Got questions about statistical analysis? Ask. We might be able to help.
There are people of all sorts on Lookout Landing. Some of you are new to this, some are old hands. Everyone should feel free to chime in with questions, answers, points of contention, etc. Just try to be nice.
1 recs | 327 comments
Links to any credible defense system would be helpful
As would instructions on how to calculate WAR
JI - December 14, 2008
http://www.insidethebook.com/ee/index.php/site/article/how_to_calculate_war/
Dewey N - December 14, 2008
Ugh.
I need a website that does it for me.
JI - December 14, 2008
It would be nice
I’m hoping fangraphs combines wRAA and UZR to make a WAR stat.
chrisisasavage - December 14, 2008
It's fairly rudimentary but I think it's pretty close:
http://spreadsheets.google.com/ccc?key=pmWvhxDox3ZipyzRW3P0jYQ
http://spreadsheets.google.com/ccc?key=pmWvhxDox3Zh2nhfbl1o9Mg
Aaron Campeau - December 14, 2008
And now there's something funky with the AL one.
Aaron Campeau - December 14, 2008
Okay fixed and actually accessible
AL WAR
NL WAR
Aaron Campeau - December 14, 2008
ASS
Two shakes…
Aaron Campeau - December 14, 2008
Okay this should work.
AL WAR
NL WAR
Aaron Campeau - December 14, 2008
What statistic are you supposed to put in each column
Dewey N - December 14, 2008
I just realized that it's using the positional adjustment from the links.
JI - December 14, 2008
.
wOBA*, <x< defensive number (i.e. I’d put 10 for Beltre because he’s a 5<x<15 defender,) positional adjustment and playing time expressed as a decimal. Tango has a brief explanation of how to figure playing time in his original calculation WAR post.
Aaron Campeau - December 14, 2008
Pujols
.458, 20, -1, .85 = 8.56
JI - December 14, 2008
A-Rod, 2003
8.22
Hmmm, seems kind of low.
*Though I did have to use the NL sheet cos the AL one is fucked up at the moment.
JI - December 14, 2008
I would use 2002
Because he was even better, but I don’t have wOBA* for 2002.
JI - December 14, 2008
Using the NL sheet
A-Rod 2005: 8.83
Jesus
JI - December 14, 2008
Bonds 2002
111.1 wRAA, +10.5 UZR
Fett42 - December 15, 2008
Yes it is.
If anyone knows how to carry rules through an entire column, that would be good to know.
Aaron Campeau - December 14, 2008
Oh ok playing time is a decimal
That’s what was throwing me
Dewey N - December 14, 2008
I just realized that more than one person editing this at a time could turn out poorly.
Aaron Campeau - December 14, 2008
I fucked it up
sorry
Dewey N - December 14, 2008
No, I never used more than one row so I never had the other rows calculate automoatically.
Aaron Campeau - December 14, 2008
Next time Statcorner does some house cleaning
they should totally add this in
JI - December 14, 2008
Silly kids
They don’t measure runs prevented on the scoreboard. Just give me a guy who hits RBI’s or hits for average, and puts his heart into it, and hustles on every play, and I’ll show you a good player. You guys valuing make believe stats like UZR and overvalue the rally-killing walk.
For pitchers give me ERA or W/L, and that’s all I need. Joe Morgan told me so.
chrisisasavage - December 14, 2008
OK this is neat
JI - December 14, 2008
It is the crowing achievement of my life to this point.
Aaron Campeau - December 14, 2008
Do pitchers
GO GO GO
JI - December 14, 2008
Those numbers are way scarier.
Aaron Campeau - December 14, 2008
I'm making it so multiple people can access this but they're offline for a few days FYI.
Aaron Campeau - December 15, 2008
This means WAR
chrisisasavage - December 14, 2008
I'm not allowed access. :-/
JI - December 14, 2008
here's a pretty good start
http://www.fangraphs.com/blogs/index.php/wrc-and-wraa
take the wRAA, add defensive value, 22.5 runs for replacement level, Tango’s positional adjustments, divide by 10 and presto
Bearskin Rugburn - December 14, 2008
Try
(wRAA * PA / 700 + 22.5 + UZR/162 (UZR/150 * 162/150) + Positional Adjustment) * PA / 700
chrisisasavage - December 14, 2008
I meant
wRAA * 700 / PA for the first bit
chrisisasavage - December 14, 2008
UZR is about it, as far as I know that's available in an easily viewable form.
Getting WAR
Matthew - December 14, 2008
Also try to avoid using non pbp metrics in assessing player defence
The outputs just aren’t enough to draw real conclusions from
Graham MacAree - December 14, 2008
Read: FRAR, FRAA
JI - December 14, 2008
Yeah pretty much
Graham MacAree - December 14, 2008
Don't forget Defensive Win Shares
Those are also bad, and almost more difficult to talk people out of because Bill James invented them.
ubelmann - December 14, 2008
This is interesting and fairly rigorous.
http://www-stat.wharton.upenn.edu/~stjensen/research/safe.html
I can’t help with the WAR. No pun intended.
two_hands - December 15, 2008
woot!
Matthew - December 15, 2008
I like how for those years it puts Arod 2nd w/ an average of +10.4 while Jeter's dead last and -13.81
Which suggests that the Yanks might have given up as much as 2 wins in a season just for giving Jeter his spot.
Fett42 - December 15, 2008
Also, why is the positional adjustment for thrid, center and second the same?
JI - December 14, 2008
oh wait they're not
JI - December 14, 2008
Oh wait, they are.
Shows what you know.
JI - December 14, 2008
Tango updated the positional adjustments recently but I lost the link.
Aaron Campeau - December 14, 2008
Here they are:
+12.5 C
+7.5 SS
+2.5 2B/3B/CF
-7.5 LF/RF
-12.5 1B
-22.5 DH
Matthew - December 14, 2008
So, 2B/3B/CF being equal strikes me as strange.
Talk me down?
JI - December 14, 2008
A lot of it has to do with talent pools.
This somewhat came up before here at LL (I think), but the notion of a defensive spectrum as Bill James laid it out is correct, but not the end-all, be-all. It’s simplistic and glosses over that positions require different skills and those skills are not linearly less important as you move down the spectrum.
In other words, a poor catcher cannot just be converted into an average SS.
Because of that, the distinctions between certain positions become muddy. We’ve seen more than a few people move between 2B and CF (Upton, Biggio, Burke), and 3B ends up as sort of an isolated position. Rarely staffed by fielders competent enough to hold down SS, but still needing a strong enough arm that 1B would be a waste.
Matthew - December 14, 2008
Link to my quick/dirty post on the defensive spectrum
Here
Graham MacAree - December 14, 2008
Ah, nice article. Clears up a few questions.
Kermit. - December 14, 2008
My concern,
Thridbase is almost always out hits second on a league wide basis, and it generally requires less range to play.
JI - December 14, 2008
Third requires a better first step and a much much better arm
Graham MacAree - December 14, 2008
You can get by at second base without a great first step or a great arm...
…but they are still awfully helpful towards increasing your range, or turning double plays. They also see 60 fewer BIZ each year, so any shortcomings are exposed somewhat less often than shortcomings at 2B are.
My intuition tells me that if third basemen outhit second basemen as a group, then the pool of players who can defensively handle third base is larger than the pool of players who can defensively handle 2B. Thus, 2B are a more scarce resource so that an average overall 2B should be more valuable than an average 3B.
