Hakim SabzevariUniversity
Data Envelopment Analysis
Mohmmad Mahdi Sahebi
mahdisahebi@hotmail.com
February 27, 2018
Hakim SabzevariUniversity
Table of Contents I
Introduction
Assessment
Production Function
Curve Fitting
Cobb Douglas
Efficiency
Absolute Efficiency
Relative Efficiency
Data Envelopment Analysis
Introduction
What is DMU?
Numerical Example
Using Graphical Analysis
DEA CCR
M. Sahebi ·DEA ·February 27, 2018 2 / 49
Hakim SabzevariUniversity
Table of Contents II
DEA Variable Returns to Scale
M. Sahebi ·DEA ·February 27, 2018 3 / 49
Hakim SabzevariUniversity
Introduction Assessment
Assessment
Introduction
How do you manage profitability of a network of hundreds or
thousands of bank branches disbursed over several states and
countries?
M. Sahebi ·DEA ·February 27, 2018 4 / 49
Hakim SabzevariUniversity
Introduction Assessment
Assessment
Introduction
How can a managed care organization manage the quality and costs
of the thousands of physicians providing health services to millions
of plan members?
M. Sahebi ·DEA ·February 27, 2018 4 / 49
Hakim SabzevariUniversity
Introduction Assessment
Assessment
Introduction
What methods would enable a government to ensure that the
multiple offices serving citizens across a country are operating at low
cost while meeting the required service quality?
M. Sahebi ·DEA ·February 27, 2018 4 / 49
Hakim SabzevariUniversity
Introduction Assessment
Assessment
Introduction
Schools: S1, S2, S3, . . .
University: U1, U2, U3, . . .
University Department D1, D2, D3, . . .
Airports: A1, A2, A3, . . .
Hospitals: H1, H2, H3, . . .
Banks : B1, B2, B3, . . .
Bank Branches: BB1, BB2, BB3, . . .
M. Sahebi ·DEA ·February 27, 2018 5 / 49
Hakim SabzevariUniversity
Introduction Assessment
Assessment
Which school is efficient?
School %Pass %Pass
Final Exam Entrance Exam
A 100% 90%
B 40% 30%
Table: Percentage of passing exams
M. Sahebi ·DEA ·February 27, 2018 6 / 49
Hakim SabzevariUniversity
Introduction Assessment
Assessment
Which airport is efficient?
Avg Service/Day
Airplan1 60
Airplan2 60
Airplan3 20
Airplan4 10
Airplan5 0.5
Table: Flight numbers
M. Sahebi ·DEA ·February 27, 2018 7 / 49
Hakim SabzevariUniversity
Introduction Assessment
The decision making based on the output parameters is not correct.
M. Sahebi ·DEA ·February 27, 2018 8 / 49
Hakim SabzevariUniversity
Introduction Production Function
Production Function
The basic relationship between the factors of production and the
output is referred to as a Production Function.
Q = f (V1, V2, . . . , Vm, U1, U2, . . . , Un)
Where:
Q = output
V1, . . . , Vm = Controllable inputs
U1, . . . , Un = Uncontrollable inputs
M. Sahebi ·DEA ·February 27, 2018 9 / 49
Hakim SabzevariUniversity
Introduction Production Function
Production Function
Why we need production functions?
Choose best Inputs
Maximize Outputs
Benchmarking Decision Making Units
...
