Originally Posted by
Wrecktangle
Miz, if the only thing you really are looking to do is play with least squares, you need more data fields than what these "packages" will give you. You need multiple factors so you can get used to understanding which will come in and which go out according to test criteria. For score prediction you will always find the single best predictor of today's game will be today's line, bar none. So unless you only want a toy regression project, you need at least a short boxscore with, for example in football, 1st downs, yds rushing & passing fumbles, penalties, TO (both types), etc.
If nothing else, scrap one of the known sites for several weeks.