The new Arkansas Poll released yesterday showed wide gaps up and down the ballot for Republican candidates, wider than anything polled anywhere else. That raised questions, particularly among those on the losing end, about methodology.
The Arkansas Poll, intended to sample the whole state, is not drawn from registered voter files, as some political polls are. Makes sense. We have a lot of people who live here who aren’t registered to vote, don’t vote or can’t vote. The survey aims at a regular sounding of the totality of the Arkansas mindset on issues political and non-political.
The poll drew its 747 responses from a random draw from a purchased list of working telephone numbers that included both landlines and mobile phones. So, when the poll reports numbers from 568 “very likely voters,” they were self-identified as voters. I asked UA prof Janine Parry, director of the poll about this. Her response:
All self-reported, yes. I suppose we could be talking to a 15 year exchange student from Honduras at any given time. There’s that chance. Seems to work, though, if you look at our track record. ..
Also, have you looked at the phone numbers in a voter file lately? A number is only supplied for perhaps every 8th person, and – in my case – it’s the land line I gave up five years ago. Plus, I suspect some voters are more likely to supply their phone numbers than other voters. Def. not a random sample … of Arkansans, or likely voters.
In the end, it’s all duct tape and bailing wire, Max. Or, rather, there’s always a chance any poll is wrong, even way wrong. And, yet, there’s the arkpoll’s history with this approach (and it’s what most organizations do, although they screen on the front end and hang up on you if you say you’re not very likely to vote … we visit with you pleasantly so we can get a robust policy sample, then narrow it to talk about what likely voters think). And there’s that reassuring consistency in the responses to the questions we ask all the time. Indeed, there’s nothing in this poll that suggests it’s wonky, except that wider-than-the-rest margin for Cotton. …
I don’t like to dangle out here, man. Trust me on that. But while there’s some evidence to suggest that I pulled a bad sample this year (chiefly other polls, almost all showing Cotton ahead just by a smaller margin), there’s more evidence to suggest that the race has widened.
Today’s cliche: The only poll that counts is Tuesday.