M. Sahebi ·DEA ·February 27, 2018 10 / 49
Hakim SabzevariUniversity
Introduction Production Function
Production Function
Curve Fitting
Min L1
Min
n∑
i=1
|yi − αxi − β|
Min L2 (LSS1)
Min
n∑
i=1
(yi − αxi − β)2
Min L∞
Min{ max
i=1,...,n
(|yi − αxi − β|)}
1
Least Square EstimationM. Sahebi ·DEA ·February 27, 2018 11 / 49
Hakim SabzevariUniversity
Introduction Production Function
Production Function
Cobb Douglas
Q = ALβ
Kα
QGeneral = x0Ax1
1 Ax2
2 . . . Axn
n
Where:
Q is total production2
L is Labor input 3
K is Captial input 4
A is total factor productivity
2
The real value of all goods produced is a year
3
The total number of person-hours worked in a year
4
The real value of all machinary, equipment, and building
M. Sahebi ·DEA ·February 27, 2018 12 / 49
Hakim SabzevariUniversity
Introduction Production Function
Figure: Image source:
https://economicpoint.com/production-function/cobb-douglas
M. Sahebi ·DEA ·February 27, 2018 13 / 49
Hakim SabzevariUniversity
Introduction Efficiency
Efficiency
Figure: Image source: https://www.tapinfluence.com
M. Sahebi ·DEA ·February 27, 2018 14 / 49
Hakim SabzevariUniversity
Introduction Efficiency
Efficiency
Absolute Efficiency
EfficiencyAbsolut =
Y
Y∗
M. Sahebi ·DEA ·February 27, 2018 15 / 49
Hakim SabzevariUniversity
Introduction Efficiency
Absolute Efficiency
Disadvantages/Advantages
Disadvantages
Global index not exist for any type of decision making units 5
In developing countries similar to Iran, calculating the Absolute
Efficiency is not beneficial
Advantage
Benchmark with global indexes is good when we want to
introduce the future path
5
DMU
M. Sahebi ·DEA ·February 27, 2018 16 / 49
Hakim SabzevariUniversity
Introduction Efficiency
Efficiency
Relative Efficiency
EfficiencyRelativek =
Yk
Xk
max{
Yj
Xj
, j = 1, 2, . . . , n}
M. Sahebi ·DEA ·February 27, 2018 17 / 49
Hakim SabzevariUniversity
Introduction Efficiency
Frontier Analysis
Stochastic Frontier Analysis
Data Envelopment Analysis6
6
DEA
M. Sahebi ·DEA ·February 27, 2018 18 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Introduction
Data Envelopment Analysis
Data envelopment analysis (DEA) is a data oriented, no-parametric
method to evaluate relative efficiency. based on pre-selected iputs
and outputs.7 DEA is an increasingly popular management tool.
DEA can handle multiple input and multiple output models.
It doesn’t require an assumption of a functional form relating
inputs to outputs.
Inputs and outputs can have very different units. For example,
X1 could be in units of lives saved and X2 could be in units of
dollars without requiring an a priori tradeoff between the two.
7
Journal ot the Operat, ioiis Itesearch, Society of Japan 2009, VoL 52,
No. 2,163-173
M. Sahebi ·DEA ·February 27, 2018 19 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis What is DMU?
Data Envelopment Analysis
DMU
x1
x2
x3
xm
DMUj
y1
y2
y3
yn
DMU = Decision Making Unit(s)
M. Sahebi ·DEA ·February 27, 2018 20 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
Data Envelopment Analysis
Numerical Example
Personal Number
Branch Transactions of Staff
Tehran 125 18
Sabzevar 44 16
Tabriz 80 17
Mashhad 23 11
Table: Bank Branches
How can we compare these branches and measure their performance
using this data?
