Keynote Paper Booklet (2) SH 22072014105003

OR56 Annual Conference - Keynote Papers
Royal Holloway University of London, 9 11 September 2014

Foreword
Foreword
On behalf of the organising committee, I would like to welcome you to the OR Societys
Annual Conference - OR56. This year we are hosted by Royal Holloway, University of
London, set in the county of Surrey.
In addition to three prominent plenary speakers, the conference is made up of presentations
covering 24 streams. The majority of these streams have keynote speakers who will present
the latest developments in the field. These cover a wide range of areas such as cloud
computing, policy analysis in the power sector, undertaking OR in the third sector and models
for ranking sports players there is certainly something for everyone.
Other exciting and thought provoking sessions include a session on Behavioural Operational
Research. This session will involve a plenary panel that brings together academic and
practitioner experts, from differing perspectives, to provide a lively forum for discussing the
development of this emerging field of OR. The conference is also holding its first debate
session. Is Good enough good enough? will discuss optimising versus satisfying approaches
to optimisation problems that are at the core of OR. The debate will consider whether good
enough solutions are closer to reality than pure optimal solutions. Why not come along and
put your opinion forward. The Making an Impact Day is also back by popular demand. The
sessions during this day are primarily focussed at practitioners who want to help make
organisations more effective. The day includes an opportunity to make some new contacts
through the speed networking session and listen to presentations on high quality case studies
that have had real impact in organisations.
Of course, an important part of the conferences is networking and being able to chat
informally with others involved in OR. As always, the social program will offer lots of
opportunities to meet up with old friends and to make some new ones.
I would like to take this opportunity to thank all the keynote speakers who provided their work
to be included in this book. Reading through each of these papers provides a great opportunity
to find out about the latest developments across the field of OR. I would also like to thank all
those who submitted their abstract for presentation in the streams. It is all these contributions
that make a successful conference.
I hope that you have a fruitful conference and that you enjoy your time at Royal Holloway.
With best regards,
September 2014
__________________________________________________________________
Susan Howick, University of Strathclyde
Keynote Handbook Editor

Contents
Contents
Data Envelopment Analysis

Setting Handicaps to Industrial Sectors in DEA: An application to Ethiopian
Industry 1
Kaoru Tone, Kidanemariam Berhe Hailu
Health
Health Care Simulation for Policy Assessment: Widening Requirements Capture
With Soft Methods19
John Powell, Navonil Mustafee
Infrastructure
Operational Research Methods for Policy Analysis in the Power Sector. 38
Afzal Siddiqui
Inventory Research
On the Value of Sharing Demand Information in Supply Chains......... 44
Mohammad Ali, John Boylan
OR Consultancy & Case Studies
The Evolution of OR in the Defence Industry in the UK: A Personal Perspective.. 57
Noel Corrigan
OR in Sport
Rankings Models in Sport.. 68
Ian McHale
Problem Structuring
Zooming in on Problem Structuring Interventions: An Affordance
Perspective. 78
L. A. Franco
Project Management
Value Management: From Wishful Thinking to Causal Certainty 91
Roger Davies
ii

Contents
Simulation
Cloud Computing for Modelling & Simulation106
Simon Taylor, Anastasia Anagnostou
Sustainable Supply Chains
A Methodological Framework for Green Logistics Networks under Periodic
Review Replenishment Policies118
Eleftherios Iakovou, Ioannis Mallidis, Dimitrios Vlachos, Rommert Dekker
Third Sector OR
Whats so Special about the Third Sector? ..........................................................................134
Ruth Kaufman
iii

Data Envelopment Analysis
Setting Handicaps to Industrial Sectors in DEA: An Application to

Ethiopian Industry
Kaoru Tone, Kidanemariam Berhe Hailu
National Graduate Institute for Policy Studies, Tokyo, Japan
tone@grips.ac.jp, doc11102@grips.ac.jp
Abstract
In the ordinary macro-economic input-output tables, the industrial sector consists of several
dozen industries and each industry in a certain sector is an aggregate of many companies in the
sector. We can apply traditional DEA models for evaluation of efficiency regarding all sectors
by means of common input and output factors. However, there remain concerns about
comparing all sectors as a scratch race. For example, some sectors are in fields with matured
technologies, while others are in emerging fields. These situations lead us to compare sectors
under a handicap race. In this paper, we propose a new DEA model based on the non-convex
frontiers that all associated sectors may exhibit and from which handicaps are derived. We
apply this model to Ethiopian industry.
Keywords: Data envelopment analysis; Non-convex frontier; Handicap; Ethiopian industry
1. Introduction
In the ordinary macro-economic input-output tables, the industrial sector consists of several
dozen industries and each industry in a certain sector is an aggregate of many companies in the
sector. The sectorial statistics are the sum of statistics of companies in the respective sector.
Usually, all sectors have the same set of inputs for producing outputs. For example, they have
labour, capital and intermediate as input and the amount of production as output. We can
apply traditional DEA models for evaluation of efficiency regarding all sectors by means of
these common input and output factors. However, there remain concerns about comparing all
sectors as a scratch race. Some sectors are in fields with matured technologies, while others
are in emerging fields. Some are labour intensive, while others are capital intensive. These
differences lead us to compare sectors under a handicap race. In this paper, we propose a new
DEA model based on the non-convex frontiers that all associated sectors may exhibit and from
which handicaps are derived. Most DEA models assume convex set frontiers. However, there
are non-convex frontiers as indicated by the S-shaped curves in production. Tone & Tsutsui
(2013) proposed a new DEA model that can cope with non-convex frontiers. They classify all
DMUs (decision-making units) into several clusters and define a new efficiency score, called
SAS (scale and cluster adjusted score), that can take account of non-convex frontiers. In this
paper, we define handicaps for each industrial sector using the SAS model. We modify inputs
(outputs) by the handicaps and re-evaluate the efficiency scores. We apply this model to
Ethiopian industry.
Several authors discussed handicap-related topics in DEA. Yang & Paradi (2006) proposed a
Handicapped Data Envelopment Analysis to adjust for corporate strategic effects for
K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)
Canadian banks. They applied the index number originally proposed by Fixler and Zieschang
(1993). Their index number is based on the tactical and strategic heterogeneity between banks.
Olesen & Petersen (2009) discussed target and technical efficiency in DEA controlling for
environmental characteristics. They extended Banker & Morey (1986) and incorporated
allowable handicap values into the model
along the same lines as specifications of assurance regions in standard DEA. Our problem and
approach differ from the preceding ones as follows. (1) We deal with industrial sectors which
have a two-layered structure, i.e., each sector consists of many companies in the sectorial
category and its inputs/outputs are the sum of these companies. (2) Although we wish to
evaluate the technical efficiency of sectors, there are handicaps among sectors, as mentioned
above, which should be identified and be accounted for in efficiency measurement. (3) For this
purpose, we first find the variable returns-to-scale (VRS) frontiers of each sector and project
companies in the sector to their frontiers. (4) Then, we find the VRS meta-frontiers for the
projected companies in all sectors. (5) If the best performer (company) in a sector is positioned
on the meta-frontiers, then we classify the sector to the no-handicap group. Otherwise, if the
best performer is off the meta-frontiers, we classify the sector to the with-handicap group.
This indicates that this sector is in emerging fields or is in unfavourable environments. (6) In
order to gauge the degree of handicaps, we apply the non-convex frontier model developed in
Tone & Tsutsui (2013) and decide the handicaps of with-handicap sectors. (7) Using
handicaps, we adjust inputs (input-oriented case) and outputs (output-oriented case), and apply
the CRS (constant returns-to-scale) model to obtain the final sectorial efficiency score.
It should be noted that we are comparing sectorial efficiency as measured by the ratio input vs.
output. For example, three sectors (Food, Textile and Motor) have the respective input
(manpower) and output (profit) exhibited in Table 1. From this table, we see that the virtual
(dual) value of input for Food is one tenth of that for Motor, but we do not intend to say that
Food should reduce its input to 1, because the environments of the three sectors are quite
different. However, this kind of comparison is necessary for understanding national or
international economics.
Table 1: Three sectors
Sector
Food
Textile
Motor
Input (Manpower)
10
5
1
Output (Profit)
1
1
1
Output/Input
0.1
0.2
1
This paper unfolds as follows. In Section 2, we introduce the basic framework and classify
sectors into non-handicap and with-handicap groups. In Section 3, we set handicaps for withhandicap sectors. Then, we redefine sectorial inputs and outputs using the handicaps and
obtain the final efficiency scores which take account of the sectorial handicaps in Section 4. In
Section 5, we apply this model to Ethiopian industry. Section 6 concludes this paper.
2. Basic framework
In this section, we describe the basic materials in the paper. Suppose that there are K sectors in
the industry and each sector k (=1,,K) has nk DMUs with m inputs and s outputs. Let us
denote a DMU in the sector k by x kj , y kj where x kj Rm ( y kj Rs ) is the input (output) vector
of the DMU. We define the set of DMUs in the sector k by X k , Yk with X k x1k , , x kn
and Yk y1k , , y kn
2.1.Evaluation of DMU within each sector and its projection onto frontiers of the
sector
We evaluate each DMU in its belonging sector by using the VRS model. In this paper, we use
the input oriented SBM (slacks-based measure (Tone 2001)). However other models, e.g.
radial models, can be applied as well. This can be attained as below.
For each DMU xok , yok (o 1, , nk ) we solve the following LP:
min
1

