Our Detailed Methodology

In addition to its large sample size, the National Annenberg Election Survey (NAES) is distinctive because it employs a rolling cross-section (RCS) design. With this method, random samples of respondents are interviewed each day of the presidential campaign period in such a way that the samples are comparable from one day to the next. Specifically, the composition of each day's interviews is balanced on various demographics characteristics. Daily interviews can thus be used to identify trends and points of change in the public's reactions to political events as they unfold over the course of the presidential campaign.

The RCS design is a series of repeated cross-sections collected with a rigorous sampling plan. This sampling plan works to ensure that each of the repeated cross-sections is composed of randomly selected members from the population under study. In the case of the NAES, the design is used to gather cross-sections of randomly selected adults in the United States during the presidential campaign. Because the composition of each cross-section is random, researchers can treat the date of interview as a chance event. Because the date of interview can be treated as a chance event, researchers can analyze the data as a single cross-section or a time series.

The date of interview is a central concern of the RCS design. The NAES rolling cross-sectional design uses "day" as the unit for the interviewing schedule in order to capture campaign dynamics and attribute changes in public opinion to particular campaign events. Unlike typical polls that release potential telephone numbers into the field at one time, the NAES design calls for a release of a set number of "replicates" (random subsamples of the sampled telephone numbers) for interviewing each night to ensure that each daily cross-section is truly random.

Strict procedures have been worked out so that each telephone number has the same chance of being selected and of producing a completed interview as any other telephone number. Serious effort has been made to increase the response rates without compromising the assumption underlying a random sample. For example, people who are initially called on a weekend are not pursued more aggressively than people who are initially called on weekdays. The NAES follows a special protocol so that the sample of respondents interviewed on any single day will be as representative of the population as possible.

The 2004 NAES uses the same protocol used in 2000. The interviewing for 2004 is being conducted by Schulman, Ronca, Bucuvalas, Inc. In the 2000 NAES, a total of eighteen call attempts were made for every telephone number that was released into the field. The call backs took place over a period of two weeks. If it was determined that a telephone number was out of service or was a non-residential number, interviews at that number were not pursued. After a telephone number was released into the field, call attempts were made as follows:

• Days 1-4: 2 attempts each day
• Days 5-14: 1 attempt each day

There was also a refusal conversion protocol. If a respondent made an initial refusal on days 1 through 6 from the phone number's release into the field, then the person was called back for potential conversion four days from the initial refusal. If the initial refusal was made on days 7 through 9, then the person was called back on day 10. If the initial refusal took place on days 10 through 13, then the respondent was called back for conversion the next day. And finally, if the initial refusal took place on day 14, then a callback for conversion was made on the same day. If a completed interview did not take place after 14 days of a telephone number having been released into the field, no further contact was initiated.

What is important to note here is that there are strict procedures in place so that no telephone number is treated differently from any of the other numbers selected. Telephone numbers released into the field on Tuesdays are not handled differently from telephone numbers released on Fridays. This protocol ensures that the probability of being interviewed is a random event. From prior research, we know that people who answer a survey on the first call attempt tend to answer differently from those who are not reached until the eighteenth call attempt. By stabilizing the proportion of respondents who completed an interview after having been called only a few times and those who completed an interview after being called numerous times, the representativeness of the daily cross-sections is maximized.