Performance

measurement: from
which perspective?
• What is performance?
• Sorting out the good from the bad.
– But, who defines what is good or bad?

– Whose interests are organizations

answering to? (Fitzgerald and Storbeck,
2002)
– Performance is socially constructed and means
different things to different people (Wholey,
1996) 3
 What is
productivity?
Hospital A has 1200 beds and
handles 100,000 patients per
year

• What is the productivity?

• What is the efficiency?
• What is effectiveness?
• Can the hospital improve these?
• How? 8
 Productivity
y
Productivity  
output
x
input
Production frontier
y

Feasible production
set : area between
x
production frontier 9

and x-axis
 Productivi
ty
y
Productivity  
output
x
y
input
yA A
Productivity A 
y AA
x

xA x 1
0
 Productivity =
Efficiency ?
y

DMU “A” is
producing at this
non-optimal point
x

1
1
 Productivity =
Efficiency ?
y
B

A A  B means
improving technical
efficiency

x
Point B is technically efficient (= max output, given a certain
input) Moving from A to B is called technical efficiency 1
2
improvement.
 Productivity =
Efficiency ?
y
B

A A  B means
improving technical
efficiency

x
Technical efficiency is to produce as much as
possible
13
with the technology currently present
 Productivity =
Efficiency ?
y
B
C
A  C means
A
changing the scale of
operations

Moving from A to C is called exploiting scale

economies 14
 Output vs. input
orientation
• Output orientation:
By how much can output quantities be
proportionally expanded without
changing the
input quantities used?

• Input orientation:
By how much can input quantities be
proportionally
reduced without changing the output
quantities 18
 Technical versus allocative
efficiency
Technical efficiency:
to obtain the maximum output, given a
certain input

Allocative efficiency (involves prices or costs):

to select a mix of inputs that produces a
given output at minimum cost.

Economic efficiency = technical  allocative

efficiency
19
 Productivi
•
ty
Productivity is easy to define when you
have 1 input and 1 output

• For multiple inputs and outputs, you

need more ratios!

• Data Envelopment Analysis (DEA)

is a technique to measure
productivity with multiple inputs
and outputs
20
 Some
definitions
• DMU: Decision Making Unit
– E.g. hospital, firm, doctor, department
• Input: production means  vector X
– E.g. labor, machine capacity, production
costs
• Output: production  vector Y
– E.g. treated patients, products, profit

21
 Example: 1 input, 1
Store output
A B C D E F G H
Employee (x) 2 3 3 4 5 5 6 8
Sale (y) 1 3 2 3 4 2 3 5

6 When is a store
5 H
better than
E
Sale

4
3
B
D G another?
s

Which store
2 C F
1 A performs the
0
0 1 2 3 4 5 6 7 8 9 best?
Employe 22
e
 Let’s look at productivity
Store ratiosA B C D E F G H
Employee (x) 2 3 3 4 5 5 6 8
Sale (y) 1 3 2 3 4 2 3 5
Sale/Employee 0.5 1 0.667 0.75 0.8 0.4 0.5 0.625

Efficient Productivity
6
frontier
5 H
(€ / employee)
E
Sale

4
B
3 D G
s

2 Envelopment area:
C F
1
A
all feasible
0 production
0 1 2 3 4 5 6 7 8 9
Employe combinations 23
e
 Statistical regression
Store lineA B C D E F G H
Employee (x) 2 3 3 4 5 5 6 8
Sale (y) 1 3 2 3 4 2 3 5

6
5 H
E Regression line (y =
Sale

4
B
3 D G 0.622x): focuses on
s

2 C F averages:
1 A
points
0
0 1 2 3 4 5 6 7 8 9 above:excellent
Employe 24
e points below:
 Data envelopment
Store analysis
A B C D E F G H
Employee (x) 2 3 3 4 5 5 6 8
Sale (y) 1 3 2 3 4 2 3 5
Sale/Employee 0.5 1 0.667 0.75 0.8 0.4 0.5 0.625

Efficient
Productivi
6
frontier ty
5 H In DEA, the best DMUs
E (B) serve as a
Sale

