Learn how to use statistics to understand the significance of the latest political polling results and to keep yourself from being duped by misleading information.
What is the Sampling Error?
First of all, notice that I said the poll result gives the percentage of the population supporting each candidate, not the percentage of the sample. That’s the whole point of the poll, after all—to figure out what the entire population is thinking. But it’s important to keep in mind that the pollster questioned only a portion of the population, and that even though she may have conducted a well thought-out and scientific poll, she could have gotten slightly different results if a slightly different portion of the population had been questioned. That sampling error is precisely the origin of the margin of error you see reported alongside polls, and it should not be ignored!
What Does the Margin of Error in Polls Mean?
The margin of error is typically reported as a plus-or-minus percentage. For example, the margin of error might be ±3%. But what does this mean? Well, let’s imagine that support for our imaginary presidential candidate A is polled to be 42% and support for B is polled at 46%, with a margin of error of ±3%. This means that the pollster is confident that if an election were held measuring the actual level of support across the entire population, candidate A would receive anywhere between 39% and 45% of the vote (that is 42% – 3% and 42% + 3%), and candidate B would receive anywhere between 43% and 49% (that is 46% – 3% and 46% + 3%).
Just how confident are pollsters with this margin of error? Well, the statistical margin of error reported with poll results is typically what’s called the 95% confidence interval. That means if the pollster created and polled 100 different samples of the population, the result would be within the original reported margin of error in 95 out of these 100 cases. In other words, the 95% confidence interval will contain the true value 95% of the time. While that’s a lot, keep in mind that the 95% confidence interval will not contain the true value in 1 out of every 20 polls.