I have a question about techniques used for projecting the final standings. I happened to be reading an article at Fangraphs about the Yankees acquiring Wood that mentioned coolstandings.com. When I looked at their projections, I noticed that it was quite different from BP's projections. A third site called sportsclubstats.com has their own set of projections. They all seem to use Monte Carlo simulations, but they have different techniques. From what I can tell, BP uses the log5 method for predicting the outcome of a given game, where the winning percentage is derived from third order wins, while coolstandings samples from distributions of runs scored and runs allowed in their MC runs.
Which method should be more accurate? I would actually favor a combination of the two: use some sort of derived value for runs, and then sample from distributions. The reason I ask is that, as a Rockies fan, I felt that their season was pretty much over after the 2-9 road trip, followed by losing two straight to the Pirates. Out of curiosity, I looked at a binomial distribution for their chances of winning 90 games. At the time, they needed to go 40-20 to make it to 90 wins. I assumed that they are at least a .550 talent level team, and found a 2.5% chance of winning 66.7% of the remaining games. The part that surprised me was that the variance seemed smaller than I would have expected, which led me to conclude that a binomial distribution, while perhaps the simplest technique, wasn't necessarily appropriate.
Any feedback is much appreciated.