Framing Inference from Mixed-Mode Surveys Using Causal Inference Framework

• Activity Number: 159
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Trivellore Raghunathan*+
• Companies: University of Michigan
• Keywords: Bayesian Inference ; Potential Outcomes ; Nonresponse ; Measurement Error ; Panel Surveys ; Mixed Modes

Abstract

To improve coverage and response rates, many surveys employ designs where the survey instruments are administered using a mixed modes such as Mail, Web, Telephone and In-person. In some designs, the modes are changed as a part of refusal conversion process. What is the appropriate approach to analyze data from such mixed mode designs? One may decide to ignore the modes, if the measurement properties are similar across the modes but, what if there are mode differences? How should one construct inferences for the population quantities based on the data collected from such survey data? This paper proposes a causal inference framework where we generate potential populations under each mode and then combine these potential populations to form a single inference. A Bayesian framework is used develop inferences and thus fully account for differences in the mode effects.The data from a longitudinal survey that used single mode design in first wave and mixed mode design in subsequent two waves will be used to illustrate the methodology. Data will be useful to highlight issues where modes may affect differentially across variables. The methodology is evaluated using a simulation study.