4.7 Potential outcomes framework (2)

  • Given a unit and a set of actions (treatment values) we associate each action-unit pair with a potential outcome (function)14
  • Q: What is the fundamental problem of causal inference? (Holland 1986)
  • Q: What would an absolute time scale emphasize?
  • Q: How many possible treatment-outcome scenarios are there for Peter with treatement variable D (0/1) and outcome variable Y (0/1)?
    • Draw a tree diagram for a particular moment in time post-treatment! (Let’s say D is = (not) taking an Aspirin; Y is = (not) having a headache)

References

Holland, Paul W. 1986. “Statistics and Causal Inference.” J. Am. Stat. Assoc. 81 (396): 945–60.

  1. In principle, these lines should be functions. In other words, for one x-value there should be only one y-value (z-value). Here we plotted the vertical line at t = 25 because it facilitates reading the graph, albeit there are no values corresponding to this vertical line. Rather the treatment variable jumps from 0 to 1.↩