Chapter 2 PRODUCTIVITY, EFFICIENCY, AND DATA ENVELOPMENT ANALYSIS 2.1 Introd !

t"on Any decision-making problem faced by an economic agent (like a consumer or a producer) has three basic features. First, there are the variables the values of which are chosen by the agent. These are the choice or decision variables in the problem. econd, there are the restrictions that define the set of feasible values from which to choose. Finally, there is some criterion function that assigns different values to the outcomes from alternative decisions. !n the conte"t of production, the decision-making agent is the firm. The choice variables are the #uantities of outputs to be produced as well as the #uantities of inputs used. The input-output combination selected by the firm must be technically feasible in the sense that it must be possible to produce the output bundle selected from the associated input bundle. For a commercial firm facing well defined market prices of inputs and outputs, the profit measured by the difference between revenue and cost serves as the criterion of choice. !t is possible, therefore, to rank the alternative feasible input-output combinations in order of the profit that results from them. $hen the criterion function has a finite ma"imum value attainable over the feasible set of the choice variables, this ma"imum value can be used as a benchmark for evaluating the efficiency of a decision-making agent. The closer the actual profit of a firm is to the ma"imum attainable, the greater is its efficiency. !t is important to recogni%e that the scope of decision-making defines what can be regarded as choice variables and the criterion function has to be appropriately formulated. For e"ample, in many practical situations, the output produced may be an assigned task that is e"ogenously determined. The producer then only chooses between alternative input bundles that can produce the targeted output. !n this conte"t, efficiency lies in minimi%ing the cost of production. This is true for many not-for-profit service organi%ations like hospitals, schools, or disaster relief agencies. &ven within a for-profit business organi%ation, as one goes down the decision-making hierarchy, the number of choice variables declines. For e"ample, at the lower end of a manufacturing firm is the production foreman on the shop floor, who is typically assigned a specific input bundle and has to manage the workers under his supervision so as to produce the ma"imum possible output from these inputs. Therefore, at this level efficiency is to be measured by a comparison of the actual output produced with what is deemed to be ma"imally possible. For the foreman, input #uantities are non-discretionary variables. The obvious payoff from efficiency measurement is that it provides an ob'ective basis for evaluating the

13

performance of a decision-making agent. The outcome at the highest level of efficiency (e.g., the ma"imum profit achievable) provides an absolute standard for management by ob'ectives. Further, comparison of efficiency across decision makers at the same level provides a basis for differential rewards. (oreover, one can assess the impact of various institutional or organi%ational changes by analy%ing how it affects the efficiency. For e"ample, the economic reforms in )hinese agriculture introduced in the post-(ao era allowed private farming to a limited e"tent. The farmers* right to appropriate the surplus (at least in part) considerably increased the output #uantities produced from the same input bundle. This increase in efficiency provides an economic 'ustification for these reforms. Any attempt to measure efficiency raises two #uestions - one conceptual and the other practical. At the conceptual level+ what do we mean by the efficiency of a decision maker? (ore specifically, where does inefficiency come from, !f the laws of production are interpreted as physical laws, identical sets of inputs must produce identical output #uantities. Therefore, if the same input bundle results in two different output #uantities on two different occasions, it must be true that differences in some other factors relevant for production but not included in the inputoutput list account for this discrepancy. !n agricultural production, for e"ample, the ma"imum output producible from a given input bundle can vary due to random differences in weather. The stochastic production frontier models allow random shifts in the frontier to accommodate such factors. -ut even after such accommodation, firms do differ in efficiency. !n the spirit of tigler (./01) one can argue that every observed input-output combination is efficient and any measured inefficiency is due to difference in e"cluded variables. Thus, if a farmer fails to attain what is considered to be the ma"imum producible level of output from a given bundle of inputs, it must be due to the fact that he did not either put in the re#uired level of effort or had a lower ability or human capital. imilarly, measured inefficiency of the production supervisor reflects a lower level or #uality of managerial input in monitoring efforts of subordinates. 2ence, a lower level of efficiency can be ascribed to lower effort, ability, or aptitude. At the practical level, the benchmark for efficiency measurement depends critically on how the feasible set of input-output bundles is specified. An input-output combination is considered feasible so long as the output #uantity does not e"ceed the value of the estimated function at the specified input #uantities. !n the absence of any clearly defined engineering formula relating inputs to outputs, this is essentially an empirical issue. A widely applied approach is econometric estimation of a stochastic production frontier. A nonparametric alternative to the econometric approach is provided by the method of 3ata &nvelopment Analysis (3&A) that builds on the pioneering work of Farrell (./40).

14

At the lowest level of decision making the ob'ective is to produce the ma"imum #uantity of output from a specific input bundle, and the benchmark is determined by the technology itself and comparison of the actual output produced with the benchmark #uantity yields a measure of technical efficiency . This is different from economic efficiency where one compares the profit resulting from the actual input-output bundle with the ma"imum profit possible. 2ere, apart from the technology, the market prices of inputs and outputs also play an important role. As will be shown later, technical efficiency is an important component of economic efficiency and a firm cannot achieve full economic efficiency unless it is technically efficient. !n this chapter, we focus on technical efficiency and show how 3ata &nvelopment Analysis can be used to measure it. 2.2 Prod !t"#"t$ and Te!hn"!a% E&&"!"en!$+ 5roduction is an act of transforming inputs into outputs. -ecause the ob'ective of production is to create value through transformation, outputs are, in general, desirable outcomes. 2ence, more output is better. At the same time, inputs are valuable resources with alternative uses. 6nspent #uantity of any input can be used for producing more of the same output or to produce a different output. The twin ob'ectives of efficient resource utili%ation by a firm are (a) to produce as much output as possible from a specific #uantity of input and, at the same time, (b) to produce a specific #uantity of output using as little input as possible. An input-output combination is a feasible production plan if the output #uantity can be produced from the associated input #uantity. The technology available to a firm at a given point in time defines which input-output combinations are feasible. Two concepts commonly used to characteri%e a firm*s resource utili%ation performance are (i) productivity and (ii) efficiency. These two concepts are often treated as e#uivalent in the sense that if firm A is more productive than firm B then it is generally believed that firm A must also be more efficient. This is not always true, however. Although closely related, they are fundamentally different concepts. For one thing, productivity is a descriptive measure of performance. &fficiency, on the other hand, is a normative measure. The difference between the two can be easily understood using an e"ample of two firms from a single input, single output industry.

