SEMANTICS ON THE APPLICATION OF CORRELATION AND REGRESSION ANALYSES

Abstract Correlation analysis at the start considers the coexistence of at least bivariate data sets without attaching notion that one is dependent nor independent. The tool is used to determine the presence of linear association between or among interrelated variables. The linear association is measured in terms of correlation coefficient. The significance of the correlation coefficient is directly associated with the sample size. For data coming from continuous distributions, the sign of the correlation coefficient is vital. The methodology allows to find proxy variables from its highly correlated variables. This is the essence of multicollinearity analysis. The nonlinear or nonmonotonic relationship is extrinsic to such analysis. The regression analysis on the other hand assumes apriori information that the existence of a dependent variable is attributed with a distinct independent variable. A linear model is estimated to forecast future value of the dependent variable given corresponding values of the independent variable. Though correlation and regression analysis may have the same decision of rejecting their respective null hypotheses, nevertheless, they are two different things. The correlation speaks to mean that there exists linear association between two variables. While the regression analysis at one hand establishes a significant model, and second, provides an inference that speak to mean that the dependent variable changes by some unit per unit change in the independent variable through its regression coefficients. However, the use of correlation analysis aids in estimating a robust regression model.