See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/227640593

Data Envelopment Analysis - Basic Models and their Utilization

Article  in  Organizacija · March 2009

DOI: 10.2478/v10051-009-0001-6 · Source: RePEc

CITATIONS READS

77 4,835

3 authors, including:

Alenka Baggia
University of Maribor
68 PUBLICATIONS   192 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ZEROTRADE View project

Pozor(!)ni za okolje View project

All content following this page was uploaded by Alenka Baggia on 21 March 2016.

The user has requested enhancement of the downloaded file.

Organizacija, Volume 42 Research papers Number 2, March-April 2009

DOI: 10.2478/v10051-009-0001-6

Data En­ve­lop­ment Analy­sis -

Ba­sic Mo­dels and their Uti­li­za­tion
Mi­lan M. Mar­ti}1, Ma­ri­na S. No­va­ko­vi}1, Alen­ka Bag­gia2

1Uni­ver­sity of Bel­gra­de, Fa­culty of Or­ga­ni­za­tio­nal Scien­ces, Jove Ili}a 154, 11000 Bel­gra­de, Ser­bia
mar­tic.mi­lan­@fon.bg.ac.yu
2Uni­ver­sity of Ma­ri­bor, Fa­culty of Or­ga­ni­za­tio­nal Scien­ces, Ki­dri­~e­va ce­sta 55a, 4000 Kranj, Slo­ve­nia

alen­ka.bag­[email protected]

Data En­ve­lop­ment Analy­sis (DEA) is a de­ci­sion ma­king tool ba­sed on li­near pro­gram­ming for mea­su­ring the re­la­ti­ve ef­fi­ciency
of a set of com­pa­rab­le units. Be­si­des the iden­ti­fi­ca­tion of re­la­ti­vely ef­fi­cient and inef­fi­cient units, DEA iden­ti­fies the sour­ces
and le­vel of inef­fi­ciency for each of the in­puts and out­puts. This pa­per is a sur­vey of the ba­sic DEA mo­dels. A com­pa­ri­son of
DEA mo­dels is gi­ven. The ef­fect of mo­del orien­ta­tion (in­put or out­put) on the ef­fi­ciency fron­tier and the ef­fect of the con­ve­xity
re­qui­re­ments on re­turns to sca­le are exa­mi­ned. The pa­per also ex­plains how DEA mo­dels can be used to as­sess ef­fi­ciency.

Key­words: Ef­fi­ciency, DEA mo­dels, Ef­fi­ciency fron­tier.

1 In­tro­duc­tion ot­her DMU­s with the sim­ple re­stric­tion that all DMU­s lay
on or be­low the ex­tre­me fron­tier. DEA is a non-pa­ra­me­
One of the most im­por­tant prin­ci­ples in any bu­si­ness is tric met­hod as it does not re­qui­re any as­sump­tion about
the prin­ci­ple of ef­fi­ciency; whe­re the best pos­sib­le eco­no­ func­tio­nal form (e.g. a re­gres­sion equa­tion, a pro­duc­tion
mic ef­fects (out­puts) are at­tai­ned with as litt­le eco­no­mic func­tion, etc.). It is a met­ho­do­logy di­rec­ted at the fron­tier
sa­cri­fi­ces as pos­sib­le (in­puts). Ef­fi­ciency can be de­fi­ned as rat­her than at cen­tral ten­den­cies. Whi­le sta­ti­sti­cal pro­ce­
the de­mand that the de­si­red goals are ac­hie­ved with the du­res are ba­sed on cen­tral ten­den­cies, DEA is a pro­cess
mi­ni­mum use of the avai­lab­le re­sour­ces. In or­der to as­sess of ex­tre­mi­ties. DEA analy­zes each DMU se­pa­ra­tely and
the re­la­ti­ve ef­fi­ciency of a bu­si­ness unit, it is ne­ces­sary cal­cu­la­tes a ma­xi­mum per­for­man­ce mea­su­re for each
to con­si­der the con­di­tions and ope­ra­tion re­sults of ot­her unit. DEA has be­co­me one of the most po­pu­lar fields in
units of the same kind and to de­ter­mi­ne the real stan­ding ope­ra­tions re­search, with ap­pli­ca­tions in­vol­ving a wide
of the re­sults of such a com­pa­ri­son. ran­ge of con­text (Tha­nas­sou­lis, 2001).
In a sim­ple case whe­re units have a sin­gle out­put and DEA is one of the most po­pu­lar fields in ope­ra­tions
a sin­gle in­put, ef­fi­ciency is de­fi­ned as their ra­tio. Ho­we­ver, re­search (Emrouz­ne­jad et al., to ap­pear; Sei­ford, 1997).
typi­cal or­ga­ni­za­tio­nal units have mul­ti­ple and in­com­men­ Sin­ce the se­mi­nal work of Char­nes, Coo­per and Rho­des
su­ra­te in­puts and out­puts. Data En­ve­lop­ment Analy­sis (1978) and sin­ce 1995, the­re was li­te­rally “ex­po­nen­tial”
was in­tro­du­ced by Char­nes, Coo­per and Rho­des (1978) to growth in the num­ber of pub­li­ca­tions. Bet­ween 1995
as­sess the re­la­ti­ve ef­fi­ciency of or­ga­ni­za­tio­nal units with and 2003, the num­ber of re­le­vant pub­li­ca­tions sta­bi­li­zed
mul­ti­ple in­puts to pro­du­ce mul­ti­ple out­puts. The aut­hors at about 225 per year. Ho­we­ver, in the last four years
of DEA de­fi­ned the ef­fi­ciency of the unit un­der eva­lua­ (2004-2007), the num­ber in­crea­sed to ap­pro­xi­ma­tely 360
tion as the ra­tio of the sum of its weight out­puts to the per year. Jour­nals such as the Eu­ro­pean Jour­nal of Ope­ra­
sum of its weight in­puts. tio­nal Re­search, Jour­nal of Pro­duc­ti­vity Analy­sis, and the
Re­cently, the Data En­ve­lop­ment Analy­sis met­hod Jour­nal of the Ope­ra­tio­nal Re­search So­ciety are the most
is be­co­ming po­pu­lar for as­ses­sing the re­la­ti­ve ef­fi­ciency uti­li­zed.
of bu­si­ness en­ti­ties. DEA is a tech­ni­que of mat­he­ma­ti­cal The pa­pers pre­sen­ted show am­ple pos­si­bi­li­ties for
pro­gram­ming that enab­les the de­ter­mi­na­tion of a unit’s using the DEA for the eva­lua­tion of the per­for­man­ce of
ef­fi­ciency ba­sed on its in­puts and out­puts, and com­pa­res bank branc­hes, schools, uni­ver­sity de­part­ments, far­ming
it to ot­her units in­vol­ved in the analy­sis. The DEA can be es­ta­tes, hos­pi­tals and so­cial in­sti­tu­tions, mi­li­tary ser­vi­ces,
des­cri­bed as data-orien­ted as it ef­fects per­for­man­ce eva­ en­ti­re eco­no­mic systems (re­gions) and ot­her things. DEA
lua­tions and ot­her in­fe­ren­ces di­rectly from the ob­ser­ved is a met­ho­do­logy of se­ve­ral dif­fe­rent in­te­rac­ti­ve ap­proac­
data and with mi­ni­mal as­sump­tions. The ef­fi­ciency of a hes and mo­dels used for the as­ses­sment of the re­la­ti­ve ef­fi­
De­ci­sion Ma­king Unit (DMU) is mea­su­red re­la­ti­ve to all ciency of DMU and for the as­ses­sment of the ef­fi­ciency

