Dmitry Arkhangelsky,Guido W. Imbens,Lihua Lei,Xiaoman Luo
Abstract:We propose a new estimator for average causal effects of a binary treatment with panel data in settings with general treatment patterns. Our approach augments the popular two-way-fixed-effects specification with unit-specific weights that arise from a model for the assignment mechanism. We show how to construct these weights in various settings, including the staggered adoption setting, where units opt into the treatment sequentially but permanently. The resulting estimator converges to an average (over units and time) treatment effect under the correct specification of the assignment model, even if the fixed effect model is misspecified. We show that our estimator is more robust than the conventional two-way estimator: it remains consistent if either the assignment mechanism or the two-way regression model is correctly specified. In addition, the proposed estimator performs better than the two-way-fixed-effect estimator if the outcome model and assignment mechanism are locally misspecified. This strong double robustness property underlines and quantifies the benefits of modeling the assignment process and motivates using our estimator in practice. We also discuss an extension of our estimator to handle dynamic treatment effects.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to estimate the average causal effect of binary treatment variables in panel data, especially in the case of complex treatment patterns. The traditional two - way fixed effects (TWFE) regression method may be inaccurate or not robust in some cases, especially when the treatment pattern is not a simple staggered adoption. The paper proposes a new estimator, which enhances the TWFE model by combining unit - specific weights derived from modeling the assignment mechanism. Specifically, the paper addresses the following issues:
1. **Improving the Robustness of Estimation**: The traditional TWFE estimator may be inconsistent when the assignment mechanism or the fixed - effect model is mis - specified. The method proposed in the paper can maintain consistency even when one of the models is mis - specified, thereby improving the robustness of the estimation.
2. **Handling Complex Treatment Patterns**: The paper pays special attention to the case of staggered adoption, that is, units enter the treatment state sequentially and permanently. In this case, the traditional TWFE method may not be able to provide accurate estimates. The method proposed in the paper can better handle such complex patterns by constructing unit - specific weights.
3. **Extension to Dynamic Treatment Effects**: In addition to static treatment effects, the paper also discusses how to extend the estimator to handle dynamic treatment effects, that is, the situation where treatment effects change over time.
4. **Theoretical Contribution**: The paper provides a strict theoretical analysis, proving the consistency and effectiveness of the new estimator when the assignment mechanism is correctly specified, and performing better when the assignment mechanism or the regression model is locally mis - specified.
### Formula Summary
- **TWFE Regression Model**:
\[
Y_{it}=\mu+\alpha_i+\lambda_t+\beta^{\top}X_{it}+\tau W_{it}+\epsilon_{it}
\]
where \(Y_{it}\) is the dependent variable of interest, \(W_{it}\) is a binary treatment variable, \(X_{it}\) is an observed exogenous characteristic, and \(\tau\) is the main parameter of interest.
- **Weighted TWFE Estimator**:
\[
\hat{\tau}(\gamma)=\arg\min_{\tau,\mu,\alpha_i,\lambda_t,\beta}\sum_{i,t}(Y_{it}-\mu-\alpha_i-\lambda_t-\beta^{\top}X_{it}-\tau W_{it})^2\gamma_i
\]
where \(\gamma_i\) is a unit - specific weight calculated according to the assignment model.
- **DATE Equation**:
\[
\mathbb{E}_{W\sim\Pi}[(\text{diag}(W)-\xi W^{\top})J(W - \mathbb{E}_{W\sim\Pi}[W])]=0
\]
where \(\xi\) is a user - specified time weight, \(J = I_T-\frac{1}{T}1_T1_T^{\top}\), and \(\Pi\) is a reshaping distribution.
### Conclusion
The new estimator proposed in the paper improves the robustness and accuracy of estimating the average causal effect in complex treatment patterns by combining the modeling of the assignment mechanism. This method is not only effective when the assignment mechanism is correctly specified, but also performs well when the assignment mechanism or the regression model is locally mis - specified. This provides a powerful tool for causal inference in practical applications.