Factor Blending


Factor Blending is performed to assist with convergence. Factors from the iteration being calculated are blended with corresponding factors from the previous iteration. This helps to avoid problems with oscillation and non-convergence, caused in particular by correlated characteristics. Since factors are calculated based on adjustment for all other characteristics, it is possible for correlated characteristics to oscillate between factors greater than 1 and less than 1, potentially in growing magnitude, resulting in unstable iterative behavior. Factor Blending helps to avoid this problem


The calculation is simply:


Blended Factor = [Credibility Weighted Factor (current iteration) x Blending Ratio] + [Previous Balanced Factor x (1 - Blending Ratio)]


Eventually, as the Factors converge, this adjustment is minimal.


The Blending Ratio is calculated based on  the number of Characteristics selected.


Blending Ratio = min(0.5, 3/characteristic count)