The article discusses how to estimate variance components in linear models using a quadratic approach. The researchers use a coordinate-free method to explain the theory and apply it to general linear regression models. They also explore the admissibility of quadratic estimates in mixed linear models with two variance components.