Вы находитесь на странице: 1из 47

Data Envelopment Analysis

(DEA)

Identifying Efficient Decision Making Units


Measuring Service Productivity

Data Envelopment Analysis (DEA)


How can corporate management evaluate the productivity of a
fast-food outlet, a branch bank, a health clinic, or an elementary
school?
The difficulties in measuring productivity are threefold.

D-2
Measuring Service Productivity

First, what are the appropriate inputs to the system (e.g., labor
hours, material dollars) and the measures of those inputs?

Second, what are the appropriate outputs of the system (e.g.,


checks cashed, certificate of deposits) and the measures of those
outputs?

Third, what are the appropriate ways of measuring the


relationship between these inputs and outputs?

D-3
Measuring Service Productivity

The measure of an organization’s productivity, if


viewed from an engineering perspective, is similar to
the measure of a system’s efficiency.

It can be stated as a ratio of outputs to inputs (e.g.,


miles per gallon for an automobile).

D-4
Measuring Service Productivity

To evaluate the operational efficiency of a branch


bank, for example, an accounting ratio such as cost
per teller transaction might be used.

A branch with a high ratio in comparison with those of


other branches would be considered less efficient, but
the higher ratio could result from a more complex mix
of transactions.

D-5
For example,
a branch opening new accounts and selling CDs would
require more time per transaction than another branch
engaged only in simple transactions such as accepting
deposits and cashing checks.

D-6
The problem with using simple ratios is that the mix of
outputs is not considered explicitly.

This same criticism also can be made concerning the mix


of inputs.

For example, some branches might have automated teller


machines in addition to live tellers, and this use of
technology could affect the cost per teller transaction.

D-7
Broad-based measures such as

profitability or
return on investment
are highly relevant as overall performance measures,

But they are not sufficient to evaluate the operating


efficiency of a service unit.

D-8
For instance, one could not conclude that a profitable
branch bank is necessarily efficient in its use of personnel
and other inputs.

A higher than-average proportion of revenue-


generating transactions could be the explanation
rather than the cost-efficient use of resources.

D-9
For instance, one could not conclude that a profitable
branch bank is necessarily efficient in its use of personnel
and other inputs.

A higher than-average proportion of revenue-


generating transactions could be the explanation
rather than the cost-efficient use of resources.

D-10
The DEA Model
Fortunately, a technique has been developed with the
ability to compare the efficiency of multiple service units
that provide similar services by explicitly considering
their use of multiple inputs (i.e., resources) to produce
multiple outputs (i.e., services).

D-11
The DEA Model
The technique, which is referred to as data envelopment
analysis (DEA), circumvents the need to develop
standard costs for each service, because

it can incorporate multiple inputs and multiple outputs


into both the numerator and the denominator of the
efficiency ratio without the need for conversion to a
common dollar basis

D-12
Thus, the DEA measure of efficiency explicitly
accounts for the mix of inputs and outputs and,
consequently, is more comprehensive and reliable
than a set of operating ratios or profit measures.

D-13
DEA is a linear programming model that attempts to
maximize a service unit’s efficiency, expressed as a ratio of
outputs to inputs, by comparing a particular unit’s
efficiency with the performance of a group of similar
service units that are delivering the same service.

In the process, some units achieve 100 percent


efficiency and are referred to as the relatively efficient
units, whereas other units with efficiency ratings of less
than 100 percent are referred to as inefficient units.

D-14
Corporate management thus can use DEA to compare a
group of service units to

identify relatively inefficient units,


measure the magnitude of the inefficiencies, and

by comparing the inefficient with the efficient ones,


discover ways to reduce those inefficiencies.

The DEA linear programming model is formulated


according to Charnes, Cooper, and Rhodes, and is referred
to as the CCR Model.
D-15
Definition of Variables

Let Ek , with k 1, 2, . . . , K, be the efficiency ratio of unit


k, where K is the total number of units being evaluated.
Let uj , with j 1, 2, . . . , M, be a coefficient for output j,
where M is the total number of output types considered.
The variable uj is a measure of the relative decrease in
efficiency with each unit reduction of output value.

