1. Have a control/baseline/placebo group.
2. Randomize allocation to study groups.
3. Ensure double-blind trials.
- Neither subjects nor experimenters should know who is in which group.
1. Measure everyone.
- Even subjects that did not participate or drop out.
1. Treat all groups equally.
2. Repeat & reproduce the trial.
- Increase the results' robustness.
Source: The Art of Statistics by David Spiegelhalter