Like I said, that’s my intuition, so I’m certainly open to arguments to the contrary. I’m still not 100% convinced by anyone’s argument what the “correct” way to do positional adjustments are.
(Further caveat: If the claim is that the 3B adjustment is +2.5 and the RF/LF adjustment is -7.5, then at most I’m thinking that 3B should move down to +0 or -2.5, and that’s a pretty small difference from +2.5, all things considered.)
ubelmann - December 14, 2008
I'm not stating my beliefs here, really
I’m just providing a possible argument as to the why
Graham MacAree - December 14, 2008
Cool
I definitely agree that that’s a possible argument and that it’s good to keep in mind that the different positions require somewhat different skills.
ubelmann - December 14, 2008
I think I heard somewhere
that 3B hitting historically is about as good as 2B hitting. I don’t know where I heard that. Tango maybe?
Edgar for Pres - December 15, 2008
Back in the old days, third and second were basically flipped on the spectrum.
JI - December 15, 2008
Put another way
There is no position where the skills necessary to excel there will also necessarily make you an excellent 3B.
Like Catcher, 3B relies on a unique combination of skills. However, those skills are not as rare in combination with good hitting as Catcher skills are, so 3B gets a smaller adjustment.
Llewdor - December 15, 2008
Are these determined through regression analysis?
I understand the concept of why they would be weighted differently, I’m asking how these numbers are arrived at.
Kermit. - December 14, 2008
To quote from Sean Smith, because he says it succintly
“These are based on defensive statistics, not offense. He looks at how players who play multiple positions perform when switching for one to another, with some consideration for handedness, as lefties do not play second, third, or short.”
Matthew - December 14, 2008
I guess I'm not totally sold on using players switching positions to determine the positional adjustment
Even other than the SSS issue, I wouldn’t use, for instance, C→2B conversions to determine the relative adjustment between positions, even though that conversion happens every now and again, because the skills for the two positions are so different.
With other positions—like 2B and 3B—there are a lot more people who play both positions, and it might be tempting to use those players to figure out the adjustment, but there are still different skills required in playing the two positions. And it seems to me that players who play both positions aren’t necessarily a representative sample of defenders at each position, but that might be wrong.
ubelmann - December 14, 2008
How much better (if any) is tRA compared to FIP or last year's RA?
Edgar for Pres - December 14, 2008
Better in which way?
Graham MacAree - December 14, 2008
What is the average percent error in year 2's RA given tRA or FIP of year 1?
Edgar for Pres - December 14, 2008
Dont have that offhand, but this raises an important point.
Namely, what’s the goal? Why should projecting RA next year be a worthwhile endeavor?
Matthew - December 14, 2008
When it comes to pitching stats...
…I always think that it’s bizarre that projection systems have traditionally been measured with correlation to ERA, when we know that ERA is so team/luck-dependent.
If tRA describes to us what the pitchers do separately from what the defense does, and we want to know how good a pitcher is going to do in year N+1 (because we’re major league GMs and need to assemble a team for next year), it seems like we should be trying to project tRA, not RA or ERA. (Also, on general grounds I object to the use of correlation coefficient over something like a chi-squared statistic since the pitchers have different innings pitched totals, so the variance in each ERA or tRA is different, but that complaint probably belongs in a different discussion.)
ubelmann - December 14, 2008
You need to check how good a stat is somehow
Just making up a stat and saying its good isn’t enough. Eventually you need to try to prove that its better than other stats already out there. You can’t just say “It should be good”. People want to know how good it actually is.
I’m not saying that tRA should be a great predictor of year N+1 but it should be better than FIP, xFIP, etc. If it isn’t then all that extra work Graham has done which should make it a better stat has in effect shown very little reward. Nobody wants to hear that and I think tRA is a better stat than anything out there right now but to validate that belief, we need to test it.
Edgar for Pres - December 15, 2008
Yes, but you have to test it the right way and against RA is the wrong way.
The correlation between FIP/tRA and RA really is meaningless. It’s like proving wOBA is good by showing that it predicts RBI totals better in year N+1 than RBI in year N does. It doesn’t make sense.
The measure is how tRA (or better tRA*) correlates with tRA(*) N+1 compared against RA and RA N+1.
Matthew - December 15, 2008
But how do you know that tRA is measuring anything that relates to runs allowed
To use your example, RBI totals in year N predict RBI totals in year N+1 well but that doesn’t mean that RBI is a valuable stat.
How can you prove to me that the weights used to calculate tRA in the end produce a stat that effectively measures the talent of a pitcher?
I’m a huge fan of tRA but I do think it needs to be related back to something rooted in reality to prove its worth.
Edgar for Pres - December 15, 2008
RA isn't a valuable stat either.
Matthew - December 15, 2008
Yes it is
Which team wins more games? The team with the higher RA or tRA. RA is an outcome stat. In the end of the day all of our stats are trying to be as good as a stat that keeps track of outcomes.
Edgar for Pres - December 15, 2008
It's not a valuable stat for a pitcher.
Matthew - December 15, 2008
I agree with that
It doesn’t tell you how good the pitcher necassarily is.
Edgar for Pres - December 15, 2008
Ok here's a question
How do you prove that FIP is a useful stat? I understand that K, BB, and HR are useful stats but how do you prove to me that the combination of the three and the weights assigned are valid and the final numbers FIP produces have actual meaning?
Edgar for Pres - December 15, 2008
You show that HR, K and BB are independent of the defense and that their repeatability yeat-to-year
is much higher than superfluous things like hits allowed. You then do modeling based on linear weighting to determine how the outcomes relate to each other in value.
Matthew - December 15, 2008
Can you tell someone how "good" FIP is?
Can you tell me if xFIP is better than FIP? Can you tell me how much better xFIP is than FIP?
Edgar for Pres - December 15, 2008
Yes, that would be the relative difference in year to year correlation with itself.
Matthew - December 15, 2008
So thats all you can do?
I really am curious how we test these stats. It seems like there must be something better.
Edgar for Pres - December 15, 2008
Hundreds of insanely smart people have been trying to improve pitching metrics for years.
Aaron Campeau - December 15, 2008
What you want is an actual number of a pitcher's talent to measure the stat against,
but if we had that actual number, we would just use that.
Matthew - December 15, 2008
Yeah I guess thats it
Too bad I guess. Next time I’ll try to phase my point better to get at this faster instead of taking 20 posts to get there.
Edgar for Pres - December 15, 2008
'the weights assigned are valid'
The weights assigned must be valid as the process of deriving the weights is giving a true comparison of the HR/BB/K outcomes and runs scored.
It’s a by definition thing, as I’ve said many times
Graham MacAree - December 15, 2008
I'm happy you are confident in them
I am too. I just like the idea of testing this type of stuff when you get to the final result. Since I guess this is impossible then I guess I’ll just have to be fine with that.
Edgar for Pres - December 15, 2008
I also want to make it clear
I know tRA doesn’t predict RA because it removes defense from the equation. I just want a method to demonstrate that the stat actually relates back to runs allowed.
How do I know that a pitcher with a 3.2 tRA is better than a pitcher with a 3.5 tRA? Do you have anything that will back that up?
Edgar for Pres - December 15, 2008
On a league-wide scale
tRA has an “error” rate of less than 1% on actual runs allowed.
Matthew - December 15, 2008
Yeah but what about for individual players?
