Numerical Dependency Analysis (NDA): A new method for estimating the statistical dependence (not correlation) of two variables

Abstract

Abstract Dependence and correlation are different statistical concepts. Although there are methods to measure linear or nonlinear correlation between two variables, measuring the statistical dependence between two variables is of great interest. The main contribution of this paper is to present a heuristic algorithmic method to estimate the measure of dependence between two variables. This method first transforms the X-Y scatter plot into functional relation plots. This procedure may have many answers for big data. Then measures the dependence of Y on X by using the new concept of successive triangles. The desirable features of this method are: This method can be applied to both numerical and categorical (nominal) variables. The presented bivariate method is distribution-free, so it can be used for non-Gaussian numerical variables. As an application of this method, it can also be used to measure the correlation. This novel and non-parametric method is validated by both simulated and clinical data. This method has other applications, such as template matching for single-dimensional patterns.