11.1 Basics 1

  • Data:
    • Outcome Y: …is observed twice or more (2+ timepoints)
    • Treatment D: …is observed twice or more (2+ timepoints)
    • Controls X: …is observed twice or more (2+ timepoints)
  • Focus on differences between time points (FD) or deviations from the individual cross-time mean (FE)
    • Estimates are based on variation within individuals (because of transformation!)
    • Variation in treatment D, outcome Y and covariates X
    • Focus on within unit variation may dramatically change the variation we are looking at
      • See Mummolo and Peterson (2018) for suggestions of how to improve interpretation of findings42
    • Stable (unobserved) confounders drop out (we’ll see this)!

References

Mummolo, Jonathan, and Erik Peterson. 2018. “Improving the Interpretation of Fixed Effects Regression Results*.” Political Science Research and Methods, 1–7.

  1. In short, explore what variation in treatment and outcome is left after the data was transformed. Based on that variation there may be more sound counterfactuals (than min. vs. max. in the original treatment variable). Seems less of a problem with dichotomous variables.↩