When Voters Don't Add Up: Non-Separable Preferences Change Vote Models

🔎 The Problem

Most multidimensional spatial models specify the systematic component of voter utility as additively separable. That assumption implies voters do not care how positions on multiple policy dimensions combine—an implication that is too restrictive in the context of mass elections and can mischaracterize vote choice.

🧭 Research Design and Model

A statistical implementation of Davis, Hinich, and Ordeshook's (1970) Weighted Euclidean Distance model is introduced to relax separability. This implementation allows estimation of both the direction and the magnitude of non-separability directly from vote choice data.

🧪 What Was Tested

• A Monte Carlo experiment evaluates how conventional separable specifications perform when true preferences are non-separable.
• Three empirical applications test for non-separability in voter preferences across economic and socio-cultural issue dimensions.

📊 Key Findings

• Conventional separable model specifications yield biased and/or unreliable estimates of the effect of policy distances on vote choice probabilities when preferences are non-separable.
• In three empirical applications, voter preferences over economic and socio-cultural issues are found to be non-separable.
• Failing to account for non-separability risks omitting crucial variation and distorting inference about how policy distances influence votes.

💡 Why It Matters

Checking for non-separability should be a routine part of robustness testing in empirical applications of multidimensional spatial models. The findings have implications beyond voting studies: any field that uses spatial specifications may face bias if non-separable preferences or utilities are ignored.