STAT 425 Week 06 - Generalized Least Squares Regression

If the errors in the linear model are heteroscedastic or correlated with known structure, or in ways that can be modeled, generalized least squares (GLS) estimation can improve on the efficiency ordinary least squares (OLS). Furthermore, coefficient standard errors from OLS may be incorrect due to the assumption of constant error variance. These notes discuss the background and implementation of GLS to address these issues.