It’s getting better, but UZR could certainly benefit from better data.

I’m not sure what the problem is.

Let’s hypothesize there is a linear relationship between WAR and the “true” number of wins. We assume that the actual number of wins will have some variance from the “true” number, which we call luck or whatever.

So you wouldn’t want the correlation to be 1, since that means that the variance in wins from luck is zero, i.e. the actual number of wins is the true number of wins. But if WAR has a high correlation, that would still be okay, depending on much variance you’re assuming actual wins have from the “true” wins.

Hell, like mnbv noted above, there’s an element of luck involved (see Tampa Bay’s 2009 season, for one). However, as fans sitting in our mom’s basements tapping away at keyboards, we’re not privy to a lot of the other stuff that goes into the game. We can speculate about how Player X is in the locker room, but we don’t really know. We can speculate that Player Y is a headcase, but we don’t really know. We can speculate that Player Z is selfish and cares only about his owns numbers, but we don’t really know. We get “hung up” on statistical analysis because it’s really all that we have to go by.

I’m not saying you or I won’t notice when guys are dogging it in the field, but at least that’s something we can see. But besides the numbers and what we see with our own eyes watching the games, there really isn’t much else for us to knowledgeably talk about.

One question for you: what do you mean by “ebb and flow”?

Baseball is, in the final analysis, a team game where results are measured in wins. What we’re trying to find is the relationship between individual performance and team wins.

When you say:

in other words…….take the shittiest player (let’s say Francoeur for argument’s sake), and let’s say he wins you 5 games in 2010 with solo walk off HRs. Does that mean he is now a stud based on WAR ?.. I mean, he WON you 5 games.

That simply isn’t accurate. That home run in isolation did not “win” those games; some combination of the offense, pitching and defense surrounding Francoeur kept the game tied until that point. If Jeff Francoer tried to play baseball by himself he’d lose every time.

So when we say a player is worth X number of wins, what we’re really referring to is the the difference in the number of games a team would win without that player.

So if the Mets didn’t have Francoeur in right field, who WOULD they have? They wouldn’t necessarily have an average baseball player – there are really more below-average baseball players than there are average and above players. But they wouldn’t simply field no baseball player – someone would have that playing time instead.

What we call “replacement level” is simply an rough sketch of what the bare-minimum alternative looks like on the typical team – a non-prospect in AAA, or a veteran that can be signed to the minor league minimum.

WAR isn’t good at predicting performance, because it is purely results based. However, if you look at underlying reasons for success or lack thereof (BABIP, LOB%, LD%, etc.), it’s easier to understand how the fluctuation may occur.

At least, so long as the past performance is somewhat skill based.

People tend to view WAR as a “skill” metric sometimes because of FIP, but it’s not supposed to be one. FIP is used not because BABIP isn’t a pitcher skill but because it’s a fielder skill – if you don’t seperate a pitcher’s run prevention from that of his defense you’re double-counting the defense’s skill in run prevention.

Though to account for one’s average batting order (regular everyday player), I suggest this formula for projected number of PAs

(569 + [18 x (9 – avg. lineup position per PA)]) x (team OPS+ / 100) x (games played / 154)

i.e. An average everyday cleanup hitter (154 games in the 4 spot) on the Washington Nationals (team OPS+ = 95) would get about 626 PAs in 2009, assuming that player is a regular. Adam Dunn has 668 PAs in 159 games played this season, so he has about 22 more PAs than projected.

As for pitchers, their numbers can simply be scaled based on their workload. It’s easy to adjust for RPs, but not as much for SPs. I suggest the following one for SPs

(IP*/GS) x 32.5

 * Don’t count the number of IP in relief outings, if any!