Network threats to causal inference: Variations in network position by participation in randomized controlled trials

Area of Science:

• Social epidemiology
• Network science
• Health services research

Background:

• Randomized controlled trials (RCTs) are crucial for causal inference in health research.
• Social networks influence individual behaviors and attitudes, potentially affecting trial generalizability.
• Differences in network positions between trial participants and non-participants are often overlooked.

Purpose of the Study:

• To evaluate the extent and impact of network position variations between RCT participants and non-participants.
• To assess how these network differences affect the generalizability of RCT findings.
• To investigate the influence of social network structure on intervention effect estimation.

Main Methods:

• A workplace-based randomized controlled trial (RCT) at a hospital served as a case study.
• Longitudinal social networks were constructed using employee cafeteria purchase data.
• Stochastic actor-oriented models (SAOMs) analyzed network position differences.
• Computational knockout experiments assessed the impact of network phenomena on effect estimates.

Main Results:

• RCT participants exhibited significantly different social network positions compared to non-participants.
• Participants co-purchased cafeteria items with more colleagues than non-participants.
• These network disparities led to downwardly biased estimates of the intervention's efficacy and effectiveness.

Conclusions:

• Social network structures can significantly impact the generalizability and interpretation of RCT results.
• Network position differences between participants and non-participants can introduce bias in effect estimates.
• Future research should account for social network context in RCT design and analysis.