From: near Wash State University
The two SIDES formulas that come with the demo produce good results... but they are HIGHLY variable. For example, DANDYON predicts about 62% ATS for 1997. But now change the rolling average from 3 weeks to 4 or 2 or 1 or 5 or 6 weeks. The thing totally falls apart.
What this tells me is that at the setting of 3 WEEKS is an ABERATION of the numbers which cannot be predicted a priori.
Would love to have the freedom to allow a computer program to allow more "non-linear" and "multi-variant" analysis. Having 30 variables which are all algebreically summed with various weighting factors is not a "multi-variant" analysis. While good, it's not enough... you can't scale for BLOWOUTS, Expected Margins of Victory vs. Actual Margins of Victory and all other kinds of popular concepts.
I'd love to see more models posted here. Thanks.