Filter posts by...
Leverage_Index
Tuesday, January 31, 2012
This is a followup to this post.
***
Guy, thank you. I prefer being lazy when I can, so thanks.
Ok, this is how it works, and I’m going to have to recalibrate things a bit so it works out to zero. It’s possible for example that David doesn’t give out WPA to pitchers on baserunning events (SB, CS, etc). Not important at the group level.
My first adjustment is to divide all the LI by 1.04 for 1980 and 1.02 for 2011.
The WPA has to be baselined as well. I have to remove 3.87 wins in 1980 and 19.71 wins. It’ll be weighted by IP x newLI.
Anyway, this is Guy’s data, recalibrated:
Year IP.... RA9 WPA LI Role
1980 11210 3.99 23.52 1.05 Relief
2011 14228 4.02 49.11 1.07 Relief
1980 26651 4.42 -23.52 0.98 Starter
2011 29299 4.44 -49.11 0.97 Starter
1980 37861 4.29 0.00 1.00 Total
2011 43527 4.30 0.00 1.00 Total
Next thing is to figure out how many runs above average each group was. That’s easy to do as the league average minus the particular group, divided by 9, times IP. We now have this:
Year IP.... RA9 WPA LI RAA Role
1980 11210 3.99 23.52 1.05 377 Relief
2011 14228 4.02 49.11 1.07 447 Relief
1980 26651 4.42 -23.52 0.98 -377 Starter
2011 29299 4.44 -49.11 0.97 -447 Starter
1980 37861 4.29 0.00 1.00 0 Total
2011 43527 4.30 0.00 1.00 0 Total
We then need to convert the runs into wins. Since both have the same run environment, we’re going to use the same multiplier. I have a quick estimator that is simply RPG+5. Which in the above case would mean 9.3 runs per win. If I use PythagenPat, I get 9.4.
Anyway, dividing the RAA by 9.3, and we get:
Year IP.... RA9 WPA LI RAA WAA Role
1980 11210 3.99 23.52 1.05 377 40.54 Relief
2011 14228 4.02 49.11 1.07 447 48.06 Relief
1980 26651 4.42 -23.52 0.98 -377 -40.54 Starter
2011 29299 4.44 -49.11 0.97 -447 -48.06 Starter
1980 37861 4.29 0.00 1.00 0 0.00 Total
2011 43527 4.30 0.00 1.00 0 0.00 Total
Finally, we apply the LI, to get leveraged wins. So:
Year IP.... RA9 WPA LI RAA WAA levW
1980 11210 3.99 23.52 1.05 377 40.54 42.57
2011 14228 4.02 49.11 1.07 447 48.06 51.42
1980 26651 4.42 -23.52 0.98 -377 -40.54 -39.73
2011 29299 4.44 -49.11 0.97 -447 -48.06 -46.62
1980 37861 4.29 0.00 1.00 0 0.00 3
2011 43527 4.30 0.00 1.00 0 0.00 5
LevWins is what we’d expect of their WPA, if those pitchers pitched at those exact levels in every situation.
Instead, what do we find? Well, let’s subtract WPA by LevWins:
Year IP.... RA9 WPA LI RAA WAA levW Diff
1980 11210 3.99 23.52 1.05 377 40.54 42.57 -19
2011 14228 4.02 49.11 1.07 447 48.06 51.42 -2
1980 26651 4.42 -23.52 0.98 -377 -40.54 -39.73 16
2011 29299 4.44 -49.11 0.97 -447 -48.06 -46.62 -2
1980 37861 4.29 0.00 1.00 0 0.00 3 -3
2011 43527 4.30 0.00 1.00 0 0.00 5 -4
I’m not going to make the final adjustment to zero it out, because the point is about to be made.
We see that in 2011, the relievers and starters have a WPA exactly matching what we expected. This would point to having no “matching” of talent to situation. Or, if there was matching (like Rivera) that was undone by bad matching.
But, look at 1980. Relievers were terribly used, getting very little win benefit. Basically, not only was there no matching, but there was severe mismatching. This points to really good relievers being used in really low LI situations.
So, back to 2011. For all the obvious Mo and Papelbon situations, we also have plenty of situations that the managers simply undid those leveraged situations.
Therefore, while they could stand to improve their 2011, they were simply abysmal in 1980.
