How accurate are the presidential pre-election polls? Not very, if you believe some high-profile pundits. But an analysis of pre-election surveys tells a dramatically different story. Polls conducted in the days before a presidential election are remarkably predictive of the outcome. And rather than getting worse, pre-election polls have been getting better, according to historical data collected by the National Council on Public Polls (NCPP). The table below compares the results of different 2004 election polls. Click on the areas outlined in red to learn more about them.

Survey-savvy journalists should keep in mind two important caveats when they report pre-election surveys:

1. Polls conducted more than a week before an election should not be interpreted as predicting the outcome.
   It is simply too early. Some people who don’t plan to vote end up doing so based on what they hear or see during the campaign’s final week, or as a result of final get-out-the-vote drives by the campaigns. Others who plan to vote decide not to bother. And still others change their preferences. In 2004, the national exit poll revealed that about one out of every 10 voters made up their mind during the final three days before the election -- and 5 percent said they didn’t decided until election day.
2. Not all polls are created equal.
   Even among the most reputable survey organizations, polls conducted earlier in the campaign are more likely to produce varying horserace results. That’s because many people have not yet firmly made up their minds about which candidate to support. Several factors -- including the wording of questions, the order of those questions and to what extent undecided voters are encouraged to declare a preference -- can produce relatively large differences in candidate preference among polls. And some pollsters simply have more experience doing pre-election polls. Which polls should you trust? Examine their track record in recent elections. A good place to start is the NCPP Web site: www.ncpp.org/.