Should we trust political polls?
Election coverage and reporting are too dependent on public opinion polls. It is almost impossible to remove partiality from polls, as question order, phrasing, and sample methodology can all introduce bias. And it is so subtle that the average person would be hard-pressed to detect it.
The media figures who report on polling can cherry-pick the polls that fit most favorably into their news coverage. Additionally, they do not explain how to read polls, frequently glossing over essential pieces of information, like the margin of error and confidence intervals.
Often, the media will report poll breakdowns, looking, for example, at how Republicans versus Democrats responded to a question about presidential job satisfaction. They usually don’t explain these questions are asked to a subset of the poll’s respondents, meaning there is a different margin of error than is attached to the whole poll. And this can skew people’s understanding of polls.
The media also uses polls to project a candidate’s chances of winning an election, aka election forecasting. But a recent study by Dartmouth College shows that this can suppress voter turnout: when people hear that their candidate has little chance of winning, they don’t vote. Ironically, this will guarantee their candidate has no chance of winning.
Polls are not deterministic. They are a snapshot of opinion at one period in time. They change as people’s opinions shift in response to new events. But the media reports them as if they are deterministic, which influences people’s thoughts about and behaviors in elections.
The 2016 U.S. Presidential election shook many Americans' faith in political polls because so many media outlets forecasted an overwhelming Clinton victory despite Trump's eventual election. People need to understand that polls are useful in understanding the likely outcome of an election and for 'taking the pulse of the electorate…[they] are a critical tool for democracy.' It's when people assume them to be accurate predictions of the future that they will be disappointed.
FiveThirtyEight, a data analysis and politics website, reported that the average margin of error for polls conducted on state-level and presidential races since 1998 is about six percentage points. They found that in 2016, the polls were about as accurate as ever, only underestimating Trump's results by 2-3 points in the average swing state. It just so happened that the race was very close--in fact, within the range of a typical polling error--yet, those 2-3 points clinched his victory. The polls correctly predicted his loss in the popular vote, and if it weren't for the unique system the U.S. uses to elect the president, Trump would have lost.
Trump's election in 2016 and other polling upsets are more of a reason to question conventional wisdom and media narratives than polling. Polls are estimates of the state of the race, but news media consumers must remember that they do not predict the future. In particular, the average of polling before a major race has consistently been close to the result. For this reason, polling as a whole is trustworthy.
- Political polling based on scientific methodology began in America in 1936 with George Gallup.
- Researchers assert that polls “...allow individuals to observe their environment, get an impression about what others think and compare their own stands on issues with the opinions of their fellow citizens.”
- In a nod to the perceived untrustworthiness of polls, Twitter has dubbed itself, “a real-time measure of public opinion.”
- A 2018 The Hill-HarrisX poll revealed that 52% of registered voters do not trust the survey reports they hear in the news.