Individual Submission Summary
Share...

Direct link:

Ideological Bias Correction Using Ordered Bayesian Aldrich-McKelvey Scaling

Sat, September 1, 2:00 to 3:30pm, Hynes, 204

Abstract

Whether the mass public in the United States is polarized (Abramowitz & Saunders 2008) or not (Fiorina & Abrams 2008) is a contested issue within political science without a clear answer. Much of this debate stems from the theoretical mismatch between how we think about polarization and how we measure it. Thus to answer the question “is the mass public ideologically polarized?” we need to consider the theoretical link between ideology as a latent construct and ideology manifested as survey responses. The Bayesian implementation of Aldrich-McKelvey scaling as a way to correct survey response bias due to differential-item functioning (DIF) by Hare et al. (2015) improves the estimation of liberal-conservative scale positions of respondents and political stimuli. Currently, much of the work in DIF assumes a continuous manifest set, however the scales commonly used to measure these political stimuli are ordinal Likert scales. We propose specifying the ideological scale as an ordered categorical variable, which results in a better match between the manifest set of responses and the model specification of latent scale. We contrast our Ordered Bayesian Aldrich-McKelvey (OBAM) approach to the standard continuous approach using simulated data with different levels of polarization, which will show the benefits of proper model specification. We also will apply OBAM to CCES and ANES data, since these data sets are commonly used in political science to answer questions regarding ideology and polarization in American politics. This allows us to reassess claims of polarization among the American public through a better remedy for the bias in survey response.

Authors