17.2 Prediction and Deviation

• Treatment should be the only thing that’s changing
• We thus need a counterfactual, e.g. before-treatment trend
• best to have short time window (short horizon)

Dealing with this trend

1. Ignore it: compare before/after
   • reasonable to believe that there’s no time effect
   • if data looks flat
   • extremely tiny time span
2. Predict after-event data using before-event data
   • extrapolate from trend
3. Predict after-event data using after-event data
   • relationships between outcome and other variables in before-event period