11.5 Estimation & Basics 2
- FE estimator identifies average treatment effect on the treated if exogeneity of time-varying idiosyncratic errors can be maintained (Gangl 2010, 34)
- Standard assumptions: Selection only observable time-variant variables (time invariant-variable should not matter) (+ linearity and additivity of treatment effect)(e.g. Keele 2015b, 10)43
- BUT causal inference with panel data (fixed effects models etc.) is an ongoing research field (K. Imai and Kim 2019a, 2019b)
- Assumptions for fixed effects models were not entirely clear (try to follow recent research)
- It was unclear how treatment/control group should be (and are) actually constructed for estimation (see e.g., PanelMatch for recent developments)
- Results were often misinterpreted (see the checklist for the interpretation of fixed effects regression results)
References
Gangl, Markus. 2010. “Causal Inference in Sociological Research.” Annual Review of Sociology 36 (1): 21–47.
Imai, Kosuke, and In Song Kim. 2019a. “On the Use of Two-Way Fixed Effects Regression Models for Causal Inference with Panel Data.” Harvard University IQSS Working Paper). Cambridge, MA. Retrieved from https …; mit.edu.
Imai, Kosuke, and In Song Kim. 2019b. “When Should We Use Unit Fixed Effects Regression Models for Causal Inference with Longitudinal Data?” Am. J. Pol. Sci. 63 (2): 467–90.
Keele, Luke. 2015b. “The Statistics of Causal Inference: A View from Political Methodology.” Polit. Anal. 23 (3): 313–35.