I'll just add SIERA to the table:

ERA/xFIP/SIERA since 2003

5.00/4.88/4.88

3.51/4.01/3.80

3.47/4.90/4.68

6.70/4.91/4.80

1.92/5.05/4.87

2.75/4.37/4.07

2.70/5.50/5.31

His SIERA seems to be lower by about 0.20 each year. This is probably a mixture of the fact that ground ball pitchers with high walk rates get more of a boost from their ground balls, and the fact that ground ball pitchers give up a lot of singles which can be turned into double plays on subsequent ground balls. Romero is actually a great example of a pitcher who SIERA does better with, based on how we modeled it.

Are there going to be 2010 projections for every pitcher's SIERA on your site?

HillMD wrote:Are there going to be 2010 projections for every pitcher's SIERA on your site?

Well, they aren't projections, though the SIERA numbers for both 2009 and for the 2010 PECOTA projections are in the 2010 Annual and I think will be on the PECOTA cards, too. The 2009 numbers will all be on the site soon (I keep being told soon. They made me it add it to the glossary today, so I think very soon). I assume it will be easier to sort and download for subscribers or something like that.

yo matt
just want to say... nice job going toe-to-toe with tango over at inside-the-book baseball. anyone here who wants to see what a real, old-fashioned sabr-debate looks like should head over there and check out their siera thread. (for those who don't know, tango is the father of fangraph's leverage index and war, and he works with mgl, the father of uzr.)

even the bp allstars have struggled in debates with tango over the years. as far as i can tell he's been a sabr heavyweight for a long time, going back to the usenet sabr boards, though i only first saw his stuff at baseball think factory (old, wild version, before baseball-reference). he and mgl and their disciples have always made my head spin. not sure if that's because they are super-sophisticated or because they have a really strange way of explaining their math or because i'm not very fast on the uptake. that you managed to figure out the nature of his objection to your work is pretty impressive.

phorever wrote:yo matt
just want to say... nice job going toe-to-toe with tango over at inside-the-book baseball. anyone here who wants to see what a real, old-fashioned sabr-debate looks like should head over there and check out their siera thread. (for those who don't know, tango is the father of fangraph's leverage index and war, and he works with mgl, the father of uzr.)

even the bp allstars have struggled in debates with tango over the years. as far as i can tell he's been a sabr heavyweight for a long time, going back to the usenet sabr boards, though i only first saw his stuff at baseball think factory (old, wild version, before baseball-reference). he and mgl and their disciples have always made my head spin. not sure if that's because they are super-sophisticated or because they have a really strange way of explaining their math or because i'm not very fast on the uptake. that you managed to figure out the nature of his objection to your work is pretty impressive.

thanks. arguing with tango is very tricky. i think some of the problem people have understanding him is that his writing is often not all that clear, so it's tough to figure out what he's doing. i think as someone who has struggled to write up mathematical ideas clearly in the past, i can following him a little better. there's also a lot of "required reading" to talk to him because he ascribes to a very different set of beliefs.

it's very challenging to argue with him, in no part because he has a lot of disciples that chime in and argue with you too. in general, he doesn't really say he's wrong but reframes the argument which he has done by basically saying "how much better is it really to do it this complicated way?" in the end. i'm not sure that his ways are even simple, but that's obviously somewhat of a judgment call.

i think it's tough to figure out how far to go arguing with him, but i think he has enough influence over sites like the hardball times, beyond the boxscore, fangraphs, etc., that it's important to at least challenge him sometimes so that the non-BP sabermetric community doesn't dismiss SIERA outright.

thanks again for the compliment. it really is a challenge to argue with him.

Matt, this is fascinating and I will try to find your crossed-SABRs with the estimable Tango.

The other night on Hot Stove they were talking about Win Probability Added for closers, noting that Brad Lidge was the worst in 2009 after being the best in 2008 (and 2004).

Any correlation between WPA and SIERA?

Buzhardt wrote:Matt, this is fascinating and I will try to find your crossed-SABRs with the estimable Tango.

The other night on Hot Stove they were talking about Win Probability Added for closers, noting that Brad Lidge was the worst in 2009 after being the best in 2008 (and 2004).

Any correlation between WPA and SIERA?

Yeah, WPA should correlate with any good metric. I haven't checked it explicitly but WPA is going to incorporate all the noise that shouldn't be in there like F-Rod getting a -0.8 wins for his popup that Luis Castillo dropped. SIERA will smooth some of that out will picking up the important stuff.

