Most Pollsters, Other Than Rasmussen, Which Is In Effect A Part of The GOP Propaganda Machine, Did Pretty Well This Cycle

Yesterday conservative Democrat Bob Etheridge conceded to reactionary psychopath Renee Ellmers, leaving just 5 House races still undecided. So although we don't know exactly how many seats the Democrats and Republicans won and lost, we do now know which pollsters did best and worst. Anzalone Liszt Research is the Blue Dog-oriented polling firm. Their House candidates for this cycle were winners Mike Quigley (IL-05), Larry Kissell (NC-08), Leonard Boswell (IA-03), Mike Doyle (PA-14), Heath Shuler (NC-11), Jason Altmire (PA-04), Luis Gutierrez (IL-04), Albio Sires (NJ-13) and Bob Brady (PA-01); plus losers Bobby Bright (AL-02), Debbie Halvorson (IL-11), Dina Titus (NV-03), Travis Childers (MS-01), John Salazar (CO-03) and Ron Klein (FL-22). They also consulted two House Members who attempted to win Senate seats, Charlie Melancon (LA-03) and Paul Hodes (NH-02). Like many of us, they asked the question, "how accurate were the polls?" Here's their report:

The benefit of elections is that pollsters are afforded the opportunity to assess the quality and accuracy of their polls. In this respect, 2010 was pretty good year for the polling community overall, with a few notable exceptions.

Looking at the national generic congressional ballot, many of the top public pollsters were very close to hitting both the final exit poll ballot margin (R+7) and the GOP percentage of the two-party vote (53%). This includes telephone surveys from Pew (R+6, 53%), NBC/WSJ (R+6, 53%), CBS/NYT (R+6, 53%), Ipsos/Reuters (R+6, 53%) and two internet-based surveys by YouGov/Polimetrix (R+7, 54%) and Zogby (R+5, 53%). A few polls overstated GOP performance, including CNN/ORC (R+10, 55%), Rasmussen (R+12, 57%), and Fox/Opinion Dynamics (R+13, 57%). The most notable and surprising outlier, however, was USA Today/Gallup, which showed Republicans with a whopping 15-point edge and 58% of the two-party vote. That would have translated into something on the order of a 90-seat pickup for the GOP, well above the projected 64-seat net pickup that the GOP will show when undecided races are called.

We noted earlier in the cycle that a pollster's choice of methodology has a clear effect on the numbers, and this was indeed the case with the top public polls in 2010. According to data complied by Mark Blumenthal at Pollster.com, the average predicted margin for surveys that sampled both landline and cell phones was R+6, compared to R+10 for surveys that sampled only landline phones. In other words, polls that included cell phones were more accurate.

The difference between phone, internet, and automated ("robo") surveys was less pronounced overall, with some exceptions. At the national level, the final Pollster.com trendline for the generic congressional ballot margin was R+7 for both telephone surveys (landline-only and landline/cell) and internet surveys, and R+8 for automated surveys. At the state level, the three most accurate polling firms each employed a different calling mode: Quinnipiac (phone, 21 polls, 3.3 points of average error); SurveyUSA (automated, 30 polls, 3.5 points), and YouGov (internet, 35 polls, 3.5 points). Rasmussen, by far the most prolific state-level pollster, was also by far the most inaccurate (105 polls, 5.8 points of average error), according to FiveThirtyEight.com.

Additionally, there was some divergence between polls released by Democratic and Republican firms this cycle. On balance, Democratic firms were more accurate at the Senate level, while Republican firms were more accurate at the House district level. At the Senate level, Democratic firms exhibited a fairly small 3.9-point average bias in favor of the Democratic candidate, compared to a 6-point bias in favor of the Republican candidate among GOP firms. At the House level, however, Republican firms showed 1 point of bias for GOP candidates compared to 9.7 points for Democratic firms-though again, some individual firms bucked this trend. This includes Anzalone Liszt Research, which was the second-most accurate among Democratic polling firms that released more than five polls this cycle.

And speaking of polls, a new one from Quinnipiac finds that 58% of Americans want to see Don't Ask, Don't Tell repealed, as opposed to 34% who don't. Among military families, 55% want to see repeal and 38% oppose repeal. Perhaps those findings have made Republicans Lisa Murkowski and John Ensign feel safe enough to say they would not vote against repealing DADT.