The method of least squares works well even if data is not normally distributed. It gives the most accurate estimates compared to other methods. The Gauss-Markov theorem proves this and its generalizations. The paper provides a list of resources to help non-experts understand the theory better.