Edgar for Pres - December 15, 2008
For individual players, RA is meaningless
Matthew - December 15, 2008
Stats that predict team outcomes well
Do not necessarily predict individual player outcomes well.
And, yeah I know RA is crappy.
Edgar for Pres - December 15, 2008
Do you have this same problem with wOBA?
It’s derived from the average outcomes of a play. The more good outcomes that a player produces, the better.
Graham MacAree - December 15, 2008
I don't have a "problem" with wOBA or tRA
I’m just wondering if there is a good way we can test these stats to make them more accurate.
Edgar for Pres - December 15, 2008
Their accuracy comes from properly weighting the stats to begin with
If they are properly weighted, they are, by definition, accurate.
Graham MacAree - December 15, 2008
SLG, for example, is an improperly weighted stat
It’s 1/2/3/4 weightings don’t correspond to actual runs scored.
Graham MacAree - December 15, 2008
I probably should have phrased it 'difficulty' or something
Graham MacAree - December 15, 2008
Why would you want to do that when you know how misleading that can be?
I mean, Washburn had a 3.2 ERA in 05, and if you made a predictoin based on his 04 numbers, you’d have to come to the conclusoin that it’s crap.
Bearskin Rugburn - December 15, 2008
The whole derivation of tRA, yes.
Graham MacAree - December 15, 2008
I believe the whole thing about tRA
I buy everything you put into it. I just think there has to be a good way to take it all and compare it back to outcomes.
Edgar for Pres - December 15, 2008
tRA is entirely derived from outcomes.
Matthew - December 15, 2008
Exactly
It just looks at the pitching outcomes before the fielding process can scramble them
Graham MacAree - December 15, 2008
Yeah got it
On a side note, do you have concerns about the stat being based on FB/LD/GB where these are slightly arbitrary? If you had hit locations and hit speed how much do you think this would improve the stat.
Edgar for Pres - December 15, 2008
10% or so would be my off the cuff guess.
It does suck to rely on somebody’s FB/LD/GB distinctions instead of having a good BIP system, but since that’s only a part of the overall stat and a majority of them wouldn’t change with increased precision, I think the overall improvement would be significant, but relatively minor
Matthew - December 15, 2008
I guess outcomes was the wrong word
I was thinking runs allowed but since the comparison between a talent stat and something like runs allowed can never be done totally unambiguously, then its a mute point.
I’m just built to think of problems from the standpoint of the scientific method and like to have many ways to test to see if my hypothesis was correct. I trust all the work you guys have done and love using tRA. If I write a paper and only have one way to back up my theory it probably won’t get published anywhere. If you can’t do anything else then that’s fine and that happens but it would be great to have a library of reasons why tRA is the shit.
Edgar for Pres - December 15, 2008
The same way that you know that OBP is measuring something that relates to runs scored
Combine OBP with the other offensive stats to see how well it predicts runs scored within that season.
With tRA, combine it with other run prevention stats (specifically, those related to defense) and see how well it predicts runs allowed within that season.
Projecting RA in year N+1 is dependent upon how will you can: a) predict the future, b) measure a pitcher’s value, and b) measure a defense’s value.
ubelmann - December 15, 2008
Too many b's
ubelmann - December 15, 2008
It measures the pitcher's role in runs allowed,
but because you can’t uncouple position players’ role in that from the number, it makes no sense to compare against it directly. If you must compare it to actual RA you would have to normalize RA to the defense behind the pitcher using whatever metrics are available and regress it against bad/good luck and park factors. But then you’re just calculating tRA from the other end.
Bearskin Rugburn - December 15, 2008
I think one of the core issues here
is are you trying to predict RA with your stat or are you trying to measure a pitcher’s quality/value. If it’s the former, well, it’s a questionable goal. If it’s the latter, correlation with RA or FIP doesn’t matter. However, a good measure of a pitcher’s skill should be fairly predictive of itself, because outside of the natural variation a pitcher remains who he is.
Bearskin Rugburn - December 15, 2008
I've only checked this for one year of tRA* back when I first came up with it
Correlation for park-unfactored tRA* to RA was over 0.70.
Need to do a longer study at some point, but that’s a really nice number.
Graham MacAree - December 14, 2008
It's based on peripherals
that have better year to year correlation. FIP is pretty good, but HR is a problem, and not all batted balls are the same.
chrisisasavage - December 14, 2008
xFIP solves a lot of the HR problem
xFIP is a good pitching stat.
chrisisasavage - December 14, 2008
It could really do with regressing the values properly rather than just going to a league average HR/fly though
Graham MacAree - December 14, 2008
I get that they are good
but how good are FIP, xFIP, tRA, and tRA*. I’d really like some numbers to evaluate the difference in accuracy between them all.
Edgar for Pres - December 14, 2008
Since hit/fx is a impossible dream
what is the likelihood that we come up with the perfect hitting statistic?
Robert - December 14, 2008
wOBA is not bad
Any linear weights type system is good for measuring hitting. The closest thing to perfect would be a regressed wOBA, which I believe wOBA* does. Park factors help too.
chrisisasavage - December 14, 2008
wOBA* is just park factored, not a regression stat.
Graham MacAree - December 14, 2008
So I had it backwards :)
chrisisasavage - December 14, 2008
Which makes sense
because as unreal as Pujols is a 0.468 wOBA cannot be a regressed statistic, except maybe for 2001-2004 era “Super-Bonds”
chrisisasavage - December 14, 2008
You'll live.
I’d like to see if I could come up with a prOPS/wOBA hybrid that weights past run values on batted ball type with league average values, which should probably be better than straight up wOBA. But good god that’ll be complex.
Graham MacAree - December 14, 2008
It would just be easier if they'd
come up with hit f/x or if there was something like Hit tracker for all batted balls.
chrisisasavage - December 14, 2008
Even then it'd be difficult
A weakass grounder to shortstop from Bengie Molina is not the same as from Ichiro, so you’d have to look at past run values for the batter too
Graham MacAree - December 14, 2008
I suspect
speed scores and handedness would be decent inputs for a regression.
I also suspect you’d still need to regress it against actual performance.
Of course I have no way of knowing that for sure.
chrisisasavage - December 14, 2008
Is there any chance we will ever see HITf/x?
BrianL - December 14, 2008
Somebody has to pay for it. That's the hold up.
Matthew - December 14, 2008
Damn.
BrianL - December 14, 2008
What is the age that pitchers and hitters are generally supposed to start declining at?
And do projection systems project declines and improvements on a case-by-case basis or just one general set of ages for everyone?
Dewey N - December 14, 2008
Depends on the system. PECOTA, being advanced, is more case-by-case
Hitters peak around 27 on average and therefore start declining at that age. However, 30 is a good mark for when hitters’ aging should start to be a concern. And that’s going to differ a lot based on the hitter’s skillset. See the USSM post about old-player skills for more.
Pitchers are a different beast entirely. Generally, pitchers velocity (and hitter’s defense) start declining as soon as the player’s body is done maturing, around his or her early 20s. But since pitching skills are so different, there’s isn’t yet a real good sense of aging curves for pitchers as a whole. Since they are the ones in command of the situation, they can take leaps forward (or back) with much greater frequency and effectiveness regardless of age.
Matthew - December 14, 2008
It's my understanding that the projection systems (besides Marcel) look for similar players and aging curves to base projections on
which is why Ichiro is always such a problem. When they’re looking for similar players historically, does this mean looking for guys with similar traditional stat results (1B, 2B, 3B, HR, SB, etc.) or with similar peripherals (K%, K/BB, LD, etc.)?
seattlebruin - December 14, 2008
Depends on the system
PECOTA uses more or less everything and weights them, I think.