37
Unauthenticated
Download Date | 12/20/15 7:32 AM
Organizacija, Volume 42 Research papers Number 2, March-April 2009

fron­tier. It sup­plies im­por­tant in­for­ma­tion for ma­na­ging im­por­tan­ce of each in­put and out­put and are de­ter­mi­ned
the ope­ra­tions of ef­fi­cient and inef­fi­cient units. This pa­per in the mo­del so that each DMU is ef­fi­cient as much as pos­
is a sur­vey of the ba­sic DEA mo­dels. Some ways in which sib­le. Gi­ven that the con­di­tion (2) is true for every DMU,
the­se mo­dels can be used are also gi­ven. it means that each of them lies on the ef­fi­ciency fron­tier
or be­yond it. If Max hk = hk* = 1, it means that ef­fi­ciency
is being ac­hie­ved, so we can tell that DMUk is ef­fi­cient.
2 DEA Mo­dels Ef­fi­ciency is not ac­hie­ved for hk* < 1 and DMUk is not
ef­fi­cient in that case. DMUk is to be con­si­de­red re­la­ti­vely
DEA met­ho­do­logy, ori­gi­nally pro­po­sed in (Char­nes et al., inef­fi­cient, if it is pos­sib­le to ex­pand any of its out­puts
1978), is used to as­sess the re­la­ti­ve ef­fi­ciency of a num­ber wit­hout re­du­cing any of its in­puts, and wit­hout re­du­cing
of en­ti­ties using a com­mon set of in­com­men­su­ra­te in­puts any ot­her out­put (out­put orien­ta­tion), or if it is pos­sib­le to
to ge­ne­ra­te a com­mon set of in­com­men­su­ra­te out­puts. re­du­ce any of its in­puts wit­hout re­du­cing any out­put and
The ori­gi­nal mo­ti­va­tion for DEA was to com­pa­re the pro­ wit­hout ex­pan­ding some ot­her in­puts (in­put orien­ta­tion).
duc­ti­ve ef­fi­ciency of si­mi­lar or­ga­ni­za­tions, re­fer­red to as Prob­lem (1) - (4) is non­li­near, non­con­vex, with a
DMU­s. The prob­lem of as­ses­sing ef­fi­ciency is for­mu­la­ted li­near and frac­tio­nal ob­jec­ti­ve func­tion and li­near and
as a task of frac­tio­nal pro­gram­ming, but the ap­pli­ca­tion frac­tio­nal con­straints. Using a sim­ple trans­for­ma­tion de­ve­
pro­ce­du­re for DEA con­sists of sol­ving li­near pro­gram­ lo­ped by Char­nes and Coo­per (1962), the abo­ve CCR
ming (LP) tasks for each of the units un­der eva­lua­tion. ra­tio mo­del can be re­du­ced to the LP form (the Pri­mal
Let xij - de­no­te the ob­ser­ved mag­ni­tu­de of i - type CCR mo­del) so that the LP met­hods can be ap­plied. In
in­put for en­tity j ( xij > 0, i = 1, 2, ..., m, j = 1, 2, ..., n) and this mo­del, the de­no­mi­na­tor has been set equal to 1 and
yrj - the ob­ser­ved mag­ni­tu­de of r-type out­put for en­tity j the nu­me­ra­tor is being ma­xi­mi­zed. The in­put orien­ted
(yrj > 0, r = 1, 2, ..., s, j = 1, 2, ..., n). Then, the Char­nes-Coo­ CCR pri­mal mo­del is:
per-Rho­des (CCR) mo­del is for­mu­la­ted in the fol­lo­wing
form for the se­lec­ted en­tity k: MODEL (M2)

MODEL (M1) (5)

s

∑u r
y rk
Ma­xi­mi­ze hk = r =1
m
(1)
(6)
∑v
i =1
i
x ik

(7)
Sub­ject to
s
(8)
∑u r
y rj
r =1 ≤1, j =1,2,..., j ,..., n (2) (9)
m k
∑v x i ij The mat­he­ma­ti­cal mo­del gi­ven abo­ve is li­near and
i=1
can be sol­ved using any of the fa­mi­liar pro­grams pac­ka­ges
ur ≥ε , r =1,2,..., s (3) for LP. Ho­we­ver, in prac­ti­ce, it is of­ten sol­ved dual task for
prob­lem (5) - (9), which is:
v i ≥ε , i =1,2,..., m (4)
MODEL (M3)
Whe­re:
n vi is the weights to be de­ter­mi­ned for in­put i; (10)
n m is the num­ber of in­puts;
n ur is the weights to be de­ter­mi­ned for out­put r;
n s is the num­ber of out­puts;
n h is the re­la­ti­ve ef­fi­ciency of DMUk; (11)
k
n n is the num­ber of en­ti­ties;
n ε is a small po­si­ti­ve va­lue.
The re­la­ti­ve ef­fi­ciency hk, of one de­ci­sion-ma­king (12)
unit k, is de­fi­ned as a ra­tio of the weigh­ted sums of their
out­puts (vir­tual out­put) and the weigh­ted sums of their (13)
in­puts (vir­tual in­put). As for the de­ci­sion-ma­king unit k,
for which a ma­xi­mum in ob­jec­ti­ve func­tion (1) is sought, The ba­sic idea be­hind DEA is best con­ve­yed in the
the con­di­tion (2) is true, mea­ning that it is ob­vi­ously 0 < dual CCR mo­del (M3), which is easy to sol­ve be­cau­se of
hk ≤ 1, for each DMUk. The weights vi and ur show the its cal­cu­la­ting size. The dual mo­del for a gi­ven unit using

