17.5 How it’s used with regressions

  • estimate one regression of the outcome on the time period before the event
  • estimate outcome on the time period after the event
  • check difference
  • doesn’t have to be linear

Outcome=β0+β1t+β2After+β3t×After+ϵ

  • more precise estimate of time trend than going day by day
  • but limited by shape
  • but need to be careful about significance testing (autocorrelation)
  • use HAC standard errors

17.5.1 Example: Improved ambulance care

  • heart attack performance (AMI)Week27

AMI=β0+β1(Week27)+β2After+β3(Week27)×After+ϵ