Graham MacAree - December 14, 2008
So do they weight traditional stats more heavily then?
I’d imagine these have a much larger sampler size historically than most play by play stats
seattlebruin - December 14, 2008
I have no idea how PECOTA really works
Graham MacAree - December 14, 2008
Is PECOTA then the best of the projection systems? What would make Marcel, Chone, or anything else as good or better?
Dewey N - December 14, 2008
Marcel is never going to be better than a 'real' projection system
It is designed to be dumb, and the benchmark of a reasonable projection system is whether it can beat Marcel.
A comparison of the major systems. From BP and small sample, so take it with a grain or two of salt, but you should get the idea.
Graham MacAree - December 14, 2008
So then why does fangraphs offer Marcel?
Dewey N - December 14, 2008
It's really easy to calculate, and it's open source.
Graham MacAree - December 14, 2008
Because Tango let them
chrisisasavage - December 14, 2008
The real question is why they post Bill James projections
They seem more optimistic than our community projections.
Edgar for Pres - December 14, 2008
They make me feel warm and fuzzy
Aaron Campeau - December 14, 2008
Here is probably where I should register my correlation coefficient complaint
It seems silly to test projection systems for pitchers by doing a correlation coefficient against ERA—we could be perfectly projecting pitchers and have all of the variance buried in year-to-year defensive variation or most of the variation could be due to year-to-year pitching changes. But we don’t know by looking at ERA.
If FIP explains pitching performance best, then projection systems should try to project a pitcher’s FIP. If tRA explaisn pitching performance best, then projection systems should try to project a pitcher’s tRA. Etc.
Also, we miss a lot of pitchers by having to set a lower limit on IP. In 2008, there were 312 pitchers with 50+ IP. There were another 182 pitchers with 20+ IP, and we wouldn’t have to throw those out if we used something like a chi-squared statistic instead of correlation coefficient.
ubelmann - December 14, 2008
It depends on what you're trying to project, right?
I mean, if you’re only interested in a player’s true talent level, then yeah, tRA* (or xFIP—pick your poison regarding which is actually getting the best defense independent measure) is probably the thing to look at.
But if you’re a GM trying to figure out what to do, then you need to account for your own park. GB% is really important for the Rockies and it’s not very important if he’s throwing in PetCo. Similarly, a GM knows exactly which defenders are on the field. The fly ball pitcher is more of a liability with a bad out field than a good one.
There’s no one-size fits all metric. tRA* seems great to me for a measure of pitching talent since it’s based on the run production associated, on average, with particular outcomes that are strictly out of the control of the defense and regressed by park factors and luck issues associated with unusual LD%, for example.
But if you want a projection metric for fantasy baseball, a metric designed to predict ERA is probably more your need.
philosofool - December 15, 2008
Even if you were interested in fantasy baseball...
…it seems like you would project tRA and defense separately and then come up with an ERA projection.
ubelmann - December 15, 2008
Depends on the skill/stat your talking about
Batting average peaks around 24, power hitting around 27, walks in the early 30s. Base stealing probably peaks for most players before they even arrive in the majors. And this all gets messy because there are stats and there are skills. Hitting for contact, pitch selection—those are skills, and they correlate with batting average, but the correlation isn’t perfect and measuring the skills is hard.
Pitching is even harder to understand. Some guys collapse as their velocity starts to drop off. Other guys adjust to the loss of velocity and remain good pitchers. Mussina, for example. Some pitchers learn control, others never do. Someone who knows more about pitching may be able to say more.
philosofool - December 15, 2008
More of a sabermetrics history question than anything
but about when did VORP fall out of favor with analysts?
BrianL - December 14, 2008
VORP for pitchers has always been bad
VORP for hitters is still a reasonable statistic to use, but over the last few years (I’d guess 2006 was when I first started hearing about this, but it may have been a point of discussion beforehand) it’s lost some lustre due to questions about the way they do replacement level and in favour of more accurate offensive stats.
Graham MacAree - December 14, 2008
Yeah, I've always been under the impression VORP was a bad tool to use with pitchers.
I was curious mostly because I got into this scene in early 2007. In all that time I’ve only seen a few mentions of VORP in use with hitters, and I’m not sure if I’ve seen anyone at LL/USSM use it during that stretch.
BrianL - December 14, 2008
Evan at BP a better stat was always Adjusted Runs Prevented (ARP).
But, for some reason they made a point of promoting VORP, and for the longest time they didn’t make ARP numbers available for starters. They wanted you to use Support-Neutral Winning Percentage or something instead. But that meant the only advanced stat they made available with which you could compare starters and relievers was VORP, so a lot of people relied on it.
Llewdor - December 15, 2008
VORP for pitchers is based on run allowed, no?
JI - December 14, 2008
Looks like it uses run average.
BrianL - December 14, 2008
Yes, it's very outcomes based.
Pitchers’ VORP is heavily influenced by luck*.
Llewdor - December 15, 2008
Damn SBN 2.0 for mangling my endnote
Llewdor - December 15, 2008
Does RBOE really prove anything? Why did you choose to add it to wOBA?
Dewey N - December 14, 2008
Speed proxy, I believe.
Aaron Campeau - December 14, 2008
It also will correlate to runs scored better than ignoring it
Graham MacAree - December 14, 2008
I guess which begs the follow-up question
are guys who are really fast disproportionately likely to reach base via error on a groundball than players who are not fast?
seattlebruin - December 14, 2008
I've seen a study done that says yes but I don't remember where it is
Graham MacAree - December 14, 2008
We didn't choose to, that's Tango's formula.
And beside being a speed proxy, it’s rewarding people that reach base.
Matthew - December 14, 2008
But in something like a PrwOBA it wouldn't be used, correct? Because you'd be trying to take the defense out of the formula?
Dewey N - December 14, 2008
*shrug*
We haven’t put the time yet into investigating it. The key aspect to developing projection systems is finding what’s repeatable year-to-year. RBOE might correlate very well year to year as part of a player’s foot speed and GB tendencies. If so, it would be worthwhile to include it, or some facsimile of it.
Matthew - December 14, 2008
Honestly, how much different is it from a hit?
JI - December 14, 2008
Not very.
Graham MacAree - December 14, 2008
Thanks Taylor
JI - December 14, 2008
It's a little bit different.
Graham MacAree - December 14, 2008
A non-response would have implied that I wasn't being stupid.
JI - December 14, 2008
I've asked this several times at USSM and Fangraphs
so at this point, if it’s a stupid question I’d rather just hear that so I can stop asking (although I’d then like to know why it’s stupid).
Is the 0 value in UZR a hypothetical constant (like replacement level hitting) or is it calculated as the average value of all players at a given position in that season (like Dewan’s +/-)
Bearskin Rugburn - December 14, 2008
Average, I am fairly certain.
Matthew - December 14, 2008
So if there were a hypothetical year
when several shortstops shifted to second, Utley had another incredible season and so on guys like Cano would be completely fucked for value. I guess this is where needing 3 years’ worth comes in.
Bearskin Rugburn - December 14, 2008
Well, yes.
Because the group of people Cano would be competing against would have gotten better.
