Collaborative Research: Smooth National Measurement of Public Opinion Across Boundaries and Levels: A View From the Bayesian Spatial Approach

点击翻译按钮获取中文摘要

This research project will measure public opinion in voting constituencies around the United States. The project will provide estimates of opinion in districts with little or no survey data, such as state legislative districts. The project's intellectual merit comes from establishing a new means for measuring public opinion that not only uses survey respondents' answers to polling questions but also incorporates important information about where respondents are located and what that implies about geographic patterns in public opinion. Coupled with population information from the U.S. Census, the project will produce stronger estimates of public sentiment when survey data are sparsely distributed. The investigators will release free software that includes user-friendly functions allowing any citizen to determine public opinion in his or her own district or in districts that have not yet cast any votes, such as proposed congressional districts in a redistricting cycle. The software will allow more sophisticated users to obtain measures for any variable (even if unrelated to public opinion) in relationship to geographic boundaries, which will have extensions to research in public health, epidemiology, economics, sociology, business, and law. The broader impact to society will be that the data and software from this project will provide more information for the news media, the public, and elected officials regarding the outlook of the nation by constituency and locale, thereby providing a better understanding of the American representation process. The project will recruit a diverse group of research assistants that will be trained in this kind of statistical analysis.Studies relating to public opinion often settle for less-than-ideal data. Frequently, researchers will measure public opinion in the 50 states or the 435 congressional districts by pooling together several surveys taken over time (losing a sense of change over time), using old measures of public opinion (which may not be consistent with current public views), or using presidential vote share to approximate public sentiment (which is prone to error because factors besides ideology affect vote choices). With smaller districts than these, such as state legislative districts, the problem is magnified considerably, because it is rare to have many survey respondents in such a small area. In this project, the investigators ask: How can survey responses and the geographic location of the respondents be used to reliably forecast constituency public opinion? To answer this question, the investigators will use the method of Bayesian universal kriging. This technique fits a training model over survey data to determine how demographic factors shape public opinion and how the portion of survey responses that cannot be explained by demographics can be explained by a geographically smoothed process. With a model like this, public opinion in constituencies can be predicted with known population demographics and the values of the geographically smoothed error process over that district.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.