, so , so
1 m sio
m i 1 xiok
subject to
X k s o x ok
Y s y
k
(1)
k
o
e 1
0, s o 0, s o 0,
where is the intensity vector, and so and so are respectively input and output slacks.
Let an optimal solution to (1) be (* , so* , so* ) . We project xok , y ok onto the efficient frontiers
of the sector k as follows:
k
xo xok so* , y o y ok so* .
(2)
Thus, we obtain the set of DMUs X , Y (k 1, , K ) which are VRS-efficient with respect to
the frontiers of the section k. See Figure 1.
Frontiers of
Sector A
Output
Frontiers of
Sector C
Frontiers of
Sector B
Input
Figure 1: Sectorial frontiers and projections
2.2.Global evaluation of the projected DMUs
which consists of n n
DMUs. We evaluate the VRS efficiency of x , y with respect to X , Y and denote its
We merge the set X , Y (k 1, , K ) and denote it by X , Y

k
o
VRS score by o . Further, we define the maximum of o among the sector k as,
k
max o .
o 1, , nk
(3)
Output Meta-frontiers
No-handicap group (A)
With-handicap group (C)

No-handicap group (B)
Figure 2: Meta-frontiers and handicaps
Input
If 1 , the best performer of the sector k is located on the global VRS frontiers (meta-
frontiers) of X , Y
. We judge this sector k has no handicap and classify k to the nok
handicap group. Instead, if 1 , the best performer of the sector k is inferior to the best
performers in the no-handicap group and we classify k to the with-handicap group. See
Figure 2 where Sections A and B belong to no-handicap group and Section C to the withhandicap group.
3. Handicaps
In this section, we describe how to set handicaps for the with-handicap group.
3.1.Sectorial inputs and outputs

Sectorial inputs and outputs can be defined as the aggregates of VRS-projected DMUs in the
sector as follows:
Input to Sector k : x i jk1 x ij (i 1,
k
, m; k 1,
Output from Sector k : y l jk1 y lj (l 1,

k
, K)
, s; k 1,
(4)
, K)
Input/output vectors of sector k are defined by

k
x x1 ,
, xm
and y y ,
T
k T
, ys
(5)
k
We deal with K DMUs defined by x , y
(k 1,
, K) .
3.2.Clustering
We classify K sectors in several clusters. First, sectors belonging to the no-handicap group go
to the cluster NHD, while a sector with-handicap holds its sector name as the cluster name.
For example, if a sector Machinery and equipment belongs to the with-handicap group, its
cluster name is Machinery and equipment.
The characteristics of industrial sectors are diverse. Some are in mature fields while others are
in emerging fields. This suggests the existence of S-shaped (non-convex) frontiers as exhibited
in Figure 3. Tone & Tsutsui (2013) proposed a method for solving non-convex frontiers based
on clusters
Output
Input
Figure 3: S-shaped frontier
3.3.Solving with the non-convex model
We solve the dataset x , y
(k 1,
, K ) with cluster name using the above non-convex
model and obtain the scale and cluster adjusted efficiency score, called SAS, which accounts
for the effect of non-convex frontiers.
3.4.Handicap
We define the handicap hk of sector k as follows:
hk 1 , if the sector belongs to the no-handicap group.
(6)
hk SAS score , if the sector belongs to the with-handicap group. (7)
4. Global issue
In this section, we redefine sectorial inputs and outputs using the above defined handicap and
obtain the overall efficiency score for industries.
4.1.Input (Output) under handicap

In the input-oriented case, we define sectorial inputs and outputs as follows:
Sectorial input ik hk jk1 xijk
(i 1,
, m : k 1,
Sectorial output lk j 1 yljk
(l 1,
, s : k 1,
nk
, K)
(8)
, K)
Further we define input/output vectors for each sector as follows:

k 1k ,
, mk and k 1k ,
T
,sk
(k 1,
(9)
, K)
See Figure 4 for an example.

Sector A A , A
Output
Sector C C , C
Sector B B , B
Input
Figure 4: Handicapped input and output
4.2.Solving the CRS model

In the non-radial SBM model case, we solve the following LP and obtain the sectional
efficiency score o* (o 1, , K ) under handicap.
o* min 1
,s ,s
1 m sio
m i 1 io
st.
K
k 1
K
k 1
k k so o
(10)
k k so o
0, so 0, so 0
Let an optimal solution to (10) be
,s
*
*
o
, so* . Then we have the projection to the SAS
efficient frontiers as follows:

Projected input: *o o so*
Projected output: *o o so* .
5. Application to Ethiopian industry

In this section, we describe the characteristics of the Ethiopian manufacturing sector and our
related data, and apply our model to 14 industrial groups which largely dominate this sector.
5.1.On Ethiopian industry

5.1.1. The manufacturing sector
Ethiopias industrial sector has not yet taken off. Its contribution to GDP has been stagnating
for a long period of time amounting to between 13 and 14 percent. Particularly, the share of
the manufacturing sub-sector remained stagnant. The manufacturing sector is dominated by
simple agro-processing activities and production of basic consumer goods. All manufacturing
exports are agriculture-based, which include clothing, semi-processed hides, footwear,
beverages and others. However, Ethiopia has great potential to jump-start its manufacturing
sector. Ethiopia has abundant low-cost labor, which gives it a comparative advantage in lessskilled, labor-intensive sectors such as light manufacturing which include textile, wood
products, leather products and apparel industries (Dinh et al., 2012; Sonobe, 2009). This
provides a good opportunity for low-cost manufacturing exports. Moreover, Ethiopia has
abundant natural resources; cattle, which can be used as an input for making leather and
leather products; forests, which can be used in the furniture industry; cotton, which can be
expanded to develop further garments industry; and land for the agribusiness industry. On the
policy side, the government, with an aim of transforming the country from agrarian to an
industrialized nation, has been preparing a conducive environment to attract foreign direct
investment and encourage domestic investment. To tap into all this opportunities, many
foreign companies such as Chinese, Indian, Japanese and Turkish companies are currently
flocking into the country.
5.1.2. The data

