Unformatted text preview: particular types of individuals you select, though it may not be valid for the population. Example: If interested in estimating how wage is related to education among individuals who have college or above education, then one can draw a random sample from those who have at least 16 years of schooling and estimate how their years of schooling affect their wages. In this case, what we estimate only applies for this subsample, but may not be generalized to the full population (including individuals who have college or above education and those who have less than college education.) Selection based on Y • Regardless, if the sample is chosen on the basis of Y, then we have sample selection bias. OLS estimators will be biased. o Example, in estimating how wage (Y) depends on education (X), if choosing only those whose wage is greater than $30,000 per year, then the OLS estimators of the slope and intercept will be biased. 3 The sample selection issue can happen when some data are missing non‐randomly, so that you end up with a non‐random estimation sample. • If any observation has missing data on on...
View Full Document
This document was uploaded on 03/11/2014.
- Spring '14