Identification of reference set and measurement of returns to scale in DEA: A least distance based framework

TL;DR

This paper introduces a novel least distance based framework in DEA for identifying reference sets and measuring returns to scale, emphasizing managerial relevance and providing a unique projection approach.

Contribution

It proposes a new lexicographic multi-objective programming method for unique closest projection and develops a linear programming model for reference set identification in DEA.

Findings

The proposed method effectively identifies reference sets using least distance projections.

It measures closest RTS (CRTS) in a two-stage process.

Numerical example demonstrates advantages over most distance frameworks.

Abstract

In data envelopment model (DEA), while either the most or least distance based frameworks can be implemented for targeting, the latter is often more relevant than the former from a managerial point of view due to easy attainability of the targets. To date, the two projection-dependent problems of reference set identification and returns to scale (RTS) measurement have been extensively discussed in DEA literature. To the best of our knowledge, nonetheless, there exists only one study which uses a closest projection for identifying reference set and accomplishes this task through a primal-dual linear programming based method. Motivated by this, we investigate the two aforementioned problems in a least distance based framework. First, we propose a lexicographic multiple-objective programming problem to find a unique closest projection for an inefficient decision making unit (DMU). Associated with the founded projection, we then identify all the possible closest reference DMUs by developing a linear programming model in the envelopment form. For an inefficient DMU, we next define closest RTS (CRTS) as the RTS of its least distance projection and measure the CRTS in two stages. Finally, we illustrate our proposed approach by a numerical example and compare the results with those found in the most distance based framework.