Abstract
Over the years following Debreu’s (1951) seminal formulation of a “coefficient of resource utilization”, a large number of indexes of technological inefficiency have been specified and a spate of papers has examined the properties satisfied by these indexes. This paper approaches the subject more synthetically, presenting generic results on classes of indexes and their properties. In particular, we consider a broad class of indexes containing almost all known indexes and a partition of this class into two subsets, slacks-based indexes and path-based indexes. Slacks-based indexes are expressed in terms of additive or multiplicative slacks for all inputs and outputs, and particular indexes are generated by specifying the form of aggregation over the coordinate-wise slacks. Path-based indexes are expressed in terms of a common contraction/expansion factor, and particular indexes are generated by specifying the form of the path to the frontier of the technology. Owing to an impossibility result in one of our earlier papers, we know that the set of all inefficiency indexes can be partitioned into three subsets: those that satisfy continuity (in quantities and technologies) and violate indication (equal to some specified value if and only if the quantity vector is efficient), those that satisfy indication and violate continuity, and those that satisfy neither. We prove two generic theorems establishing the equivalence of these two partitions: all slacks-based indexes satisfy indication and hence violate continuity, and all path-based indexes satisfy continuity and hence violate indication. We also discuss the few indexes that do not belong to either of these two sets. Our hope is that these results will help guide decisions about specification of the form of efficiency indexes used in empirical analysis.
Similar content being viewed by others
Notes
Vector notation: \(\bar x\) ≥ x if \(\bar x_i\) ≥ x i for all i; \(\bar x\) > x if \(\bar x_i\) ≥ x i for all i and \(\bar x\) ≠ x; and \(\bar x \gg x\) if \(\bar x_i >x_i\) for all i.
In other words, ∂(T) is the set of weakly efficient production vectors belonging to technologies satisfying (i).
See Charnes et al. (1994).
Under our assumptions—notably, non-emptiness, closedness, and boundness of P(x) for all \(x \in {\bf{R}}_ + ^n\)—Eff(T) is non-empty (see, e.g., Shephard, 1970, p.15).
It is a straightforward matter of renormalization to convert an efficiency measure (typically mapping into the (0,1] interval) into an inefficiency measure.
Note that, to focus on the salient issues at hand, we restrict the domain of I to positive values of input and output quantities, thus avoiding some distracting boundary issues (see Levkoff et al. (2012) for an analysis of boundary problems).
Of course, proportional slacks are conversely converted to additive slacks by s i = (1 − α i )x i for all i and t j = ((1 − β j )/β j )y j for all j.
Note that, under our restrictions on T, the constraint set in the maximization problem (1) for a given \(\left\langle {x,y} \right\rangle \) is compact; hence the maximum exists.
See Russell and Schworm, 2011 for an in-depth comparison of these indexes and those that follow below.
Charnes, Cooper, Golany, Seiford, and Stutz also specified the “Additive DEA Model” (eschewing the weights), but that index is not independent of units of measurement and hence has been superceded by the Quantity-Weighted Additive Index.
The directional distance index is adapted from the shortage function of Luenberger (1992) to the measurement of efficiency by Chung et al. (1997). Both restrict g only to the non-negative orthant, but restriction to the positive orthant improves the properties of the index (see Russell and Schworm, 2011). Under the restriction \(g \gg 0\), the shortage function is a weighted \(\left\| \cdot \right\|_\infty \) distance function, where the weights are inverses of the respective components of g (see Briec, 1998).
Briec (2000) derives this index from the directional-distance function by using the definition of I DD with the direction g = \(\left\langle {x,y} \right\rangle \).
Other than the “Additive Model” of Charnes et al. (1985), which has been superseded by the Weighted Additive Index formulated in the same paper for the explicit purpose of establishing unit invariance.