Sunday, January 29, 2012
This article came up in a search, and given all the new readers around here, I wanted to highlight it:
http://www.hardballtimes.com/main/article/crucial-situations
For those interested in part 2, and part 3, as well as the LI chart:
http://insidethebook.com/articles.shtml
Sunday, January 01, 2012
Great find:
Shawn Chacon got 35 saves as the “closer” for the Rockies in 2004. He had an ERA of 7.11. He had zero saves before that season. He got one more save for the rest of his career.
Chacon was a full-time starter for all but two seasons, one being the aforementioned season. He had a -2.3 WPA in 2004, which is really all you need to know about his effectiveness. He was as bad at home as he was away. He was as bad in save situations as non-save situations. He had as many K as BB, which is a recipe for a disaster of a season, and add in a worse-than-average HR rate, and you get the mother of all closer seasons.
And yet, 35 “saves”. In those 35 saves, he had a WPA of +3.1, or +.09 wins per save. In his other 31 games, he was -6.1 wins (or -.20 wins per game).
If we take a bad Mariano Rivera season to compare: in 2007 he had 30 saves, with +3.7 WPA, or +.12 wins per save. In his other 37 games, he was -1.3 WPA (or -.04 wins per game).
There is one good piece of information to go along with those 35 saves: he had a 1-9 record. As you can see, when Chacon would blow a save, he would blow it really big.
There have been six seasons where a reliever has: saved at least 30 games, won at most 2, and lost at least 8, including Chacon. Four of those pitchers had a respectable ERA, but one, Brad Lidge, had perhaps the worst relief season of all-time. 31 saves, 0 wins, 8 losses, and a 7.21 ERA. His WPA that year was -4.6 wins.
Thursday, December 22, 2011
I don’t remember this, but this was from two years ago, and talked about again today.
Basically, when you calculate the value of a reliever, you multiply whatever value you get by the Leverage Index he merits based on his talent. This would be analogous to figuring out a player’s talent level as a fielder, and then multiplying it by the opportunities he sees (and the opps would be tied to his talent, such that a better fielder will find himself at SS, 2B, 3B, CF, and a worse fielder in the corners or 1B).
If you have a crappy fielder in CF, you can’t punish him too harshly because the team happens to put him there. Think Junior near the end of his career.
Saturday, November 26, 2011
Cardinals and Rangers. That’s how I interpret the CLI stat in the 2012 THT Annual. LI is Leverage Index. CLI is Championship Leverage Index, which basically means how much did each of the 30 team’s chance of making the playoffs change day to day.
So, Studes calculated it, and figured that CLI for the Cards at 2.22, to lead the league. In last place was the Astros. Results are in the sidebars of page 83-84. And by the way, I love the sidebars. Just a great feature and tremendous layout. Much better in this small format, 9x7 book (just a bit bigger than The Book).
I think it’s a tremendous story stat, deserving of tracking on an annual basis. I hope SOMEONE does it, because it’s pretty cool.
Wednesday, November 23, 2011
Twitter is blocked at the office, so I only can get on there for like two minutes a day. I just posted a couple of points regarding leverage, but, frankly, twitter is the worst place to talk about this. If someone is following me, and there are responses to my posts on leverage, please post them below, and I’ll answer them here.
Monday, October 31, 2011
As you guys know, I’m big on WPA/LI (aka Situational Wins), even if I can’t articulate it well enough. I’ll get there one of those years.
Similar to WPA/LI (which is the change in win expectancy by game state, divided by the leverage for that game state) is RE24/boLI (which is the change in run expectancy by the 24 base-out states, divided by the leverage for that base-out state).
One set of boLI numbers can be found here. Basically, the highest leverage is with bases loaded, and lowest leverage is with bases empty. You can get a double in either case, but the impact of your double will be felt more with the higher-leveraged state.
So, in order to put both PA on a level playing field, you “deleverage” it by divided the change in run expectancy by the leverage index for that base-out state. Or, so I thought.
I wrote this to Sean Forman (who calculated RE24/boLI on his site) with regards to a Tyler Clippard play:
Sean,
Ok, this is what I’m thinking. Let’s look at this event for Clippard:
game_id event_num inning date_game
ATL201105100 67 8 5/10/2011
There’s runners on first and second, the LI for that base-out state is 1.94. A HR is hit, 3 runs score. The RE delta is -2.38. We would divide that by 1.94 to “neutralize” the leverage.
