Featured on The Hill, by John Zogby
A good political pollster’s job should be easy enough to understand: draw the best probability sample; ask balanced, fair-worded questions; be able to offer the best-informed interpretation; and find the vectors that point to trends that could emerge. But if the client or pundits are looking for the final result of an election with certainty, including right up to the day before voters cast their ballots, they’d better visit a soothsayer or tarot card reader.
When I first started in this profession in 1984, things were much simpler. There were relatively fewer political independents, up to 90 percent of voters had their minds made up mainly due to their party identification, and even how the undecideds would break in the end seemed to follow a clear pattern.
That is not so today.
In any typical presidential election, as many as 10 percent to 15 percent tell us that they make up their mind on Election Day, and more likely voters than ever before choose to not vote at all. Thus, as someone who has been polling and explaining polls for four decades, please allow me to offer some advice on what polls tell us — and it is a lot — and what they cannot do.
Polls are a snapshot of a moment in time, and there are multiple possibilities for error. They do not and cannot predict an outcome all the time, especially far in advance of an actual vote.
Polls missed the outcomes of the Reagan landslide in 1980, for instance, just as they missed the huge Republican victory in 1994. In both cases they did not estimate the degree that voters in the losing parties would simply decide to not vote. Polls can only tell us how things look today, where a candidate is strong or weak, what undecideds look like demographically, and what messages or events can alter a person’s decision (or lack thereof).
Collecting a series of polls, however, can point out directional signs, upward or downward ticks, and who has real potential or is just whistling Dixie. Someone polling at 6 percent or 7 percent may initially be overlooked, but it depends on whether that candidate was polling 1 percent, then 3 percent and then 7 percent versus whether he/she started at 22 percent, then 13 percent and is now at just 7 percent. Directional signals are vital readings.
Pollsters just don’t look at fleeting opinions on issues; the best of us try to attach a value to each issue and tweak out which value is strongest as a driver for how voters will vote. The same person who says that terminating a pregnancy is tantamount to manslaughter can strongly feel that a woman has the right to choose. Which value is more important in that voter’s decision? Which message and bonding persona of a candidate can call that value to the fore?
In other words, good polling should identify the values that are sacrosanct and can really reveal how a voter may act.
Don’t mix different pollsters and try to find trends. Rather, follow each pollster’s trend lines. Pollsters use some different methodologies and different Election Day turnout models; you get a much better picture by following each pollster.
Look closely at the turnout models chosen by each pollster. I am always wary when I see individual polls and pollsters who oversample Democrats over Republicans to the tune of 8 to 10 points’ advantage for Democrats. That is just not the United States of America. Unless there is a strong reason why one party or the other is falling apart, there are generally fewer Republicans than Democrats in the national electorate — but not by a wide margin.
One of the most troubling myths leftover from the 2016 election was that somehow the pollsters got it wrong. I did not poll in 2016 and I am going to be gracious to my competitors. The polls were just fine. It was the talking heads missed the mark.
If we look back at the Real Clear Politics compilation of tracking polls, nearly every poll showed Hillary Clinton winning the popular vote by 2-4 percentage points. One prominent poll for the Los Angeles Times had Donald Trump in the lead — but enough to make me look closer. The average Clinton lead in the polls the day of the election was 2.4 percentage points — her actual winning popular vote margin.
Perhaps of greater significance — but somehow lost in the shuffle — was how Clinton’s substantial leads in the battleground states dissipated from two Sundays before the election until Election Day. Her 9-point lead in New Hampshire, 10-point leads in Pennsylvania, Wisconsin and Michigan, 6-point lead in Wisconsin, and even slight lead in North Carolina steadily evaporated to a tie in New Hampshire, only a 3-point lead in Pennsylvania, virtual ties in Wisconsin and Michigan, and a 2-point deficit in North Carolina. These prompted me on Saturday through the late morning of Election Day to warn folks to not jump the gun on a Clinton victory.
I believe what happened is that the talking heads selectively screened out a possible victory of Donald Trump, and that columnists, bloggers and pundits of both parties simply missed the downward trend for Clinton.
So don’t dismiss the polls. Read them and examine them thoroughly. The industry has been challenged by new technologies (cellphones, call screeners, near-universal access online) and significant social changes (including not answering the telephone) but for the most part we pollsters have adjusted like many industries have done.
Just remember that good polls will tell you what can happen, even what more likely will happen. But don’t always expect to have them tell you what exactly is going to happen.