D-16
Definition of Variables
Let vi , with i 1, 2, . . . , N, be a coefficient for input i,
where N is the total number of input types considered.
The variable vi is a measure of the relative increase in
efficiency with each unit reduction of input value.
Let Ojk be the number of observed units of output j
generated by service unit k during one time period.
Let Iik be the number of actual units of input i used by
service unit k during one time period.

e is the index of the unit being evaluated

D-17
The objective is to find the set of coefficient u ’s
associated with each output and of v ’s associated with
each input that will give the service unit being
evaluated the highest possible efficiency.

This function is subject to the constraint that when the


same set of input and output
coefficients ( uj ’s and vi ’s) is applied to all other
service units being compared, no service
unit will exceed 100 percent efficiency or a ratio of
1.0.
D-18
THE DEA MODEL
Fractional Form
Objective Function

u1O1e  u2 O2e    u M OMe


max Ee 
v1 I1e  v2 I 2e    v N I Ne

Constraints

u1O1k  u2 O2 k    u M OMk
 10
. k  1, 2, , K
v1 I1k  v2 I 2 k    v N I Nk

j  0 j =1,2,…,M

i  0 i= 1,2,…,N
D-19
DEA in Standard LP Form
(Scaling inputs to sum to 1.0)

STANDARD LP FORM(scaling inputs to sum of 1.0)

max E e  u1O1e  u2 O2 e     u M O Me

SUBJECT TO:
v1 I1e  v2 I 2 e    v N I Ne  1

u1O1k  u2 O2 k    u M O Mk   v1 I 1k  v2 I 2 k    v N I Nk   0 k  1, 2, , K

where:

uj  0 j  1, 2,  , M
vi  0 i  1, 2,  , N

RECOMMENDED SAMPLE SIZE: K  2 N  M 


D-20
Example Problem
An innovative drive-in-only burger chain has established
six units in several different cities.
Each unit is located in a strip shopping center parking
lot. Only a standard meal consisting of a burger, fries,
and a drink is available.
Management has decided to use DEA to improve
productivity by identifying which units are using their
resources most efficiently and then sharing their
experience and knowledge with the less efficient
locations.

D-21
The next slide summarizes data for two inputs:
labor-hours and
material dollars
consumed during a typical lunch hour period to generate an output of 100
meals sold.

Normally, output will vary among the service units, but in this example, we
have made the outputs equal to allow for a graphical presentation of the units’
productivity.

D-22
Burger Palace Example
SUMMARY OF OUTPUTS AND INPUTS FOR BURGER PALACE

Service unit Meals sold Labor-hours Material dollars


1 100 2 200
2 100 4 150
3 100 4 100
4 100 6 100
5 100 8 80
6 100 10 50

LP MODEL FOR EVALUATION OF SERVICE UNIT 1

max E ( S1 )  u1100

subject to:
u1100  v1 2  v2 200  0
u1100  v1 4  v2 150  0
u1100  v1 4  v2 100  0
u1100  v1 6  v2 100  0
u1100  v1 8  v2 80  0
u1100  v110  v2 50  0
v1 2  v2 200  1
u1 , v1 , v2  0
D-23
For this simple example, we can identify efficient units by
inspection and see the excess inputs being used by
inefficient units (e.g., S 2 would be as efficient as S 3 if it
used $50 less in materials).

To gain an understanding of DEA, however, we will


proceed to formulate the linear programming problems for
each unit, then solve each of them to determine efficiency
ratings and other information.

D-24
Burger Palace Productivity
Frontier
200 S1 (2,200)
Material dollars

150 S2 (4,150)

100 S3 (4,100) S4 (6,100)

S5 (8,80)

C (5.3, 88.9)
S6 (10,50)
50

0
0 2 4 6 8 10
Labor hours D-25
To gain an understanding of DEA, however, we will
proceed to formulate the linear programming problems for
each unit, then solve each of them to determine efficiency
ratings and other information.

