Sat, Aug 08, 2015 - Page 9 News List

Navigating the US presidential polls without being fooled

While polls are increasingly becoming an integral part of US presidential elections, they are often misread or misunderstood. Pay attention to them if you want, but do it smartly

By Brendan Nyhan  /  NY Times News Service

Illustration: Lance Liu

The onslaught of US presidential polls has already begun. You may be tempted to avoid the polling deluge, but the results of these surveys do influence the campaign. That is why we want to show you how to read, or ignore, the polls like a professional.


Polls with surprising or novel results can be irresistible to journalists and the public alike. It is newsworthy if public attitudes seem to have changed in some unexpected way. As a result, these findings tend to attract the most public attention and media coverage. Unfortunately, they are the most likely to be spurious.

What looks like a shift in public opinion is often just random statistical variation. First, all polls should come with an associated margin of error or some other estimate of uncertainty. Take it seriously. With the sample sizes conventionally used in polling, changes in support of one or two percentage points can not be distinguished from random variation. Second, given the number of polls that are conducted, outliers are likely to be common. Approximately one in 20 polls of US President Barack Obama’s approval rating, for instance, will produce a statistically significant change from the last estimate even if nothing changed.


Instead of being seduced by potential outliers or statistical noise in individual polls, rely on polling averages like the ones provided by the Huffington Post’s Pollster or Real Clear Politics, which better distinguish between genuine changes in public opinion and random noise. When an average picks up shifts in opinion across multiple polls, we can be more certain that the views of the public are changing.

When people ignore averages and focus on an individual poll with an extreme value, they are often led astray. For example, a Gallup article last year highlighted former US secretary of state Hillary Rodham Clinton as having by far the best net favorable ratings of the potential presidential candidates tested: +19 (55 percent favorable versus 36 percent unfavorable). However, the poll was an outlier compared with both Gallup’s previous poll, which had her at +11 and other polls of Clinton’s favorability conducted at about the same time. Sure enough, the next Gallup poll had her at, you guessed it, +11, suggesting that her advantage over the Republican candidates tested (many of whom had net favorables of +6 to +12) was less clear than the original articles suggested.


At this stage in the campaign, approximately six months before the first primaries, eventual nominees have traditionally been among the top three or four candidates in the polls.

However, if it is usually important for a candidate to be among the leaders, it is virtually irrelevant who the actual leader is. Early polls often reflect name recognition more than anything else. Donald Trump, for instance, has come in first or second in a number of recent national polls of Republican candidates, but he is hardly a top-tier candidate.

What should you pay attention to instead of polls? Endorsements. Political science research suggests that endorsements from party elites are better predictors than polls. First, these officials can help candidates win by rallying supporters to their side and providing financial and organizational help. And party elites observe the contenders closely and can often anticipate which candidates are most likely to succeed.

Comments will be moderated. Keep comments relevant to the article. Remarks containing abusive and obscene language, personal attacks of any kind or promotion will be removed and the user banned. Final decision will be at the discretion of the Taipei Times.

TOP top