It is not possible to show that either class of indexes satisfies strict monotonicity: \(\left\langle {x,y} \right\rangle \in T,\left\langle {x{\prime},y{\prime}} \right\rangle \in T\), and \(\left\langle {x{\prime}, - y{\prime}} \right\rangle \ge \left\langle {x, - y} \right\rangle \) imply I(x′, y′, T) > I(x, y, T). One particular specification, the Weighted Additive Index, does satisfy this condition, but it has the disadvantage of requiring the use of arbitrary weights to correct for the dependence of the “Additive Model” on unit changes.
The proof follows closely the continuity proof of selected slacks-based indexes in Russell and Schworm (2011, pp. 155–156).
In fact, the Hölder distance function is obtained by extending Ω to a set-valued map, \(\Omega :\langle x,y,\lambda ) \mapsto \left\langle {x,y} \right\rangle + B_p(0,1]\), in which case the Hölder distance function is given by \(D_p(x,y) = {\sum} \left\{ {\left. \lambda \right|}\Omega (x,y,\lambda ) \subset T + ( {\bf{R}}_ + ^n \times {\bf{R}}_ - ^m ) \right\},\) where B p (0, 1] is the closed unit ball with norm p. See Briec (1998).
In a private communication, Walter Briec points out an interesting feature, unrelated to our axiomatic structure, of our binary taxonomy.: “The path based indexes are strongly related to the characterization of the technology. The Farrell measure can be extended to the whole nonnegative Euclidean vector space. In such a case, it is essentially the inverse of the Shephard’s distance function. This is the same thing with the hyperbolic measure and the directional distance function. Though these distance functions may yield infeasibilities, they characterise the technology. Moreover, under convexity assumptions they are dual the the usual cost, revenue and profit functions. The things are much more complicated with the slack-based measures. It is difficult to extend them over \(R_ + ^{m + n}\). Hence they do not characterize the technology, at least under their standard forms. Moreover, their dual properties are not very clear.” (See, e.g., Luenberger (1992) and Färe and Primont (1995) for the analysis of duals to the path-based measures identified above.)
In fact, the hyperbolic (in)efficiency index is a straightforward generalization of the Debreu–Farrell index (which takes a radial path to the boundary of the input-requirement set), and the Färe–Grosskopf–Lovell index is a straightforward generalization of the Färe–Lovell index.
References
Banker R, Cooper WW (1994) Validation and generalization of DEA and its uses. TOP 2:249–314
Briec W (1998) Hölder distance function and measurement of technical efficiency. J Prod Anal 11:111–131
Briec W (2000) An extended Färe-Lovell technical efficiency measure. Int J Prod Econ 65:191–199
Charnes A, Cooper WW, Golany B, Seiford L, Stutz J (1985) Foundations of data envelopment analysis for Pareto-Koopmans efficient empirical production functions. J Econom 30:91–107
Charnes A, Cooper WW, Lewin AY, Seiford LM (1994) Data envelopment analysis: theory, methodology, and application. Kluwer Academic Publishers, Boston
Chung YH, Färe R, Grosskopf S (1997) Productivity and undesirable outputs: a directional distance function approach. J Environ Manag 51:229–240
Coelli T (1998) A multi-stage methodology for the solution of orientated DEA models. Oper Res Lett 23:143–149
Cooper WW, Park KS, Pastor JT (1999) RAM: a range adjusted measure of inefficiency for use with additive models, and relations to other models and measures in DEA. J Prod Anal 11:5–42
Cooper WW, Pastor JT (1995) Global efficiency measurement in DEA. Working paper. Depto Est e Inv. Oper, Universidad Alicante, Alicante, Spain.
