Yes, I've kept at it. Here's what's new compared to the version I put out this morning:
- Home/Away correction. There's a big difference between teams playing at home and on the road; the variance between home and away performances averaging 0.5 goals scored per game. It's now accounted for.
- Point expectancy. We don't care that much about wins in the league, we care about points. Multiply win expectancy by three and add in draw expectancy, and suddenly you have a measure of points expected in a game. You can also apply this to individual events, thus determining exactly how many points a given goal or red card was worth.
- More transparency. Included with every graph is all the inputs used to derive it, making it clear exactly what went into constructing the charts.
Let's take a look at one of the craziest games in Stamford Bridge's history after the jump.
Figure 1: Aston Villa at Chelsea, December 26th 2007. Click for full size.
I don't even know where to begin describing this game, but it had everything. Goals, sendings off, lead changes - if you went to stress-test a win/point expectancy chart, this is the game to do it on. So, what does everyone think? What can be improved for version 3.1? It's simple to just extract one graph or the other rather than using the whole giant block, so I think I'll use the point expectancy chart in recaps and link it to the big multi-graph page.