The least squares principle allows us to estimate unknown parameters based on minimising the sum squared residuals (recall: the residuals are the deviations between model and observations). But what if we want to assign different weights to our observations, because we think that some observations are better? That’s where weighted least squares comes into play.