Michael Barone has an excellent piece on the way that polling works (or doesn’t) in the 21st Century. It’s looking increasingly like the old techniques of political polling aren’t coping well with technological change. Voters are less likely to have landline phones, and less likely to have time to bother with answering a pollster’s questions. Barone casts some doubt on the results of many of the polls that have come out for this election cycle:
If you could go back in history and conduct polls, I don’t think you’d find any, and certainly not many, two-year periods when the balance in party identification shifted from even to having one party 12 percent ahead of the other.
At this stage of the campaign, pollsters try to screen their respondents and report only those who answer a series of questions in ways that suggest they are actually going to vote. Many polls find that a higher proportion of Democrats than Republicans pass the screen. Others find similar proportions do. But pollsters of both parties will admit that polls do a poor job at projecting turnout.
I don’t buy for a second that the Democrats have shifted partisan ID by twelve points in two years. If that were true, the Democrats would be ahead in nearly every race by huge margins. However, the balance of power in Congress remains close — even if the Democrats win, it seems unlikely to be by more than a handful of seats.
It could be that the problems with polling is due to methodologies not keeping up with the times. It could be due to unconscious bias. It could be unconscious bias. Whatever it may be, the polls seem to be diverging more and more from the actual face of the electorate. When we’re talking about a potential 12 point bias towards the Democrats in an election that is already looking to be quite tight, it’s entirely possible that Election Night could be quite surprising…