Geographic Inaccuracy of Cellphone Samples

Abstract

The geographic accuracy provided in cellphone samples has been an issue of particular concern for telephone surveys. Surveyors use this information for geographic sample stratification and analysis; however, the location generated using random digit dialing (RDD) methods is typically the rate center in which the phone was activated and not the owner’s current residence. Researchers have found the location is fairly accurate at the regional (census region) level. However, the error rate increases when moving to the state and local levels. Cellphone error rates are higher among ‘cell-only’ (cellphone users who do not have a landline telephone) respondents than for ‘dual-users’ (cellphone users who have a cellphone and landline telephone). Among respondents sampled from the landline frame, the error rate (difference between location on the sample file and current location of residence) is smaller. Past research estimates the state error rate is about 1% for a landline sample and 12% for the ‘cell-only’ group (Skalland and Khare 2013). This paper builds upon previous research by providing state level estimates of inaccuracy for both ‘cell-only’ and ‘dual-users.’ It also examines the trend of state-level inaccuracy estimates over 18 months using data from the Gallup Daily Tracking survey. Finally, we examine the characteristics of ‘cell-only’ and ‘dual-users’ and implications of geographic inaccuracy of cellphone samples on the state level dual-frame RDD estimates.