SB Nation User Blog: Edgar for Pres

WE by count

Edgar for Pres — Sun, 22 Jun 2008 05:43:41 -0000

Just was surfing the web and stumbled along this. I think its pretty cool and I'm happy to see that somebody has already done it. Josh Kalk (writer for THT) has posted the WE by count on his blog.Here is the link to his blog and the WE table.

For example:

Tie game in the bottom of the 9th and bases are loaded with 2 outs

Count / WE

0-0 / 65.8

0-2 / 60.1

3-0 / 76.1

Thats a pretty big difference just based on the count. I was hoping the difference would be bigger than it is for more situations. In most cases the count would only change the WE by fractions of a point within an AB.

Who's your favorite player?

Edgar for Pres — Tue, 01 Apr 2008 06:05:46 -0000

I know we've all got our own mancrushes on the team. Who are they?

And no I don't think this is gay cause if loving a guy like Brett Favre or Michael Jordan makes me gay then I don't want to be anything else. I just threw everybody in the poll but if you pick somebody stupid then you miss the chance to support your favorite player.

How good was Griffey's Defense

Edgar for Pres — Wed, 26 Mar 2008 22:24:34 -0000

Over at THT Sean Smith has been going through old data and putting together historic defensive stats. You can go download the spreadsheet that has all the info.

I have always felt like Griffey was a great defender but I was young and he dove in CF so maybe I had overvalued his play over the years. Looking at him now and the stats we have currently he looks horrible. I went through and grabbed what Sean had on Griffey.

player	pos	Chances	Runs	year
grifk002	cf	454	-12	1989
grifk002	cf	491	-4	1990
grifk002	cf	526	-2	1991
grifk002	cf	522	-7	1992
grifk002	cf	477	-12	1993
grifk002	cf	352	-9	1994
grifk002	cf	280	2	1995
grifk002	cf	560	5	1996
grifk002	cf	581	6	1997
grifk002	cf	615	6	1998
grifk002	cf	547	8	2000
grifk002	cf	307	-10	2001
grifk002	cf	153	-11	2002
grifk002	cf	143	-5	2003
grifk002	cf	267	-9	2004
grifk002	cf	452	-23	2005
grifk002	cf	375	-17	2006
grifk002	rf	422	0	2007

When you look over the numbers you'll see that he wasn't that good when he was young. Probably still pretty rough because of lack of experience. He was a fine CF for around 5 yrs from 1995 to 2000 and then injuries caught up with him and killed his abilities to play defense.

I think this jives pretty well with how I believe he actually was. Am I way off? I'm a little surprised he wasn't a better defender when he was younger and that he was still decent around 2000.

PS. I do like how I can just copy excel stuff in and it looks good now.

What FA are left and who should we sign

Edgar for Pres — Sun, 10 Feb 2008 00:23:41 -0000

In an attempt to make a diary that relates to baseball, I figured I'd try to go through the FA that are left. I think we all agree that we probably need at least a bench bat and/or maybe a 4th OF. We could use a little insurance because if the four headed monster of Ibanez/Wilkerson/Sexson/Vidro has the collapse they are doomed to have them we'll be screwed. I also wouldn't be surprised if we signed a RP. With the retarded Cairo signing I'm guessing we won't be looking at any infielders.

So here we go in no particular order:

1B/DH
Tony Clark - Padres (1yr/900k)
Doug Mientkiewicz
Mike Piazza
Mike Sweeney
Russell Branyan
Jeff Cirillo
Corey Koskie?
Barry Bonds
Shea Hillenbrand

IF
Tony Graffanino

OF
Kenny Lofton
Kevin Mench
Trot Nixon
Corey Patterson
Shannon Stewart
Reggie Sanders
Sammy Sosa
Preston Wilson

RP
Jeff Weaver
Armando Benitez
Shawn Chacon
Rudy Seanez
Bob Wickman

I wasn't gonna do a bunch of analysis of all the options but there are still a fair number of guys out there that could help the club.

