Full Citation
Title: Dynamic Estimation of Latent Opinion from Sparse Survey Data Using a Group-Level IRT Model
Citation Type: Miscellaneous
Publication Year: 2014
ISBN:
ISSN:
DOI:
NSFID:
PMCID:
PMID:
Abstract: Recent advances in the modeling of public opinion have dramatically improved scholars' ability to measure the public's views on important issues. For instance, Bayesian item-response theory (IRT) models provide a flexible framework for placing survey respondents in a low-dimensional space, while the combination of multilevel modeling and poststrati cation (MRP) improves small-area estimation of public opinion. However, it has been di$fficult to extend these techniques to a broader range of applications due to computational limitations and problems of data availability. In this paper, we develop a new group-level Bayesian IRT model that overcomes these limitations. Rather than estimating opinion at the individual level, we propose a hierarchical IRT model that estimates mean opinion in groups de fined by demographic and geographic characteristics. Opinion change over time is accommodated with a dynamic linear model for the parameters of the hierarchical model. The group-level estimates from this model can be re-weighted to generate estimates for geographic units. This approach has substantial advantages over an individual-level IRT model for the measurement of aggregate public opinion. It is much more computationally efficient and permits the use of sparse survey data (e.g., where individual respondents only answer one or two survey questions), vastly increasing the applicability of IRT models to the study of public opinion and representation. We demonstrate the advantages of this approach for the study of the American public's policy preferences in both the modern and mid-20th century periods. We also demonstrate a potential application of our model for the study of judicial politics.
User Submitted?: No
Authors: Caughey, Devin; Warshaw, Christopher
Publisher: Dept of Political Science, MIT
Data Collections: IPUMS USA
Topics: Methodology and Data Collection
Countries: