The Estimated Value Drift Of Linear Regression By LS&TLS Method

Linear Regression Analysis is used to establish a linear model to reflect correlation between the dependent variable and one or more independent variables,as well as between the regression coefficients,etc.The classical Least Squares method(LS method)and the Total Least Square method(TLS method)can be used to set up linear regression models or equations.However,many experiments show that the results by either LS method or TLS method can significantly deviate from the true values of the regression parameters.Using the Least square method through the example and the simulation experiment of the linear regression,the problem of the estimated value drift of regression coefficients is illusitrated,and the results only with multiple correlation coefficient s and multiple decision coefficients to judge whether the valuation deviates from the truth are unreliable.Through one to five dimensional linear regr ession simulation experiments,the drift index of the estimated value is concluded,and the judgement conditions are drawn,and relative valuations such as observations of variance are discussed and analysed.Through the example and the simulation experiment of the Total least square mathod in three kinds of random error model,the problem that posterior variance of observations is too small and the obvious difference between the relative true error and the relative mean square error of the regression coefficient are shown,furthermore,the reason is discussed and analised.