Premium
When Should We Use Unit Fixed Effects Regression Models for Causal Inference with Longitudinal Data?
Author(s) -
Imai Kosuke,
Kim In Song
Publication year - 2019
Publication title -
american journal of political science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.347
H-Index - 170
eISSN - 1540-5907
pISSN - 0092-5853
DOI - 10.1111/ajps.12417
Subject(s) - causal inference , matching (statistics) , econometrics , directed acyclic graph , outcome (game theory) , estimator , inference , nonparametric statistics , causal structure , identification (biology) , regression , causal model , computer science , fixed effects model , confounding , mathematics , panel data , statistics , artificial intelligence , algorithm , physics , botany , mathematical economics , quantum mechanics , biology
Many researchers use unit fixed effects regression models as their default methods for causal inference with longitudinal data. We show that the ability of these models to adjust for unobserved time‐invariant confounders comes at the expense of dynamic causal relationships, which are permitted under an alternative selection‐on‐observables approach. Using the nonparametric directed acyclic graph, we highlight two key causal identification assumptions of unit fixed effects models: Past treatments do not directly influence current outcome, and past outcomes do not affect current treatment. Furthermore, we introduce a new nonparametric matching framework that elucidates how various unit fixed effects models implicitly compare treated and control observations to draw causal inference. By establishing the equivalence between matching and weighted unit fixed effects estimators, this framework enables a diverse set of identification strategies to adjust for unobservables in the absence of dynamic causal relationships between treatment and outcome variables. We illustrate the proposed methodology through its application to the estimation of GATT membership effects on dyadic trade volume.