How different would elections be if there were no polling? Would it affect your vote, if you didn’t know who was ahead and who was behind? Might it even change the outcome?
In recent months, political scientists and professional pollsters around Massachusetts have been engaged in a tense, even occasionally heated debate about just these questions, with broad implications for newspapers, campaigns, and indeed our whole democratic system.
What’s the argument?
Over the summer, when Martha Coakley and Steve Grossman were battling for the Democratic nomination for governor, political science professor Jerold Duquette raised concerns about polls showing Coakley far ahead. Her large lead seemed implausible to the Central Connecticut State professor, which made him think the polls might be wrong.
He also worried that the polls themselves were affecting the race.
“If there were no media polls in the race for the Democratic gubernatorial nomination so far there is little chance that Martha Coakley would be considered the lone frontrunner. Media coverage of the race between the Attorney-General and State Treasurer Steve Grossman would almost certainly reflect a very close and competitive fight between two Democrats who have already won statewide office.”
His worry was that the polls themselves had produced a “bandwagon effect,” where people — and donors — favor the lead candidate simply because she is leading. When that happens, it turns the polls into a kind of self-fulfilling prophecy.
Did these claims attract much attention?
Duquette’s arguments were met with limited skepticism at the time (said Boston Magazine writer David Bernstein, “If I want a baseless claim that a Dem frontrunner is in trouble I’ll read the Herald”).
And the whole thing may have disappeared if Duquette hadn’t made a specific prediction, built around the following argument:
1) The polls have given Coakley an air of inevitability
2) This sense of inevitability will reduce turnout
3) The people who do vote will be “high information” voters, with strong feelings about the race
4) “High-information” voters are less likely to vote for Coakley
“If I’m right,” he concluded, “the margin between Coakley and Grossman should be considerably less than 20 percent.” The final margin was 6 points.
How did the pollsters respond?
Professional pollsters like Steve Koczela at MassINC and John Della Volpe at SocialSphere (which handles the Globe’s polls) have remained largely unconvinced by Duquette’s argument. They acknowledge that the final outcome was closer than the polls had indicated and cite a number of other explanations.
Among them: voters simply changed their minds at the last minute. The race for attorney general was transformed by a big, late break in favor of Maura Healey, and it’s possible something similar happened in the governor’s contest. The trouble with this argument is that there were some very late polls, and they didn’t show any shift.
Pollsters are also pointing to voter turnout. All polls are built around assumptions about who is likely to vote. It may be that fewer people turned up than pollsters expected, and those who did were less supportive of Coakley. You don’t have to follow the full path of Duquette’s logic about bandwagon effects and high-information voters to reach this conclusion.
Has their been a counterattack?
Koczela has sometimes used the 2006 primary election as a counterexample to Duquette’s claims. By his count, there were actually more polls that year than in 2014, without the feedback effects Duquette has cited.
During the spring, Deval Patrick actually trailed Tom Reilly in the polls, often by 15 points or more. Yet, Reilly’s front-runner status didn’t make his candidacy seem inevitable.
Support for Patrick surged during the summer. By primary day, Patrick was ahead by 20 points. And yet, despite his large lead, there was no complacency, no reduction in turnout, and no clear advantage for “high-information voters.”
What does the research suggest?
Unfortunately, there’s not much hard evidence about how polls affect outcomes. Researchers can’t easily compare “elections with polling” to “elections without polling” because all elections have polling. Even those countries that try to limit polling do so in a very narrow way — say, by preventing newspapers from publishing polls in the last few days before an election.
What researchers can do, instead, is develop mathematical models that mimic election-related behaviors. Maurice Cunningham, a political science professor at the University of Massachusetts at Boston who blogs with Duquette at MassPoliticsProfs, has referred to one such paper which suggests that polling can indeed affect fund-raising and campaign strategy. But without real-world evidence, it’s not clear how accurate this model really is. What is more, when it comes to assessing the impact of polls on voters’ decision-making, even these authors conclude that the “evidence is at best inconclusive.”
So, who’s right?
Perhaps someday, we’ll discover a planet exactly like ours, except that they have elections without polling. And then we’ll get the evidence we need to really understand whether polls do, indeed, have a meaningful effect on election results.
Until then, perhaps the best we can do is follow Duquette’s lead and test our intuitions by making some predictions. Here’s mine: even though the governor’s race looks quite tight, turnout will still be low.