4.18 ATE: Decomposition

  • ATE can be decomposed as a function of 5 quantities (e.g. Keele 2015b, 4)20
  • \(\pi\) proportion of sample assigned to treatment (2/6 = 0.33..)
  • E[Yi1|Di = 1]: Average outcome under treatment (Yi1) for units | in treatment condition (Di = 1)21
  • E[Yi1 - Yi0] = \(\pi\) (E[Yi1|Di = 1] - E[Yi0|Di = 1]) + (1 - \(\pi\))(E[Yi1|Di = 0] - E[Yi0|Di = 0])
  • Q: Which parts do we have and which parts don’t we have?
Unit Di (Aspirin: Yes/No) Yi1 (Pain | YesAspirin) Yi0 (Pain | NoAspirin)
Simone 1 0 ?
Julia 1 1 ?
Paul 0 ? 1
Trump 0 ? 0
Fabrizio 0 ? 0
Diego 0 ? 0

References

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


  1. Start again by explaining columns of table.

  2. Red: Average outcome under treatment for units (Yi1) | in treatment condition (Di = 1); Blue: Average outcome under control for units (Yi0) | in treatment condition (Di = 1)