close to 0.  The correlation of strand rate to FIP is going to be fairly strong.  The strand rate is a byproduct of a great FIP and a great BABIP with men on base.”

I’m pretty sure the best pitchers in baseball also can change their BABIP with men on base.  I checked this for the top 5 pitchers in WAR over the last three years (and spot checked some hall of famers, too).  They affect BABIP by lowering LD-rate.

http://www.beyondtheboxscore.com/2011/1/14/1936033/pitcher-influence-on-line-drives-revisted-also-clutchness

So, the best pitchers in baseball increase their strand rate because they influence LD-rate in high leverage situations, thereby depressing BABIP in high leverage situations.

In his “DIPS Revisited”, MGL does notice a high correlation in LD frequency, which matches what you are saying here.

I prefer “Tigger with a calculator and pocket protector.” Woo-hoo-hoo-hoo.

Reminds me of the first time my 2 year old was at the zoo, “Daddy, that tiger has spots.” Classic.

Not on topic, but I thought this might be an interesting read for followers of this blog:

http://lanekenworthy.net/2011/01/11/is-winner-take-all-bad-or-good-for-the-middle-class-evidence-from-baseball/

14/BWoodrum, that is a better way to say it, but even the term career norms is misleading because skills diminish over time that may make a person’s expected strand rate next year lower than his historical strand rate.  It’s all about expected K rate, BB rate, and G/L/F splits.  Some pitchers can have an expected strand rate of 77% and some others only 68% (I use those numbers intentionally because they are outside of the typical boundaries that you set of 69-76%, yet they still might be the expected levels for a given high-skilled or low-skilled pitcher).

Here’s the best way to describe it.  Go here and do this:

http://tangotiger.net/markov.html

You MUST follow along, because this is the only way people are going to believe it.

1. Note that we have 1 HR, 13 other times on base (and 27 outs), meaning all this data is on a per game basis
2. Click CALCULATE
3. Runs scored per game is 4.905
4. That means of the 13 runners on base, they scored 3.905 runs (we removed the HR), or 30% of runners scored, or 70% strand rate.

5. Now, click BACK, and change the BB to 0
6. Click CALCULATE
7. We now have 3.515 runs per game
8. With 9 runners on base, and 2.515 of those scoring, that’s 28% of runners scored, or 72% strand rate

9. Now, click BACK, and change H to 7 and AB to 34 (maintaining 27 outs).  We now have only 6 non-HR runners on base.
10. Click CALCULATE
11. That’s 2.439 runs, of which 1 is a HR
12. So, we have 1.439 runs on 6 runners or 24% of runners scored, or 76% strand rate.

The pattern is unmistakeable and the direction (if not magnitude) is entirely consistent with our expectation.  If you allow a runner on base, but you don’t allow a great deal of runners to get on base, no one will push him across. 

That’s why the strand rate for Mo and Pedro and Maddux and Clemens and RJ will be much higher than for other pitchers.  The strand rate is a byproduct of their talent level.

This is NOT true, or not true to any great extent, with their skill at getting outs on balls in play.

23/ This makes perfect sense. Which, naturally, leads me to more questions.

The final situation has essentially an expected LOB% (xLOB% anyone?) of 76%. Isn’t it possible for it—in a real game—to end up 68% or something lower? Would that be luck (i.e. random variation) or is there something else?

Also, and this may open a wholly different can, but how do inherited runners play into this (with starters)? Many starters get pulled after one or two guys get on base.

First of all: Does that affect the starter’s strand rate, or does that get recorded as the reliever’s? (I’m not sure.)

Secondly (if it does indeed affect the starter): Wouldn’t a starter’s LOB% thereby be partly a product of the relievers following him—possibly moving him back towards league average? And how big is this effect? (I’m guessing not terribly big.)

Right, as I said:

“The strand rate is a byproduct of a great FIP and a great BABIP with men on base. “

The calculator at my site presumes a standard line across all splits situations.  But, if you have a .000 OBP with men on base, and a .500 OBP (all home runs even) with bases empty, guess how many runners you will strand: 100%.

Cliff Lee v Roy Halladay last year is a good example.  They had somewhat similar seasons, but Lee stranded only 68% of runners, and Doc stranded 83%.  That’s entirely due to their performance with men on base.

But in 2008-2009, they both stranded 77% of runners.  Basically, it’s mostly luck that Doc stranded 83 instead of 77, and Lee stranded 68 instead of 77.  But, it is NOT luck that they strand 77 to begin with.  (Well, some of it is luck.)

It’s a bit like saying RBI is 100% luck.  Well, no, it’s not.  You start with talent, you add in the talent of the runners ahead of you, and build in split performance with men on base.

Strand Rate is analogous to that.  Cliff Lee and Roy Halladay start with a high expected stranded rate because they are awesome, then you add in their fielders with men on base, and their own split performance with men on base, and voila, you get 83% for Doc and 68% for Lee.

If you do a followup video, you can maybe use Doc and Lee, showing how they start high based on their 2008-2009, and see how they diverge (based on luck) in 2010.

AND THEN, do the same using two bad pitchers, and showing how they diverge in 2010, enough to put one guy ahead of Cliff Lee.  I can look for those pitchers for you if you like.

"That’s why the strand rate for Mo and Pedro and Maddux and Clemens and RJ will be much higher than for other pitchers.  The strand rate is a byproduct of their talent level.”

Again, that’s only part of the story.  As I showed in the link above, the best pitchers actually decrease their BABIP via decreasing their LD-rate in high leverage situations.  I just looked up the 5 HoF pitchers you mention above.  RJ, Pedro, Roger, and Greg all have decreased LD-rate in high leverage situations(Greg by a little, RJ by a ton).  Funny enough, Mo doesn’t.  Don’t know why, perhaps because he’s always pitching in high lev. situations?

Preferential decrease of LD-rate in high lev. situations will increase strand rate above what FIP would predict.