You are on page 1of 41

1 2007-05-30 Ana Camanho

Performance Measurement using Data


Envelopment Analysis (DEA) and
Malmquist indices:
Issues, Challenges and Applications.
Ana Camanho
University of Porto, Portugal
(acamanho@fe.up.pt)
Accounting and Management Science Seminars
Norwegian School of Economics and Business Administration
30-05-2007
2 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
The School of Engineering
3 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
University of Porto - School of Engineering
15 Schools
60 graduate programs
120 master programs
100 doctoral programs
2200 lecturers and researchers
1600 administrative staff
27000 students, of which 3500
postgraduate
7 Departments
10 graduate programs
25 master programs
13 doctoral programs
450 lecturers and researchers
250 administrative staff
6000 students, of which 1000
postgraduate
4 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Contents
Introduction do Data Envelopment Analysis (DEA)
Concept efficiency measurement
Input or output orientation
Returns to scale
By-products of a DEA analysis
DEA formulation (linear programming model)
Additional topics in DEA
DEA in practice: case study of retailing stores
- Efficiency analysis adjusting for environmental factors
- Target setting using network DEA
Productivity change over time
Introduction to Malmquist index
DEA in practice: case study of retailing stores
- Analysis of productivity change using Malmquist index
- Comparison of performance of stores with different configurations: The use of a
new Malmquist-type index.
Conclusions
5 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Introduction to DEA
Model of efficiency analysis
Objective of a Data Envelopment Analysis (DEA) assessment:
Comparison of performance of homogeneous decision making units
(DMUs) that use multiple inputs for the production of multiple outputs.
The efficiency measure compares the ratio output/input of the DMU
assessed with the value of this ratio observed in the other DMUs
analysed.
Decision Making Units
Decision Making Units Decision Making Units
Inputs
Outputs
6 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Introduction to DEA
Graphical illustration of the DEA concept
Single input and single output, assuming Constant Returns to Scale (CRS)
Output
Input
Inefficiency
A
E
x
*
E
Efficient frontier
Efficiency =
*
xE
xE
7 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Introduction to DEA
Output or input oriented analysis
Input oriented measures keep output fixed
- Input oriented efficiency indicates by how much can input quantities be proportionally reduced
holding output constant.
Output oriented measures keep input fixed
- Output oriented efficiency indicates by how much can output quantities be proportionally increased
holding input constant.
Output
A
E
x
*
E
Efficient frontier
Output
efficiency =
*
xE
xE
y
Input
efficiency =
yE
yE
* *
* *
E
Input
Scope for output
augmentation: EE*
Scope for input
reduction: EE**
8 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Introduction to DEA
Returns to scale assumptions
Constant Returns to Scale (CRS) or Variable Returns to Scale (VRS)
Output
Input
A
Efficient frontier (CRS)
Efficient frontier (VRS)
Increasing returns to scale
Decreasing returns to scale
Input efficiency
(CRS) =
yE
yE
*CRS
Input efficiency
(VRS) =
yE
yE
*VRS
E
E*
CRS
E*
VRS
Scale efficiency =
VRS *
*CRS
yE
yE
CRS efficiency = VRS efficiency Scale efficiency
y
9 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Introduction to DEA
Output or input oriented analysis
Choice depends on analysts view over which variables (inputs or outputs) it is
believed managers can exercise control.
Input and output orientation will estimate the same frontier.
Input and output oriented measures of efficiency are equivalent under CRS.
Under VRS, input and output oriented analysis will give different measures of
efficiency for DMUs with efficiency < 1.
Constant returns to scale or variable returns to scale
Analyst must understand the constraints of the sector analysed.
Choice depends on the purpose of analysis and whether short-run or long-run
efficiency is examined.
A VRS assessment implies that firms are only compared to others firms of
roughly similar size.
VRS produces efficiency scores greater than or equal to CRS efficiency scores.
10 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Introduction to DEA
What else can we learn from efficiency analysis?
Cost efficiency
target
Input 2 /output
DMU E
A
C
B
D
Isocost line
Efficient
frontier
Technical
efficiency target
Benchmarks
O
E
T
Input 1 /output
Production possibility set
E
C
Technical
efficiency =
OE
OE
T
Cost
efficiency =
OE
OE
C
Allocative
efficiency =
T
C
OE
OE
Tech. Eff Alloc. Eff
= Cost Eff
11 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Formulation of the DEA model
DEA is based on linear programming [Charnes et al., 1978]
Efficiency measure with one input and one output:
with multiple inputs and outputs
But, firm outputs cannot be added together directly, and the same for the
inputs
If we know the output weights and input weights, the job is done. There is no
need for sophisticated analysis. A major contribution of DEA is to offer insights
about the value of weights.
Input
Output
= Efficiency
... Input Input Input
.... Output Output Output
3 3 2 2 1 1
3 3 2 2 1 1
+ + +
+ + +
=
weight weight weight
weight weight weight
Efficiency
... Input Input Input
.... Output Output Output
3 2 1
3 2 1
+ + +
+ + +
= Efficiency
12 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Formulation of the DEA model
For each DMU, we have a model that maximizes the efficiency score,
subject to all other DMUs having efficiencies less than or equal to one.

