Tuesday, November 13, 2012

Polling Problems

During the 2012 campaign, Republican pollsters erred badly by overestimating the GOP vote.   At National Journal, Reid Wilson writes:
Some answers are easy to implement: Pollsters should fix voter screens, used to weed out of their samples irregular voters who aren't likely to vote. Including only likely voters often leads to a more Republican-heavy sample. But in an era of fine-tuned turnout machines and get-out-the-vote drives, even those irregular voters are likely to show up. Polling all registered voters, rather than those most likely to make it to the polls, would at least give Republicans an idea of the worst-case scenario.
Pollsters should also control more for age, gender, and race than for party identification. One prominent party pollster pointed to a late survey conducted for Indiana Republican Richard Mourdock that showed him leading Democrat Joe Donnelly by 2 points. That survey, conducted by McLaughlin & Associates, showed that 56 percent of Indiana's electorate would be over age 55. Exit polls revealed that number to be vastly overstated; only 43 percent of the electorate was over 50.' 
The party-identification question gets to the heart of another misperception that pollsters make. Tell almost anyone that Romney would have won self-identified independent voters by 5 points and logic would dictate that Romney would win a clear victory. But Democratic pollsters say that metric is flawed, and that many Republicans remain so disaffected by their own party that they refuse to identify with it. Instead, some say that pollsters should look at self-described ideology, rather than party identification. Indeed, Obama beat Romney among the 41 percent of voters who call themselves moderate by 15 points.
Pollsters also recognize that Americans' daily routines are changing, something that has an impact on their surveys. About one-third of all households do not have a landline, according to the National Health Interview Survey, meaning that a significant swath of the electorate is available to pollsters only by cell phone. The percentage of younger Americans who don't have a landline is almost double that. Pollsters who don't include a sufficient number of cell-phone respondents in their surveys risk missing out on younger voters -- voters most likely to back Democrats, thus skewing polls to the right.