Lecture - 2: Data Envelopment Analysis - Ii
Lecture - 2: Data Envelopment Analysis - Ii
Lecture - 2: Data Envelopment Analysis - Ii
Learning objective
To explain Data Envelopment Analysis for multiple inputs and
multiple outputs case in the form of linear programming
input
and
single
output.
Let
us
see
below
an
illustration.
The owner of a chain of coffee shops Coffees and more wants to evaluate the
efficiency of the shops in one city. The items sold as outputs in his store are in two
categories: Beverages Coffee, Tea and other drinks and Snacks Sandwiches,
Biscuits and so on. The inputs leading to the sale of these items are: Available floor
space in the store and the number of employees. The data regarding inputs and outputs
are given below in Table 11.6.
Table 11.6: Input and output data for Coffees and more
Store Location
Number of
employees
Floor space in
100m2
Units of
Snacks
Units of
Beverages
x1
40
38
50
54
44
110
66
x2
302
262
320
336
316
510
470
y1
200
300
320
360
188
460
440
y2
180
100
110
144
132
180
176
To find the efficiency of each of the units and to determine the relative efficiency, we
need to give weights to each input and to each output. These weights are the
coefficients for each input and output variable. The coefficients pertaining to output
variables measure the relative decrease in efficiency with each unit reduction of output
variable. The coefficients pertaining to input variables measure the relative increase in
efficiency with each unit reduction of input variable. These weights can be either
provided by the user or can be determined optimally by using some optimization
technique like DEA.
(1200)+(2180)
(140)+(2302)
The efficiency of other units are determined in similar fashion which are presented in
Table 11.7. Relative efficiency of each unit can be determined using the ratio of
efficiency of that unit and efficiency of most efficient unit as shown in Table 11.7.
1.47
1.33
1.14
1.31
1.09
0.94
1.21
0.9
0.77
0.89
0.74
0.64
0.82
x
X = 21
x
m1
x12
x22
xm 2
x1n
x2 n
xmn
y
Y = 21
y
s1
y12
y22
ys 2
y1n
y2 n
y sn
Now, the efficiency of each of the DMU j (Decision Making Unit the coffee store in
this case), with j = 1,2,.n is to be measured so that we can determine relative
efficiency and identify inefficient units. If there are n DMUs, we need n
optimizations, one for each DMU using following notations.
DMU o : DMU which will be evaluated in a particular trial, with o=1,2,,n. This
notation will be used for the DMU in the objective function. The same DMU in the
constraints will follow the notation DMU j as defined next.
DMU j : DMUs with j= 1,2,..n
v i : Coefficient for input i, with i = 1,2,..m
u r : Coefficient for output r, with r = 1,2,..s.
The DEA model can be presented to maximize the efficiency of DMU o , 0 ,by writing
the objective function of oth DMU as given below.
Max o =
The model will be subjected to the constraints; (1) efficiency of all other DMUs will be
less than or equal to 1 and (2) non negativity constraints. The model will take fractional
programming form as given below.
Objective function:
u y + u y + ....... + u s yso
(for oth DMU)
o = 1 1o 2 2o
Max
v1 x1o + v1 x1o + ........ + vm xmo
Subject to constraints:
u1 y1 j + u2 y2 j + ....... + u s ysj
v1 x1 j + v1 x1 j + ........ + vm xmj
1 ( j = 1, 2, ......., n)
v1 , v2 ,......., vm 0
u1 , u2 ,......., u s 0
The fractional programming model can be converted into a linear programming model.
This is done by scaling each of the inputs to 1 and rewriting the constraints as
mentioned below.
Objective function
Max o = u1 y1o + u2 y2 o + ....... + u s yso
Subject to
v1 x1o + v1 x1o + ........ + vm xmo = 1
u1 y1 j + u2 y2 j + ..... + u s ysj v1 x1 j + v1 x1 j + ..... + vm xmj
( j = 1, 2, ......., n)
v1 , v2 ,......., vm 0
u1 , u2 ,......., u s 0
The presented model finds the best set of weights or coefficients pertaining to each
input and output variable while maximum rating of efficiency is assigned to oth DMU.
After solving linear programming model for DMU o , the oth DMU will be efficient only
if the model results in:
Super labs are in Research and Development of electronic goods. The founders, in the
year 2001 decided to set-up their own company to produce consumer electronic goods
that use the state-of-the art technologies that have been developed by them. The
company has set up business in North America and quickly moved to Asia, Europe,
Africa and South America. The main products of the company are its impressive range
of Smart phones, Tablet PCs and Laptops.
