SABR FAQ - First Draft

[rluzinski]

Well, we got the separate "Statistical Analysis" thread. Hopefully we can develop it into a great resource for not only general SABR information, but hopefully a nice source of Brewer related statistical analysis as well.

 

There's some great links here already; Let's add some more. Eventually we'll clean this thread up and have everything nice and organized.

[pogokat]

sounds great, get crackin'

[brettac1]

I think that is a great idea. If someone doesn't have a handle on what a stat is trying to describe, they could post a question in that forum rather than wasting a bunch of posts in a regular thread in the MLB forum.

 

EDIT: Also, it would be a good place to maybe post articles/links/new thoughts that those of us who are stat-leaning individuals, where they would probably be more easily discussed, i.e. it wouldn't fall off the 1st page of topics so quickly.

[EdgarDiazRocks]

Oh and please inform us all on who exactly the mysterious "Replacement Player" is . . . .http://forum.brewerfan.net/images/smilies/wink.gif

 

And why PECOTA's formula is a secret akin to Nuclear Launch Codes . . .

[tomwopat]

Baseball-Reference.com's "Bullpen" Wiki and Wiki Gonzalez, the Baseball Think Factory encyclopedia are good SABR-ish resources for this kind of information.

 

But I'd be all for having a BF.net authored wiki of our own.

 

edit- fix formatting

[rluzinski]

Oh and please inform us all on who exactly the mysterious "Replacement Player" is

 

Why he's 80% of an average guy of course... so like 5-7 and 120 lbs.

 

Eckstein!

 

Seriously, a seperate forum sureeee would be nice. Atleast we wouldn't waste everyone's time posting the same thing over and over again. We could just link to a thread.

 

It's nice to have Brewer-centric SABR posts. Like showing Lee's RBI's are mostly a function of opportunity, for instance. It would be nice having one pinned thread with just links to SABR-centric websites, like the one's above.

[brettac1]

From Keith Woolner's old site: www.stathead.com/articles...r/vorp.htm

 

Based on research I conducted and published in Baseball Prospectus 2002, replacement level is set at 80% of the positional average rate of offense for most positions (85% is used for catchers, 75% is used for 1B/DH). Replacement level for pitchers is figured separately for starting and relieving, as detailed in BP2002:

 

Starting pitcher replacement level = 1.37 * League RA - 0.66

 

Relief pitcher replacement level = 1.70 * League RA - 2.27

 

...

 

Replacement-level delta is set at 70 points of OPS below league positional average (based on previous studies I've done, but have not published -- I may finally get around to it this year). The same delta is used for all positions (this is a possible area for future research).

[EdgarDiazRocks]

Quote:

---

Based on research I conducted and published in Baseball Prospectus 2002, replacement level is set at 80% of the positional average rate of offense for most positions (85% is used for catchers, 75% is used for 1B/DH). Replacement level for pitchers is figured separately for starting and relieving, as detailed in BP2002:

 

Starting pitcher replacement level = 1.37 * League RA - 0.66

 

Relief pitcher replacement level = 1.70 * League RA - 2.27

 

...

 

Replacement-level delta is set at 70 points of OPS below league positional average (based on previous studies I've done, but have not published -- I may finally get around to it this year). The same delta is used for all positions (this is a possible area for future research).

---

 

 

Zzzzzz.. . . . .Z....Zz.......Z..z..z....zzz . . .*snork*

 

Uh . .huh? Did you say something? I drifted off there.

 

Oh yeah Corey Sullivan sucks because he's hitting 5.3(%?) worse than a nebulous mathmatical formula that changes every time a CF comes to the plate.

 

Yep, makes perfect sense. So sorry

 

Let me suggest another book

 

Zero: The Biography of a Dangerous Idea

 

As in Zero point.

[brettac1]

Oh yeah Corey Sullivan sucks because he's hitting 5.3(%?) worse than a nebulous mathmatical formula that changes every time a CF comes to the plate.

 

Sullivan's VORP is -5.7, actually.http://forum.brewerfan.net/images/smilies/wink.gif But it's not % points, it's Runs Below the mythological Replacement Player. It's cumulative, but an easier way to show his VORPr (his VORP/gm.) extrapolated over 162 games, as I am want to do. Sullivan's VORPr is -.092, which means he costs his team a full run (in comparision to the RP) per 10 games or so, or about 15 runs a season, using the rule of 10 - 1.5 wins. You can wake up now.http://forum.brewerfan.net/images/smilies/wink.gif

[EdgarDiazRocks]

Runs Below the mythological Replacement Player. It's cumulative, but an easier way to show his VORPr (his VORP/gm.) extrapolated over 162 games, as I am want to do. Sullivan's VORPr is -.092, which means he costs his team a full run

 

Well of course it's runs.

 

Because not only do we get to make up "Player X", but then we get to "translate" his made up batting stats into made up runs scored and in the case of WARP include made up fielding and baserunning stats.