=

=
=
=
=
r r
m
1 i
i ij
s
1 r
r rj
m
1 i
i ij
s
1 r
r rj
j
,
n 1,..., j DMU each for 1
x
y
: subject to
x
y
Max
0
0
0
v u
v
u
v
u
e

=
=
=

= =
=
=
r r
m
1 i
i ij
s
1 r
r rj
m
1 i
i
0
ij
s
1 r
r rj
,
n 1,..., j , 0 x y
1 x
: subject to
y Max
0 0
v u
v u
v
u e
j
analysis under DMU j
score efficiency
output to attached weight
j unit for r output of quantity y
i input to attached weight
j unit for i input of quantity x
0
r
rj
i
ij
0
=
=
=
=
=
=
j
e
r u
v
Input oriented DEA model with
Constant Returns to Scale
Linear programming model
[Charnes et al., 1978]
13 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Formulation of the DEA model
DEA input oriented model (with CRS).
j , 0
s 1,..., r , y y
m 1,..., i , x x
: subject to
Min
j
n
1 j
rj j rj
n
1 j
ij j ij 0
0 j
0
0
0

=
=
=

=
=


e
weights formulation
Envelopment formulation

=
=
=

= =
=
=
r r
m
1 i
i ij
s
1 r
r rj
m
1 i
i
0
ij
s
1 r
r rj j
,
n 1,..., j , 0 x y
1 x
: subject to
y Max
0 0
v u
v u
v
u e
score efficiency
output to attached weight
j unit for r output of quantity y
i input to attached weight
j unit for i input of quantity x
0
r
rj
i
ij
=
=
=
=
=
j
e
r u
v
Duality of Linear Programming

14 2007-05-30 Ana Camanho (acamanho@fe.up.pt)


free is
,
n 1,..., j , 0 x y
1 x
: subject to
y Max
r r
m
1 i
i ij
s
1 r
r rj
m
1 i
i
0
ij
s
1 r
r rj j
0 0
w
v u
w v u
v
w u e

= +
=
+ =

= =
=
=
Formulation of the DEA model
DEA input oriented model (with VRS) [Banker et al., 1984]
1
s 1,..., r , y y
m 1,..., i , x x
: subject to
Min
1
j
n
1 j
rj j rj
n
1 j
ij j ij 0
0 j
0
0
0
=
=
=
=

=
=
=
n
j
e

weights formulation
Envelopment formulation
score efficiency
output to attached weight
j unit for r output of quantity y
i input to attached weight
j unit for i input of quantity x
0
r
rj
i
ij
=
=
=
=
=
j
e
r u
v
Duality of Linear Programming

15 2007-05-30 Ana Camanho (acamanho@fe.up.pt)


