Beane listens to his scouts, for sure. But when it comes to analyzing defense, he prefers cold, hard data to knee-jerk conclusions. Even his own.

The question for me is what kind of proprietary system do teams like the A's use, and is it any good?

The question for me is what kind of proprietary system do teams like the A's use, and is it any good?

I think it was described in Moneyball--assuming it's the same system. Essentially the pitcher receives credit for the expected outcome of the batted ball based on trajectory and speed, while the fielder is credited with the difference between that outcome and the actual outcome.

Isn't the rabbit in a concealed compartment in the hat?

The question for me is what kind of proprietary system do teams like the A's use, and is it any good?

1. who knows?
2. Presumably they think it's better than UZR and Fielding Bible to which they have full access.

My best guess on #1 is that it would be pbp stats supplemented with video analysis. The only other real alternative is access to better (i.e. less prone to measurement error) pbp data but I don't know who they'd get that from. I could imagine them having reasonably elaborate setups in their own ballparks that might give them more accurate pbp data but that would only give them half a season of data on their own players and nothing else.

Of course I'm not convinced a blindfolded JP Ricciardi tossing darts at a dartboard isn't as accurate a measure of defense as what we've got. :-)

My best guess on #1 is that it would be pbp stats supplemented with video analysis.

Do you think teams will ever release their old data? Or is this stuff locked away from view forever?

and one American League GM described Holliday as “brutal” with a glove.

He's never going to be able to live down getting hit in the dick by that liner, is he?

Essentially the pitcher receives credit for the expected outcome of the batted ball based on trajectory and speed, while the fielder is credited with the difference between that outcome and the actual outcome.

So if you expect it to be an out and it isn't, the fielder gets blamed; if you expect it to be a hit and it isn't, the fielder gets credit. In some ways I actually like that better than assigning a percentage credit based on the average result, although I think you have to make an allowance for positioning; you would expect different results in certain areas of the field when Ryan Howard is batting than you do when Manny Ramirez is batting.

-- MWE

Of course I'm not convinced a blindfolded JP Ricciardi tossing darts at a dartboard isn't as accurate a measure of defense as what we've got. :-)

Given the quality of the Jays' defense over the past couple years, a blindfolded Ricciardi might actually be better than what we've currently got - evaluating defense seems to have been one of the few things he was very good at.

He's never going to be able to live down getting hit in the dick by that liner, is he?

I've never seen a player's defensive reputation change so radically because of one play. He will never be forgiven for the sympathy pain he inflicted on an unsuspecting male public.

You take one game-losing ball to sack and you're branded a terrible defender. It's unfair! A ball to the groin is its own punishment!

I've never seen a player's defensive reputation change so radically because of one play. He will never be forgiven for the sympathy pain he inflicted on an unsuspecting male public.

I find it funny that the only two plays I can distinctly remember Matt Holliday making in the outifled are the dick ball, and Brian Giles' double over his head that he totally misplayed in the Game 163 a couple years ago.

Holliday must be the poster boy of the perception vs. reality debate in terms of defence. He's a slugger who totally muffed two balls on national television. I can see how a lot of people would just assume he's a bad defender and damn the numbers.

Holliday must be the poster boy of the perception vs. reality debate in terms of defence. He's a slugger who totally muffed two balls on national television. I can see how a lot of people would just assume he's a bad defender and damn the numbers.

I can see why people consider him Lonnie Smithesque because of those plays.

I've watched Holliday on the Rockies quite a bit and saw virtually every game that he started with the Cardinals. He *does* look absolutely brutal in the field, always breaking the wrong way and stumbling around awkwardly while squinting up as though he's lost the ball in the lights. His fielding looks like nothing so much as a Manny Ramirez clone who's trying really, really hard.

That fielding metrics believe him to be not only an average fielder, but even a good one, is mind-boggling. They shouldn't be discounted, but they're still mind-boggling.

I don't follow Mike - how different are the two - doesn't almost everything use a variant of sum of (e(outcome)-outcome)?

I don't follow Mike - how different are the two - doesn't almost everything use a variant of sum of (e(outcome)-outcome)?

I think he's talking about systems where a ball that is caught and was expected to be caught with probability .45 is credited 55% to the fielder and 45% to the pitcher. Is that right, Mike?

I took it as: +.55 fielder or -0.45 fielder. This is a little different than some zone rating type stuff, where you handle in zone (implied 100% of the time) and out of zone (implied 0% of the time, but also in a different accounting) - and I guess I'm assuming some sort of balancing to make it a zero-sum game....

Fine but the existing tools are still better than "I watch the games ... and let me tell you Derek Jeter can suck the chrome off a 1957 Chrysler 300"

What I was talking about where systems where the fielder is credited with an expectation based on the average result of all balls hit in the zone - e.g if a ball is caught 80% of the time and goes for a hit 20% of the time, the fielder gets credit for the 20% delta if he catches the ball and is debited for the 80% delta if he misses.

