Because this appears to be the successor to the Dugout, I figured this might be the place to pose the following question:

How do WARP or other similar stats account for defense? Specifically, if you were able to determine a player’s offensive contribution (say, something like Runs over Replacement), (a) how would/is defense measured and (b) how is this incorporated into an overall value?

Yes, I am thinking of DMB here—specifically trying to determine how much to weigh defense in assessing various players.

Very good question, and sadly I can’t answer it. WARP is a BP statistic, and while I love BP’s work they don’t release their formulas (to my knowledge).

I wasn’t trying to refer specifically to WARP; rather, I’m really interested in finding how to determine the role of defense in general and how to incorporate into an overall assessment.

For instance, I could hypothetically determine:

Overall Contribution = Offense + (X * Assists) + ( Y * Putouts) - (Z * Errors)

Of course, I could include other variables (such as double plays, etc.) Is this a fair scheme and, if so, how would I determine X, Y, Z etc.?

I suppose you could run a regression with WARP or the overall statistic of your choice and a defensive metric. That would be able to give you a feeling for how much defense (overall defense, that is) is a part of a player’s total contribution.

I hope I’m helping here…do I understand your question correctly? Or do you actually want to determine X, Y, Z, and so on?

I suppose you could run a regression with WARP or the overall statistic of your choice and a defensive metric. That would be able to give you a feeling for how much defense (overall defense, that is) is a part of a player’s total contribution.

I hope I’m helping here…do I understand your question correctly? Or do you actually want to determine X, Y, Z, and so on?

Yeah, sort of. My end goal is to take the results of simulated DMB seasons and determine a player’s overall value (even if my methods are quite rough and “back of the envelope”). I can figure out an offensive contribution by taking, say, Runs Created and figure out how much they are over a replacement level.

What I really want to do is to incorporate defense somehow to determine an overall player value. Through DMB, I can determine assists, putouts, DPs, total chances, errors, passed balls, etc.—how do I take it from there? (I suppose that I could hypothetically and arbitrarily figure that all errors cost, say, 0.25 runs, and simply subtract them from the offensive contribution, but of course this doesn’t take into account the player’s range, total chances, etc.)

Put another way: When I postulated Overall Value = RCoverRepl + (X * putouts) + (Y * assists), etc.—what stats should I be considering?

Is this the right kind of thinking (using an equation like this) or should I be going in a completely different direction?

So you’re pretty much trying to find “defensive linear weights” ???

I just ran a multiple regression, attempting to find a rough estimate for you (very rough—just numbers from 2001-2005, not park adjusted or anything). I found that you shouldn’t use double plays—it doesn’t have much effect on the data. So, here it goes:

(numbers as coefficients, x y and z)

Assists- 0.1571
Putouts- 0.1203
Errors- -0.4763

Of course, this doesn’t take total chances into account.

Thanks, Saber Head, that really helps a lot. I also figure that if you know assists, putouts, and errors, you basically have total chances, no?

Really, though, I was first trying to ask a more fundamental question: Is something like “defensive linear weights” even a valid basis for measuring an overall contribution?

If it isn’t, what is?

How do the various means of assessing overall value (Win Shares, WARP, or some other metric) measure/weigh defense?

I really like you’re question, and now you’ve got me thinking about it. FIP, a popular defense-indpendent metric, weights its components. So why not the same for the rest of pitching/defense? I have a hard time imagining that defense is not linear…

I’ll collect my thoughts and post back later…I’m glad I could help you.

I’ve been thinking about this, and I’ve decided that I’d like to work on it a bit. If you right-click and select “save target as” on the link below, you’ll download a spreadsheet listing the top 65 players in this new fielding model (it’s the stat on the far right). Now, one way for me to determine how it’s working (and what changes I have to make) is to see if the top players are in fact considered go fielders—or at the least they aren’t bad players. Please take a look at it, scim through the players listed, and let me know if there are any “oddballs.”

Soriano is an odd choice for the list, as is Jeter. Also, I never thought that Konerko was much of a glove man.

One thing that strikes me about the list is that it’s dominated by middle infielders. I suppose any sort of linear weighting would do that simply because they have more chances—so that a top LF wouldn’t rate as highly as an average SS. For my purposes, though, that’s probably right. (I’d compare LF by position, not overall.)

Soriano is an odd choice for the list, as is Jeter. Also, I never thought that Konerko was much of a glove man.

One thing that strikes me about the list is that it’s dominated by middle infielders. I suppose any sort of linear weighting would do that simply because they have more chances—so that a top LF wouldn’t rate as highly as an average SS. For my purposes, though, that’s probably right. (I’d compare LF by position, not overall.)

Good point. Yeah, if you set it up so you’re looking above average (or replacement) at a certain position it will probably work a bit better. I’m going to have to take another look at it.

There is also an issue with catchers. Also, you have to adjust for K tendencies of the pitching staff. The more pitchers stirke out jitters - the fewer BIP fo the fielders to get.

I just started reading “Win Shares” and the calculation is very complex. Each position has a different criteria. If I get a chance this weekend maybe I can post a summary of how Win Shares were originally calculated.