As far as data are concerned, the data used in this application were extracted from the Central
Statistical Agency (CSA) of Ethiopia database. CSA conducts annual census (covering the
entire population) of large and medium size power driven manufacturing establishments of the
country that employ 10 persons or more. We use data from 2008 census that comprises
relevant outputs and inputs. For the purpose of this application, data were aggregated to 2-digit
level industries according to the International Standard Industrial Classification of All
Economic Activities (ISIC). Furthermore, to avoid data fluctuations, we also smooth the data
using a four-year moving average1. Based on the literature and data availability, we use a
single-output and 3-input production technology for the Ethiopian manufacturing. Output is
measured by the gross value of all output produced by the firm. The inputs include (i) number
of employees measured by the sum of permanent and temporary workers, (ii) capital input
measured by the net value of fixed assets at the end of the survey year, (iii) intermediate inputs
aggregated as the sum of the values of raw materials, fuel and lubricating oil, electricity, wood
and charcoal for energy for each establishment and other industrial costs. The types of
industries for which our model applies include manufacture of food and beverages, textiles,
wearing apparel, tanning, leather and footwear, wood and wood products, paper and printing,
fabricated metal products, non-metallic mineral products, chemical and chemical products,
rubber and plastics products, basic iron and steel, machinery and equipment, motor vehicle,
and furniture.
5.2.Main statistics
We begin by presenting the original (non-handicapped) dataset in Table 2. The summary
statistics of this dataset is exhibited in Table 3. Note that the summary statistics in Table 3
represents the Min/Max inputs and outputs of individual DMUs, not that of the industry.
The 2008 data are part of the big dataset (covering 2000 to 2009) we had. We first made a fouryear moving average adjustment using the 2000 to 20009 dataset and then extracted the 2008 data for
the purpose of this application.
Table 2: Original dataset from CSA of Ethiopia for the year 20082
Industrial group
Food and beverages
Textiles
Wearing apparel
Tanning, leather &footwear
wood
Paper and printing
Chemicals
Rubber and plastics
Non-metals
Basic iron and steel
Fabricated metals
Machinery and equipment
Motor vehicles
Furniture
(I)labor
35391.1458
18363.7916
4815.79165
7887.5625
1959.81248
7967.64585
6426.50001
7083.16665
10300.75
1821.60418
2482.89584
1303.5
1214.41666
4529.18749
(I)capital
1788168221
381057469
104353732
363361747
6871141.31
180550018
286834161
361385672
677356667
212223001
88309646.7
92018536.2
60968765.8
93204031.8
(I)intermediate
1495232246
395188377
65318032.1
561064968
20176600.1
320157164
482231522
417657829
576183151
633340164
189549991
67075028.3
233431786
87966624.3
(O)production
4068003497
692187774
129750825
890647967
56729660.1
638558017
818780598
727078679
1377392696
1002609408
342615766
116429874
327031850
189933023
Table 3: Summary statistics of DMUs in their respective industry (non-handicapped data)

Statistics
Minimum
Average
St. Dev.
Maximum
(I)Labour
(I)Capital
(I)Intermediate
(O)Production
1214.417
7967.698
9133.455
35391.15
6871141.31
335475915
454726832
1788168221
20176600.1
396040963
378329529
1495232246
56729660.1
812696402
1014420298
4068003497
5.3.Evaluation of DMU within each sector and projection

First, we utilized the scheme described in Section 2.1 to find sectorial frontiers and
projections. The development of the handicap model begins with utilization of the projected
inputs and output where all the DMUs are on their efficient frontier in their respective
industry. In Table 4, we provide the Min/Max of the projected inputs and output of DMUs in
their respective industry. The summary does not represent the Min/Max of the industry.
I and O stand for input and output, respectively. The value of capital, intermediate inputs and
output are measured in Ethiopian Birr (ETB), Ethiopian currency. The current exchange rate of ETB
against USD is about Birr 19.41 for 1 USD. Labor is measured by the number of annual temporary and
permanent workers. Moreover, to consider price changes in our study, inputs (except labor) and output
were deflated by their respective implicit sectorial GDP deflator at 2000 price.
2
10
Table 4: Summary statistics of the projected data of DMUs in their respective industry
Statistics
Minimum
Average
St. Dev.
Maximum
(I)Labour
(I)Capital
(I)Intermediate
(O)Production
942.0572
4461.147
4508.271
18689.76
3140337.5
158798402
226084166
870850327
18952775
285544233
219213081
821789723
57420822.6
871889484
1150917694
4544356485
5.4.Global evaluation of the projected DMUs

The efficiency scores in Table 4 were calculated according to the procedure outlined in
Section 2.2. After merging the projected inputs and output, we evaluated DMUs with the VRS
model from which the maximum score of DMUs in their respective sector was defined in
order to judge whether the industry belongs to the no-handicap or with-handicap group.
Accordingly, we see in Table 5 that the Wearing apparel, Tanning, leather and footwear, Paper
and printing, and Machinery and equipment industries are in the with-handicap group while
the remaining are in the no-handicap group.
Table 5: No-handicap and With-handicap groups before non-convex adjustment

k
Industry
Food and beverages
Textile
Wearing apparel
Tanning, leather and footwear
Wood
Paper and printing
Chemicals
Rubber and plastics
Not-metals
Fabricated metals
Motor vehicle
Furniture
max o
o 1, , nk
1
1
0.343418
0.909345
1
0.488351
1
1
1
1
1
0.447216
1
1
5.5.Handicaps
5.5.1. Sectorial inputs and outputs
Table 6 presents sectorial inputs and outputs aggregated from the VRS-projected DMUs in the
sector according to Section 3.1.
11
Table 6: Aggregates of inputs (output) of VRS-projected DMUs in the sector

Industry
Food and beverages
Textile
Wearing apparel
Wood
Paper and printing
Chemicals
Rubber and plastics
Not-metals
Fabricated metals
Motor vehicle
Furniture
(I)labor
18689.76
5477.866
5713.199
5005.315
2325.809
5693.758
2788.107
2374.122
6230.622
2052.191
942.0572
1046.282
1313.834
2803.14
(I)capital
870850327
37291624
139074482
186313400
3140337.5
88431187
82517926
174408612
379273000
74160812
46086368
60668080
50655770
30305709
(I)intermediate
821789723
191300303
112307222
469860107
18952775
267162146
442951464
174512573
490968503
223054964
392698767
59152487
262797476
70110758
(O)production
4544356485
363220956
220631685
800649505
57420822.6
614123385
1028226007
956744221
1782276024
460355315
659006974
114955660
397401292
207084441
5.5.2. Solving non-convex model