Matthew - December 14, 2008
Past M's
Is there any way to find the defensive values of the 2001 M’s? I can’t find them anywhere
Or Ken Griffey Jr. value as a M? or the value of Buhner’s amazing sliding catches. haha.
ghostsalmon - December 14, 2008
Only on a team basis, and it's a rough measure
inverse BABIP
JI - December 14, 2008
I think
JI - December 14, 2008
Yea
defensive efficiency. I think you can even find IF/OF splits using fangraphs, and I seem to remember someone going through this exercise but haven’t the initiative to go find it.
Bearskin Rugburn - December 14, 2008
BP has park adjusted
Defensive Efficiency which is a little less rough
Trenchtown - December 14, 2008
Moneyball
I’m interested in sabremetrics, but not well versed. I read Moneyball, which describes Bean’s view on defense (i.e. that it doesn’t matter that much). On this site and on ussmariner, the authors talk often about how important defense is.
Question: Would Beane agree? Has he changed his view or did I just mis-read the book?
solap1 - December 14, 2008
He's changed his view.
Matthew - December 14, 2008
Beane
Just goes to show how great GMs will listen to others and even change their views. Not just have one stance and not change no matter what.
ghostsalmon - December 14, 2008
In addition
Defensive Metrics have come along way even since Moneyball, especially in what is publicly available. So Beane felt it was easier to find efficiencies in things that were easily quantified in value, once those undervalued things became properly valued it was time to move on to something else
Trenchtown - December 14, 2008
Beane has always been a "If it matters, measure it" kind of guy.
As a corollary to that, “If you can’t measure it, you can’t know to what extent it matters.”
In 2001, you couldn’t really measure defense. Beane knew enough to realise that the tools he had at his disposal were crap, so he ignored them. The functional result of this was to ignore defensive value completely, because he couldn’t measure it. He had no way to choose between two players based on defensive skills, so he made what is always the rational choice in the face of total uncertainty and deemed them equivalent. In all cases.
Rational decisions are often really counter-intuitive. When Moneyball came out I was only 3 years removed from finishing my academic work in Rational Decision Theory, so it all seemed really obvious to me, but I recognised that most people would disagree with me (and Billy).
Llewdor - December 15, 2008
I think that’s overstating it.
JI - December 15, 2008
Yeah
it mentioned in the book using a form of zone rating to measure Johnny Damon as the best defensive center fielder in the game but it didn’t make up for his particularly awful offensive year. The book glossed over the fact that was considerably below Damon’s talent level and improvement was almost insured
Trenchtown - December 15, 2008
Beane's all about targeting players that are undervalued
Flashy defence was overvalued at the time of Moneyball, which is why he wasn’t willing to pay for it. The market has shifted the other way for a while though.
Graham MacAree - December 14, 2008
You bastard
JI - December 14, 2008
Do we have any evidence that this is actually true?
Or rather, that the “flashy” correlated with “defense” at all?
marc w - December 15, 2008
I think it was that OBP was the great cheap resource
and he was going to use it.
When OBP became mainstream, he shifted his focus more towards run prevention.
JI - December 14, 2008
see under
Cust, Jack; outfielder.
Bearskin Rugburn - December 14, 2008
The A's were pretty beat up.
JI - December 14, 2008
They had options
Bearskin Rugburn - December 14, 2008
Cust was still an above average OF making league minimum
His UZR/150 was -15… which is bad, but still 6 runs better than Ibanez was last year.
+25 for his bat, -15 for his glove, and -7.5 for position still leaves him 2.5 runs above average. Even if that A’s planned on playing him in LF for the whole season, it would still be a net positive, which is what matters in the end.
MrIncognito - December 18, 2008
Another question sort of...
Does anyone have an examples of any fielder having a major shift in their defensive numbers in a good direction or bad direction?
ghostsalmon - December 14, 2008
Over how long a stretch of time?
BrianL - December 14, 2008
Yuni?
Matthew - December 14, 2008
Good direction, short term
Hanley Ramirez. But one year doesn’t mean all that much, and you almost never have a sea change over just one season unless a guy was playing through injury or something
bad direction, long term
everyone
Bearskin Rugburn - December 14, 2008
Jason Bay
seems to have dropped off a cliff as a result of his knee issues. Was an average defender, now terrible. It’s hard to be completely sure because the sample with Pitt this season is small, he changed parks and the park he changed to is totally wonky (Green Monster). Still, he was enough plays below average that expecting him to go all the way back to average isn’t reasonable. (Besides, only about 1/4 of his games were in Boston.)
philosofool - December 15, 2008
It's such an outlier that I would say it's reasonable he comes back up to average.
It’s a really small sample, there are extenuating circumstances and it’s way out of whack with his career numbers. That’s not to say it should be expected, but it’s certainly within the realm of possibility.
Aaron Campeau - December 15, 2008
I was just thinking...
someone either sucking one year then improving the next or vice versa. Also someone like Vizquel was he over/underrated?
ghostsalmon - December 14, 2008
Please use the reply feature.
Vizquel was and remained a pretty good defender for an absurdly long time. He was probably correctly rated.
As for others, Yuni going rapidly downhill is an easy example. As for getting better, you’d have to be looking more at young players or players shifting to a new position. Defense degrades from an early age.
Matthew - December 14, 2008
UZR never loved him
but yeah. As for improvement, Jeter had a good year (read average) defensively according to UZR PMR and Dewan.
Bearskin Rugburn - December 14, 2008
Vizquel could be both
He’s not Ozzie Smith, like the MSM says, but UZR loved late 30s early 40s Vizquel, so I have to imagine his younger self was +10 – +15 annually.
JI - December 14, 2008
Wow
I was going through the 02 Indians’ UZRs on account of Vizquel rated a -8 that year looking for clues. Milton Bradley was a +5 CF. Oh, what could have been.
Bearskin Rugburn - December 14, 2008
Probably an anomaly
JI - December 14, 2008
which Milton or Vizquel?
cause Omar nets -1 for that three year stretch. Could be because overall SS talent was high that year, and that’s an issue I’ve been kinda struggling with for a while (see above). Whatever, if anything it shows the limitations of these metrics and the value of evaluating fielders the way they do for the fielding bible awards.
Bearskin Rugburn - December 14, 2008
Vizquel
Bradley had good speed at one point, it’s not crazy to think that he was between 0 and +5 when he was young.
JI - December 14, 2008
Just curious...
Would Yuni losing weight help his defense or is he just a lost cause at SS?
Fin - December 15, 2008
I think his ceiling is dramatically lower
than it once was, but he certainly looks like he could be a bit lighter on his feet with more conditioning.
Bearskin Rugburn - December 15, 2008
If he lost some weight, he would at the very least gain incremental first step speed, quickness, and ability to get down for balls. His whale flops will be less comical, however.
abender20 - December 15, 2008
Pitchers and wins above average
Can I simply use runs above average (based on pBAA, or FIP*IP/9 or whatever my runs above average metric is) divided by 10.5 to figure the wins above average for a starting pitcher, or does something else need to be taken into account? How many runs/IP is replacement level for starters? For relievers?
What’s the proper way to account for high leverage for relievers?
philosofool - December 15, 2008
Generally, replacement level for pitchers is taken as
150IP at 5.50 tRA/FIP/RA
and the rest of IP at 4.5 tRA/FIP/RA for RP
of course, then you have to credit the SP back some value if he exceeds 150IP for reducing the burden on the bullpen or tax the SP for going under 150IP
Matthew - December 15, 2008
Tangentially related
Pitcher WAR:
Pitcher tRA = A
League tRA = B
(A + B) ^ 0.28 = C
(B/A) ^ C = D
D / (D + 1) = E
E = pitcher Win%
E – .38 = F
F * (xOuts / 27) = pitcher WAR
Jeff Sullivan - December 15, 2008
I hate math.
