Thursday, September 15, 2011

Naive PLAYOFFS+ MVP awards since 1995

by Jason Wojciechowski

On Monday, I wrote that a variety of theories of value might be considered valid ways to figure out who the MVP of a league is. The most controversial one, as predicted, was what I called PLAYOFFS+. The idea underlying PLAYOFFS+ is that wins that get a team into the playoffs are more valuable than wins that move a team from 70-92 to 76-86.

There are more or less sophisticated ways that PLAYOFFS+ could be implemented, as I mentioned in that post. One amusing way to do it is to use the following criteria:

  1. The player's team must make the playoffs (because making the playoffs is what ultimately matters if PLAYOFFS+ is the theory you hold)

  2. The player's individual WAR must be higher than the margin by which the team made the playoffs (this weeds out players that you could, theoretically, remove from the team and still have a playoff team -- this includes high performers on teams that dominate their divisions and low performers on all playoff teams)

  3. Of that group of players, sort by total WAR to get the top ten for the ballot (there's no other sensible way to order the players without whom their teams would not make the playoffs -- using some sort of question about the gap between individual WAR and the team's lead in the playoff race results in absurd answers on both ends)

To that end, I made a Google Doc that shows the MVP ballot for each league going back to 1995. The WAR that I've used here is rWAR, found at Baseball-Reference. I have no loyalty to this particular version of WAR -- it was just conveniently placed along with team standings, actual MVP voting, and so forth. This is all just for fun anyway.

You can hit that doc for the entire rundown. Here's a summary and discussion.

Let me list off the successes first, in terms of the system's results matching up with real-life voting:

  • Edgar Martinez, third in both fake and real life in 1995;
  • Ken Caminiti and Mike Piazza finish first and second on this ballot in 1996, just as in real life;
  • Ken Griffey Jr. wins both in 1997, with Jeff Bagwell third;
  • Nomar Garciaparra first on the ballot, second in real life in 1998;
  • Sammy Sosa nailed for first in 1998;
  • Pedro Martinez and Jeff Bagwell both first on the ballot, second in real life in 1999;
  • Alex Rodriguez, Jason Giambi, and Frank Thomas in the top three in both in 2000 (shuffled orders, though);
  • The 2001 NL is pretty awesome down-ballot: Luis Gonzalez 2nd (fake)/3rd (real), Albert Pujols 4th/4th, Lance Berkman 6th/5th, Chipper Jones 7th/8th, Jeff Bagwell 9th/7th;
  • Barry Bonds runs away with both in 2002, and Jeff Kent finishes 4th/6th;
  • Vlad Guerrero wins both in 2004 in the AL, and Adrian Beltre finishes 1st/2nd in the NL;
  • Alex Rodriguez wins both in 2005, with David Ortiz 3rd/2nd, and Mariano Rivera, weirdly, 9th/9th;
  • In the NL that year, Andruw Jones goes 2nd/2nd, Morgan Ensberg 3rd/4th;
  • Albert Pujols goes 1st/2nd in 2006;
  • Alex Rodriguez again wins both in 2007;
  • Matt Holliday is 1st/2nd the same year in the NL, with Jimmy Rollins 3rd/1st;
  • Kevin Youkilis was 2nd/3rd in 2008 with Carlos Quentin 3rd/5th;
  • Manny Ramirez's abbreviated tenure with the Dodgers in 2008 wasn't worth consideration by standard individual-only metrics, but the system put him tied for tenth and he finished fourth in real life. Similarly, CC Sabathia takes second place in the system and ninth in real life;
  • The system nails both awards in 2009, with Joe Mauer and Albert Pujols, and Troy Tulowitzki takes 3rd/5th as well. To be fair, Pujols led his league in rWAR and Mauer was second only to Zack Greinke, so we didn't need a fancy new way of thinking about the MVP to predict this voting;
  • Robinson Cano gets 2nd/3rd in 2010, with Joey Votto 2nd/1st in the NL.

The tops of the ballots end up looking pretty good, or at least defensible, in many years, but the system also results in a lot of aberrant ballots:

  • Of the thirty two league-seasons voted on by the system, eleven contained incomplete ballots because there were not ten players on playoff teams whose individual rWAR was greater than the team's margin into the playoffs.

  • Worse yet, the 2000 NL resulted in no qualifiers at all -- the Mets margin of eight games for the Wild Card was the closest race in the league that year.

  • There were, despite that, only a handful of times that the winner took home the award with fewer than seven rWAR -- Mike Piazza in 1995, Brady Anderson in 1996, Sammy Sosa in 1998, Darin Erstad in 2002, Mark Prior in 2003, and John Danks and Chase Utley in 2008. None finished below six rWAR.

    Danks did not even receive a Cy Young vote in 2008, and Utley finished 14th in the MVP, Mark Prior finished 9th in 2003, Darin Erstad did not place in 2002, and Brady Anderson was also 9th in 1996. On the other hand, Sammy Sosa did actually win in 1998, and Mike Piazza was fourth in 1995. I'd say that five ballots out of 32 with truly ugly first-place results in a better rate than the BBWAA has managed. (Hello six guys who voted for Dante Bichette in 1995.)

The major place that the ballots look funny is down past first and second place. There you see things like Mike Stanley and Steve Reed in 1995, Ismael Valdez in back to back years, Stan Javier in 1997, a Top Ten Cubs list in 1998 (literally), Matt Clement in 2003, all-Angels in 2004 (Robb Quinlan 10th -- no really), Jose Contreras in 2008, and all-Twins in 2009 (Joe Crede!).

It isn't clear that this model is actually useful, but the purpose of this was to be amusing, not predictive, so we shouldn't be surprised that, without doing any statistical testing, eyeballing the data says that it won't get us any closer to a predictive model for MVP voting than a simple list of WAR or any other naive system you might want to develop. (Any of which would probably be less cumbersome than this to cobble together, though I admit that I probably could have done this in 20 minutes in a database if I had that set up.)

What I hope is that my next implementation of PLAYOFFS+, which will grant credit for pushing a team toward the playoffs but not dismiss the contributions of players who just miss, might give us some more sensible (and possibly even predictive?) results.

1 comment:

Chris U. said...

I think the key is not necessarily making the playoffs but being in contention. A player who makes the difference between being in contention and not is very valuable to his franchise, because more meaningful games typically mean more fan interest.

I would define being in contention as making the playoffs OR being within X games of first place in the division or wildcard as of September first (or some other late-season arbitrary date). This will eliminate the players from teams that were never really in it but went on a winning streak after being eliminated (or conversely, finished the season fairly close but only due to the first place team resting down the stretch (see: 2000 Yankees)), while also including teams that outperformed for five months and were thus in contention but faded in September. How many games should X be? Whatever one could consider "striking distance" to be as of the selected date.