Peter Enns is a professor of government at Cornell University who studies quantitative research methods. He comments on how to interpret polls in the presidential race.
He says:
“National polls typically have a margin of error of about plus or minus 3 percent. For this reason, if a lot of polls are conducted, it is not surprising to find one or two polls that differ. Sometimes headlines focus on the one poll that differs the most, but if this difference reflects sampling error, this is actually the least informative poll.
“Most of the current ‘horse race’ polls are based on estimates of ‘likely voters.’ However, the polling firms often do not report exactly how they identify likely voters. Without this information, it is impossible to identify how much these calculations of who is a likely voter affects the estimates. A look at the current Gallup poll suggests this might be quite a lot. Their estimate of registered voters is 48 percent Romney and 47 percent Obama, a statistical tie. Their estimate based on likely voters, however, is 51 percent Romney, 45 percent Obama.
“Both of these concerns – sampling error and a firm’s calculation of who is a likely voter – can be alleviated by viewing the patterns across multiple polls. Thus, my recommendation is to focus on the patterns and not the deviations of any particular poll. “Another important point is that the polls always fluctuate during the campaign. The polls increasingly approach the actual vote margin as Election Day nears; historically the polls from the final week of the campaign are quite accurate.”
Enns research paper on polling accuracy:http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1643043
Contact Syl Kacapyr for information about Cornell's TV and radio studios.
RSS