ECON6300/7320/8300 Advanced Microeconometrics Quantile Regression
Christiern Rose 1University of Queensland
Practical 8 April 2019
1/4
Introduction
This class will review:
Plotting quantile functions
Quantile regression
Bootstrap standard errors
Simultaneous quantile regression and hypothesis tests Plots of quantile coefficients
We begin with a demonstration following Microeconometrics using STATA chapter 7.
We move on to a practical based on estimating Engel curves for medical expenditure in Vietnam.
2/4
Practical (1)
We have data from the World Bank’s 1997 Vietnam Living Standards Survey for 5,006 households with positive medical expenditures in the previous year. The data are qreg0902.dta.
The variables are age and gender of household head, whether the household is a farm, whether it is urban, the household size, log total household expenditure and log household expenditure on medicine.
We are interested in estimating Engel curves for medical expenditure
The outcome of interest is log household expenditure on medicine, the covariate of interest is log total household expenditure and the remaining variables are controls.
3/4
Practical (2)
1. Load the data, summarize and describe
2. Plot the quantiles of the outcome of interest (log household
expenditure on medicine)
3. Estimate the quantile regression at the median and
compute the average marginal effect of age on total
medical expenditure (not log total medical expenditure)
4. Compare the OLS estimates of the coefficient on log total
expenditure with the quantile estimates q = 0.1, 0.5, 0.9. Is
there evidence of heterogeneity at different quantiles?
5. Are your findings consistent with heteroskedasticity in the
standard linear regression model (OLS)? Perform a test of
heteroskedasticity to verify this.
6. Test equality of the coefficient on log total expenditure at
quantiles 0.1,0.5,0.9 (i.e. H0 : β.1 = β.5 = β.9)
7. Plot the quantile regression coefficients against the
quantiles. Include confidence intervals for the quantile regression coefficients. Include the OLS coefficient and confidence interval for comparison.
4/4