MattS wrote:
thanks. arguing with tango is very tricky. i think some of the problem people have understanding him is that his writing is often not all that clear, so it's tough to figure out what he's doing. i think as someone who has struggled to write up mathematical ideas clearly in the past, i can following him a little better. there's also a lot of "required reading" to talk to him because he ascribes to a very different set of beliefs.

it's very challenging to argue with him, in no part because he has a lot of disciples that chime in and argue with you too. in general, he doesn't really say he's wrong but reframes the argument which he has done by basically saying "how much better is it really to do it this complicated way?" in the end. i'm not sure that his ways are even simple, but that's obviously somewhat of a judgment call.

i think it's tough to figure out how far to go arguing with him, but i think he has enough influence over sites like the hardball times, beyond the boxscore, fangraphs, etc., that it's important to at least challenge him sometimes so that the non-BP sabermetric community doesn't dismiss SIERA outright.

thanks again for the compliment. it really is a challenge to argue with him.

tango is a treasure, but more like myrrh than frankencense.
and he and his gang (i was going to mention them in my previous post... i agree, they are more annoying than the man himself) will never, ever forgive bp for the paywall and secret formulas.

i agree that the communication problem is a combination of the obtuse writing style (which i should forgive... i'm worse than either of you) and the perspective. perspective wise, i think it's an issue of inverse modeling (e.g. regression) versus forward modeling (eg markov chains). i've seen seismologists get pretty darn angry about the same clash of perspectives.

my take, and my research approach, is to combine the two. using forward modeling to help identify important regression parameters and to test the robustness of a regression model. if you want to tango-proof your model, you'll probably need to do some of that.

tango's forward modeling is actually too simplistic for my tastes. a good workstation can do a lot more with the right software, and a little cluster (does bp have one?) can do even more. if you want/need any hints, pm me (i don't check my pm box much, but i'll keep an eye open this week).
i'm too lazy/inefficient to do sabr-stuff in addition to my day job, but i would love to do a little coaching if that would be of any use to anyone.

for those looking for the sabr-board (tango) discussions i mentioned...

tango's first siera thread

tango's second siera thread

yeah, i think the anger is really a lot about what they call the "paywall" which is precisely why tango's website is actually one big advertisement for his book, i guess. the concept of calling it a paywall is just funny. THT sells a book and has batted ball reports that are pay only. Bill James' website is pay only. Lots of people sell things that other people want. I never understood the concept of paying for things being some evil paywall or why sabermetrics was automatically supposed to be not-for-profit.

they also do like there component based modeling where they build things up from individual events, but i think that really does miss a lot of things, and can't be used exclusively. i'm sure that this negative quadratic term in groundball rate is very real in SIERA since it's consistently true, but i'm not sure that it could all be explained by components. why is it that going from a 40% to 50% GB pitcher doesn't drop your ERA as much as going from a 50% to 60% GB pitcher. It's true in every subset of data we run. it's a large effect that is significant. and it probably belongs largely to situational pitching and other things like that. that will get lost in their way of doing things. you need both, i think, like you said.

i don't actually know what clustering is, so let me know if you think it would be something i could do that help my sabermetrics work. i've realized since all the recently added people joined BP that we all have really specialized abilities in statistics and i'm clueless about a number of different techniques.

Looks intriguing... the analysis of strikeouts and groundouts having accelerating/decelerating returns is certainly pretty novel. Just curious, how significant did those quadratic terms turn out to be when you estimated the model?

TheBrig wrote:Looks intriguing... the analysis of strikeouts and groundouts having accelerating/decelerating returns is certainly pretty novel. Just curious, how significant did those quadratic terms turn out to be when you estimated the model?

yeah, i really was surprised by that too. the 3rd article is here. they were VERY significant. t-values of 1.97 and 3.94 for the strikeout and ground ball squared terms respectively (p-values .049 and .000). they also held up as we used different subsets of data very consistently.

MattS wrote: i'm sure that this negative quadratic term in groundball rate is very real in SIERA since it's consistently true, but i'm not sure that it could all be explained by components. ...
i don't actually know what clustering is, ...

forgot to say, i absolutely, totally love seeing the quadratic terms. the "linear" part of linear regressions and linear weights in baseball has been bugging me for years. lots of the universe is linear to a good approximation... but lots isn't. and a big share of what isn't approximately linear is approximately quadratic instead.

there is such a thing as cluster analysis, but in this case i was talking about hardware... linking a bunch of pc's together to make a mini-supercomputer. combine that with parallel compilers and a properly parallelized code and you can either invert a very big matrix very fast or run a whole bunch of model tests simultaneously. 16 pc's (leftover after an office-wide upgrade at bp?) will work nicely. i have a bright undergrad building one for me right now.

my main reason for doing forward modeling is that i don't trust p-values or variance reduction or "clinical assumptions" when it comes to determining which terms are and aren't significant.

Hey Matt, how many BP Idol contestants actually got hired by BP? Was it just you and Ken Funck?

