Skiace
Member
The bold critique in particular doesn't make much sense to me. Silver gives every poll a weighting in his averaging, based on a number of factors that could be considered "quality of the data." Such factors include the history of the given polling firm (accuracy as well as precision), the sample size of the poll, whether the poll used live operators or robo-calls, whether the poll contacted cell phones or only land-lines, how recent the poll was, whether it took place all on one day or over several, etc. He has written blog posts about all of these factors.Tony Crocker":24yx4ypw said:Harold doesn't think much of Nate Silver. His political background was as an Obama volunteer in the primary season of 2008. He has never actually worked in political polling. He puts poll number results into a computer without adequately analyzing the quality of the data, then accepts average or aggregated results. Thinking you can aggregate national numbers from state polls doesn't work if the state polls weren't done properly.
Competent polling methodology is expensive. Questions have to be designed carefully in both content and sequence to minimize bias. Response rates have declined steadily with unlisted numbers, cell phones, etc. Now we have more early voting as an additional complication. Time and money saving shortcuts can easily introduce bias, particularly with the low response rates. National organizations like Gallup have more resources to try to improve accuracy.
Harold says Nate Silver does not know his political history well, has made numerous false statements about elections before his time. As in the financial arena it helps if your historical perspective extends back 30 years, not just to 2008. Harold worked full time in political polling design and analysis from ~1976-1981.
Today Silver says Romney's probability of winning is 21%. Harold thinks it's at least 50%.
It makes sense that averaging many many polls from different polling firms and applying some moderate weighting to them ought to produce as good a prediction as anything else, and certainly better than any single polling firm can do on their own.
But his track-record speaks for itself thus far. 49 states correct in 2008 and what looks to be 50 states correct in 2012. He also called the 2010 GOP house takeover, though he was fairly late to make that call. I'm sure he will post a summary in the coming days or weeks assessing his own accuracy as well as other poll aggregation sites.
For now, here's Silver's predictions of the seven states that NYT called "toss-ups" and how they compare to results currently in:
(It should be noted that he had Florida as a Romney lean for weeks, but flipped it to just barely Obama-lean (50.3% chance) in the last day or two.)
Code:
State Silver Actual %reporting
CO 50.8-48.3 51-47 94
FL 49.8-49.8 50-49 97
IO 51.1-47.9 52-47 99
NH 51.4-47.9 52-47 99
OH 51.3-47.7 50-48 90
VA 50.7-48.7 51-48 99
WI 52.4-46.9 53-46 99