2.' The S"n(%e O tp t, S"n(%e Inp t Te!hno%o($)

uppose that firm A uses xA units of the input x to produce yA units of the output y. firm B, on the other hand,

15

produces output yB from input xB . Then the average productivities of the two firms are

AP A =
and

yA xA yB xB

for firm A

AP B =

for firm B.

!f APA > APB we conclude that firm A is more productive than firm B. $e can even measure the productivity inde" of firm A relative to firm B as

A,B =

AP A = y A / x A . yB / xB AP B

!f this productivity inde" e"ceeds ., firm A is more productive than firm B. The higher it goes above unity, the more productive is firm A relative to firm B. Assuming that ( xA, yA = (!",# and ( xB, yB = ("$,% ,

AP A =
Thus,

# % and AP B = . !" "$ !' = !&% . %

A,B =

2ence, firm A is ..0 times as productive as firm -. An important point to note is that in the single output, single input case we do not need to know the technology to measure either the absolute or the relative productivity of a firm. !n this respect, APA or APB merely describes the performance of the individual firm without evaluating such performance. 7f course, the productivity inde" does provide a comparison between the firms. 8evertheless, it uses no reference technology for a benchmark. 8ow suppose that we do know that the technology is described by the production function y = f (x & ('&!

Then, y(A = f (xA is the ma"imum output producible from input xA. imilarly, y(B = f (xB is the ma"imum output that can be produced from xB. $e can measure the technical efficiency of a firm by comparing its actual output with the ma"imum producible #uantity from its observed input. This is an out)ut-oriented measure of efficiency. For firm A, the output-oriented technical efficiency is

16

*+ , =

yA yA
(

! .

('&'a

imilarly, for firm B,

*+ , =

yB yB
(

! .

('&'b

!f firm A produced the ma"imum producible output ( y(A from input xA its average productivity would have been
( AP A =

yA xA

whereas at the observed input-output level, its productivity is

AP A =

yA xA

Thus, an alternative characteri%ation of its output-oriented technical efficiency is

*+ , =
imilarly,

yA
( yA

yA / xA AP A = . ( ( yA / xA AP A

('&#a

AP B =
and

yB xB

*+ , =

AP B . ( AP B

('&#b

!n this sense, the technical efficiency of a firm is its productivity inde" relative to a hypothetical firm producing the ma"imum output possible from the same input #uantity that the observed firm is using. Thus,

*+ , = A,A(
and

('&$a

*+ , = B,B( &
17

('&$b

!n Figure 9.., we measure input x along the hori%ontal a"is and output y up the vertical a"is. 5oints PA and PB represent the input-output bundles of firms A and B respectively. Average productivity of A is e#ual to the slope of the line ,PA. imilarly, slope of ,PB measures the average productivity of B. -ecause the input-output combinations of the two firms are actually observed, we know that these two are feasible points. 3ifferent information is necessary, as noted above, to measure productivity and efficiency. First, in order to measure the average productivities of the two firms and to compare their productivities, we do not need to know anything beyond these two points .. !n particular, we do not need to know what other input-output bundles are feasible. That is, no knowledge of the technology is necessary. !n order to determine the efficiency of A, we need the point P(A showing the ma"imum output y(A producible from A*s input #uantity xA. imilarly, point P(B provides a benchmark for firm B. :ocation of these two reference points depends upon the functional form and parameters of the production frontier f (x . For firm A,

*+ , =

yA
( yA

P A x A slope of 75 A = . ; ( P A x A slope of 75 A

imilarly, for firm B,

*+ , =

yB
( yB

P B x B slope of 75 = ; . ( P B x B slope of 75 -

These ratios are measures of output-oriented technical efficiency. The graph of the production function y = f (x is the frontier of the production possibility set defined by the underlying technology. 5oints P(A and P(B are vertical pro'ections of the points PA and PB onto the frontier. !n both cases, we hold the observed input bundle unchanged and e"pand the output level till we reach the frontier. This is known as the output-augmenting or output-oriented approach. An alternative is the input-saving or in)ut-oriented approach. This is shown in Figure 9.9. !n this case, the output level (yA or yB) remains unchanged and input #uantities are reduced proportionately till the frontier is reached. For firm A, the input-oriented pro'ection on to the frontier would be the point ) ; A , where output yA is produced from
; input x ; A . imilarly, for firm B, the input-oriented pro'ection is the point ) B showing the output level yB being

18

; produced from input x B

The pair of input-oriented technical efficiency measures for the two firms are as follows+

*+ =
and

A -

x; A xA

*+ =

B -

; xB

xB

! .

As before,
A *+ - =

slope of 75A < A, A; slope of 7 ) ; A

and
B *+ - =

slope of 75< -, B; . ; slope of 7 ) B

!n practice, whether the input- or the output-oriented measure is more appropriate would depend on whether input conservation is more important than output augmentation. =enerally the input- and output-oriented measures of technical efficiency of a firm will be different. The e"ception is in the case of constant returns to scale ()> ) when both approaches yield the same measure of efficiency. uppose that the observed input-output combination is (x., y. . Further, the ma"imum producible output from x. is y.( while the minimum input #uantity that can produce y. is x.(& Thus, both (x., y.( and (x.(, y. are technically efficient points lying on the frontier. For the input- and output-oriented technical efficiency measures to be e#ual, we need

x? y = ?; . x? y?

This is e#uivalent to

y? x?
;

y = ? . x?

19

Thus, the average productivity at two different points on the frontier remains the same. This, of course, implies constant returns to scale. -efore we elaborate on the case of )> , we note that a firm may be more productive without being more efficient than another firm . uppose that

f (x = x &
Then,
( ( y A = !" = $ and y B = "$ = / &

Thus,

*+ , =
and

yA
( yA

# $

*+ , =

yB
( yB

% . /

)learly, firm B is more efficient that firm A. At the same time,

AP A =

yA xA

# yB % > = = AP B & !" x B "$

Thus, A is more productive without being more efficient than B. uppose that firm A actually produces y(A rather than yA from input xA. !n that case, both are e#ual to unity. imilarly, if B also produced y(B instead of yB from input xB, both been unity. 8evertheless,
A A *+ , and *+ -

B B *+ , and *+ - would also have

AP A =

yA xA

>

yB xB

= AP( B .

