13.6 Validating assumptions

  • Continuity and local randomization assumptions inherently untestable but “empirical implications” testable (Skovron and Titiunik 2015, 26)
  • General problem: Researcher has no control over assignment
    • If cut-off known to units → danger of score change/manipulate
    • Q: Can you think of an example of score manipulation?

  • Qualitative tests of assumptions
    • Explore how manipulable score/assignment are
      • e.g., institutional appeal possibility (changing of scores)
      • e.g.a Administrative process of score assignment

  • Quantitative tests of assumptions
    1. Density of the running variable around the cutoff: Local N below/above cutoff very different? (Q: Why?)
    2. Treatment effect on pre-treatment covariates or placebo outcomes: Are treated units near the cutoff similar to control units? (Q: Why?)
    3. Treatment effect at alternative fake/placebo cutoff values (Q: Why?)
  • The three quant. tests differ between continuity- and randomization-based approach

References

Skovron, Christopher, and Rocıo Titiunik. 2015. “A Practical Guide to Regression Discontinuity Designs in Political Science.”