M. Sahebi ·DEA ·February 27, 2018 21 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
Data Envelopment Analysis
Ratio
Personal Number Personal transactions
Branch Transactions of Staff per staff member
Tehran 125 18 6.94
Sabzevar 44 16 2.75
Tabriz 80 17 4.71
Mashhad 23 11 2.09
M. Sahebi ·DEA ·February 27, 2018 22 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
0 1 2 3 4 5 6 7
Tehran 6.94
Sabzevar 2.75
Tabriz 4.71
Mashhad 2.09
M. Sahebi ·DEA ·February 27, 2018 23 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
Data Envelopment Analysis
Relative Efficiency
Personal transactions
Branch per staff member Relative Efficiency
Tehran 6.94 100(6.94/6.94) = 100%
Sabzevar 2.75 100(2.75/6.94) = 40%
Tabriz 4.71 100(4.71/6.94) = 68%
Mashhad 2.09 100(2.09/6.94) = 30%
M. Sahebi ·DEA ·February 27, 2018 24 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
Data Envelopment Analysis
Extending the example
Personal Business Number of
Branch transactions transactions staff
Tehran 125 50 18
Sabzevar 44 20 16
Tabriz 80 55 17
Mashhad 23 12 11
M. Sahebi ·DEA ·February 27, 2018 25 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
Data Envelopment Analysis
Extending the example Ratio
Personal transaction Business transactions
Branch per staff per staff
Tehran 6.94 2.78
Sabzevar 2.75 1.25
Tabriz 4.71 3.24
Mashhad 2.09 1.09
M. Sahebi ·DEA ·February 27, 2018 26 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Numerical Example
Data Envelopment Analysis
Extending the example Ratio
Personal transaction Business transactions
Branch per staff per staff
Tehran 6.94 2.78
Sabzevar 2.75 1.25
Tabriz 4.71 3.24
Mashhad 2.09 1.09
A 1.23 2.92
B 4.43 2.23
C 3.32 2.81
D 3.70 2.68
E 3.34 2.96
M. Sahebi ·DEA ·February 27, 2018 27 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis Using Graphical Analysis
Business transactions per staff
Tehran
Tabriz
Sabzevar
Mashhad
0 1 2 3 4 5
0
1
2
3
4
5
6
7
M. Sahebi ·DEA ·February 27, 2018 28 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA CCR
Data Envelopment Analysis
TECCR:
Min θ
Subject to:
ΣλiXi ≤ θX0 i = 1, . . . , m
ΣλiXi ≥ Y0 i = 1, . . . , m
λ ≥ 0
M. Sahebi ·DEA ·February 27, 2018 29 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Data Envelopment Analysis
TEVariableReturnstoScale:
Min θ
Subject to:
ΣλiXi ≤ θX0 i = 1, . . . , m
ΣλiXi ≥ Y0 i = 1, . . . , m
Σλi = 1
λ ≥ 0
M. Sahebi ·DEA ·February 27, 2018 30 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Data Envelopment Analysis
Example 1
Personal Business Number of
Branch transactions transactions staff
Tehran 125 50 18
Sabzevar 44 20 16
Tabriz 80 55 17
Mashhad 23 12 11
M. Sahebi ·DEA ·February 27, 2018 31 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Problem:
Min θTehran
Subject to:
18L1 + 16L2 + 17L3 + 11L4 ≤ 18θ
125L1 + 44L2 + 80L3 + 23L4 ≥ 125
50L1 + 20L2 + 55L3 + 12L4 ≥ 50
L1 , L2 , L3 , L4 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 32 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
M. Sahebi ·DEA ·February 27, 2018 33 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Problem:
Min θSabzevar
Subject to:
18L1 + 16L2 + 17L3 + 11L4 ≤ 16θ
125L1 + 44L2 + 80L3 + 23L4 ≥ 44
50L1 + 20L2 + 55L3 + 12L4 ≥ 20
L1 , L2 , L3 , L4 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 34 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
M. Sahebi ·DEA ·February 27, 2018 35 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Problem:
Min θTabriz
Subject to:
18L1 + 16L2 + 17L3 + 11L4 ≤ 17θ
125L1 + 44L2 + 80L3 + 23L4 ≥ 80
50L1 + 20L2 + 55L3 + 12L4 ≥ 55
L1 , L2 , L3 , L4 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 36 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
M. Sahebi ·DEA ·February 27, 2018 37 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Note that DMUs Tehran and Tabriz are overall efficient and DMUs
Sabzevar and Mahshad are ineficient with an efficiency rating of
xxxxxxx.
