Causal Analysis

Question 1

causal relationship only if _ assumption holds

Answer 1

exogeneity

Question 2

unobserved factors (ui) unrelated to regressor X

Answer 2

exogenous regressor

Question 3

Xi and ui correlated

Answer 3

endogenous regressor

Question 4

techniques to overcome endogeneity

Answer 4

randomisation and IVs

Question 5

limitations of randomisation in OLS

Answer 5

unethical, estimate only those who chose to take part, limited duration

Question 6

assumptions of an IV

Answer 6

relevance and exogeneity

Question 7

relevance assumption

Answer 7

Z must be correlated with X

Question 8

exogeneity assumption (exclusion restriction)

Answer 8

Z must not be correlated with error term u

Question 9

instrument is used to

Answer 9

observe exogenous variation in Xi through movements in Zi to identify causal effect of Xi on Yi

Question 10

IV estimation can be done by

Answer 10

2SLS

Question 11

2SLS stages

Answer 11

1. regress X on Z to get predicted Xs

2. regress Y on predicted Xs

Question 12

weak instruments problem

Answer 12

2SLS estimator will be biased towards OLS estimator if correlation is close to zero
doesn’t explain sufficient variation in endogenous variable X

Question 13

testing strength of instrument

Answer 13

F-stat

Question 14

F-stat (actual value, not p-value)

Answer 14

<10 = weak
>10 = strong

Question 15

testing validity of instrument

Answer 15

can only use IV if same number of instruments and endogenous regressors

Question 16

LATE

Answer 16

Local Average Treatment Effect - effect of receiving treatment on w/e

Question 17

ITT

Answer 17

Intention to Treat effect - captures causal effect of receiving treatment = underestimate

Question 18

causal effect =

Answer 18

ITT/difference in compliance rates btw treatment and control group

Question 19

Regression Discontinuity (RD) designs

Answer 19

situation in which treatment D depends on an observed continuous variable Q (running variable)

Question 20

example of RD design

Answer 20

babies w/ both weight below certain value receiving extra neonatal care

Question 21

Sharp RD design

Answer 21

treatment status (D) is a deterministic function of q. e.g MLDA = 21 - status changes at some point in continuous function

Question 22

Fuzzy RD design

Answer 22

treatment status (D) is NOT a deterministic function of q. change in the probability of treatment. e.g raising of school leaving age affect on earnings = indirect effect

Question 23

IV

Answer 23

correlated with one of the variables but not correlated with anything in error term that affects dependent variable