New research published in the Journal of Survey Statistics and Methodology suggests approaches to improve how polling uncertainty is measured, giving decision-makers a clearer sense of the precision of poll results. By introducing a “margin of total error” that includes nonresponse and measurement bias, which can often be larger than sampling error, the approach offers a more realistic view of poll accuracy. Westat’s J. Michael Brick, PhD, was among the coauthors of the article.
The study finds that regression and mixed-effects models best predict total error using benchmark datasets like election results. Other methods, such as simply doubling the margin of error, work inconsistently and perform poorly outside of election polling.
These models allow pollsters to tailor uncertainty estimates based on similar past surveys, leading to more transparent and reliable polling insights.
Learn More
Toward a Margin of Total Error
Authors: Sharon L. Lohr, Andrew Mercer, Courtney Kennedy, J. Michael Brick