You're going to give us your professional opinion on whether this algorithm is up to snuff?
That’s impossible unless I can see specific “factors”, coefficients, and weights. Ideally the algorithms. Like an auto-regression?
Most importantly, I think based on what people posted, I get the gist of how F+ calculates power rating. I’ve skimmed and not yet deep dived…but would also like to view its predictive power, if any.
FIFA openly publishes its ELO system. Obviously it’s easier to assess. But it is sometimes really off in using it to predict wins.
My gut is F+ could be useful as a power rating. I’m not sure however on the part the measures offensive efficiency. How well does an offense move vs a defense. Somebody noted a long drive with a fumble loss at the 1 yard line is inefficient?
That is debatable. Sounds like a pretty sophisticated model overall. But I’ve seen extremely complicated models do no better than simpler ones. Reducing factors is actually called Principal Component Analysis (PCA).
For example, you have an age, weight, height, favorite color, NCAA football team, etc. Let’s predict your earnings for the next year. PCA reduces factors like favorite color that the data will show to not affect earnings like your education can.
Same in football. Is a made up factor like drive efficiency meaningful? Even if so, how heavily should it be weighed vs other principal factors. All this by the way is for the most part computed by drawing a line that best fits predicting a dependent variable, like a game result, based on many independent variables…all plotted on a graph.
Even neural networks use weights at nodes. Indicating which path to take. It’s all based obviously on statistical scoring. Again I ultimately need a deeper look.
Like I said I’m skeptical of many of these models to be highly predictive when the phenomena is too complicated…and when it’s a tight prediction, like ND vs Army as opposed to ND vs Penn State. Kinda common sense. Just computes odds, as we all know.