4
B
3 D G benchmark
s

2
C F Efficiency
1
A others:
sale /
0 0 
employee
sale B/
0 1 2 3 4 5 6 7 8 9  1
Employe employee B 25
e
 Data envelopment
Store analysis
A B C D E F G H
Employee (x) 2 3 3 4 5 5 6 8
Sale (y) 1 3 2 3 4 2 3 5
Sale/Employee 0.5 1 0.667 0.75 0.8 0.4 0.5 0.625

sale /
Efficiency others: 0 sale 
employee
B/ 1
 employee B
For example: efficiency F is: 0.4/1 =
40% And: 1 = B > E > D > C > H > A =
G > F = 0.4
This relative efficiency measure is units
invariant, whereas 2
6
 Example: 2 inputs, 1 output
Store A B C D E F G H I
Employee 4 7 8 4 2 5 6 5.5 6
Floor area 3 3 1 2 4 2 4 2.5 2.5
Sale 1 1 1 1 1 1 1 1 1

Production possibility Efficient frontier: C-

set: D-E
5
4 E
A
G C: the best w.r.t. floor
H B
area/Sales

3 area E: the best w.r.t.

2
F I employee D: best
D C
Floor

1 combination
0
0 1 2 3 4 5 6 7 8
9 2
Employee/Sales 7
 Example: 2 inputs, 1
output 5

4 E A G
H B
3
area/Sales P
2 F I
D
C
Floor

0
0 1 2 3 4 5 6 7 8
9
Employee/Sales

• Efficiency of e.g. A is: OP  0.8571, or

OA
85.71%
• A must multiply both inputs by 85.71% to
become technically efficient (input
orientation) 28
 Example: 2 inputs, 1
output 5

4 E A G
H B
3
area/Sales P
2 F I
D
C
Floor

0
0 1 2 3 4 5 6 7 8
9
Employee/Sales

• D and E are the peers / reference set of A

– A must consider D and E as examples to become
technically efficient
• C and D are the peers / reference set of F,
etc. 29
 Example: 1 input, 2
outputs
Store A B C D E F G
Employee 1 1 1 1 1 1 1
Customers 1 2 3 4 4 5 6
Sale 5 7 4 3 6 5 2
B
7
6 E Efficient frontier: B-E-
5 A F F-G
B: best w.r.t.
output2/

4 C
sales E: best
input

3 D
2 Productio G combination F:
1 n best combination
possibilit
0
0 1y set
2 3 4 5 6
G: best w.r.t.
7 customers 3
output1/input 0
 Example: 1 input, 2
outputs B
7
6 E
5 A F

output2/
4 C P

input
3 D
2 G
1
0
0 1 2 3 4 5 6
7
output1/input
• A, C, and D are inefficient
• Efficiency of D = OD / OP = 0.75 =
75%
• Reference set of D is {F, G}
31
 Example: 1 input, 2
outputs B
7
6 E
5 A F

output2/
4 C P(16/3,4

input
3 D )
2
(4,3)
1 G
0
0 1 2 3 4 5 6
7
output1/input
• Efficiency of D is 75%
• D must multiply its outputs by 1/0.75 =
1.33 to become technically efficient
(output orientation)
32
1.33 x (4,3) = (16/3,4) = P
 Technical vs. mix
Q B
inefficiency 7
• D  P eliminates 6 E
inefficiency without 5 A F

output2/
P
changing the 4 C

input
D
output/input proportions. 3
2 G
D is called technically
1
inefficient. 0
0 1 2 3 4 5 6
• A  Q shows that A 7
output1/input
is technically
However, Q’s output1 can be improved to B,
inefficient.
without modifying the input. This is called mix
inefficiency.
 A is both mix and technically inefficient
Q is techn. efficient, but not mix efficient. B is 33

both.
 Example: 2 inputs, 2
outputs
Hospital A B C D E F G H I J K L

Doctors 20 19 25 27 22 55 33 31 30 50 53 38

Nurses 151 131 160 168 158 255 235 206 244 268 306 284

Outpatients 100 150 160 180 94 230 220 152 190 250 160 250

Inpatients 90 50 55 72 66 90 88 80 100 100 147 120

When is a hospital better than

another? Which hospital performs
the best? 34
 Example: 2 inputs, 2
Hospital outputs
A B C D E F G H I J K L

Doctors 20 19 25 27 22 55 33 31 30 50 53 38

Nurses 151 131 160 168 158 255 235 206 244 268 306 284

Outpatients 100 150 160 180 94 230 220 152 190 250 160 250

Inpatients 90 50 55 72 66 90 88 80 100 100 147 120

Possibility: assign weights to inputs and

outputs:
E.g Weight for doctor : weight for nurse =5:
. Weight for outpatient : weight for 1
inpatient
Drawback: these fixed weights are =1:
arbitrary 3