!n that case, firm A is more productive without being more efficient than firm B. $e now consider the case of constant returns to scale. For a single-output, single-input technology, the )>

20

frontier is a ray through the origin as shown in Figure 9.@. 2ere the production function is of the form f (x = k x, k > .&

Along this frontier (i.e., at every point on this frontier) average productivity is the constant k. As before,

*+ , =
and

y A slope of 75A = ( y A slope of 75; A

x A slope of 7 PA . A *+ - = = x A slope of 75 A
imilarly,

;;

*+ , =
and

yB
( yB

slope of 75 . slope of 75; -

x B slope of 7 PB . B *+ - = = x B slope of 75 B

;;

-ut, points

P ;; A , P A , P

;;

, and P(B

are all on the same ray through the origin. 2ence,

A A B B *+ , = *+ - and *+ , = *+ - &

Thus, when the technology e"hibits )> , input- and output-oriented measures of technical efficiency are identical. Further,
A ( *+ , AP A / AP( AP A & AP A A = = . B ( ( *+ , AP B / AP B AP B AP B

-ut, under )> , AP(A = AP(B. 2ence, when the technology e"hibits constant returns to scale,

21

A,B =

A AP A *+ , *+ = = . B B AP B *+ , *+ -

('&0

Therefore, higher productivity always implies greater efficiency only under )> . 2.* M %t"p%e Inp t, M %t"p%e O tp t Te!hno%o($)

7nce we step outside the simplified world of single-input, single-output production, the concept of average productivity measured by the output-input #uantity ratio breaks down. &ven in the relatively simple case of oneoutput, two-input production, we can no longer discuss average productivity in an une#uivocal manner. Assume that firm A uses x!A of input . and x'A of input 9 to produce the scalar output yA. imilarly, firm B produces output yB using x!B of input . and x'B of input 9. 8ow we have two different sets of average productivities+

AP A =
and

yA x!A

, AP ' A=

yA x 'A

for firm A

AP B =

yB x!B

, AP ' B=

yB x 'B

for firm B.

!t is inappropriate to treat firm A as more productive than firm B whenever AP!A e"ceeds AP!B because it is possible that at the same time AP'B e"ceeds AP'A. A firm*s average productivity in respect of one input depends on the #uantity of the other input as well. Therefore, measuring a firm*s productivity relying on a single input disregarding other inputs is wrong. 6nfortunately, this was the common practice in the 6. . -ureau of :abor tatistics and other important agencies for many years. (a'or business economists often compare output per man-hour across regions or over time to study Aproductivity changesA in manufacturing. -ut unless one includes the #uantities of capital, energy, and other inputs, such productivity measures fail to reflect total factor )roducti1ity . !n the single-output, multiple-input case, we need to aggregate the individual input #uantities into a composite input. $e can then measure productivity by the ratio of output #uantity to the #uantity of this composite input. $hen multiple outputs are involved, a similar aggregate measure of output is also needed. 7ne practical approach uses market prices of inputs for aggregation. uppose that r! and r' are the prices of the two inputs. Then,

22

3 A = r. x!A 2 r9 x 'A
and

('&"a

3 B = r. x!B 2 r9 x 'B

('&"b

are the aggregate input #uantities for A and B, respectively. !n that case,

AP A =
and

yA 3A

r. x. A

yA + r9 x 9 A

('&%a

AP B =

yB 3B

r. x.B

yB . + r9 x 9 B

('&%b

-ut, obviously, the aggregate input bundles represent the input costs of the two firm s. Thus, a firm *s average productivity is merely the inverse of its average cost (A)). That is,

AP A =

! A4 A

and

AP B =

! A4 B

&

8ow suppose that each firm produced two outputs+ y! and y' .The output prices are 5! and 5', respectively. Then the aggregate outputs of the two firm s are measured as follows+

6 A = 5. y !A 2 5 9 y 'A for firm A

and

6 B = 5. y!B 2 5 9 y 'B for firm B.

!n that case,

AP A =
and

6 A = 5. y!A 2 5 9 y 'A r. x!A 2 r9 x 'A 3A

('&/a

AP B =

6 B = 5. y!B 2 5 9 y 'B . r. x!B 2 r9 x 'B 3B

('&/b

Thus, a firm *s average productivity is merely its (gross) rate of return on outlay. The firm with a higher rate of return is deemed to be the more productive one.

23

$hile this approach is simple and appealing from the perspective of a competitive market, input and output prices are not always available. This is especially true in the service sector (like education, public safety, etc.) where prices are seldom available for outputs. (oreover, in the presence of monopoly the market prices of inputs or outputs would be distorted. $hat we prefer, therefore, is a measure of productivity that would not re#uire the use of market prices. )onsider, again, a single-output, multiple-input production technology. Assume further that constant returns o scale holds. :et

x A = ( x!A , x 'A , &&&, x nA be the input bundle and y A the (scalar) output level of firm A.

Assume, further, that y = f (x is the production function showing the ma"imum output ( y( producible from the input bundle technical efficiency of firm A is

x&

Then, the

A *+ =

yA yA = & ( f ( xA yA

('&7

-ut, under )> ,

f ( x = f i xi ,
i=! n

where,

fi

f ( x xi

& Thus, it is possible to construct the

aggregate input #uantity as

3 A = f i ( x A xiA &
i=!
!n this case,

('&!.

AP A =

yA 3A

yA f ( xA

&

('&!!

imilarly, for firm B producing output y B from the input bundle

xB ,

24

AP B =

yB 3B

yB f ( xB

&

('&!'

As was pointed out earlier, in this case of )> , the productivity inde" of firm B relative to firm A is merely the ratio of their respective technical efficiency levels. !t may be noted that when market prices are actually available optimi%ing behavior of competitive firms would result in the prices of individual inputs being e#uated to the corresponding values of their marginal products. Thus,

ri = 5 f i 8

( i = ! , ' ,&&&, n

('&!#

where ri is the price of input i and 5 is the output price. !n that case,

AP A =

5 yA

r
i=!

