Data Envelopment Analysis of Firm Strategic Efficiency

Regression fits an average relationship and measures deviations around it, whereas DEA builds a frontier of best practice and measures each firm's distance from that frontier. DEA needs no functional form and handles many outputs at once, but it is deterministic: as Charnes, Cooper and Rhodes formulated it, every gap from the frontier is treated as inefficiency, with no random error term. Stochastic frontier analysis sits between the two, fitting a parametric frontier while splitting deviations into noise and inefficiency. The choice depends on whether you trust a functional form and how much measurement noise you expect.

The CCR model of Charnes, Cooper and Rhodes assumes constant returns to scale, so it measures overall technical efficiency. The BCC model of Banker, Charnes and Cooper adds a convexity constraint (the peer weights sum to one) to allow variable returns to scale, isolating pure technical efficiency. Dividing the CCR score by the BCC score yields scale efficiency. Using both together tells a strategist whether a firm underperforms because of how it manages resources or because it operates at the wrong size relative to the most productive scale.

Because DEA is nonparametric and self-weighting, discrimination collapses when the sample is small relative to the number of inputs and outputs. A common rule of thumb is to have at least two to three times as many firms as the total count of inputs plus outputs. Falling short triggers the curse of dimensionality, in which many firms are rated efficient simply because there are too few comparators to challenge them, and the scores become uninformative for benchmarking.