DEA: evolution of the state-of-the-art
(Seiford, 1996, Journal of productivity analysis)
Antecedents:
Farrell (1957)
Pareto (1927)
Koopmans (1951)
Shephard (1970)
Malmquist (1953)
DEA model
[Charnes et al,1978]
VRS model
[Banker et al, 1984]
Non-radial models
[Charnes et al., 1985]
Models with weight restrictions
[Dyson and Thanassoulis, 1988]
Models with non-discretionary variables
[Banker and Morey, 1986]
Analysis of panel data
(Malmquist index) [Fare et al., 1994]
Comparison of groups
(Program efficiency) [Charnes et al., 1981]
Statistical foundation for DEA [Banker, 1996]
and Confidence intervals on DEA
efficiencies using Bootstrapping [Simar and
Wilson, 1998].
Stochastic Frontiers
Aigner et al. (1977)
Target setting using Network DEA models
(DEA models that account for interrelations
between DMUs and sub-DMUs).
[Thanassoulis and Dyson, 1992; Fare et al., 1997]
Other
enhancements
16 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
analysis of a retailing organization
The structure of the retailing organisation
(largest Portuguese retailing organisation)
Case study: 70 stores (14 hypermarkets and 56 supermarkets)
Objectives of the performance assessment
To analyse the impact of exogenous conditions (competition and population) on store performance.
To define targets for sales maximization, allowing for the reallocation of area among sections
To analyse the evolution of performance over time.
To compare the performance of Heavy Bazaar sections with different configurations.
T70 HB70 LB70 P70 G70 Store 70
T1 HB1 LB1 P1 G1 Store 1
T2 HB2 LB2 P2 G2 Store 2

Textiles
section
Heavy Bazaar
section
Light Bazaar
section
Perishables
section
Grocery
section
Commercial
management
Operational
management
T70 HB70 LB70 P70 G70 Store 70
T1 HB1 LB1 P1 G1 Store 1
T2 HB2 LB2 P2 G2 Store 2

Textiles
section
Heavy Bazaar
section
Light Bazaar
section
Perishables
section
Grocery
section
Commercial
management
Operational
management
17 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
Why consider non-discretionary variables in efficiency
assessments?
To allow fair comparisons: DMUs facing unfavourable exogenous
conditions should not be penalised for producing less output or
consuming more inputs than the other DMUs.
Examples of non-discretionary variables:
Competition and population density (e.g., affect bank branches,
supermarkets, restaurants):
Cultural and economic level of families (e.g., school results)
Fixed production quotas and farming areas (agriculture and fisheries)
18 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
The model by Banker and Morey, 1986.
Treats ND variables as a different set. For example, for an input
oriented analysis with ND inputs, the equiproportional input
reductions are only looked for controllable inputs:

=
=


= =
= =
0 , 1 , ,
, , , ,..., 1 , |
1 1
1 1
j
n
j
j io
n
j
ij j
io o
n
j
ij j ro
n
j
rj j o
ND i x x
D i x x s r y y
Min


=


= =
= =
0 , ,
, , , ,..., 1 , |
1 1
1 1
j io
n
j
j
n
j
ij j
io o
n
j
ij j ro
n
j
rj j o
ND i x x
D i x x s r y y
Min


VRS:
CRS:
19 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
Multi-stage models
Ray (1988, 1991): objective is to explain the differences in efficiency
scores based on the effect of non-discretionary variables.
- 1st stage: run a DEA model without ND variables
- 2nd stage: regress DEA scores on ND variables and correct the efficiency
score based on the impact of ND variables estimated with regression.
Grosskopf et al. (1997): objective is to adjust the controllable input
and output variables according to the effect of ND factors
- 1st stage: regression analysis is used to adjust the controllable input and
output variables according to the effect of ND factors
- 2nd stage: run DEA with the adjusted variables
Various further methodological proposals no generally
accepted method.
20 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
Ruggiero (1996):
The PPS is defined only by controllable variables
The comparison among DMUs is fair because it is ensured that the
peers will always have ND factors that are equal or less favourable
than those of the DMU assessed.
In practice, this model constructs several different frontiers,
according to the values of the ND factors of the DMUs assessed.
C H
E
F
B
G
A
D
0
2
4
6
8
10
0 2 4 6 8 10 12
Input
O
u
t
p
u
t
frontier 1
frontier 2
frontier 3
frontier 5
frontier 4
VRS
DMU Efficiency Peers
A 100% A =1
B 100% B=1
C 100% C=1
D 100% D=1
E 100% E=1
F 75% E=1
G 89.58% A =0.83, C=0.17
H 75% C=1
x
1
(D) x
2
(ND) y
A 8 8 8
B 6 4.6 5
C 3 1.9 2
D 10 9 9
E 6 3.6 4.5
F 8 3.6 4.5
G 8 9 7
H 4 1.9 2
21 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
Model of retailing activity at the store level
Objective: Maximize sales
The store assessment is based on the construction of a production possibility set defined
only by controllable factors. The effect on NC factors is taken into account by restricting
the peers of each DMU to stores with similar or worst environmental conditions, as in
Ruggiero, 1996. The ND factor considered is the ratio population/competition.
Sales
Operational costs
Store
Stock
Staff costs
Area
Products spoiled
Population / competition
22 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
,...,n j ,
ND , i x x then if
D ,r y y
D ,i x x
j
n
j
j
ij ij j
rj
n
j
j rj
ij
n
j
j ij
o
o
o
1 0 , 1
0
Max
1
1
1
) , (
j
= =
>