Debreu G (1951) The coefficient of resource utilization. Econometrica 19:273–292
Färe R, Grosskopf S, Lovell CAK (1985) The measurement of efficiency of production. Kluwer-Nijhoff, Boston
Färe R, Lovell CAK (1978) Measuring the technical efficiency of production. J Econ Theory 19:150–162
Färe R, Primont D (1995) Multi-output production and duality: theory and applications. Kluwer Academic Publishers, Boston
Farrell MJ (1956) The measurement of productive efficiency. J R Stat Soc Ser A Gen 120:253–282
Fukuyama H, Weber WL (2009) A directional slacks-based measure of technical efficiency. Socioecon Plann Sci 43:274–287
Koopmans TC (1951) Analysis of production as an efficient combination of activities. In: Koopmans TC (ed) Activity analysis of production and allocation. Physica-Verlag, Heidelberg
Levkoff SB, Russell RR, Schworm W (2012) Boundary problems with the ‘Russell’ graph measure of technical efficiency: a refinement. J Prod Anal 37:239–248
Luenberger DG (1992) New optimality principles for economic efficiency and equilibrium J Optim Theory Appl 75:221–264
Pastor JT, Ruiz JL, Sirvent I (1999) An enhanced DEA Russell graph efficiency measure. Eur J Oper Res 115:596–607
Russell RR (1990) Continuity of measures of technical efficiency. J Econ Theory 35:109–126
Russell RR, Schworm W (2009) Axiomatic foundations of efficiency measurement on data-generated technologies. J Prod Anal 31:77–86
Russell RR, Schworm W (2011) Properties of inefficiency indexes on 〈Input, Output〉 space. J Prod Anal 36:143–156
Shephard RW (1970) Theory of cost and production functions. Princeton University Press, Princeton
Tone K (2001) A slacks-based measure of efficiency in data envelopment analysis. Eur J Oper Res 130:498–509
Zieschang KO (1984) An extended farrell technical efficiency measure. J Econ Theory 33:387–396
Acknowledgements
We thank Walter Briec, Rolf Färe, Antonio Peyrache, and Valentin Zelenyuk, as well as the referees of the JPA, for comments and suggestions. The paper has also benefited from discussions during seminars at Lecce University and the University of Queensland.
Author information
Authors and Affiliations
Corresponding authors
Ethics declarations
Conflict of interest
The authors declare that they have no competing interests.
Appendix: Proof of the Theorem
Appendix: Proof of the Theorem
1.1 Proof of (a) Slacks-based indexes violate continuity and satisfy indication
It suffices to prove that I s satisfies (I), since the R–S Impossibility Result then implies that this class of indexes violates (C).
Suppose that \(\left\langle {x,y} \right\rangle \)∈T, with \(\left\langle {x,y} \right\rangle \gg 0^{[n + m]}\), is not efficient so that there exists a production vector \(\left\langle {x{\prime},y{\prime}} \right\rangle \in T\) satisfying \(\left\langle {x{\prime},y{\prime}} \right\rangle \gg 0^{[n + m]}\) and \(\left\langle {x{\prime}, - y{\prime}} \right\rangle < \left\langle {x, - y} \right\rangle \). If \(\left\langle {s{\prime},t{\prime}} \right\rangle \) is the solution to
then \(\left\langle {x{\prime} - s{\prime},y{\prime} + t{\prime}} \right\rangle \) is a boundary point of T. As \(\left\langle {x{\prime} - s{\prime},y{\prime} + t{\prime}} \right\rangle \le \left\langle {x{\prime}, - y{\prime}} \right\rangle < \left\langle {x, - y} \right\rangle \), there exists an \(\left\langle {s^o,t^o} \right\rangle \in {\bf{R}}_ + ^{n + m}\) such that \(\left\langle {s^o,t^o} \right\rangle >\left\langle {s{\prime},t{\prime}} \right\rangle \) and \(\left\langle {x - s^o,y + y^o} \right\rangle = \left\langle {x{\prime} - s{\prime},y{\prime} + t{\prime}} \right\rangle \in T\). As ψ is increasing in \(\left\langle {s,t} \right\rangle \), it must be that \(I^s(x,y,T) >I^s\left( {x{\prime},y{\prime},T} \right) \ge 0\).
