ECON 8686 Problem Set #3- Measurement Error
Part I. Simulation
1) Create a simple measurement error simulation:
set obs 100000
gen z=uniform()
gen a=uniform()
gen b=uniform()
gen e1 = invnorm(a)
gen e2 =invnorm(b)
gen y=3*z+e1
gen x=z+e2
2) Using the output from the following command:
corr z e2, cov
Calculate the bias of the coefficient from the regression of y on x (for simplicity, ignore
the covariance term).
3) Confirm your answer to question #2 by regressing y on x
4) Now:
replace b=.5*b
replace e2=invnorm(b)
replace x=z+e2
5) Repeat questions 2 and 3. What happened? Why?
Part II. Measurement Error and Panel Data
This part of the problem set is based on:
McKinnish, Terra. 2008. “Panel Data Models and Transitory Fluctuations in the
Explanatory Variable.” Advances in Econometrics Vol.21
The data set birth.dta contains data for all 50 states plus D.C. for the years 1972-93. The
variables are:
births: birth rate, white women ages 20-24
afdcben: real monthly AFDC Benefit
earn: real earnings per capita
state: state ID code
year
1) Regress the logged birth rate on logged AFDC benefits, logged per capita earnings,
and state and year fixed-effects.