Here, we solve the non-convex nature of the data using the data in Table 5 and classify the
scale adjusted scores (SAS) of each sector. In Table 7 an a in the Cluster column
represents the non-handicapped group. We found that all with-handicap sectors belong
to non-convex (S-shaped) frontiers.
Table 7: No-handicap and With-handicap groups after non-convex adjustment
Industry
Food and beverages
Textile
Wearing apparel
Tanning, leather and
footwear
Wood
Paper and printing
Chemicals
Rubber and plastics
Not-metals
Fabricated metals
Machinery and
equipment
Motor vehicle
Furniture
(I)labor
18689.8
5477.87
5713.2
(I)capital
870850327
37291624
139074482
(I)intermediate
821789723
191300303
112307222
(O)production
4544356485
363220956
220631685
Cluster
a
a
WA
SAS
1
0.5931
0.768
5005.31
2325.81
5693.76
2788.11
2374.12
6230.62
2052.19
942.057
186313400
3140337.5
88431187
82517926
174408612
379273000
74160812
46086368
469860107
18952775
267162146
442951464
174512573
490968503
223054964
392698767
800649505
57420822.6
614123385
1028226007
956744221
1782276024
460355315
659006974
TLF
a
PP
a
a
a
a
a
0.9923
1
0.9715
1
1
0.7429
0.664
1
1046.28
1313.83
2803.14
60668080
50655770
30305709
59152487
262797476
70110758
114955660
397401292
207084441
ME
a
a
0.5433
0.6274
0.6622
12
Note: WA= Wearing apparel, TLF=Tanning, leather and footwear, ME= Machinery and
equipment
5.6.Handicap
The final (after non-convex adjustment) handicap scores are reported in Table 8.
Table 8: Handicap score after non-convex adjustment
Industry
Handicap
Food and beverages
1
Textile
1
Wearing apparel
0.768
0.9923
Wood
1
Paper and printing
0.9715
Chemicals
1
Rubber and plastics
1
Not-metals
1
Fabricated metals
1
1
0.5433
Motor vehicle
1
Furniture
1
5.7. Global issue

5.7.1. Input (output) under handicap
In this section, we apply the scheme outlined in Section 4.1 to obtain the handicapped data.
Since we are using the input-oriented model, the inputs of the original dataset of each sector
were multiplied by the handicap scores given in Table 8 to arrive at the data in Table 9.
13
Table 9: Handicapped Inputs and output

Industry
Food and beverages
Textile
Wearing apparel
Wood
Paper and printing
Chemicals
Rubber and plastics
Not-metals
Fabricated metals
Motor vehicle
Furniture
(I)labor
35391.15
18363.79
3698.528
7826.828
1959.812
7740.568
6426.5
7083.167
10300.75
1821.604
2482.896
708.1915
1214.417
4529.187
(I)capital
1788168221
381057469
80143666.2
360563862
6871141.31
175404343
286834161
361385672
677356667
212223001
88309646.7
49993670.7
60968765.8
93204031.8
(I)intermediate
1495232246
395188377
50164248.7
556744768
20176600.1
311032684
482231522
417657829
576183151
633340164
189549991
36441862.9
233431786
87966624.3
(O)production
4068003497
692187774
129750825
890647967
56729660.1
638558017
818780598
727078679
1377392696
1002609408
342615766
116429874
327031850
189933023
5.7.2. Solving the CRS model

So far, we have been adjusting the original data to account for the handicapped industry. In
Table 10, we report the efficiency scores of each industry obtained after making handicap
adjustments. To obtain the efficiency scores reported in Table 9, we used the dataset given in
Table 9.
14
Table 10: Efficiency score with handicap model

DMU
Food and beverages
Score
0.8357
Rank
6
Textile
0.5192
14
Wearing apparel
0.5727
13
0.5836
12
Wood
Paper and printing
0.8722
Chemicals
0.6640
Rubber and plastics
0.5899
11
Non-metals
0.7839
Fabricated metals
0.834
Motor vehicle
Furniture
0.6019
10
5.7.3. Comparisons with the no-handicap model

The scores in Table 11 were obtained using the original (non-handicapped) dataset reported in
Table 2. Figure 5 compares the scores from the no-handicap and with-handicap models where
the heading (H) indicates a with-handicap sector.
15
Table 11: Efficiency sore with no-handicap model

Industry
Food and beverages
Textile
Wearing apparel
Wood
Paper and printing
Chemicals
Rubber and plastics
Non-metals
Fabricated metals
Motor vehicle
Furniture
WithHandicap
Score
1
0.5359
0.5037
0.5776
1
0.8918
0.6943
0.6216
0.9304
1
0.9215
0.5401
1
0.6627
Rank
1
13
14
11
1
7
8
10
5
1
6
12
1
9
NoHandicap
1.2
1
0.8
0.6
0.4
0.2
0
Figure 5: Comparison of handicap and no-handicap models

5.8.Observations
In Figure 5, we see that of the 3 handicapped industries, Wearing apparel, Tanning, and
Machinery and equipment have seen improvements in efficiency after the handicap adjustment
was made, with the machinery and equipment industry becoming efficient. There is a slight
decline in the efficiency score in the Paper and printing industry (handicapped) and the no-
16
handicap industries as compared to the no-handicap model. The decrease of efficiency scores
in paper and printing and other sectors might have been caused by the increase in the
machinery and equipment score (it has the smallest handicap and is now efficient). The
emergence of this efficient sector affects all other sectors.
5.9.Policy implications
Besides the contribution of this study to handicap modelling in DEA, its application to the
Ethiopian industrial sector has policy implications. From a policy perspective, the study of the
efficiency in the industrial sector may be helpful in allocating resources in a more efficient and
productive way. This study can also be used as an input for policy makers to formulate
policies that are sector specific by identifying the best and worst performing industries. For
instance, despite the strong belief that Ethiopia can be an industrial hub in labor-intensive light
manufacturing such as textile, wearing apparel, leather and leather products and furniture
industries, the performance of these sectors remained poor in the period under study. Many
factors such as poor input supply chain, lack of international exposure and poor technology
might have contributed to the poor performance of the mentioned sectors. Given that Ethiopia
has a comparative advantage in providing ample inputs to boost these sectors, we suggest that
the government needs to reform the input industries to achieve maximum efficiency in the
sectors. As Ethiopian domestic investors are less competitive in the international arena,
foreign direct investment may be helpful in bringing international experience, thereby
improving the efficiency and competiveness of the sectors. The poor performance of the
sectors might also indicate that firms in these sectors are using some older techniques of
production (outdated machinery) in the production process. Hence, improving the existing
techniques of production and/or introduction of modern machinery may improve the
efficiency level. In particular, Machine and equipment sector has the potential to be the top
industry, if innovation in inputs were executed.
6. Conclusions
In this paper, we proposed a handicap setting method for fair evaluation of industrial sectors
and applied it to Ethiopian industry. The manufacturing industry is composed of many sectors
which include many companies in the category. Thus, there is a two-layered structure. The
statistics of a sector is the sum of those of its member companies. In order to evaluate the
relative efficiency of industrial sectors, we need to take account of the performance of their
membership companies. For this purpose, we evaluated sectorial frontiers and projected
member companies to their respective frontiers. We then merged the projected companies and
found the meta-frontiers of all projected companies in the industry. If a member of a certain
sector is on the meta-frontiers, we classified this sector to the no-handicap group, whereas if
all members of a sector are off the meta-frontier, we classified the sector to the with-handicap
group. Then we applied the non-convex model proposed in Tone & Tsutsui (2013) for
deciding handicaps of with-handicap sectors. Most of the sectors belonged to no-convex (Sshaped) frontiers. We modify inputs (in the input-oriented case) or outputs (in the outputoriented case) using the handicaps and re-evaluate the sectorial efficiency. With respect to the
17
developing industry of Ethiopia, several sectors are in emerging fields. We found four sectors
belonged to the with-handicap group; (1) Wearing apparel (handicap=0.768), (2) Tanning,
leather and footwear (handicap=0.9923), (3) Paper and printing (handicap=0.9715), and (4)
Machinery and equipment (handicap=0.5433). The most handicapped sector is Machinery and
equipment. If this sector could be improved by innovation, this sector could become the top
sector, while the other three handicapped sectors remain inefficient even after taking account
of handicaps. The sectors in the no-handicap group could not increase their relative efficiency.
This might be caused by the emergence of Machinery and equipment sector as the most
efficient sector.
Further areas for research include cost, revenue and profit-related extensions of this approach.
References
Banker RD and Morey RC (1986). Efficiency Analysis for Exogenously Fixed Inputs and
Outputs. Operations Research 34(4): 513-521.
Dinh H, Palmade V, Chandra V, and Cossar F (2012). Light Manufacturing in Africa:Light
Manufacturing Private Investment and Create Job. World Bank.
Fixler DJ and Zieschang KD (1993). An Index Number Approach to Measuring Bank
Efficiency: an Application to Mergers. Journal of Banking and Finance 17 9943799450.
Olesen OB and Petersen NC (2009). Target and Technical Efficiency in DEA: Controlling for
Environmental Characteristics. Journal of Productivity Analysis 32: 27-40.
Sonobe T, Akoten J and Otsuka K (2009). An Exploration into the Successful Development
of the Leather-Shoe Industry in Ethiopia. Review of Development Economics 13(4):
719736.
Tone K (2001). A Slacks-based Measure of Efficiency in Data Envelopment Analysis.
European Journal of Operational Research 130: 498-509.
Tone K and Tsutsui M (2013). How to Deal with S-shaped Curve in DEA. Journal of
Optimization Theory and Applications (forthcoming), GRIPS Report 13-10. http://rcenter.grips.ac.jp/gallery/docs/13-10.pdf.
Yang Z and Paradi JC (2006). Cross Firm Bank Branch Benchmarking Using "Handicapped"
Data Envelopment Analysis to Adjust for Corporate Strategic Effects. Proceedings of
the 39th Hawaii International Conference on System Sciences pp 1-10.
18

