Quote:
Originally Posted by Ether
Mathematically speaking, it's an exact (and unique) solution to a least-squares problem.
Good description, but replace the phrase "match scores" with whatever alliance metric is being analyzed.
No iteration is necessary since the set of simultaneous equations is linear in the parameters. It can be solved directly without iteration by forming the normal equations for the system and then using matrix methods (this is how all the solvers I have seen here on CD do it). It can also be solved by using matrix methods directly on the design matrix without forming the normal equations. This latter method is numerically more stable and provides greater precision, but it is more computation-intensive and is overkill for this application.
For those interested in the math, a short explanation of the matrix math required to solve the normal equations is given here.
|
Thanks for clarifying where I was misleading - I know my knowledge of OPR calculations is somewhat lacking... I was hoping I could clear up some confusion and initiate someone like you, Bongle, or Ed Law clearing up where I was confused.
Thanks for all you guys bring to those of us who use and appreciate the stats but lack the software and statistical experience to generate them.
Nathan