• Presentation

Evaluating and Modeling Early Cooperator Bias in RDD Surveys

Citation

Biemer, P., & Link, M. W. (2006, January). Evaluating and Modeling Early Cooperator Bias in RDD Surveys. Presented at Second International Conference on Telephone Survey Methodology, Miami, FL.

Abstract

Response rates to RDD telephone surveys continue to plummet and methods for increasing them have been largely ineffective. Now survey methodologists are focusing on the other component of nonresponse bias – viz., the difference between respondents and nonrespondents – and developing approaches for minimizing these differences in the achieved sample. Recent research suggest the early cooperator bias (ECB) may be small for RDD surveys; i.e., persons who cooperate with fewer call attempts do not differ appreciably from those who require many more call attempts to be contacted and interviewed. This would explain why a survey with a 40% response rate, representing mostly early cooperators, would show no greater nonresponse bias than the same survey pushed to a 70% response rate. If the ECB is negligible for the key characteristics in a survey, then there should be no reason to continue to expend resources trying to achieve a higher response rate. To that end, this paper examines the ECB for a number of health characteristics using data from the Behavioral Risk Factor Surveillance Survey and for a range of truncated callback rules (i.e., the maximum number of call attempts to reach a final disposition for a telephone number). To limit the risk of a large ECB in some estimates, we propose a weight adjustment scheme for reducing the ECB which does not rely on external auxiliary data. Rather, the approach is based upon a model for the ECB that takes into account the number of call attempts required to reach a final disposition for a case. An evaluation of the performance of the ECB adjustment method for the BRFSS data is also presented.