However, part of that RE delta is the batter himself. He himself is not part of the “extra” leverage. His HR would contribute exactly -1 run to the delta RE regardless of what the boLI is.
So, what really should happen is that you divorce the delta RE between the runners and the batter. The neutralizing of the delta RE should only apply to the runners, not the batter.
I still haven’t thought through it all, but this is where I’m coming from.
I’ll talk to you later.
Tom
So, in the case of my Clippard HR illustration, the -2.38 RE delta was really -1.38 for the runners and -1.00 for the batter. We deleverage the runners by a factor of 1.94 (to get it to -0.71), and we leave the batter untouched (at -1.00), for a total deleveraged RE24 of -1.71.
Compare that to -2.38/1.94= -1.23.
I’m 90% convinced I’m right, but I’d like to hear from the Straight Arrows among you.
(And, if you had questions regarding deleveraging, be it here or with WPA/LI, now’s your chance as well.)
Thursday, October 13, 2011
Rivera did better in low-leverage situations. (Look at tOPS+ if you need one number, not that I’m limiting this to one number.)
Hoffman did the same in both.
Wagner did much better in low-leverage.
Percival did a bit better in low-leverage.
These are the first four I picked out.
This writer picked out some names… randomly, or just the exceptions?
Valverde did better in high-leverage situations. Papelbon did better in high-leverage as well.
The right thing to do is pick out a large enough list of relievers. Fangraphs makes it easy enough. That’s all relievers with at least 300 IP, from 1993-2011. Sort by WPA/LI to get the best relievers on top (that’s your unbiased list ordered).
The “clutch” column tells you if he pitched better in high-lev or low-lev situations. A negative number means he pitched better in low-lev. So, a mix of some closers who pitch better in low-lev and some who pitch better in high-lev.
For every Mariano Rivera who got better numbers in low-leverage situations than he did in high-leverage situations, you have a Joe Nathan who was better in high-lev than low-lev.
Basically, you can prove anything with numbers if you are allowed to cherry-pick your players. BR.com and Fangraphs.com make the presentation of numbers so ubiquitous, it makes some bloggers dangerous.
Look hard enough, and you’ll find splits that can support any theory you want, especially if you can pick and choose the data points you want.
Wednesday, September 21, 2011
Apparently I missed some twitter back and forth between Colin and David and others, trying to figure out how fWAR has him at 39 wins, while rWAR has him at 56 wins.
Here’s how you can come up with the answer fairly quickly. fWAR is FIP at its core. And Fangraphs gives his career FIP- as 62. Since Rivera is at 2.76, that sets the league average at 2.76/.62=4.45 for ERA. Divide by .923 to get into RA9, and we have a league average of 4.82. Replacement level is about 7% higher (for relievers) or 5.16. Because of his short time at starting, the replacement level is really 5.20. Rivera’s 2.76 FIP on ERA scale is 2.99 on RA9 scale. Rivera is therefore +2.21 runs better per 9 IP (5.20 - 2.99). With 1209 IP, that puts us at +297 runs. That’s around 29.7 wins, depending on the runs to win converter.
His LI is 1.87, so we give him 1.435 for WAR purposes. 29.7 x 1.435 = 43 wins. fWAR has him at 39 wins.
Rally however uses runs allowed in rWAR as his core. Both BR and FAngraphs has his ERA at 49% of league average. Since his ER/RA is around the standard league average, we can use his ERA- as a proxy for his RA-. Anyway, so his RA9 is 2.40, and if we divde by .49 we get a league average of 4.90. A league average of 4.90 means 5.24 reliever replacement. Bump that up to 5.28 because of his starts, and we have Rivera at a whopping +2.88 runs per 9IP. With 1209 IP, that’s +387 runs, or 38.7 wins. Multiplying by the 1.435 LI multiplier gets us to 56 WAR. rWAR is showing 56 WAR.
The really big difference originates with FIP v RA9. Rivera is a BABIP machine. It’s a huge difference.
***
While you can make the argument that BABIP has little to do with the pitcher at the seasonal level, you can’t make the same argument at the career level.
