Small Area Estimation by Combining Two Surveys with Special Consideration of Cell Only Households

• Activity Number: 437 - Session in Honor of Jim Lepkowski's Retirement
• Type: Invited
• Date/Time: Wednesday, August 2, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Dawei Xie* and Qiang Pan and Van Parsons and Trivellore Raghunathan and Benmei Liu and Eric Feuer and Nathaniel Schenker
• Companies: University of Pennsylvania and and National Center for Health Statistics and University of Michigan School of Public Health and National Cancer Institute and National Cancer Institute and Retired
• Keywords: small area estimation ; BRFSS ; NHIS ; cell-only households

Abstract

Cancer surveillance research requires estimates of the prevalence of cancer risk factors and screening for small areas such as states and counties. To obtain such estimates, Raghunathan et al (2007) has developed an approach to combine two national surveys, the BRFSS and NHIS where the population in each small area was divided into two strata, households with landline telephones and those without. Estimates from 1997-2000 were obtained from a Bayesian hierarchical model. The potential noncoverage and noresponse biases in the BRFSS were took into account. When applying this approach to newer data in the year of 2004-2006, we had to consider another stratum of the population, the cell-only households. We extended the above approach to obtain estimates for three sub-populations for each small area and weighted them by the proportions in each sub-population. The proportion of cell only households, landline households, and no landline or cell phone households for the small areas were obtained and the variances of these estimates were incorporated into the final small area estimates on cancer risk factors and screening via Markov Chain Monte Carlo method.