Methods to Account for Classification Error in County Assignment for a Periodic Dual-Frame Telephone Survey
Marcus E. Berzofsky
RTI International
Caroline W. Blanton
RTI International
Kimberly C. Peterson
RTI International
Bo Lu
Division of Biostatistics, Ohio State University
Howard Speizer
RTI International
Timothy Sahr
Ohio Colleges of Medicine Government Resource Center
State based surveys are often designed to make estimates at the county or county-group levels. For cellphone samples no good sub-state identifiers previously existed on the frame requiring survey methodologists to only draw random samples of cell numbers from the entire state; thus, making it difficult to target areas within a state. To overcome this shortcoming, sample vendors have recently been able to identify a cellphone number's rate center and determine the county from which the rate center is located. We demonstrate that rate centers can be used to identify a respondent's county of residence, but may have large classification error rates. These high rates make it difficult to accurately devise a sample allocation using a cellphone frame. However, for periodic surveys, such as the OMAS, we show how past iterations of the survey can be used to estimate the overall classification error rate and the probability that the cellphone owner will actually reside in the county assigned. We propose the Rate Center Plus method which uses rate centers to stratify and allocate the desired respondent sample to counties plus incorporates the classification error rates to improve accuracy.