Child pages
• Total Effects on Target vs Correlation with Target Node

Contents

Question

When using Target Analysis which approach should be used for what purpose? Total Effects on Target vs. Correlation with Target Node: The former computes  standardized Total Effects, the latter yields the relative significance, right?

The Standard Target Mean Analysis (STMA: Analysis | Visual | Target Mean Analysis | Standard) returns the Total Effect function. The STMA uses the mean value evidence (based on the MinXEnt) to go through the variation domain of the variable X and for measuring the impact of X on the target’s mean.

Total Effect (TE) is the derivative of this Total Effect function (TEf), taken at the a-priori mean value of X. For a linear TEf, TE reflects the overall strength of the relation with the target node. If TEf is nonlinear, TE is just a local measure of the strength of the relation at the mean value of X.

The Direct Effect Target Mean Analysis (DETMA: Analysis | Visual | Target Mean Analysis | Direct Effect) returns the Direct Effect function. The DETMA uses the mean value evidence to go through the variation domain of X and for measuring its impact on the target’s mean while holding fixed the probability distributions of all the other variables, except:

• The Not-Observable variables belonging to the Class “Factor
• The variables belonging to the Class “Non_Confounder” .

Similarly to TE, Direct Effect (DE) is the derivative of the Direct Effect function (DEf), taken at the a-priori mean values of the variables. The nonlinearity of the DEf has then the same consequences on the scope of the measure.

The Correlation with Target Node (Analysis menu – Report – Target Analysis) is based on Mutual Information. Mutual Information (MI) is defined as the difference between the Marginal Entropy H(T) of the target variable and its Conditional Entropy H(T|X). The form of the dependency between and X and T does not have any impact on MI, which is then always global.