35
 DEA: Data Envelopment
Analysis
Hospital
(Charnes,
A B C
Cooper,
D E
Rhodes,
F G H
1978)
I J K L
Doctors 20 19 25 27 22 55 33 31 30 50 53 38
Nurses 151 131 160 168 158 255 235 206 244 268 306 284
Outpatients 100 150 160 180 94 230 220 152 190 250 160 250
Inpatients 90 50 55 72 66 90 88 80 100 100 147 120

• DEA uses variable weights

• Each hospital chooses weights
individually, such that:
– Their productivity is maximized, and
between 0 and 1
– All other DMU productivities are also 36

between 0 and 1
 DEA notation
Entities:
N DMUs
(index i, j) K
inputs
M outputs
Parameters:
xi = K 1 input
vector for DMU i
yi = M 1 output
vector for DMU i
Variables:
37
v = K 1 input
 DEA notation: example
Hospital A B C D E F G H I J K L

Doctors 20 19 25 27 22 55 33 31 30 50 53 38

Nurses 151 131 160 168 158 255 235 206 244 268 306 284

Inpatients 90 50 55 72 66 90 88 80 100 100 147 120

K = 2 inputs, M = 1 output
xi = K1 = 21 input vector for
DMU i
yi = M1 = 11 output vector for
DMU i 38
 Constant Returns to Scale (CRS) DEA
model
CRS: linear relation between input and output

For each DMU i, we want to obtain a vector v

of input weights and a vector u of output
weights, such that the weighed ratio of
outputs and inputs is maximal:
u yi
T
Weighed
max T output
v xi Weighed
input
Where u, v are variable vectors, and xi, yi are
resp. input and output parameter vectors

39
 CRS-DEA model
Basic form (model solved for each
DMU j):
u yi
T
max T
v xi
s.t. uT y
i 1
vT x i i 
u,v  0
In other words: find values for the weight
vectors u and v, in such a way that the
efficiency measure for this DMU j is
maximized 40
 CRS-DEA model
If (u*,v*) is a solution, then (  u*,   v*) is
also a solution  there are infinitely
many solutions!
Solution: add constraints: vTxi = 1, and
rewrite the model as follows:
Ratio form ofyDEA
u max  i
T T
model Multiplier form of DEA model
max T i

v xi y
s.t. u y T
s.t.  T
x 1
i  1 i

v T x i i  T

 y i  xi  0 i
T

u,v   ,  41

0 0
 CRS-DEA model
v and u are replaced by  and 
respectively, to discern between
these two models
Ratio form of DEA Multiplier form of DEA
model model
u yi T
max T i
max T
v xi y
s.t. u y
T
s.t.  T
x 
i 1 1
vT x i  y  x i  0
T T

i  i i

u,v i  ,  42

0 0
 Example: 2 inputs, 1
output
DMU A B C D E F
Input: x1 4 7 8 4 2 10
x2 3 3 1 2 4 1
Output: y1 1 1 1 1 1 1

Production possibility
set:
5 E
4 A
x 2/
y1

B
3
2
D
1
C F
0
0 1 2 3 4 5 6 7 8 9
10 4
x1/y1 3
 Example: 2 inputs, 1
output
DMU A B C D E F max  T
i
Input: x1 4 7 8 4 2 10
ys.t  T xi 
x2 3 3 1 2 4 1
.T
Output: y1 1 1 1 1 1 1
 y i 1 T xi  0  i

for DMU max  1

 , 
B:
0
71  3 2    41  22
1  4  3
1 2 D 21  4 2
A 7 1  3 2 E101 
 F
B   2
44
  81  2 ,  0
 Example: 2 inputs, 1
Optimaloutput
solution for B:
1 = 0.0526, 2 = 0.2105,  = 0.6316 = * = efficiency
Reference set: {C, D} (equality occurs for constraints (C)
and (D)) Note: 2 / 1 = 4  input2 is four times as
important to B as input1
5
4 E
A
x 2/

3
y1

B
2
D
1
C F
0
0 1 2 3 4 5 6 7 8 9 10
x 1/
y1 45
 Example: 2 inputs, 1
output
Optimal solution for F:
1 = 0, 2 = 1,  = 1 = * =
efficiency

