THE BOOK--Playing The Percentages In Baseball

Friday, November 19, 2010

LI and WPA used to determine most exciting games

This is a great way to find out the best games, and Max provides a handy list of games at the bottom at the article.

Tuesday, October 19, 2010

Upset at not bringing in Rivera in a low leverage situation?

Neyer is absolutely correct:

This time, there was a tomorrow. Two tomorrows. You want to guess how many times in 2010 Rivera has pitched on three straight days?

Zero. He hasn’t done it, not even once. And while the Yankees scored so many runs that save opportunities were scarce in 2010, it’s a fact that Rivera didn’t pitch on three straight days even once. And Girardi might be loath to ask him to start now. Especially with two big games coming up, in both of which Girardi might want the greatest postseason relief pitcher in history to toss a couple of innings.

I’m usually a big proponent of being aggressive with your best relievers in big games. But with two terribly important games right around the corner, this might have been the time to keep it in the holster and live to fight another day. Or two.

The Leverage Index of being down by two entering the top of the 9th is 0.3.  You would like to bring in your ace reliever when the LI is at least 1.5, and preferably at least 2.0.  Anything under 0.7 is a definite non-ace condition.  The chance of winning the game is only 6.6%.

Now, if there were no games the next day, then, sure, you can use that bullet, because you get to replenish that bullet.  But, with two more games to go, it’s not clear that that bullet will get replenished in time.  Or, as Neyer notes in his article, bringing in Rivera down by 1 and with runners on base would have been defensible.  The LI there would at least have been over 1.0.

Putting in Rivera would have been a desperation move at a point in time when you don’t need to be desperate.

Friday, October 01, 2010

Fangraphs and Pitcher WAR

Just now, I’ve opened up my reader.  I’ve been fascinated with the PECOTA percentiles until now.  Dave has a post about Pitcher WAR at Fangraphs.  Two in fact:
http://www.fangraphs.com/blogs/index.php/why-our-pitcher-war-uses-fip/
http://www.fangraphs.com/blogs/index.php/why-our-pitcher-war-uses-fip-part-two/

Let’s see what he says… Ok, his first link sets the landscape for anyone coming into this thing late, or not fully versed.  Good job in the details here.  Now on to the second post…

Those are two perfect posts.  They are exactly on target, and shows that in sabermetrics, we can actually, gasp, have an opinion.  The very thing that we have been lambasted for (sticking to the numbers, and not watching the game) is exactly the reason Dave argues for why we need to have two WAR for pitchers.  That it’s not easy to get it right, and there’s alot of things to consider.

By the way, when I said earlier that I’ve been bugged by Fangraphs pitching WAR being all FIP, I am just as bugged by Rally’s position of pitching WAR too.  I’m still in the exploratory stages.  And note, my bugging is limited to “past performance” and Cy Young talk.  In terms of true talent, (single-season) FIP should be overweighted compared to the other components.  The more years you have, the less you need to overweight.

Friday, June 11, 2010

Game-state version of wOBA

This is going to require alot of effort from you, to read and to follow.  It’s a step-by-step kind of thread.  And anyone out there that has ever disparaged, questioned, or just been plain flummoxed by wOBA or WPA/LI, well, this thread may be for you.  I hope all your questions will have been answered after you read this.  (Though it is possible that new questions you hadn’t considered will pop up!)

It took me about an hour to write this.  I know I’m asking alot for your time and patience in return.  The reward should be there.  Let’s go…

Read More

Tuesday, June 01, 2010

Forcing the opposing manager to walk Pujols

Poz asks:

It was a hustle play for Ludwick, exactly the sort of hustle play that we admire in our players. He took advantage of an opponent’s mistake, got himself into scoring position, did exactly what you would want him to do.

Then again — as you have no doubt considered — there is another factor here. Albert Pujols was hitting next. And, of course, the Reds promptly walked Pujols. In other words, by taking second base, Ludwick took the bat out of Albert Pujols’ hands.