Efficient level of InputsSabzevar:
0.285217391
[
18
]
+ 0.104347826
[
17
]
=
[
6.907826087
]
Efficient levels of OutputsSabzevar:
0.285217391
[
125
50
]
+ 0.104347826
[
80
55
]
=
[
44
20
]
M. Sahebi ·DEA ·February 27, 2018 38 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Tehran
Tabriz
Sabzevar
Mashhad
0 1 2 3 4 5
0
1
2
3
4
5
6
7
SabzevarEfficient
M. Sahebi ·DEA ·February 27, 2018 39 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Problem:
Min θMashhad
Subject to:
18L1 + 16L2 + 17L3 + 11L4 ≤ 11θ
125L1 + 44L2 + 80L3 + 23L4 ≥ 23
50L1 + 20L2 + 55L3 + 12L4 ≥ 12
L1 , L2 , L3 , L4 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 40 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
M. Sahebi ·DEA ·February 27, 2018 41 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Data Envelopment Analysis
Example 2
DMU Inputs Outputs
1 5 14 9 4 16
2 8 15 5 7 10
3 7 12 4 9 13
Table: 3 DMUs with 2 inputs and 3 outputs
M. Sahebi ·DEA ·February 27, 2018 42 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Data Envelopment Analysis
LP Program for DMU1
Problem:
Min θ
Subject to:
5L1 + 8L2 + 7L3 ≤ 5θ
14L1 + 15L2 + 12L3 ≤ 14θ
9L1 + 5L2 + 4L3 ≥ 9
4L1 + 7L2 + 9L3 ≥ 4
16L1 + 10L2 + 13L3 ≥ 16
L1, L2, L3 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 43 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Data Envelopment Analysis
LP Program for DMU2
Problem:
Min θ
Subject to:
5L1 + 8L2 + 7L3 ≤ 8θ
14L1 + 15L2 + 12L3 ≤ 15θ
9L1 + 5L2 + 4L3 ≥ 5
4L1 + 7L2 + 9L3 ≥ 7
16L1 + 10L2 + 13L3 ≥ 10
L1, L2, L3 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 44 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Data Envelopment Analysis
LP Program for DMU3
Problem:
Min θ
Subject to:
5L1 + 8L2 + 7L3 ≤ 7θ
14L1 + 15L2 + 12L3 ≤ 12θ
9L1 + 5L2 + 4L3 ≥ 4
4L1 + 7L2 + 9L3 ≥ 9
16L1 + 10L2 + 13L3 ≥ 13
L1, L2, L3 ≥ 0
M. Sahebi ·DEA ·February 27, 2018 45 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Figure: DMU1 LP Solved
M. Sahebi ·DEA ·February 27, 2018 46 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Figure: DMU2 LP Solved
M. Sahebi ·DEA ·February 27, 2018 46 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Figure: DMU3 LP Solved
M. Sahebi ·DEA ·February 27, 2018 46 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
Note that DMUs 1 and 3 are overall efficient and DMU 2 is ineficient
with an efficiency rating of 0.733333.
Efficient levels of Inputs:
0.261538
[
5
14
]
+ 0.661538
[
7
12
]
=
[
5.935
11.6
]
Efficient levels of Outputs:
0.261538


9
4
16

 + 0.661538


4
9
13

 =


5
7
12.785


M. Sahebi ·DEA ·February 27, 2018 47 / 49
Hakim SabzevariUniversity
Data Envelopment Analysis DEA Variable Returns to Scale
@articlemurillo2004economic, title=Economic efficiency and frontier
techniques, author=Murillo-Zamorano, Luis R, journal=Journal of
Economic surveys, volume=18, number=1, pages=33–77, year=2004,
publisher=Wiley Online Library
M. Sahebi ·DEA ·February 27, 2018 48 / 49
Hakim SabzevariUniversity
Thank you for your attention!

Data envelopment analysis

  • 1.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Mohmmad Mahdi Sahebi mahdisahebi@hotmail.com February 27, 2018
  • 2.
