Q&A: Are polls trustworthy?

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility: In the final stretch to the midterm elections, polls have come out tracking candidates' likelihood of victory at the local, state and federal levels. According to Tim Servinsky, project manager at Penn State Harrisburg's Center for Survey Research, polls built with the right questions and asked to the right population can be accurate and a critical tool for policymakers.

However, particularly in the context of political polling, there are important details to keep in mind when reading and interpreting the most recent polling projections. In the following Q&A, Servinsky, a polling data expert who has been in the field for 20 years, explained what goes into building a credible poll, why polling matters and what to look for to confirm a poll's accuracy or reliability. Servinsky: Polls focus on a sample of people from a population of interest and ask them preselected questions.

By asking the respondents the same set of survey questions, pollsters can analyze the results to learn more about the views or beliefs of that population. Different factors go into building a reliable poll. Let's say you want to make an inference about all of Pennsylvania—this means your population of interest is Pennsylvanians.

You might select 20,000 people from all of Pennsylvania, and that's your target sample. The final group that you get results from—let's say it's 1,000 people that complete the survey—would be your sample size. Deciding a final sample size is usually a balancing act between the goals and budget of the poll, alongside the level of risk you're willing to take with the size of your sample.

For example, if I survey 400 people, that results in what's called a margin of sampling error of about 4.9%, with 95% confidence. That means that if I did this survey 100 times, I would expect the results for each question to vary by only about 5% in either direction in 95 of those 100 times. Those other five times would be outlying responses.

So, when I conduct my survey, I'm thinking about what this margin of error will mean when I go to interpret my results. Crafting good questions is also important. The perfect poll questions should be unbiased, succinct, short and easily understood.