I don't think we can.

I think we can get something to confirm suspicions like, "Ozzie WAS really good," etc. I'd like to see how it translates to the error-rich times of yore.

I think a wisdom of the crowds approach is best for older fielders; especially if you can find observers that Totalzone agrees with and then find predecessors that agree with those observers and so on and so forth.

What #1 said. *Actually* I think it can be done, but the level of scrutiny has to be nearly infinitely higher. And of course, there will still be limits.

To be clear - I don't think we can have a single method that covers all of baseball history that we'd feel comfortable using for modern periods. I do think that someone will come up with a weaker version of TotalZone for olden times.

Basically I think we can get to letter grades with around an 85% confidence that we're correct.

There's a potential sanity check that I haven't seen applied for any of the defensive metrics. Basically check how pitchers changing teams are affected.

For me, the Holy Grail of fielding statistics is going to be a method that can cover the whole of baseball history.

BTW, I'm not sure if current batting metrics, let alone fielding metrics, cover the 19th Century well.

I think a wisdom of the crowds approach is best for older fielders.

To a point. The systems that have evolved tend to be based on the same underlying set of data, and we don't know the extent to which recent ball-in-play data (both types of balls in play and distribution of balls in play) is representative of historical ball-in-play data. Furthermore, most of the systems are vetted against either ZR or UZR, and to the extent those systems are biased because of the nature of the data set, the wisdom of the crowds approach will reflect the biases.

-- MWE

I do think that someone will come up with a weaker version of TotalZone for olden times.

I do have it, worked on the outfielders last night. But it's not very good, nothing that I'd use if I had alternative and better data. That being said, high error rates are to our advantage. When a good fielder has 7 errors and a bad one 18, then error rate does little to show the difference between them, compared to some estimate of how many hits allowed. But if a good fielder has 25 errors and a bad one has 80, then that's a big difference which does explain a lot (though certainly not all) of their defensive worth.

Outfielders are a pain in the A, because not only do you not have innings played, you've got guys who split their time among three positions and all you have listed is "OF". It is a huge benefit that the Lahman database has games played by OF position going back to 1871. But it still requires some mathematical gymnastics to estimate how much time the OF played at each spot, and how much of his PO, A, E should be estimated to be at each spot.

Soon I'll be able to give an estimate, and so far it looks like it passes the smell test - players who were well-regarded defensively score well, those who weren't do not, and the ranges in runs are in the same ballpark as the modern defensive metrics show (I would have scrapped it if I saw players rating as +70 or -70 for a season or something like that). Still, the holy grail is for retrosheet to publish play by play accounts going all the way back. They have completed the boxscore accounts for all of the 1920s so one can hope.

Didn't realize my post would kill the thread.

I need some help in rating the best outfielders from the first part of last century

Here are the players to be rated, click the link above for more information, but I'm just asking for a rating of 1 to 5 (1 being worst) for their defensive play. I will count all responses made here, my blog, or on Tango's blog.

The left fielders:
Sherry Magee
Zach Wheat
Duffy Lewis
Bobby Veach
Charlie Jamison
Goose Goslin
Heinie Manush
Al Simmons
Joe Medwick
Bob Johnson
Ted Williams
Ralph Kiner

center field:
Ty Cobb
Clyde Milan
Dode Paskert
Tris Speaker
Max Carey
Cy Williams
Edd Roush
Lloyd Waner
Sam West
Earl Averill
Doc Cramer
Joe DiMaggio
Dom DiMaggio

Right field:
Sam Crawford
Harry Hooper
Babe Ruth
Sam Rice
Harry Heilmann
Paul Waner
Mel Ott
Chuck Klein
Wally Moses
Bill Nicholson
Enos Slaughter

No love for Bob Meusel?

Feel free to add Bob in if you want to submit a ballot. I probably forgot him because he split time between LF/RF, so didn't rank high on career games for either. Same for Babe Ruth, but I added the Babe back in since he's so obvious.

Furthermore, most of the systems are vetted against either ZR or UZR, and to the extent those systems are biased because of the nature of the data set, the wisdom of the crowds approach will reflect the biases.

Mike, I wasn't talking about the wisdom of THIS crowd. I was thinking more about the wisdom of Shirley Povich, Fred Lieb, Red Smith, Casey Stengel, and/or Red Barber. Folks like that.

the ranges in runs are in the same ballpark as the modern defensive metrics show (I would have scrapped it if I saw players rating as +70 or -70 for a season or something like that)

I don't know...in an age when there were more balls in play, more errors, most likely a greater spread of talent, I think the ranges should certainly be larger than today.

#15, I agree that if we had reliable PBP data we'd expect a wider spread. On the other hand it makes all sorts of sense to be conservative when you know that your error bars have to be huge.

Maybe a little bigger, but huge ranges in the ratings are indicative of flawed methods. Like finding the average SS in Ozzie 1980's innings had 500 assists, Ozzie had 600, so he's +100 plays or +75 runs. A method has to be better than that or else I'm not going to get anywhere near it.

BTW, I'm not sure if current batting metrics, let alone fielding metrics, cover the 19th Century well.

I think we have a pretty good idea of pre-modern ordinal measurement, just not cardinal measurement. A lack of reliable cardinal measures for 19th century frustrates inter-era comparisons, but otherwise I think that baseball historians have a pretty good handle on who were the best from that era in pretty much ever facet of the game.

And frankly the deadball game was so radically different than the modern form, I don't think it would possible to construct an valid inter-era metric even if we had PBP data for the 19th century. IMHO, those hoping to develop a historical defensive system that allows comparisons to current players should just start with the 1920s (or possibly post-WWII for data considerations).

I think we have a pretty good idea of pre-modern ordinal measurement, just not cardinal measurement.

Right. It wouldn't make as much sense to trot out Roger Connor's WARP3 or Albert Spalding's ERA+ as it would were we talking about players 100 years later.