*9 A *4 A

('&!$

xiA

where

*9 A and *4 A refer to the total revenue and the total cost of firm

A& imilarly, for firm B producing output

yB from input xB,

AP B =

5 yB

r x
i i=!

iB

= *9 B & *4 B

('&!0

This, it may be noted, is the return to the dollar criterion proposed by =oergescu->oegen (./4., p .?@). 7f course, one cannot take this approach when market prices are not available. !n fact, even when prices e"ist, they may not be appropriate weights for aggregation. For e"ample, a firm with higher market power may have higher output prices relative to a firm without market power. !n such cases, using actual prices for aggregation will e"aggerate productivity or efficiency of the former. $hen market prices cannot or should not be used, we need to construct shadow )rices of inputs for aggregation. For a competitive profit-ma"imi%ing firm, the price of any input deflated by the output price e#uals the marginal productivity of the input. Therefore, we can use these marginal productivities as shadow prices. 6nder )> , the production function is homogeneous of degree . in inputs. Thus, the

25

input #uantity inde" (like 3A and 3B) is also homogeneous of degree .. 6nlike the market prices, the shadow prices of inputs are not uniform across firms. >ather, these shadow prices depend on the input bundle at which the marginal productivities are evaluated. !n order to measure the technical efficiency of any observed input-output bundle one needs to know the ma"imum #uantity of output that can be produced from the relevant input bundle. 7ne possibility is to e"plicitly specify a production function. The value of this function at the input level under consideration denotes the ma"imum producible output #uantity. The more common practice is to estimate the parameters of the specified function empirically from a sample of input-output data. -ecause the least s#uares procedure permits observed points to lie above the fitted line, in a stochastic frontier model one includes a composite error, which is a sum of a one-sided disturbance term representing shortfalls of the actually produced output from the frontier due to inefficiency and a two-sided disturbance term representing upward or downward shifts in the frontier itself due to random factors. For the econometric procedure re#uires one must select a particular functional form (e.g., the )obb 3ouglas ) out of a number of alternatives. At the same input bundle x. the value attained by f(x. will depend on the functional form chosen. Further, the parameter estimates are also sensitive to the choice of the probability distributions specified for the disturbance terms. 3ata &nvelopment Analysis (3&A) is an alternative nonparametric method of measuring efficiency that uses mathematical programming rather than regression. 2ere, one circumvents the problem of specifying an e"plicit form of the production function and makes only a minimum number of assumptions about the underlying technology. Farrell (./40) formulated a linear programming model to measure the technical efficiency of a firm with reference to a benchmark technology characteri%ed by constant returns to scale. This efficiency measure corresponds to the coefficient of resource utili%ation defined by 3ebreu (./4.) and is the same as hephardBs distance function.(./4@). !n 3&A, we construct a benchmark technology from the observed input-output bundles of the firms in the sample. For this we make the following general assumptions about the production technology without specifying any functional form. These are fairly weak assumptions and hold for all technologies represented by a #uasi-concave and weakly monotonic production function.

(A.) All actually observed input-output combinations are feasible. An input-output bundle (x, y is feasible when the output bundle y can be produced from the input bundle x. uppose that we have a sample of : firms from an industry

26

producing m outputs from n inputs. :et x; =(xi;, x';,<,xn; be the input bundle of firm ; (; = !,',<,: and y; = (y!;, y';,<, ym; be its observed output bundle. Then, by (A.) each (x;, y; (; =!,',<, : is a feasible input-output bundle.

(A9) The production possibility set is conve". )onsider two feasible input-output bundles (xA, yA and (xB, yB . Then the (weighted) average input-output bundle ( x , y ) ,where x = x A + (. ) x B and y = y A + (. ) y B for some

satisfying ? ? is also feasible.

(A@) !nputs are freely disposable. !f (x., y. is feasible, then for any x

x., (x, y.

is also feasible.

(AC) 7utputs are freely disposable. !f (x., y. is feasible, then for any y

y., (x., y

is also feasible.

!f additionally we assume that constant returns to scale holds, (A4) !f (x, y is feasible, then for any k ., (kx, ky is also feasible.

!t is possible to empirically construct a production possibility set satisfying assumptions (A.-A4) from the observed data without any e"plicit specification of a production function.

D, y D ) where )onsider the input-output pair ( x

D = ; x ; , y D = ; y ; , ; = ., and x
. . .

D, y D ) is feasible. ?( ; = .,9,..., : ). -y (A.-A9), ( x

D , ( x, y D ) is also feasibly. 8e"t, by (AC), if y y D , ( x, y ) is feasible. Thus, using (A.-AC), 8ow, by (A@), if x x D, y D ) is feasible. !f, additionally, constant returns to scale is assumed, ( kx D , ky D) the input-output combination ( x
D for some k ?. Then, by construction, y = ky D and E is also a feasible bundle for any k .&& 3efine E x = kx

y k ; y ; and E x k ; x ; . 8e"t define ; = k ; . Then ; ? and E

. . .

= k . -ut k is only

restricted to be non-negative. 2ence, beyond non-negativity, there are no additional restrictions on the ; s. Therefore, based on the observed input-output #uantities and under the assumptions (A.-A4), we can define

27

the production possibility set or the technology set as follows+

* == ( ," y)+" ' " G y ' y G ' ?G ( '<., 9,. , 8 )F.

4 ' '<. '<.

'

('&!%

2ere the superscript 4 indicates that the technology is characteri%ed by constant returns to scale. 8ow consider the output-oriented technical efficiency of firm t producing output y t from the input bundle x t. $e want to determine what is the ma"imum output (y( producible from the same input bundle x t.. uppose that ( is the ma"imum value of such that (x t, y t lies within the technology set. Then y ; = ; y t and the output-oriented technical efficiency of firm t is
t t t *+ , = *+ , ( " , y ) <

yt . = ; . ('&!/ ; y

The linear programming (:5) problem for measuring the output-oriented technical efficiency is formulated in the following section. !n order to evaluate the input-oriented technical efficiency of any firm, we e"amine whether and to what e"tent it is possible to reduce its input(s) without reducing the output(s). This is #uite straightforward when only one input is involved. !n the presence of multiple inputs, a relevant #uestion would be whether reducing one input is more important than reducing some other input. $hen market prices of inputs are not available, one way to circumvent this problem is to look for e5ui-)ro)ortionate reduction in all inputs. This amounts to scaling down the observed input bundle without altering the input proportions. The input-oriented technical efficiency of firm t is ( where

( = min + (xt, yt *4.

8ote
!

('&!7
) 2ence, ( k " t , k ; y t ) T .

that

( " t , ; y t ) T ) .

etting

k=

!
(

we

get

t t ) " , y ) T .

