Imagine you fit a straight line to data that arrive over time. The error from this month might spill into next month’s error. Ordinary least squares (OLS) still gives estimates that are right on average, but they can be less precise than they could be—and the standard errors that standard software prints can mislead you. This chapter shows a smarter tool, Generalized Least Squares (GLS), that works better when we know something about how errors are related, especially when they are autocorrelated (correlated with their own past).