38
Unauthenticated
Download Date | 12/20/15 7:32 AM
Organizacija, Volume 42 Research papers Number 2, March-April 2009

in­put and out­put va­lues of ot­her units tries to con­struct a bi­na­tion of DMU­s, which are in the set (tho­se that have
hypot­he­ti­cal com­po­si­te unit out of the exi­sting units. If it po­si­ti­ve va­lue for λ in the op­ti­mal so­lu­tion). The DMU
is pos­sib­le, the gi­ven unit is inef­fi­cient, ot­her­wi­se it is ef­fi­ ope­ra­tes un­der va­riab­le re­turns to sca­le if it is sus­pec­ted
cient and lies at the ef­fi­ciency fron­tier. The ef­fi­ciency fron­ that an in­crea­se in in­puts does not re­sult in a pro­por­tio­nal
tier is a set of seg­ments in­ter­con­nec­ting all the ef­fi­cient chan­ge in the out­puts. The con­ve­xity con­straint en­su­res
DMU­s and it acts as an en­ve­lo­pe for inef­fi­cient units. An that the com­po­si­te unit is of si­mi­lar sca­le size as the unit
inef­fi­cient unit can be en­ve­lo­ped be­low (in­put-orien­ted being mea­su­red. The BCC mo­del yields a mea­su­re of
mo­del) or abo­ve (out­put-orien­ted mo­del). pure tech­ni­cal ef­fi­ciency that ig­no­res the im­pact of the
Be­cau­se the prob­lems des­cri­bed by mo­dels (M2) and sca­le size by only com­pa­ring a DMU to a unit of si­mi­lar
(M3) are as­so­cia­ted and be­cau­se of the dua­lity theo­rem sca­le. Of­ten, small units are qua­li­ta­ti­vely dif­fe­rent from
in li­near pro­gram­ming, DMUk is ef­fi­cient if and only lar­ge units and a com­pa­ri­son bet­ween the two may di­stort
if con­di­tions for op­ti­mal so­lu­tion (λ*, s+*, s-*, Zk*) are mea­su­re­ments of com­pa­ra­ti­ve ef­fi­ciency. The mea­su­red
ac­com­plis­hed for the prob­lem (10)-(13): ef­fi­ciency is al­ways at least equal to the one gi­ven by the
CCR mo­del. The en­ve­lop­ment sur­fa­ce ob­tai­ned from the
Zk* = 1 (14) BCC mo­del re­sults in a con­vex hull.
s+* = s-* = 0 in all al­ter­na­te op­ti­ma (15) The DEA mo­del can be in­put or out­put orien­ted.
The in­put orien­ted mo­del con­tracts the in­puts as far as
Using the op­ti­mal so­lu­tion (λ*, s+*, s-*, Zk*) of the pos­sib­le whi­le con­trol­ling the out­puts. In an in­put orien­
prob­lem (10) - (13), it can be de­ter­mi­ned:
ted mo­del, an inef­fi­cient unit is made ef­fi­cient through
the pro­por­tio­nal re­duc­tion of its in­puts, whi­le its out­puts
X’’k = Zk*Xk - s-* (16)
pro­por­tions are held con­stant. The out­put orien­ted mo­del
Y’’k = Yk + s+* (17) ex­pands the out­puts as far as pos­sib­le whi­le con­trol­ling
the in­puts. In an out­put orien­ted mo­del, an inef­fi­cient unit
It can be shown that af­ter CCR pro­jec­tion (16), (17),
DMUk with al­te­red in­puts X’’k and out­puts Y’’k be­co­mes is made ef­fi­cient through the pro­por­tio­nal in­crea­se of its
ef­fi­cient. The dif­fe­ren­ce ∆Xk = Xk - X''k and ∆Yk = Y''k out­puts, whi­le the in­puts' pro­por­tions re­main unc­han­ged.
- Yk res­pec­ti­vely shows the es­ti­ma­ted amount of in­put An inef­fi­cient DMU can be made more ef­fi­cient by pro­
and out­put inef­fi­ciency. Thus it can be seen for inef­fi­cient jec­tion onto the fron­tier. Mo­del orien­ta­tion de­ter­mi­nes
DMUk, how to chan­ge its in­puts and out­puts, so it would the di­rec­tion of the pro­jec­tion for inef­fi­cient DMU­s. In an
be­co­me ef­fi­cient. We should emp­ha­si­ze that, for each in­put orien­ta­tion, one im­pro­ves ef­fi­ciency through the pro­
DMUj (j = 1, 2, ..., n) ta­ken as DMUk, an ap­pro­pria­te por­tio­nal re­duc­tion of in­puts, whe­reas an out­put orien­ta­
li­near pro­gram­ming prob­lem is sol­ved (10) - (13). Hen­ce, tion re­qui­res pro­por­tio­nal aug­men­ta­tion of the out­puts.
we should sol­ve n li­near pro­gram­ming tasks with the form The in­put and out­put mea­su­re­ments are al­ways the
(10) - (13), with (s+m+1) va­riab­les and (s+m) con­straints same in the CCR mo­del, but fre­quently dif­fer in the BCC
per task. mo­del. Thus, if we are using the CCR mo­del, we can sol­ve
The CCR mo­dels (dual and pri­mal) with in­put orien­ one mo­del and give eit­her in­ter­pre­ta­tion. If we sol­ve the
ta­tion are still the most wi­dely known and used DEA BCC in­put mo­del, we can only give an in­put in­ter­pre­ta­
mo­dels des­pi­te the nu­me­rous mo­di­fied mo­dels that have tion and we must sol­ve the BCC out­put mo­del for an out­
ap­pea­red. The CCR mo­dels as­su­me con­stant re­turns to put in­ter­pre­ta­tion. Anot­her dif­fe­ren­ce bet­ween the BCC
sca­le. DMU ope­ra­tes un­der con­stant re­turns to sca­le if an and CCR mo­dels lies in the sca­lar trans­for­ma­tions of all
in­crea­se in the in­puts re­sults in a pro­por­tio­na­te in­crea­se data for a gi­ven DMU. The ef­fi­ciency mea­su­re in the CCR
in the out­put le­vels. The­se mo­dels cal­cu­la­te an ove­rall mo­del is unc­han­ged by sca­lar trans­for­ma­tions, sin­ce the
ef­fi­ciency in which both its pure tech­ni­cal ef­fi­ciency and ef­fi­ciency ra­tio of the sca­led DMU is unc­han­ged. On the
its sca­le ef­fi­ciency are ag­gre­ga­ted into a sin­gle va­lue. The ot­her hand, the sca­lar trans­for­ma­tions of a gi­ven DMU
en­ve­lop­ment sur­fa­ce ob­tai­ned from the CCR mo­del has chan­ge the sca­lar size and could ea­sily af­fect the ef­fi­ciency
the sha­pe of a con­vex cone. The ef­fi­cient DMU­s would lie mea­su­re­ments from the BCC mo­del.
on top of the struc­tu­re, whi­le the inef­fi­cient ones would Nu­me­rous ex­ten­sions of the ba­sic DEA mo­dels are
be co­ve­red un­der the cone. In a sin­gle in­put and out­put pre­sen­ted in the li­te­ra­tu­re (Char­nes et al., 1995; Coo­per et
case, the ef­fi­ciency fron­tier is re­du­ced to a straight line. al., 2005; Tha­nas­sou­lis, 2001). Some of the ex­ten­sions are:
The CCR mo­del yields the same ef­fi­cien­cies re­gard­less of n con­straints are pla­ced on the weights for par­ti­cu­lar
whet­her it is in­put- or out­put-orien­ted. in­puts and out­puts (Tha­nas­sou­lis, 2001),
The most im­por­tant ex­ten­sion of the ori­gi­nal CCR n con­straints are pla­ced on the amount of par­ti­cu­lar vir­
mo­dels is gi­ven in Ban­ker et al. (1984) whe­re an ad­di­tio­nal tual in­puts and out­puts (Tha­nas­sou­lis, 2001),
con­straint was in­tro­du­ced in mo­del (M3): n in­puts and out­puts that can­not be con­trol­led are
n brought into analy­sis (Ban­ker and Mo­rey, 1986),
∑λ j
=1 (18) n ca­te­go­ri­cal va­riab­les are brought into the mo­del (Ban­
j =1 ker and Mo­rey, 1986),
This con­straint enab­les va­riab­le re­turns to sca­le and n mo­dels for ran­king re­la­ti­vely ef­fi­cient DMU­s are
pro­vi­des that the re­fe­ren­ce set is for­med as a con­vex com­ de­ve­lo­ped (An­der­sen and Pe­ter­sen, 1993).

