Look: I am eager to learn stuff I don't know--which requires actively courting and posting smart disagreement.

But as you will understand, I don't like to post things that mischaracterize and are aimed to mislead.

-- Brad Delong

Copyright Notice

Everything that appears on this blog is the copyrighted property of somebody. Often, but not always, that somebody is me. For things that are not mine, I either have obtained permission, or claim fair use. Feel free to quote me, but attribute, please. My photos and poetry are dear to my heart, and may not be used without permission. Ditto, my other intellectual property, such as charts and graphs. I'm probably willing to share. Let's talk. Violators will be damned for all eternity to the circle of hell populated by Rosanne Barr, Mrs Miller [look her up], and trombonists who are unable play in tune. You cannot possibly imagine the agony. If you have a question, email me: jazzbumpa@gmail.com. I'll answer when I feel like it. Cheers!

Monday, August 12, 2013

W.A.R. - The Nonsense Side of Sabermetrics

I'm all for statistical analysis.  Properly applied it helps us make sense of things that happen in the real world.  Such is the case with most of the baseball stats known as Sabermetrics.  Improperly applied, though, you get gibberish, like the category of Sabermetric statistics collectively known as "Wins Above Replacement." [W.A.R.]  You can read about it's various guises here.

The basic idea in any version of W.A.R is that you gather a bunch of statistics on a player, and use some algorithm to calculate how many more wins the team accrues because of his play, as compared to an entry level replacement just called up from the minors.

It's nonsense.

Consider the case of an offensive player.  [W.A.R. is also calculated for pitchers, but we'll ignore that.]   According to the chosen algorithm, these stats are aggregated and massaged into a single number that represents that alleged contribution to team wins for that player.   The problem is that statistics are probabilistic, but wins and losses happen in specific instances, where statistics are meaningless.

This is such a fundamental and obvious flaw, that I'm astounded that anyone even came up with the idea.

Consider the just concluded three game series with the Tigers visiting the Yankees.  Game 1, Game 2, Game 3.   In this series, Miguel Cabrera, arguably the best hitter in the game today, "went 7-for-13 with two walks and a strikeout for the series, and homered in three consecutive games, each to a different part of the field."  Also he became "the first player in history to homer off Rivera in consecutive at-bats."  [Quotes from Game 3 link.]

Quite an amazing feat for anyone at any time, and even more so for a hitter who is hobbled with a nagging hip flexor injury.  But how much did all of this contribute to wins?  Sadly, not at all.  His two pops off Rivera in games 1 and 3 were dramatic clutch hits, one tying the game and the other setting the stage for Victor Martinez' game tying blast. But in each case, the Yankees went on to pull out the victory.  In game two, a 9-3 Tigers romp, Miggy's HR was just one RBI among many.

This was great hitting overall, most particularly in dramatic clutch situations, but, alas, a wins contribution of exactly zero.

Just a few days earlier in Cleveland, Don Kelly provided the winning punch with a three run homer in a 5-1 Tigers victory.   What do you imagine Kelly's and Carbera's W.A.R. values are? 

 If you're interested, they are 0.9 and 7.3, respectively.

On the pitching side, 17-1 Max Scherzer, with an ERA of 2.84, 9.95 K/9 and 0.8 HR/9 has a WAR of 4.8.

Really.

So - not only is a win the result of clutch play by an individual in a specific circumstance, it's a team result that depends not only on what the rest of the team is doing, but just as much on what the other team has done on that particular day.

Wins Above Replacement.

Nonsense on stilts.



No comments: