I had some free time this morning, so for kicks I ran a quick analysis looking at the projected correlations between a couple of Cal's early games and the rest of their season. Results after the jump.
|Fresno / Washington Result||W / W||W / L||L / W||L / L||Any|
|Total # Pairings||705||796||299||700||2500|
|Win % vs|
In other words, sweep Fresno and Washington and a bowl is likely. Split and it's possible but tough. Get swept and the season is pretty much lost. And yes, I know that Colorado comes before Washington... but that's Cal's easiest road game by far and almost certainly a must-win as it is. Washington is the one more like a toss-up, and as such the more interesting game to look at.
It's also interesting to note that with a win against Fresno, the Washington game becomes much closer to a tossup about 700 / 1500, over 45%). With a loss to Fresno, the Washington game becomes a much bigger hurdle (goes down to almost exactly 30%). So while both games are important, it's VERY important to not lose to Fresno in week one, because that says bad things about Cal's team and its chances in 2011.
On a more technical level, each additional win against the Fresno-Washington pair is worth about one expected win against everyone else. While there really isn't any data out there to really check that against, intuitively it makes some sense. An upset loss to Fresno could be because Cal is about to have a poor season, or it could be because Fresno is about to have a good season, or it could simply be a fluke result. The same story in reverse is true of a potential upset win at Washington. It makes sense that a win in an early season game would create optimism for the rest of the year, and it also makes sense that the effect of any single game on future expectations should be somewhat limited.
I should probably emphasize at this point that this is the first year I've been doing pre-season Compu-Picks projections, and as a consequence there are a lot of things that only time will tell if they're really working out or not (and the displayed correlations are a function of the underlying assumptions, which have only been back tested against historical data as opposed to really analyzed going forward, i.e. caveat emptor to a substantial degree at this point). But I thought that this was an interesting result, and given the occasional stats slant on the blog, figured that some people here would find interesting as well.