Which leads to our question: In the grand scheme of things, was Ludwick’s taking second base a good play or not?

Poz goes through the exercise to show that you’d rather have Orlando Cabrera with the bases loaded, than Pujols with runners on 1B and 3B.  Let me show you a more theoretical answer, and it starts with the average run expectancy chart, which should at this moment be in your back pocket.  For those of you who still have not printed it, or memorized it, go here.  (At one point in the future, I will no longer link to that page.  That chart should be in your wallet, opposite to the picture of your kid.  And if you have two kids, well, shrink their pictures so they fit on one side.  This is baseball dammit, and it should make up half your life.)

Anyway, with the bases loaded and two outs, and an average hitter at bat, that is worth 0.815 runs to the end of the inning. 

With runners on the corners and two outs, that’s 0.538 runs to the end of the inning?  What about with Pujols and runners on the corners and two outs? 

Pujols is a career .435 wOBA hitter, which is about 100 points above the league average.  To convert wOBA to runs, you divide by around 1.15 or 1.20.  So, +.100 / 1.15 = +.087.  So, on average, Pujols is worth about +.087 runs per PA above the average hitter, in an average base/out situation.  You can also look at his WPA/LI career total of 57.5 wins in 6312 PA, or +.009 wins, which is around +.090 runs per PA.  Fangraphs also has his wRAA (runs above average) as +566 runs, which comes out to +.090 runs per PA.  Any way you shake it, on average, Pujols is +.090 runs per PA.

Now, you might be tempted to simply take the 0.538 runs that an average hitter generates with runners on the corner and two outs, and add in Pujols’ +.090, and be done with it.  However, runners on the corner and two outs is not an “average” base/out situation.  In order to figure that part out, you need to know the “leverage” of that base/out situation, or boLI (base/out Leverage Index).  There’s also a simple way to figure out the boLI, and you need to go to this chart.  If you go down to the 1b/3b/2out line, you will see the run value of the out is -.54 runs, which you compare to around -.30 runs for the typical value of the out.  That sets the boLI at around 1.8.  That is, everything that happens in the base/out situation is magnified by 1.8.  You can see the full chart here.

If Pujols really is worth +.090 runs per PA on average, he’s probably worth around .090 x 1.8 = .162 runs per PA with runners on the corners and two outs.  (One could just as well use the LWTS figures for that base/out state and apply Pujols’ specific stats to get a more accurate number.  That would be more correct.  My point here however is to try to find a simpler way so that you can be more comfortable to use this process for other situations.) So, .538 + .162 = .700 runs per PA. 

And that becomes our estimate: Pujols with runners on the corners and two outs plus the average batters coming up generate .700 runs.  An average hitter with bases loaded, 2 outs plus the same average batters coming up generate .815 runs.  And that’s why you don’t see Pujols being IBB with runners on the corners and two outs.  It would have to be a pretty bad hitter for that to happen. 

How bad?  Well, it would have to be -.115 runs per leveraged PA.  And since the boLI was 1.8, that means -.064 runs per average PA.  Which is around -.075 in wOBA.  And if the league average wOBA hitter is .335, that would make our breakeven hitter around .260.  And that’s pretty much the worst true hitter possible in MLB, outside of pitchers-as-batters.  Therefore, in order to walk the best hitter in MLB with runners on the corners and two outs, you have to have the worst hitter in MLB on deck.

(This analysis also ignores the game situation, meaning inning/score.  That requires more leverage index talk.  And if the larger the game LI, the more it makes sense to walk Pujols with runners on the corners and two outs.  That’s the subject of another post.)

Sunday, May 09, 2010

Since I was asked…

...by more than one person.

Bill James once said don’t multiply or divide two numbers unless there’s a g-dd-mn good reason that you need to multiply or divide them. And, that’s what I thought about here.

So, you need to make the case for what you did.  I can see the ratio of Shuts to Melts.  I can see the ratio of IP*LI to G.  I do NOT see the reason to multiply the two sets of results other than more-is-better.  I also don’t see the need to divide by xFIP other than less-is-better.  In the end, what is the unit? 

I don’t want to discourage you in trying to come up with something.  But I do want to discourage you to just slapping things together like that. 

Saturday, May 08, 2010

Confirming validity of Shutdowns, Meltdowns

Rational Pastime makes a good case:

Friday, May 07, 2010

Shutdown and Meltdown lists

I love it when other people do the hard work.  One of the good ways to evaluate a stat is if it matches your preconceived notions to some extent.  Bill James had a great line that says that a stat that never surprises is probably useless, and a stat that always surprises is probably wrong.  What you want is for 80% confirmation, so that you can at least learn something. 

Does Shuts and Melts do that?  If it does, then it’s a decent stat.  If it doesn’t, then it’s not.

Tuesday, May 04, 2010

Goodbye Saves, Hello… Douses?

I don’t have a good word for it yet.  “Fires extinguished” would be what I’m looking for.  As Jeff notes, the idea of saves can be improved upon.  So, what can we track?

How about any game in which a reliever enters (regardless of score, ahead/behind) and he comes out with a WPA of at least +.05, we call that a “douse” (or whatever better word you can think of).  And if he leaves the game -.05 or worse, that’s an arson (or whatever better word you can think of).  Last year, Mariano Rivera had 38 douses and 5 arsons.  Just a matter of playing around with the threshholds to see what kind of results we’d like to get.

There’s a certain comfort in counting things in binary form like this.

Two points:
1. ignore if it’s sellable
2. we need a good name

Leverage Index to end of game

There’s something that’s been bothering me about Leverage Index (LI).

Let’s take the top of the 8th, bases loaded, 2 outs situation, home (pitching) team up by 1.  The LI is 6.8.  The LI says that the change in win expectancy is about 7 times as impactful as a random situation.  And, it’s easy enough to see.  But, by the time the top of the 9th comes up, the LI can be anything from 2.9 (if no runs score) to close to 0 (if there was lots of runs that had scored in the 8th).

What would be more interesting therefore is to see what the average LI would be from that point in the 8th inning to the end of the game (or at least to the end of the next inning, which is about when a reliever is expected to pitch… that is, up to 2 innings, with one sit-down half inning). 

I kinda tackled this issue in The Book already, where I compare the win% of a star pitcher and an average pitcher from some point in the game to the end of the game.  And the results were pretty much similar to the standard way I do LI.  But, I’m thinking more of these oddball high-leverage cases (basically men on base, two outs).  I don’t think I’d get similar results.

Anyway, if the top of the 8th situation had a 6.8 LI, it’s possible that from that point onwards, to the end of the top of the 9th, the average LI will be say 4.0.  It’s still a super high LI of course.  It’s just not the big LI that the 6.8 suggests (when you are thinking of it in terms of the manager to bring in his reliever).  The 6.8 is still true in terms of PA-by-PA.

Just something I’ve been thinking about, and I’ll generate the LI-to-end-of-next-inning one day.

Tuesday, April 27, 2010

Daily reliever usage charts

I’d like to see this for every reliever, for every season, as part of his charts:

And more.  Here’s what I propose: For every game, mark the highest gmLI for ANY reliever who entered that game in red, and the gmLI THIS reliever in blue (if he even played).  This way, you can tell how this reliever was used relative to other relievers that game, and if he was even used at all.  Maybe for THIS reliever, have it a blue bar, and the ANY reliever as the red dot.

Monday, March 22, 2010

Ace relievers

In a response to the Bill James note regarding relievers from pre-2003

1. We excerpted the “three-run lead” part of the relief chapter in SI when The Book first came out.  There’s some good stuff in there.  I’d say it’s worth five minutes of your time.

