What Is the Science Behind Polls?
Why shouldn't polls always be trusted? Jules tackles this question here.
In terms of media bombardment, it has been a particularly brutal election year, with a seemingly relentless torrent of information coming in fast, each and every day. Smack in the middle of each spin cycle are the latest poll numbers and survey results. But can they be trusted? Trace Dominguez looks at the science of the polling process in today's DNews report.
Scientific polling organizations typically survey 1,000 to 1,500 people, by phone, using an established set of protocols. The number of people polled tends to stay in this range, since it's logistically feasible for fast turnaround, and adding more people doesn't change the accuracy considerably. At 1,000 to 1,500 people, the margin of error is around three percent.
Pollsters can get that figure down to one percent, but it requires around 10,000 people surveyed. That takes too much time (and money) for time-sensitive public polling. The goal is to reach a broad cross-section of the public that is sufficiently randomized to produce meaningful numbers.
The Pew Research Center -- one of the world's busiest polling services -- typically finds its participants by assembling a list of geographical areas. The number of calls made in each area is proportional to its share of phone numbers across the entire country. For randomization, the pollsters use the area code and first five digits, then randomize the final two numbers.
The idea is to get totally random representatives from across the nation, although with more people switching to cell phones, it's become difficult to establish where respondents actually live. From there, pollsters deploy several other strategies to ensure the most accurate results possible. For instance, when asking specifically about a head-to-head election, pollsters will ask whether the person plans to vote at all. If so, candidate choices are given in random sequence, and the voting question is asked before any policy questions.
Research studies over the years have helped pollsters further refine their technique. Keep in mind that these kind of precautions are only taken by objective scientific polls. Informal polls -- especially online polls -- are famously inaccurate because they make no effort to secure a representative sample.
Trace has more details in his report, or for some real hard numbers, click on over to the National Council on Public Polls, which keeps records of how accurate each poll was after every election.
How Stuff Works: How Political Polling Works