=
=
=

1 0
1 0
<
+ >

ND ), i ( x x then If
o
ij ij j
{ } 001 . 0 1 0
1 0
1
=

<
+
, ,
,

ND ), i ( x x
j
j j j j
ij j ij
o
The criteria for peer selection is having a ratio population/competition equal or lower than 1.2
times the value of this ratio in the store assessed. This correspond to =20% in the new DEA model.
23 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
Results: example for one store, comparing observed inputs and outputs
and DEA targets, accounting for ND factors.
Output:
Inputs:
Original value
Objective
Original value
Objective
Original value
Objective
Sales
Area
Stock Op. costs staff costs prod. spoiled
24 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
models with non-discretionary variables
Is the impact of NC factors on efficiency significant?
The differences in the inefficiency distributions corresponding to the model accounting for
ND factors and the model only with controllable factors were tested using the K-S test.
The test revealed that the NC factors have a significant effect on store activity (p=0.004).
When the NC factors are included in the model, the efficiency value increases in 50
stores. For these stores, the efficiency values increasing up to 15% (with the exception of
two stores, with efficiency increase around 25% and 35%).
0
5
10
15
20
25
30
0% 5% 10% 15% 20% 25% 30% 35%
Di fference between the resul ts of the model wi thout NC
factors and the new model wi th NC factors (%)
F
r
e
q
u
e
n
c
y
25 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Target setting using Network DEA
Model of retailing activity at the section level
Objective: Maximize store sales taking into account the inter-
relations among sections of the same store
The sales target for each section should be defined allowing for the
reallocation of floor space among sections within the store.
Model used in based on Fare et al. (1997)
Sales
No. references
Store sections
(grocery, perishables,
heavy bazaar, light
bazaar, textiles)
Stock
Area
Products spoiled
26 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Target setting using Network DEA
Network DEA model with output orientation and Constant Returns to Scale.

=
=
=
=
=
=
=
=

+ + + +
n
1 j
j textils] [area j j textils] [area
n
1 j
j bazaar] light [area j j bazaar] light [area
n
1 j
j bazaar] heavy [area j j bazaar] heavy [area
n
1 j
j groceries] [area j j groceries] [area
n
1 j
j s] perishable [area j j s] perishable [area
sections all
j section] [area j area] [store
n
1 j
spoiles]j [products j spoiled]j [products
n
1 j
[stock]j j [stock]j
n
1 j
s]j [reference j s]j [reference
] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [
x
x
x
x
x
x
x x
x x
x x
: subject to
Max
0
0
0
0
0
0 0
0
0
0
0 0 0 0 0


v
v
v
v
v
v
y y y y y
j tex tex j lb lb j hb hb j gro gro j per per
0
0
j 0
y
y
y
y
y
z] [section
z] section [area
n
1 j
textils]j [sales j j textils] [sales [tex]
n
1 j
bazaar]j light [sales j j bazaar] light [sales [lb]
n
1 j
bazaar]j heavy [sales j j bazaar] heavy [sales [hb]
n
1 j
j groceries] [sales j j groceries] [sales [gro]
n
1 j
s]j perishable [sales j j s] perishable [sales [per]
0
0
0
0
0

=
=
=
=
=






v
y
y
y
y
y
j
27 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Target setting using Network DEA
Results: example of targets for one store, comparing standard DEA with
the Network DEA model:
Original value
Objective DEA
Objective Network DEA
Original value
Objective DEA
Objective Network DEA
Original value
Objective DEA
Objective Network DEA
Original value
Objective DEA
Objective Network DEA
Grocery Perishables Heavy B. Light B. Textiles
Grocery Perishables Heavy B. Light B. Textiles
Products spoiled Stock
No. references
28 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Productivity change over time
The standard approach to the measurement of productivity
change over time is the Malmquist index.
(Caves et al, 1982; Fare et al, 1994)
For use when we have panel data.
Decomposes productivity change into efficiency change (firms moving
closer to the frontier) and technological change (shifts in the
frontier).
No need for price data, no need for assumptions of cost minimisation
or revenue maximisation.
Input-based or output-based Malmquist index.
Based on input or output distance functions.
Calculates Total Factor Productivity (TFP) using DEA models.
29 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Productivity change over time
Productivity change between 2 data points is calculated by ratios
of distances of each point relative to a common technology.
The Malmquist index is a geometric mean of two indices, evaluated with
respect to period t and period t+1 technologies (Fare et al, 1994).
For output orientation: M
O
>1 Productivity growth
M
O
<1 Productivity decline
Decomposition of the index:
2 / 1
1 t
O
1 1
1 t
O
t
O
1 1
t
O
1 1 O
) , (
) , (
) , (
) , (
) , , , ( M

