I understand when firms are polling weeks before Election Day that asking campaign questions to “likely voters” will likely produce a much better read on possible ballot box results. But does the preferred “likely voter” pool of respondents work when asking more generic, non-campaign questions? Or won't those results be skewed by only asking “likely voters”?

For instance, with a new Battleground Poll out today from George Washington University, Politico stresses the results from the survey which deal with prominent media figures, such as Bill O'Reilly (very popular), Jon Stewart, and Rachel Maddow (not so much.)

The poll also asks respondents what their main sources of news are [emphasis added]:

The poll found that 81 percent of those polled get their news about the midterm elections from cable channels, like Fox News, CNN, MSNBC, or their websites, compared with 71 percent from national network news channels, such as ABC, NBC or CBS, and their websites.

Among cable news channels, Fox was the clear winner, with 42 percent of respondents saying it is their main source, compared with 30 percent who cited CNN and 12 percent who rely on MSNBC.

I don't think you have to be a polling guru to see the problem with asking only “likely voters” media-related questions, when all indications are today that the “likely voters” pool leans solidly to the conservative side. (It also skews very old.) I think the GWU poll would have been more useful if, when asking the non-horse race questions, it had expanded beyond the world of “likely voters.”