( 7bviously, under )> , =

!
(

& That is, the input- and output-oriented technical efficiency measures are identical

28

in this case. 2.* Data En#e%op+ent Ana%$,", )harnes, )ooper, and >hodes (./0H, ./H.) introduced the method of 3ata &nvelopment Analysis (3&A) to address the problem of efficiency measurement for decision making units (3(6s) with multiple inputs and multiple outputs in the absence of market prices. They coined the phrase decision makin> units in order to include non-market agencies like schools, hospitals, and courts, which produce identifiable and measurable outputs from measurable inputs but generally lack market prices of outputs (and often of some inputs as well). !n this book, we regard a ?@A as synonymous with a firm& uppose that there are 8 firms each producing m outputs from n inputs. Firm t uses the input bundle
t t " < ( " it , " 9t ,.., " nt ) to produce the output bundle y < ( y.t , y9t .., ymt ) . As noted above, measurement of

average productivity re#uires aggregation of inputs and outputs. -ut no prices are available. $hat we would need in this situation is to use vectors of AshadowA prices of inputs and outputs. 3efine
t t u < ( u.t , u 9t ,.., u nt ) as the shadow price vector for inputs and v < ( v.t , v9t ,.., vmt ) the

shadow price vector for outputs. 6sing these prices for aggregation we get a measure of average productivity of firm t as follows+

APt =

1y
r=! n i=!

rt rt

ux

it it

v *y = t t u *"

t t

('&'.

8ote that the shadow price vectors used for aggregation vary across firms. Two restrictions are imposed, however. First, all of these shadow prices must be non-negative, although %ero prices are admissible for individual inputs and outputs. econd, and more importantly, the shadow prices have to be such that when aggregated using these prices, no firm*s input-output bundle results in average productivity greater than unity. This, of course, also ensures that APt ! for each firm t. These restrictions can be formulated as follows+

29

v * y r<. AP ; = t ' < n u *"

i <.

t '

vy u"

rt r'

.G ( ' < . , 9 ,..., t ,.., 8 ) G

('&'!

it i'
.

u it . 8 ( i = ! , ' ,&&&, n 8 1 rt . 8 ( r = ! , ' ,&& , m

!n general, there are many shadow price vectors (u t, v t) satisfying these restrictions. 7ut of them we choose one that ma"imi%es APt as defined above. This is a linear fractional functional programming problem and is #uite difficult to solve as it is. There is, however, a simple solution 9. 8ote that neither the ob'ective function ( APt ) nor the constraints are affected if all of the shadow prices are multiplied by a non-negative scale factor k (>.). Therefore, define

wit = k u it ( i = ! ,' ,&& , n

and

('&''a

) rt = k 1 rt ( r = ! , ' ,&&&, m

('&''b

Then the optimi%ation problem becomes

ma"

t t p *y t t w *"

s.t.

p *y . , ( ' < . , 9 ,..., 8 )G t ' w *" p ? G w t ? .

t

'

('&'#

8ow, set

!
i=!

u it xit

('&'$

30

Then, wtBxt = ! and the problem becomes

ma" prt yrt

r <.

s . t . prt yr' - wit "i' ? G ( ' < . , 9 ,... , t , .. , 8 )G

r <. i <.

wit "it < .G

i<.

('&'0

prt ? G ( r < . , 9 ,..., m ) + wit ? G ( i < . , 9 , ..., n ) .

This is a linear programming (:5) problem and can be solved using the simple" method. everal important points re#uire emphasis. First, the shadow prices of inputs cause the value of the observed input bundle xt of the firm under evaluation to e#ual unity. As a result, the value of the output bundle itself ( )tByt becomes a measure of its average productivity. econdly, at prices ( ) t,w t ) the observed input-output bundle of no firm in the sample would result in a positive surplus of revenue over cost. !f one interpreted the input prices as the imputed values of these scarce resources, then if the prices chosen are such that the imputed value of any input bundle is less than the imputed valuation of the output bundle it produces, then clearly the resources are being under-valued and the imputed input prices should be revised upwards. imilarly, if the output prices reflect the cost of the inputs drawn away from other uses to produce one unit of the output produces, then a total imputed value of the output bundle e"ceeding the total imputed cost of the input bundle used would imply that the output bundle is over valued. Finally, when constant returns to scale is assumed, the efficient production correspondence

31

C(x, y = . is homogeneous of degree %ero. Thus,

('&'"

C xi

C xi + y ; y ; = ?. ;

('&'%
C y;

Further, under competitive profit-ma"imi%ation @, price of output ; is proportional to

while the price of input i is

proportional to the negative of of the firm is %ero.

C xi

. 2ence, when shadow prices are derived from the technology, the imputed profit

This constraint applies to every firm including firm t, the one under consideration. As a result, the ma"imum value of the aggregate output 6t is unity implying that

t =

6t = 6 t = pt * yt . ( 6t

('&'/

Thus, the optimal solution of this :5 problem yields a measure of the output-oriented technical efficiency of firm t. For simplicity, consider the two-input, two-output case. :et y t = (y!t , y 't and xt = (x!t , x 't . Then the :5 problem becomes

ma" p. t y. t I p9 t y 9 t s. t. p. t p. t y.. I p 9 t y9. - w. t ".. - w 9 t " 9. ? G y.9 I p 9 t y 99 - w. t " 9. - w 9 t " 99 , ? G . . . . . . . ('&'7a p. t y. t I p9 t y 9 t - w. t ". t - w 9 t " 9 t ? G . . . . . . . I p9 t y9 8 - w. t ". 8 - w 9 t " 9 8 ? G w. t ". t I w 9 t " 9 t < . G p. t , p 9 t , w. t , w 9 t ? .

p. t y. 8

32

The dual of this :5 is the problem

min s ..t . . y.. I 9 y.9 I ... I t y. t I ... I 8 y. 8 y. t G

. y 9. I 9 y 99 I ... I t y 9 t I ... I 8 y 9 8 y 9 t G x.t - . x.. - 9 x.9 - ... - t x.t - ... - 8 x. : ? G x 9t - . x 9. - 9 x 99 - ... - t x 9 t - ... - 8 x 9 : ? G free , ' ? , ( ' < . , 9 ,... , 8 ) .

