Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 18

Data Envelopment

Analysis via Linear


Programming
Frameworks and Applications

Suri Gurumurthi, Ph.D. (2020) 1


Extending to multiple outputs ...
Ex: Consider 8 bank branches in a city working for the same 160 hrs. in a month.
Each serves business and individual customers.

Which ones are most “productive”?


Branch # Business Customers # Individual Customers
1 48 68
2 12 80
3 35 76
4 31 71
5 20 70
6 20 105
7 36 53
8 15 65

Note: There is some “efficient” trade-off between the number of individual


versus business customers that any one bank can achieve, but what is it?
Extending to multiple outputs ...
Ex: Consider 8 bank branches in a city working for the same 160 hrs. in a month.
Each serves business and individual customers.

Which ones are most “productive”?


Branches 1 & 6 dominate the others!
# Individual Customers # Business vs Individual
120 Customers
100
6

80 2 3
5 4 1
60
8
7
40

20

0
0 10 20 30 50 60

40

# Business Customers
Note: There is some “efficient” trade-off between the number of individual
versus business customers that any one bank can achieve, but what is it?
Extending to multiple outputs ...
Branches 2, 3, 4, 5, 7, 8 are less than 100% “efficient”

73.4% distance to frontier


Efficiency Score (#5)=73.4% Branches 1 & 6 dominate the others!
# Business vs Individual
120 Customers
100
6

80 2
COMPOSITE BRANCH or ENTITY
3
5 4 1
60
8
7
40

20

0
0 10 20 30 40 50 60

# Individual Customers

“Pareto-Koopman efficiency” along the frontier - cannot increase an output (or decrease
an input) without compensating decrease in other outputs (or increase in other inputs).
Extending to multiple outputs ...
“Nearest” efficient points define a reference set and a linear combination of the
reference set inputs and outputs defines a hypothetical composite unit (HCU)
Reference set for #5 = {1,6}

# Business vs Individual Customers


120
6
73.4% distance to frontier 100

Efficiency Score=73.4% 80 2 3
5 4 1
60
8
7
40

20

0
0 10 20 30 40 50 60

# Individual Customers

“Pareto-Koopman efficiency” along the frontier - cannot increase an output (or decrease
an input) without compensating decrease in other outputs (or increase in other inputs).
DEA and Efficient
Frontiers
• DEA uses an efficient frontier to define multiple I/O productivity
• Frontier defines the (observed) efficient trade-off among inputs
and outputs within a set of DMUs.
• Relative distance to the frontier defines efficiency
• “Nearest point” on frontier defines an efficient comparison unit
(hypothetical comparison unit (HCU))
• Differences in inputs and output between DMU and HCU define
productivity “gaps” (improvement potential)

How do we do this analysis systematically?


Productivity or Efficiency
Measure Productivity OR Ef/iciency
$! " # 𝑤!𝑌!)
𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦 𝐸 = 𝑊𝑒𝑖𝑔ℎ𝑡𝑒𝑑 𝑆𝑢𝑚 𝑜𝑓 𝑂𝑢𝑡𝑝𝑢𝑡𝑠 ( ∑
𝑊𝑒𝑖𝑔ℎ𝑡𝑒𝑑 𝑆𝑢𝑚 𝑜𝑓 𝐼𝑛𝑝𝑢𝑡𝑠(∑&%" 𝑣% 𝑋
#
%)

Inputs Outputs

equipment #type A cust.


facility space Technology #type B
+
server Decision Making cust. quality
labor index
mgmt. labor $ oper. profit
Linear Programming
Formulation:
Data
K # operting units (DMUs) k  1,...,
K N # inputs i  1,..., N
M # outputs j  1,..., M
observed level of output j from DMU
O jk
k
I ik
Model variables
observed level of input i from DMU k
vi weight on input i
u weight on output j
j efficiency of DMU (0 -
E 100%) M

u O
k
j
Ek  1N j jk

v I
i1
i ik
To evaluate a give unit, e, choose nonnegative weights to
solve ...
max
Ee s.t.
Ek �
100, k
 1,...,
K$
max Q 𝑢! 𝑂! '
Which can be formulated! " #
𝑠. 𝑡.
& Normalize weighted input of
Q 𝑣% 𝐼%' = 1 e to one
%"#
$ &

Q 𝑢! 𝑂! ( ≤ Q 𝑣% 𝐼% 𝑘=
!"# %"# 1, . . . , 𝐾
( ,
𝑢 ! ≥ 0, 𝑗=
𝑣% ≥ 1, . . . , 𝑀
0, 𝑖=
1, . . . , 𝑁
Using the results: Efficiency Vs. Profit
Matrix
High Profit
Under-performing Best practice
potential leaders comparison group

Low High
Eff. Eff.