Health
Health Care Simulation for Policy Assessment: Widening

Requirements Capture with Soft Methods
John Powell and Navonil Mustafee
University of Exeter Business School, Streatham Court, Rennes Drive Exeter, Devon,
EX4 4ST
j.h.powell@exeter.ac.uk, n.mustafee@exeter.ac.uk
Abstract
A simulation study consists of several well-defined stages, e.g., problem formulation, model
implementation and experimentation. The application of multiple techniques in the model
implementation stage is referred to as hybrid simulation, which we distinguish in this paper
from a hybrid M&S study, the latter referring to studies that apply methods and techniques
from disciplines like Operations Research, Systems Engineering and Computer Science to one
or more stages of a simulation study. We focus on the first stage of a simulation study (and by
extension a hybrid M&S study), viz., eliciting the system requirements, and conduct a review
of literature in Soft Systems Methodology for healthcare operations management. We discuss
the potential for the use of Qualitative System Dynamics as an additional soft OR method,
complementing (rather than supplanting) existing approaches, which can further aid the
understanding of the system in the problem formulation/conceptual modelling stage of a
Hybrid M&S study.
Keywords: Simulation; Health systems; Policy; Soft Systems
1.
Introduction
Simulation methods enable stakeholders to analyze and evaluate strategies for effective
management of complex systems, providing an experimental test bed or environment for
different stakeholders to assess the effectiveness of putative, sometimes emergent policies.
There are two separate areas of investigation which they aid, namely policy investigation and
cognitive investigation (after Powell and Coyle 2005). By policy investigation we mean the
exploration of the (predicted) effects of emergent, developing policy by interested parties.
Here the scope of the system under investigation, including performance measures, is held
constant; or is at least undebated. This allows exploration of a variety of policy options, which
may be operational or strategic in nature, aimed at managing the undebated system to produce
optimal, desirable or at worst acceptable results. In the simplest case there is one, dominant
system owner who has the authority, de iure or de facto to represent any other stakeholders in
assessing system output. More often there are a number of stakeholders vying for control over
the system performance through the application of the candidate policies.
In many cases, and in health systems as an example, the stakeholder set not only disagrees
about the desirability or optimality of system outputs, but about the very definition of the
19
J. Powell and N. Mustafee / OR56 Keynote Papers, pp.19-37 (2014)
system under scrutiny. In most cases these disagreements are cognitive in nature since the
divergence among the stakeholders perceptions of the system are shaped through their
unique experiences of working with the system. In other cases the differences are structural in
the sense that different stakeholders will define the system itself or parts of the system
differently from their points of view (and in rare cases existential), but we label them here
collectively as requiring cognitive exploration in contrast with the need to explore policy
options. In both cases simulation can assist in that exploration, but the cognitive case requires,
we submit, a degree of extra support.
Health systems have particular characteristics in terms of requirements capture, including an
unavoidable dissonance between the ontologies of the simulated system (positivist, single
reality) and that of the surrounding socially constructed, multiply-defined recipient and policy
contexts. Within healthcare studies, simulation techniques such as Monte Carlo Simulation
(MCS), Discrete-Event Simulation (DES), System Dynamics (SD) and Agent-based
Simulation (ABS) are in wide use (Brailsford et al. 2009; Katsaliaki and Mustafee 2011).
Frequently, however, these techniques have been applied in isolation one from another, in
dissonance with good system practice, (Jackson 2000) which stresses the usefulness of hybrid
methods in order to overcome the unavoidable limitations of any single approach. The search
for the best possible representation and analysis of the system under scrutiny has, then, led to
an increasing number of studies that combine simulation techniques/Operational Research
methods. These combined methods allow application of multiple techniques in the model
implementation stage of a simulation study, thereby enabling synergies across techniques to
engender improved insights. However, while this is a critically important phase, a M&S study
comprises several other well-defined stages (Maria 1997), for example, problem formulation
stage/conceptual modelling (Robinson 2011), data collection, validation and verification,
model execution, output data analysis and recommendations. It is, then, appropriate to explore
the use of multiple techniques in the wider perspective of a study. In this paper we distinguish
between hybrid simulation and a hybrid M&S simulation study, the latter referring to studies
that apply methods and techniques from disciplines like Systems Engineering, OR and
Computer Science to one or more stages of a simulation study (Section 2).
Next, we focus at the starting point of a M&S/hybrid M&S study with the objective of
identifying
commonly
used
multi-disciplinary
techniques
in
the
problem
formulation/conceptual modelling stage. We were able in a literature review to identify the
application of systems engineering approaches and tools (e.g., SysML: Systems Modeling
Language Eldabi et al. (2010)), problem structuring methods (PSM/soft OR) (e.g., Soft
Systems Methodology Lehaney and Paul (1994, 1996), Kotiadis, Tako, and Vasilakis (2014);
Group Model Building Brard (2010) and concepts from information systems (e.g., Issue
Maps/Issue-Based Information System (IBIS) Eldabi et al. (2010)) in the problem
formulation phase. Our review of literature also revealed the prominence of the Soft Systems
Methodology (SSM) (Checkland and Scholes, 1990) in healthcare studies in general,
especially those that involved multiple stakeholders in the decision making stage. We
therefore considered it pertinent to conduct a literature review on SSM in the context of its
application in the wider area of healthcare operations management (i.e., we were not restricted
to SSM-M&S studies alone).
20
The literature review identified the lack of use of complementary soft OR techniques (which
could potentially be used together with SSM) in the problem formulation stage of a hybrid
M&S study. In Section 5 we critique this segmented approach and argue the need for soft OR
techniques that not only provide insights to the wicked or messy problems but which further
support the development of the quantitative simulation models and could potentially be used
together with SSM. In this line of thought we discuss Qualitative system dynamics (QSD), a
modeling and analysis technique which bridges the numerical, positivist requirements of
simulation and the socially constructed world of its surrounding context, is a subset of System
Dynamics (SD), a well-known system representation and analysis method (Sterman, 2000). In
addition to its almost ubiquitous numerical form, SD has of recent years expanded its
embodiment into a qualitative form (Powell and Coyle, 2005) which in its analytical phase
replaces the detailed simulation-type system output with a structural analysis, concentrating on
predicting desirable managerial interventions by direct consideration of the dominant
dynamics of the system in focus. As such we find its ability to incorporate system knowledge
other than that of a numerical nature highly conducive to the requirements capture of
simulations in the present context.
The concluding section of our paper (Section 6) presents future work and draws the paper to a
close.
2.
Hybrid M&S Study
The existing literature in hybrid simulation comprises mainly discursive papers comparing
different techniques (e.g., Brailsford, Desai and Viana 2010), together with case studies,
integration frameworks and studies that have used these frameworks/framework extensions to
create hybrid models (e.g., Viana et al. 2014, Chahal and Eldabi 2010, Chahal and Eldabi
2008; Zulkepli, Eldabi, and Mustafee 2012). Specific techniques that have been applied
include: the combined application of SD and DES (Viana et al. 2014; Zulkepli, Eldabi, and
Mustafee 2012; Chahal 2010; Chahal and Eldabi 2008), hybrid M&S using ABS, SD and DES
(Djanatliev and German 2013; Viana et al 2012), combined application of ABS and load plan
construction heuristics (Mustafee et al. 2012), hybrid simulation with loosely coupled SD and
agent-based models (Djanatliev et al 2012) and application of integer programming with
simulation (Lee et al. 2012). The majority of these papers focus on the model implementation
phase of a simulation study.
Our review of literature provides evidence of the application of methods and techniques from
disciplines like Computer Science, Systems Engineering, OR, Information Systems and
Distributed Computing to the various other stages of a M&S study. For example, in problem
formulation stage approaches from soft OR, systems engineering and information systems
have been used (refer to the previous section for specific examples). For faster simulation
execution in the experimentation stage authors have used techniques from computer science,
e.g., General-Purpose computing on Graphics Processing Units (GPGPUs) (Perumalla 2006;
Park and Fishwick 2010), parallel and distributed simulation (Lendermann, Gan, and
McGinnis 2001; Mustafee et al. 2009) and distributed computing solutions like simulation
execution over desktop grids (Mustafee and Taylor 2009; Taylor et al. 2011) and, in the future,
cloud computing resources. Model formalisms based on Discrete Event System Specification
21
(DEVS) (e.g., Dynamic Structure Discrete Event System Specification (DSDEVS) - Barros
1995) and meta-modelling using UML (Traor 2003) have been used. Further, it is arguable
that statistical and computing approaches presently being developed in the field of business
intelligence, big data and analytics (Demirkan and Delen 2013) could be applied in M&S study
stages pertaining to the collection of input data and output data analysis.
Hybrid M&S Study