('&'7b

3efine

and

;=

; . Then minimi%ation of is e#uivalent to ma"imi%ation of . !n terms of the

redefined variables, the :5 problem now becomes

33

ma" s . t . ' y. ' y. t G

'<. 8

' y9 ' . < y9 t G

'<.

' ". ' J ". t G

'<. 8

('&#.

' "9 ' "9 t G

'<.

free G ' ? G ( ' < . , 9 ,..., 8 ) .

Thus, clearly,
!
(

from this problem e#uals

( ( from the previous problem. Further, by standard duality results,

e#uals p t; y t .

+xam)le'&! . Table . reports the observed input and output #uantities for 1 firms.

Firm inputKoutput output . (y.) output 9 (y9) input . (".) input 9 ("9)

A C 9 9 @

B / C 0 4

4 1 @ 1 0

? H 1 4 H

+ 0 4 H C

C .. H 1 1

34

Table 9. ..

!n order to evaluate the technical efficiency of firm 4, we solve the following :5 problem+

ma" s . t . C A I / - I 1 ) I H 3 I 0 & I .. F - 1 ? G 9 A I C - I @ ) I 1 3 I 4 & I H F - @ ? G 9 A I 0 - I 1 ) I 4 3 I H & I 1 F @ A I 4 - I 0 ) I H 3 I C & I 1 F A , - , .... , C ? G free . 1 G 0 G

('&#!

8ote that the output #uantities of firm 4 appear as coefficients of - in the left hand sides of the ine#ualities whereas its input #uantities appear on the right hand sides of the constraints. The optimal solution of this problem is
( ( ( ( ( ( A= ! 8 C = .&""% 8 B= 4= ? = . 8 = !&//7 &

This means that if we construct a reference firm (say 4() by combining 11.0L of the input-output bundles of firm C with the input-output bundle of firm A, then this new firm would produce ...@@ units of y! and 0.@@ units of y' using 1 units of x! and 0 units of x'. )omparison of this potential output bundle with the actual output levels of firm 4 reveals that output y! can be e"panded by a factor of ..HH/ while output y' can be increased by a factor of 9.CCC. 8ote that this new firm does not re#uire more of any input than is actually used by firm 4. Thus, it is possible to e"pand e1ery out)ut by at least the factor ..HH/. This is measured by ( in the optimal solution. 2ence, a measure of technical efficiency of firm 4 is

*+ ( 4

! = .&0'7 & !&//7

This technical efficiency measure, unfortunately, fails to reflect the full e"tent of potential increases in all of the outputs individually. !n the present case, although y! can be increased by HH./L only, y' can be e"panded all the way up by .CCL. 8or does it show any potential reductions in individual inputs that are feasible simultaneously with increases in outputs although such is not the case here. These :5 models yield radial measures of efficiency. $hile it is true that for any individual firm, say firm t, the largest output bundle with the same output-mi" as

35

( yt!, yt' that can be produced from the input bundle of firm t, is (( y(!, ( y(' , it is often possible to e"pand individual (although not all) outputs by a factor larger than (. imilarly, we may not be entirely using up all the individual components of the observed input bundle of the firm under consideration in order to produce the e"panded output bundle.
( ( ( ( 8 & 3efine Take another look at (9.@?). uppose that the optimal solution is ( !, ' ,&&&, :

( y!t =
('&#' Then

;=!

( ( ; y!; 8 y't =

;=!

( ; y'; 8

( x!t =

;=!

( ; x!; 8 x('t =

; x'; &
;=!

; ; ; ; ; ( ( ( ( y t < ( y.t , y9t ) can be produced from "; t < ( ".t , " 9t ) . 8ote that y!t y!t and y't y't& imilarly,

x!t x!t( and x't x't(& Thus,

y; y; ( = min . t , 9 t . y .t y9 t

('&##

3efine the output slack variables s!2 = y!t( -(y!t and s'2 = y't( - (y't& The input slack variable scan be similarly defined as s-! =x!t D x!t( and s-' = x't D x't(& !t may be recalled that an input-output bundle (x,y is regarded as Pareto efficient only when (a) it is not possible to increase any output without either reducing some other output or increasing some input and (b) it is not possible to reduce any input without increasing some other input or reducing some output. Thus, (xt(, yt( is Pareto efficient . -ut (xt, t(yt is not unless all output and input slacks are e#ual to %ero.

!ncluding appropriate slack variables in the constraints, we get, at the optimal solution

36

( ( ( ; y!; - y!t = s2 ! .8 ;=! :

( ( ( ; y '; - y 't = s2 ' .8 ;=!

('&#$)

-( x!t - ; x!; = s! . 8 ;=! :

-( x't - ; x '; = s ' . & ;=!

( 2( 2ere ( s2 ! , s' -( ( are the output slacks and ( s! , s'

are input slacks at the optimal solution. $henever any

output slack is strictly positive, it is possible to e"pand that particular output by the amount of the output slack even after it has been e"panded by a factor ( ( !). uppose that in a particular application we get ( = !&'0. This means that we can increase both outputs by 94L. !n this case, technical efficiency of the firm is ?.H?. 8ow suppose that s!2( = !.& This implies that we can further increase output . by .? units. 2ence, ?.H? does not fully reflect the e"tent of its inefficiency. (oreover, if any one of the input slacks is strictly positive, the implication is that above e"pansion of the output bundle can be achieved while reducing individual inputs at the same time. !n a revision of their original model )harnes, )ooper, and >hodes (./0/) introduced penalties in the ob'ective function for strictly positive input and output slacks. Their revised output-oriented model was+

37

ma"

I < I ( s. I sI 9 I s. I s 9 ) 8

I s.t. ' y. ' - s. < y. t G '<. 8

I ' y 9 ' - s 9 < y 9 t G '<. 8

' ". ' I s. < ". t G '<. 8

' " 9 ' I s9 < " 9 t G '<. I ? ( ' < . , 9 ,..., 8 ) G s. , sI ? G free . 9 , s. , s 9 '

('&#0

2ere is an infinitesimally small positive number (selected by the researcher). -y including input and output slacks in the ob'ective function, we ensure that

E > ( whenever any of slack variables is strictly positive at the optimal

( e#uals . and all the slacks are e#ual to ? at the

solution. Thus a firm will be rated as fully efficient only when

optimal solution. 7therwise, its efficiency will be less than unity even when )onsider the revised form of the input-oriented model+

( e#uals ..

38

min

E < - (s s.t.

I .

