Linking Survey Items to Real Preferences: Hierarchical Models That Reveal Polarization

🔎 What This Model Does

Presents a class of hierarchical item response models that integrate measurement and substantive analysis of multi-item opinion surveys. Individual item responses are modeled as arising from a latent preference whose mean and variance may depend on observed covariates.

🧭 How Responses Are Modeled

• Individual responses to multiple items are treated as manifestations of a single latent preference.
• Both the latent preference mean and its variance can be functions of observed covariates, allowing heterogeneity in levels and dispersion across groups.
• The hierarchical structure links item-level measurement to higher-level analysis in one unified model.

📈 Advantages Compared to Two-Step Approaches

• Reduces bias relative to the common two-step practice of first constructing a composite and then analyzing it.
• Increases statistical efficiency in estimating relationships between covariates and latent attitudes.
• Facilitates direct comparison across surveys that cover different sets of items.

🔬 What This Enables Researchers To Investigate

• How preferences differ among social and demographic groups
• Regional variation and how preferences evolve over time
• Levels, patterns, and trends of attitude polarization
• Degrees and patterns of ideological constraint

🛠 Software Availability

An open-source R package, hIRT, is available for fitting the proposed hierarchical item response models.

⚖️ Why It Matters

By integrating measurement and analysis, this approach yields less biased, more efficient inferences about public opinion and makes cross-survey comparisons and studies of polarization and constraint more reliable.