Insights on bias and information in group-level studies

Area of Science:

• Epidemiology
• Biostatistics
• Social Sciences

Background:

• Group-level studies (e.g., ecological, aggregate) analyze data at a group level.
• These studies often aim to infer effects at the individual level, despite a disconnected analysis and inference level.
• This disconnection introduces unique biases, including cross-level and model specification bias.

Purpose of the Study:

• To explore biases in group-level studies when inferring individual-level effects.
• To identify conditions for valid cross-level inference.
• To analyze the impact of study design on exposure effect estimation.

Main Methods:

• Conceptual analysis of bias in group-level studies.
• Specification of assumptions for valid cross-level inference.
• Discussion of the interplay between exposure, study design, and bias.

Main Results:

• Cross-level inference is valid under the assumption that between- and within-group exposure effects are equal.
• Group-level analyses rely solely on between-group comparisons for information.
• Models incorporating even minimal within-group information are susceptible to model specification bias.

Conclusions:

• Valid individual-level inference from group-level studies necessitates strong assumptions about exposure effects.
• Model specification bias is a significant concern, particularly when group-level models are not derived from individual-level models.
• Careful consideration of study design and potential biases is crucial for accurate inference in group-level research.