ECON6300/7320/8300 Advanced Microeconometrics Instrumental variables
Christiern Rose 1University of Queensland
Practical 4 March 2019
1/6
Introduction
This class will review:
Instrumental variables
2SLS and GMM estimation
Tests for endogeneity of regressors
Tests for weak instruments
Tests of overidentifying restrictions (instrument validity)
We begin with a demonstration from Microeconometrics using STATA Chapter 6 looking at the effect of employer/union sponsored health insurance on drug expenditures
We move on to a practical looking at the the returns to schooling
2/6
Demonstration (1)
We analyse data on health, retirement and private insurance
The data is from the Medical Expenditure Panel Survey
We explore the effect of employer/union sponsored health
insurance (hi_empunion) on (log) drug expenditure
(ldrugexp)
We treat insurance as endogenous as it is a choice
variable. Those who expect high future medical expenses are more likely to take out insurance.
3/6
Demonstration (2)
We analyse data on health, retirement and private insurance
We consider as instruments:
The proportion of total income that comes from social
security (ssiratio)
An indicator for low income status (lowincome)
The size of the individual’s firm’s labour force (firmsz)
Whether the firm operates in more than one location (multlc)
As exogenous regressors we include number of chronic conditions (totchr), age, female, black/hispanic indicator (blhisp) and log income (linc)
The data are mus06data.dta and the do file is mus06p1iv.do
4/6
Practical (1)
We use data from Kling (2001) : mus06klingdata.dta are wage data collected in 1976
We want to estimate the impact of years of schooling (grade76) on wages (wage76)
Other covariates in the wage equation could be black, south76, smsa76, reg2-reg9, smsa66, age76 agesq76. (Use desc)
Years of schooling (grade76) is endogenous! Those with higher ability tend to have more schooling, and ability ought also to determine wages.
Possible instruments could be proximity to a 4 year college (col4) or family education levels (e.g. daded momed)
5/6
Practical (2)
Load the data
Describe, summarise and choose a baseline instrument(s)
Estimate the 2SLS model and optimal two-step GMM. Assume heteroskedastic errors. Interpret your results
Compare your results with the OLS estimator.
Are your results sensitive to the choice of instrument(s)?
Is there evidence of endogeneity of the regressors in your wage equation?
Are your instruments weak?
Are your instruments valid?
6/6