JI - December 15, 2008
Once you set up a spreadsheet it's really quite simple
Jeff Sullivan - December 15, 2008
I hate setting up the spreadsheet
JI - December 15, 2008
Setting up the spreadsheet is really quite simple
Jeff Sullivan - December 15, 2008
Not if you're stupid like me.
Aaron Campeau - December 15, 2008
We could just add this to SC
Matthew - December 15, 2008
I can't help but feel I am being mocked.
JI - December 15, 2008
Where could I find a derivation of this model
Thanks
danduke - December 15, 2008
`
WAR
Matthew - December 15, 2008
OP should have said "pRAA" not "pBAA"
oops.
philosofool - December 15, 2008
What are the thoughts on "The Book, Playing the Percentages in Baseball"
I am going on vacation nest week and wanted to add it my read list, but if there are better books let me know.
InSpokane - December 15, 2008
It's fantastic.
Matthew - December 15, 2008
There are no better books.
Llewdor - December 15, 2008
And just to clarify. I am going on vacation next week.
InSpokane - December 15, 2008
The Book is very mathy. It's not an easy read, by any means.
If that’s a problem, ask for other suggestions.
Llewdor - December 15, 2008
Agreed. It's not the first book on the subject I would try to read.
I have around as a reference, but I’ve never been able to just sit down and plow through it.
Aaron Campeau - December 15, 2008
The Book
I agree that The Book won’t necessarily be an enjoyable experience as you are reading it, but it should be satisfying in the end. It’s like vegetables. I would recommend The Hardball Times Annual (any of the last three were excellent… the earlier ones are also good) before reading The Book.
tangotiger - December 15, 2008
I think it's completely indispensable.
It just assumes a higher level of understanding of the underlying concepts (as it absolutely must to serve the purpose that it serves) and isn’t really intended to be read for pleasure by most people.
Aaron Campeau - December 15, 2008
I think it all depends on how fluent you are with numbers.
The Book was my first real foray into truly advanced statistics and I really enjoyed it, but I’m a math major.
Jack Moore - December 15, 2008
I am ok with numbers. I was a geology Major, which required a fair amount of math.
InSpokane - December 15, 2008
Yeah, if you can follow numbers relatively well you should be fine
I suggest maybe reading the “toolshed” chapter (that’s the first one) twice, because if you can follow that chapter, the rest of the book will make much more sense, since most of the results in the book are derived from those tools somehow.
Jack Moore - December 15, 2008
If someone more defense-stat savy could help me on my fanpost, it'd be greatly appreciated.
http://www.lookoutlanding.com/2008/12/15/692693/speculative-new-look-at-ro
It’s a bit of a read, but thanks to anyone who can help.
Terminator_X - December 15, 2008
How quickly does a team's defense change?
If I know a team has a good defense this year, what is the chance it will have an average defense next year? How often does the defense stay good? How much does defense bounce around? Is it cyclical? How well does it correlate to team age (which usually is cyclical)?
Edgar for Pres - December 15, 2008
Dude. Check out Tampa 07 and Tampa 08
-54 UZR to +69. That’s… that’s a lot of wins on defense alone.
Bearskin Rugburn - December 15, 2008
Also
Tigers 07 to 08 for the opposite direction. But it’s all correlated with major personnel changes, if you’re asking how much defense can fluctuate with the same players, I think it’s actually quite low, barring injury. The Phillies have had a good defense for a few years thanks to Rollins, Utley, Victorino and company.
Bearskin Rugburn - December 15, 2008
There was also the Rockies' big decline from '07 to '08, mostly due to Tulowitski's injury and Brad Hawpe's general awfulness.
Benne - December 15, 2008
Actually, upon closer inspection, the Rox weren't all that great in '07, either.
-3.4 team UZR, good for 23rd in the league. It’s funny how much the MSM can pound assumptions and predetermined reputations in your head.
Benne - December 15, 2008
FWIW, again, UZR is just one measurement
and one year’s worth of data, yada yada yada
Matthew - December 15, 2008
Of course, no disagreement there.
Can anyone provide me a link to PMR? One of these days I’ll subscribe to Bill James’ site so I can get +/-.
Benne - December 15, 2008
PMR is hosted (kinda hard to find) on baseballmusings.com
Matthew - December 15, 2008
Try
Here
Graham MacAree - December 15, 2008
Thanks.
Benne - December 15, 2008
Team UZR should be pretty accurate though right?
I thought it was just player UZR that was very variable.
Edgar for Pres - December 15, 2008
For an entire team I think it's ok
I mean, it’s like 10 player years, right?
Bearskin Rugburn - December 15, 2008
It's still one measurement's idea of value
and we all know how much defensive metrics agree with each other. For instance, COL’s BABIP that year was just .294. In a field like Coors, that’s pretty good.
Matthew - December 15, 2008
And xRR has them at +15 on the year
Graham MacAree - December 15, 2008
How does xRR compare to team UZR and other team defensive metrics?
Edgar for Pres - December 15, 2008
Very good correlation to THT's team defensive numbers last year, which is all I've looked at.
Graham MacAree - December 15, 2008
Any big outliers?
Not totally sure why it might be interesting but it might for some reason.
Edgar for Pres - December 15, 2008
Not that I recall
Teams were pretty much in the same order.
Graham MacAree - December 15, 2008
How are things like +/- and UZR calculated?
HARRYP09 - December 15, 2008
`
Tango’s summary
Matthew - December 15, 2008
Ok, so its more of a play-by-play observation thing?
HARRYP09 - December 16, 2008
Yes.
A bunch of people watching video to determine where the ball was hit.
Teej - December 16, 2008
Defense question
UZR figures are in runs, while Dewan’s +/- numbers are in “plays.” From what I understand, a play is worth roughly 0.8 runs on average, but is there a difference in the run value of an outfield play compared with an infield play?
Basically, when we’re looking at Dewan’s numbers, is a +15 outfielder saving the same number of runs as a +15 infielder.
Teej - December 15, 2008
"Is there a difference in the run value of an outfield play compared with an infield play?"
Yes
Graham MacAree - December 15, 2008
Generally 0.75 for infielders and 0.85 for outfielders
Jeff Sullivan - December 15, 2008
Thanks.
Teej - December 15, 2008
It seems like there would be a pretty dramatic difference within IF
A 2B/SS play usually would result in a single while a 1B/3B would often result in a double. This might only be worth 0.01runs but I could see a play at 3B being worth more that 0.8 maybe.
Edgar for Pres - December 15, 2008
Does anyone know where you could find +/- numbers
I have only been able to find a few players
Flamefox111 - December 15, 2008
Bill James online has them
but you have to pay $3 a month for the site.
philosofool - December 15, 2008
I've been paying him $3/month for quite a while
and I think I’ve accessed the site 3 times.
Llewdor - December 16, 2008
This seems like a wise expenditure of $36 annually
pdb - December 16, 2008
I'm thinking about it, honestly.
Teej - December 16, 2008
If you had to judge a player's season by one stat...
What would be the best to choose for hitters, and then the best for pitchers?
