522 – Survey Data Collection: Going Online or Going Off-Track? Exploring Statistical Issues in Online Nonprobability Panels Research
Compositional Model Inference
J.Michael Brick
Westat
The ability to conduct surveys using opt-in Web respondents has raised concerns about whether these samples are valid. Probability sampling theory is not applicable because the units are not subject to being sampled with a known and non-zero probability of selection. Frameworks have been proposed for Web opt-in surveys, but these generally have features that are not well suited for general-purpose surveys. This paper proposes a model-based framework for making inferences from non-probability samples that we refer to as a compositional approach. The paper outlines the assumptions required for making inferences from these types of samples, and suggests some evaluation measures for assessing the assumptions.