At the beginning of the year 2011, the CEO- global operations of the company
quickly pulled reports of Inputs and Outputs that are being used and produced at various
regions in which company operates.
He arrived at the following figures:
INPUTS ($ in millions)
DMU
Product
Development
Marketing
N.America
14
16
Europe
10
18
Asia
16
10
Africa
12
S.America
15
10
14
The data tells him how much each of the regions consume for the two main activities,
namely Product development and Marketing, and how many units of its products are
being sold. However, he is unable to figure out, which are the best performing units and
how much of increase or decrease the Regional directors are to be advised to make on
investments.
One of his analysts recommends the use of Data Envelopment Analysis to analyze the
performance of the different units and then make a conclusion. The analyst also
mentioned that the tool can be best used as the inputs and outputs to the various
Decision Making Units are the same units ($ spent of product development, marketing
and number of units of each product sold).
The analyst quickly arrived at the following formulation:
For each of the inputs, assign weights u1 , u2 and assign weights v1 , v2 and v3 for the
outputs. Therefore the formulation for the five DMUs would be as follows:
For Unit 1 (N.America):
Objective function
Maximize 9v1 + 4v2 + 16v3
Subject to the constraints:
5u1 + 14u2 =
1
10u1 + 18u2 =
1
(9v1 + 4v2 + 16v3 ) (5u1 + 14u2 ) 0
(8v1 + 2v2 + 9v3 ) (10u1 + 18u2 ) 0
(9v1 + 4v2 + 10v3 ) (9u1 + 16u2 ) 0
(6v1 + 1v2 + 8v3 ) (7u1 + 12u2 ) 0
(9v1 + 4v2 + 16v3 ) (5u1 + 14u2 ) 0
(10v1 + 4v2 + 14v3 ) (9u1 + 16u2 ) 0
u1 , u2 0
v1 , v2 and v3 0
9u1 + 16u2 =
1
7u1 + 12u2 =
1
(9v1 + 4v2 + 16v3 ) (5u1 + 14u2 ) 0
(8v1 + 2v2 + 9v3 ) (10u1 + 18u2 ) 0
(9v1 + 4v2 + 10v3 ) (9u1 + 16u2 ) 0
(6v1 + 1v2 + 8v3 ) (7u1 + 12u2 ) 0
(9v1 + 4v2 + 16v3 ) (5u1 + 14u2 ) 0
(10v1 + 4v2 + 14v3 ) (9u1 + 16u2 ) 0
u1 , u2 0
v1 , v2 and v3 0
For Unit 5 (S.America):
Objective function
Maximize 8v1 + 2v2 + 9v3
Subject to the constraints:
10u1 + 18u2 =
1
u1 , u2 0
v1 , v2 and v3 0
The above presented models can be solved using any optimization software or using
excel. On solving the above linear programming models in Excel for all the units, the
following efficiencies for the DMUs were found.
DMU
Efficiency
N.America
Europe
0.67
Asia
0.875
Africa
0.75
S.America
Now, CEO likes to compare the efficiency of the unit in Europe (inefficient unit) with
other efficient units. It is found that the shadow prices of the efficient units that are N.
America and S. America to be 0.157 and 0.657 respectively. So, N. America and S.
America forms the reference set for Europe. The linear combination of a particular
input for two efficient units using shadow prices as multiplier, will give the point on the
efficient frontier where the inefficient unit can be projected. That is Europe should
brought down the product development expenses from 10 to 6.698 (5 x 0.157 + 9 x
0.657).
Inputs and
North America
Outputs
South America
Values
Weights
Total 1
Values
Weights
Total 2
Sum of Total
1 and 2
Europe
Excess used
by the unit
Europe
Product
Development
0.157
0.785
0.657
5.913
6.698
10
3.302
Marketing
14
0.157
2.198
15
0.657
9.855
12.053
18
5.947
Smart Phone
0.157
1.413
10
0.657
6.57
7.983
0.017
Tablets
0.157
0.628
0.657
2.628
3.256
-1.256
Laptops
16
0.157
2.512
14
0.657
9.198
11.71
-2.71
Thus on comparing the inputs and outputs of Europe with that of N and S America,
Europe to reduce Product development and Marketing costs by 3.302 and 5.947 (000s
of dollar value) and increase sales of Tablets and Laptops by 1.256 and 2.71 units
respectively to reach the values of the efficient units. The sale of Smart phones is
however to the level of the efficient units N and S America. Thus the changes to be
made to the other inefficient DMUs can be analyzed.