So, let's say you put together a roster of 25 players, according to a particular set of guidelines. (For purposes of this thought experiment, let's call you...DJL44.) You look over the roster you've assembled and think, "Man, this group of players could totally win the pennant!"
But then some other guy comes along. (Let's call him...me.) He says, "Dude, what are you smoking? That team would be lucky to break .500."
What's your response? Do you have a response that doesn't ultimately boil down to "nuh-uh"?Well, to some degree there isn't any certain way short of buying an MLB franchise, making all the trades, and then playing the season to see if you're right. But let's say you don't need 100% incontrovertable proof; you just need something defensible that isn't just, "They're awesome because I put them together and I'm awesome."
Most of us would listen to numbers.
If you had a set of numbers that showed that your club was at least as likely as not to win, say, 93 games, with 93 games being a pretty baseline division-winning total these days, I'd say you'd have made your point -- you've got a team that could, and given their composition probably would be expected to, reach the playoffs.
So what numbers do you use? If, like me, you decide to go in pursuit of some numbers, you find that a lot of them have...challenges. Issues that make them less than ideal as a system of evidence. Let's say you decide to use:
1. Projection systems
You want to know how good a player is likely to be in 2011? We've got access to a lot of projection systems that map out the contours of what a player's expected 2011 season will be. Pick the one that seems most accurate and go with it, right?
Not so fast. Projection systems are actually pretty darned good at telling you what a players numbers will likely look like in 2011 -- even Marcel the Monkey gets credit for over 70% accuracy when projecting stats. But what the projection systems don't generally do is tell you how valuable that stat line is going to be. Say you're looking at a player and trying to figure out if his .293, 12 HR, 81 RBI projection is going to help you win a playoff spot. Is he a second-baseman or a DH? Is he Joe Mauer? Is he a Yankee, where those numbers are just going to get lost in the crowd? Is he playing in Fenway or Petco? Is he batting second or eighth?
OK, that's probably overkill -- even most value-based systems don't differenciate between batting order positions. But the point still stands: how do you really know how much value, in terms of wins, that player is going to bring to your team? Based solely on projection systems, you really don't. So you move into the field of value stats, and find:
2. Wins Above Replacement (WAR)
These days, everybody loves WAR. Sportswriters are using it in their articles, baseball color guys are mentioning it during broadcasts. It seems like WAR has the potential to become that short-hand, single-number summation of a player's full value contribution to his team. Except...well, I won't get into the argument over how accurate or useful WAR is when looking at players, but I will ask this question:
If you took every player on every team in MLB last year and added their WAR scores together based on the team they played for when assembling that WAR score, which team would finish with the highest 'team WAR'? Philadeplhia, right? After all, they led the majors in wins.
Well, let's do the math (according to FanGraphs, since baseballreference doesn't seem to sum their WAR totals by team):
Philadelphia had 23.6 batting WAR, 18.4 pitching WAR, and...well we don't know what their fielding WAR was but as a team they finished with negative UZR so it can't have added much to their win total. That's a total of 42 WAR which, when added to the baseline definition of replacement level (a team with 0 WAR should win 48 games) totals 90. OK, a bit shy of their actual 97 win season, but not too shabby, right?
The problem is that the Phillies finished 14th in baseball in batting WAR and 7th in pitching WAR. The Giants, Twins (!), and Red Sox (!!) finished ahead of the Phillies in both categories, and two of those clubs even had more UZR runs than did Philly. So by this measure not only were the Phillies not better than the team that beat them in the NLCS (which makes sense), but they weren't as good as a team that didn't even get out of the Division Series (less sense) or one that didn't even make the post-season (not much sense).
There's some utility here, but there's also too much wiggle-room for WAR to be a convincing statistic in favor of our argument.
3. Win Probability Added
Yes, I know that WPA isn't intended to be a measure of player value, but rather just a nice way of seeing who had a good year (even if some folks used to write about it a lot). But WPA is team-normalized -- for each team that finished with a positive WPA after a game, a team finishes with the same negative WPA. Perhaps WPA has a value at the team level that it doesn't have at the individual level?
Short answer: yes, but not in any way you can actually use.
Longer answer: WPA does do a great job of telling you who won games -- the Phillies and Rays lead in team WPA, in that order, as you'd expect. And superficially, WPA even eerily predicts wins in that the Phillies finished with a team WPA of just over 16 (which when added to 81 wins gives you their actual total of 97) while the Rays finished with exactly 15 team WPA (which results in their total of 96).
The problem with WPA is that, as awesome as it seems to be as a team measurement tool, it's almost impossible to predict what a given player's WPA will be in advence. Case in point: Here are Michael Cuddyer's WPA totals from the last five seasons (which you can actually get from baseballreference.com):
2.6, 0.6, 0.0, 0.7, -0.8
Any idea what Cuddy might do this year? Me neither.
Ultimately, I boiled it down to a measure that actually works both ways we need it to:
4. Win Shares
The stat is derived starting from the actual wins of the player's team, so when you add the win shares back up again, you get actual wins. Also, offensive win shares, at least, are as consistent as runs created, which is the primary way offensive win shares are calculated. (Pitching and defensive win shares are a bit more hazy, but for our purposes, they work as well as WAR, which is all we need; we're just trying to come up with a believable level of proof, not dead certainty.)
So the goal is to come up with a team that's expected to earn at least 279 win shares (93 wins * 3). As an added hurdle, I'll limit myself to just American League players. Let's see if I can put it together before the All Star break...