Estimating the Bias Resulting from the Exclusion of Cell Phone-Only Respondents
Levine, B., Davies, C., & Shook-Sa, B. E. (2011, July). Estimating the Bias Resulting from the Exclusion of Cell Phone-Only Respondents. Presented at JSM 2011, Miami Beach, FL.
More than thirty percent of the population has zero probability of selection in most landline RDD surveys due to the exclusion of cell phone only households, zero banks, and households with no phone service. This coverage error is problematic since cell phone only respondents are different than landline respondents; thereby, converting coverage error into coverage bias. To mitigate the coverage error many telephone surveys have moved to a dual-frame design consisting of both cell phone and landline numbers. For surveys that implement a repeated cross-sectional design, comparing estimates of population parameters before and after the change from a landline RDD to a dual-frame RDD can be problematic because differences in estimates might be a consequence of the different sampling frames. This paper presents methodology used to adjust for bias caused by the exclusion of cell phone only respondents; thereby, enabling the comparison of estimates from repeated cross-sectional studies that change methodology from exclusively sampling from landlines to sampling from both landlines and cell phones.