Assessing interchangeability at cluster levels with multiple-informant data.

Area of Science:

• Social epidemiology
• Statistical modeling
• Public health research

Background:

• Studies on neighborhood social disorder and health frequently use multiple informants.
• A key assumption is the interchangeability of latent constructs from multi-informant data.
• Current methods struggle to distinguish individual-level uncertainty from neighborhood-level uncertainty.

Purpose of the Study:

• To propose a novel statistical model for analyzing multi-informant data in neighborhood health studies.
• To accurately delineate uncertainty at individual and neighborhood levels.
• To address limitations in existing methods for assessing neighborhood effects on health.

Main Methods:

• Development and application of a multilevel variance component factor model.
• Utilizing data from a representative sample of children in Detroit and surrounding suburbs.
• Comparing the proposed model with existing methodologies.

Main Results:

• Informant-level models can overestimate neighborhood effects due to individual-level variations.
• The proposed multilevel model effectively distinguishes between individual and neighborhood influences.
• The new model provides a more accurate assessment of social disorder's impact on health.

Conclusions:

• The multilevel variance component factor model is recommended for studies aggregating multi-informant data at the neighborhood level.
• Accurate measurement of neighborhood effects requires models that account for individual differences.
• This approach enhances the reliability of findings in social epidemiology and public health.