June 20, 2017 – mmr
Old Things Made New: Shapley Value Regression
Sometimes it amazes me to look back at the past decade or so and see how the ready availability of computing power has led to the emergence or reemergence of some OLD mathematical concepts… like the dominance of Discrete Choice Modeling and Max Diff, which have completely supplanted old style, traditional Conjoint.
But then, I’m probably just ‘geeking a bit’.
Anyway, like me, Shapley Value Regression (SVR) is OLD… it’s actually been around since the 1950s, coming out of game theory space, as so many mathematical constructs have and was then actually “reinvented” in the ‘90s by Kruskal. SVR is an approach to “estimating the incremental contribution of any item to a set of items.”
What’s NEW is the greater ease of application afforded by modern computing power and the growing recognition of SVR’s potential in Marketing Research analyses! Among others, SVR has potential application in areas like Product Line Expansion and Optimization (alongside of alternatives like TURF Analysis); and SVR has particular potential in Drivers Analyses, particularly in cases (often) when our old friend, Multi-Collinearity is likely to be an issue.
So, the question we should ask is… Should I be using SVR as the ‘standard’/ ‘best-in-class’ approach?
Stay tuned for a research-on-research case study to be posted in Summer 2015.