I sI 9 I s. I s 9 )

y
' '<. ' 9 '

.'

I - s. < y. t G

y
'<.

- sI 9 < y9 t G

"
' '<.

.'

I s. < ". t G

"
' '<.

9 '

I s"9 t G 9 <

?(
'

I ' < . , 9 ,..., 8 ) G s. , sI ? G free . 9 , s. , s 9

('&#"a The dual of this :5 problem is

ma" p. t y. t I p 9 t y9 t s . t . pit y. ' I p9 t y9 ' - w. t ". ' - w 9 t " 9 ' ? G ( ' < . , 9 ,..., 8 ) G w. t ". t I w 9 t " 9 t < . G p. t G p 9 t G w. t G w 9 t .
The only difference between this problem and its earlier specification is that now we have a lower bound on the shadow prices. 7n solving the primal problem we obtain the input and output bundles
;; ; t ;; t -; I; " t < " - st G yt < y I s t .

('&#"b

('&#%

The pair

;; ; ( "; t , y t ) is a 5areto efficient production plan.

2owever, using the optimal value of the ob'ective function from one of the revised models (either or

E )

39

would be problematic. )omputationally,

! E and E will not be e"actly e#ual. )onceptually, inclusion of the slacks in

the ob'ective function raises a problem of aggregation because unlike or , the input and output slacks are not unit free. Finally, the efficiency measure obtained would not be invariant to the numerical value of chosen by the analyst. At present, the overall consensus in the literature is that presence of positive slacks in the optimal solution should be interpreted as merely signifying that the efficient radial pro'ection of ( x t, y t is not 5areto efficient. -eyond that, the revised ob'ective function value should not be used to obtain a scalar measure of technical efficiency. 7ne should rather report the slacks separately along with the radial efficiency measure. !n a later chapter we will return to the #uestion of incorporating slacks in a scalar measure of efficiency. 2.* An E-a+p%e o& O tp t.Or"ented D.E.A. on SAS) +xam)le'&' Table 9.9 reports the output and input levels of a sample of @? electric utilities from Morea. The output is measured by megawatt hours of power generated. The three inputs are kilowatt hours of installed capacity, labor (man-years), and fuels (tonnes of oil e#uivalent). For the 3&A models the data were rescaled C by dividing each input and output variable by its sample mean and multiplying by .???. The appropriate :5 problem (in A ) for firm 1 is shown in &"hibit 9A. 8ote that is included in the left hand side of the ine#uality for the output. The output ine#uality is of the Agreater than or e#ual toA type. The input ine#ualities, on the other hand, are of the Aless than or e#ual toA type. 7utput and input #uantities of all firms appear on the left-hand sides of the restrictions. The right hand side includes the #uantities of the firm under evaluation (firm1 in this case). &"hibit 9- reports the optimal solution of the :5 problem specified in e"hibit 9A. The ob'ective function value (..@?.H11) shows that the #uantity of power generated by this firm can be e"panded by @?.?./L The outputoriented technical efficiency of firm firm1 is ?.01H (which is the inverse of the optimal value of ). !n the Avariable summaryA section firms 0 and 94 have AactivityA greater than ?. Thus, at the optimal solution, only % and '0 will be strictly positive. The hypothetical comparison unit for firm 1 is a firm that uses 4.919L of the input bundle of firm 0 and 1?.490L of the inputs of firm 94 to produce a similarl linear combination of the output levels of these two firms. This reference firm would produce @?.?./L more of the output compared to the actual performance of firm 1. The negative Areduced costA associated with any non-basic firm shows how the ob'ective function would be affected if it entered the basis. The rows identified as 7- N. through 7- N@ are the input slack variables. 8ote that there is a

40

positive slack (@0..@C9 units) associated with the capital input (capacity). 8o slack e"ists in thelabor or fuel inputs, however This implies that the @?..H0L increase in the output can be achieved while reducing the capacity input by these amount of the slack at the same time.& Finally, the Aconstraint summaryA section shows that the AactivityA levels for labor anf fuel are e#ual to the A>2 A value. Thus, these input constraints are binding. The dual activity associated with them are the shadow prices of these inputs. 7n the other hand, the AactivityA level for capacity is 141.4H9 while the A>2 A is .?90./9C. This results in the slack of @0..@C9 units shown earlier. 2./ S ++ar$) 5roductivity of a firm is measured by the ratio of the output produced to the input used. $e do not need to know the production technology in order to measure productivity. &fficiency, on the other hand, compares the actual output from a given input with the ma"imally producible #uantity of output. Thus, knowledge of the reference technology is critical for efficiency measurement. !n the multiple-input, multiple-output case, individual inputs and outputs need to be suitable aggregated. !n the absence of market prices, one can employ the method of 3&A, which endogenously generates Ashadow pricesA of inputs and outputs for aggregation.

Guide to the Literature:

3ebreu (./4.) addressed the #uestion of resource utili%ation at the aggregate level. ubse#uently, hephard (./4@) introduced the 3istance function as an alternative characteri%ation of the technology. Farrell (./40) defined technical and allocative efficiency as two separate components of the economic efficiency of a firm and developed the formal linear programming model for measuring technical efficiency. !ntroduced by )harnes, )ooper, and >hodes (./0H, ./H.) the method of 3&A generali%ed FarrellBs measure of technical efficiency from the single output to the multiple output case. ee Forsund and arafoglou (9??9) for an overview of the developments the literature subse#uent to FarrellBs paper that led to the introduction of the 3&A methodology. )harnes, )ooper, :ewin, and eiford (.//C) offer a brief over view of the primal and dual specifications along with a number of e"tensions of the basic ))> model. They also trace the chronology of development the literature subse#uent to the seminal ))> paper through an interesting flow chart. Ali (.//C) offers an in depth discussion of the computational aspects of 3&A in the same volume.

