This is a variant of ordinary least squares in which both the response variable and predictor variables contain measurement errors. The goal is to find a linear model that minimizes the sum of the squared distances between the observed data and the fitted model.