Winexp 5: RoY, Cy Young & MVP

by Apricot on Sep 17, 2005 1:37 AM PDT

As you may recall, I last sent Baby Winexp (v46) to her room for telling me Rincon was doing well and that Blanton wasn't as good as I thought he was. Well, she used her time wisely and had some updates for me.

(What are these numbers? You can look at background here. They were computed using some clever routines I wrote to convert game play-by-play information into information about changing game win probabilities. The software is buggy, but most of the truly embarrassing bugs are gone. These measure offense only, not defense.)

Rookie of the Year

Here is a ranking of some frontrunners for RoY as of today. I think Baby Winexp won't get an argument from us that Street is the RoY by a mile.

Team Name WXC Events
OAK Huston Street* 3.532 298
CHW Tadahito Iguchi 2.095 540
OAK Dan Johnson 1.836 383
TOR Gustavo Chacin* 1.282 786
OAK Joe Blanton* 1.209 765
OAK Nick Swisher 0.298 474
NYY Robinson Cano -0.532 475

The only complaint might be Cupcakes's low ranking. Fair enough, if you want to forget May (what's that?) and start counting from June, the rankings now go:

OAK Huston Street* 3.026 186
OAK Dan Johnson 2.276 361
OAK Joe Blanton* 2.181 539
OAK Nick Swisher 1.073 360
CHW Tadahito Iguchi 0.728 337
TOR Gustavo Chacin* 0.568 528
NYY Robinson Cano -0.054 393

MVP

BOS     David Ortiz     7.628    642
NYY     Alex Rodriguez    6.158    664
HOU     Roger Clemens*    6.065    777
ATL     Chipper Jones     5.888    387
STL     Chr Carpenter*    5.883    883
FLA     Dontre Willis*    5.846    839
FLA     Carlos Delgado    5.325    560
ARI     Tony Clark      4.867    343
ANA     Vladi Guerrero    4.52    556
CHC     Carl Zambrano*    4.459    834

On sheer clutch offense, Ortiz wins easily in the AL. In the NL, it's too close to call between three great pitchers and Chipper Jones(!).

For perspective, here are the A's best WXC players to date:

OAK     Huston Street*    3.532    298
OAK     Rich Harden*      2.669    499
OAK     Eric Chavez     1.868    644
OAK     Dan Johnson     1.836    383
OAK     J Duchscherer*    1.663    326
OAK     Joe Blanton*      1.209    765
OAK     Bobby Kielty      1.105    420
OAK     Barry Zito*     1.070    888
OAK     Mark Kotsay     0.970    583
OAK     Adam Melhuse      0.800    94
OAK     Jay Payton      0.590    239
OAK     Mark Ellis      0.573    422
OAK     Eric Byrnes     0.551    220
OAK     Kiko Calero*      0.440    226
OAK     Nick Swisher      0.298    474
OAK     Kirk Saarloos*    0.271    619

Huston ranks 26th in the majors.

AL Cy Young

I hesitated to post this section, because who would believe someone who claimed that :d was the best active pitcher in the AL?

TOR     Roy Halladay*     4.323    567
OAK     Huston Street*    3.532    298
ANA     Fra Rodriguez*    3.164    249
CHW     Cliff Politte*    3.123    242
ANA     Bartolo Colon*    2.911    838
CHW     Mark Buehrle*     2.812    891
MIN     Johan Santana*    2.748    806
OAK     Rich Harden*      2.669    499

I'll just leave it at that.

Nostalgia

Also for the nostalgia of it all:

NYY     Jason Giambi      3.663    487
STL     Mark Mulder*      3.155    812
ATL     Tim Hudson*     2.14    738
CLE     Arthur Rhodes*    0.73    177
BOS     Chad Bradford*    0.108    88
BAL     Miguel Tejada     -0.358    633
TOR     Ted Lilly*      -1.702    512
PIT     Mark Redman*      -1.049    754
BOS     Keith Foulke*     -2.184    214

COL     Eric Byrnes     0.236    62
OAK     Eric Byrnes     0.551    220
BAL     Eric Byrnes     -0.841

2004

There's so much I'd like to say about 2004, but I'm tired. I can't resist listing the top WXC in baseball in 2004:

SF      Barry Bonds     13.372    614
STL     Jim Edmonds     5.917    616
STL     Albert Pujols     5.908    680
COL     Todd Helton     5.806    667
LA      Adrian Beltre     5.726    672
TEX     Mark Teixeira     5.429    614
HOU     Lance Berkman     5.162    690
CIN     Adam Dunn     5.108    679
COL     Jeromy Burnitz    5.079    603
MIN     Johan Santana*    5.011    888
ANA     Vladi Guerrero    4.856    695

That Bonds dude was a cartoon.