Imperialism32 - December 15, 2008
And yes, I realize using only one stat is almost always a bad idea
It’s just going to be a lot easier for what I’m planning on doing to use one pretty good stat.
Imperialism32 - December 15, 2008
actually
If you chose from among the several good metrics for hitting and pitching performance, they’re mostly in agreement. With an entire seasons worth of data, wOBA, EqA, and GPA are all quite good. wOBA is theoretically the most sound (as I understand these metrics) but in practice each of them gives very similar results. Same with xFIP, FIP, tRA* and tRA. (FIP and tRA tend to give the pitcher more credit for his circumstances than he deserves though.)
For defense, the very best thing you can do is take the average run value according to all the play-by-play metrics you have available (UZR, PMR, +/-). Defense is much more difficult to quantify and there are several models for doing it. Integrating all these models will probably produce the most reliable results. Using just one for defense is a bad idea.
(People will tell you I’m wrong about this, but I actually think that integrating info from FRAA and RZR is worth doing. If I were to set up a spreadsheet for this, I would take the average run value from the three PBP metrics, weight them .8 and add the average of RZR runs and FRAA weighted .2 as the best measure of a players skill. I won’t argue for that here, I’ll just throw it out and let people jump on me for it if they want to. Sometimes I learn a lot that way. :-))
philosofool - December 15, 2008
wOBA for hitters, tRA* for pitchers
Matthew - December 15, 2008
I prefer Game-Winning-RBIs and Wins myself
After all, the most important thing in baseball is to win, right?
Fett42 - December 15, 2008
I prefer tRA for performance rather than tRA*
Graham MacAree - December 15, 2008
Me too
Jeff Sullivan - December 15, 2008
The downfall with tRA* is that it's strictly a one-year calculation, yes?
oc - December 15, 2008
Meaning?
Matthew - December 15, 2008
Wondering why tRA is a more favorable tool than tRA*
oc - December 15, 2008
Meaning on the one-year?
Matthew - December 15, 2008
oc - December 15, 2008
Amazingly, I am aware of what's published on my own website.
I am asking you to clarify your question.
Matthew - December 15, 2008
It's like this...
What does that mean?
Is that because you would have trouble trying to regress them back to a new league or something?
Graham prefers tRA as an evaluation tool over tRA* and I’m just wondering why. Does tRA* not do something that tRA does?
I don’t mean to sound like a ballsqueeze, I’m just trying to understand the whole ‘league-regression’ thing.
oc - December 15, 2008
The way I understand it, tRA* has more predictive value while tRA is better for performance evaluation.
Aaron Campeau - December 15, 2008
Yes
Matthew - December 15, 2008
It means the inputs are from the current year only.
The regression formulas themselves are based on ~20 years of data.
Matthew - December 15, 2008
Any reason you prefer tRA over tRA*?
oc - December 15, 2008
Height
Bearskin Rugburn - December 15, 2008
Go Sexson!
Bearskin Rugburn - December 15, 2008
Well...
There are really two different things you might want to do. One is evaluate how the exact outcomes a player had were related his teams performance. Another is trying to judge the “true talent” of a player, where you try to isolate skill from luck and assess just skill.
If you’re looking for actual contribution of pitchers where you’re not concerned to isolate skill from luck, tRA. If you’re looking to isolate skill, tRA*.
For batters, luck is a lot less a part of the equation and wOBA is good for both. The biggest thing to look for in batters as a luck indicator is a BABIP (batting average on balls in play, = (H-HR)/(AB – SO – HR): if BABIP significantly deviates from LD% + .12 (by more than .02), the batter has probably been lucky getting hits.
Also, and this is very important, the defensive skills of position players matter a lot. Erase the word “hitter” from your mind. There are position players and pitchers, and position players should be assessed by considering the value of their offensive and defensive performance. Adam Dunn is a great hitter but he is not even a good baseball player because his defense is that bad. More or less the same can be said about Raul Ibanez, Pat Burrel, and even Manny Ramirez. (Ramirez is such a great hitter that he’s a good baseball player despite his defense, but the important point is that in evaluating a player you cannot simply ignore the quality of his fielding skills.)
philosofool - December 15, 2008
Yeah, if Manny was an elite defensive CF, he might be the greatest player of all time
I don’t have wOBA for the ’50s and ’60s on hand, but Manny has a career OPS+ of 155 and Willie Mays has a career OPS+ of 156. Defense is very, very, very, very important.
ubelmann - December 15, 2008
Thanks a lot, everyone
I suppose it might help if I clarified what I was gathering data for: Doing draft analysis stuff, and was looking for the easiest way to rank players’ seasons.
Imperialism32 - December 15, 2008
Could someone tell me
1. Why the translation from runs above replacement to wins above replacement is linear? Why do people always say that 20 RAR = 2 WAR? From an economist’s perspective, shouldn’t there be decreasing marginal returns to run scoring/prevention? For example, the win-loss impact on a team that jumps from 600 to 800 runs in one year is, all else equal, much greater than the win-loss impact on a team that jumps from 800 to 1000 runs in a year.
2. I am always very suspicious of metrics that do not strongly consider volatility involved in player performance. To say a guy is a 2 WAR player is one thing, but it would be better, for example, to have a 95% confidence interval for his performance. Has there been much work done on the variance of player performance from year to year?
danduke - December 15, 2008
This exact question came up
in the Fangraphs Q&A with Tom Tango. I forget exactly his explanation, but it’s some fairly simple number crunching with an enormous sample size. You can count on it. I’ll post the link when I have 15’ to look for it.
Bearskin Rugburn - December 15, 2008
This is basically just contingent
In principle it’s non-linear, but in practice it gets pretty linear. There reason is that the relation from runs score and runs allows to wins is approximately RS2/(RS2+RA^2) and for the particular scoring environment of baseball, those numbers are between 600 and 1000 for almost every team in modern baseball. Non-linearity raises it’s head if you start playing with extreme numbers, but when you’re working within ordinary run creation/prevention enviroments, for all practical purposes it’s linear.
philosofool - December 15, 2008
Minor self-correction
Win% approximately equals RS2/(RS2+RA^2)
philosofool - December 15, 2008
me and the damned typos
RS ^ 2 / (RS ^ 2 + RA ^ 2)
which should sound like "runs scored squared divided by (runs scored squared plus runs allowed squared.)
philosofool - December 15, 2008
Yeah...
…the linear approximation just turns out to work well here, and I can see why someone might be skeptical, but the proof is in the pudding. Linear approximations turn out to be surprisingly good in many places. For instance, strictly speaking, force is not linearly proportional to acceleration, but it was good enough for Newton.
ubelmann - December 15, 2008
And good enough for intrasystem space flight too
Graham MacAree - December 15, 2008
Absolutely
ubelmann - December 15, 2008
Re: point #2
PECOTA projections do this to some degree, though I don’t know how accurate they are. For instance, Jose Vidro’s 10th percentile EQA this year was .218 and his 90th percentile EQA this year was .294. As you can see, though, the variance tends to be rather large, and I’m not sure exactly how useful it would be on an individual player basis. More useful, probably, are the breakout/improve/attrition/collapse rates. For instance, Vidro had a 1% breakout rate and a 38% collapse rate. I would guess that this implies the distribution of likely future outcomes is a rather skewed distribution, so I don’t know that one number can really explain the variance around the most likely outcome very well. With younger players, you’re generally going to have a larger chance of overperformance and in old players you’re generally going to have a larger chance of underperformance.
