Up until now I have viewed Gallup and Rasmussen as equally reliable in their polling. Generally they have been closer to each other than either was to the other pollsters that publish public polling results. Recently that has changed.
Rasmussen and Gallup are now both producing daily head to head tracking polls on Romney and Obama. Over pretty much the same period Rasmussen shows Romney up 5 and Gallup shows Obama up 7, a 12% difference well outside of either polls MOE. It’s also interesting that, over the last 7-10 days, the Gallup poll has shown a lot more movement than the Rasmussen poll. Pretty strange why are they different? It appears that the reason is that the two polls are using very different assumptions about voter demographics.
Rasmussen is kind enough to publish their demographic assumptions. Rasmussen conducts a three month moving average of the party demographics that each month produces data for a sample set of 45000 voters. So as of the three months from January – March of 2012 their sampling indicated the following party affiliation for the electorate: 36.1% Republican, 32.8% Democrat and 31.1% unaffiliated. This is by far the most comprehensive analysis of voter demographics by party affiliation of any polling organization.
Gallup is not as forthcoming about their demographic assumptions, but it is possible to work backwards from their reported support by party and come up with a very close ( less than +-0.5%) estimate of their sampling assumptions. The problem with working this backwards is that rounding of the actual reported sample induces some error preventing a perfect fit and at the edges of the sample sizes you have the same error margins. However one can come close enough that we can establish the sample sizes well within the MOE of the poll. Working Gallup’s results backwards they are using a weighted sample that is about 27% Republican, 33.5% Democrat, and 39.5% Independent.
If you take the party specific results of the Gallup poll and apply the demographic assumptions used by Rasmussen you get 44.8% for Obama, 47.7% for Romney and 7.5% undecided. Given the poll’s respective MOEs the difference between the adjusted Gallup results and those for Rasmussen is not statistically significant. What is significant is that Obama goes from up 7% (outside that poll’s MOE) to down 3% (inside that poll’s MOE). Further, when Rasmussen’s demographic assumptions are used, both polls show Obama well under the 48% level that is presumed to be the lower bound for reelection.
The Gallup poll did not provide a break down of the responses to the presidential approval numbers by party so it was not possible to adjust those results for the differences in demographics.
Rasmussen, who publishes their demographic assumptions has not reported the reversal of position shown by Gallup, who does not report on their demographics. Further this reversal has occurred rather dramatically after Gallup was the target of criticism by the white house. Nothing of note has occurred during the week of so in which Gallup’s numbers flipped.
So at this point it comes down to whose demographics do you believe.