5
4 E
A
x 2/

3
y1

B
2
D
1
C F
0
0 1 2 3 4 5 6 7 8 9 10
x 1/
y1 46
 Pareto-Koopmans
efficiency
Definition:
A DMU is fully efficient if and only if it is
not possible to improve any input or
output without worsening some other
input or output

47
 CRS-DEA model: envelopment
form
Dual form of multiplier model is the
envelopment form:
Multiplier form of DEA (input oriented version)
Envelopment form of DEA
model model
max T yi min 
s.t. 1 dual:  s.t.
 xi 
T
 y i  Y 
T 1
 y i  xi  0 
T N duals:  0

  ,  i
 0 xfree,
i  X
0
0
1+N
constraints
> K+M 4
8
constraint
 Projection and slacks
The projection of x2/ Example
y A
the inefficient
DMUs is: A
’
A to A’
B
B to B’ C B
However, it is ’ D
questionable
whether A’ is an
0 x1/
efficient point, since its y
be reduced
input x2 can without changing the output (A’ 
C).
This is known as input slack. Similarly, there can
50
also be output slack. Slack = mix inefficiency!
 Projection and slacks

Envelopment form

min 
s.t.
 yi  Y  0  
 Y  Output
yi slack
 X  Input
xi  X  0  slack
xi
 free,   0 51
 Example: 2 inputs, 1
output
DMU 1 2 3 4 5 x
y
/ (Example
2
from Coelli et
6
Input: x1 2 2 6 3 6 al. page
5
1 143,...)
x2 5 4 6 2 2 4
3
Output: y 1 2 3 1 2 2 4
2
x1 / y 2 1 2 3 3 3 5
1
x2 / y 5 2 2 2 1 0
0 1 2 3 4 5 x1/
For DMU 3: 6
min  y
min
s.t. s.t  y 3  (y11  y22  ...  y55 )
 yi  Y
. x13 0 (x111  x122  ...  x155 )
x0  X 
i x230 (x211  x222  ...  x255 ) 

0 0
1 2 3 4 5 5
 free,    free, ( , , , , )0 2
 Example: 2 inputs, 1
DEAoutput
DMU      1 2 4 5 IS1 IS2 OS
result 3
s: 1 0.5 - 0.5 - - - 0.5 -
-
2 1.0 - 1.0 - - - - -
-
3 0.833 - 1.0 - 0.5 - - -
x2/ -
6
y 4 0.714 - 0.214 - - 0.286 - - -
5 2 is 100%
1
4 5 1.0 -
efficient - - 1.0 - - -
3
Reference set / targets /
2 4 - peers for 3 are {2,
2
3 5 5}
1 1 is 50% efficient, and has input
0 slack (= technical and mix
0 1 2 3 4 5 x1/ inefficiency)
5
6
y 3
 Variable Returns to Scale
• (VRS)
CRS: k times input means k times output.
This may not be realistic (e.g. constraints on
finance)!
• VRS: DEA only considers the convex hull of
observed DMUs (DMUs are only
benchmarked against DMUs of similar
size)
5 E 5
E
x /y
42 1 G x /y
4 2 1 G
A B
H A B
3 3 H
2
D
F I 2
D
F I
C
1 C 1
0 0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
x1/y1 x1/y1
Variable Returns to Scale
(VRS)
CRS DEA model
VRS DEA model
min 
min  s.t.
s.t.  y i  Y  0
 y i  Y  0 xi  X  0
xi  X  0 1T   1
 free,   0  free,  
0
  i Convex
combination of 55

inputs and
 Scale efficiency
Example (1 input, 1 output):
CRS Frontier
y VRS
Frontier
Q S

CRS projection
PCRS TECRS = APCRS/AP
A
P
R

x
56
 Scale efficiency
Example (1 input, 1 output):
CRS Frontier
y VRS
Frontier
Q S

VRS projection
P CRS P VRS TEVRS = APVRS/AP
A
P
R

x
57
 Scale efficiency
Example (1 input, 1 CRS Frontier
output):
VRS
y S
Frontier

A PCRS PVRS
P
R

x
Scale efficiency (SE): / APVRS = TECRS/TEVRS
58
APCRS
 Scale efficiency
This means that: TECRS = TEVRS x SE

In other words, CRS technical efficiency can

be decomposed in “pure” technical efficiency,
and scale efficiency.

Furthermore: the scale efficiency can be

computed from the CRS and VRS solutions

Scale efficiency (SE): / APVRS = TECRS/TEVRS

59
APCRS