Tweet Comment 21 comments | 0 recs |

Do you like this story?

Comments

first thing that sticks out

tejada -.36? wow.

A's v Giants "is kind of like the difference between going to see the Ramones and going to see the Bee Gees. A's fans will go see the Ramones." -BB 07/27/05

by xbhaskarx on Sep 17, 2005 2:40 AM PDT reply actions

Tejada

yeah, I'd like to understand how that's possible. Granted he's really come back to earth in the last six weeks. But this doesn't smell right. He has a .952 OPS w/RISP, but a negative winexp contribution?

Perhaps you've discovered some hidden reason that Billy Beane was really a genius for letting Miggy go. But right now I'd put my money on a bug - meaning no offense.

With Eric Chavez batting, Jason Kendall scores on defensive indifference.

by matthias on Sep 17, 2005 3:41 AM PDT up reply actions

and no pun intended

on "no offense"

With Eric Chavez batting, Jason Kendall scores on defensive indifference.

by matthias on Sep 17, 2005 3:42 AM PDT up reply actions

I find it bizarre too.

I'll look into it. The last Baby Winexp result i found bizarre (Crosby = teh suck) was completely validated by looking at performance in close games (3 runs or less margin) with runners in scoring position. I'll check that out for Migs.

It also hurts Miggy that the O's have lost a lot of games...

by Apricot on Sep 17, 2005 9:29 AM PDT up reply actions

a piece of the puzzle

That's a running total of WXC.

No time to check out more right now... maybe someone else interested can see if Miggy did indeed crater since the end of July.

by Apricot on Sep 17, 2005 9:53 AM PDT up reply actions

Miggy since July

he definitely did drop off big time starting in August. He had a .724 OPS in August and has a .692 for September. He had two homers on Aug 23, but has had only one since then (!). It's after Aug 23 that your graph really drops him in a big way.

With Eric Chavez batting, Jason Kendall scores on defensive indifference.

by matthias on Sep 17, 2005 5:24 PM PDT up reply actions

like i've been saying

IT's a ridiculously flawed metric. remember crosby as the worst A next to Ryan Glynn? And check out Ellis' ranking.

And 3 relievers in the top 4 of all AL pitchers, including KRod????

This thing's a curiosity-- with little or no meaning.

oaktoon

by oaktoon on Sep 17, 2005 3:31 PM PDT up reply actions

Kielty above Zito?

Melhuse is high on the list too. I assume playing time is not taken into account.

Joe Blanton is phat

by gojohn10 on Sep 17, 2005 2:49 AM PDT reply actions

what's with the "baby" stuff

by Reg on Sep 17, 2005 8:01 AM PDT reply actions

just seemed right

It didn't seem right to call it just Win Expectancy, because there are different ways of calculating it and,

My program is young and pukes on me.
It is pretty kickass for such a recent arrival (began Labor Day weekend). It's growing fast, now that tracked down play-by-play for 2004 and automated the downloading of PxP for 2005. I have also discovered the unbelievable Retrosheet archives (not readable yet).
It doesn't know that much yet, but I learn from it every time we play.
I wouldn't want it operating heavy machinery yet.
I think we've reached the stage where 95% of the processing bugs are due to actual mistakes in play-by-play from sportsline.com. And I'm amazed how many mistakes there are. Thus, it's not Infant Winexp, but a real Baby Winexp now.
I've programmed half of it with an actual baby on my lap.

I figure when it gets more mature I'll start calling it Teen Winexp.

by Apricot on Sep 17, 2005 10:44 AM PDT up reply actions

Better if you called it

DOA Win Expectancy-- because that's what its value is.

oaktoon

by oaktoon on Sep 17, 2005 3:32 PM PDT up reply actions

I like it, Joe.