What I understand Beane's system to be is a more sharply defined than that; if the expected result is a single based on speed and trajectory, the pitcher is credited (or debited) with allowing a single, and the fielder would be credited with the difference between what you would expect following a single (including any typical runner advancement) and the actual result of the play. So if the ball is hit into an area with a runner on first and no out where the expected result would be first and second and no out, the pitcher would be credited with first and second, no out. If the fielder caught the ball, the fielder would get credit for the difference between first and second, no out and runner on first, one out. If the fielder was slow in cutting the ball off so that the runner on first went to third, the fielder would be debited the difference between first and second and first and third. This has the effect of removing routine plays made, and penalties for outstanding plays not made, from the calculations. It also would take context into consideration as well; missing a routine play with runners on base costs a team more than missing a routine play with bases empty. And I suppose it could be (and perhaps is) more finely-tuned to account for positioning and degree of hittability as well; a fielder would not be penalized quite as much for not making a play when the pitcher is being hammered, or not making a play on a GB hit to the left side of the infield by Ryan Howard when the shift was on, as he would be when the pitcher was pitching well or when he was positioned normally.

-- MWE

Thanks Mike. I really wish we could get a peak at the results they come up with.

What I understand Beane's system to be is a more sharply defined than that; if the expected result is a single based on speed and trajectory, the pitcher is credited (or debited) with allowing a single, and the fielder would be credited with the difference between what you would expect following a single (including any typical runner advancement) and the actual result of the play.
But do you mean:
- the expectation is a single
or
- the expectation is: single (80%), out (20%)

I follow the rest.

So if you expect it to be an out and it isn't, the fielder gets blamed; if you expect it to be a hit and it isn't, the fielder gets credit. In some ways I actually like that better than assigning a percentage credit based on the average result, although I think you have to make an allowance for positioning; you would expect different results in certain areas of the field when Ryan Howard is batting than you do when Manny Ramirez is batting.

-- MWE

it's certainly a lot better than expecting a left fielder in fenway to be 3-6x taller than yao ming to catch a ball...

But do you mean:
- the expectation is a single

Yes, that's what I mean. If you don't expect the fielder to catch the ball (and since 80% of the time he won't I would assume that's the expectation), then you assume that when he does make that play it's because he's a good fielder and credit him with the full 100% difference rather than 80% of the difference.

-- MWE

JP might have done a good job evaluating defenders, but blindly following UZR could have led to the same results. It's not like Scott Rolen being a great defender was any kind of secret.

Jays defense really took a hit in the 2nd half. They had Dh's in the corners with Lind and Snider, a corner guy playing center, and at 3b, Encarnacion. As a fielder, he's a pretty good hitter. Now ss is up in the air. No wonder Halladay wants out.

But do you mean:
- the expectation is a single
or
- the expectation is: single (80%), out (20%)

Mike, I don't see where you get the idea that the A's do the former.

I also don't see why it's preferable. Say there are two opportunities, one that is caught 10% of the time and one that is caught 48% of the time; the former method doesn't distinguish between them, but it does see an opportunity that is caught 52% of the time is drastically different.

For what it's worth, Probabilistic Model of Range (PMR)previously has rated Holliday below average in LF. Here is the address for PMR for LF in 2008:

http://baseballmusings.com/?p=29077

Holliday is in the bottom half of MLB left fielders, according to PMR, just one step above Manny Ramirez.

Billy Beane never should have invented UZR.

When Holliday got to Oakland and I started watching all his games, I was shocked by how unathletic he looks. He's stiff. He's duck footed, and he shuffles more than runs (unless he's going full out). His weight is also distributed more to the lower half than a typical great athlete, and the overall effect is very heavy and slow looking feet. Can definitely see why a scout would think he looks terrible. Still, I don't really remember ever thinking he should have got to a ball that fell, so it's totally believable that he's moving quicker than he appears.

Mike, I don't see where you get the idea that the A's do the former.

Tha's my intepretation of the comment in #2 that the A's credit the pitcher with the expected result based on speed and trajectory of the ball. I would think of that as a binary decision - either you expect the play to be made or you expect it not to be made.

I also don't see why it's preferable.

It's related to the other comment. Fielding has, essentially, binary outcomes; either a play is made or a play isn't. And the determination about what you expect on a play is IMO ALSO binary; either you expect the play to be made or you don't.

The biggest issue that I have with "pure DIPS" is in a sense the biggest issue that I have with most PBP-based fielding systems; such systems try to allocate credit or blame for every ball in play to the fielders, ignoring the role that the pitcher plays. As I said elsewhere, "fielder-dependent" does not mean "pitcher-independent", and we need to recognize that on some subset of balls in play the credit/debit belongs to the pitcher and not to the fielders. The A's approach is preferable to me because there are clear definitions of responsibility; the pitcher gets the credit for the expected results and the fielder gets the credit for the difference between the expected results and what actually happened. I also prefer a binary "yes/no" on both sides to a percentage system because I believe it's easier to account for situational changes in fielder positioning that way - when you assign based on percentage of the time a play is made in a particular area you almost of necessity have to aggregate numbers together from unlike game situations, and the final percentages may not accurately reflect the expectation for the particular game situation.

A year ago, I looked at the 10 least likely plays-to-be-made by second basemen in Pinto's PMR system. Something like half of them were ground outs hit into the teeth of a shifted infield, all of which were fairly routine given where the 2B was positioned at the time. I would think that a system like the A's system would properly recognize those plays as being routine.

-- MWE

I don't see why you're tying binary v. probabilistic to account for game state / positioning / "pitcher responsibility issues." Isn't the calculation of the A's "yes/no" decision (provided that that's what they are doing) subject to some of the same issues?