Using unit fixed effects regression models for causal inference with longitudinal data has become a common practice. However, this approach makes two critical assumptions about dynamic relationships between treatments and outcomes that may not always hold.

Data & Methods

The study introduces a novel nonparametric matching framework to analyze how standard unit fixed effects estimators implicitly compare observations across time periods. This methodology establishes equivalence with weighted regression approaches for causal inference in longitudinal settings.

Key Findings

Researchers using these models should be aware of two crucial assumptions:

1️⃣ Past treatments do not directly influence current outcomes

2️⃣ Past outcomes do not affect current treatment status

These implicit comparators can yield misleading results when dynamic relationships exist between variables.

Real-World Implications

The framework demonstrates how standard methods may fail to account for time-varying confounders or feedback effects, potentially leading researchers astray in their causal claims.