Wednesday, August 24, 2011
This article on Shutdowns and Meltdowns noted that Venters already has 43 shutdowns and only 3 meltdowns. Which got me to ask: which reliever has had the highest WPA, without the benefit of getting a bunch of saves. Turning to the ultra-quick Play Index, I asked for all pitchers with at most 10 saves (and some other filter to remove starters or swingmen), sorted by WPA.
#1 was Mariano Rivera’s lone season as setup guy. That one was a WPA of +5.4. #2 is Rafael Betancourt in 2007 at +5.4 as well. And #3 was Mark Eichorn and his 157 innings in relief (and 10 saves, so just managing to satisfy my arbitrary filter).
After that? We have Venters and Clippard, both this year, both at +4.8 wins. Clippard is especially noteworthy, in that he has no saves at all so far. With just over a month to go, it’s possible that one of these two relievers is going to supplant Mariano Rivera for best relief season by a non-closer (as measured by WPA).
Monday, June 06, 2011
J-doug points out:
In yesterday’s game against the Arizona Diamondbacks, Washington Nationals closer Drew Storen had what could only be described as a meltdown appearance. If not for the triage work of Todd Coffey, Sean Burnett and Henry Rodriguez, as well as a five-run 11th courtesy of a Rick Ankiel sacks-juiced walk and a Mike Morse grand slam, Storen would have squandered the curly-W.
All in all, Drew Storen’s butcher job amassed a dismal -0.449 Win Probability Added (-0.457 WPA if you ask FanGraphs instead of Baseball Reference).
He also earned the hold.
Wait, what’s that you say? Storen earned the hold even though it was his runners who tied the game? He earned the hold even though he was on the hook for the loss should Coffey allow another inherited runner to score?
Yes, that’s correct. In fact, Storen’s HD is the 10th worst in WPA terms since the stat was invented, and the 4th worst in a winning game. It’s easily the most detrimental hold this season.
And had the Nationals lost in the 9th, Storen would have kept his HD and earned the loss, while Coffey would get the blown save and the Nationals would drop another game below .500. How is this possible?
He also gives us this juicy tidbit:
As further evidence of the absurdity of the hold stat, a whopping 3,278 holds classify as meltdowns since the hold was invented, compared to four saves.
The meltdown that he mentions is the thing I cooked up last year or so. That there is such a non-link between saves and meltdowns, but such a huge link goes to show that a “hold” is often just in name only.
I agree with Jesse that Dewan was obviously well-intentioned. But, does this mean that the mainstream has to continue to accept the Hold stat?
***
A great use of WPA by the way. WPA is great for turning a word description of the game into a nice cold number. It makes analysis much easier.
Thursday, May 26, 2011
Leverage Index describes the “level of fire” at a particular point in time. It understands where you are in a game, and has no idea what is about to happen, other than to expect “average” things will happen. So, we have this great insight from a 19-inning game:
11 plays had a leverage index of 4.00 or above (meaning it was a really important play);
only one of them involved a hit.
Pitcher Hitter Inn. Outs Base Score Play LI
Herndon Hernandez 11 2 123 4-4 Ramon Hernandez grounded out to pitcher (Grounder). 6.86
Masset Polanco 9 2 123 3-3 Placido Polanco reached on fielder’s choice to shortstop (Grounder). Jimmy Rollins out at second. 6.38
Masset Brown 9 1 123 3-3 Domonic Brown fouled out to catcher (Fly). 5.71
Fisher Ibanez 19 1 123 5-4 Raul Ibanez hit a sacrifice fly to center (Fly). Jimmy Rollins scored. 5.71
Halladay Bruce 7 2 123 3-3 Jay Bruce singled to right (Grounder). Miguel Cairo scored. Drew Stubbs scored. Joey Votto advanced to 2B. 4.66
Halladay Rolen 7 1 123 3-1 Scott Rolen struck out swinging. 4.66
Romero Phillips 11 1 12_ 4-4 Brandon Phillips picked off. 4.62
Romero Bruce 11 2 12_ 4-4 Jay Bruce walked. Joey Votto advanced to 3B. Scott Rolen advanced to 2B. 4.43
Masset Mayberry 9 1 12_ 3-3 Chase Utley advanced on a wild pitch to 2B. 4.31
Masset Rollins 9 1 _23 3-3 Jimmy Rollins was intentionally walked. 4.15
Fisher Howard 19 1 _23 4-4 Ryan Howard was intentionally walked. 4.15
This sounds like sudden-death overtime in hockey, where there’s a high stress level on every pass and shot, and the goalie turns back every shot, makes every save. Until the last one goes through.