 

And then all this is throwm into one big hopper and out spits a one size fits all number that anyone can use.

 

-5.7.

 

That's much worse than 28.8. Fire Him!

 

These "one size fits all" numbers like Win Shares, Warp3 and VORP do really drive me nuts.

 

It's like going shopping with the wife for a pair of pants.

 

I go to the men's section, collect the right series of data for my body and buy a pair of pants. The inches around my waist the inches of inseam

 

She goes to the wall picks out the arbitary number somebody assigned despite any sort of acknowledgment that girls might have different body shapes.

 

5'2" with wider hips? thats a "12(?)" 6'4" with thin hips? Also a "12" . . .

 

It's why I can get a pair of pants in 10 minutes & a then have to wait an hour as she tries on 60 pairs to "find one that fits"

 

It's the same with these numbers . . .

 

In 1987 Tony Gwynn had a WARP3 of 10.8 and Eric Davis 10.3.

 

Wow how exciting. They both were 10+ runs better than "Player X!"

 

Despie the facts that they had totaly different seasons and are totally different players.

 

Nothing is better than Tony Gwynn raw numbers to wallow around in.

 

Thanks to Retrosheet I see that . . .

 

In '87 Gwynn hit .370 /.447/.511 wit a bb/k ratio of 82/35(!!) in 157 games

 

Didn't really care which hand you threw with

 

RHP .376 /.457/.541

LHP .361/.433 /.470

 

and despite hitting only 7 HR and having nobody but John Kruk behind him, scored 119 times.

 

Davis on the other hand hit .293/.399/.593 with 37(!!) HR and a bb/k rate of 84/134 in a injury shortened (read average Eric Davis) season of 129 games.

 

Though RHP could handle Davis a lot better . . .

 

RHP .272/.380/.526

LHP .340/.440/.741

 

And thanks to having hitters like Buddy Bell, Kal Daniels and Dave Parker behind him Davis scored 120 times! One more than Gwynn!

 

I often hear how RBI is a useless stat, but in my opinon "Runs Scored" is overvalued by SABER formulas.

 

removing the times that they scored themselves (HR), Gwinn reached bases a whopping 109 times more than Davis, yet had his teammates leave him out to dry on the bases.

 

So Davis' WARP3 rewards him for having teammates who can drive him in.

 

At least more than "Player X" teammates can.

 

Also this just in, there is a rumor that "Player X" is secretly Scott Podsednik's older brother Rex.

http://absoluteanime.com/speed_racer/_rex.jpg

[rluzinski]

So yeh, anyway... Brian, can we get a "Statistical Analysis" forum please? It might cut down on some of these little bickersand could prove to be a valuable resource for Brewer-centric statistical analysis.

[brettac1]

EDR, just so you know (I'm pretty sure you do), I'm in agreement with you on VORP, as I have said a bunch of times before. I once in a while do like to use VORPr, though.

[batman]

I don't necessarily disagree with you, EDR, but consider that the distinction between those two players in WARP3 is due to defense, not offense. Davis' historically-adjusted equivalent average was .330 compared with Gwynn's .342, so the BP metrics definitely recognize the superiority of Gwynn at the plate. On defense, Gwynn gets credited as being four runs above average as a right-fielder while Davis is 13 runs above average as a center-fielder. That's why their WARP scores are similar. How much you decide to trust BP's defensive stats is up to you.

 

~Bill

[pogokat]

rluz...i can't give you a forum, but i can pin a thread to the top, i think...

[rluzinski]

batman supplied a link to an explanation for EQA...

[1992casey]

I like the idea of original writing by Brewerfan.net paritcipants, with links to bibliographic material. I think the way to handle this is to make this thread a rough draft for something that could be used as a more polished FAQ. I see a two pronged approach. One would be to list links to potentially useful articles. The other would be ask questions we'd like to see answered.

 

I'll get it going with some links:

 

General

• Bill James Primer
• Tango on Baseball
• Cyril Morong's Sabermetric Research
• Sabernomics Blog
• Dan Agonistes Blog

Baseball Players and Age

• The Effects of Age on Hitting
• Is Sosa Finished?
• Peak ages for hitters
• Age and Established Win Shares
• The Science of Forecasting
• ESPN.com: MLB - Cleveland's contracts could backfire (Dec., 2001)
• Age and Hitting I
• Age and Hitting II
• Age and Hitting III
• What is a Normal Career Trajectory?

Productive Outs

• Yet Another Productive Outs Article
• Productive Outs Are Not Productive
• Productive Outs I
• Productive Outs II
• The Truth About Productive Outs
• Productive Outs (Dan Agonistes Blog)

The Effect of Ks

• Neyer - Swing and a miss: Not Milwaukee's biggest problem
• Neyer - Why pitchers desperately need K's
• Do Strikeouts Matter?
• Just Another Out?

