2008 NL MVP: Pujols v. Howard
Thursday, November 20th, 2008This comes from an email exchange with my friend JTT. My email:
Here is another article and counter article by two really good
sportswriters, Thomas Boswell and Joe Posnanski. Definitely read
these, Boswell is making points very similar to what you were making
earlier this year, JTT. Thing is he's getting blasted by commenters
and other bloggers/writers.
http://voices.washingtonpost.com/baseball-insider/2008/11/mvps_howard_k-rod_not_pujols_p.html
http://joeposnanski.com/JoeBlog/2008/11/19/life-of-boswell/
JTT's email response:
specific judgment of Howard over Pujols (as indicated in my prior
e-mail), but I don't think it's crazy. I strongly agree with Boswell
on this though: the VORP discrepancy between Howard and Pujols is just
so great that the obvious inference to draw here is that this is one of
those things VORP gets badly wrong. That's not the end of the world;
it doesn't mean VORP is a bad stat. Even good stats will get things
wrong, occasionally. Here is one time when VORP really screws up.
And, truth be told, everyone really knows it's misleading here. Even
the people attacking Boswell so harshly here surely don't believe that
Pujols is 3 times more valuable than Ryan Howard,
which is what you would have to believe if you accepted their VORP
ratings. If the Boswell-critics really do believe this, we need to
bring them into our fantasy league and EWB leagues, so I can exploit
the hell out of them. I'll gladly trade them my 1 Pujols for their 3
Ryan Howards, and we'll just see who fares better.
number of baseball stats you could consider. Take slugging percentage
for example. To calulate it, you take ((singles x 1) + (doubles x 2) +
(triples x 3) + (HR x 4))/AB. Offhand, it seems obvious why you would
select just these values — why a single should get 1 point, a double 2
points, and so on. But in truth, I would bet my life that these aren't
the "real values" of these different sorts of hits. That is, I would
bet my life that 1 HR isn't "equivalent in value" to 4 singles, or to 2
doubles, or to 1 triple + 1 single, etc. Here is a quick argument for
the conclusion: Suppose you get to choose what your team's offensive
production will be today, which option do you prefer (a) 1 HR, no other
hits, or (b) 1 triple, 1 single (not necessarily consecutive, just over
the course of the game), and no other hits? I guarantee you that (a)
on average will deliver your team more runs, because (a) guarantees at
least *one* run, the guy who hit the HR (in addition, there may or may
not be runners on base). In contrast, (b) doesn't guarantee you at
least one run. If the triple takes place in the 2nd inning and the
single in the 7th, it doesn't really help you at all. Obviously, there
are possible scenarios where (b) leads to more runs: the triple
happened right after you got two walks, and the single happened when
there was a runner on 2nd thanks to an error, etc. Still, I guarantee
that (a) will have the higher expected run-value than (b).
it: ((singles x 1) + (doubles x 2) + (triples x 3) + (HR x 4.2))/AB.
Call this SLUG* (slug-prime). The way it differs from original SLUG is
that SLUG* assigns a value of 4.2 to a home run, instead of just 4.
This is to reflect our latest conclusion, that a HR is really worth
more than 1 triple + 1 single. (According to SLUG*, a HR is worth 4.2,
while a triple + single is worth 4.) Well, from that first step, you
can see how (literally, infinitiely) many different revisions might
go. Why not assign a single the value of .9, or .91, or .911, or
.9111, etc., infinitely. I say let a thousand flowers bloom, and use
each statistic so-generated: so you have SLUG, SLUG*, SLUG**, SLUG***,
and so on, for infitenly many different stats. And that's just
limiting our attention to variations on slugging percentage — here
alone we get infinitely many different statistical measures.
valuations of players. So for instance, Pedroia has a worse SLUG than
Youkilis. But, suppose that if you use SLUG*****, Pedroia is better
than Youkilis. So then, the question is, which statistic is a better
measure of true value, SLUG or SLUG*****? If it's SLUG, then Youkilis
should get the MVP; if it's SLUG*****, Pedroia should get the vote.
But how do you find out which statistic is better?
"commonsense" . If you want me to back up my commonsense verdicts,
with "stats," I can do that all day. Playing the role of Boswell,
Pujols is better than Howard at VORP, but Howard is better than Pujols
at VORP x RBI. (VORP x RBI isn't a pre-existing stat, I just made it
up: it's VORP multiplied by RBI total.) So question, why should we
care about VORP more than VORP x RBI? Why not vote on the basis of
VORP x RBI and thus give the award to Howard over Pujols? Hey, that
way, we can claim to be honoring both Sabermetrics (by including VORP) and traditional stats (by including RBIs). Everyone should be happy.
no way to figure out which stat is bettern between SLUG vs. SLUG****,
between VORP vs. VORP x RBI, than by appealing to commonsense. There's
no way to assess it by appealing to more stats. For, even if you do
show me some stat that indicates VORP is better, I'll just invent my
own stat that says VORP x RBI is better. (And again, Sabermetrics
people ultimately ground their claims in commonsense anyway, e.g., when
they discredit RBIs. They argue that RBIs depend on too many
extraneous factors. Let me stipulate that they're right about this.
Even so, it's a commonsense appeal, not statistics.)

