THE BOOK--Playing The Percentages In Baseball

Awesome. Just plain awesome.

Let the comparisons begin!

For all the accolades that Bonds piled up for is run between ‘01-’04, Yaz’s 1967 year is better than any of them.

This site is amazing. Now if only there was a sort function…

Someone sent me the list below, and I think it’s just a matter of time until Rally gives us a comprehensive similar list.  This is a semi-random list of players.

Brooks Robinson comes out as a HOF.  Looks like the “guaranteed” HOF is at 75 WAR.  At 55-74 WAR, you are in the “Gray area” and the closer you are to 75, the more likely you are to make the HOF.  Below 55, and you have to have a special set of circumstances to make it.

Willie Bloomquist is at a career almost +1 WAR in almost 2 full seasons. 

WAR Name Notes
175 Bonds, Barry
142 Mays, Willie 3 seasons missing
141 Aaron, Hank 1 season missing
115 Henderson, Rickey
108 Schmidt, Mike
106 Robinson, Frank
104 Morgan, Joe
97 Rodriguez, Alex active
97 Yastrzemski, Carl
93 Boggs, Wade
92 Ripken, Cal
90 Brett, George
84 Rose, Pete
83 Griffey, Ken Jr active
82 Carew, Rod
79 Thomas, Frank active
79 Molitor, Paul
76 Yount, Robin
75 Jackson, Reggie
74 Robinson, Brooks
72 Whitaker, Lou
70 Murray, Eddie
70 Larkin, Barry
70 Smith, Ozzie
70 Bench, Johnny
70 Ramirez, Manny active
70 Walker, Larry
70 Raines, Tim
69 Grich, Bobby
69 Martinez, Edgar
68 Fisk, Carlton
68 Trammell, Alan
68 Rodriguez, Ivan active
67 Pujols, Albert active
67 Biggio, Craig
67 Santo, Ron
67 Edmonds, Jim active
66 Lofton, Kenny
66 Smith, Reggie
66 McCovey, Willie
66 Carter, Gary
65 Nettles, Graig
64 McGwire, Mark
64 Hernandez, Keith
64 Sheffield, Gary
64 Randolph, Willie
63 Evans, Dwight
63 Bell, Buddy
62 Williams, Billy
62 Sandberg, Ryne
61 Winfield, Dave
60 Davis, Willie
60 Allen, Dick
60 Bando, Sal
59 Guerrero, Vladimir active
59 Evans, Darrell
58 Piazza, Mike
58 Bonds, Bobby
57 Boyer, Ken
57 Stargell, Willie
56 Wynn, Jimmy
54 Torre, Joe
54 Cash, Norm
53 Cruz, Jose
53 Clark, Jack
52 Simmons, Ted
52 Aparicio, Luis
51 Cedeno, Cesar
50 Perez, Tony
50 Downing, Brian
49 Lynn, Fred
48 Staub, Rusty
48 Butler, Brett
47 Burks, Ellis
47 Suzuki, Ichiro active
47 Colavito, Rocky
46 White, Roy
44 Fregosi, Jim
44 Murphy, Dale
43 Strawberry, Darryl
43 Rice, Jim
43 Fernandez, Tony
43 Mattingly, Don
43 Freehan, Bill
43 Munson, Thurman
42 Singleton, Ken
42 Brock, Lou
42 Maris, Roger
41 Otis, Amos
40 Parker, Dave
40 Belle, Albert
39 Wills, Maury
38 Salmon, Tim
37 Sanders, Reggie
35 Concepcion, Dave
33 Guerrero, Pedro
29 McGee, Willie
27 Mazeroski, Bill
26 White, Frank
17 Carter, Joe
-4 Mendoza, Mario

Rally posts his leaderboard here:
http://www.baseballthinkfactory.org/files/newsstand/discussion/wins_above_replacement_1955_2008/#3090943

And two posts later, OCF tells us how Hall Of Merit voted:
http://www.baseballthinkfactory.org/files/newsstand/discussion/wins_above_replacement_1955_2008/#3091005

Indeed, one could even argue that the Hall of Merit has NO value added to what Rally does all by himself… insofar as this covers only the Retrosheet players.

At random I looked up Dante Bichette, to discover he was only 2.8 Wins Above Replacement for his entire career!

In 1999, when he was .298/.354/.541 (not park adjusted), with 34 HR and 133 RBI (I know, meaningless), he was actually nearly 3 wins BELOW replacement level, because of bad range and poor arm (although the strength of his arm was certainly high). 

Wow.  Colorado would have won 3 more games by putting a Quad A player out there for Bichette?  Dante made $7.25 million that year.

I’m not sure how much I trust that -30 in defense for Dante that year.  It is out of place with the rest of his career, though he wasn’t ever much of a fielder.  Even giving him a zero only brings him to about replacement level as an average hitter playing left field and costing his team runs on the bases.

It’s strange that some of the flukiest fielding numbers come from the dataset that should have the most detailed data - the 1990’s when project scoresheet zones are included.  In some cases the coding of fly vs line seem to be questionable at best.

Another example is Tony Gwynn at -28 runs for the 1989 Padres and +30 two years later.  It’s a good thing that those two years cancel out, and otherwise Gwynn’s career ratings appear reasonable - good ratings in the 80’s when he was young and fast, not so good in the mid 90’s when he got fat.

"In some cases the coding of fly vs line seem to be questionable at best. “

Right, I question those ALOT.  I do WOWY five ways:
1. control for identity of pitcher and hand of batter
2. control for park, plus the two hands
3. control for identity of batter plus the two hands
4. control for the base/out plus the two hands
5. control for GB, FB, LD, Pop, plus the two hands

And I trust the results (without testing for it), in that order.

For the batting portion, I believe Rally uses Baseruns to generate custom linear weights for each team. I do think that using single-season linear weights generated by Baseruns is better than using single-season empirical linear weights. Rally also chooses to generate linear weights for each team. So he doesn’t have to worry about making park adjustments. Finally, I believe that he removes pitcher hitting before inputting the statistics into Baseruns to generate the linear weights.

Everything you say is correct, except for the park adjustments: I still use them.  Baseruns will get the total of all the stats to add up to, say 960 runs in a Coors field season.  But then I need to adjust that 960 by the park factor.

Perhaps I could just adjust the out value and leave runs as is - but that would also require a park factor.

I could be wrong here, but I thought that if you used custom linear weights for each team, then you would not have to make park adjustments. However, the disadvantage to using custom linear weights for each team is that you now have to use a custom runs to wins figure for each team. Hopefully someone else could comment on whether it is necessary to use a park factor with team-specific linear weights.

Terps: see if this gives you a different perspective:

http://www.insidethebook.com/ee/index.php/site/comments/quick_park_factors/

Terps, I agree that custom LWTS for each team + custom R/W converter eliminates the need for park adjustment.  But I think that for Rally’s presentation, it’s helpful to do the park adjustment, as it allows you to compare the batting runs across players. 

In the end, as long as you are consistent in your process, custom LWTS + custom RPW converter should give you just about the same wins number as standard LWTS + park factor.  I think it’s really just a matter of whether you want the intermediate runs above baseline results to be comparable across parks, or if you don’t care about that and are only interested in the final wins above baseline result.

I will disagree with Patriot, on the expectation that we must be talking about two different things.

The reason that it doesn’t eliminate the need for the park adjustment is listed in my link of post 12.  In short, you need to know how many of the team’s wins to give to the offense and how much to the defense, and you can only do that once you know how hard or easy it is to score runs there.

The customLWTS only translates components to runs.  But the conversion from runs to wins requires our knowing how to split the responsiibility between off/def.

Thanks for responding quickly Patriot and Tango. I have to meet with my therapist at 3:00 (yes I see a therapist and I hate every minute of it). I’ll take a look at Tango’s link in #12 when I get back.

Tango is right, because the kind of custom R/W converter I am had in mind would have to incorporate the park factor.

What I was talking about (but didn’t explain at all) would be something along the lines of: Larry Walker is +75 runs in Coors Field weights...Coors Field has a 1.2 park factor, and thus the RPG is 6 rather than the league average 5, and so the RPW is 12 rather than 10, and so his 75 runs convert to 6.25 wins.  The PF is still there in some form or another, but it’s just absorbed into the RPW, and you don’t have to explicitly say that you are using it...because from the wins perspective, I really don’t care what the park effect is per se, just what the runs:wins relationship looks like.

That being said, the Bill James argument on the quick PFs that Tango discusses in #12 is still interesting 25 years later.  From a certain definition of value, you might not care about splitting the credit between offense and defense, and want a 50/50 split.

But for the purposes of this discussion, Tango and I are in agreement as he surmised.

Which of these methods is more correct? Or would they both yield similar/identical results in terms of wins above average or wins above replacement.

1. Applying league average linear weights and then park-adjusting the runs above average/runs created. When converting runs to wins, you would use the same runs-per-win figure for all teams.

2. Applying team specific linear weights and then park adjusting the runs above average/runs created. When converting runs to wins, you would use a different runs-per-win figure for all teams.

I could be off here (I’m just mulling things over at work), but you have to be careful with your baselines here. You’d have to use LWTS baselined to zero and then convert those to RAA or RAR - otherwise you have offensive performance baselined to team average, which isn’t what we’re interested in. At the same time, you have to feed the runs/wins converter RAA or RAR, otherwise you end up with total wins in excess of actual wins.

Colin, when you use team-specific linear weights, you do not have to force them to zero for each team (nor should you force them to zero for each team).

When you input the team statistics into Baseruns, you end-up with linear weights that will approximately sum to runs scored. I like to call the linear weights generated by Baseruns “R/O linear weights.” You would then need to convert the R/O linear weights into “runs above average linear weights” by subtracting out the team’s runs-per-out from each of the out values. For the most part, the RAA for good offenses will end up being positive while the RAA for bad offenses will end up being negative.

Ignore what I said in #19. If you generate custom linear weights for each team using Baseruns, and subtract out the team’s runs-per-out, the linear weights will sum to zero for each team.

Rally gives us his top 300 players:
http://www.baseballprojection.com/war/top300.htm

Every eligible player at 72 WAR and above is in the HOF. 

There’s a host of players from 60-72 WAR where the argument has been there for the HOF.  It basically comes down to if are a big-hall or small-hall kind of guy.

There’s a host of players in the 50-60 WAR category that you are left thinking “if only...”, or “I dunno… maybe should be a little higher”.

But, all-in-all, this has got to be one of the, if not the, best non-opinion lists around.

Another interesting list is to convert Rally’s WAR to WAA.

(RAR-REP)*WAR/RAR does the trick, I think. May (or may not) be better for HOF.

Another thing you can do is convert the whole thing into a rate stat - say if all the extra wins were on offense, what would the player’s WOBA have to be?

Once you have a rate stat you can figure how many standard deviations a player is from average, from replacement, or whatever baseline you want.

This will give an edge to high peak players.