I'd love to see the team sign Mike Sweeney but that's probably me being irrational. As sad as it is, Mientkiewicz might be as good as Sexson next year (and have the hardest name to spell ever). Bonds would be great but we'll never sign him. Branyan and Piazza would be interesting if nothing else. I'd also really like so get Koskie but from what it sounds like he might never play again (or at least not until part way through the season) because of the concussion issues.

In the OF, I like Patterson the most probably (if Bonds doesn't count). After that its basically everyone else. Half of them will probably do fine but the others will probably suck. The bar has to be set by asking if they are better than Wlad/Reed. It would be nice to have somebody who could be a defensive sub (Patterson).

I wouldn't be extremely pissed if we signed a RP. With Sherrill gone it isn't the worst move as long as we don't overpay and don't sign somebody who isn't actually any good.

Anyway, thoughts? Since this Bedard trade went down, a push for this year needs to be made by the front office. If I took over Bavasi's job right now I'd probably try to grab somebody useful with a trade.

If you need something else to talk about here is a random name: Hee Seop Choi.

Using OPS+: Steroids Era is Over!

Edgar for Pres — Fri, 11 Jan 2008 07:51:50 -0000

So a while back I was wondering how the variance of OPS+ has changed with time. The average of OPS+ is set at 100 by definition but there is nothing in the formula that makes sure the stat falls under the same distribution all the time. An OPS+ of 120 is always 20% better than average but it doesn't tell you what percentile you were.

My first thoughts about this came from trying to compare OPS+ from different generations. A fairly wide feeling is that as the competion in the league increases we should see the variance in the league decrease. Ty Cobb might have been much like Ichiro but he played in a drastically different league for example.

Not surprisingly, I don't have amazing database skills and the info wasn't availible instantly. After dreading it for a while, I went through baseball reference's database using their PI tool and took down all the players with more than 100 PA for a bunch of different years. Then I found the PA weighted OPS+ which came out to be about 100 for every case (good sign!). After that I found the PA weighted standard deviation in OPS+. I also found the skewness but I don't think that tells me a bunch.

So anyway, what did I see?

Well, looking at the graphs you see that it looks surprisingly noise free. I was pretty shocked. I didn't expect this. I was basically expected to not see anything interested. Oh yeah, don't complain about me not having every year. It takes a while to strip the info for me and I got lazy. So anyway, we see that the variance looks to decrease slowly which is what we expect. I kinda would have thought the increase might have been larger but I've been wrong plenty of times before.

The most interesting features of the graph is the blip around 1970 and 1999. I'm not 100% sure why there is the bump around 1970. I think a small error I made when I was collecting data was that I did not exclude pitchers and they tend to have pretty horrible OPS+. I don't think it caused this spike but it might explain a little of it. This wasn't smart to do but I don't think it destroys everything (hopefully). From 1990 to 2007 there were only 6 times that 100 PA was broken by a pitcher so the recent times should not be affected by this at least.

And to the blip I think is most interesting. The one that goes from around 1990 to 2005 fits amazingly well to the steroids era. The most interesting part I think is that the variance is back down to the level it was before 1990. The first question I asked myself is what happens when you take a few of the top hitters out for those years? Maybe its just that a couple players like Barry Bonds are just skewing it. Well I tried that and it doesn't get rid of the whole peak. Does this mean that that the effects of steroids and performance enhancers are gone or does it mean that everybody is using HGH now? Does it mean anything at all? I'm not really sure. I'm guessing that the testing baseball has now has really done something.

WE Rating System: How good was that game?

Edgar for Pres — Tue, 13 Nov 2007 07:08:04 -0000

For a little while I wanted to try to use WE to quantify how "exciting" a game was. It should be pretty simple. In general, the more spiky the graph, the better the game and blowouts tend to be big yawnfests.

Using some data Jeff gave me (thanks), I was able to do a little number crunching and come up with some simple formulas that give you some idea of how good a game was.

