Good discussion. I'm a 'stats' guy. Love WAR for the weighted system it offers. I'm not quite on board for how the metrics are kept for defense (UZR) just yet. 

 

As many others pointed out, positional weighting can really put a player in a very good light when the position is particular weak for that year. e.g. Donaldson, Gordon.

 

For those who don't think Gordon is a good defender - the guy impresses the crap out of me. One hop off the fence, no hesitation, on a rope, right on the bag. Time after time. He probably turns 60-70 doubles a year into singles or 7-4 put out. If he was a catcher, his 'pop time' would be under 1.4 secs (for catchers, pop time is pitch from ball received to 2B). Heyward also...a 6'4" galloping ghost. Basically a faster Ken Griffey, JR. (also tall, at 6'3"). Now are their values as high as they should be for WAR? I don't know. I don't think so. But they are elite defenders from the eye test. 

 

SABR doesn't like the SB...but I do. I wish there was a better way to metrically track going 1st to 3rd base on a single to the outfield for a runner. Stuff like that. To be an elite SB player, e.g. 50+ SB talent at 90%+ success rate is going to cause more damage than most metrics will currently track.

 

I like war, but you have to appreciate linear weights, weighted outcomes, and so forth to get at all excited about WAR. If those don't give you a mental 'up' then WAR won't. I'd then resort, for just being impressed with defense for what it is and then, for hitters, go with OPS+. 

If you still believe in ERA, then a debate about WAR is going to accomplish nothing.

 

 

Probably just as accurate as WAR. That is to say, not very.

Couldn't agree more. Very few things deal in binary yes/no.  Baseball fans have it easy with out/no out, hit/no hit, on base/not on base.  If people don't want to step outside the comfort zone of these binary statistics, no one is forcing them.

 

That said, nearly every science and statistical analysis cannot work on yes/no.  Everything is gray area and uncertainty (thousands of non-baseball examples out there, all of which are probably off-topic).  As a baseball example, I really liked the homerun vs. out comparison in the Fangraphs article you linked.  The result is binary (homerun/no homerun), but, if we want to apply this binary result to something more subjective (Davis's skill at hitting the ball), then we run into something that necessarily requires error bars.

The authors of these statistics gloss over the sampe sizes needed for reliability, and margin of eror. Once it is said into the blogshere, it is taken as gospel. BABIP is said to be random. I wish I could find the study  that said BABIPnormalizes after about 1000 innings for a pitcher. That is also about the time when all of the predictos of what the ERA will be matches the era of the pitcher. Molina saves 50 runs a year yet the staff ERA is not the best despite the quaity of ptchers. They  gloss over that there is little correlation on pitch framing and actual runs not being scored. Defensive metics I never looked at.

It is a racket. Pubish articles, get people to your website, make moey. Scientific, PFFFFFFFFFFFFFFFT.

 

Technology and baseball is a diferent story. The tools like trackman and pitch f/x provide real data 

Let us not forget the humble confidence interval, either.

 

Thanks, my brain was not working yesterday, that is the concept I was trying so hard to reference......btw, I kind of miss the weeping angels. I told my wife last weekend if we ever bought another house, I'd want at least one weeping angel in the backyard. She found it odd I'd want to constantly challenge myself not to be freaked out......

Fangraphs posted this article yesterday to get community feedback on the defensive component of WAR.  The genesis for this debate was from Jeff Passan (Yahoo!), a big Fangraphs supporter with some reservations about defense in WAR.  Fangraphs talks about continuing wanting to push the model forward and polls users at the end. 

 

Passan nails my views here:

Defense does have its place in WAR. Just not in its present incarnation, not until we know more. Not until we can account for positioning on the field. Not until we can find out the exact speed a ball leaves a bat and how quickly the fielder gets a jump and the angle on the ball and the efficiency with which he reaches it. Not until we understand more about fielding, which will allow us to understand how to properly mete out value on a defensive play, which may take years, yes, but look how long it took us to get to this point, where we know more about hitting and pitching than anyone ever thought possible.

 

 

In a related podcast, Dave Cameron identifies 3 options to reduce the impact of defensive metrics with the data that we do have today (skip to the 14 min mark):

1. 'We don't trust these numbers' so, we constrain the values by regressing towards zero.  Cameron says the amount of regression would be fairly arbitrary, but mentions reducing them by half.

2. 'We want to be closer to the player's true talent level' so, we regress towards the player's average over a larger sample size.  For example, if a player has averaged +20 UZR the last three years but is at a 0 UZR this year, his value would adjust towards his previous performance.  This approach targets the issues around a single season of defensive data.

3. 'We want a bigger sample' so, we use an average over a larger sample.  Cameron mentions using three years, which also targets the issues with single season defensive data.

 

The current model gives pitching 43% of all available WAR and 57% to hitters/position players.  Run creation accounts for half of the game and all of that value is given to the hitter.  Run prevention accounts for the other half and that value is currently considered 85% pitcher and 15% defense (hence the 43-57 split).  This is his main issue with decreasing the amount of value placed on defense -- it breaks that distribution and forces more value to the pitchers -- which isn't supported by the available data.

 

I think options 2 and 3 could be done in a way that does not disrupt the distribution of value, but provides for a more accurate depiction while simultaneously reducing the defensive value outliers (aka Alex Gordon). 

 

Any thoughts?

Edited September 9, 201412 yr by jay

I loved Passan's article.  I thought it was one of the best pieces of baseball writing I've read in awhile.

Lots of interesting debate in the comments on the FG article.

 

The idea of error bars that was previously mentioned here is brought up:

"Error bars would be a very interesting concept. We could say within absolute confidence that a player is worth ___ WAR with a degree of error of +/- _____. As of right now, WAR is very precise, but not entirely accurate."

 

Cameron seems dead set on maintaining that 100% of value is assigned somewhere.  The counter that has developed is to use some sort of an "other" bucket to put value that we aren't sure about.  The idea garners support from Tom Tango.

"Really though, why do we insist on giving out 100% of the wins (minus replacement level)? We all acknowledge that there are things that aren’t measured by WAR as well as things that WAR claims to measure but may do so inaccurately. Why not have, for example, 54% wins for position players, 43% wins for pitchers, and 3% unassigned because they’re indeterminable?"