39
Unauthenticated
Download Date | 12/20/15 7:32 AM
Organizacija, Volume 42 Research papers Number 2, March-April 2009

3 DEA Mo­del Uti­li­za­tion be­co­me re­la­ti­vely ef­fi­cient. For each inef­fi­cient DMU
(one that lies be­low the fron­tier), DEA iden­ti­fies the sour­
In the first part of this sec­tion, we dis­cuss how DEA ces and le­vel of inef­fi­ciency for each in­put and out­put. The
mo­dels can be used to as­sess DMU­s. A key sta­ge in a le­vel of inef­fi­ciency is de­ter­mi­ned by com­pa­ri­son with a
DEA as­ses­sment is the iden­ti­fi­ca­tion of the in­put/out­ sin­gle re­fe­rent DMU or a con­vex com­bi­na­tion of ot­her
put va­riab­les per­tai­ning to the units being as­ses­sed, see re­le­vant DMU­s lo­ca­ted on the ef­fi­cient fron­tier that uti­li­
(Bous­so­fia­ne et al., 1991). Sin­ce DEA is used to eva­lua­te ze the same le­vel of in­puts and pro­du­ce the same or hig­her
per­for­man­ces by di­rectly con­si­de­ring in­put and out­put le­vel of out­puts. In the pre­vi­ous sec­tion, we have seen that
data, the re­sults will de­pend on the in­put/out­put choi­ce we can reach the le­vel of inef­fi­ciency using the op­ti­mal
for the analy­sis and the num­ber and ho­mo­ge­neity of the so­lu­tion of mo­del (M3) and re­la­tions (16) and (17). We can
DMU­s to be eva­lua­ted. In this sta­ge it is im­por­tant to ar­ri­ve at si­mi­lar in­for­ma­tion through sen­si­ti­vity analy­sis of
con­sult the peo­ple wor­king in the units that are to be eva­ the op­ti­mal so­lu­tion in mo­del (M1). The­se re­sults are very
lua­ted, so ma­jor in­puts and out­puts can be iden­ti­fied pro­ im­por­tant to ma­na­gers, be­cau­se they in­di­ca­te the sour­ces
perly. It is im­por­tant to en­ve­lop all the im­por­tant in­puts of inef­fi­ciency for re­la­ti­vely inef­fi­cient DMU­s.
in the analy­sis, na­mely all the re­sour­ces used, and all the Im­pro­ve­ment of ef­fi­ciency in not only inef­fi­cient but
im­por­tant out­puts (the pro­ducts and ser­vi­ces pro­du­ced). also the ef­fi­cient units can be at­tai­ned by iden­tif­ying an
A lar­ge num­ber of in­puts and out­puts com­pa­red to the ef­fi­cient ope­ra­ting prac­ti­ce. It can usually be found in the
num­ber of units to be eva­lua­ted may re­du­ce the dis­cri­mi­ re­la­ti­vely ef­fi­cient units. Ho­we­ver, among the re­la­ti­vely
na­ting ten­dency of the met­hod. The lar­ger the num­ber of ef­fi­cient units, some are bet­ter than ot­hers at set­ting a
in­puts and out­puts com­pa­red to the num­ber of units to good exam­ple. A need to di­stin­guish the re­la­ti­vely ef­fi­
be eva­lua­ted, the grea­ter the chan­ces that the units will cient units and find a good ope­ra­ting prac­ti­ce, emer­ges
al­lo­ca­te ap­pro­pria­te weights to a sin­gle sub­set of in­puts from the es­sen­ce of a DEA mo­del that al­lows a unit to
and out­puts that will make them ap­pear ef­fi­cient. In or­der se­lect the weights that will show it as ha­ving ma­xi­mum
to pre­ser­ve the dis­cri­mi­na­ting po­wer of the met­hod, the ef­fi­ciency. In this way, the units may ap­pear ef­fi­cient
num­ber of the units to be eva­lua­ted should be much lar­ be­cau­se all very small in­put sub­sets will be ig­no­red wit­hin
ger than the num­ber of in­puts and out­puts. Some aut­hors their choi­ce of weights. Mo­reo­ver, the in­puts and out­puts
sug­gest from ex­pe­rien­ce that the num­ber of units in the as­sig­ned lar­ger weights could be gi­ven a se­con­dary im­por­
DMU should ex­ceed the num­ber of in­puts and out­puts by tan­ce whi­le tho­se that are ig­no­red could be as­so­cia­ted
at least twi­ce. Bous­so­fia­ne et al., (1991) pro­po­se te­sting with the units' main func­tions.
