9.2 Basics 1

  • Panel data: Observing units (individuals, countries etc.) several times (Q: Balanced panel?)
    • Stored in wide- or longformat
  • Time-series data: Countries/country-level measures across time
    • Q: Imagine you would replace Hans and Peter in Figure 9.1 with Germany and Italy? What additional assumptions would we need for comparisons? (Hint: Composition?)
  • Time as an additional dimension in the joint distribution (Figure 9.2)
  • Identifcation strategies (e.g. Keele 2015b, 10)
    • Differences-in-differences (DID)
    • Fixed effects (FE), First differences (FD)
  • Classic approaches currently reinterpreted from causal inference perspective
    • e.g. Assumption that pre-treatment values of treatment/outcome do not matter has to be added (i.e. control for them) (Imai and Kim 2016)
  • Q: Remember the exercise we did once “Finding the counterfactual” (see Section 4.13)

Figure 9.1: Panel of individuals

References

Imai, Kosuke, and In Song Kim. 2016. “When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data?” Working Paper.

Keele, Luke. 2015b. “The Statistics of Causal Inference: A View from Political Methodology.” Polit. Anal. 23 (3): 313–35.