D-26
Burger Palace Example
SUMMARY OF OUTPUTS AND INPUTS FOR BURGER PALACE

Service unit Meals sold Labor-hours Material dollars


1 100 2 200
2 100 4 150
3 100 4 100
4 100 6 100
5 100 8 80
6 100 10 50

LP MODEL FOR EVALUATION OF SERVICE UNIT 1

max E ( S1 )  u1100

subject to:
u1100  v1 2  v2 200  0
u1100  v1 4  v2 150  0
u1100  v1 4  v2 100  0
u1100  v1 6  v2 100  0
u1100  v1 8  v2 80  0
u1100  v110  v2 50  0
v1 2  v2 200  1
u1 , v1 , v2  0
D-27
Similar linear programming problems are formulated
(or, better yet, the S1 linear programming problem is
edited) and solved for the other service units by
substituting the appropriate output function for the
objective function and substituting the appropriate input
function for the last constraint.

Constraints 1 through 6, which restrict all units to no


more than 100 percent efficiency, remain the same in all
problems.

D-28
This set of six linear programming problems was solved
with Excel Solver in less than five minutes by editing
the data file between each run.

Because the output is 100 meals for all units, only the
last constraint must be edited by substituting the
appropriate labor and material input values from the
input table shown in the slide prior to the previous slide
for the unit being evaluated.

D-29
The figure in the next slide shows, service units S1 , S3 ,
and S6 have been joined to form an efficient-production
frontier of alternative methods of using labor hours and
material resources to generate 100 meals.

As can be seen, these efficient units have defined an


envelope that contains all the inefficient units—thus the
reason for calling the process “data envelopment
analysis.”

D-30
Burger Palace Productivity
Frontier ( Figure K)
200 S1 (2,200)
Material dollars

150 S2 (4,150)

100 S3 (4,100) S4 (6,100)

S5 (8,80)

C (5.3, 88.9)
S6 (10,50)
50

0
0 2 4 6 8 10
Labor hours D-31
The linear programming results for each unit are shown in
Table 2 and summarized in Table 3.

In Table 2, we find that DEA has identified the same units


shown as being efficient in Figure K . Units S2 , S4 , and S5
all are inefficient in varying degrees.

Also shown in Table 2 and associated with each inefficient


unit is an efficiency reference set.

Each inefficient unit will have a set of efficient units


associated with it that defines its productivity.
D-32
Summary of DEA Results (TABLE 2)
Service Unit Efficiency Efficiency Relative Relative
Rating (E) Reference Set Labor- Material
Hour Value Value (V2)
(V1)
S1 1.000 N.A. 0.1667 0.0033
S2 0.857 S1 (0.2857) 0.1428 0.0028
S3 (0.7143)
S3 1.000 N.A. 0.0625 0.0075

S4 0.889 S3 (0.7778) S6 0.0555 0.0067


(0.2222)
S5 0.901 S3 (0.4545) S6 0.0568 0.0068
(0.5454
S6 1.000 N.A. 0.0625 0.0075
Calculation of Excess Inputs Used by Unit S4 (TABLE 3)
Outputs Reference Set Composite Excess
and Reference Inputs
Inputs Unit C Used
S3 S6 S4
Meals (0.7778) x 100 + (0.2222) x 100 = 100 100 0
Labor-hours (0.7778) x 4 + (0.2222)x 10 = 5.3 6 0.7
Material ($) (0.7778)x 100 + (0.2222)x 50 = 88.9 100 11.1
As Figure K shows for inefficient unit S4 , the efficient units
S3 and S6 have been joined with a line defining the efficiency
frontier.

A dashed line drawn from the origin to inefficient unit S4 cuts


through this frontier and thus defines unit S4 as inefficient.

D-35
In Table 2, the value in parentheses that is associated with
each member of the efficiency reference set (i.e., .7778 for
S3 and .2222 for S6 ) represents the relative weight assigned
to that efficient unit in calculating the efficiency rating for S4
.

D-36
These relative weights are the shadow prices that are
associated with the respective efficient-unit constraints in the
linear programming solution.

