Following the previous session, Rachel provides a complete recap of the methods of randomization (see session 3), pointing out their advantages and drawbacks according to different contexts. Also, the randomization methodology we choose is important for the level of randomization we will use (individual, household, community, institution, etc.): the lower the level of randomization, the bigger the statistical power of the outcomes.
Throughout the description of the Balsakhi Program Rachel gives an example of multiple treatment, necessary when we want to answer several questions with one unique experiment. The difficulty is to keep the assignment of the program random, isolating the different impacts we want to estimate. While assignment must be random, the outcomes analysis can be split into subgroups.
Stratification is always an useful tool when we randomize, a must in case of small sample size. Stratification has to occur before the randomization and, especially, on variables we think have a significant impact on the final outcome and on subgroups we are particularly interested in.
Finally, when we randomize we need a list of the whole sample (treatment and control) we want to study. Then, we are able to randomize, usually using a random number generator (with Excel or Stata).