Every presidential election cycle, the pollsters selling their product to a predominantly Democrat news media play the same games.
Most polls before the party conventions survey non-voting adults or registered rather than likely voters with enormous 6% to 13% over-counts of Democrats over Republicans.
After the conventions, these polls start erecting likely voter screens, which still often over-count Democrats, but not as dramatically as during the preceding Spring and Summer.
Then, in order to maintain some semblance of credibility, these polls employ fairly realistic likely voters screens and partisan weighting so their final survey results come close to the final election results.
And in every presidential election cycle, Democrat consumers of this polling product get a false sense of confidence like Charlie Brown expecting that Lucy will hold that football for him, while Republican consumers needlessly fret and worry. For those who prefer to live in a reality-based world, polling is fairly easy to correct with a simple back of the envelope calculation:
1) Go to the very beginning or end of the poll’s news release and look for the percentage of Democrats, Republicans and Independents surveyed in the poll. This is the partisan weighting used by the poll. (Sometimes the poll only gives the total number of respondents and then the numbers of Democrats, Republicans and Independents. You will have to derive your own percentages from those numbers.)
2) Employ a realistic partisan weighting for 2012 voters:
For a Democrat leaning result, use the historical Edison election day exit polling since 2004.
2004: 37% Dem / 37% GOP / 26% Ind
2006: 38% Dem / 36% GOP / 26% Ind
2008: 39% Dem / 32% GOP / 29% Ind
2010: 36% Dem / 36% GOP / 27% Ind
For this model, I assume that:
1. A percentage point of GOP voters left the party after 2004 and became Independents.
2. The remaining GOP voters are as enthusiastic as and the Democrats are less enthusiastic than in 2004.
Applying these assumptions to the historical Edison exit results, I am seeing a partisan turnout very much like 2010 – 36% Democrat, 36% Republican and 27% Independent.
For a GOP leaning result, use the Rasmussen rolling polling of thousands of Americans asking whether they are registered Democrats, Republicans or Independents, which currently finds that registered voters are 35.4% Republican, 34% Democrat, and 30.5% Independent. This result conforms with other findings that Democrats have been leaving the party during the Obama administration and becoming Independents.
3) Reweight the poll results.
Assume all GOP voters are supporting Romney, all Democrat voters are supporting Obama and Independents are split evenly.
While there is a small percentage of partisans who will vote for the opposing candidate, Democrat and GOP voters overwhelmingly support their own nominees. The support for the opposing candidates is basically a wash and is too small to be of concern for this simple reweighting.
The Independents are the real swing voters in this election. Although recent polling suggests that Independents prefer Romney by roughly 50% to 38%, I am going to assume for these purposes that the horserace poll results themselves account for this lead.
Now, let us reweight the latest Washington Post/ABC News poll as an example:
The Washington Post/ABC News poll uses a partisan weighting of 33% Democrat, 27% Republican and 36% Independent (with 4% not responding) to find that likely voters prefer Obama over Romney by 49% to 48%.
Using the Edison model of 36% Democrat, 36% Republican and 27% Independent, the Democrat weighting is increased by 3%, the Republican weighting increased by 9% and the Independents lowered by 9% (eliminating the 4% not responding).
Romney’s poll support of 48% gains 9% with the increase in Republicans and then loses 4.5% in the decrease of Independents for a revised support of 52.5%.
Obama’s poll support of 49% gains 3% with the increase in Democrats and then loses 4.5% in the decrease of Independents for a revised support of 47.5%.
Under the Edison model reweighting, Romney now leads Obama 52.5% to 47.5%.
Using the Rasmussen model of 35.4% Republican, 34% Democrat, and 30.5% Independent, the Democrat weighting is increased by 1%, the Republican weighting increased by 8.4% and the Independents lowered by 5.5% (eliminating the 4% not responding).
Romney’s poll support of 48% gains 8.4% with the increase in Republicans and then loses 2.75% in the decrease of Independents for a revised support of 53.65%.
Obama’s poll support of 49% gains 1% with the increase in Democrats and then loses 2.75% in the decrease of Independents for a revised support of 47.25%.
Under the Rasmussen model reweighting, Romney now leads Obama 53.6% to 47.2%.
My suggested reweighting is only a simple back of the envelope correction meant to give you a general idea of the actual support for each candidate in a particular poll and not to predict the actual outcome of the election. Despite these limitations, you can see that polls corrected with a realistic partisan weighting offer far different results than the biased spin we are being fed by the media.