In the run-up to the Nov. 4 presidential election, the News Office has asked MIT experts to weigh in on the presidential candidates, their policy ideas and aspects of the campaign. In this installment in the series, Adam Berinsky, associate professor in the Department of Political Science, looks at the issue of polling.
Q: Has the election process become too dominated by polling?
A: Polling has been around in one form or another since the 19th century. In the 1930s, a number of firms started conducting the random sample polls we see today. Over time, the technology has changed. In the 1930s, all polls were conducted through face-to-face methods. In the 1970s, telephone polling emerged as the dominant method. And today we see Internet polls. One thing that has changed is that it is easier and cheaper today to conduct polls than ever before, so we have seen a proliferation of polls. So there may be more polls today, but polls -- and the dominance of polls as a means of predicting elections -- have been around for a while.
Q: How can the general public better understand polling? What questions should we ask ourselves when we look at a poll?
A: You want to look at a few things. First, look at how the poll was conducted. Some pollsters -- most notably Zogby -- use Internet panels where people choose to respond to surveys. This is a poor way to conduct Internet surveys, and the Zogby polling shows it. Polls in which people "self-select" into surveys completely subvert the process of random sampling that is the basis for polling. These polls are essentially like the call-in polls you sometimes see on TV. People who are interested in a topic respond and they may be very different than the average American. I should note, however, that some Internet polls can be useful.
You also want to look at who is being surveyed. Now that we are close to the election, many firms are surveying "likely voters" -- those people they think are most likely to turn out -- rather than the full population. Since every firm uses a different standard to determine "likely voters," some of the polls are different. But the important thing is to contrast the results to those of the full public, or "registered voters."
Finally, take a look at how pollsters word their questions. The general presidential choice question is straightforward, but once you start asking about issues or characteristics of the candidates, small differences in the wording of questions can have large effects on the answers you get.
Q: Does polling indicate that negative campaigning works in terms of how the electorate will actually vote?
A: The research evidence on this is mixed. Some say that negative campaigns turn off voters, but others have argued that negative campaigns help mobilize a candidate's supporters. The jury is still out.
Q: Since the system of the Electoral College determines the president, do national polls have real significance? Shouldn't state-by-state polls be considered more significant?
A: The nice thing about the 2000 election is that it showed us that presidential elections turn on state-level results. So there has been a lot of state-level polling conducted in the last two presidential elections. True state polls matter most, but often these polls have smaller sample sizes and are conducted less often than national-level polls.
The national polls can give us a sense of the larger dynamics in the race. After this year's Republican convention, for example, McCain had a large bump in the national polls. The different states reacted differently -- some traditionally Republican states became more pro-McCain and some traditionally Democratic states held steady -- but in general, McCain's bounce was felt across the nation. The national polls gave us a window into the overall change in the race faster than the state-level polls. So it's important to look at both sets of polls.
Q: Why are polls sometimes wrong?
A: Well, they are right more than they are wrong. There are two issues here. First, any individual poll could be wrong and sometimes a host of polls is wrong (in the New Hampshire Democratic primary this year, for example) but often if we look at the results of a number of polls together, we get the right answer. Pollster.com and realclearpolitics.com both present the full range of polls with averages. In 2004 and 2006, the average of the polls near the election was pretty much spot on. Second, the polls today tell us how people would vote if the election was today, but we're not voting until November. Once we get closer to Election Day, after the candidates have finished their debates and the campaign has played out, the polls will be more predictive. But just because polls taken in August and September sometimes fail to predict the winner doesn't mean they are wrong; they still give us a sense of where the race stands today.
A version of this article appeared in MIT Tech Talk on October 8, 2008 (download PDF).