How accurate are polls?
Historically, individual presidential polls, such as Gallup, have missed by an average of about 3% off the final vote tally. But if you average together multiple polls, the historical accuracy is about 1% off the final total. You can see a current average here. Scott Keeter of the Pew Research Center sums it up: “In 2004, nearly every national pollster correctly forecast that Bush would win in a close election, and the average of the polls predicted a Bush total within a few tenths of a percent of what he achieved. Among statewide polls in races for governor and U.S. Senate, 90% correctly forecast the winner, and many that did not were still within the margin of sampling error. The record in 2000 was similar, though that was an even closer [presidential] election.” You can see a chart of state polls in the 2004 election; the polls correctly predicted 13 of the 14 battleground states. In the most controversial state, Ohio, the polls predicted a 2.1% win for Bush, and the final vote matc
Exit polls during the Virginia Republican primary predicted that George W. Bush would defeat John McCain by 12%. In the final result, Bush won by a hair less than 9%. So, if exit polls, which poll only people who actually voted, can be wrong by 3%, the other polls can be even worse. Return to Institute of Election Analysis Home Page Contact: Joshua Leinsdorf Note: The Institute of Election Analysis would like to thank Fox 5 news for the inspiration for this analysis of polling inaccuracy.
The NCPP analyzed final presidential election polls conducted by the national media dating back over 50 years. When compared with actual election outcomes, average poll error for presidential elections between 1956 and 1996 has been declining. Average poll error on each candidate during this period was 1.9 percentage points. Important to this analysis of accuracy is that most of these polls were conducted within days or even hours before election day. Polls conducted 1-2 weeks before election day or even longer by local newspapers and TV stations cannot usually be expected to closely match election outcomes. Earlier polls are intended to monitor the success of campaigns and to identify the issues or events which will influence voter preferences on election day.
We can’t state anything with certainty, because anything can happen. What we can do is measure a moment in time and then keep measuring moments and project what may happen. Do you think the media misuses polls or takes them out of context? Sometimes they do, or they make the same mistakes that consumers do. Like, you can’t mix and match different pollsters. So if Company A has McCain up by seven points and Company B has him down by three, you can’t say that he’s up by four. Another mistake is not accounting for a margin of error. If Company A has Obama up by two and Company B has McCain up by one, they’re actually saying the same thing, because a margin of error exists. In general, consumers should look at many different polls and observe the overall trends. On Election Day 2004, you said that John Kerry would win. Can we trust you? That was a bit of hubris on my part. A couple of states such as Ohio were too close to call, so I made an assessment based on pre-election polls and exit p
The owner of a company that conducted political polls pleaded guilty on charges of wire fraud after it was found out that the company was falsifying poll information. Poll data was altered to meet customer requirements and in some cases the data was totally made up. The company’s customers included the campaigns of President Bush and that of Sen. Joe Liberman.
