Definitions Etc

Question 1

SUTVA

Answer 1

Stable unit treatment value assumption

Question 2

Sutva is not satisfied if…

Answer 2

1) If treatment varies eg. Different people giving different doses or styles of session
2) there are no interferences, so treatment applied to one group doesn’t affect success of the other group. This might happen if the treatment group teach the control group what they learned or the treatment group raise the grade boundaries/ boundaries of entry/ take all jobs in the town etc.

Question 3

Z works as a variable because it fulfils which two assumptions

Answer 3

Exclusion restriction (So covariance between z and u is 0 because z is random) 
Strong instrument (So covariance between z and t is not 0, since the random allocation affects whether person gets treatment)

Question 4

What is the equation for calculating ca-itt?

Answer 4

B= cov(z,y) ÷ cov(z,t)

B = CA itt
Z= random allocation of ppts
Y= outcome
T= did they recieve treatment

Question 5

What is cov(z,y) in the ca-itt equation?

Answer 5

Cov(z,y) is ITT because it is the association between random allocation and the outcome

Question 6

What is cov(z,t) in the ca-itt equation?

And how do you calculate this?

Answer 6

Cov(z,t) is the association between being assigned to the treatment group and receiving treatment.

It is calculated as % who were assigned to T and got T minis % assigned to C who got T.

Question 7

Name the 4 methods of dealing with contamination mentioned in lecture 3

Answer 7

As treated
Per protocol
ITT (intention to treat)
CA-ITT (contamination adjusted intention to treat)

Question 8

CA itt means that the treatment effect is not dragged down by contamination. However it is not without cost. Name a disadvantage of ca-itt related to it’s ability to keep the treatment effect estimate high with increasing contamination

Answer 8

With greater contamination you get larger confidence intervals. Therefore increasing uncertainty

Question 9

CA-ITT assumes what about the effect on each ppt

Answer 9

The effect on those treated will be the same as the effect on those not treated. However, this assumption does not always hold

Question 10

What do SATE AND PATE stand for

Answer 10

Sample average treatment effect

Population average treatment effect

Question 11

Which type of validity are RCTs bad at?

Answer 11

External validity

Question 12

Which validity are observational studies bad at?

Answer 12

Internal validity

Question 13

What does RDD stand for?

Answer 13

Regression discontinuity design

Question 14

What can you do given only observational data to try and raise internal validity?
What is the disadvantage of this?

Answer 14

Propensity score matching. Eg. Matching a kid who did the treatment with a kid from a similar background who didn’t

Because you exclude those who do not have a match in the control group from the treatment group the external validity does down. Now your sample is not fully representative because you kicked some people out.

There is a toss up between external and internal validity

Question 15

When will SATE not equal PATE?

Answer 15

If there are:

heterogeneous treatment effects and non random samples

Question 16

What are the disadvantages of RDD

Answer 16

Only gives treatment effect for two ppts. Meaning it is internally valid but not externally valid