    Hakim SabzevariUniversity Table ofContents I Introduction Assessment Production Function Curve Fitting Cobb Douglas Efficiency Absolute Efficiency Relative Efficiency Data Envelopment Analysis Introduction What is DMU? Numerical Example Using Graphical Analysis DEA CCR M. Sahebi ·DEA ·February 27, 2018 2 / 49
  • 3.
    Hakim SabzevariUniversity Table ofContents II DEA Variable Returns to Scale M. Sahebi ·DEA ·February 27, 2018 3 / 49
  • 4.
    Hakim SabzevariUniversity Introduction Assessment Assessment Introduction Howdo you manage profitability of a network of hundreds or thousands of bank branches disbursed over several states and countries? M. Sahebi ·DEA ·February 27, 2018 4 / 49
  • 5.
    Hakim SabzevariUniversity Introduction Assessment Assessment Introduction Howcan a managed care organization manage the quality and costs of the thousands of physicians providing health services to millions of plan members? M. Sahebi ·DEA ·February 27, 2018 4 / 49
  • 6.
    Hakim SabzevariUniversity Introduction Assessment Assessment Introduction Whatmethods would enable a government to ensure that the multiple offices serving citizens across a country are operating at low cost while meeting the required service quality? M. Sahebi ·DEA ·February 27, 2018 4 / 49
  • 7.
    Hakim SabzevariUniversity Introduction Assessment Assessment Introduction Schools:S1, S2, S3, . . . University: U1, U2, U3, . . . University Department D1, D2, D3, . . . Airports: A1, A2, A3, . . . Hospitals: H1, H2, H3, . . . Banks : B1, B2, B3, . . . Bank Branches: BB1, BB2, BB3, . . . M. Sahebi ·DEA ·February 27, 2018 5 / 49
  • 8.
    Hakim SabzevariUniversity Introduction Assessment Assessment Whichschool is efficient? School %Pass %Pass Final Exam Entrance Exam A 100% 90% B 40% 30% Table: Percentage of passing exams M. Sahebi ·DEA ·February 27, 2018 6 / 49
  • 9.
    Hakim SabzevariUniversity Introduction Assessment Assessment Whichairport is efficient? Avg Service/Day Airplan1 60 Airplan2 60 Airplan3 20 Airplan4 10 Airplan5 0.5 Table: Flight numbers M. Sahebi ·DEA ·February 27, 2018 7 / 49
  • 10.
    Hakim SabzevariUniversity Introduction Assessment Thedecision making based on the output parameters is not correct. M. Sahebi ·DEA ·February 27, 2018 8 / 49
  • 11.
    Hakim SabzevariUniversity Introduction ProductionFunction Production Function The basic relationship between the factors of production and the output is referred to as a Production Function. Q = f (V1, V2, . . . , Vm, U1, U2, . . . , Un) Where: Q = output V1, . . . , Vm = Controllable inputs U1, . . . , Un = Uncontrollable inputs M. Sahebi ·DEA ·February 27, 2018 9 / 49
  • 12.
    Hakim SabzevariUniversity Introduction ProductionFunction Production Function Why we need production functions? Choose best Inputs Maximize Outputs Benchmarking Decision Making Units ... M. Sahebi ·DEA ·February 27, 2018 10 / 49
  • 13.
    Hakim SabzevariUniversity Introduction ProductionFunction Production Function Curve Fitting Min L1 Min n∑ i=1 |yi − αxi − β| Min L2 (LSS1) Min n∑ i=1 (yi − αxi − β)2 Min L∞ Min{ max i=1,...,n (|yi − αxi − β|)} 1 Least Square EstimationM. Sahebi ·DEA ·February 27, 2018 11 / 49
  • 14.
    Hakim SabzevariUniversity Introduction ProductionFunction Production Function Cobb Douglas Q = ALβ Kα QGeneral = x0Ax1 1 Ax2 2 . . . Axn n Where: Q is total production2 L is Labor input 3 K is Captial input 4 A is total factor productivity 2 The real value of all goods produced is a year 3 The total number of person-hours worked in a year 4 The real value of all machinary, equipment, and building M. Sahebi ·DEA ·February 27, 2018 12 / 49
  • 15.
