Saturday, April 14, 2007

Abstract statistical analysis in action

This post is about video games, baseball, video game baseball, and how much of a dork I truly am. If that bothers you, you should skip this one.

I've been rereading one of my favorite baseball books recently, Baseball Between the Numbers, a hefty tome regarding statistical analysis and its use in modern major league baseball. Essentially, a team of writers invent roughly eight thousands new stats (and this is baseball, a sport already drowning in stats) to explain various aspects of the game. In several chapters, they take conventional baseball wisdom and use their shiny new stats to chuck it into the toilet.

According to them, these new strategies would help teams win games. I've decided to test two of those strategies, with the help of MLB 2K7, on the PS2. I'm going to run a season with the Astros according to Baseball Between the Numbers methodology. We'll see how it works.

I'm not going to play every single game – that would take forever, and would also distort the results, since the game is often depressingly easy, even on its hardest difficulty setting. Instead, I'll use the Manage option for each game – I'll make all of Phil Garner's decisions, while the computer will simulate the game itself. This is perfect, because the two theories I'm testing are managerial decisions.

Theory 1: Batting order doesn't matter. They do exhaustive research and number crunching and prove that, at best, the difference between an optimal lineup and a least optimal lineup gets is two or three wins a year. Their choice for an optimal lineup? The nine hitters ranked in descending order of on-base percentage. The higher you are in the order, the more at-bats you get; the higher your OBP, the more you get on base; the more runners you get on base, the more runs you score; the more runs you score, the more games you win.

So: this season, I'll send the Astros out to bat in order of their OBP. I'll start of the season with last year's number, and then adjust it periodically. (I'll leave it the way it is for the first ten games of the season, to give the stats a chance to develop, then start shifting after each game.)

In descending order of on-base percentage, then, here is your counter-intuitive Astros starting lineup.

  1. Lance Berkman, 1B (.421)
  2. Luke Scott, RF (.418)
  3. Morgan Ensberg, 3B (.393)
  4. Carlos Lee, LF (.359)
  5. Chris Burke, CF (.326)
  6. Brad Ausmus, C (.302)
  7. Craig Biggio, 2B (.298)
  8. Adam Everett, SS (.286)
  9. Pitcher's spot

Weird, huh? Berkman certainly isn't your prototypical leadoff man. A couple things: I thought about using career OBP, rather than last year's, but decided against it because of Craig Biggio. His career numbers are skewed by his earlier, better seasons, and his .354 career OBP is clearly not representative of his current ability. Also, it's very possible that Brad Ausmus's surprisingly high OBP is a result of frequently batting in front of either the pitcher or Adam Everett – if the guy hitting behind you is an automatic out, you're far more likely to draw walks. His batting average last year was only .230, which would seem to prove this. But the theory is the theory, so he remains in the six spot.

Theory 2: Only using your closer in a save situation is idiotic, and a waste of your relief ace. The theory says that the most important pitching situation in a game is rarely in the ninth inning – it might happen in the eighth, or the seventh, or even the sixth. Since managers (and closers, too) are stuck in the Saves mindset, inferior pitchers will be used in these situations, and cost you games.

So Brad Lidge (oh, man, I can hear you grumbling and grinding your teeth when I write his name – listen, listen: Brad Lidge. Ooh, that's creepy.) will be used earlier and more often in non-save situations. I'll spare you the deluge of numbers on how and why he'll be used, but there's a giant two-and-a-half page chart (pages 66-68) that explains exactly when he should be brought in, depending on the inning, the outs, the runners on, and the score. Also, it varies depending on when he pitched last, and if there's a game tomorrow.

I know what you're saying ('cause I can hear it, remember): "Brad Lidge sucks!" Certainly seems that way. But the computer-simulated MLB 2K7 Lidge isn't quite as ineffectual as his real-life counterpart, so don't count him down yet. And if it doesn't work out, I can always trade him.

Other than that, I'll try to adhere as closely as possible to standard baseball logic and strategy, at least as I interpret it. Will it be an interesting experiment? Or a boring wankathon? Your guess is as good as mine. (I'm guessing wankathon.)

(There's another theory I badly wanted to test: that a four-man starting rotation is better than a five-man, if properly used and monitored. Unfortunately, the video game is designed for a five-man rotation, and starters are pretty much required to get four days of rest before they can be used again. Oh well.)

The first four games

The real Astros were beaten in each of the four games of the year. We'll see how the fake Astros do.

I check the Team Pulse before starting the season to check my team's morale with the new lineup. Everyone seems happy, except, hilariously, Brad Ausmus, who's been moved up in the order. (He probably sees through his high OBP, too.)

Berkman makes me look a genius by walking to lead off the bottom of the first against the Pirates, but the next three guys go in order and make me look stupid again. And I continue to look dumber as the game goes on, as the Astros can't get anybody on base at all. I end up losing 5-0, only collecting five hits – and four of them are by the bottom third of my order. (Berkman and Lee each drew a walk, as did Ausmus and Everett.)

The second game does me no favors, either, as my bullpen collapses in spectacular fashion: Chad Qualls gives up six runs in one inning of work, tanking the game completely. Following the book's rules for use of the closer (he hasn't worked in more than five days, and there is an off-day in the next two days), I bring in Lidge for the ninth; he does exactly what you think he would do, giving up a run and four hits. Not that it matters, because I lose 11-4. The Pirates collected 23 hits. That's bad, obviously, but I managed 10, plus four more walks and two hit batsmen – the Astros are certainly putting people on base, they're just having trouble getting them home.

(At this point, I'm interrupted by a phone call from FRINAN. I share my experiment with him, and he informs me of another problem with it: I'm playing as the Astros. Too true, too true.)

Game three is even uglier, somehow. I finally get the bats working, with seven runs and 19 hits, but my pitching is, um, shall we say, god-awful. I lose 15-7. To the Pittsburgh Pirates.

And then I lose in unceremonious fashion to the Cardinals, 6-0. I've been shutout twice in four games, and never even been close to winning any of them. Good work, Baseball Between the Numbers!

More on this experiment as it continues. If it continues. It's kind of frustrating.

No comments: