How Voter Registration Records Sharpen MRP Estimates to County and District Levels

🔎 What's Changing in Survey Research

Declining telephone response rates have pushed survey practice toward cell-phone supplements, nonprobability samples, and greater reliance on model-based inferences. At the same time, new statistical methods and large administrative data sources open opportunities to address some of these challenges.

📊 How Voter Registration Databases Are Used with MRP

Voter registration databases provide political variables at scale and without the typical survey biases that complicate analysis. These databases enable methodologists to combine external administrative information with survey data to improve post-stratified estimates.

Key features of these databases:

• Party registration and past voting behavior available at individual or aggregate levels
• Very large coverage across registered voters, enabling projection to full voter rolls
• Free from overreporting bias or endogeneity that can affect self-reported survey responses

🧭 Illustration: Producing 2012 Vote-Choice Estimates for 195 Million Voters

A general process is developed to integrate voter files with multilevel regression and poststratification (MRP). This approach is illustrated by producing vote-choice estimates for the 2012 presidential election and projecting those estimates to 195 million registered voters in a postelection context.

Findings from the illustration:

• Estimates remain stable and reasonable for demographic subgroups within small geographies
• Reliable inference extends down to county and congressional district levels
• These projections can supplement exit polls, which have become increasingly problematic and are not available in all geographies

⚠️ Problems, Limits, and Open Research Questions

The approach offers clear advantages but also raises important caveats. Limitations include concerns about coverage (only registered voters), variable availability across jurisdictions, and further methodological work needed to handle remaining biases and integration challenges. Several open areas for research are discussed to guide future improvements and broader application.