Randomisation

Question 1

How does randomization solve selection bias (i.e. make comparisons ceteris paribus)

Answer 1

Ensures that on average treatment and control group are similar, only the treatment differs between both groups, outcome in the absence of treatment will be the same

Question 2

Difference between lab and field experiment (+ natural)

Answer 2

Lab: controlled environment, artificial randomization
Field: real environment, artificial randomization
Natural: real environment, coincidental randomization

Question 3

Different methods of randomization

Answer 3

Oversubscription (assign treatment randomly among eligible individuals, not enough budget for all)
Randomized phase-in (gradual implementation across eligible areas/organization - who gets treatment is not random but order is)
Within-group randomization (within areas/organizations, some individuals/units randomly receive treatment)
Encouragement design (randomize encouragement to take up treatment - treatment is not random but encouragement is)

Question 4

Potential issues with randomization

Answer 4

Ethical issues
External validity
Compliance (some individuals are offered treatment but do not take it)
General equilibrium effects (difference when scaling up)
Spillover effects (intervention benefits also non-treated)
Randomly generated imbalance, especially when sample size is small

Question 5

How can you estimate the treatment effect in the case of randomization, why no need for control variables?

Answer 5

Randomization makes sure that when T has an effect on Y, and there is another explanatory variable X, that X does not have any influence or correlation with T