Problem Formulation / Conceptual Modelling
Cognitive Mapping
Input/Output
Data and
Analysis
Business
Intelligence
Big Data and
Analytics
Issue Mapping
Group Model Building
Model development (this

can be Hybrid Simulation)
SSM
...
Model
Formalism
Agent-based
Simulation
System
Dynamics
DEVS
Discrete-Event
Simulation
Monte Carlo
Simulation
UML
...
...
...
Experimentation
GPGPU
Execution
Parallel and Distributed

Simulation
Grid/Cloud-based
Simulation Experimentation
...
Figure 1: Conceptual representation of a Hybrid M&S study ( denotes other methods)

The discussion above indicates the multi-disciplinary nature of a M&S study. Unlike a hybrid
simulation that focuses on individual M&S techniques in the model implementation stage, a
hybrid M&S study recognizes and deploys the use of inter-disciplinary methods at various
other stages of a simulation study. Figure 1 shows our conceptualization of a hybrid M&S
study, identifying a number of inter-disciplinary methods that have been used (or can
potentially be used) in specific stages of a M&S study. Our conceptual representation is not
exhaustive (indeed not all stages of a simulation study are depicted; stages pertaining to input
data and output data analysis have been combined; model formalism has been introduced as a
stage). Figure 1 includes the model development stage in the center and depicts four
simulation techniques which can be used either in isolation (as in the conventional studies) or
can be combined to implement a hybrid simulation. The techniques are represented in grey
boxes so as to distinguish them from inter-disciplinary methods (in white boxes) that are/can
be applied to stages other than simulation model development, such as Model
Conceptualization, Input Data, Output Data Analysis, Simulation Experimentation and Model
Formalism.
22
Multiple Simulation
Techniques
Single Simulation
Technique
Model Implementation Stage of

Simulation Study
Not a Hybrid M&S Study
Hybrid M&S Study
Hybrid Simulation
Hybrid Simulation
Not a Hybrid M&S Study
Hybrid M&S Study
Not a Hybrid Simulation
Not a Hybrid Simulation
2
Well-defined methods from other
disciplines have been used in one
or more stages of the study
Methods from other disciplines have

not been used
Stages of a Simulation Study

(apart from model implementation)
Figure 2: Hybrid M&S study versus Hybrid Simulation
Figure 2 clarifies the differing scope of hybrid simulation vis--vis hybrid M&S. It shows that
a hybrid M&S simulation study will apply well-defined methods from disciplines outside
M&S in one or more stages of the study (Quadrant 2; Figure 2). A hybrid M&S study will also
be a hybrid simulation when multiple simulation techniques have been used in the model
implementation stage (Quadrant 1; Figure 2). However, an implementation of hybrid
simulation without the application of inter-disciplinary methods in the wider study will
disqualify it from being a hybrid M&S study (Quadrant 4; Figure 2). Quadrant 3 represents the
traditional studies which have used only one modelling technique and in which methods from
other disciplines have not been used.
3.
Methodology
In order to identify papers for our methodological review of literature we conducted a search
using the ISI Web of Science (WOS) database. WOS is one of the largest databases of
quality academic journals and provides access to bibliographic information pertaining to
research articles published from 1900 onwards (for Science Citation Index Expanded). It
indexes approximately 9800 high impact research journals spread across more than 200
disciplines. Further, WOS now includes two conference proceedings citation databases, for
Science (CPCI-S) and Social Science & Humanities (CPCI-SSH) respectively, which indexes
the most significant academic conferences (from 1990 onwards) in a wide range of disciplines.
We used various combinations of keywords and varied parameters like specific WOS
databases used and timespan for the search, reviewed the retrieved results and subsequently
decided on the specific search parameters for the study. We used the keywords ("soft system*
method*") AND (health*), with * being the wildcard character which allowed us to use the
23
keyword derivatives in the search condition, to conduct a topic search on title, abstract, author
keywords and keyword plus. All WOS databases were included in the search. We did not
restrict the year of publication and selected timespan as all years. It is to be noted that we did
not use the acronym SSM since our preliminary searches using this keyword resulted in a
number of irrelevant hits. We subsequently tried to apply a search condition which would
restrict results to SSM in the context of healthcare, however, this did not produce the desired
outcome since several papers on Superficial Spreading Melanoma (which is the most common
form of skin cancer) were retrieved.
Our search criteria retrieved a total of 68 papers for the initial analysis. The next stage
involved reading the abstracts and identifying papers that applied SSM in the context of
healthcare operations management. Thus we excluded, for example, four papers that used
SSM for designing healthcare information systems (including papers on eliciting user
requirements for such systems) and further two papers on using SSM for planning and
management of information security in healthcare, one paper on racial equality in health
provisions, one paper on SSM-based evaluation of electronic library for healthcare
professionals and one paper on using SSM to identify impact and barriers to information
access and use in health. 49 papers were selected subsequent to the abstract reading phase of
the study and this comprises the corpus of scholarly articles for the literature review part of
this paper.
4.
Review of Literature in SSM
In this section we present some preliminary findings from our literature review, emerging
from a key word search on soft systems and health systems and which will be reported upon in
a later publication. The variables captured in analyzing the articles included, the aim of the
study, the specific application context, the country in which the SSM study was conducted,
whether SSM was used as per the original methodology (e.g., the seven stage SSM process;
(Checkland 1981)) or extensions/improvements to the methodology were proposed, software
used (e.g., for the purposes of data analysis, illustration of rich pictures), the SSM concepts
and tools that were used (e.g., holons, rich pictures, root definition, CATWOE, conceptual
model, etc.), the primary stakeholders, the composition of the stakeholder group who
contributed to the development of the conceptual model(s) and/or who were responsible for
recommending and/or implementing the results through action research, outcome of the
study/findings, how SSM contributed to the outcome of the study, and whether the SSM study
included elements of mixed methodology (considering the scope of this paper, this variable is
of particular interest to us and will be further explored). Further to this, for studies which had
used SSM in the context of M&S, we collected variables describing the type of simulation
model (e.g., DES, SD) and whether the paper included the implementation aspects of the
aforementioned model.
Approximately 80% of the 49 papers selected for review (40 papers) have been published in
journals (one exception to this is a poster which was published in the journal Tropical
Medicine & International Health); the remaining nine papers were published as conference
proceedings with Winter Simulation Conference accounting for six papers. In relation to
24
journals, 11 articles have been published in the Journal of the Operational Research Society,
followed by three articles each in the Journal of Advanced Nursing and Health Services
Management Research (refer to Table 1).
It is to be noted here that our dataset included review papers (paper number 10 and 38 see
Table 1), papers which proposed methodological extensions to SSM (17, 48) and frameworks
(28), discussion and viewpoint papers (27, 41), and empirical papers reporting on SSM
studies. With regard to the latter, most of the studies were based in the UK (20 studies); this is
hardly surprising since the founder of SSM (Peter Checkland) and his colleagues were based
in Lancaster University (UK) and spent decades working on approaches that finally
culminated in the formalization of SSM as a problem structuring approach. Further, it is
widely recognized in the international OR/M&S community that Britain has significant
strengths in soft OR (and applications in healthcare) (EPSRC/ESRC, 2004). Studies that
were conducted outside the UK include, three studies in USA (in Chicago - 31, multiple
regions - 38, 35), one study each in Australia (State of Victoria) (18), South Africa (KwaZulaNatal) (19), Ireland (23), Norway (33), Brazil (36), Greece (37), Iran (39) and Canada (41).
A few studies have proposed an extension to SSM. (12) introduce SSM+ that consists of 16
stages and is positioned for the use of policymakers, clinicians and managers in health care
circles. (32) propose the extension of SSMs graphical elicitation with graphical modelling
approaches from other disciplines. (40) propose the inclusion of Performance Measurement
Model (PMM) in SSM. However, the majority of the studies have either religiously adhered
to the seven stage process of SSM (e.g., 25, 20) or have adopted a subset-of SSM concepts and
tools (e.g, there are references to using SSM as a guiding principle, using SSM-inspired
tools, using modified version of the methods from SSM). Our dataset also consisted of at
least eight papers that claimed to have conducted a SSM-study, however, the process itself
was not adequately described/the use of SSM was not apparent, and in some cases only a
general reference to SSM was made. It is arguable that a couple of these studies used the term
SSM as an umbrella term since their case studies included multiple stakeholders with a need to
form consensus/cooperatively implement actions. We had two papers in which a plan for
action was proposed, i.e., the SSM study was not yet implemented and the authors articulated
the future course of action (44, 35).
Out next analysis is on context of application or the specific theme of study. Some of the
themes we identified included, use of SSM for professional and organizational learning
including development of curriculum in health sciences (15, 21, 33), evaluation of decisionmaking in inter-agency/intra-agency, interdisciplinary/unidisciplinary teams (16, 34), use of
SSM for improving patient care (18, 23, 30), critical systems inquiry into the complexities of
implementing healthcare provision (19, 20, 25), use of SSM for conceptual modelling (32), the
use of SSM for multi-agency planning of disaster (12), etc.
Our final analysis focuses on the complementary methods that have been used together with
SSM in the development of conceptual models (it is to be noted that the development of the
conceptual model is the fourth step in the seven stage SSM process; for a M&S study that uses
SSM, the SSM conceptual model informs the conceptual model of the simulation to be
developed). Of the 49 studies in our literature review, we could find evidence of multiple
25
methods in only a handful of studies. The study by (16) advocated two approaches for
evaluating the success of integrated, multiprofessional healthcare teams in the UK SSM and
a pluralistic framework; however, these approaches were not used together but in two separate
case studies. Thus it is arguable that this particular study has not implemented a mixed
methodology. The study by (19), however, contributes to the growing literature on the
combination of systems methods by illustrating how conceptual models of SSMs purposeful
human activity were constructed from the participatory use of Concept Maps and Sign-graph
Diagrams. The study by (26) has used SSM with a project management framework
(PRINCE2); this is an example of using a concept from Operations Management (project
management) together with soft OR. (31) has used a social ecological model together with
SSM, wherein the former was used to understand the interaction between various personal and
environmental factors, and the latter helped understand the multiple perspectives of
stakeholders which in this case were also a source of resistance to change and conflict. In
(34), appreciative inquiry was combined with SSM to form an evaluation approach that was
designed to capture individual as well as organizational learning. In (35) the authors developed
models of workflow and information flow using Hierarchical Task Analysis and SSM. Thus,
only a handful of studies have explored the use of complementary approaches that can be used
together with SSM, and this, we believe, is an area of future research.
Table 1: Papers selected for the literature review

N
1
Authors
Title of the Paper
Source Title
Year
Lauri S
Using a computer simulation program to assess

the decision-making process in child health care.
Computers in
nursing
1992
Lehaney,B
et al.
Using soft systems methodology to develop a

simulation of out-patient services
Journal of
the Royal
Society of
Health
1994
Wells JSG
Discontent without focus? An analysis of nurse

management and activity on a psychiatric inpatient facility using a 'soft systems' approach
Journal of
Advanced
Nursing
1995
Hindle T et
al.
Developing a Methodology for Multidisciplinary

