Selzer contacted me to let me know that this is NOT the case. They have been using random digit dialing of landlines (as almost all pollsters do) for their general election polling, and have not been calling a supplementary sample of cellphone numbers.
The reason I got the idea that they had been calling cellphones was because of this article at Pollster.com in October, in which Mark Blumenthal wrote that:
"Selzer also informs us via email that their completed interviews included a small number of voters interviewed on their cell phones."There WERE a small percentage of cellphone numbers included in Selzer's Iowa caucus polling, because these polls were based on registered voter lists purchased from the state of Iowa, and some voters had left their cellphone number as their primary point of contact on those lists. So while there was no deliberate attempt to reach cellphone voters, she was able to reach some anyway via this list-based sampling. For their general election polling, however -- in Iowa as well as Indiana and Michigan -- Selzer is using random digit dailing of landlines, and so no cellphones will be included.
With the Selzer polls taken out of the sample, and refreshing my original analysis based on the most current data, I now show a cellphone effect of 2.3 points rather than 2.8 points. The cellphone variable remains statistically meaningful at the 95 percent significance threshold.
Yesterday, the Pew Research Center issued a detailed study on cellphones and the 2008 election. They found that Barack Obama performed a net 2-3 points better between three of their recent polls when a cellphone sample was included. They also found that cellphone-only voters were significantly more likely to support Obama than non-cellphone voters of the same age.
I apologize to Mrs. Selzer as well as our readers for any confusion. We will have another post up momentarily with some additional thoughts based on my conversation with Mrs. Selzer.