CrashburnAlley wrote:Hey Matt, how many BP Idol contestants actually got hired by BP? Was it just you and Ken Funck?

Tim Kinker writes occasionally and Jeff Euston just got hired too as PEV bought up Cot's Contracts. Euston and I are the only two writing weekly as near as I can tell, but I think Kniker and Funck may when they are less busy.

phorever wrote:even the bp allstars have struggled in debates with tango over the years. as far as i can tell he's been a sabr heavyweight for a long time, going back to the usenet sabr boards, though i only first saw his stuff at baseball think factory (old, wild version, before baseball-reference). he and mgl and their disciples have always made my head spin. not sure if that's because they are super-sophisticated or because they have a really strange way of explaining their math or because i'm not very fast on the uptake. that you managed to figure out the nature of his objection to your work is pretty impressive.

BPro-ers have generally avoided debating with me. They are not interested in getting to the truth or having their readers learn anything. Matt is different, if for no other reason that he's been a long-time poster on my blog.

I'm from Baseball Boards.com and later Baseball Primer.

I'll say we have a strange way (for you and maybe most people) of explaining things. Strange of course does not mean bad. Just very different.

MattS wrote:thanks. arguing with tango is very tricky. i think some of the problem people have understanding him is that his writing is often not all that clear, so it's tough to figure out what he's doing.

Interestingly, it's perfectly clear in my head. I can accept that something is getting lost in the translation to the written word.

i think it's tough to figure out how far to go arguing with him, but i think he has enough influence over sites like the hardball times, beyond the boxscore, fangraphs, etc., that it's important to at least challenge him sometimes so that the non-BP sabermetric community doesn't dismiss SIERA outright.

I would hope that even if I have no influence whatsoever that I should be challenged on the merits.

To the extent that I have influence, it's completely based on my going under the hood and stripping everything down to the bone so that I can completely understand something, and give my seal of approval (or disapproval as the case may be). Basically, people trust me. I'm guided by the facts, and perform all my testing. I'm unbiased, and I'll admit I'm wrong when I'm wrong.

And I confront any and all challenges.

thanks again for the compliment. it really is a challenge to argue with him.

As I tell my kid learning piano: If it was easy, everyone would do it! If it is a challenge, it's because I demand alot of whoever I'm sparring with.

In the end, the entire point of the debate is to advance knowledge to the point where the interested reader will learn something and can move forward in some direction.

I believe that without my thread, SIERA would simply exist the way QERA has existed and MORP and WARP and VORP has existed: those other metrics had problems, clear problems, simple-to-correct problems. But, it seems that I'm the only one that bothers to point them out enough that others take notice. Where does SIERA stand? I don't know yet, and that's the point of the thread.

he and his gang (i was going to mention them in my previous post... i agree, they are more annoying than the man himself) will never, ever forgive bp for the paywall and secret formulas.

I can't speak for "my" gang (they are actually unaffiliated to me), but I have no problem with their paywall. I have 100% problem with secret, and not-so-secret formulas that don't necessarily work. And I have some problem with having all my legitimate concerns in the metrics being unanswered.

Matt is a breath of fresh air in this regard.

tango's forward modeling is actually too simplistic for my tastes.

A Markov chain is too simplistic?

yeah, i think the anger is really a lot about what they call the "paywall"

No anger from me. If you mean it from others, that's fine. And I would say this is only a minority of my readers, given that a substantial number of them are subscribers of BPro.

which is precisely why tango's website is actually one big advertisement for his book, i guess.

Actually, the blog part of the website exists on its own, and is not an advertisement in any way for the book. Indeed, any time I reference The Book, I tell people to read it for free from Amazon's Look Inside. And MGL doesn't care about advertising, since all the money he makes he's donated to Retrosheet. (Notice how the last two years Dave Smith of Retrosheet made an announcement for money that they don't need the money? Thank MGL for that.)

the concept of calling it a paywall is just funny. THT sells a book and has batted ball reports that are pay only. Bill James' website is pay only. Lots of people sell things that other people want. I never understood the concept of paying for things being some evil paywall or why sabermetrics was automatically supposed to be not-for-profit.

No one said it was an evil paywall. It's a paywall. Bill James has a paywall. THT has a book that you have to pay for. The term "wall" is used because if you click to it, you will be stopped from going through. I'd be happy to use another single-word or two-word term to describe to the reader what happens if they click a link to an article they can't read in full, if you want to recommend that term.

they also do like there component based modeling where they build things up from individual events, but i think that really does miss a lot of things, and can't be used exclusively.

I never said to use anything exclusively.

why is it that going from a 40% to 50% GB pitcher doesn't drop your ERA as much as going from a 50% to 60% GB pitcher. It's true in every subset of data we run. it's a large effect that is significant. and it probably belongs largely to situational pitching and other things like that. that will get lost in their way of doing things.

