Abstract
A criterion of equalizing residual squares is considered for regression modeling, and its results are compared with the ordinary least squares and least absolute deviations regression models. Equalized residuals yield a model with the minimum mean difference of the error squared or the most efficient residual variance. Absolute errors' minimum mean difference is also tried, and its results are close to those of the squared equalized deviations. Solutions obtained with these nonlinear optimization criteria can be more adequate and better interpretable than the results of the regular least squares and least absolute deviations regression models.