the cor­re­la­tion bet­ween the in­puts and out­puts, as one To di­stin­guish the re­la­ti­vely ef­fi­cient DMU­s, Bous­so­
pos­sib­le way to re­du­ce their num­ber. If a pair of in­puts fia­ne et al. (1991) sug­ge­sted the fol­lo­wing met­hods (or a
is po­si­ti­vely cor­re­la­ted then one may be omit­ted wit­hout com­bi­na­tion of the­se):
any im­pli­ca­tions on the ef­fi­ciency to be ra­ted. The same n cross ef­fi­ciency ma­trix,
ap­plies to out­puts. The avai­la­bi­lity of data may also af­fect n the di­stri­bu­tion of vir­tual in­puts and out­puts,
the choi­ce of in­puts and out­puts in prac­ti­ce. If the data on n weight re­stric­tion,
an in­put or out­put is not avai­lab­le then the pos­si­bi­lity of n the fre­quency by which an ef­fi­cient unit ap­pears in
using a sub­sti­tu­te for which such data is eit­her avai­lab­le or the peer groups.
can be ob­tai­ned re­la­ti­vely ea­sily should be chec­ked. Ba­sic DEA mo­dels eva­lua­te the re­la­ti­ve ef­fi­ciency of
DEA is a met­ho­do­logy with se­ve­ral dif­fe­rent in­te­rac­ DMU­s but do not al­low any ran­king of the ef­fi­cient units
ti­ve ap­proac­hes and mo­dels used for the as­ses­sment of them­sel­ves. This is the main weak­ness of the ba­sic DEA
the re­la­ti­ve ef­fi­ciency of a DMU and for the as­ses­sment mo­dels. One way to rank ef­fi­cient DMU­s is to mo­dify
of the ef­fi­ciency fron­tier. It sup­plies im­por­tant in­for­ma­ ba­sic DEA mo­dels. One of the­se has been for­mu­la­ted
tion for ma­na­ging the ope­ra­tions of both ef­fi­cient and by An­der­sen and Pe­ter­sen (1993). The ba­sic idea is to
inef­fi­cient units. For each inef­fi­cient unit, DEA iden­ti­fies com­pa­re the unit un­der eva­lua­tion with a li­near com­bi­na­
a set of re­la­ti­vely ef­fi­cient units, thus ma­king a peer group tion of all the ot­her units in the sam­ple, i.e., the ob­ser­ved
for the inef­fi­cient unit. The peer set for an inef­fi­cient unit DMU is exc­lu­ded from the peer group. Ef­fi­cient units
con­sti­tu­tes the units with the same op­ti­mum weights as may pro­por­tio­nally in­crea­se the va­lue of the in­put vec­tor
the inef­fi­cient unit, but with a re­la­ti­ve ef­fi­ciency ra­ting of whi­le pre­ser­ving ef­fi­ciency. The units ob­tain an ef­fi­ciency
1. Such peer units are iden­ti­fied fairly ea­sily by the fact sco­re abo­ve 1. This sco­re ref­lects the ra­dial di­stan­ce of
that they all have a po­si­ti­ve va­lue for λ in the op­ti­mum the DMU un­der eva­lua­tion from the pro­duc­tion fron­tier,
so­lu­tion to (M3) for an inef­fi­cient unit. The iden­ti­fi­ca­tion es­ti­ma­ted with that DMU exc­lu­ded from the sam­ple. This
of peer groups should be very use­ful in prac­ti­ce. Peer ap­proach pro­vi­des an ef­fi­ciency ra­ting for ef­fi­cient units
units can be used to high­light the weak as­pects of the that is si­mi­lar to the ra­ting of the inef­fi­cient units abo­ve.
per­for­man­ce of the cor­res­pon­ding inef­fi­cient unit. The The se­cond part of this sec­tion re­la­tes to two of the
in­put/out­put le­vels of a peer unit can also so­me­ti­mes pro­ most im­por­tant ap­pli­ca­tions of DEA in Ser­bia. DEA is
ve use­ful tar­get le­vels for the inef­fi­cient unit. used very suc­cess­fully for the com­pa­ra­ti­ve analy­sis and
From the so­lu­tion of any DEA mo­del, we can get in­for­ ran­king of 30 re­gions in Ser­bia (Mar­ti} and Savi}, 2001)
ma­tion on how much a re­la­ti­vely inef­fi­cient unit should and in as­ses­sing the re­la­ti­ve ef­fi­ciency of 20 in­vest­ment
re­du­ce their in­puts or in­crea­se their out­puts in or­der to pro­grams in agri­cul­tu­re (Mar­ti} et al., 1996).

