Difference-in-differences methods

Difference-in-differences analysis is a technique for establishing causal relationships using quasi-experimental data. This approach addresses scenarios where treatment and control groups may differ before treatment, making direct comparisons invalid. The method works by measuring both groups before and after treatment, calculating the difference between groups at each time point, and then finding the difference between these differences to estimate the causal effect. This analysis is typically implemented through a regression model with a treatment variable, a time variable, and their interaction, where the interaction coefficient represents the treatment effect. The validity of difference-in-differences analysis depends critically on the parallel trends assumption, which states that without treatment, the treatment group would have developed similarly to the control group over time. This assumption must be justified both conceptually (through theoretical arguments) and empirically (by demonstrating similar pre-treatment trends). Another important consideration is the non-independence of observations, as the same subjects are measured twice. This issue is typically addressed using cluster robust standard errors in large samples. While difference-in-differences is a powerful tool for causal inference in non-equivalent control group designs, it requires careful attention to these assumptions to produce valid causal estimates. Link to the slides: https://osf.io/m6tkn