Candidates for Under-performing


closure possibly profitable

Low Profit
DEA
Design
• “Ambivalence” about inputs and outputs - all should be
relatively important!
• “Approximate similarity” among DMUs
• objectives
• technology
• Provides relative efficiency only
• choice of units to include matters
• inclusion of “global leader” unit
may be desirable
• Experimenting with different I/O combinations may be
necessary
DEA
• Summary
Addresses fundamental productivity • Role of DEA
measurement problems due to ... • “data mining” to generate hypotheses
• complexity of service outputs
• evaluation/measurement
• variability in service outputs
• benchmarking to identify “best practice”
• Takes advantage of service operating units
environment
• large numbers of similar facilities • Caveats
• diversity of • “black box” - No information on root
practices/management/environment causes of inefficiency
• Provides useful information • Be aware of assumptions (e.g.
• objective measures of productivity linearity)
• reference set of comparable units • Can be sensitive to selection of
• excess use of inputs measure inputs/outputs
• returns to scale measure
Computing Efficiency of
DMUs
TO COMPUTE EFFICIENCY OF BRANCH # 1
MAXIMIZE Efficiency of #1 = (48 x Weight_Bus + 68 x Weight_Individual)/(160 x Weight_Labor)
DECISION VARIABLES: Weight_Bus, Weight_Individual, Weight_Labor
SUBJECT TO CONSTRAINTS: Branch # Business # Individual
Customers
• 160 x Weight_Labor = 1 (Normalize the denominator to calculate efficiency) Customers

• 0 <= (48Weight_Bus + 68Weight_Individual)/(160Weight_Labor) <= 1 1 48 68


• 0 <= (12Weight_Bus + 80Weight_Individual)/(160Weight_Labor) <= 1 2 12 80
• 0 <= (35Weight_Bus + 76Weight_Individual)/(160Weight_Labor) <= 1 3 35 76
• 0 <= (31Weight_Bus + 71Weight_Individual)/(160Weight_Labor) <= 1 4 31 71

• 0 <= (20Weight_Bus + 70Weight_Individual)/(160Weight_Labor) <= 1 5 20 70

• 0 <= (20Weight_Bus + 105Weight_Individual)/(160Weight_Labor) <= 1 6 20 105


7 36 53
• 0 <= (36Weight_Bus + 53Weight_Individual)/(160Weight_Labor) <= 1
8 15 65
• 0 <= (15Weight_Bus + 65Weight_Individual)/(160Weight_Labor) <= 1
• Weight_Bus >= 0; Weight_Individual >= 0; Weight_Labor >= 0

Suri Gurumurthi, Ph.D. (2020) 13


Computing Efficiency of DMUs:
Translating to a Linear Programming
Problem
TO COMPUTE EFFICIENCY OF BRANCH # 1
MAXIMIZE Efficiency of #1 = 48 x Weight_Bus + 68 x Weight_Individual
DECISION VARIABLES: Weight_Bus, Weight_Individual, Weight_Labor
SUBJECT TO CONSTRAINTS: Branch # Business # Individual
Customers
• 160 x Weight_Labor = 1 (Normalize the denominator to calculate efficiency) Customers

• 48Weight_Bus + 68Weight_Individual <= 160Weight_Labor 1 48 68


• 12Weight_Bus + 80Weight_Individual <= 160Weight_Labor 2 12 80
• 35Weight_Bus + 76Weight_Individual <= 160Weight_Labor 3 35 76
• 31Weight_Bus + 71Weight_Individual <= 160Weight_Labor 4 31 71

