November 2008 exit poll methodology

About the 2008 general election exit polls

Conducted for The Associated Press and other members of the National Election Pool (ABC, CBS, CNN, FOX and NBC) by Edison Media Research and Mitofsky International

Coverage: Representative surveys of voters nationally and in each of the 50 United States and District of Columbia in the 2008 general elections. The surveys cover early/absentee voters by phone and Election Day voters in person as they exit the polling place. The sample is designed so that everyone who votes in the general election has a known probability of being included.

How many interviewed: About 80,000 voters nationwide for survey analysis, tens of thousands more in larger sample to help determine election outcomes.

Nearly 70,000 reached in traditional "exit polls" as they exit the voting booth
More than 12,000 early voters reached the week before the election in telephone surveys

Exit polls: Stratified probability sample, taking into account population size and past voting history, of more than 1,300 voting precincts nationwide -- as many as 50 per state, and 300 in the sample for the national survey. The sample is designed so that everyone who votes in the general election has a known probability of being included.

At each sampling location an interviewer approaches voters at a specified interval -- for example, every fifth voter -- as he or she exits the polling place. The interval helps ensure the randomness of the sample.
Respondents check answers on a paper questionnaire prepared by the National Election Pool, including questions about demographics and issues related to the person's vote choices.
Interviewing starts when the polls open and continues until about an hour before polls close.
In places with large Hispanic populations, Spanish-language versions of the questionnaires are available.
Participation is voluntary and anonymous.

Telephone polls: NEP commissioned telephone surveys to reach pre-Election Day voters in 18 states with high rates of early or absentee voting or where nearly all voting is done by mail. Early voters also were sampled in other states for inclusion in the national sample. Digits in the phone numbers dialed were generated randomly to reach households with unlisted and listed landline phone numbers.

In Alaska, Arizona, Arkansas, California, Florida, Georgia, Iowa, Michigan, Montana, Nevada, New Mexico, North Carolina, Ohio, Tennessee and Texas the phone data are being blended with exit poll results, with the phone poll results weighted to represent the expected share of early vote in the electorate.
There are phone polls only, with no exit polls, in Colorado (with a large majority voting early and much Election Day voting at consolidated voting centers rather than precincts) and Oregon and Washington (where all or nearly all voting is done by mail).

Weighting: Survey results are adjusted to reflect the different probabilities of selecting a sample precinct and people attending each, as well as by the observed sex, race and estimated age of voters who refuse to participate or whom interviewers miss.

Sampling error: As with any survey, the results could vary because of chance variations in the sample. The larger the sample, the lower the sampling error. For example, in an exit poll of 1,000 voters, no more than one time in 20 should chance variations in the sample cause each result in the poll to vary by more than plus or minus 5 percentage points from the answers that would be obtained if voters in a given election were polled. For an exit poll of 1,500 voters the sampling error would be 4 points; for a survey of 2,000, 3 points.

In exit polling, sampling error also depends on how many poll sites have voters with the characteristic of interest. For example, black or high-income voters may be found clustered only in certain sample precincts. Sampling error may be up to three times larger for clustered characteristics.

Polls are subject to other sources of error, such as from question wording or order.

Additional information: Edison/Mitofsky site