***
A point of clarification with this comment:
Note that Jay Bruce’s solo home run in the tenth earned just a 2.30 LI and Ryan Howard’s solo home run to tie the game back up in the bottom half earned a 3.42 LI.
Note that the LI is assigned PRIOR to the event occurring. It is a description of the state that the batter and pitcher find themselves. So, Jay Bruce found himself in a state where the LI was 2.30. What you “earn” are wins (not LI), meaning that in this situation, rather than a HR being worth around +.14 wins, they were paying off at about 2.3 times that. (He earned +.342 wins.) The change in win expectancy is how much more relief the Reds fans got (and the level of despondency that Phillies fans experienced). Ryan Howard’s HR spun things around the other way.
Thursday, April 21, 2011
Bard:
“That was the game right there. You guys have heard me talk about it time and time again that the game can be won in the sixth or seventh. For me, that was it. He came in and stopped it,” Francona said.
I asked Bard what his title should be. If Jonathan Papelbon is the closer, what is he?
“Maybe you can call me the stopper,” he said.
Bard said he has spent some time on Fangraphs.com reading the work those guys have done on a stat called Shutdowns and Meltdowns. It’s a simple way to evaluate relievers.
“It’s just another way to look at things,” Bard said.
Bard got +0.21 wins of WPA in that game, easily clearing the bar for a shutdown. Papelbon got +0.13 wins.
Glove-slap: Jeff.
Saturday, April 09, 2011
In 2010, Feliciano had a pretty good season, according to the seasonal numbers: above average ERA, FIP, xFIP, WPA. But he had 17 shutdowns and 13 meltdowns. The average is 2:1, and anything close to 1:1 is replacement level.
We can look at his gamelog, export to Excel, and sort by WPA. In his 17 shutdowns, he had a total of +1.8 WPA. In his 13 meltdowns, it was a total of -2.2 WPA. In all his 62 other games (basically classified as “neutral” as far as shutdowns and meltdowns), he was at +0.8 wins. Ideally, he would have been at zero. But, he had alot of “good” games, but not “shutdown” games, but the shutdown stat gives him no credit for it. This is what happens when you set arbitrary threshholds, and give out binary outcomes.
But, if you set it too low, then all those great +.20 win games get the same value ("1") as those barely above average +.02 win games.
I agree with Raymond that it’s dangerous to rely on SD and MD to give you too strong a picture of a performance. Like anything, you have to be careful how you use it. Use SD and MD to start your analysis, and then see why it doesn’t seem to add up.
Wednesday, April 06, 2011
Great saber-filled but mainstream-ish article on the case for shutdowns and meltdowns.
My preference is SD minus MD, or SD minus MD per game. Nonetheless, that’s a quibble.
Monday, March 28, 2011
Good stuff from Steve, giving us the lay of the land.
With regards to the age thing, I’d like to see a corresponding line that shows the WPA/LI of the pitchers, because I would presume the older pitchers (those allowed to pitch in MLB) must be the really good pitchers. Nonetheless, I think the point would still remain that a young pitcher is not going to put thrown into the fire too quickly.
Thursday, March 17, 2011
Cool chart, that shows how a pitcher is used in terms of save opportunities on the x-axis and the leverage index on the y-axis. The outlier points (top left corner) are the setup guys who come into the game in high-leverage situations, but are not given the chance to close out the game. Chances are, these are your closers of tomorrow. Guys in the bottom left corner, well, that’s ALOT of relievers, and that’s where most of them will stay in their careers.
Wednesday, January 19, 2011
Colin concludes his primer on WE and Leverage with:
Which brings me to the power of myth-making. Metrics that incorporate a relief pitcher’s leverage but ignore his ability to create leverage for others paint a skewed picture of how relief pitchers create value. Teams that hold their most valuable bullpen arms in reserve waiting for save chances may be winning more close games (although not as many as they may think), but the cost may well be staying closer in fewer games to begin with.