=
+
+ +
+
+ +
+ +
t t
t t
t t
t t
t t t t
x y d
x y d
x y d
x y d
x y x y
2 / 1
1 t
O
t
O
1 1
1 t
O
1 1
t
O
t
O
1 1
1 t
O
1 1 O
) , (
) , (
) , (
) , (
) , (
) , (
) , , , ( M

=
+
+ +
+
+ + + +
+
+ +
t t
t t
t t
t t
t t
t t
t t t t
x y d
x y d
x y d
x y d
x y d
x y d
x y x y
Efficiency change
Technological change
30 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Productivity change over time
For output oriented analysis:
Distance function = DEA efficiency
Output 2 /input
a
O
Output 1 /input
Frontier period t
Y
t+1
Y
t
d
c
b
Frontier period t+1
2 / 1
1 t
O
t
O
1 1
1 t
O
1 1
t
O
t
O
1 1
1 t
O
O
) , (
) , (
) , (
) , (
) , (
) , (
M

=
+
+ +
+
+ + + +
+
t t
t t
t t
t t
t t
t t
x y d
x y d
x y d
x y d
x y d
x y d
Efficiency change
Technological change
1/2
t
1 t
1/2
t
t
1 t
1 t
t
1 t
O
Oa
Ob
Oc
Od
Oa
Oy
Od
Oy
Ob
Oy
Oa
Oy
Od
Oy
Oc
Oy
Oa
Oy
Od
Oy
M

=
+
+
+ +
<1:
efficiency
decline
>1:
technological
improvement
31 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Retailing example: productivity change
Model of retailing activity at the store level
Analysis of store productivity change between 2002 and 2004
The Malmquist index was decomposed further to identify scale efficiency changes
(Fare et al., 1994).
2 / 1
1 t
O
t
O
1 1
1 t
O
1 1
t
O
t
O
1 1
1 t
O
t
O
1 1
1 t
O
1 1 O
) , (
) , (
) , (
) , (
) , (
) , (
) , (
) , (
) , , , ( M

=
+
+ +
+
+ + + +
+
+ +
+
+ +
t t
t t
t t
t t
t t
t t
t t
t t
t t t t
x y d
x y d
x y d
x y d
x y S
x y S
VRS x y d
VRS x y d
x y x y
Technical efficiency
change (VRS)
Technological change
Scale efficiency
change
Sales
Operational costs
Store
Stock
Staff costs
Area
Products spoiled
32 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Retailing example: productivity change
Efficiency status of the stores:
Productivity change:
(no. of stores with an index greater than 1 [improvement], equal to 1, or smaller than 1 [decline])
10 stores promoted innovation (5 supermarkets and 5 hypermarkets)
These stores define the frontier in t+1 and are beyond the production possibility
set of period t.
Year 2002 Year 2004
No. efficient stores 17 16
No. inefficient stores 53 54
No. stores remained in the frontier between 2002 and 2004
Average efficiency (for inefficient stores) 87% 87%
11
Malmquist
index
Technological
change
Efficiency
change (VRS)
Scale efficiency
change
improve (index >1) 12 7 13 4
14 hypermarkets decline 2 2 1 6
equal 0 5 0 4
improve 24 21 16 14
56 supermarkets decline (index <1) 32 29 37 41
equal 0 6 0 1
33 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Comparison of performance of sections with different configurations
Objective:
To explore the differences in performance between two groups of
heavy bazaar sections (selling electrical appliances, electronics,)
Although all sections are from the same organisation:
- Sections in Group A are within hypermarkets and large supermarkets
located in large cities
- Sections in Group B are within smaller supermarkets, located in small
cities/towns.
- Sample of 18 stores in each group
The research question is: which group of stores performs better?
- Better performance implies having:
- Less efficiency spread within the group
- A more productive frontier.
34 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Comparison of performance of stores with different configurations
Malmquist-type index for group comparisons: [Camanho and Dyson, 1996]
The new index (I
AB
) compares the performance of groups of DMUs operating
under different conditions (A and B).
- The index focuses on comparisons in a static setting (i.e., for a given moment in
time).
- The index handles directly all the observations corresponding to individual DMUs.
This index can be decomposed into two sub-indexes:
- Comparison of efficiency spread between groups (IE
AB
).
- Comparison of productivity differences between the group best-practice frontiers
(IF
AB
).
AB AB AB
IF IE I =
35 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Comparison of performance of stores with different
configurations
Overall Malmquist-type index for comparison of group
performance
(index I
AB
, with DMUs in group A and DMUs in group B)
For an output oriented analysis, a value of I
AB
> 1 indicates that group A
performs better than group B.
2 / 1
/ 1
1
/ 1
1
/ 1
1
/ 1
1
) , (
) , (
) , (
) , (