    Hakim SabzevariUniversity Introduction ProductionFunction Figure: Image source: https://economicpoint.com/production-function/cobb-douglas M. Sahebi ·DEA ·February 27, 2018 13 / 49
  • 16.
    Hakim SabzevariUniversity Introduction Efficiency Efficiency Figure:Image source: https://www.tapinfluence.com M. Sahebi ·DEA ·February 27, 2018 14 / 49
  • 17.
    Hakim SabzevariUniversity Introduction Efficiency Efficiency AbsoluteEfficiency EfficiencyAbsolut = Y Y∗ M. Sahebi ·DEA ·February 27, 2018 15 / 49
  • 18.
    Hakim SabzevariUniversity Introduction Efficiency AbsoluteEfficiency Disadvantages/Advantages Disadvantages Global index not exist for any type of decision making units 5 In developing countries similar to Iran, calculating the Absolute Efficiency is not beneficial Advantage Benchmark with global indexes is good when we want to introduce the future path 5 DMU M. Sahebi ·DEA ·February 27, 2018 16 / 49
  • 19.
    Hakim SabzevariUniversity Introduction Efficiency Efficiency RelativeEfficiency EfficiencyRelativek = Yk Xk max{ Yj Xj , j = 1, 2, . . . , n} M. Sahebi ·DEA ·February 27, 2018 17 / 49
  • 20.
    Hakim SabzevariUniversity Introduction Efficiency FrontierAnalysis Stochastic Frontier Analysis Data Envelopment Analysis6 6 DEA M. Sahebi ·DEA ·February 27, 2018 18 / 49
  • 21.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Introduction Data Envelopment Analysis Data envelopment analysis (DEA) is a data oriented, no-parametric method to evaluate relative efficiency. based on pre-selected iputs and outputs.7 DEA is an increasingly popular management tool. DEA can handle multiple input and multiple output models. It doesn’t require an assumption of a functional form relating inputs to outputs. Inputs and outputs can have very different units. For example, X1 could be in units of lives saved and X2 could be in units of dollars without requiring an a priori tradeoff between the two. 7 Journal ot the Operat, ioiis Itesearch, Society of Japan 2009, VoL 52, No. 2,163-173 M. Sahebi ·DEA ·February 27, 2018 19 / 49
  • 22.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis What is DMU? Data Envelopment Analysis DMU x1 x2 x3 xm DMUj y1 y2 y3 yn DMU = Decision Making Unit(s) M. Sahebi ·DEA ·February 27, 2018 20 / 49
  • 23.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example Data Envelopment Analysis Numerical Example Personal Number Branch Transactions of Staff Tehran 125 18 Sabzevar 44 16 Tabriz 80 17 Mashhad 23 11 Table: Bank Branches How can we compare these branches and measure their performance using this data? M. Sahebi ·DEA ·February 27, 2018 21 / 49
  • 24.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example Data Envelopment Analysis Ratio Personal Number Personal transactions Branch Transactions of Staff per staff member Tehran 125 18 6.94 Sabzevar 44 16 2.75 Tabriz 80 17 4.71 Mashhad 23 11 2.09 M. Sahebi ·DEA ·February 27, 2018 22 / 49
  • 25.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example 0 1 2 3 4 5 6 7 Tehran 6.94 Sabzevar 2.75 Tabriz 4.71 Mashhad 2.09 M. Sahebi ·DEA ·February 27, 2018 23 / 49
  • 26.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example Data Envelopment Analysis Relative Efficiency Personal transactions Branch per staff member Relative Efficiency Tehran 6.94 100(6.94/6.94) = 100% Sabzevar 2.75 100(2.75/6.94) = 40% Tabriz 4.71 100(4.71/6.94) = 68% Mashhad 2.09 100(2.09/6.94) = 30% M. Sahebi ·DEA ·February 27, 2018 24 / 49
  • 27.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example Data Envelopment Analysis Extending the example Personal Business Number of Branch transactions transactions staff Tehran 125 50 18 Sabzevar 44 20 16 Tabriz 80 55 17 Mashhad 23 12 11 M. Sahebi ·DEA ·February 27, 2018 25 / 49
  • 28.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example Data Envelopment Analysis Extending the example Ratio Personal transaction Business transactions Branch per staff per staff Tehran 6.94 2.78 Sabzevar 2.75 1.25 Tabriz 4.71 3.24 Mashhad 2.09 1.09 M. Sahebi ·DEA ·February 27, 2018 26 / 49
  • 29.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Numerical Example Data Envelopment Analysis Extending the example Ratio Personal transaction Business transactions Branch per staff per staff Tehran 6.94 2.78 Sabzevar 2.75 1.25 Tabriz 4.71 3.24 Mashhad 2.09 1.09 A 1.23 2.92 B 4.43 2.23 C 3.32 2.81 D 3.70 2.68 E 3.34 2.96 M. Sahebi ·DEA ·February 27, 2018 27 / 49
  • 30.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis Using Graphical Analysis Business transactions per staff Tehran Tabriz Sabzevar Mashhad 0 1 2 3 4 5 0 1 2 3 4 5 6 7 M. Sahebi ·DEA ·February 27, 2018 28 / 49
  • 31.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA CCR Data Envelopment Analysis TECCR: Min θ Subject to: ΣλiXi ≤ θX0 i = 1, . . . , m ΣλiXi ≥ Y0 i = 1, . . . , m λ ≥ 0 M. Sahebi ·DEA ·February 27, 2018 29 / 49
  • 32.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Data Envelopment Analysis TEVariableReturnstoScale: Min θ Subject to: ΣλiXi ≤ θX0 i = 1, . . . , m ΣλiXi ≥ Y0 i = 1, . . . , m Σλi = 1 λ ≥ 0 M. Sahebi ·DEA ·February 27, 2018 30 / 49
  • 33.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Data Envelopment Analysis Example 1 Personal Business Number of Branch transactions transactions staff Tehran 125 50 18 Sabzevar 44 20 16 Tabriz 80 55 17 Mashhad 23 12 11 M. Sahebi ·DEA ·February 27, 2018 31 / 49
  • 34.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Problem: Min θTehran Subject to: 18L1 + 16L2 + 17L3 + 11L4 ≤ 18θ 125L1 + 44L2 + 80L3 + 23L4 ≥ 125 50L1 + 20L2 + 55L3 + 12L4 ≥ 50 L1 , L2 , L3 , L4 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 32 / 49
  • 35.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale M. Sahebi ·DEA ·February 27, 2018 33 / 49
  • 36.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Problem: Min θSabzevar Subject to: 18L1 + 16L2 + 17L3 + 11L4 ≤ 16θ 125L1 + 44L2 + 80L3 + 23L4 ≥ 44 50L1 + 20L2 + 55L3 + 12L4 ≥ 20 L1 , L2 , L3 , L4 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 34 / 49
  • 37.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale M. Sahebi ·DEA ·February 27, 2018 35 / 49
  • 38.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Problem: Min θTabriz Subject to: 18L1 + 16L2 + 17L3 + 11L4 ≤ 17θ 125L1 + 44L2 + 80L3 + 23L4 ≥ 80 50L1 + 20L2 + 55L3 + 12L4 ≥ 55 L1 , L2 , L3 , L4 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 36 / 49
  • 39.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale M. Sahebi ·DEA ·February 27, 2018 37 / 49
