Are telephone surveys with low response rates usable?

Telephone surveys are an efficient mechanism for large scale data collection. Unfortunately, they are frequently plagued by low response rates and the associated potential for non-response bias. This study presents a theory based method to assess the impact of telephone survey non-response. English speaking women 18 to 44 years of age who were receiving services at a federally funded clinic in a small New York State city were recruited for participation in a women's health study. At the end of the study, the women were asked if they would participate in a future telephone survey about health and health behaviors if randomly telephoned by researchers. Based on their answer, women were classified as potential telephone survey responders or potential survey non-responders. Measures of association (specifically prevalence ratios and associated confidence intervals) between respondent characteristics and health behaviors were calculated using the whole sample and the potential responder sub-sample. Measures of association did not vary when calculated with data from the whole sample or the potential responder sub-sample. These results suggest that non-response may not significantly impact the ability to assess associations related to general health behaviors within a group. As all studies have less than perfect response rates, using diverse methods to better understand non-response bias is needed. Given the difficulty of collecting information about health behaviors and characteristics from persons who are unwilling to participate in a telephone survey, studies of intent provide an additional method to assess and understand the impact of non-response on results from survey research.