Owing to inadequate information on a multifactor nonadditive object, the investigator postulates a model that does not correspond to the actual relations between the variables. A method is presented which lets us eliminate systematic errors that arise in the linear regression coefficients.