41

Table 2.2 Input-Output Data for Korean Electr c !t l t e"

firm 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Capacity 706.70 1284.90 1027.92 1027.92 1027.92 1027.92 2055.85 2055.85 2055.85 51.40 51.40 51.40 51.40 1669.35 308.38 308.38 256.98 256.98 1027.92 642.45 1027.92 1027.92 385.47 865.64 906.03 256.98 256.98 2878.19 2878.19 2569.81 Labor 643.39 1142.20 1749.44 1019.30 1033.76 527.72 1048.22 1055.45 1062.68 86.75 101.21 93.98 101.21 1612.09 910.87 903.64 1178.34 1185.57 1366.30 751.83 838.57 824.12 1655.46 809.66 780.74 1069.91 1033.76 1828.96 1821.73 1763.90 Fuel 648.95 1101.65 531.19 640.32 640.41 448.10 2136.09 2140.03 2140.18 111.28 91.63 91.92 92.24 1585.23 344.51 344.48 273.29 273.28 1185.60 699.30 1090.23 1090.26 362.30 559.96 554.62 221.73 228.01 3509.60 3510.85 3352.76 Output 614.66 1128.39 533.52 611.80 619.68 404.99 2276.89 2278.26 2172.23 71.72 73.40 73.88 73.83 1548.44 260.83 258.85 181.65 179.92 1076.19 586.16 959.15 958.38 278.13 660.53 673.12 246.69 252.86 3708.16 3709.64 3528.04

Source !able 1 of S.". #ar$ a%& '.(. Le)our& * +%ter%atio%al 'our%al of #ro&uctio% ,co%omic)* -ol 63* 2000* pp 59.67. /ote) 0a1 +% t2e ori3i%al )ource* capacity i) mea)ure& i% $ilo.4att 2our)* labor i% ma%.year)* fuel i% to%e) of oil e5ui6ale%t* a%& output i% me3a.4att 2our). +% t2i) !able* eac2 i%put or output 6ariable 2a) bee% )cale& by it) )ample mea% a%& multiplie& by 1000.

42

,72ibit 28. Output.orie%te& 9,8 L# #roblem for Firm 6 Firm : 1 : 2 : 3 : 4 : 5 : 6 : 7 : 8

capital 706.698 1284.90 1027.92 1027.92 1027.92 1027.92 2055.85 2055.85 labor 643.389 1142.20 1749.44 1019.30 1033.76 527.72 1048.22 1055.45 fuel 648.946 1101.65 531.19 640.32 640.41 448.10 2136.09 2140.03 output 614.660 1128.39 533.52 611.80 619.68 404.99 2276.89 2278.26 ob;ecti6e 0.000 0.00 0.00 0.00 0.00 0.00 0.00 0.00 : 9 : 10 : 11 : 12 : 13 : 14 : 15 : 16 : 17

2055.85 51.396 51.396 51.3962 51.396 1669.35 308.377 1062.68 86.749 101.207 93.9782 101.207 1612.09 910.865 2140.18 111.276 91.632 91.9232 92.244 1585.23 344.508 2172.23 71.720 73.405 73.8759 73.834 1548.44 260.830 0.00 0.000 0.000 0.0000 0.000 0.00 0.000 : 18 256.98 1185.57 273.28 179.92 0.00 : 27 256.98 1033.76 228.01 252.86 0.00 : 19 1027.92 1366.30 1185.60 1076.19 0.00 : 20 : 21 : 22 : 23 : 24

308.377 256.98 903.636 1178.34 344.483 273.29 258.852 181.65 0.000 0.00 : 25 : 26

642.452 1027.92 1027.92 385.47 865.640 906.033 256.98 751.825 838.57 824.12 1655.46 809.658 780.742 1069.91 699.303 1090.23 1090.26 362.30 559.963 554.623 221.73 586.162 959.15 958.38 278.13 660.532 673.120 246.69 0.000 0.00 0.00 0.00 0.000 0.000 0.00 : 29 2878.19 1821.73 3510.85 3709.64 0.00 : 30 2569.81 1763.90 3352.76 3528.04 0.00 p2i 0.000 0.000 0.000 .404.985 1.000 <type< => => => ?> ma7 <r2)< 1027.92 527.72 448.10 0.00 .

: 28 2878.19 1828.96 3509.60 3708.16 0.00

43

,72ibit 2( S8S Output of Output.orie%te& CC@ 9,8 Ao&el for Firm 6 !2e L# #roce&ure Solutio% Summary Ob;ecti6e -alue -ariable Summary -ariable Col /ame Statu) !ype 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 COL1 COL2 COL3 COL4 COL5 COL6 COL7 COL8 COL9 COL10 COL11 COL12 COL13 COL14 COL15 COL16 COL17 COL18 COL19 COL20 COL21 COL22 COL23 COL24 COL25 COL26 COL27 COL28 COL29 COL30 p2i <O(S1< <O(S2< <O(S3< <O(S4< #rice 8cti6ity @e&uce& Co)t .0.319551 .0.352614 .0.672975 .0.455153 .0.441649 .0.301866 0 .0.009117 .0.274152 .0.128571 .0.082295 .0.078966 .0.082727 .0.680701 .0.557802 .0.559748 .0.686168 .0.693319 .0.775215 .0.555766 .0.621882 .0.618105 .0.85466 .0.055598 0 .0.35681 .0.342487 .0.123406 .0.119913 .0.160087 0 0 .0.000398 .0.002437 .0.002469 9ual 8cti6ity 1.3018660735

(8S+C

(8S+C

(8S+C (8S+C

/O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0.052621 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0.6052773 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 0 0 /O/./,B 1 1.3018661 SL8CC 0 371.34196 SL8CC 0 0 SL8CC 0 0 S"@#L"S 0 0 Co%)trai%t Summary SDS Col @2) 8cti6ity

Co%)trai%t @o4 /ame !ype 1 2 3 4 5 <O(S1< <O(S2< <O(S3< <O(S4< <O(S5< L, L, L, B, O(',C!-,

32 1027.9237 656.58174 0 33 527.72356 527.72356 0.0003978 34 448.10376 448.10376 0.0024368 35 0 0 .0.002469 . 0 1.3018661 .

44

45

46

T# " " true onl$ n t#e " n%le-output& " n%le- nput ca"e. '#en (ult ple nput" an)*or output" are n+ol+e)& ,e (a$ nee) to u"e t#e tec#nolo%$ for a%%re%at on. 2 This approach was introduced earlier by )harnes and )ooper (./1H). 3 )onsider the profit ma"imi%ation problem ma"

)
;

y ; wi xi sub'ect to the constrain C ( x, y ) = ?. The :agrangian takes the form

i ; i

E ( x, y , ) = ) ; y ; wi xi C ( x, y )

and the first order conditions for a ma"imum are ) ; = C ; and wi = Ci .

4

$e e"amine the effect of data transformation on the 3&A efficiency score later in )hapter C.