An interesting tweet this weekend from Nate Silver asks a pretty pertinent question:
Odd that Rasmussen has polled the SD and ND at-large House districts -- tough Democratic holds -- but not the tough GOP hold in DE.
It is odd, because when you are polling a Senate race in a state with only one Congressional district, it doesn't require any additional effort to poll the House race, as well. When Daily Kos opened up our new polling partnership with PPP, we did so in Delaware. And, as you can see (PDF file), we made sure to poll the House race.
Perhaps the outcome of that poll (we had Democrat John Carney up double digits on either Republican) explains why Rasmussen stayed away from it. It would seem that it would be an ethical morass for them to sit on data that wasn't amenable to Republicans. But they could certainly choose not to poll it. And that would appear to be what they have done, in this case.
Over at Balloon Juice, the writer there offers a pretty succinct answer to Nate's question, and it is hard not to think that it is the correct one:
Early polls of voters in R+9 and R+10 states is an easy way to drive a “wave” narrative, and Rasmussen got what he wanted when he polled South Dakota and North Dakota.
Another way to drive a "wave" narrative is to not get exposed. Notice how reluctant Rasmussen has been to poll primaries this cycle, or the one major statewide contest in 2010. Rasmussen did a rather strange exit poll of sorts in Nevada's high-profile Senate primary, but did not poll the actual primary (according to this polling compendium). Indeed, after practically living in Nevada in the earlier part of the year (five polls in the state from January to April), the pollster disappeared for the six weeks prior to the primaries, only returning the day after the primary.
In the case of the Massachusetts' special election, they polled more than a week out, and then never came back to the race, save for a bizarre national poll on the day of the election about who the country wanted to see win the race (as if that were relevant to anything other than narrative-setting). Therefore, they do not take a hit for their errant result, because it happened with quite a bit of time left on the clock of a very volatile race.
Cherry-picking races, to say nothing of avoiding polling immediately before elections, seems to be a rather strange way for a firm that gets heaps of praise as one of the most important pollsters in the business to behave.
Or, as Nate put it: odd.