The first we'll call Blowout Rating (BR). This tells you how big of a blowout a game was (big shocker!). It ranges from 100 to -100. 100 represents a "perfect" blowout where we win. -100 represents the opposite. 0 tells us the WE was 50% the whole game.

The formula: BR = 100*[2*average(WE)-1]

This basically uses the area under the WE curve to figure out whether the game was mostly hopeless or a guaranteed win.

The next two formulas are a little more difficult (not too bad) and use a little calculus (oh no!). They are the average absolute 1st and 2nd derivatives of the WE curve.

The average absolute 1st derivative (AA1D) mathematically is the slope of the WE curve. Absolute means if the slope is negative then we'll make it positive. This gives us the average change in WE with each plate appearance. In other words, if it is 10%, then whenever there was a batter up the WE changed on average 10% (in either direction). I'd say this would signify how exciting the game is because it shows us how much the WE changes.

The average absolute 2nd derivative (AA2D) is a little bit tougher to nail down. Its the absolute derivative of the 1st derivative (which was not absolute). When the slope of the WE changes directions this results in a large value for the 2nd derivative. This tries to tell us how much the slope of the WE varies and gives us a percent change in the slope of the WE. In other words, this shows us how much the momentum varies in a game.

I tried to make a model game where we'd expect to see the WE jump around as much as is possible. It probably isn't the theoretical maximum but its probably close enough. I found that the AA1D(max) = 22% and AA2D(max) = 19%. Once again, these are a little too low so maybe we'll just say that they are both 25%.

These are the games Jeff supplied me with data for.
Sep 08
Aug 15
Jul 7
May 09
Apr 21

I think the games I was given gave me a pretty good mix to work with. So what did I get?

Blowout Rating (BR)
4_21    -72
5_9    43
7_7    69
8_15    -19
9_8    -41

So what does this tell us? Well the April game pretty much sucked. It sucked so much Jeff didn't write anything up for it. On the other hand, the June game was also a pretty big blowout in our favor as the game was in hand the whole time. The best of the bunch according to this was the August game which was back and forth a little until the last few innings where it slipped away. The biggest shortcomings of this is that it doesn't take into account the timing of the game. It would be kinda nice if it was weighted so that the last 1/3 of the game was more important but for now I think it is pretty decent and shows what you want in one number. Once again, this doesn't really show us how "exciting" a game was, just how big of a blowout it was.

Absolute Average 1st Derivative (AA1D)
4_21    2.1%
5_9    2.2%
7_7    1.7%
8_15    4.4%
9_8    3.6%

Absolute Average 2nd Derivative (AA2D)
4_21    2.6%
5_9    3.2%
7_7    2.4%
8_15    6.1%
9_8    5.2%

I kinda lumped these two together because they kind of show similar things in the end. They show that the last two games were clearly more exciting than the first three. Looking at the WE graphs this isn't surprising. Also, again as with the BR, there is no extra weighting for the time things occur but I think WE takes care of this since it varies much more near the end of the game which increases both metrics. One nice tidbit you can tease out from this that isn't in the BR metric is that the September game was a more exciting game than the May game even though the BR is roughly the same magnitude for both.

One thing I didn't look into just because I didn't have the info that would be interesting is looking at the average leverage index for a game. This would be a pretty easy way to show how many close situations there were in a game. I'd like to look at this one and see how it'd compare to the other metrics I came up with.

Oh I also thought it'd be good to include this one that Jeff says he likes. As a quick and dirty way he says he looks at a game to see how many plays change the WE by more than 10%.

Plays where WE changed my more than 10%
4_21    2
5_9    2
7_7    1
8_15    5
9_8    6

Appears as if Jeff's method is pretty good and goes along right along with my derivatives. This method appears to be surprisingly good. Once again, Jeff appears to be pretty smart.

This stuff is pretty rough but I kind of wanted to get the idea out there. It should be pretty easy to formula a good rating system and be able to come up with a list of the most exciting games ever.

Bill James Projections

Edgar for Pres — Fri, 09 Nov 2007 20:23:27 -0000

I was debating whether to put this up but since Bill James' projections are availible on fangraphs (how I got them). Go over and take a look.