DOA Winexp. It's got a ring to it. Like a bathtub.

You'll forgive me if I refer to you as "Joe" from now on, right?

Unlike you, I find analysis and statistics interesting when they challenge my assumptions and give me new insight, not for how much they confirm my pre-determined ideas.

by Apricot on Sep 17, 2005 3:53 PM PDT up reply actions

You might be right,

oaktoon - but if this is worthless I'd like to understand why it's worthless, since intuitively it seems initially like it should make some good sense. I find it interesting. If you don't, there are a lot of other good threads around here to comment on.

With Eric Chavez batting, Jason Kendall scores on defensive indifference.

by matthias on Sep 17, 2005 5:16 PM PDT up reply actions

i've already punctured it enough

No ability to measure defense;
arbitrary/rearview mirror assessment of what acts help win games;
no appreciation for context-- i.e, did a starter who lasted 7 innings and thus saved the bullpen in a losing game matter anywhere near as much as a closer who comes on for 1 inning in a 2 run game?

The thing is bogus. OK??

I'll stop commenting on it.

oaktoon

by oaktoon on Sep 17, 2005 7:57 PM PDT up reply actions

we've hashed this out in past threads

I still find each of your objections irrelevant or unconvincing and do not display any understanding of how win expectancy is calculated. I have a host of objections to win expectancy, but none of them are the ones you list.

We'll have to agree to disagree.

by Apricot on Sep 18, 2005 2:01 AM PDT up reply actions

Sample size for the situations?

Now for a serious question: do all the situations that are used to calculate winexp values have reasonably large sample sizes?

I'm wondering about the following hypothetical. Player X comes up in the bottom the 13th with his team down three runs, the bases loaded and two out. In the winexp database there's only one instance of that exact situation ever happening before, and it just happens that in that one instance the player at bat hit a grand slam. So the winexp here is 1.0. But our Player X now strikes out. He gets dinged heavily for bringing the winexp down from 1.0 to 0.0.

So have you checked, what is the smallest sample size ever used?

With Eric Chavez batting, Jason Kendall scores on defensive indifference.

by matthias on Sep 18, 2005 8:57 AM PDT reply actions

excellent questions

There are a lot of situations with small sample sizes. You have to decide at which point to cut it off.

I wrote in a couple of thresholds: one for which you asterisk the WX because of small sample size (set right now at 100 games) and one for which you refuse to use the data at all (set right now at 30 games). Most of the SSS data happens now with huge leads, like 6+ run difference, so these are in situations where the WX is very close to 0 or 1. I don't have an a priori way of figuring out these thresholds... it would be interesting to come up with one.

Extra innings from historical data are currently calculated from 9th inning data.

by Apricot on Sep 18, 2005 9:48 AM PDT up reply actions

the upshot

(I forgot to finish) is that I currently chuck small sample sizes, which means that play gets no WX assigned and the following play gets no WX change assigned. One can (and should) argue about what constitutes "small". In 99% of these cases, it's in a blowout where very little WXC is being assigned.

by Apricot on Sep 18, 2005 10:08 AM PDT up reply actions

in blowouts...

they really don't matter, it almost flatlines.

see: http://www.athleticsnation.com/story/2005/8/18/17733/5847 for more details on that.

the sample size in the 8th and 9th usually leads to it be way off from the algorithms; i tend to believe it's somewhere in themiddle of the two, but i really don't know...

Hell is sold out... but some people are re-selling their tickets on Stub-Hub.com, or you can just sit on Mt. Davis. - McFood

by Jjjsixsix on Sep 18, 2005 10:49 AM PDT up reply actions

Perhaps...

you could consolidate certain situations to get larger sample sizes. For one, consider all extra innings identical, as for all practical purposes they are. For another, treat all very large leads as if they were the same - e.g. any lead more than seven runs is treated as a seven run lead. Would that help?

With Eric Chavez batting, Jason Kendall scores on defensive indifference.

by matthias on Sep 18, 2005 7:38 PM PDT up reply actions

interesting idea

I'm not sure you gain a lot because all such situations are basically indistinguishable from probability 1 or 0. All the situations where the outcome is in doubt are very well represented in the data.

by Apricot on Sep 18, 2005 9:03 PM PDT up reply actions