40
Unauthenticated
Download Date | 12/20/15 7:32 AM
Organizacija, Volume 42 Research papers Number 2, March-April 2009

In Mar­ti} and Savi} (2001), DEA is used to es­ti­ma­te Ni­sa­va, the City of Bel­gra­de and the re­gion of Jab­la­ni­ca
how well re­gions in Ser­bia uti­li­ze their re­sour­ces. Ba­sed uti­li­ze their re­sour­ces most ef­fi­ciently.
on the data for 4 in­puts (Arab­le area, Ac­ti­ve fi­xed as­sets, In Mar­ti} et al. (1996), it is shown how DEA could
Con­sump­tion of elec­tri­city and Po­pu­la­tion) and 4 out­puts be used to pro­vi­de in­for­ma­tion con­cer­ning ef­fi­cient and
(Gross do­me­stic pro­duct, To­tal num­ber of physi­cians, inef­fi­cient in­vest­ments, and also to rank the ef­fi­cient
To­tal num­ber of pu­pils in pri­mary school and To­tal num­ in­vest­ments and in­di­ca­te how to im­pro­ve the ef­fi­ciency
ber em­plo­yed in the so­cial sec­tor), an out­put-orien­ted of the­se inef­fi­cient in­vest­ments. An exam­ple il­lu­stra­tes
CCR DEA mo­del is ap­plied and it ap­pears that 17 out the ap­pli­ca­tion of the ap­proach pro­po­sed to as­sess the
of the 30 re­gions are ef­fi­cient. 5 out of 7 re­gions from Voj­ re­la­ti­ve ef­fi­ciency of 20 in­vest­ment pro­grams in agri­cul­tu­
vo­di­na and 12 out of 18 re­gions from cen­tral Ser­bia are re and their ran­king. The agri­cul­tu­re bank ma­na­ge­ment
ef­fi­cient, whi­le all the re­gions from Ko­so­vo and Me­to­hia of­ten fa­ces a prob­lem in mea­su­ring the ef­fi­ciency of new
are inef­fi­cient. Li­near Dis­cri­mi­na­tion Analy­sis (LDA) is in­vest­ments. Nor­mally, the go­vern­ment is ef­fec­tua­ting its
also ap­plied to de­ter­mi­ne the re­gions. The com­pa­ri­son bet­ in­vest­ments in the agri­cul­tu­ral sec­tor through the ban­king
ween the DEA and the LDA re­sults in­di­ca­ted that LDA system, by no­mi­na­ting one or se­ve­ral banks to hand­le the
could be a use­ful tool for chec­king the DEA re­sults. in­vest­ment loans to va­ri­ous agri­cul­tu­ral firms com­pe­ting
The aut­hors analy­zed which chan­ges in in­puts and for funds. The banks are fa­cing the tre­men­dous prob­lem
out­puts inef­fi­cient re­gions should make in or­der to be­co­ of de­ci­ding, un­der the con­straints of li­mi­ted funds and
me ef­fi­cient. Ho­we­ver, it was shown that goals ob­tai­ned the wish to ma­xi­mi­ze the eco­no­mic re­turn, which firms to
using the ba­sic CCR mo­dels and DEA mo­dels with exo­ se­lect for their in­vest­ment port­fo­lio. Furt­her, banks have
ge­nously fi­xed in­puts are not qui­te rea­li­stic. It is dif­fi­cult to take care of the uni­form de­ve­lop­ment of all re­gions,
to ex­plain that some re­gions have goals to re­du­ce the size unem­ploy­ment le­vels, eco­lo­gi­cal fac­tors, etc.
of the po­pu­la­tion or arab­le area. For inef­fi­cient re­gions, The first step in as­ses­sing the re­la­ti­ve ef­fi­ciency of a
we ex­pe­ri­men­tally de­ter­mi­ned rea­li­stic goals wit­hout set of in­vest­ment pro­grams in the agri­cul­tu­ral sec­tor is the
any re­duc­tion of its in­puts. The­se goals could be ac­hie­ved de­fi­ni­tion of the sets of com­mon in­puts and com­mon out­
more ea­sily but are not in the pro­duc­tion pos­si­bi­lity set. puts. By re­vie­wing the stan­dard prac­ti­ce of eva­lua­ting the
In or­der to rank the 17 ef­fi­cient re­gions, an out­put- in­vest­ment pro­grams that is used by the in­vest­ment banks,
orien­ted ver­sion of An­der­sen-Pe­ter­sen’s DEA mo­del and 4 in­puts (Re­qui­red loan amount, La­bour costs, Pro­duc­tion
a cross ef­fi­ciency ma­trix are used. The same or si­mi­lar costs and Energy con­sump­tion) and 3 out­puts (Ex­pec­ted
rank was ob­tai­ned for 12 re­gions. A com­pa­ri­son of the va­lue of do­me­stic sa­les, Ex­pec­ted va­lue of ex­ports, So­cial
ob­tai­ned ranks shows that the re­gions were ran­ked more ju­sti­fia­bi­lity and en­vi­ron­men­tal ac­cep­ta­bi­lity) have been
rea­li­sti­cally with the cross ef­fi­ciency ma­trix. Ne­vert­he­ in­ve­sti­ga­ted. The de­fi­ni­tions and cor­res­pon­ding units
less, the same 8 re­gions are ran­ked in the top ten in both of mea­su­re are ob­vi­ous for all in­puts and the first two
ap­proac­hes. The ranks ob­tai­ned sho­wed that the re­gion of out­puts li­sted. Ho­we­ver, so­cial ju­sti­fia­bi­lity and en­vi­ron­

Tab­le 1: In­put and out­put va­lues for 20 in­vest­ment pro­grams com­pe­ting for loans

