Data Envelopment Analysis

(DEA)
Which Unit is most productive?

DMU labor hrs. #cust.

1 100 150
2 75 140
3 120 160
4 100 140
5 40 50

DMU = decision making unit

Extending to multiple outputs ...
Ex: Consider 8 M.D.’s working at a hospital for the same 160 hrs. in
a month. Each performs exams and surgeries.

Which ones are most “productive”?

Doctor #Exams #Surgeries

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

surgeries and exams that any one M.D. can do in a month, but
what is it?
 DEA (Charnes, Coopers & Rhodes ‘78)
A multiple-input, multiple-output productivity measurement tool

Basic intuition DMU labor hrs. #cust. #cust/hr.

(DMU = decision making unit) 1 100 150 1.50
2 75 140 1.87
3 120 160 1.33
#cust.
4 100 140 1.40
5 40 50 1.25

200 7
1.8 x
e =
p x
s lo
x DMU’s 1,3,4,5 are
x dominated by DMU 2.
100
x

labor hrs.
50 100
 Efficient M.D.’s: These two
Scatter plot of outputs: M.D.’s (#1 and #6) define the
most efficient trade-off between
the two outputs.
120 #6
effi
100 cien
t fron
tier
80 #1
#Surgeries

60

40

20 These points are dominated

by #1 and #6.
0
0 10 20 30 40 50 60
#Exams
How bad are the inefficient M.D.s and where
are the gaps?
120

100

80 #5
#Surgeries

Performance “gap”
60

40
73.4% of distance to frontier
20

0
0 10 20 30 40 50 60
#Exams

Efficiency score = 73.4%

“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
120 #6 is {1,6}

100

80 #5
#Surgeries

#1
60

40
HCU
20

0
0 10 20 30 40 50 60
#Exams
DEA summary so far:
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
Conceptually ...

Productivity = Outputs
Inputs
Reality: more complex ...

Inputs Outputs
equipment #type A cust.
facility space Technology
#type B cust.
+
server labor Decision Making quality index
mgmt. labor oper. profit
Operating Units Differ
● Mix of customers served
● Availability and cost of inputs
● Facility configuration
● Processes/practices used
● Examples
– bank branches, retail stores, clinics, schools, etc.
Questions:
– How do we compare productivity of a diverse set of
operating units serving a diverse set of markets?
– What are the “best practice” and under-performing units?
– What are the trade-offs among inputs and outputs?
– Where are the improvement opportunities and how big are
they?
Some approaches
● Operating ratios
– e.g. Labor-hrs/transaction, sales/sq.-ft.
– Good for highly standardized operations
– Problem: Does not reflect varying mix of inputs and outputs
found in more diverse operations
● Financial approach: Convert everything to money!

Monetary Inputs Monetary Outputs

● Problems?
– Some inputs/outputs cannot be valued in Rupee (non-profit)
– Profitability is not the same as operating efficiency (e.g.
variances in margins and local costs matter as well)
Profitability vs. efficiency

● Profitability is a function of 3 elements …

– Input prices (costs)
– Output prices
– Technical efficiency (How much input is required to generate
the firm’s output.)

● Improving operations requires understanding

technical efficiency not just overall profitability.
 LP Formulation:
Data
K # operting units (DMUs) k = 1,..., K
N # inputs i = 1,..., N
M # outputs j = 1,..., M
O jk observed level of output j from DMU k
I ik observed level of input i from DMU k
Model variables
vi weight on input i
uj weight on output j
Ek efficiency of DMU (0 - 100%)
M

∑u O
j =1
j jk

Ek = N

∑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

Which can be formulated

M
max ∑ u j O je
j =1

s.t.
N
Normalize weighted input of

∑v I
i =1
i ie =1 e to one

M N

∑u O
j =1
j jk ≤ 1* ∑ vi I ik ,
i =1
k = 1,..., K

u j ≥ 0, j = 1,..., M
vi ≥ 0, i = 1,..., N
 Output analysis

λk dual variable associated with DMU k

λk > 0 ⇒ DMU k is in the reference set of DMU e

These dual variables can be used to construct an efficient
hypothetical composite unit (HCU) with
K
Oˆ j = ∑ λk O jk , j = 1,..., M Output j of HCU
k =1

K
Iˆi = ∑ λk I ik , i = 1,..., N Input i of HCU
k =1

Satisfying
Oˆ j ≥ O je , j = 1,..., M

Iˆi ≤ I ie , i = 1,..., N
HCU can be used to measure excess use of inputs and
potential increase in outputs

∆Output = Oˆ j − O je ≥ 0, j = 1,..., M

∆Input = I ie − Iˆi ≥ 0, i = 1,..., N

Refer to spreadsheet examples.

Using the results: Eff.-Profit Matrix
High Profit

Under-performing Best practice

potential leaders comparison group

Low High
Eff. Eff.

Under-performing Candidates for

possibly profitable closure

Low Profit
Designing DEA Studies
● #Inputs/Outputs
K > 2(N+M)
● “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 measurement
problems due to ...
– complexity of service outputs
– variability in service outputs
● Takes advantage of service operating environment
– large numbers of similar facilities
– diversity of practices/management/environment
● Provides useful information
– objective measures of productivity
– reference set of comparable units
– excess use of inputs measure
– returns to scale measure
DEA Summary (cont.)

● Role of DEA
– “data mining” to generate hypotheses
– evaluation/measurement
– benchmarking to identify “best practice” units
● Caveats
– “black box” - No information on root causes of inefficiency
– Be aware of assumptions (e.g. linearity)
– Can be sensitive to selection of inputs/outputs