• 20Weight_Bus + 70Weight_Individual <= 160Weight_Labor 5 20 70

• 20Weight_Bus + 105Weight_Individual <= 160Weight_Labor 6 20 105


7 36 53
• 36Weight_Bus + 53Weight_Individual <= 160Weight_Labor
8 15 65
• 15Weight_Bus + 65Weight_Individual <= 160Weight_Labor
• Weight_Bus >= 0; Weight_Individual >= 0; Weight_Labor >= 0

Suri Gurumurthi, Ph.D. (2020) 14


Computing Efficiency of DMUs:
Translating to a Linear Programming
Problem
TO COMPUTE EFFICIENCY OF BRANCH # 2
MAXIMIZE Efficiency of #2 = 12Weight_Bus + 80Weight_Individual
DECISION VARIABLES: Weight_Bus, Weight_Individual, Weight_Labor
SUBJECT TO CONSTRAINTS: Branch # Business # Individual
Customers
• 160 x Weight_Labor = 1 (Normalize the denominator to calculate efficiency) Customers

• 48Weight_Bus + 68Weight_Individual <= 160Weight_Labor 1 48 68


• 12Weight_Bus + 80Weight_Individual <= 160Weight_Labor 2 12 80
• 35Weight_Bus + 76Weight_Individual <= 160Weight_Labor 3 35 76
• 31Weight_Bus + 71Weight_Individual <= 160Weight_Labor 4 31 71

• 20Weight_Bus + 70Weight_Individual <= 160Weight_Labor 5 20 70

• 20Weight_Bus + 105Weight_Individual <= 160Weight_Labor 6 20 105


7 36 53
• 36Weight_Bus + 53Weight_Individual <= 160Weight_Labor
8 15 65
• 15Weight_Bus + 65Weight_Individual <= 160Weight_Labor
• Weight_Bus >= 0; Weight_Individual >= 0; Weight_Labor >= 0

Suri Gurumurthi, Ph.D. (2020) 15


Computing Efficiency of DMUs:
Translating to a Linear Programming
Problem
Branch # Business # Individual Branch Efficiency
Customers
Customers
1 48 68 1
2 12 80 0.761905
3 35 76 0.930163
4 31 71 0.851902
5 20 70 0.733696
6 20 105 1
7 36 53 0.765217
8 15 65 0.64538
Suri Gurumurthi, Ph.D. (2020) 16
Computing Efficiency of DMUs:
Multiple Ouputs with Different Input
Levels
TO COMPUTE EFFICIENCY OF BRANCH # 1
MAXIMIZE Efficiency of #1 = 48 x Weight_Bus + 68 x Weight_Individual
Branch # Business # Individual # Input
DECISION VARIABLES: Weight_Bus, Weight_Individual, Weight_Labor Customers Hours
Customers
SUBJECT TO CONSTRAINTS:
1 48 68 198
• 198 x Weight_Labor = 1 (Normalize the denominator to calculate efficiency)
2 12 80 192
• 48Weight_Bus + 68Weight_Individual <= 198Weight_Labor
• 12Weight_Bus + 80Weight_Individual <= 192Weight_Labor 3 35 76 135

• 35Weight_Bus + 76Weight_Individual <= 135Weight_Labor 4 31 71 185


• 31Weight_Bus + 71Weight_Individual <= 185Weight_Labor 5 20 70 188
• 20Weight_Bus + 70Weight_Individual <= 188Weight_Labor 6 20 105 143
• 20Weight_Bus + 105Weight_Individual <= 143Weight_Labor 7 36 53 135
• 36Weight_Bus + 53Weight_Individual <= 135Weight_Labor 8 15 65 163
• 15Weight_Bus + 65Weight_Individual <= 163Weight_Labor
• Weight_Bus >= 0; Weight_Individual >= 0; Weight_Labor >= 0

Suri Gurumurthi, Ph.D. (2020) 17


Computing Efficiency of DMUs:
Multiple Outputs with Different Input
Levels Branch # Business # Individual # Input Branch
Customers Hours Efficienc
Customers y
1 48 68 198 0.909091
2 12 80 192 0.56746
3 35 76 135 1
4 31 71 185 0.667642
5 20 70 188 0.56151
6 20 105 143 1
7 36 53 135 1
8 15 65 163 0.567748

Suri Gurumurthi, Ph.D. (2020) 18

You might also like