Nobody who follows US politics is unaware of the accusations of bias in the pollster Rasmussen. Neither can anyone be unaware of the logic behind such accusations. Yet still, Scott Rasmussen’s apparently tireless work for the RNC continues to do its damage through the media who fail to understand how insidious manipulated polls can be.
Here’s a couple examples of Rasmussen polling’s evident blindness to reality.
● In the final weeks of the 2010 governor’s race in California, Rasmussen’s results suggested, using more polling “data” than anyone else, that the race was a toss-up. Right up to until Election Day, Rasmussen flooded the media with the view that Meg Whitman was going to pull off one of the greatest upsets in US voting history. So what actually happened when voters went to the polls? Whitman lost to Jerry Brown by 13 points. This is despite the $140,000,000 of her own money she used to lather the airwaves with advertising. This was not a “close race” or “tossup”—it was a slaughter.
● Rasmussen also had an apparently out-of-thin-air rosy view of Sharron Angle’s Senate candidacy the same year in Nevada. If he had set out to stir the chattering classes to accept the possibility of Harry Reid’s defeat, he sure got that to happen. Press organs from coast to coast were all eager to participate in the “scoop” of Harry Reid, the Majority Leader of the US Senate, losing his seat. Yet once again, Rasmussen’s polling proved to be entirely off the rails. Reid beat Angle 51% to 44%. Not even close.
So how is it that Rasmussen, with his dismal track record, continues to make such a stir in the media? Well, the problem lies with scientific poll trackers such as Nate Silver at 538.com. Silver will agonize in print over the possibility that Rasmussen is deliberately biased, yet Silver will continue to use all Rasmussen’s results in his aggregate polling figures, which throws everything out of balance. Here’s how the dirty work gets done.
Rasmussen’s polls, being automated “robocalls”, constitute a huge fraction of the aggregate polling data for any period of time. No outfit comes anywhere near producing the number of polls he does. So anyone doing an aggregate of all polls over a, say, one week period will find much more data from Rasmussen than anyone else.
Let’s look at a scenario. Let’s suppose there are four major pollsters producing data for an election cycle, Pollster A, B, C, and D. Pollster A produces two polls within a two week period leading up to an election, Pollsters B and C each produce one poll, and Pollster D produces five polls. Here’s a schematic of their results:
Candidate #1 Candidate #2
Pollster A 50% 47%
Pollster A 51% 47%
Pollster B 49% 48%
Pollster C 53% 43%
Pollster D 42% 49%
Pollster D 44% 50%
Pollster D 43% 50%
Pollster D 41% 49%
Pollster D 42% 48%
So even though three pollsters show Candidate #1 with a substantial lead (50.75% to 46.25%), an aggregate of all polling done, including the 5 polls from the only pollster who shows Candidate #1 losing, is that Candidate #2 is ahead by roughly 48% to 46%. Not only does this skew the actual support between the candidates, it causes Pollsters A, B, and C to wonder if their methodology is somehow wrong, and in many cases they adjust their future results to better conform to this aggregate figure.
The simple way to cure this “flood-polling bias” would be to take aggregate results from each pollster and only then aggregate the totals. The numbers would then look like this:
Candidate #1 Candidate #2
Pollster A 50.5% 47%
Pollster B 49% 48%
Pollster C 53% 43%
Pollster D 42% 48%
Then the true aggregate would be: Candidate #1 49%, Candidate #2 46.5%. The bias from Pollster D would still taint the overall result, but not so drastically as to forcefully misrepresent the actual relative strength of the two candidates.