Survey Introduction and Methodology 1998

view survey | view press release

The information contained in this report is based on 810 in-depth telephone interviews conducted with United States heads of household. Household selection on this project was accomplished via a computer-generated pure unweighted (EPSEM) random digit dial (RDD) telephone sample which selected households on the basis of telephone prefix. This method was used because it ensured a randomly selected sample of area households proportionately allocated throughout the sample universe. This method also ensured that all listed and newly listed telephone households were included in the sample.

This survey employed a multi-stage sampling process. The first step was to stratify the sample by state according to the current population residing in each state. Telephone households were then selected within each state using the RDD methodology. A probability sample developed in this manner draws samples proportionately relative to a state's proportional distribution of the U.S. population. This strengthens the ability of the sample to be compared with Census data and other demographic information.

The questionnaire used in this study was designed by BRC in conjunction with The Campaign for Home Energy  Assistance (see appended questionnaire). After approval of the preliminary draft questionnaire, it was pre-tested with a randomly selected cross-section of 20 United State residents. The pre-test focused on the value and understandability of the questions, adequacy of response categories, questions for which probes were necessary, and the like.

All of the interviewing on this project was conducted between June 25 and July 1, 1998 at the Center's central location computer-assisted telephone interviewing (CATI) facility where each interviewer worked under the direct supervision of BRC supervisory personnel. Behavior Research Center uses the ACS-QUERY CATI system. The CATI system is a computer controlled interview that uses a tightly-integrated branching pattern to control cuing and display of contingent questions. This system allows for a more relaxed interview environment, while reducing the risks of coding error typically found with hard copy survey instruments. The system also performs internal consistency checks on survey variables and prompts interviewer staff to ask probe questions or clarify answers.

All of the interviewers who worked on this project were professional interviewers of the Center. Each had prior experience with BRC and received a thorough briefing on the particulars of this study. During the briefing, the interviewers were trained on: (a) the purpose of the study; (b) sampling procedures; (c) administration of the questions; and (d) other project-related factors. In addition, each interviewer completed a set of practice interviews to ensure that all procedures were understood and followed.
 
Interviewing on this study was conducted during an approximately equal cross-section of evening and weekend hours. This procedure was followed to ensure that all households were equally represented, regardless of work schedules. Further, during the interviewing segment of this study, up to four separate attempts, on different days and during different times of day, were made to contact each selected resident. Only after four unsuccessful attempts was a selected household substituted in the sample. Using this methodology, the full sample was completed, and partially completed interviews were not accepted nor counted toward fulfillment of the total sample quotas.

One hundred percent of the completed interviews were edited, and any containing errors of administration were pulled, the respondent re-called, and the errors corrected. In addition, 15 percent of each interviewer's work was randomly selected for validation to ensure its authenticity and correctness. No problems were encountered during this phase of interviewing quality control.

As the data collection segment of this study was being undertaken, completed interviews were turned over to BRC's in-house coding department. The coding department edited, validated and coded the interviews. Upon completion of coding, a series of validity and logic checks were run on the data to insure it was "clean" and representative of the sample universe.

When analyzing the results of this survey, it should be kept in mind that all surveys are subject to sampling error. Sampling error, stated simply, is the difference between the results obtained from a sample and those which would be obtained by surveying the entire population under consideration. The size of sampling error varies, to some extent, with the number of interviews completed and with the division of opinion on a particular question.
An estimate of the sampling error range for this study is provided in the following table. The sampling error presented in the table has been calculated at the confident level most frequently used by social scientists, the 95 percent level. The sampling error figures shown in the table are average figures that represent the maximum error for the sample bases shown (i.e., for the survey findings where the division of opinion is approximately 50%/50%). Survey findings that show a one-sided distribution of opinion, such as 70%/30% or 90%/10%, are usually subject to slightly lower sampling tolerances than those shown in the table.

The overall sampling error for this study is approximately +/- 3.5 percent when the sample is studied in total (i.e., all 810 cases). However, when subsets of the total sample are studied, the amount of sampling error increases based on the sample size within the subset.