4.12 ITE: Estimation

  • Individual-level Treatment Effect: Difference in potential outcome for unit i at time t post-treatment
  • ITE is unobservable because units can not receive both treatment and control (i.e., all values of treatment variable)
    • Aspirin → headache: Simone has headache and takes aspirin
Unit Di (Aspirin: Yes/No) Yi (Pain: Yes/No) Yi1 (Pain | Aspirin) Yi0 (Pain | NoAspirin)
Simone 1 0 0 ?
  • ITE forces us to think about this process in terms of counterfactuals and individual cases (Lam 2013)
  • Table above divides the Y column into columns for treatment and control group
    • Q: What would the table look like if the treatment had more levels?
  • Estimation requires filling in/replacing the missing counterfactual

References

Lam, Patrick Kenneth. 2013. “Estimating Individual Causal Effects.” PhD thesis, Harvard University.