Stat Descriptions

• Introduction to VORP: Value Over Replacement Player
• Ultimate Zone Rating (UZR), Part 1
• Ultimate Zone Rating (UZR), Part 2
• WHIP it good ... statistically (covers 8 different stats)
• And the Winning Number is... OPS

Misc.

• Stats vs. Scouts: The Great Debate
• Resisting the feud (critiques above article)
• 13 Reasons Why Pro Baseball Can't Teach Hitting by Rob Ellis

[1992casey]

One category missing from the above is clutch hitting. I can find the stuff, but I'm now too tired. Maybe another time unless someone beats me to it.

 

This has always bugged me. When we evaluate players, we always talk about 10 runs being equal to a win. I've seen it so much I accept it as fact. But I'd love to see a couple links to more information about it.

[bjamesw22]

That is awesome that you took the time to put all of those in. Good work.

[pogokat]

www.baseballgraphs.com/

 

Good stuff on pythagorean's here...

[rluzinski]

hardballtimes.com

[Frisbee Slider]

I have seen this posted on bf.net before, but I am interested in seeing a table that lists "expected runs" for different situations during a baseball game (ex. runners on 1st and 2nd with 1 out).

 

Could someone post a link on where I could find a table with that information?

[brettac1]

Right here:

 

www.tangotiger.net/RE9902.html

 

There may be a newer one, but this is usually the one I reference when I mention Run Expectancy.

[rluzinski]

Run Expectancy - 1999-02

 --------OUTS--------- 0 1 2 Empty 0.555 0.297 0.117 1st 0.953 0.573 0.251 2nd 1.189 0.725 0.344 3rd 1.482 0.983 0.387 1st&2nd 1.573 0.971 0.466 1st&3rd 1.904 1.243 0.538 2nd&3rd 2.052 1.467 0.634 Loaded 2.417 1.65 0.815 

 

LINK

 

Run Frequency - 1999-02

 

The following table presents the rate of the number of runs that scored, from that base/out state, to the end of that inning. All data is from 1999-2002.

 Base Outs ----------------Runs------------------- 0 1 2 3 4 5+ Empty 0 0.707 0.154 0.074 0.035 0.016 0.013 Empty 1 0.827 0.101 0.042 0.017 0.007 0.005 Empty 2 0.923 0.051 0.017 0.005 0.002 0.001 1st 0 0.563 0.176 0.132 0.067 0.034 0.028 1st 1 0.717 0.123 0.091 0.040 0.017 0.013 1st 2 0.864 0.062 0.049 0.016 0.006 0.003 2nd 0 0.368 0.348 0.142 0.076 0.035 0.030 2nd 1 0.594 0.230 0.098 0.045 0.018 0.014 2nd 2 0.777 0.147 0.049 0.017 0.006 0.003 3rd 0 0.136 0.542 0.164 0.090 0.035 0.033 3rd 1 0.338 0.478 0.106 0.045 0.018 0.014 3rd 2 0.737 0.187 0.050 0.017 0.006 0.004 1st&2nd 0 0.359 0.219 0.165 0.127 0.070 0.059 1st&2nd 1 0.574 0.161 0.110 0.088 0.038 0.028 1st&2nd 2 0.769 0.106 0.058 0.044 0.015 0.008 1st&3rd 0 0.124 0.417 0.174 0.142 0.076 0.067 1st&3rd 1 0.345 0.370 0.119 0.092 0.042 0.031 1st_3rd 2 0.715 0.151 0.061 0.049 0.016 0.008 2nd&3rd 0 0.144 0.249 0.307 0.147 0.079 0.074 2nd&3rd 1 0.305 0.285 0.218 0.101 0.053 0.038 2nd&3rd 2 0.724 0.054 0.141 0.049 0.021 0.011 Loaded 0 0.128 0.255 0.211 0.143 0.134 0.130 Loaded 1 0.330 0.252 0.151 0.106 0.093 0.068 Loaded 2 0.675 0.092 0.105 0.055 0.048 0.025 

 

LINK

 

I use the second table alot. For instance, it tells me:

 

* Lead off double scores 63% of the time

* Runner at 3B and 1 out scores 66% of the time

[Frisbee Slider]

Thanks. Exactly what I was looking for.

 

Out of curiousity only...why are the only tables available from 1999-2002? Of course, to create a table like this takes a lot of resources, but I would think there would be different tables with larger/different samples available, too.

 

Thanks again.

[Jimbo]

This has always bugged me. When we evaluate players, we always talk about 10 runs being equal to a win. I've seen it so much I accept it as fact. But I'd love to see a couple links to more information about it.

 

Two ways to come up with the magic number of 10 additional runs = 1 additional win.

 

Use linear regression. Let wins be your dependent variable while run differential is the independent variable. Use data from any year.

 

Or for those not statistically inclined, take out a sheet of ordinary graph paper. Let the Y-axis be wins, the X-axis run differential. Take the 14 or 16 teams from any league or any year and plot. Draw a straight line that closest to connecting most of the dots. That line should have a slop of about "10".