26.4 Standard Errors
Serial correlation is a big problem in DiD because (Bertrand, Duflo, and Mullainathan 2004)
- DiD often uses long time series
- Outcomes are often highly positively serially correlated
- Minimal variation in the treatment variable over time within a group (e.g., state).
To overcome this problem:
- Using parametric correction (standard AR correction) is not good.
- Using nonparametric (e.g., block bootstrap- keep all obs from the same group such as state together) is good when number of groups is large.
- Remove time series dimension (i.e., aggregate data into 2 periods: pre and post). This still works with small number of groups (See (Donald and Lang 2007) for more notes on small-sample aggregation).
- Empirical and arbitrary variance-covariance matrix corrections work only in large samples.
References
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. “How Much Should We Trust Differences-in-Differences Estimates?” The Quarterly Journal of Economics 119 (1): 249–75.
Donald, Stephen G, and Kevin Lang. 2007. “Inference with Difference-in-Differences and Other Panel Data.” The Review of Economics and Statistics 89 (2): 221–33.