Nate Silvers has mathematically shown the reliability of polling organizations. It turns out Rasmussen is but a mediocre poll skewed to Republicans while Research 2000 polling has been much more accurate. In that light while President Obama’s Approval Rating under Rasmussen is under 50%, Research 2000 has it at 54%. Who should on believe? Do you need to ask?
Blast from Rasmussen’s Past
by Nate Silver @ 2:23 PM
Have to be briefer than I’d like here, but I’m in the midst of putting together a new set of pollster ratings. In addition to populating the database with 2008 General Election results (as well as contests held in 2009 and 2010), I’m also doing a fair amount of backfilling to try to make the database, which goes back to 2000, more comprehensive. Previously, for instance, we did not have Rasmussen’s polling in place for 2000, which can be hard to come by. But thanks to the magic of The Internet (as well as a tip from Chris Bowers) I was able to track it down.
Here are the results of statewide polling in the 2000 Presidential Election. Although the pollster ratings will eventually involve some fancier math, the numbers you see below are as simple as it gets: I’ve simply looked at the average error of the last poll issued by each pollster in each state in terms of projecting the margin between Bush and Gore. (A cut-off is established 21 days prior to the election.)
While 2000 was generally a fairly rough year for pollsters, who had to deal with an unenthusiastic electorate, some third-party challengers, and some late-breaking developments like Bush’s DUI charge, Rasmussen was the worst of the lot, missing by an average of 5.7 points. They also called 7 states wrong.** Some of this was the result of bias, as they were 3.5 points too high on Bush’s margin in the states they surveyed, on average.
Although Rasmussen has certainly gotten a lot of things right, their high pollster rating was mostly based on their strong performance in 2004 and 2006. Their rating is likely to go down now that we’ve found their 2000 data, and are adding in the 2008 data, when their performance was mediocre.
** The figure for ‘miscalls’ in the chart assigns half a point when the pollster’s final survey has the race in the state exactly tied.
p.s. If anyone happens to have Harris Interactive statewide polling for 2000, which seems to have disappeared from the Internet, please drop me an e-mail.
