Examining the bias in landline only surveys: How does the cell phone only population differ from the landline population on health indicators, and are estimates from landline surveys biased?

Method: Data from a RDD survey of cell phone only respondents in Colorado were merged with data from the Colorado Behavioral Risk Factor Surveillance System (BRFSS) corresponding to the same data collection period. We compared point estimates from the BRFSS only data, using BRFSS sample weights, with estimates from the merged data, which were adjusted for the probability of selection for both landline and cell phone only respondents, and poststratified by age, race/ethnicity, and sex.

Results: The estimated prevalence of ever having had an HIV test increased by 3.6 percentage points (38.3% vs. 41.9%) when including the cell phone only population. The bias was more severe among non-whites (39.4% vs. 45.7%), 18-34 year olds (39.3% vs. 47.0%), and those making less than $50,000 (39.1% vs. 45.3%). Similarly, the smoking prevalence was 1.3 percentage points higher after merging cell phone only population. The bias was more severe among non-whites (22.5% vs. 25.8%) and 18-34 year olds (23.7% vs. 26.5%). The change in point estimates for number of days with poor mental or physical health, and having health care coverage were small, but may not be inconsequential.

Conclusion: Our results provide evidence that RDD surveys that exclude the cell phone only population are biased for some health indicators, and that this bias is most severe among groups with high wireless substitution rates (e.g., young adults, non-whites).