Correction of regression model parameters based on expert information on changes in the significance of predictors in the background

The paper proposes an algorithmic method of accounting for regression modeling of complex objects of any nature, along with statistical expert information on changes in the significance of predictors in the background. The method of mixed estimation and the continuous form of the maximum consistency method are used as methods for calculating unknown parameters of the model. The first of them is based on a combination of the methods of smallest modules and anti-robust estimation, each of which "works" on "its own" subsample of the original data sample. The second one is designed to increase consistency in changing the calculated and set values of the dependent variable. The implementation of the joint use of statistical and expert information in modeling for these methods is reduced to solving the corresponding linear programming problems. The proposed method is implemented in the construction of a linear regression model of patent activity in Russia