Research on the Some question about selection of independent variables

The paper studies four methods of selecting independent variables in multivariate analysis. In general condititon, advanced statistical method and backward statistical method can not obtain the best subset of independent variables, which may be affected by the orders of variables or associations among variables. With the ill conditioned multicollinearity, the method was not effective in spite of stepwise regression and optimal selecting method of total subsets. In respect of this, the paper proposes a new method which integrates variable deletion and ingredient analysis.