4.21 Assumptions: Independence Assumption (IA)

  • IA = Treatment status is independent of potential outcomes (Yi1, Yi0 \(\perp\) Di)
    • i.e., assignment status unrelated to potential outcomes…
    • …whether person gets/takes aspirin is independent of what the person would have under treatment/control (whether pain or not)
  • Under IA “expectation of the unobserved potential outcomes is equal to the conditional expectations of the observed outcomes conditional on treatment assignment(Keele 2015b, 5)
  • IA allows us connect unobservable potential outcomes to observable quantities in the data
  • IA is linked to the “assignment mechanism”22
  • Why does IA (+ SUTVA below) identify causal effect? (next slide)

References

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


  1. see Imbens and Rubin (2015), 14 for a recent discussion.