tag:blogger.com,1999:blog-6922495112008511600.post3069639349632947719..comments2014-10-18T13:03:51.031-04:00Comments on Dave in Fla's Poll Analysis: Dave inFlanoreply@blogger.comBlogger3125tag:blogger.com,1999:blog-6922495112008511600.post-57096229769462053302012-11-03T18:51:56.425-04:002012-11-03T18:51:56.425-04:00I finally spotted this item on your blog, and had ...I finally spotted this item on your blog, and had a chance to review it in detail. I can follow what unskewedpolls.com, which you reference, is doing, but I don't see where your own formula comes from, how you can make your adjustments using just one party affiliation, and how your stated assumptions come into it.<br /><br />Taking the example from unskewedpolls.com, it seems like a better assumption is that 90% of R voters will go Romney, while 90% of D voters will go for Obama. If we let "R" be the percentage of voters that are Republican, and likewise for "D", then Romney would get .9*R+.1*D votes from affiliated voters, while Obama would get .1*R+.9*D. The difference comes out to 0.8*(R-D). If we define the turnout differential, T=R-D, (the difference between the Republicans that vote and the Democrats that vote), that formula comes down to 0.8*T.<br /><br />Anyone could use that formula to make a first-cut at "adjusting" a skewed poll. If a poll is "D+10", for example, but you think the turnout will be "D+0", you could simply subtract 0.8*10=8.0 from the poll's published result.<br /><br />But what about Independents? It seems clear that they are about 30% of the voters. That means that the contribution to the final result due to independent is 0.3*dI, where "dI" is the difference between the percentage of Independent voters going for Romney, minus the percentage going for Obama.<br /><br />Putting it altogether, the voting result (making the 90% assumption above) is V = 0.8*T + 0.3*dI. For example, if you go with a voter mix of R/D/I of 36/34/30 (Roughly what Ramussen says it is), and the Romney/Obama split among Independents was 55/45, you would get V = 0.8*(36-34)+0.3*(55-45) = 0.8*2 +0.3*10 = 1.6+3 = +4.6 (favoring Romney).<br /><br />It seems to me that the way to do this would be to first estimate dI, by calculating dI = (Poll_result-0.8*T_poll)/(fraction of independents in poll), where T_poll is the turnout reflected in the poll (the negative of the "D+" number). Basically, you just need the split among Independents. If the the internals give you that more directly, that would be best (rather than doing the above).<br /><br />Then calculate V = 0.8*T + 0.3*dI (where, again, T=R-D).<br /><br />Anyway, this formula shows the importance of GOTV. Improving your turnout by 1% gives you +0.8%, while improving your split among independents by 1% only gives you +0.3%.<br /><br />-OptimizerAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6922495112008511600.post-10847538419199932772012-10-28T10:45:39.300-04:002012-10-28T10:45:39.300-04:00Neat!Neat!susieq2cutehttp://www.blogger.com/profile/12497231846355866149noreply@blogger.comtag:blogger.com,1999:blog-6922495112008511600.post-69961959636839509522012-09-27T08:55:00.097-04:002012-09-27T08:55:00.097-04:00COOL!!!COOL!!!Erichttp://www.blogger.com/profile/09671465601763164161noreply@blogger.com