Thanks. This is really helpful to know :-). I haven’t run SRS yet as a system (as I just happen to have got into least squares, and weighted least squares more easily – also my initial go at SRS didn’t have things converging). I’m just keen to try various rating systems (I like putting a number on things – it works for me it’s nice to be able to back up a feeling). I just wondered if you’d done a normal probability or QQ plot against the error from the SRS (e.g. is it normally distributed – you’d need a couple of seasons of data for this).

I have all the data for the FBS so I’m going to have to give this a go (I’ll probably try capped and uncapped as it will be easy enough to do but I think capping is right for College Football because of the issues with teams like Oregon – taking the first team off in the 2nd Quarter etc. – or some of the complete mismatches such as the Oklahoma State game early in the season). Thanks for the advice.

Don’t think capping would cause any problems. And if I’m understanding you correctly, this isn’t that hard.

The SRS got one game perfect. It had Maryland at 37.3 and NCST at 42.3, so projected the Wolfpack to win by 2 at Maryland, which is exactly what happened.

It was most off on Kansas State/West Virginia, with the expected rsult being K-State winning by 6.1 points, not 41 points.

Of the 56 games at the FBS level this weekend, the average margin between the actual difference and the SRS projections was 12.1 points. The standard deviation was 9.1.