41
Unauthenticated
Download Date | 12/20/15 7:32 AM
Organizacija, Volume 42 Research papers Number 2, March-April 2009

men­tal ac­cep­ta­bi­lity de­ser­ve furt­her ex­pla­na­tion. Na­mely, tion of each in­put and out­put to the re­la­ti­ve ef­fi­ciency of
any in­vest­ment pro­gram must be so­cially ju­sti­fiab­le and the sub­ject unit in com­pa­ri­son with the ot­her units. The
en­vi­ron­men­tally ac­cep­tab­le. The term so­cially ju­sti­fiab­le op­ti­mal weights for each of the 4 in­puts and 3 out­puts
en­com­pas­ses a num­ber of fac­tors such as the unem­ploy­ of the ef­fi­cient units are ob­tai­ned as the va­lues of dual
ment le­vel, re­gio­nal le­vel of de­ve­lop­ment and si­mi­lar. va­riab­les in (M2). Ba­sed on that, the va­lues of the vir­tual
Each of the in­vest­ment pro­grams pro­po­sed has been as­sig­ in­puts and out­puts are com­pu­ted in­di­ca­ting how each of
ned a so­cial ju­sti­fia­bi­lity and en­vi­ron­men­tal ac­cep­ta­bi­lity the ef­fi­cient units at­tai­ned their ma­xi­mum ef­fi­ciency. The
le­vel using a sca­le from 0 to 100. Thus, an in­vest­ment pro­ in­vest­ment pro­grams P3 and P4 have the ma­xi­mum ef­fi­
gram that is fully so­cially ju­sti­fiab­le and en­vi­ron­men­tally ciency due to the sa­tis­fac­tory va­lue of mix y2/x4, be­cau­se
ac­cep­tab­le has an as­sig­ned le­vel of 100. Con­ver­sely, an the lar­gest pro­por­tion in the vir­tual out­put co­mes from y2
in­vest­ment pro­gram that can­not be so­cially and/or en­vi­ and from x4 in the vir­tual in­put. The in­vest­ment pro­gram
ron­men­tally ju­sti­fied at all, has a le­vel of 0. P5 is re­la­ti­vely ef­fi­cient, mostly due to out­put y2 be­cau­se
An il­lu­stra­ti­ve exam­ple, ba­sed on the data for 20 of its best mix y2/x1 and the sa­tis­fac­tory va­lue of mix y2/
in­vest­ment pro­grams com­pe­ting for loans, is analy­zed x4. A si­mi­lar rea­so­ning ex­plains the re­la­ti­ve ef­fi­ciency of
here. Their in­put and out­put va­lues are pre­sen­ted in Tab­le the ot­her ef­fi­cient units.
1. The 20 li­near pro­grams of type (10) - (13) are for­mu­la­ Furt­her, it is pos­sib­le to analy­ze the pos­sib­le pro­por­
ted, con­si­sting of 28 va­riab­les (m+n+s+1=28) and 7 con­ tio­nal chan­ge of in­puts and/or out­puts nee­ded to make
straints (m + s =7). The re­sults were ob­tai­ned using the E- an inef­fi­cient in­vest­ment pro­gram into an ef­fi­cient one,
DEA pro­gram­ming pac­ka­ge (see Mar­ti} & Savi}, 2001)), es­tab­lis­hing the sen­si­ti­vity of the ef­fi­ciency to chan­ges
de­ve­lo­ped at the Fa­culty of Or­ga­ni­za­tio­nal Scien­ces from in in­puts and/or out­puts. By using the op­ti­mal so­lu­tion
Bel­gra­de and are also pre­sen­ted in Tab­le 1. of mo­del (M2) and re­la­tion (16) and (17), set tar­gets can
The so­lu­tion ob­tai­ned al­lows the clas­si­fi­ca­tion of the be de­ter­mi­ned for a re­la­ti­vely inef­fi­cient unit to gui­de
in­vest­ment pro­grams into a set of ef­fi­cient pro­grams with them to­wards im­pro­ved per­for­man­ce. This set of tar­gets is
a re­la­ti­ve ef­fi­ciency of 1 and a set of inef­fi­cient pro­grams in­put-orien­ted as the main chan­ges are to the in­put le­vels.
with a re­la­ti­ve ef­fi­ciency less than 1. The num­ber of re­la­ The ne­ces­sary va­lue of the in­puts and out­puts that make
ti­vely ef­fi­cient in­vest­ment pro­grams is 10. For each inef­fi­ each inef­fi­cient in­vest­ment pro­gram be­co­me ef­fi­cient are
cient pro­gram, the list of peer pro­grams is gi­ven. gi­ven in Tab­le 2.
Bous­so­fia­ne et al., 1991, pro­po­ses that a sim­ple count The re­sults ob­tai­ned, show that DEA can be suc­cess­
of the fre­quency that an ef­fi­cient unit ap­pears in dif­fe­rent fully used in sup­por­ting the de­ci­sion ma­king pro­cess of
peer groups could be used as an al­ter­na­ti­ve in­di­ca­tor of in­vest­ment banks. The re­sults of the ap­pli­ca­tion of the
good ope­ra­ting prac­ti­ce. This count in­di­ca­tes the ex­tent DEA met­hod to as­sess the re­la­ti­ve ef­fi­ciency of in­vest­
that a re­la­ti­vely ef­fi­cient unit is a self eva­lua­tor or an eva­ ments in agri­cul­tu­re pro­vi­de:
lua­tor of ot­her units. Ac­cor­ding to the fre­quency count, n a mea­su­re of the grea­test re­la­ti­ve ef­fi­ciency that each
the re­sults ob­tai­ned in­di­ca­te that the in­vest­ment pro­gram in­vest­ment pro­gram can ac­hie­ve ac­cor­ding to the fac­
P10 ap­pears in all the peer groups. A sa­tis­fac­tory ra­ting is tors and fields inc­lu­ded in the analy­sis,
gi­ven to the in­vest­ment pro­grams P18 and P7, ap­pea­ring in n in­for­ma­tion about the most im­por­tant in­puts and out­
5 peer groups, as well as P14 and P17 with 4 ap­pea­ran­ces. puts for each ef­fi­cient in­vest­ment,
One of the ways to com­pa­re re­la­ti­vely ef­fi­cient units n the list of ef­fi­cient in­vest­ments that form a peer group
is the di­stri­bu­tion of vir­tual in­puts and out­puts. Vir­tual for each inef­fi­cient in­vest­ment and
in­puts and/or out­puts are ob­tai­ned by mul­tipl­ying their n a re­port on the ex­cess of in­puts and the lack of out­
mag­ni­tu­des with the cor­res­pon­ding op­ti­mal weights vi puts when the re­la­ti­ve ef­fi­ciency of an in­vest­ment is
*
and ur. Thus, ur ∗ yrj is a vir­tual out­put r for unit j, whe­re less than 1.
*
ur de­no­tes the op­ti­mal va­lue for ur in (M1). The mag­ni­ In Mar­ti}, Kr­~e­vi­nac & Pe­tri}, it ������������������
is also shown how
tu­des of the vir­tual in­puts and out­puts show the con­tri­bu­ the DEA met­hod could be used for ran­king a set of ef­fi­

Tab­le 2: The va­lue of the in­puts and out­puts that make each inef­fi­cient in­vest­ment pro­gram be­co­me ef­fi­cient

