Geographical Screening in Cell Phone Surveys: Sampling Variability and Bias in the March 2011 National Flu Survey

• Activity Number: 44
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Nicholas Davis*+ and Tammy Santibanez and Lin Liu and Kennon R. Copeland
• Companies: NORC and CDC and NORC and NORC
• Keywords: Survey ; Cell Phone ; Design Effect ; Bias ; Geography

Abstract

Surveys of cell-phone users targeting specific geographical areas are subject to inaccuracy in the estimated location of survey respondents. The county of cell-phone activation or "wire center county" on which many cell-phone sampling frames are stratified has an estimated county-level error rate of 62% and a state-level error rate of approximately 10%. Survey estimates based on sampling area are subject to bias due to the inclusion of outside residents, and survey estimates based on true area of residence with no screening are subject to increased sampling error when the probability of selection differs by sample area. In this paper we introduce a method for geographically screening cell-phone survey respondents, and describe the effects of alternative methods on sampling variability (design effects) and bias in survey estimates. We test our hypotheses using data collected by the March 2011 National Flu Survey, conducted by NORC at the University of Chicago on behalf of the Centers for Disease Control and Prevention. The survey utilized a dual-frame stratified sample designed to produce estimates of influenza vaccination rates nationally and among 20 sub-state sampling areas.