I guess I'll just quickly summarize them.

Ichiro: repeat of 2006
Beltre: nice progression
Sexson: big rebound
Lopez: I guess its good, kinda...
Yuni: Same old
Jones: I'm happy with it.
Ibanez: Look at LL/USSM projections for 2007
Guillen: Predictable
Kenji: WOW!!!! All star??
Wlad: Adam Jones light

SP: Boring with Felix looking better but not great

So I don't know if much is amazing beside Kenji. (.301/.357/.497 really?)

Another RC formula

Edgar for Pres — Fri, 27 Jul 2007 22:49:37 -0000

I know, everyone and their mother has made a RC formula or two. I was just messing around in some huge spreadsheet quite a while ago and kinda just stumbled across this. The thing that started me messing around is that I don't like complicated equations and OPS is nice because its so simple and uses numbers I already have a good sense for.

As we all know OPS is a decent judge of runs created but it isn't as good as we would like. We also know it undervalues OBP.

OPS = OBP + SLG = (AVG + ISOpa) + (AVG + ISOpo) = 2*AVG + ISOpa + ISOpo

So my idea was basically to take a bunch of data and fit the expression RC/PA = a*AVG + b*ISOpa + c*ISOpo and do a pretty straightforward regression. This actually gave pretty good results. RC/PA = 7.674*AVG + 6.357*ISOpa + 14.945*ISOpo but the R^2 was only 0.8 which isn't very good when compared against team RS results from the past 40 years. After looking at some of the residuals it was apparent that it was too simple to represent the AVG term as just a single order term.

To improve it I changed it to RC/PA = a*AVG + b*AVG^2 + c*ISOpa + d*ISOpo. I know this is getting away from the simplicity and that made me sad. This gave great results though. RC/PA = 1.4088*AVG^2 - 0.1412*AVG + 0.3446*ISOpa + 0.2701*ISOpo and has a R^2 of 0.96 when compared against team RS results. Also, when compared against Baseruns it has an R^2 of 0.999 which basically means that it gives you baseruns without the complex formula that baseruns has.

I like it. I don't know if I'll use it. I just thought it was surprising that it did so well when compared against something as advanced as Baseruns.

Oh, I looked at including things like SB or CS but when using team RS to create a correlation the impact of SB and CS is just not significant and gives pretty strange results sometimes.

WPA and Pitch Counts

Edgar for Pres — Sat, 12 May 2007 17:56:56 -0000

I think everybody around here loves WPA and Jeff’s graphs and maybe even Fangraphs’ live versions. They are fun, interesting, and WPA is actually a decent stat at the end of the season to look at.

Right now WPA only looks at outcomes whenever the runners move or outs are made. This leads to only maybe 40 outcomes per game and makes the graphs look pretty rough. There is also much more information about the game that WPA is ignoring.

In particular, the number of balls and strikes. I think if WPA was redone so that it could be computed for not only the game state (runners and outs) but for the specific count it would result in more information being transmitting to the viewer. Now instead of 40 data points for each team, you may have 120.

The nice thing about this in my mind is that implementing this wouldn’t affect the final WPA for the players at all. If a hitter comes in and gets a hit, the change in WPA is independent of how the WPA changes as balls and strikes are thrown.

So what is the advantage of taking data as pitches are thrown and no immediate outcomes are occurring like hits or outs? Well I think there may be a lot of data that WPA is just missing. The most immediate impact would be smoother WPA graphs since we probably at least triple the amount of data used to calculate them. Also, it’s the hitter’s job to hit the ball but it also helps if he works the count in his favor he should receive some type of credit for it just like if a pitcher gets good counts on hitters. From this we could look at a bunch of different things like how much of a difference working the count for a batter actually has on winning the game. We could also look at how much of a pitcher’s success is derived from what he does before the batter sends a ball in play.