(Note in Table 2 that for unit 4, these weights appear as


opportunity costs for S3 and S6 .)

D-37
The values for v1 and v2 that are associated with the inputs
of labor-hours and materials, respectively, measure the
relative increase in efficiency with each unit reduction of
input value.

For unit S4 , each unit decrease in labor-hours results in an


efficiency increase of 0.0555. For unit S4 to become
efficient, it must increase its efficiency rating by 0.111
points.

This can be accomplished by reducing labor used by 2


hours (i.e., 2 hours 0.0555 0.111).
D-38
Note that with this reduction in labor-hours, unit S4 becomes identical
to efficient unit S3 .

An alternative approach would be a reduction in materials used by


$16.57 (i.e., 0.111/0.0067 16.57).

Any linear combination of these two measures also would move unit
S4 to the productivity frontier defined by the line segment joining
efficient units S3 and S6 .

D-39
Burger Palace Productivity
Frontier (Figure K)
200 S1 (2,200)
Material dollars

150 S2 (4,150)

100 S3 (4,100) S4 (6,100)

S5 (8,80)

C (5.3, 88.9)
S6 (10,50)
50

0
0 2 4 6 8 10
Labor hours D-40
Table 4 contains the calculations for a hypothetical unit C,
which is a composite reference unit defined by the weighted
inputs of the reference set S3 and S6 .

As Figure K in the previous slide shows, this composite


unit C is located at the intersection of the productivity
frontier and the dashed line drawn from
the origin to unit S 4.

D-41
Thus, compared with this reference unit C, inefficient unit
S4 is using excess inputs in the amounts of 0.7 labor-hour
and 11.1 material dollars.

D-42
DEA offers many opportunities for an inefficient unit to
become efficient regarding its reference set of efficient units.

In practice, management would choose a particular approach


on the basis of an evaluation of its cost, practicality, and
feasibility;

However, the motivation for change is clear (i.e., other units


actually are able to achieve similar outputs with fewer
resources).

D-43
Summary of DEA Results
SUMMARY OF DEA RESULTS

Efficiency Efficiency Relative labor-hour Relative material


Service unit rating (E) reference set value  v1  value v2 
S1 1.000 N.A. .1667 .0033
S2 0.857 S1 (.2857) .1428 .0028
S3 (.7143)
S3 1.000 N.A. .0625 .0075
S4 0.889 S3 (.7778) .0555 .0067
S6 (.2222)
S5 0.901 S3 (.4545) .0568 .0068
S6 (.5454)
S6 1.000 N.A. .0625 .0075

CALCULATION OF EXCES INPUTS USED BY UNIT S4

Composite
Outputs and reference Excess
inputs Reference set unit C inputs used
S3 S6 S4
Meals (.7778) × 100 + (.2222) × 100 = 100 100 0
Labor-hours (.7778) × 4 + (.2222) × 10 = 5.3 6 0.7
Material $ (.7778) × 100 + (.2222) × 50 = 88.9 100 11.1 D-44
DEA and Strategic Planning

High
Under-performing Benchmark
potential stars group
Profit

Problem Candidates
Low
Branches for divestiture

Low High
Efficiency

D-45
Mid-Atlantic Bus Lines
Depot City Ticket Freight Labor Cost
1 Philadelphia 700 300 40 500
2 Baltimore 300 600 50 500
3 Washington 200 700 50 400
4 Richmond 400 600 50 500
5 Raleigh 500 400 40 400
6 Charleston 500 500 50 500
7 Savannah 800 500 40 600
8 Jacksonville 300 200 30 400
D-46
Mid-Atlantic Bus Lines Questions
1. Use DEA to identify efficient and inefficient
terminal operations.
2. Using the appropriate reference set of efficient
terminals, make recommendations for change in
resource inputs for each inefficient terminal.
3. What recommendations would you have for the
one seriously inefficient terminal in regard to
increasing its outputs?
4. Discuss any shortcoming in the application of
DEA to Mid-Atlantic Bus Lines.

D-47

Вам также может понравиться