=
=
=
=

j
B
j
B
j
B
j
A
j
A
j
B
j
B
j
B
j
A
j
A
j
A
j
A
AB
y x d
y x d
y x d
y x d
I
AB AB AB
IF IE I =
36 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Comparison of performance of stores with different configurations
Comparison of efficiency spread between groups
(index IE
AB
, with DMUs in group A and DMUs in group B)
For an output oriented analysis, a value of IE
AB
> 1 indicates that
the efficiency spread is smaller in DMUs of group A than in group B

=

=

1
1 j
1
1 i AB
B group of frontier the to B group in DMU of Distance
A group of frontier the A to group in DMU of Distance
IE
i
i

/ 1
1
/ 1
1
AB
) , (
) , (
IE

=
=
j
B
j
B
j
B
j
A
j
A
j
A
y x d
y x d
37 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
DEA in practice:
Comparison of performance of stores with different configurations
Comparison of productivity differences between the group frontiers
(index IF
AB
, with DMUs in group A and DMUs in group B)
For an output oriented analysis, a value of IF
AB
> 1 indicates
greater productivity of the frontier of group A than the
frontier of group B
2 / 1
/ 1
1
/ 1
1
/ 1
1
/ 1
1
AB
) , (
) , (
) , (
) , (
IF

=
=
=
=

j
B
j
B
j
A
j
B
j
B
j
B
j
A
j
A
j
A
j
A
j
A
j
B
y x d
y x d
y x d
y x d
38 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Model of heavy bazaar activity:
Heavy bazaar sections from group A have better performance (I
AB
>1).
- The frontier of group A is more productive (IF
AB
>1), but the efficiency spread is larger (IE
AB
<1)
which indicates that there is still scope for efficiency improvements in group A.
Non-parametric tests were used to test the statistical significance of the differences
between groups, captured by the indices IE
AB
and IF
AB
.
- The Kolmogorov-Smirnov test showed that the position of the frontiers and the efficiency spreads
are different between the stores of groups A and B.
DEA in practice:
Comparison of performance of stores with different configurations
Sales
Stock
No. of references
Area
Products wasted
I
AB
IE
AB
IF
AB
Heavy Bazaar
Output
Input
Group A
Group B
39 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Conclusions:
Advantages and disadvantages of DEA
Advantages:
Easy to use
Allows multiple inputs and multiple outputs
Does not require specification of functional form for the frontier
Does not require a priori specification of weights for inputs and outputs
Inputs and outputs can be expressed in different measurement units
Disadvantages:
No account for measurement error / random noise (all shortfall in the
input-output ratio of a DMU is inefficiency).
Sensitive to outlier data
40 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
Software available:
EMS: Efficiency Measurement System, version 1.3 University of Dortmund, by Holger Scheel.
Available freely at http://www.wiso.uni-dortmund.de/lsfg/or/scheel/ems/
DEAP version 2.1 Centre for Efficiency and Productivity Analysis, University of New England,
Australia, by Tim Coelli
Available freely at http://www.uq.edu.au/economics/cepa/software.htm
Frontier Analyst, version 4 Banxia Software in Glasgow
Commercially available at http://www.banxia.com/famain.html
Performance Improvement Management (PIM DEA SoftV1.), by Emmanuel Thanassoulis & Ali
Emrouznejad (developers of Warwick DEA software)
Commercially available at http://www.deasoftware.co.uk/
Thanassoulis, E. (2001) Introduction to the theory and Application of Data Envelopment Analysis: A
foundation text with integrated software, Kluwer Academic Publishers.
DEAFrontier Joe Zhu
Details at http://www.deafrontier.com/software.html
Zhu, J. (2003) Quantitative models for performance evaluation and bechmarking: Data Envelopment