The reason that Hoffman and Rivera et al have a career LI close to 2.0 in relief is that they let the rest of the team do all the hard work, and they come in to finish the job. Basically, they were given a lottery ticket with 5 of the 6 winning numbers pre-selected, and he just has to work his magic to get the 6th number. But, the reason they were given 5 of those numbers is because they are great pitchers. If they were only average pitchers, they’d only be given 4 of the 6 winning numbers.
So, we have two things going on:
1. A situation where you can make alot of money has been created for them, like buying stocks on margin, or borrowing money. (For example, you have 1000$, and someone loans you another 1000$ at 50% interest, so you have 2000$. You bet that money on a horse, and the horse wins. You win 4 times your bet, so that’s 8000$ in winnings, and 10,000$ in your pocket. You give your friend back the 1000$ plus 500$ in interest, and now you have 8500$ in your pocket, where you once started off with 1000$. You actually made 8.5 times your money.)
2. But the only reason you are in that position is because you are good and trustworthy. (For example, the only reason your friend even lent you 1000$ is because he trusted you to give him his money back at 50% interest.)
If you were untrustworthy, your friend might have demanded 200% in interest. So, after giving him back his 1000$, you also give him 2000$ in interest, for 3000$. That leaves you with 7000$, or 6000$ profit.
In a complex method called “chaining” (which may have originated with Patriot, or with me I can’t remember, but did originate back at the old Baseball Boards RIP), we try different kinds of gamblers because SOMEONE has to bet on that horse. It’s going to happen.
When you work it out with respect to relievers, you will find that the leverage component “earned” by the player’s perceived talent level to find himself in that situation is about halfway between the actual leverage and average leverage. This is why when we see Rivera or Hoffman with an LI of 2.0, we “credit” them with an LI of 1.5. Because that’s the equivalent to the chaining process.
Friday, January 14, 2011
I was aghast:
Let’s first look at Leverage numbers, a metric tracked at Baseball Prospectus. 1.00 is the Leverage situation at the start of the game when the first pitch is thrown, and then from there it’s driven by Win Expectancy.
I’ve already dispelled why BPro’s LEV should not be used years ago, and makes the above statement false. Colin’s done great work in cleaning up most of the issues I’ve had with the math at BPro. Presumably, he’ll get to this one at some point.
In any case, since the writer is having an issue with a Fangraphs writer, why not quote the leverage numbers from Fangraphs? (Note that both Fangraphs and B-R.com use Leverage Index data provided by me.)
***
Anyway, back to Farnsworth: he had a .468 BABIP with men on base in 2009. Not only that, but he gave up far more walks with men on base than bases empty. Before you do a comparison, you have to decide how relevant that piece of data is in forecasting him in 2011.
Thursday, January 13, 2011
Marty Noble:
The save rule needs to be amended in a way so we can measure the degree of difficulty or the challenge involved. Saves ought to be weighted, i.e., a save achieved in a three-run game would be worth one point, a save achieved in a one-run game would be worth three points, a two-run save would warrant two points.
More could be done involving the number of outs required, inherited runners vis-a-vis the score. But simply weighting the saves, by itself, probably would produce a greater disparity in the save-points totals than the disparity that exists now when 12 closers have between 35 and 45 saves. So we’d likely see a conspicuous difference between the 33 saves Closer A achieved and the 35 Closer B earned.
Leverage Index in the top of the ninth to start the inning:
up by 3: 0.8
up by 2: 1.6
up by 1: 2.9
And in the bottom of the 9th:
up by 3: 1.0
up by 2: 2.0
up by 1: 3.6
Basically, this is the pattern that Noble is proposing, and it is captured by Leverage Index (LI) already. He further suggests outs and runners on base without elaborating. But, since we’ve already been able to match his needs on bases empty 0 outs with LI, presumably he can easily buy in on what the rest of the chart shows for runners on base and outs.
LI is basically a quantification of your qualitative feelings on how much impact a certain context has. I think this is how you can hook in someone like Noble, to show him that we already have a process that matches exactly what he’s thinking on things he’s come to a conclusion on, and so, he should be able to buy in to the rest of the stuff he hasn’t made a decision on yet.
If he insists on integers to assign points (and presumably capping it), that’s fine, we can come up with some basic rules based on half-innings, score, runners, and outs.
from Amazon
Recent comments
Older comments
Page 1 of 320 pages 1 2 3 > Last »Complete Archive – By Category
Complete Archive – By Date