How should we think of compliance in a phase-in RCT design? What are the assumptions recovering the LATE by instrumental variables in this case?

Details: In a traditional (simultaneous) RCT, in case of non-compliance, the LATE (local average treatment effect) can be recovered by using the random assignment as an instrumental variable if two assumptions hold:

  • monotonicity: random assignment weakly increases (or decreases) program participation
  • exogeneity: random assignment itself has no direct effect on Y, except via participation in the program

What are the assumptions required to recover the IV in case of a phase-in design? Is this covered in any paper or textbook?

The reason I ask is because participants may have a preference for receiving the program in a specific time period.

For example:

Setting of the RCT:

  • 3 periods: t={1,2,3}

  • N=100 participants randomized into receiving the RCT in either:

    • Treatment: 50 receive program in t=1
    • Control: : 50 receive program in t=3
  • Outcome Y compared between treatment and control at the end of period 2

  • Non-COmpliance (assumed I know this, which in fact can not be observed):

    • 20 subjects only accept treatment in t=1
    • 10 subjects only accept treatment in t=3
    • 70 subjects are compilers (accept treatment when offered)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.