5.1 Basics

  • Strategy: Randomly assign units to treatment and control
    • “Gold standard”: Researcher controls assignment of units (Independence Assumption = valid)
    • Treated and control groups will be identical in all respects, observable and unobservable on average (long run)
  • Identifies average treatment effect within the study population (ATE = ATT)
  • Data: Normally “cross-sectional”28
  • Assumption(s): Independence asssumption, Sutva
  • Q: What do the concepts internal/external validity and generalizability designate (Campbell, Stanley, and Gage 1963)?
    • Valid extrapolation requires either random sampling or additional assumptions
  • Q: How realistic is the idea of a fully randomized experiment? Think of the examples we discussed before (AIDS treatment, Jobtraining program etc.)!
  • Q: What is a “manipulation check”?

References

Campbell, Donald Thomas, Julian C Stanley, and Nathaniel Lees Gage. 1963. Experimental and Quasi-Experimental Designs for Research. Houghton Mifflin Boston.

  1. In reality, there are always time differences in terms of when variables are measures. Even if they are seemingly measured at the same moment in time. Hence, the term “cross-sectional” can be misleading.↩