Based On The Composition Data Analysis Methods

Compositional data is widely used in data analysis of the fields in economy, technology and geosciences. For example, it could be used to analyze the changing theory of the products'market share, or presume industrial structure's tendency of an area, or speculate expenditure ratio of city residents'domestic life of an area and so on.In this paper, we improved the linear models of compositional data and introduced some characters of the parameter estimate; Still the paper expounded the justification, advantages and problems of logratio approaches to the compositional data analysis; It also analized the unadaptability of traditional principal component analysis on composition data, and discussed the appliance steps of the principal component analysis on composition data.Based on the traditional linear regression method, a partial least-squares regression analysis method is introduced. In this paper, the theoretical analysis is given to show the feasibility and rationality of this method.The paper still pointed out the shortcomings to solve the problem of compositional data with partial least squares, a new model of linear regression of small variance principle composition was introduced, and a specific example given.Finally, this paper introduced logcontrast PLS path modeling of multiple compositional data by combining centered logratio transform-ation with PLS path modeling; The methodology study is based on the special algebraic theoretical system of compositional data and the for-mula expression of logcontrast latent variables of compositional data are derived from this model, which theoretically justifies the proposed method; Then, the validity and practicability of this model was verified by an empirical results.