A number of recent conversations on this site (as well as Ducksnorts) have concerned the notion of the "Expected Win-Loss" (X W-L) differential, specifically in its capacity to predict the "Actual Win-Loss" (W-L) differential for any given team.  The formula itself takes into consideration how many runs a team has scored (Runs Scored, or RS), as well as how many it has given up (Runs Allowed, or RA).  The key assumption is that a team's W-L record is significantly correlated with its RS-RA differential.  Hence, a team that scores a lot of runs and gives up few should have a better record than a team that scores few runs and gives up a ton.  I doubt too many people would have a problem with this, especially given a large enough sample size (as Dex noted in a recent post).  In fact, I found out that at the end of the 2006 season the W-L for the league was very highly correlated with the X W-L numbers (r = .91).  Again, no big surprise.

The problem as I see it is that (1) the season is only 162 games, and (2) the division races are usually very tight (last year, 4 of the 6 division races were decided by 4 games or less).  In this case, over the span of 162 games, you would hope that the X W-L formula would have a margin of error less than that, right?  Well, as it turns out, if you look at the distribution of actual-predicted wins for the league in 2006, you would observe a standard deviation of over +- 4 games, which essentially means that only about 68% of the league would come within about +- 4 wins of their predicted totals (assuming a normal distribution - in reality, 53% of the teams came within 3 or fewer wins, and 73% came within 4 or fewer).  Roughly 95% of the league came within +- 8 wins.

Furthermore, the X W-L formula only predicted 2 of the 6 division winners.  And this is after the season has finished!  I can't even imagine what its accuracy was at the All-Star Break or the July 31st trade deadline when a GM might actually need this formula.

So what does all of this mean?  Well, while X W-L is certainly correlated to a high degree with W-L (no shit?), its value in the grand scheme of things is questionable at best.  I don't see how it could play a significant role in whether or not you pick up that "big bat" you've been eyeing, or in an evaluation of the "depth" of your bullpen.  As I see it, there's always room for improvement.