  • 40.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Note that DMUs Tehran and Tabriz are overall efficient and DMUs Sabzevar and Mahshad are ineficient with an efficiency rating of xxxxxxx. Efficient level of InputsSabzevar: 0.285217391 [ 18 ] + 0.104347826 [ 17 ] = [ 6.907826087 ] Efficient levels of OutputsSabzevar: 0.285217391 [ 125 50 ] + 0.104347826 [ 80 55 ] = [ 44 20 ] M. Sahebi ·DEA ·February 27, 2018 38 / 49
  • 41.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Tehran Tabriz Sabzevar Mashhad 0 1 2 3 4 5 0 1 2 3 4 5 6 7 SabzevarEfficient M. Sahebi ·DEA ·February 27, 2018 39 / 49
  • 42.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Problem: Min θMashhad Subject to: 18L1 + 16L2 + 17L3 + 11L4 ≤ 11θ 125L1 + 44L2 + 80L3 + 23L4 ≥ 23 50L1 + 20L2 + 55L3 + 12L4 ≥ 12 L1 , L2 , L3 , L4 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 40 / 49
  • 43.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale M. Sahebi ·DEA ·February 27, 2018 41 / 49
  • 44.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Data Envelopment Analysis Example 2 DMU Inputs Outputs 1 5 14 9 4 16 2 8 15 5 7 10 3 7 12 4 9 13 Table: 3 DMUs with 2 inputs and 3 outputs M. Sahebi ·DEA ·February 27, 2018 42 / 49
  • 45.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Data Envelopment Analysis LP Program for DMU1 Problem: Min θ Subject to: 5L1 + 8L2 + 7L3 ≤ 5θ 14L1 + 15L2 + 12L3 ≤ 14θ 9L1 + 5L2 + 4L3 ≥ 9 4L1 + 7L2 + 9L3 ≥ 4 16L1 + 10L2 + 13L3 ≥ 16 L1, L2, L3 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 43 / 49
  • 46.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Data Envelopment Analysis LP Program for DMU2 Problem: Min θ Subject to: 5L1 + 8L2 + 7L3 ≤ 8θ 14L1 + 15L2 + 12L3 ≤ 15θ 9L1 + 5L2 + 4L3 ≥ 5 4L1 + 7L2 + 9L3 ≥ 7 16L1 + 10L2 + 13L3 ≥ 10 L1, L2, L3 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 44 / 49
  • 47.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Data Envelopment Analysis LP Program for DMU3 Problem: Min θ Subject to: 5L1 + 8L2 + 7L3 ≤ 7θ 14L1 + 15L2 + 12L3 ≤ 12θ 9L1 + 5L2 + 4L3 ≥ 4 4L1 + 7L2 + 9L3 ≥ 9 16L1 + 10L2 + 13L3 ≥ 13 L1, L2, L3 ≥ 0 M. Sahebi ·DEA ·February 27, 2018 45 / 49
  • 48.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Figure: DMU1 LP Solved M. Sahebi ·DEA ·February 27, 2018 46 / 49
  • 49.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Figure: DMU2 LP Solved M. Sahebi ·DEA ·February 27, 2018 46 / 49
  • 50.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Figure: DMU3 LP Solved M. Sahebi ·DEA ·February 27, 2018 46 / 49
  • 51.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale Note that DMUs 1 and 3 are overall efficient and DMU 2 is ineficient with an efficiency rating of 0.733333. Efficient levels of Inputs: 0.261538 [ 5 14 ] + 0.661538 [ 7 12 ] = [ 5.935 11.6 ] Efficient levels of Outputs: 0.261538   9 4 16   + 0.661538   4 9 13   =   5 7 12.785   M. Sahebi ·DEA ·February 27, 2018 47 / 49
  • 52.
    Hakim SabzevariUniversity Data EnvelopmentAnalysis DEA Variable Returns to Scale @articlemurillo2004economic, title=Economic efficiency and frontier techniques, author=Murillo-Zamorano, Luis R, journal=Journal of Economic surveys, volume=18, number=1, pages=33–77, year=2004, publisher=Wiley Online Library M. Sahebi ·DEA ·February 27, 2018 48 / 49
  • 53.