|
Abstract:
|
Address-Based Sampling (ABS) frames are well suited for linking to auxiliary data, either by geocoding frame addresses or by direct address matching. Survey researchers commonly append publicly available census data to ABS frames. Additionally, the frame can be linked to proprietary data sources compiled and sold by large data brokers such as Experian, Acxiom, Epsilon, and CoreLogic. When these auxiliary data are appended to the ABS frame and combined with reported values and response paradata from a previous survey, researchers can develop address-level models that predict demographic characteristics and respondent behaviors, helpful in sample designs for surveys of targeted subpopulations. These models can be used for stratification, decisions about data collection protocols, and weight adjustments, but this paper focuses on stratification and sampling for eligible subpopulations. This paper reviews the general principles of predictive modeling with ABS and gives examples from our experience.
|
