Presented at the 1996 Annual Meetings of the American Association for Public Opinion Research, Salt Lake City, UT, May 17, 1996.
Implementation of an RDD telephone survey requires decisions about how to recruit a respondent from eligible household members. To guide such methodological decisions, we develop a behavioral model for response rates and related quantities (e.g., gender mix) based on method and characteristics of potential respondents. We use data from a random assignment experiment conducted within a local health behavior survey to specify parts of this model. We contrast three common selection methods: interviewing the first available adult, selection of the household member with the most recent birthday, and randomized choice of a household member after an inventory of adults.
Our analysis of 1,100 attempted interviews compares various types of refused and completed interviews. How much difference does randomized respondent selection make? Within-household respondent selection can be expected to reduce cooperation by 14%, change the distribution of household types by no more than 3% in any category, reduce the percentage of women by 7%, and increase the proportion of respondents less likely to answer the phone by 30%. We found no significant differences in cooperation between men and women. We did find that hand-offs to new respondents did not occur as frequently in the experiment as the model predicts, and this shortfall varied by gender of respondent and selection method.
We would like to acknowledge the cooperation and support of the University of Virginia Health Sciences Center, Martha Jefferson Hospital of Charlottesville, VA, and the Thomas Jefferson Health District, for allowing their project to be adapted to the purpose described here. The Centers for Disease Control, U.S. Department of Health and Human Services provided technical assistance. Lea Cunningham, Joseph Spear and Douglas E. Loyd made valuable contributions to the execution of this project.