Lecture notes 11 errors in variables Flashcards
What is a measurement error in regressions?§
Instead of observing true X you observe X*
How is measurement error calculated?
What are the assumptions?
U = X* - X
Cov(U,X) = 0
Cov(u, epsilon) =0
When would mismeasurement occur intentionally?
When people are asked about something they have an incentive to overreport eg earnings.
How do you do a regression with mismeasurement error?
U= X* - X
Y = B0 + B1X + epsilon
sub in for X
Y = B0 + B1(X*-U) + epsilon
Y= B0 + B1X* - B1U + epsilon
This gives OLS estimate of
B1 + COV(X,V)/ Var(X)
Then to check the
b1 is approx B1 Var(X)/ Var(X) + Var(U)
As U increases the varianced term decreases so as mismeasurement increases b1 becomes biased
What is the direct impact of mismeasurement on OLS estimate of B1
Bigger mismeasurement b1 is smaller and biased.
What is the impact of mismeasurement error in Y
under the asusmption that Cov(u,X) = 0
it causes no issues on b1
But the variance changes from normally
sigma^2/ sum of (xi-xbari)^2 vs sigma squared + var(u)/ sum of i=1 to n (Xi -Xibarn)