Exit poll methodology -- Super Tuesday primaries
Voter surveys in the Feb. 5, 2008, presidential primaries were conducted for the National Election Pool (The Associated Press, ABC, CBS, CNN, FOX and NBC) by Edison Media Research and Mitofsky International. NEP members prepared the questionnaires.
Election day surveys were conducted in the following number of precincts per state:
The polling places are a stratified probability sample of each state, factoring in size and past voting history. The sample is designed so that everyone who voted in the state's primaries Tuesday has a known probability of being included.
At each sampling location an interviewer approached voters at a specified interval -- for example, every fifth voter -- as he or she exited the polling place. The interval helps ensure the randomness of the sample. Voters were asked which primary they had just voted in and were given a paper questionnaire specific to that primary.
In Arizona, California and Tennessee -- states with expected high incidence of absenteee or early voting -- telephone polls also were conducted over the past week, screening for people who had already voted or said they were certain to vote before Tuesday. Results from the telephone interviews were blended into the election day samples, weighted to the estimated proportion of early voting among the overall electorate.
All samples are approximations. A measure of the approximation is called the sampling error. Sampling error is affected by the design of the sample, the characteristic being measured and the number of people who have the characteristic. If a characteristic is found in roughly the same proportions in all precincts ("non-clustered") the sampling error will be lower. If the characteristic is concentrated in a few precincts the sampling error will be larger. Gender would be a good example of a characteristic with a lower sampling error. Characteristics for minority racial groups will have larger sampling errors.
The table below lists typical sampling errors for this exit poll for non-clustered characteristics at the 95 percent confidence interval. No more than one time in 20 should chance variations in the sample cause the results to vary by more than this amount for non-clustered characteristics. However, non-sampling factors such as question wording and order and voter non-response may increase total error.
Respondents Margin of sampling error 97 - 102 plus or minus 15 percentage points 103 - 118 plus or minus 14 percentage points 119 - 138 plus or minus 13 percentage points 139 - 163 plus or minus 12 percentage points 164 - 196 plus or minus 11 percentage points 197 - 239 plus or minus 10 percentage points 240 - 299 plus or minus 9 percentage points 300 - 384 plus or minus 8 percentage points 385 - 511 plus or minus 7 percentage points 512 - 714 plus or minus 6 percentage points 715 - 1,067 plus or minus 5 percentage points 1,068 - 1,764 plus or minus 4 percentage points 1,765 - 3,457 plus or minus 3 percentage points 3,458 - 9,604 plus or minus 2 percentage points 9,605 - up plus or minus 1 percentage point