Dual-Frame Weighting Of RDD And Cell Phone Interviews At The Local Level Michael P. Battaglia 1 , Donna Eisenhower 2 , Stephen Immerwahr 2 Kevin Konty 2 , 1 Abt Associates, 55 Wheeler St., Cambridge, MA 02138 2 New York City DOHMH, Epidemiology Services, 125 Worth St., New York, NY 10013 Abstract National dual frame telephone surveys use National Health Interview Survey estimates of the size of telephone service groups. How does one go about weighting a local-level dual frame survey? Telephone service control totals were calculated in two ways. A model- based method fit a multinomial regression model to data from the NHIS. The model was applied to New York City data from the American Community Survey to estimate telephone service totals. A second approach used direct estimates from the NYC Housing and Vacancy Survey, an in-person survey. We compared estimates for health risk factors. Differences in overall and subgroup prevalence were small. The results suggest that the model-based weighting approach yields prevalence estimates that are close to the estimates based on weighting to local independent estimates. Implications for local dual frame surveys are discussed. Key Words: Dual Frame Sample Design, Poststratification 1. Introduction Dual frame telephone sample designs are now widely used to provide coverage of adults with telephone service in the U.S. (Srinath et al. 2004). In an overlapping dual frame design a list-assisted random-digit-dialing sample is drawn from the landline sampling frame. A random sample of 10-digit telephone numbers is also drawn from dedicated cellular telephone exchanges. For households sampled in the landline sample one adult is typically randomly selected. In the cell phone sample the cell phone is treated as a personal communication device and the adult using the phone is interviewed, or one adult is randomly selected from among the adults in the household that share the cell phone. All adults in the cell phone sample are eligible to be interviewed. A non-overlapping design follows the same approach except that in the cell phone sample only adults who only have a cell phone are eligible to be interviewed. For both types of dual frame designs the final step in the weighting methodology often involves poststratification to socio-demographic control totals and telephone service control totals. For national surveys the telephone service control totals are generally obtained from the NHIS for three categories: cell-only adults, landline-only adults, and dual service (landline and cell service) adults. Poststratification to control totals in telephone surveys is used to reduce nonresponse bias and noncoverage bias. The ACS or the Current Population Survey can be used to develop control totals for socio- AAPOR 5816