Quote:
Originally Posted by Itse
In my limited experience, they do notably better than than corsi, and actually better than combined with corsi which was surprising.
I have not done in any way a statistically valid analysis, just toyed with them a little though. I encourage people to make their own experiments, especially since there's lots of ways to do this stuff.
The experience has made me personally highly sceptical of the usefulness of corsi for these predictions. I consider it to be like plus/minus for players; uninteresting / irrelevant on its own, but if you limit the context right and combine corsi with other information, it's a neat addition to what we already have.
It's not as good as zone starts, which is another stat that's started to gain attention at the same time as corsi. Zone starts puts a number on something that I at least have a really hard time tracking just by eyeballing it.
|
I'll stop with the thread derail after this but I checked last years standings to still better try and understand it all. The thing with using goal differential, and forgive me if I'm doing a bad job explaining this, is wining percentage is a dependent event. The team that scores more than it's opposition wins more. But since they're directly related it doesn't necessarily tell the whole story. A few bounces can win you the game and it'll increase your goal differential at the same time.
I started by looking at the 5v5 CF%, Goal Differential and Actual standings at thanksgiving last year and compared them to the final standings at the end of the season. What I found for last year is that goal differential and the actual standings are in fact the best way at predicting the final standings. Goal differential was within 4.6 positions and the standings were withing 4.7. My standings are only ranked on points and don't include tie breakers at this moment so I'd call that difference statistically insignificant. Corsi however was off by 7.3 - far worse.
I then thought that it's not truly fair to use that as a predictor since the results to date, approximately 25%, are influencing the final results. So I decided to see how the predictors (5v5 CF%, Goal Diff, Standings) predict the rest of the season. It turns out none of them are very good. Goal Diff and Standings drop to 6 and 6.1 respectively and Corsi is 7.8.
What's interesting is the biggest outlier of all of them were the Flames last year and particularly using goal diff as a predictor. Their goal diff was 30th at thanksgiving. After thanksgiving they were the 8th best team. Another interesting note was last years Oilers. At thanksgiving they were the 7th best CF%, 7th best goal differential and 7th in the standings. After thanksgiving they were the 7th best team and they finished 8th overall.
TL;DR - the current standings are actually the best predictor of the final standings. How boring.