I just came across a paper on LARS, the linear model selection algorithm that's sweeping the nation. The mathematically and/or masochistically inclined may view it here.*
Ok, so it's not quite that popular, but it is being heralded as one of the biggest advances in linear modelling in a few decades - and that's saying a lot for a field that has been relatively unchanged for almost two centuries. Essentially, it's a new take on model selection, or determining which independent variables should be included when modeling a response variable.
Oh, and the lasso is a form of restricted regression model which is quite compatible with the LARS algorithm.
I'm going to stop writing now.
*The 2004 article published in the Annals of Statistics is here, but it has blurry grayscale images. My link is to a 2003 version, but hey - it's in color!
{ 1 comment… read it below or add one }
We used this technique to great effect to determine the most critical factors for user retention at a large social networking site.