Unignorable: Tackling Non-ignorable Non-response in Survey Research
Michael Bailey
Half Day, 1:30 PM – 5:30 PM
Los Angeles Convention Center, 410
Survey researchers typically deal with non-response via weighting, quota sampling and multilevel regression and post-stratification. These tools are powerful, but do not address non-ignorable non-response, the kind of response that occurs when non-response is directly related to the content being surveyed.
Ironically, non-ignorable non-response is often ignored, a pattern this course seeks to counteract by exploring survey research through the lens of non-ignorable non-response. This entails understanding first how ignorable and non-ignorable non-response have been important in the history of polling, including in the highly fluid contemporary era.
Second, this involves thinking deeply about why non-ignorable non-response poses such dangers for polling, especially modern polling that is typically based either on opt-in internet samples or random samples with very low response rates.
The course ends on a constructive note. We need not be passive or fatalistic in the face of potential non-ignorable non-response. There is a broad and growing toolkit for dealing with non-ignorable non-response. Using this toolkit makes new demands on the data, but not unreasonable ones.
The goal is that participants emerge with a stronger understanding of this important potential source of survey error and a grasp of the tools to help tame it.
Course Objectives
1) Understand the history of polling up to contemporary era in light of ignorable and non-ignorable non-response.
2) Understand the intuition behind the distinctive nature of non-ignorable non-response bias, why weighting and related tools do not address it and the type of data needed to diagnose and counteract it.
3) Be introduced to a statistical toolkit for diagnosing and reducing or eliminating non-ignorable non-response bias.