Currently, the National House Generic Ballot at Pollster.com shows Republicans with a narrow 41.5%-40.7% advantage. Some may object to this, pointing out that without Rasmussen, Democrats actually hold a sizable 43.2%--38.8% advantage, showing that a single polling firm is heavily influencing the overall figure. Since a House generic ballot is a reasonably accurate measure of which party leads in the national House popular vote (though, admittedly, is not a very accurate predictor of seat totals), it is worth asking which figure is more accurate. Which major party is actually ahead right now.
Here are some thoughts on how to solve this problem:
What I conclude from this is that Rasmussen should be included in the overall figures, but that it is still skewing the results too heavily. Assuming for a moment that additional polling firms would produce national generic ballot results equal to the simple mean of the four non-Rasmussen polling firms (Dem 42.25%--40.00 Rep%), here is what the national generic ballot would currently look like with more polling firms included in the measurement:
- First, removing Rasmussen from the figures altogether does not appear justifiable. Despite their huge error in 2000, their 2004, 2006 and 2008 polling appeared perfectly sound. Whatever problems it may have had a decade ago, automated IVR polling has emerged as a methodologically viable alternative to live-interview polls.
- Second, in the fall of 2010, significantly more polling firms will be measuring the national generic ballot than currently are doing so. In 2006, seven polling firms conducted generic ballot tests in the final week of the campaign. In 2008, eight polling firms published generic ballot tests (seven here, plus Rasmussen). By way of comparison, in August of 2009, only five polling firms, including Rasmussen, conducted national generic ballot tests.
Six polls: Dem 41.1%--40.5% Rep
Seven polls (2006 level): Dem 41.4%--40.4% Rep
Eight polls (2008 level): Dem 41.5%--40.4% Rep
Nine polls: Dem: 41.7%--40.3% Rep
These hypothetical projections all fall into the same range, showing a narrow Democratic advantage. They also match-up well with the current median of the five firms that have published national generic ballots, the Pew poll showing Democrats ahead 45%-44%.
Given that the mean error on generic ballot simple means since 1998 has been 1.8%, projecting a 1.0% Democratic advantage instead of a Republican advantage of 0.8% doesn't change much. This shift is rendered even less significant by the inaccuracy of predicting specific seat totals in the House from the national popular vote. Within that range, either party could still end up with 230 seats (I hope to have an article explaining why this is the case next week).
Still, I hope that this analysis adds some depth to discussions of the national popular vote. I doubt it is as simple as either discarding Rasmussen entirely, or thinking that Rasmussen's numbers should be given such significant weight simply because they poll more often.