Identification of geochemical factors in regression to mineralization endogenous variables using structural equation modeling

Abstract Structural equation modeling (SEM) is a statistical technique for testing and estimating causal relations using a combination of statistical data and qualitative causal assumptions. It can be considered as an extended statistical hypotheses testing with pre-defined structural model. However, when the SEM is applied as an exploratory technique in data mining, the traditional model depicts a fundamental limitation of incapacity of generating and refining structural model. A new SEM method is proposed in this paper, which combines the principles of cluster and regression analysis. Thus, the new mathematical model can not only generate factors to form model structure but also ensure the optimum relationship to the objective variables. The method is applied to a data set of lake sediment geochemical compositions for identifying gold mineralization associated factors in Southern Nova Scotia, Canada. A SEM model consisting of three measurement sub-models and one structural sub-model is created on the basis of the concentration values of 16 elements from 671 lake sediment samples. The calculated results show that the factors obtained by the new SEM model represent three geochemical factors that are associated with As and dominated by Cu , Zn and W , respectively.