How Bayesian Scaling Fixes Bias in Aldrich–McKelvey Ideology Estimates

Information about the ideological positions of political actors is central to questions of representation, polarization, and voting behavior. Surveys that ask respondents to place actors on a common ideological scale are common, but respondents often introduce systematic biases into those placements.

🔍 What the Paper Examines

The paper focuses on two common respondent-level distortions in scaling: rationalization bias and differential item functioning (DIF). Aldrich–McKelvey (AM) scaling provides a widely used correction for DIF but does not account for rationalization bias. As a result, AM-type approaches can produce misleading estimates when rationalization bias is present.

🧪 How the Claim Is Tested

• Monte Carlo simulations are used to evaluate performance of AM-type models versus an alternative approach.
• The alternative is a unified Bayesian scaling model that simultaneously estimates DIF and rationalization bias.

📈 Key Findings

• Respondents introduce two distinct problems when placing actors on a common ideological scale: rationalization bias (systematic distortion toward respondents' own views) and DIF (items functioning differently across respondents).
• AM scaling effectively addresses DIF but ignores rationalization bias.
• Monte Carlo evidence shows AM-type models can give inaccurate results when rationalization bias exists.
• The Bayesian scaling approach that jointly models DIF and rationalization bias performs better than AM-type models in the presence of rationalization bias.

💡 Why It Matters

Better measurement of actors' ideological positions matters for inferences about political representation, polarization, and voting behavior. A unified Bayesian scaling approach reduces bias from respondent behavior and yields more reliable ideological estimates when rationalization tendencies are present.