2. James was mostly right, but partly wrong, regarding leverage, especially the tie-game.  Really, it just comes down to the 3-run lead in the ninth inning.  It sometimes comes down to letting your ace pitch two innings with a 1-run or 2-run lead in the 8th.  It often comes down to your ace reliever pitching in blowout games, as he’s been sitting around waiting for a “save” situation that never came, and so, he’s going to pitch in games with no game impact just to keep him fresh. 

3. Rob Wood did a fantastic piece, that should be re-read annually.

Tuesday, February 02, 2010

More hockey ties

Gabe:

NHL teams are obviously aware of these incentives, and it should come as no surprise that a record high percentage of games have been tied at the end of regulation time this season.  I know of no other sport that works like post-lockout hockey - as long as teams don’t decide to game system any more than they already do, it can probably continue.  But the incentive is to play for the tie whenever you can, and the system can easily fall completely out of its unsteady equilibrium.

I presume the chart is by home team.  If he showed it by home or away, the gap would be even wider.

Tuesday, September 29, 2009

Leverage Index in 1925

Brandon goes through the daily boxscores, makes his best estimates for base/out situations, and comes up with Firpo Marberry’s Leverage Index (LI), when entering a game.

The perfect use of LI to illuminate the past.

Thursday, August 20, 2009

Educating a fan on… reliever impact

A fan said this:

No matter how you look at the math, the closer never comes close in value.

I would not use the numbers I’m about to present as-is (they require adjustments and change in baselines), but as a first step, Eric Gagne in 2003 is in the discussion:

WPA
+5.6 Gagne
+5.4 Pedro
+5.2 Halladay (that guy’s always in the discussion)
+4.9 Loaiza
+4.5 Schmidt
+4.2 Donnelly
+4.1 Wagner
+4.1 Prior

As I said, you have to change the baseline to something other than average, and you have to change the leverage aspect to realize that the reliever should only get partial credit for it.  How do you do the adjustments? 

First figure the LI of each pitcher.  Starters are usually very close to 1, so you might as well leave them be.  For Gagne, he had an LI of 1.8.  We give him credit for 1.4 (halfway between the LI he found himself in and the average… we do that because we credit him for having the talent to leverage the situation, but realizing that if he wasn’t there, someone else, worse than him, would be there anyway… it’s a way to approximate the chaining effect).  So, his 5.6 WPA, based on 1.8 LI, is worth 4.3 on 1.4 LI. 

The replacement level for a reliever is .030 wins per 9 IP below average.  So, you give every reliever that much, times his adjusted LI.  For Gagne, we add +0.8 wins times 1.4, or +1.1 wins.  That gives us an adjusted and rebaselined WPA of 5.4 WAR.

For starters, the replacement level is .120 wins per 9 IP below average.  So, you give each starter that much more.  For your typical workhorse starters of 210 IP (pretty much what Prior and Schmidt got), that adds 2.8 wins.  That puts those guys at around 7 WAR.

So, I would not say that an ace reliever never comes close in value.  But, he’s got a huge hill to overcome, that’s for sure.

FWIW, the way Fangraphs calculates WAR has Prior at 7.6, Schmidt at 6.7, and Gagne at 4.5.

Glove-slap: Repoz.

Friday, June 12, 2009

Kyle Farnsworth

See, this is exactly the kind of article I like from Poz and exactly the reason that MGL likely won’t:

He has struck out more than one batter per inning, he has a better than 2-to-1 strikeout to walk ratio, he allows fewer than one hit per inning, the league is hitting .250 against him in his career … and yet his ERA+ is less than 100 and he has blown 32 of the 59 save opportunities*. Now, seriously, how’s that possible?

*Farnsworth is one of only three pitchers in baseball history — along with Ryne Duren and Oliver Perez — to strike out more than one batter per inning and have a below average ERA+. Farnsworth is the only one of the three to have a better than 2-to-1 strikeout to walk ratio.

How’s is possible? I don’t know, really.

Kyle Farnsworth has a career 4.44 ERA.
Kyle Farnsworth has a career 4.44 FIP.

Kyle Farnsworth may have a great K rate, and a good K/BB ratio, but when you allow 114 HR on 722 hits (16%, when the league average is just over 10%), well, that explains everything.

What I commend Joe in his story telling.  He painted a great picture, he looked in alot of places.  But, when the ERA = FIP, then there is no puzzling piece.  FIP is made up of K, BB, and HR.  And Joe only looked at two of those pieces.  It’s a little thing that I’ll overlook, and we’ll see if MGL thinks it’s too much of a gaffe to accept.

Anyway, it is a great find that Joe has where he shows that Farnsworth is almost always used in low-leverage situations.  The Fangraphs page quantifies this with an LI of 0.55.  That is worse than mop-up duty.  And yet, somehow, he managed to have a clutch score of -2.7 wins.  In 21 IP!!! 

For those who are not used to the clutch score, the league trailer in worse clutch wins is usually in the -2 wins range.  But after a full season.  The current leaders, after Farnsworth, is 4 pitchers tied at -1.3 clutch wins (Brandon Lyon, Fausto Carmona, Masa Kobayashi Maru… sorry Indians fans, and Luis Ayala).  If you think of those pitchers depresses you with his performance in the clutch this year, imagine having two of them, but stuck into the same body.  That, is Kyle Farnsworth.

I would not be surprised if Farnsworth will end up with the worst Clutch season for any pitcher in the Retrosheet era, for someone who pitches in as few high leverage situations as he does.

Friday, May 08, 2009

LI in Star-Ledger

The Star-Ledger is the main newspaper in Northern NJ (it’s also the one that was delivered at the Soprano’s doorstep every morning).  Leverage Index is quoted there, and used effectively:

On April 13 against Tampa Bay, with the Yankees already down nine runs, Coke gave up three unearned runs in an inning of mop-up work. Under the average leverage index scale, a 1.0 is an average pressure situation. That day, Coke’s appearance earned a 0.0.

In his next appearance, on April 16, Coke retired Grady Sizemore with the bases loaded to preserve what to that point had been a tie game. His average leverage index that day was a whopping 4.11.

Coke said he tries not to get too caught up in which role he is used.

“I feel my role is in the seventh inning, and I’ve got to come in in the second because we have no one else, then I’ve done nothing for my team except hurt my team right out of the chute because of my mental preparation,” Coke said. “As a bullpen guy, you’ve got to be willing to do what you’ve got to do.”

Friday, April 24, 2009

Year-by-year LI in the Retro years

Courtesy of Rally. 

He used my LI numbers, so you should see strong agreement with his numbers, Fangraphs, and b-r.com.

The 20 highest LI (min 200 PA):

Five lowest:

Tuesday, April 21, 2009

Nervous closers

In one of Studes’ batted ball reports (sub required), he introduced “Nervous Closers”.  The concept is remarkably simple:
1. Figure out the Leverage Index (LI) when he comes into the game
2. Figure out if his LI ever goes above his initial LI

That’s it.  So, a reliever can get 2 outs, allow a runner on 1B, and then get the 3rd out and not make you nervous.  But, that same reliever can start with the runner on 1B and then get 3 outs.  He made you nervous.

Studes makes a few adjustments along the way, but that’s the basic idea.  He concludes:

No matter how we cut the data, it’s clear that K-Rod likes to make things, um, interesting.

Monday, March 30, 2009

Championship Leverage Index

Sky gives it to us:

Like Tango’s individual game Leverage Index, the Championship Leverage Index doesn’t exactly tell you anything new, but just quantifies a game’s importance into a useful number. It can be useful in analyzing players’ performance in “big games” as well as looking at things like attendance or TV ratings. It’s also fun just to realize in quantitative terms exactly how much each game matters.