Watch for sudden indecision in polls

Please forgive your pollster if he or she seems a little groggy or cranky today. We’re full-on into the world of nightly tracking polls, and time for sleep is hard to come by. Even when your head hits the pillow for a few hours as dawn approaches, your brain continues to parse the numbers you’ve been staring down. Your subconscious prepares to deliver the news to clients, even as you try to sleep.

And it’s not just your own polls you have to worry about these days. It has fallen to campaign pollsters to read the cross-tabs of public polls — SurveyUSA, Rasmussen, the local newspaper or TV station — to try and make sense of their results for your campaign clients, reconciling differences among the Tower of Babel of polls out there every week, theirs as well as yours. 

Every few nights or days, some weird results pop up. These are what a former client of mine described as “wild-hair” polls. There are many reasons why wild-hair polls occur. Sampling theory itself allows for one in every 20 random samples to have a larger margin of error than is typical. Random sampling has limits. Other times you can spot skews in the composition of a sample that explains the result, finding an obvious partisan, age, social status or other imbalance. But through the years, I have encountered some wild hairs that are not so easily explained. 

One type of weird poll that I often see, and that readers might want to look out for, is one with a sudden sharp rise in undecided voters. You may be trundling along with 80 percent of the electorate choosing the Republican or Democrat, with only 20 percent being undecided, when “wham!” — the percent of undecided voters suddenly spikes to 30 percent or more. As you get closer to Election Day, you expect the percentage of voters who are undecided to gradually winnow. But two days or two weeks before the election, the undecided voters spike for a day or two. Why?

This happens so often that I anticipate it. My frequent experiences have led me to form some untested hypotheses that explain the phenomenon. One potential explanation is an unfortunate side effect of tracking mania. Call centers are digesting so much work that they get indigestion. Interviewers — live ones, at least — are pressed to work faster and longer. This can diminish their passion for wheedling a ballot choice from respondents. If the voter dawdles, rather than waiting for a candidate name to be coughed up, the interviewer hits the undecided button and moves on. If the script asks whether the undecided voter is leaning either way, an interviewer’s bearing and manner can induce a choice, or not. So being in a hurry might cause an unexpected explosion of voter waffling.

But real-world events matter, too. I find that the percentage of undecided voters can rise around the time of key events, like debates, powerful new attack ads, after major editorial or organizational endorsements, or even the release of startling new poll results. Particularly where voters have modest knowledge of candidates, and their initial choices were based mostly on party labels or geographical affinity, new information can cause some voters to hit a mental reset button that takes the race back to day one of the contest. Debates provide voters with a different type of information than they get from TV ads or local news coverage. Hardcore attack ads can startle some voters into a reassessment of their favorite. Endorsements can present discrepant information that causes reconsideration of a candidate choice based on little thought. Even polls can remind a voter that an initial choice is out of step with friends and neighbors, causing the voter to rethink his or her preference.

David Hill has been a Republican pollster since 1984. This cycle he is polling for gubernatorial campaigns in four states.