05/03/2019
We have run Shapley Value Regression (SVR) and compared it to Shapley Owen Value decomposition Regression (SOVR) using 3 of 15 Variables (n=2500), and we find in the second method that there is a loss of 1-2% of the explanation in the dependent variable. The big advantage of SOVR is that it can run on smaller samples. We usually recommend n=30 per independent variable as a minimum. So, with 15 variables (the recommended maximum number of variables for this method) you need n=450, whereas SOVR needs only n=90 when we use sets of 3 independent variables as we run SVR on three variables at a time.