We humbly suggest Silver doth protest too much. For the fundamental issue of polling data is not so much the statistical vagaries introduced into the polling method as they are the tendency to over stretch polling model purpose, highlighted by the increasing challenge of predicting turn-out.
Silver spends a great deal of time denigrating Rasmussen methodology. Never mind that he has outperformed more established pollsters in past elections. The simple fact is that there are challenging and downright incriminating statistical issues with building accurate polls regardless of the approach.
The largest structural challenge in using polling data as a predictive model as opposed to a wind socket is estimating voter turn-out. Rasmussen, perhaps the most active pollster in tweaking its model for this variable, subjects himself to extreme success or failure in this regard. That doesn't make him a bad pollster.
click to enlarge, double click to reduce
The fact remains that the Democratic ground machine increasingly demonstrates their ability to affect races even in highly charged elections. Would anyone dispute that Angle + Dem organizers = Nevada win by 5%+ going away? Let us see the polling numbers that dispute that, Mr. Silver. Without them, it is difficult to poke a hole too far into Rasmussen's numbers.
Until the Tea Parties bolster their emotion and ideology with tactical muscle on the ground, they will continue to under-perform. If anything, Rasmussen and other pollster data over estimated turn-out in their models, and implicitly under-estimated the ability of the Democratic machine to affect key races.
That a few polling models are vanilla enough to both over-estimate these effects in the previous election, and under-estimate them in this one, does not vindicate their prowess. It just makes them vanilla during a changing political landscape.
No comments:
Post a Comment