So why isn’t this done already? My guess is that it’s more work and doesn’t really affect the WPA for any players at the end of the season. There is also the issue that there might not be a big enough sample to do an accurate job of figuring out the probabilities. I mean there are only so many times there has been a game in the bottom of the sixth for the home team where there are two outs, a runner on first and third, and a full count. If you don’t have enough games to make accurate predictions for all counts then this won’t work.

So just an interesting thought I was tossing around and figured I'd throw it up here.

Inherited Runners And ERA

Edgar for Pres — Sun, 06 May 2007 05:51:35 -0000

First off, to get to the point I’m proposing a new baseball statistic that I think would improve ERA while keeping the character of the stat. I've already talked about this some but I'm interested in what people have to say about it.

I think we will agree that ERA is one of the most used and deeply flawed stats available. It’s not a good at predicting future pitcher performances and there are other tools that do a much better job at that. I realize that there are plenty of stats out there like FIP but these don’t really tell us what the pitcher actually accomplished. They are more measures of how well a pitcher may have pitched. Sometimes I don’t want the feel good stats that try to tell us how good Jonny was doing out there and I just want to know how Jonny actually did and how many runs he actually allowed. The nice thing about ERA is that it attempts to simply quantify a pitcher’s past performances in one easy to understand stat. Sure there are other stats like WE that also provide this information but these seem too linked to the leverage of the situation that is partially out of the control of the pitcher. ERA is nice because it’s simply telling us how many earned runs the pitcher has allowed (per game pitched).

The two major flaws that I see in ERA which I think could be fixed revolve around inherited runners and fielding errors. (It might be only one major flaw because the fielding errors are an inherited runner problem too.)

When a starting pitcher leaves a game it is quite common that he might leave part way through an inning with a runner on base. Presently ERA just gives all the responsibility of what this runner does to the starting pitcher. In my mind this is not fair at all to the starting pitcher because if he leaves the game with a runner on 1B and two outs, the chance of that runner scoring is quite small but if the reliever comes in and allows the runner to score then the starter gets penalized with one ER and the reliever gets nothing. This also happens all the time when relievers are switched out too. My thought is to spread the blame for allowing inherited runners between the two pitchers. You could probably imagine a few ways of doing this but I think I’ve settled on my favorite by using a scoring probability chart (I think TangoTiger may have developed one) that gives the chance a runner scores in a particular situation. There might be a better way of assigning blame too. All you really need is the probability that a runner will score based on the bases occupied and the outs. But anyway that would allow that ER to be divvied up between the two pitchers fairly.

The second part of correcting ERA is to fix how fielding errors are connected to earned runs. I’m guessing that the creators of ERA didn’t have a clue what to do about fielding errors and instead just threw up their hands and said, "Screw it. Let’s just blame anything the runner does on that fielder." This approach makes some sense because that runner would probably be out but is now on base. This is much like the inherited runner case. Sure a runner is now on base that isn’t really the pitcher’s fault but the pitcher still has some control if that runner comes around and scores. I’d suggest that the fielder is credited with an ER value equal to the change in the scoring probability. The rest of the ER would be credited to the pitcher.

So as an example:
The bases are empty and there are no outs. From the run expectancy chart, the RE is 0.555 runs. Pitcher A allows a groundball to the SS. The SS throws wide of 1B and the runner reaches 1B on an error. This raises the run expectancy to 0.953 runs. The difference in run expectancy is 0.398 runs. The pitcher then allows a double and the runner from 1B scores. The fielder made an error that effectively cost the team 0.398 ER. This means that the pitcher is responsible for 0.602 ER (1 - 0.398 = 0.602).
This would work the same way with an inherited runner when a pitching change is made.

There is a little quirk with this that I could see. Should this be done for every situation runners are inherited or should it only be done when an inherited runner scores? I’m not sure if I have a good answer to this.

