Nov 5 Lecture
Experiments in Democracy
how we use experiments to learn about the political world*
milk or tea first? Scientist’s (man and women) met, she insisted that she
could tell the order of milk put in (first or after); they decided to run an
experiment to see if she could actually tell; 8 cups, half milk in first, other half
tea in first, randomize and give to girl; identifies all 8 correctly: “Was she
lucky?” “How could we know?”
The Fisher Exact Test
- obviously, this is why today we know about this example
- the probability that she could get all cups correct, by chance, is 1 in 70
So what’s important?
- we can learn from experiments
- we can test peoples claims
- treatment and control
Some important questions
- Does federalism lead to better political representation?
- Are voters superficial?
- How do we stop patronage and instead get public goods? Does contact
from political campaigns matter in an election
- Treating/ Conditions
- Assignment to treatment or control/ comparison groups is random and
the process of randomization is known
- Ex-post measurement of results
- Example: does Prof. Kopstein teach better when he wears a tie?
- A treatment is a stimulus
- Anything that is thought to have an effect on some outcome
- It is applied to some people and not to others
- Or, it is applied at some times and not others
- Example: whether Prof. Kopstein wears a tie to class to teach
- We put some subjects in treatment and some in control
- We do it randomly
- What does it mean that this is done randomly?
- Example: we could flip a coin before every class to see whether Prof.
Kopstein would wear a tie
Ex-post measurement of results
- For an experiment to be complete, we need to measure the results
- What we want to measure are casual effects
- Example: we could measure exam performance at the end of the year Example
- Question: does wearing a tie make Prof. Kopstein teach better
- Treatment: wearing a tie or not
- Random assignment: flip a coin before class
- Measurement: exam results
What if we didn’t have one of the parts?
Treatment: wearing a tie (needs to have variation)
Random assignment: Kopstein decided on his own when to wear a tie
(wouldn’t work- his decision)
Measurement: no measurement after exams (wouldn’t know results- no
All necessary for an experiment
To make a causal inference, we must have all three parts
Causal inference: if there’s x, theres y (if you exercise more frequently, you
will be healthier)
- difference in outcomes between two states of the world, treated versus
- Why is this problematic?
- We cannot observe an given subject in both its treated and untreated
- “fundamental problem of causal inference”
- Random assignment enables us to create two groups whose treated and
untreated states are the same in expectation
- exam results (tie) – exam results (without a tie)
Why do an experiment?
- Because we want to make casual inferences
- But we want to make them about important things
1. Does federalism increase representation?
2. Can politicians win by providing public goods rather then patronage?
3. Are elections won for superficial reasons?
- Field experiments
- Natural experiments
- Survey experiments
- Laboratory experiments