Here There and Everywhere

Can we trust the polls this year? That’s a question many people have been asking as we approach the end of this long, long presidential campaign. As a recovering pollster and continuing poll consumer, my answer is yes — with qualifications.

Martin Kozlowski
To start with, political polling is inherently imperfect. Academic pollsters say that to get a really random sample, you should go back to a designated respondent in a specific household time and again until you get a response. But political pollsters who must report results overnight have to take the respondents they can reach. So they weight the results of respondents in different groups to get a sample that approximates the whole population they’re sampling.

Another problem is the increasing number of cell phone-only households. Gallup and Pew have polled such households, and found their candidate preferences aren’t much different from those with landlines; and some pollsters have included cell-phone numbers in their samples. A third problem is that an increasing number of Americans refuse to be polled. We can’t know for sure if they’re different in some pertinent respects from those who are willing to answer questions.

Professional pollsters are seriously concerned about these issues. But this year especially, many who ask if we can trust the polls are usually concerned about something else: Can we trust the poll when one of the presidential candidates is black?

It is commonly said that the polls in the 1982 California and the 1989 Virginia gubernatorial races overstated the margin for the black Democrats who were running — Tom Bradley and Douglas Wilder. The theory to account for this is that some poll respondents in each case were unwilling to say they were voting for the white Republican.

Further Reading

Tom Bradley Didn’t Lose Because of Race – Voters rejected his liberal policies.
By Sal Russo 10/20/2008

It’s not clear that race was the issue. Recently pollster Lance Tarrance and political consultant Sal Russo, who worked for Bradley’s opponent George Deukmejian, have written (Mr. Tarrance in RealClearPolitics.com, and Mr. Russo on this page) that their polls got the election right and that public pollsters failed to take into account a successful Republican absentee voter drive. Blair Levin, a Democrat who worked for Bradley, has argued in the same vein in the New York Times. In Virginia, Douglas Wilder was running around 50% in the polls and his Republican opponent Marshall Coleman was well behind; yet Mr. Wilder won with 50.1% of the vote.

These may have been cases of the common phenomenon of the better-known candidate getting about the same percentage from voters as he did in polls, and the lesser-known candidate doing better with voters than he had in the polls. Some significant percentage of voters will pull the lever for the Republican (or the Democratic) candidate even if they didn’t know his name or much about him when they entered the voting booth. In any case, Harvard researcher Daniel Hopkins, after examining dozens of races involving black candidates, reported this year, at a meeting of the Society of Political Methodology, that he’d found no examples of the “Bradley Effect” since 1996.

And what about Barack Obama? In most of the presidential primaries, Sen. Obama received about the same percentage of the votes as he had in the most recent polls. The one notable exception was in New Hampshire, where Hillary Clinton’s tearful moment seems to have changed many votes in the last days.

Yet there was a curious anomaly: In most primaries Mr. Obama tended to receive higher percentages in exit polls than he did from the voters. What accounts for this discrepancy?

While there is no definitive answer, it’s worth noting that only about half of Americans approached to take the exit poll agree to do so (compared to 90% in Mexico and Russia). Thus it seems likely that Obama voters — more enthusiastic about their candidate than Clinton voters by most measures (like strength of support in poll questions) — were more willing to fill out the exit poll forms and drop them in the box.

What this suggests is that Mr. Obama will win about the same percentage of votes as he gets in the last rounds of polling before the election. That’s not bad news for his campaign, as the polls stand now. The realclearpolitics.com average of recent national polls, as I write, shows Mr. Obama leading John McCain by 50% to 45%.

If Mr. Obama gets the votes of any perceptible number of undecideds (or if any perceptible number of them don’t vote) he’ll win a popular vote majority, something only one Democratic nominee, Jimmy Carter, has done in the last 40 years.

In state polls, Mr. Obama is currently getting 50% or more in the realclearpolitics.com averages in states with 286 electoral votes, including four carried by George W. Bush — Colorado, Iowa, New Mexico and Virginia. He leads, with less than 50%, in five more Bush ’04 states with 78 electoral votes — Florida, Missouri, Nevada, North Carolina and Ohio. It’s certainly plausible, given the current state of opinion, that he would carry several if not all of them.

Of course, the balance of opinion could change, as it has several times in this campaign, and as it has in the past. Harry Truman was trailing Thomas E. Dewey by 5% in the last Gallup poll in 1948, conducted between Oct. 15 and 25 — the same margin by which Mr. Obama seems to be leading now. But on Nov. 2, 18 days after Gallup’s first interviews and eight days after its last, Truman ended up winning 50% to 45%. Gallup may well have gotten it right when in the field; opinion could just have changed.

We have no way of knowing, since George Gallup was just about the only public pollster back then, and he decided on the basis of his experience in the three preceding presidential elections that there was no point in testing opinion in the last week. Now we have a rich body of polling data, of varying reliability, available.

Second, the exit polls in almost all the primaries this year showed Mr. Obama doing better than he actually did. The same respondent bias — the greater willingness of Obama voters to be polled — which apparently occurred on primary days could also occur in the exit poll on Election Day, and in the phone polls of early and absentee voters that Edison/Mitofsky will conduct to supplement it.

The exit poll gives us, and future political scientists, a treasure trove of information about the voting behavior of subgroups of the electorate, and also some useful insight into the reasons why people voted as they did. And the current plethora of polls gives us a rich lode of information on what voters are thinking at each stage of the campaign. But political polls are imperfect instruments. Reading them right is less a science than an art. We can trust the polls, with qualifications. We will have a chance to verify as the election returns come in.