3. Experiments

Question 2

What is the bias if we randomlly asign treatment?

Answer 2

Bias in sample is expected to be 0

Thus

E(potential outcome in absence of treatment given treated indv.) = E (potential outcome in absence of treatement, given control group)

E(y0i I D=1) = E(y

Question 3

What is the notation for a potential outcome?

Answer 3

y1i = potential outcome if treatment 1

y0i = potential outcome if treatment 0

Question 4

Given no bias, what is the Average Treatment Effect (ATE)

Answer 4

E(y1i - Ey0i)

E( potential outcome of treatment - potential outcome in absence of treatment)

ATE

Question 5

How do we justify that treatment is randomly assigned

Answer 5

Comparing characteristics of treatment and control individuals

IF they look the same on observable characteristics, so its reasonable to claim they would be similar on unobservable characteristics –> thus no bias

Question 6

How do we compare characteristics of treatment and control individuals

Answer 6

Through a T-test

e.g. Xbar,1t - Xbar0t

evaluate if treated and control look the same

Question 7

How would you do a comparison of means of treatment and control?

Answer 7

Null Hypoth: Diff in means is 0

T test done by dividing the difference in means by standard error for the difference in means

Question 8

What does Var (X-Y) = ?

If so calculate the the SE(Xbar1 - XBar)

Answer 8

Question 9

unbiased def

Answer 9

avg. estimate equals the parameter of interest

Question 10

What can the observed difference in average outcomes between treated and control individuals be written as

Answer 10

= sample ATT + bias

Question 11

Implications of Randomisation

Answer 11

1. Unbiasedness
2. ATT = ATE; as no bias

Question 12

Do we see elimination of bias via randomisation in small samples

Answer 12

In small samples it is likely that bias is non-zero in the sample

Question 13

What is the Law of Large Numbers (LLN)

Answer 13

The larger the sample, the closer we expect the sample average to be to the expected value.

as n goes to infinity, sample average goes to population average

Question 14

Answer 14

C

Question 15

Would randomly assigning nets reduce bias?

Answer 15

Question 16