42
Unauthenticated
Download Date | 12/20/15 7:32 AM
 Organizacija, Volume 42 Research papers Number 2, March-April 2009

cient in­vest­ment pro­grams in agri­cul­tu­re using An­der­sen- Char­nes, A., Coo­per W. & Rho­des E. (1978). Mea­su­ring the
Pe­ter­sen’s DEA mo­del and a cross ef­fi­ciency ma­trix. The Ef­fi­ciency of De­ci­sion Ma­king Units. Eu­ro­pean Jour­nal of
re­sults ob­tai­ned are analy­zed and com­pa­red. Ope­ra­tio­nal Re­search, 2: 429-444.
Coo­per, W., Sei­ford, L. & Tone, K. (2005). In­tro­duc­tion to Data
En­ve­lop­ment Analy­sis and its Uses: With DEA-Sol­ver Soft­
wa­re and Re­fe­ren­ces. New York: Sprin­ger.
4 Conc­lu­sions Coo­per, W., Sei­ford, L., Tha­nas­sou­lis, E. & Za­na­kis, S. (2004).
DEA and its Uses in Dif­fe­rent Coun­tries. Eu­ro­pean Jour­nal
DEA is a non-pa­ra­me­ter met­ho­do­logy for eva­lua­ting of Ope­ra­tio­nal Re­search, 154: 337-344.
the ef­fi­ciency of non-pro­fit DMU­s. It con­tains so­lu­tions Emrouz­ne­jad, A., Par­ker, B. & Ta­va­res, G. (to ap­pear). Eva­lua­
for se­ve­ral mu­tually con­nec­ted li­near pro­gram­ming mat­ tion of re­search in ef­fi­ciency and pro­duc­ti­vity: A sur­vey and
he­ma­ti­cal mo­dels for each of the DMU­s. Whi­le each of analy­sis of the first 30 years of scho­larly li­te­ra­tu­re in DEA.
So­cio-Eco­no­mic Plan­ning Scien­ces.
the­se mo­dels ad­dres­ses ma­na­ge­rial and eco­no­mic is­sues
Mar­ti}, M., Kr­~e­vi­nac S. & Pe­tri} J. (1996). An Ap­pli­ca­tion of
and pro­vi­des use­ful re­sults, their orien­ta­tions are dif­fe­rent Data En­ve­lop­ment Analy­sis to As­sess Re­la­ti­ve Ef­fi­ciency
and, more im­por­tant, they ge­ne­ra­li­ze and pro­vi­de con­tact of In­vest­ments in Agri­cul­tu­re. Stu­dies in Re­gio­nal&Ur­ban
with the­se dis­ci­pli­nes and con­cepts. Thus, the mo­dels may Plan­ning, 4: 141-152.
fo­cus on in­crea­sing, de­crea­sing or con­stant re­turns to sca­ Mar­ti}, M. & Savi} G. (2001). An Ap­pli­ca­tion of DEA for Com­
le as found in eco­no­mics, which are here ge­ne­ra­li­zed into pa­ra­ti­ve Analy­sis and Ran­king of Re­gions in Ser­bia with
the form of mul­ti­ple out­puts. Re­gards to So­cial-Eco­no­mic De­ve­lop­ment. Eu­ro­pean Jour­
nal of Ope­ra­tio­nal Re­search, 129: 344-355.
The ex­ten­si­ve but pro­bably in­com­ple­te bib­lio­graphy Sei­ford, L. (1997). A Bib­lio­graphy for Data En­ve­lop­ment
(Emrouz­ne­jad et al., to ap­pear) is in­ten­ded to do­cu­ment Analy­sis (1978-1996). An­nals of Ope­ra­tions Re­search, 73:
the dif­fu­sion and growth of DEA usa­ge. The bib­lio­graphy 393-438.
shows DEA ap­pli­ca­tions in a wide ran­ge of con­texts, such Tha­nas­sou­lis, E. (2001). In­tro­duc­tion to the theory and ap­pli­ca­
as edu­ca­tion (pub­lic schools and uni­ver­si­ties), health care tion of data en­ve­lop­ment analy­sis: a foun­da­tion text with
(hos­pi­tals, cli­nics and physi­cians), ban­king, the ar­med for­ In­te­gra­ted Soft­wa­re. Bo­ston: Klu­wer Aca­de­mic Pub­lis­her.
ces (re­crui­ting and air­craft main­te­nan­ce), au­di­ting, sports,
mar­ket re­search, mi­ning, agri­cul­tu­re, re­tail out­lets, or­ga­ni­ Mi­lan M. Mar­ti} ob­tai­ned his doc­to­ral de­gree in the field
za­tion ef­fec­ti­ve­ness, trans­por­ta­tion (fer­ries and high­way of Or­ga­ni­za­tio­nal Scien­ces from the Uni­ver­sity of Bel­gra­de.
main­te­nan­ce), pub­lic hou­sing, in­dex num­ber con­struc­tion, He works as an As­so­cia­te pro­fes­sor at the Uni­ver­sity of
bench­mar­king, etc. Bel­gra­de, Fa­culty of Or­ga­ni­za­tio­nal Scien­ces in the field of
Ope­ra­tions Re­search. His re­search in­te­rests inc­lu­de Data
En­ve­lop­ment Analy­sis, Per­for­man­ce Mea­su­re­ment, Op­ti­mi­
Re­fe­ren­ces za­tion Met­hods and Pro­ject Sche­du­ling.

An­der­sen, P. & Pe­ter­sen, N. (1993). A Pro­ce­du­re for Ran­king

Ma­ri­na S. No­va­ko­vi} ob­tai­ned her ma­ster de­gree in the
Ef­fi­cient Units in Data En­ve­lop­ment Analy­sis. Ma­na­ge­
field of Or­ga­ni­za­tio­nal Scien­ces from the Uni­ver­sity of
ment Scien­ce, 39: 1261-1264.
Bel­gra­de. She works at the of­fi­ce of stu­dent af­fairs and
Ban­ker, R., Char­nes A. & Coo­per W. (1984). Some Mo­dels for
con­ducts all ad­mi­ni­stra­ti­ve af­fairs re­la­ted to teac­hing and
Es­ti­ma­ting Tech­ni­cal and Sca­le Inef­fi­cien­cies in Data En­ve­
stud­ying on gra­dua­te le­vel. Her re­search in­te­rests inc­lu­de
lop­ment Analy­sis. Ma­na­ge­ment Scien­ce, 30:1078-1092. Ope­ra­tions Re­search, Pub­lic
�����������������������������������
Re­la­tions and Bu­si­ness ����
Com­
Ban­ker, R. & Mo­rey, R. (1986). Ef­fi­ciency Analy­sis for Exo­ge­ mu­ni­ca­tion.
nously Fi­xed In­puts and Out­puts. Ope­ra­tions Re­search,
34(4): 513-521.
Bous­so­fia­ne, A., Dyson G. & Tha­nas­sou­lis E. (1991). Ap­plied Alen­ka Bag­gia ob­tai­ned her ma­ster de­gree in the field of
Data En­ve­lop­ment Analy­sis Eu­ro­pean Jour­nal of Ope­ra­tio­ Or­ga­ni­za­tio­nal Scien­ces from the Uni­ver­sity of Ma­ri­bor. She
nal Re­search, 52: 1-15. works as a teac­hing as­si­stant at the Uni­ver­sity of Ma­ri­bor,
Char­nes, A., Coo­per W., Le­win, A. & Sei­ford, L. (1995). Data Fa­culty of Or­ga­ni­za­tio­nal Scien­ces in the Soft­wa­re Qua­lity
En­ve­lop­ment Analy­sis: Theory Met­ho­do­logy and Ap­pli­ca­ and Te­sting La­bo­ra­tory. Her re­search in­te­rests inc­lu­de pro­
tion. Bo­ston: Klu­wer Aca­de­mic Pub­lis­her. duc­tion and per­son­nel sche­du­ling al­go­rithms and systems.

43
Unauthenticated
Download Date | 12/20/15 7:32 AM
View publication stats