Analysis with Spreadsheets and DEA excel solver, Kluwer Academic Publishers.
DEA-Solver-PRO version 5 SAITECH, Inc.
Commercially available at http://www.saitech-inc.com/Products/Prod-DSP.asp
Cooper, W.W., Seiford, L.M., Tone, K. (2006), Introduction to Data Envelopment Analysis and Its Uses:
With DEA-Solver Software and References, Springer.
OnFront version 2 Economic Measurement and Quality AB (EMQ AB)
Available at http://www.emq.com/software.html
41 2007-05-30 Ana Camanho (acamanho@fe.up.pt)
References:
Aigner, D., Lovell, C. A. K., and Schmidt, P. (1977). Formulation and estimation of stochastic frontier production function models. Journal of
Econometrics, 6(1):2137.
Banker, R. D., Charnes, A., and Cooper, W. W. (1984). Some models for estimating technical and scale inefficiencies in Data Envelopment
Analysis. Management Science, 30(9):1078-1092.
Banker, R. D. and Morey, R. C. (1986). Efficiency analysis for exogenously fixed inputs and outputs. Operations Research, 34(4):513-521.
Banker, R. (1996). Hypothesis tests using Data Envelopment Analysis. Journal of Productivity Analysis, 7(2-3):139-159.
Caves, D. W., Christensen, L. R., and Diewert, W. E. (1982). The economic theory of index numbers and the measurement of input, output and
productivity. Econometrica, 50:13931414.
Camanho, A. S. and Dyson, R. G. (2006). Data envelopment analysis and Malmquist indices for measuring group performance. Journal of
Productivity Analysis, 26(1):3549.
Charnes, A., Cooper, W., and Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research,
2(6):429-444.
Charnes, A., Cooper, W.W., Rhodes, E. (1981). Evaluating program and managerial efficiency: an application of data envelopment analysis to
program follow through. Management Science, 27(6):668697.
Charnes, A., Cooper, W.W., Golany, B., Seiford, L., Stutz, J. (1985). Foundations of Data Envelopment Analysis for Pareto-Koopmans Efficient
Empirical Production Functions. Journal of Econometrics, 30 (1/2), 91-107.
Dyson, R., Thanassoulis, E. (1988). Reducing weight flexibility in Data Envelopment Analysis. Journal of the Operational Research Society, 39
(6), 563-576.
Fre, R., Grabowski, R., Grosskopf, S., and Kraft, S. (1997). Efficiency of a fixed but allocatable input: A non-parametric approach. Economics
Letters, 56:187-193.
Fare, R., Grosskopf, S., Lindgren, B., Roos, P. (1994) Productivity developments in swedish hospitals: a Malmquist output index approach. In:
Charnes A, Cooper WW, Lewin AY, Seiford LM (eds). Data envelopment analysis: theory, methodology and applications. Kluwer Academic
Publishers, Boston, pp 253272.
Grosskopf, S., Hayes, K., Taylor, L., and Weber, W. (1997). Budget-constrained frontier measures of scale equality and efficiency in schooling.
The review of Economics and Statistics, 79:116124.
Ray, S. C. (1988). Data Envelopment Analysis, nondiscretionary inputs and efficiency: an alternative interpretation. Socio-Economic Planning
Sciences, 22(4):167176.
Ray, S. C. (1991). Resource-use efficiency in public-schools - a study of Connecticut data. Management Science, 37(12):16201628.
Ruggiero, J. (1996). On the measurement of technical efficiency in the public sector. European Journal of Operational Research, 90:553565.
Simar, L. and Wilson, P.W. (1998). Sensitivity analysis of efficiency scores: how to bootstrap in nonparametric models. Management Science,
44(1):4961.
Thanassoulis, E. and Dyson, R. (1992). Estimating preferred target input-output levels using Data Envelopment Analysis. European Journal of
operational research, 56 (1), 80-97.

You might also like