Has anyone tried to analyse voting blocs from polling data? (important since Democratic Caucases require a candidate to get at least 15% in a district to count). I used Samplemiser to produce daily figures in Iowa for each of the candidates from late September to Novemeber 29 and then correlated candidates shares to find which candidates had the strongest negative correlation with each other (a negative correlations implies that their supporters are more likely to move between each other):
I found:
Biden voters ---> Obama
Richardson ---> Clinton
Clinton ---> Edwards & Obama
Edwards ----> Clinton
Obama ----> Clinton
Undecided & Fringe ----> Edwards
Assuming that 25% of Clinton, Obama and Edwards voters, 50% of Biden and Richardson voters and all undecided/fringe are forced to move. I project a Clinton victory if the election were held today (although Obama is projected a lead in the polls).
For more info click the link:
http://thepoliticaltipster.wordpress.com/2007/12/03/obama-leads-by-174/nb: Mods, if you think that this should go somewhere else feel free to move.