Next suppose that I s(x,y,T) > 0. Then \(\left\langle {s,t} \right\rangle \) > 0[n+m], so that there exists a point \(\left\langle {x{\prime},y{\prime}} \right\rangle \in T\) satisfying \(\left\langle {x{\prime},y{\prime}} \right\rangle \gg 0^{[n + m]}\) and \(\left\langle {x{\prime}, - y{\prime}} \right\rangle < \left\langle {x, - y} \right\rangle \). Therefore, (x, y) is inefficient.
Consequently, I s satisfies indication (I) and therefore violates (C). ♢
1.2 Proof of (b) Path-Based Indexes Satisfy Continuity and Violate Indication
It suffices to prove that I p satisfies (C), since the R–S Impossibility Result then implies that this class of indexes violates (I).
Consider a sequence \(\left\{ {x^\nu ,y^\nu ,T^\nu } \right\}_{\nu = 1}^\infty \) converging (in the topology of closed convergence) to \(\left\langle {x^o,y^o,T^o} \right\rangle \). We need to show that I p(x ν, y ν, T ν) converges to I(x o, y o, T o).
Before addressing directly the issue of continuity of I p, we formally establish two (intuitively obvious) facts.
-
(i)
The intersection of the curve \(\left\{ {\Omega (x,y,\lambda )\left| {\lambda \in \Lambda } \right.} \right\}\) and the frontier ∂(T) is a singleton for each \(\left\langle {x,y,T} \right\rangle \)∈ Ξ (and hence for each element of the sequence \(\left\{ {x^\nu ,y^\nu ,T^\nu } \right\}_{\nu = 1}^\infty \) and its limit \(\left\langle {x^o,y^o,T^o} \right\rangle \) ).
Suppose not: for some \(\left\langle {x,y,T} \right\rangle \) ∈ Ξ, \(\left\{ {\Omega (x,y,\lambda )\left| {\lambda \in \Lambda } \right.} \right\} \cap \partial (T)\) contains two points, \(\left\langle {x{\prime},y{\prime}} \right\rangle \) and \(\left\langle {\hat x,\hat y} \right\rangle \). Since \(\Omega _i^{in}\) is decreasing in λ for all i and \(\Omega _j^{out}\) is increasing in λ for all j, either \(\left\langle {x{\prime}, - y{\prime}} \right\rangle \gg \left\langle {\hat x, - \hat y} \right\rangle \) or \(\left\langle {\hat x, - \hat y} \right\rangle \gg \left\langle {x{\prime}, - y{\prime}} \right\rangle \). As these are frontier points, each inequality violates the free disposability assumption (FD), which proves the result.
-
(ii)
Since Ω is continuous, the sequence of paths \(\left\{ {\Omega \left( {x^\nu ,y^\nu ,\lambda } \right)\left| {\lambda \in \Lambda } \right.} \right\}_{\nu = 1}^\infty \) converges to the path \(\left\{ {\Omega \left( {x^o,y^o,\lambda } \right)\left| {\lambda \in \Lambda } \right.} \right\}\) (in the topology of closed convergence).
Consider a sequence, \(\left\{ {\Omega \left( {x^\nu ,y^\nu ,\lambda ^\nu } \right)} \right\}_{\nu = 1}^\infty \), converging to Ω(x o, y o, λ′), where (by continuity of Ω) \(\lambda {\prime} = {\mathrm{lim}}_{\nu \to \infty }\,\lambda ^\nu \). As Λ is a closed set, λ′ ∈ Λ, establishing that \(\Omega \left( {x^o,y^o,\lambda {\prime}} \right) \in \left\{ {\Omega \left( {x^o,y^o,\lambda } \right)\left| {\lambda \in \Lambda } \right.} \right\}\).
Now consider a point Ω(x o, y o, λ′) with λ′ ∈ Λ and construct the sequence \(\left\{ {\Omega \left( {x^\nu ,y^\nu ,\lambda {\prime}} \right)} \right\}_{\nu = 1}^\infty \). Clearly, \(\left\{ {\Omega \left( {x^\nu ,y^\nu ,\lambda {\prime}} \right)} \right\} \in \left\{ {\Omega \left( {x^\nu ,y^\nu ,\lambda } \right)\left| {\lambda \in \Lambda } \right.} \right\}\) for all ν and \({\mathrm{lim}}_{\nu \to \infty }\Omega \left( {x^\nu ,y^\nu ,\lambda {\prime}} \right) = \Omega \left( {x^o,y^o,\lambda {\prime}} \right)\), establishing convergence.
