Abstract

Ideal point estimation methods face a significant challenge whenlegislators engage in protest voting - strategically voting againsttheir party to express dissatisfaction. Such protest votes introduceattenuation bias in conventional estimators, making ideologicallyextreme legislators appear artificially moderate. We propose a novelstatistical framework called emRIRT (EM-based robust IRT method) thatextends the fast EM-based estimation approach ofImai, Lo and Olmsted (2016) to handle protest votes. Our methodintroduces shift parameters for all votes, effectively employing l0regularization to systematically identify protest votes. Throughsimulation studies, we demonstrate that emRIRT maintains estimationaccuracy even with high proportions of protest votes, while beingsubstantially faster than MCMC-based methods. Applying our method tothe 116th U.S. House (2019-2020), we successfully recover the extremeliberal positions of progressive Democrats known as "the Squad", whose protest votes had caused conventional methods to misclassifythem as moderates. For instance, while conventional methods rankOcasio-Cortez as more conservative than 69% of Democrats, our methodplaces her firmly in the progressive wing, aligning with herdocumented policy positions. This approach provides both robust idealpoint estimates and systematic identification of protest votes,facilitating deeper analysis of strategic voting behavior inlegislatures.