The Marcel projections have a reliability number, but I don’t know that that really translates to the sort of confidence interval that you seem to be seeking.
If you look at the ZiPS projections (for say, the 2009 Mariners), Dan Szymborski provides some breakdowns into how likely certain outcomes are for a few players. It seems apparent from looking at those that anything encompassing a 95% confidence interval would be rather worthless, approximately equivalent to saying “this player will play major league baseball next year.” Maybe like +/- 15% or +/- 30% around the mean would be interesting, but the 95% confidence intervals are probably too large for them to be useful.
ubelmann - December 15, 2008
PECOTA's ranges are broad
but they still help show you the shape of the curve (though it would be better if they actually showed us the curve).
Llewdor - December 16, 2008
can someone explain to me`
how Adrian Beltre is an “average” 3B in ZiPS, but Russel Branyan is “very good” at 3B?
outtathequestion - December 15, 2008
I was confused over this, too.
— There are just four rankings, and there’s nothing between “average” and “very good,” so it’s odd phrasing to say the least.
— ZiPS projects Branyan to be a slightly better hitter than Beltre on a per-at-bat ratio.
— Even though ZiPS lists range, I’m assuming the “average” and “very good” classifications don’t factor it in.
— Beltre and Branyan appear to both be right on the dividing line between “average” and “very good,” and I guess Branyan was just a touch higher than the bar.
If all those things are true, it makes sense to me. Here’s my previous question and some other people’s comments.
Teej - December 15, 2008
I still have trouble getting down into the nitty gritty of statistics
Enough to where I can confidently come up with my own observations on talent.
I recently purchased The Book, Baseball Between the Numbers, and THT Annual 2009 (also threw in Moneyball and “Cheater’s Guide” by DMZ for good measure). Which book should I read first and will they help me get a better understanding on how I can properly evaluate talent?
ThundaPC - December 15, 2008
Baseball Between the Numbers first, then go with the THT Annual.
Think of the Book like you would a textbook; crucial for understanding the specifics but not to be attempted until you have a solid understanding of the basics. Cheater’s Guide is awesome but it’s a pretty easy read that wont really enhance your ability to evaluate talent. Moneyball is also a great read, but don’t take it too seriously; it was written with a lot more importance placed on the narrative than highlighting the underlying concepts Beane has used over the years.
Aaron Campeau - December 15, 2008
I'd actually go 'Moneyball' first
While not great at giving you the fundamentals of deep analysis it gives you an excellent motive to be open-minded about it.
Graham MacAree - December 15, 2008
The only reason I suggested otherwise
was because I read it and took it as gospel and set myself back a few years. But if I had gone in knowing what I know now about Beane’s philosophy, I probably wouldn’t have had that problem.
Aaron Campeau - December 15, 2008
The overriding process is correct
but many of their conclusions are wrong
JI - December 15, 2008
Right.
I didn’t see it that way when I read the book because of a combination of my own stupidity, the A’s being really good and the way the book was written. Because they’ve obviously changed the way they go about things, I think I would probably get it if I was reading it for the first time.
Aaron Campeau - December 15, 2008
Yeah, I re read some of it recently, and you can kinda laugh at some of their ideas
But, my word, that book is one of the easiest reads there is. Whenever I put it up I can’t put it down.
JI - December 15, 2008
Yes, that would violate the laws of space-time.
Llewdor - December 16, 2008
Cool.
Thanks acblue, Graham.
ThundaPC - December 16, 2008
Wouldn't it be cool if WPA was weighted against wOBA to give an indication of who really is "clutch" in comparison to their regular performance?
Or is that already included?
Dewey N - December 15, 2008
Or is that the clutch stat on fangraphs?
And if so, how does it work?
Dewey N - December 15, 2008
So WPA for an average batter is zero, right?
You expect Albert Pujols to put up a high WPA because, well, he’s ungodly good. If you correlate whatever ‘goodness’ measure you’re using to WPA, you can predict WPA based on OPS/wOBA/whatever. Fangraph’s ‘clutch’ score is just the difference between the expected WPA and the actual. I don’t know what stat they’re using to generate expected WPA though.
Make sense?
Graham MacAree - December 15, 2008
Cool
Dewey N - December 15, 2008
Isn't it just using a win expectancy matrix?
I always assumed that WPA was determined by looking at the win expectancy before a plate appearance and the win expectancy afterward and subtracting the former from the latter to get the change in the probability of a win.
philosofool - December 15, 2008
No more posting without careful reading...
the “expect” in “expected WPA” was important. Yeah, I’d love to know how they get that.
philosofool - December 15, 2008
I believe they use WPA/LI
the clutch stat is (WPA/pLI) – WPA/LI
Jack Moore - December 15, 2008
Thanks for the 'put
PECOTA’s cool, but I’m sure how seriously I should take it.
Another question – have any of you read any of the following books:
The Diamond Appraised – Craig Wright and Tom House
The Book on The Book – Bill Felber
Understanding Sabermetrics – Costa, Huber and Saccoman
Diamond Dollar$ – Vince Gennaro
Curve Ball – Jim Albert and Jay Bennett
Are they worthwhile? I have a solid background in math/stat/econ so those are not barriers to my enjoyment.
danduke - December 15, 2008
Take it seriously
But remember that the very best projection metrics are falible. Probably what’s most useful about these is the shear volume of players that they cover, not the individual projections, which won’t ordinarily blow away the average opinion of say, ten reasonably informed and not too biased baseball fans.
philosofool - December 15, 2008
PECOTA was designed by FiveThirtyEight branchild Nate Silver
If he can call an election with near-deadly accuracy, I have faith in his baseball projection system.
Benne - December 15, 2008
I wouldn't go that far
Baseball ability is a lot harder to measure than public opinion.
philosofool - December 16, 2008
If someone has a better system, they should share with the class
PECOTA is very good.
ubelmann - December 16, 2008
I read Understanding Sabermetrics because the guy who wrote it teaches the sabermetrics course at my school
It’s really not that great compared to the other books mentioned here… very simplistic, the best it gets into is a little bit of info on linear weights.
Fett42 - December 16, 2008
Would pRAA weighted by leverage be more useful for
figuring out a pitcher’s impact. It wouldn’t tell us about their talent but it would tell us how much they contributed or how “valuable” they were. Is this totally wrong?
Edgar for Pres - December 15, 2008
Does anybody have..
the year to year correlations for stats such as BABIP, GB%, etc? I had a pretty handy chart with all that on there at one point but I can’t seem to find it anywhere.
CKel - December 16, 2008
`
From prior article
Matthew - December 16, 2008
Just what I was looking for, thanks.
CKel - December 16, 2008
I keep seeing F.A.T. over at BtB
but i can’t find out what this stands for. can someone please ’splain?
DyeLongJustice - December 16, 2008
Freely available talent
Jeff Sullivan - December 16, 2008
Freely available talent.
Basically guys you can call up from the minors or trade a meaningless minor leaguer for.
Teej - December 16, 2008
FINALLY
thank you so much, its about damn time. i’ve google and btb searched that twice now.
DyeLongJustice - December 16, 2008
Fortnightly applied turnips
Graham MacAree - December 16, 2008
featured abbreviation: the
Matthew - December 16, 2008
Freely Avrabendoubled Talent
Dewey N - December 16, 2008
You must Login with your SB Nation account and be a member of Lookout Landing to post a comment.