B1422
Title: Finding alignment between social media and survey responses
Authors: Frederick Conrad - University of Michigan (United States) [presenting]
Abstract: It may sometimes be possible to substitute data from social media for survey responses, e.g., in place of a survey wave. The experience is reported measuring alignment, i.e., the corresponding change over time between survey responses and social media posts, a precondition for substitution. The testbed consists of (1) a data set, the Census tracking survey, a nonprobability web survey that measured American public opinion about the US Census Bureau and the 2020 census, and (2) tweets about the Census Bureau and its studies. There is no reason to expect alignment for all or even most survey questions. Nonetheless, alignment is found for some survey questions, some of which are only evident when certain filters are applied to the social media corpus. The experience is reported by selecting tweets that are semantically related to a survey question and further filtering tweets based on the stance opinion they express so that comparing patterns of posts with patterns of survey responses is apples to apples. The value of applying these filters is also discussed when the goal is not to find alignment per se but to understand public opinion better.
