Randomised Control Trials Flashcards
What are the two denotions for the two outcomes for individual i and what do they mean
Y0,i = i’s outcome if she DOESN’T receive treatment
Y1,i = i’s outcome if she DOES receive treatment
i..e 1 and 0 dnote whether or not treatment has been given
What is the fundamental issue with treatment vs no treatment?
we never observe both potential outcomes for the same individual
What is the counterfactual?
The potential outcome that is not observed
what is the difference in group means between treatment and control?
average causal effect of program on participants + selection bias
What is selection bias?
natural differences of pre-existing differences between the two groups in absence of the programme
What is the mathematical expectation of Yi, E[Yi]?
equivalent to the sample average in an infinitely large population
law of large numbers tells us that the average of Yi gets v close to E[Yi] as the number of observations gets large
How does random assignment to treatment work in an experiment?
Random assignment to treatment: eligibility for program is literally determined at random e.g. via pulling names out of hat
law of large numbers - the sample averages can be brought as close as well like to the population average just by enlarging the sample
So should we just randomise everything?
External validity - health insurance experiments on a low income group, small group not sure about the geographic representation and more general applicability
Hard to do in large numbers - participants are often volunteers and thus can be self selected - threatens internal validity and selection bias
ethical issues - if a negative treatment like air pollution, many people might reject the esperiement threatening external validity
What is a randomised experiment?
where treatment is randomly assinged
What is the assumption in a radomised experiment?
If the assumption is valid what is the effect
treatment group and control group are similar in the absence of the treatment
if valid selection bias is zero and the difference in group means is the average causal effect
What is the difference between internal and external validity?
Internal validity = the casual effect is valid within the experimental sample
External validity = does the same effect apply in another sample or context?
How do you ensure that treatment and control groups are similar?
look at the pre-existing charactersitics and assess whether in all observed dimensions the two groups are similar to each other