-
(iii)
Continuity of I p.
Assume I p is not continuous. Then there exists a sequence \(\left\{ {\left\langle {x^\nu ,y^\nu ,T^\nu } \right\rangle } \right\}_{\nu = 1}^\infty \) converging to \(\left\langle {x^o,y^o,T^o} \right\rangle \) for which the sequence \(\left\{ {I^p\left( {x^\nu ,y^\nu ,T^\nu } \right)} \right\}_{\nu = 1}^\infty \) does not converge to I p(x o, y o, T o).
Let λ ν = I p(x ν, y ν, T ν) for ν = 1…,∞ and λ o = I p(x o, y o, T o) so that the sequence \(\left\{ {\lambda ^\nu } \right\}_{\nu _k = 1}^\infty \) does not converge to λ o. Therefore, there exists a subsequence \(\left\{ {\lambda ^{\nu _k}} \right\}_{\nu _k = 1}^\infty \) such that
for all elements of the subsequence.
By the definition of path-based indexes (2), Ω(x, y, I p(x, y, T)) is the intersection of the paths \(\left\{ {\Omega (x,y,T)\left| {\lambda \in \Lambda } \right.} \right\}\) and ∂(T). Moreover, (i) above implies that this intersection consists of a single point. Finally, condition (ii) above implies that
and
Since Ω is continuous and \(\lambda ^\nu \not \to \lambda ^o\), the sequence \(\left\{ {\Omega \left( {x^o,y^o,\lambda ^\nu } \right)} \right\}_{\nu = 1}^\infty \) does not converge to Ω(x o, y o , λ o).
To ease the notation, denote Ω(x ν, y ν, I p(x ν, y ν, T ν)) by \(\left\langle {\tilde x^\nu ,\tilde y^\nu } \right\rangle \) for all ν and Ω(x o, y o, I p(x o, y o, T o)) by \(\left\langle {\tilde x^o,\tilde y^o} \right\rangle \).
The condition (c) on Ω implies that
for all elements in the subsequence \(\left\{ {\nu _k} \right\}_{k = 1}^\infty \).
As T ν → T o and \(\left\{ {\Omega \left( {x^\nu ,y^\nu ,\lambda } \right)\left| {\lambda \in \Lambda } \right.} \right\}_{\nu = 1}^\infty \) converges to \(\left\{ {\Omega \left( {x^o,y^o,\lambda } \right)\left| {\lambda \in \Lambda } \right.} \right\}\), there exists a sequence \(\left\{ {\bar x^\nu ,\bar y^\nu } \right\}_{\nu = 1}^\infty \subset \partial \left( {T^\nu } \right)\) converging to \(\left\langle {\tilde x^o,\tilde y^o} \right\rangle \). Hence, there exists a subsequence of {ν k } denoted \(\left\{ {\nu _{k_j}} \right\}\) such that
for all elements of the subsequence.
As \(\left\langle {\tilde x^\nu ,\tilde y^\nu } \right\rangle \) and \(\left\langle {\bar x^\nu ,\bar y^\nu } \right\rangle \) are contained in the frontier of T ν, either of these strict inequalities violates free disposability (FD).
This contradiction implies that I p(x ν, y ν, T ν) converges to I p(x o, y o, T o) so that I p satisfies continuity (C), in which case it must violate indication (I). ♢
Rights and permissions
About this article
Cite this article
Russell, R.R., Schworm, W. Technological inefficiency indexes: a binary taxonomy and a generic theorem. J Prod Anal 49, 17–23 (2018). https://doi.org/10.1007/s11123-017-0518-2
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11123-017-0518-2