That is an interesting finding. I'll wait to see your article to see some empirical data to see the extent that this is true.

....


I watch the games.

holy shit, that just happened. okay, let's go:

tangotiger wrote:BPro-ers have generally avoided debating with me. They are not interested in getting to the truth or having their readers learn anything.

I think the reason is that you don't ever concede when they are right, and the debate generally descends into methodological assumptions. I have a hard time with the argument that BP doesn't want their readers to learn anything. I would argue that BP is as responsible for casual baseball fans knowing sabermetrics as anybody but Bill James and Michael Lewis. Using VORP instead of RBI is a bigger gap than VORP to VORP with EqR instead of RC.

Matt is different, if for no other reason that he's been a long-time poster on my blog.

I wouldn't call that causal.

To the extent that I have influence, it's completely based on my going under the hood and stripping everything down to the bone so that I can completely understand something, and give my seal of approval (or disapproval as the case may be). Basically, people trust me. I'm guided by the facts, and perform all my testing. I'm unbiased, and I'll admit I'm wrong when I'm wrong.

And I confront any and all challenges.

See, that is not why you have influence. You have influence because your blog is a meeting place for sabermetricians because the way to become a well known sabermetrician outside of BP Idol is to write articles that get linked on your blog. You create traffic for smaller websites by picking the articles you find interesting and linking to them. It's not about stripping things down, because if you used BP's methods and BP used your methods, people would use EqA and VORP because that is the way to get linked on your blog. I'm not talking about elite sabermetricians, I'm talking about still learning ones. It's easier to get noticed if you get linked on your blog.

A Markov chain is too simplistic?

It makes assumptions just like regression does. It makes assumptions that are not all accurate but are close enough in many case. Regression makes assumptions that can be manipulated to your favor if you know how to manipulate the correlation between the regressors and the noise term.

yeah, i think the anger is really a lot about what they call the "paywall"

No anger from me. If you mean it from others, that's fine. And I would say this is only a minority of my readers, given that a substantial number of them are subscribers of BPro.

Less are subscribers because of your negativity towards BP. You did a poll and a large fraction weren't. Ultimately, I don't see how you can consider yourself a serious sabermetrician unless reading sabermetrics isn't about doing a literature review. And you can't complete the literature review of barely any topic in sabermetrics without a subscription to BP. Most topics are hard without reading The Book too, but BP is part of any literature review, and seeing how you encourage that type of thing, you should encourage subscribing. How about instead of 'behind the paywall' you say, 'complete your literature review at three cents an article.'

I think that BP would have way more subscribers if people didn't refuse to pay out of protest. It's a typical irrationality of economic decision making where people won't pay for something out of principle like that when their utility for the articles is much higher than what they are paying for them. Many people don't want to pay for baseball on the internet. They'll spend money on hats and tshirts and everything else baseball related, but they have an aversion to paying for something online. This is exploited when you convince people they are getting ripped off.

which is precisely why tango's website is actually one big advertisement for his book, i guess.

Actually, the blog part of the website exists on its own, and is not an advertisement in any way for the book. Indeed, any time I reference The Book, I tell people to read it for free from Amazon's Look Inside. And MGL doesn't care about advertising, since all the money he makes he's donated to Retrosheet. (Notice how the last two years Dave Smith of Retrosheet made an announcement for money that they don't need the money? Thank MGL for that.)

MGL has an option to keep the money and spend it on a vacation. MGL has an option to donate the money to RetroSheet and enjoy the feeling of giving to a cause he believes in as well as helping it. He chooses the latter because he likes it even more than spending the money on a vacation. Regardless, the money is spent on a way of his choosing, and therefore the incentive to make money is there all the same. Good for him that his preferences are such that he places a high value on RetroSheet. But there is still an incentive to make money from The Book and that's my point.

It's an advertisement for the book in the sense that it encourages people to buy the book. That you encourage people to read it through Amazon is great, but I never figured out how to do that so I ordered a used copy.

why is it that going from a 40% to 50% GB pitcher doesn't drop your ERA as much as going from a 50% to 60% GB pitcher. It's true in every subset of data we run. it's a large effect that is significant. and it probably belongs largely to situational pitching and other things like that. that will get lost in their way of doing things.

That is an interesting finding. I'll wait to see your article to see some empirical data to see the extent that this is true.

The ((GB-FB-PU)/PA)^2 coefficient is statistically significant at the 99.9% level. That's the implication of that finding.

MattS wrote:holy $#@!, that just happened. okay, let's go:

Sounds like something one would say after being punched or pushed really hard.

or seeing Tom fucking TangoTiger post a gigantic fucking response on your diddly poo Phillies board