Regression-based negative control of homophily in dyadic peer effect analysis

Area of Science:

• Social epidemiology
• Biostatistics

Background:

• Peer effects and social contagion are studied in social science.
• Homophily bias, where similar individuals connect, challenges causal inference in these studies.
• Previous analyses of health conditions (e.g., obesity) and psychological states (e.g., happiness) spreading through social networks are criticized due to homophily bias.

Purpose of the Study:

• To develop a novel regression-based approach to identify and estimate contagion effects.
• To address and mitigate the impact of homophily bias in social science research.
• To evaluate the peer effect of obesity using the developed methodology.

Main Methods:

• A regression-based approach is developed.
• The method utilizes a negative control exposure for robust identification and estimation.
• Contagion effects are estimated on both additive and multiplicative scales.

Main Results:

• The proposed method effectively accounts for homophily bias.
• Contagion effects can be accurately estimated in the presence of homophily.
• The peer effect of obesity was evaluated in the Framingham Offspring Study.

Conclusions:

• The developed regression approach provides a valid method for estimating contagion effects.
• This approach allows for more reliable causal inference regarding peer influences.
• The study successfully applied the method to demonstrate its utility in real-world data analysis.