If you adjust the ER value whenever any inherited runners exist, more information is being recorded that will probably lead to a better description of how well a pitcher pitched. This technique could also lead to a negative ER value in some cases. Let’s say an error is committed that allows a runner to reach first base. The fielder is given an ER value of +n and the pitcher is given an ER value of –n. If the pitcher goes on to finish with no runs in the outing he’d finish with –n ER in the game and a negative ERA. This doesn’t make sense in real life but it does tell you more about how well the pitcher pitched.

On the other hand maybe the ER value should only be adjusted when any of the inherited runners actually score. This way ERA will conform more to what we traditionally think. This would get rid of the problem of negative ERA values.

So why am I putting this out there? Well I don’t really know how to implement it myself. I’m pretty good with excel but I don’t think I have good enough programming skills to actually do the work of going through play by play data to generate the actual results. I’ve thought about maybe sending some emails to the places I visit often like THT, fangraphs, baseballreference, ect. I’m not sure if something like this would pique their interest or not.

And the most important part. Would this actually tell us anything that ERA doesn’t already tell us? I’d say yes (otherwise I probably wouldn’t have written all this). Think about all those inherited runners you’ve seen come around and score because of that crappy reliever. I’m a Mariners fan so I think of guys like Bobby Ayala, Matt Thornton, and Julio Mateo. I’m pretty sure that this stat would expose these crappy relievers and show their true colors. They are making an impact on the amount of runs scored in a game and I think should be held accountable. The same can be said I think when an error is committed. Sure a runner might be on first base that isn’t your fault but that home run you gave up the next at bat isn’t helping the final score of the game either. This would also remove the ambiguity between R/9 and ERA because this stat would do a better job of dealing with errors.

I’m not sure if you agree with what I have above. I can see the case for not implementing any of this and instead not putting any of the responsibility on the pitcher for inherited runners but at some point the pitcher should share some responsibility for allowing inherited runners to score.

Here is another example:

Los Angeles - Bottom of 5th    April 21, 2007
Horacio Ramirez pitching for Seattle
O Cabrera walked.
V Guerrero walked, O Cabrera to second.
G Anderson doubled to left center, O Cabrera scored, V Guerrero to third.
J Mateo relieved H Ramirez.
S Hillenbrand grounded out to shortstop.
R Quinlan singled to left, V Guerrero and G Anderson scored, R Quinlan to second advancing on throw.
M Napoli singled to left, R Quinlan to third.
M Izturis popped out to third.
E Aybar reached on infield single to second, R Quinlan scored, M Napoli to second.
G Matthews walked, M Napoli to third, E Aybar to second.
O Cabrera grounded into fielder's choice to shortstop, G Matthews out at second.

So Ramirez comes in with the score 3 – 0 and gives up one run and leaves two runners on. Then Mateo comes in and relieves him. He inherits a runner on 2B and 3B with no outs. Then allows all the inherited runners to score and gives up one more of his own.

RE (run expectancy)
Empty / 0 outs: 0.552
2B + 3B / 0 outs: 2.052
RE difference: 1.5
This is the increase in the RE due to runners on base. This means that it is likely that 1.5 more runs will score in this situation compared to if there were no inherited runs.

So it’s expected that Mateo will allow 1.5 runs score in this situation. Mateo ends up allowing the two inherited runners score plus another of his own.
ER = ER(No Inherited Runners) + ER(Inherited passed on) – ER(inherited allowed to score)
ER(Mateo) = 1 - 1.5 +2= 1.5 runs (he actually had 1 ER in the boxscores)
ER(Ramirez) = 4 + 1.5 = 5.5 runs (he actually had 6 ER in the boxscores)

It’s not a huge difference but Mateo comes out 0.5 runs worse and Ramirez 0.5 runs better. If we look at Mateo over the entire season and adjust his line we will find that he should have 1.749 more ER (0.5 + 1.647 – 0.398). It also just happens that all the runners he inherited have scored and all his runners he has left on have scored so it doesn’t matter what method is used of the two that were proposed above. Before correction he has 5 ER in 12 IP for an ERA of 3.75. After correction he has 6.749 ER in 12 IP for an ERA of 5.06. The second line I think represents his performances much more accurately.

Poll