6.4 Ideal experiments: Benefits

  • Serve as benchmark to uncover weaknesses in identification strategy
  • Example from the ‘harsh’ reality: Yearly panel data (2004-2008)
    • Random sample of swiss households (several individuals in each household)
    • Treatment: Have you been insulted or threatened verbally since last interview? (0,1)
    • Response: Would you say that most people can be trusted or that you can’t be too careful in dealing with people? (0-10)
  • Ideal (thought) experiment reveals and emphasizes..
    • Potentially non-random assignment of treatment (bias!)
    • Non-random sample of individuals
    • Sampling on household-level (Sutva?!)
    • Repeated measures but no control over timing of measurement
    • Potential attrition of units
    • Surveys = self-reports