2. Run a multiple regression model using the Efficiency Scores (from the first step) as a dependent variable and all the inputs and output variables used in the data envelopment analysis in the first step as independent variables.
I am wondering about the soundness of this methodology.
From Paul@21:1/5 to All on Sat Aug 6 13:12:53 2016
I doubt any good would come of it. First, since the efficiency scores are essentially ratio scores, a linear regression model is likely not to fare too well. You could perhaps cobble together a nonlinear model that would fit.
Second, depending on your sample of units being evaluated, you might find precious little correlation between predictors and scores. A unit with lots of output 1 and no output 2 could get a high rating, as could a unit with lots of output 2 and no output
1. Replicate across a bunch more units, and you have variables with high causal relation to the scores but no significant correlation (Pearson or Spearman).
If you care to articulate your goal, perhaps someone can weigh in with alternative suggestions. Also, you might want to consider asking on OR Exchange, which is a fair bit busier these days than sci.op-research.