engleski

Overcoming Multicolinearity by Orthogonal Transformation of the Explanatory Variables

This paper deals with overcoming the situation when the multicolinearity appears as dominant problem in serious statistical-econometric research. Result of using standard statistical-econometric methods (as stepwise technique) is excluding a numerous explanatory variables with significant influence on dependent variable. Even more, multicolinearity becomes a barrier for specification of any influence of "removed" variables. This is especially relevant in the cases of analyzing total effects of the entire set of explanatory variables on the dependent variable. Moreover, indirect effects must not be ignored in the situation when they are dominant component of total effect. The whole procedure of relaxing this problem is illustrated by a practical example of comparing performance indicators of all manufacturing enterprises in the Split-Dalmatian County in 2004. The data base consists of a wide range of performance indicators for 1744 manufacturing enterprises, among which twelve are selected as representative ones. Using principal components method four factors have been extracted, i.e. all selected variables (performance indicators) have been meaningfully grouped in to factors: activity, liquidity, leverage, economic efficiency. It is shown that principal component method doesn’ t stands for a transformation method only, but it can be used as explicit modelling approach. These uncorrelated common factors are used to overcome multicolinearity by orthogonal transformation of explanatory variables in multiple regression model. The essential part of analysis is establishing of direct, indirect and overall effects of each explanatory variable on return on equity as chosen dependent variable. This work builds up a complete procedure of standardized coefficient estimation for each extracted factor, comparing their relative influence as well as their weights specification. Analytical hierarchy process is used as the technique of measuring inconsistency of assigned factor weights.

common factors; orthogonal transformation; principal component method; multicolinearity; direct and indirect effects on dependent variable; analytical hierarchy process

nije evidentirano