Action Research: A Case Study
JORS
Lehaney B
et al.
Simulation modelling for resource allocation and

planning in the health sector
Journal of
the Royal
Society of
Health
1995
Lehaney B
The use of soft systems methodology in the

development of a simulation of out-patient
JORS
1996
26
1995
et al.
services at Watford General Hospital
Hernando S
Promoting library services to qualified nurses:

towards a market-led approach.
Health
libraries
review
1997
Connell
NA et al.
Patient-centred performance monitoring systems

and multi-agency care provision: a case study
using a stakeholder participative approach.
Health
services
management
research
1998
Hindle T et
al.
Specialty location decisions in the reformed NHS:

a case study.
Health
services
management
research
1998
10
Jun JB et
al.
Application of discrete-event simulation in health

care clinics: A survey
JORS
11
Lehaney B
et al.
A case of an intervention in an outpatients

department
JORS
12
Gregory
WJ et al.
Planning for disaster: developing a multi-agency

counselling service
JORS
Loo GS et
al.
A Soft Systems Methodology model for Clinical

Decision Support Systems (SSMM-CDSS)
Workshop on
Database &
Expert
Systems
Applications
13
1999
1999
2000
2001
14
Connell
NAD
Evaluating soft OR: some reflections on an

apparently 'unsuccessful' implementation using a
Soft Systems Methodology (SSM) based approach
15
Clarke CI
et al.
Professional and organizational learning:

Journal
analysing the relationship with the development of of Advanced
practice
Nursing
Cook G et
al.
Decision-making in teams: issues arising from

two UK evaluations.
Journal of
Interprofessional
Care
2001
Braithwaite
J et al.
Introducing soft systems methodology plus

(SSM+): why we need it and what it can
contribute.
Australian
Health
Review
2002
16
17
27
JORS
2001
2001
O'Meara P
Would a prehospital practitioner model improve

patient care in rural Australia?
Emergency
Medicine
Journal
2003
Luckett et
al.
A critical systems intervention to improve the

implementation of a district health system in
KwaZulu-Natal
Systems
Research and
Behavioral
Science
2003
Kalim K et
al.
The role of soft systems methodology in

healthcare policy provision and decision support
Conference
on Systems,
Man &
Cybernetics
2004
Stokes PJ
et al.
Information-seeking behaviour of nurse teachers

in a school of health studies: a soft systems
analysis
Nurse
Education
Today
2004
Fahey DK
et al.
Applying systems modelling to public health
Systems
Research and
Behavioral
Science
2004
Wells JSG
Hospital-based industrial therapy units and the

people who work within them - an Irish case
analysis using a soft-systems approach
Journal of
Psychiatric
and Mental
Health
Nursing
2006
Kotiadis K
et al.
Combining PSMs with hard OR methods: the

philosophical and practical challenges
JORS
Kalim K et
al.
An illustration of whole systems thinking.
Health
services
management
research
26
Checkland
P et al.
Process and content: two ways of using SSM
JORS
27
Kuljis J et
al.
Can health care benefit from M&S methods in the

same way as business and manufacturing has?
WSC 2007
28
Eldabi T et
al.
Towards a framework for healthcare simulation
WSC 2007
29
Sachdeva R
Mixing methodologies to enhance the
JORS
18
19
20
21
22
23
24
25
28
2006
2006
2006
2007
2007
2007
30
31
32
33
et al.
implementation of healthcare operational research
Reed Jan et
al.
Specialist nurses for older people: implications

from UK development sites
Journal of
Advanced
Nursing
2007
SuarezBalcazar Y
et al.
Introducing systems change in the schools: the

case of school luncheons and vending machines
American
Journal of
Community
Psychology
2007
Kotiadis K
et al.
Conceptual modelling: Knowledge acquisition

and model abstraction
WSC 2008
Fougner M
et al.
I had a rich picture ... : Insights into the use of

"soft" methodological tools to support the
development of interprofessional education
Journal of
Interprofessional
Care
2008
Carr Susan
M et al.
Leadership for health improvement-implementation and evaluation.
Journal of
health
organization
and
management
2009
Journal of
the American
Medical
Informatics
Association
2009
34
Unertl KM
et al.
Describing and Modeling Workflow and

Information Flow in Chronic Disease Care
35
36
37
38
39
Tako A et
al.
A participative modelling framework for

developing conceptual models in healthcare
simulation studies
WSC 2010
Eldabi T et
al.
Model driven healthcare: Disconnected practices
WSC 2010
Mingers J
et al.
A review of the recent contribution of systems

thinking to operational research and management
science
EJOR
Ribesse N
et al.
Technical assistance for health system

strengthening in fragile state: learning from
experiences through systemic approach
Tropical
Medicine &
International
Health
29
2008
2010
2010
2010
2011
40
Dalkin SM
et al.
Understanding integrated care pathways in

palliative care using realist evaluation: a mixed
methods study protocol
BMJ OPEN
2012
41
Barra M et
al.
Using a Soft Systems Methodology framework to

guide the conceptual modeling process in discrete
event simulation
WSC 2012
2012
Sgourou E.
et al.
Using Soft Systems Methodology as a systemic

approach to safety performance evaluation
Symposium
on Safety
Science &
Technology
2012
Hodges S
et al.
If We're Going to Change Things, It Has to Be

Systemic: Systems Change in Children's Mental
Health
American
Journal of
Community
Psychology
2012
44
Heyrani A
et al.
Clinical governance implementation in a selected

teaching emergency department: a systems
approach
Implementati
on Science
2012
45
Kotiadis K
et al.
Using a model of the performance measures in

Soft Systems Methodology (SSM) to take action:
a case study in health care
JORS
2013
46
Price M et
al.
Provider connectedness and communication

patterns: extending continuity of care in the
context of the circle of care
BMC Health
Services
Research
2013
Sinclair E
et al.
Developing stroke-specific vocational

rehabilitation: a soft systems analysis of current
service provision
Disability
and
Rehabilitatio
n
2014
48
Kotiadis K
et al
A participative and facilitative conceptual

modelling framework for discrete event
simulation studies in healthcare
JORS
2014
49
Vandenbro
eck P et al.
Recommendations for the organization of mental

health services for children and adolescents in
Belgium: Use of the soft systems methodology
Health
Policy
2014
42
43
47
5.
Complementarity requirements
It is clear from our review of the literature that there is little evidence of the mobilization of
complementary soft OR techniques in the problem formulation/conceptual modelling stage of
30
the hybrid M&S approach. There are dangers in such a segmented approach, particularly in
respect of the need for inclusion in the model building (at the requirements capture stage) of:
a) The maximally available information about the behavioural dynamics of the system
under consideration, particularly as the very raison detre of a simulation (at least in
this context) is accurately to extrapolate the dynamic effects of system mechanics as
conditioned by the proposed policies being assessed. It would seem wise at the
requirements capture stage to maximize this input and indeed to confirm anticipated
dynamics with informed parties. If this is done in an explicit manner there is an
additional advantage of being able to validate, together with the stakeholder group, the
functional behaviour of the simulation on which the whole assessment turns.
b) The value perspectives of stakeholders other than the system owner in respect of the
benefits or otherwise of the results of the simulation. At first glance it may seem that
the multiple valuation of system outputs can be respected ex post facto, by giving voice
and weight to the system outputs in the weighing of the benefits of the proposed policy
or policies. To some extent this can be done, but there is a danger that in declaring and
designing the output representations within the simulation, that implicit valuations are
made inadvertently, rendering less visible, or even invisible, factors valued by a
stakeholder group which happens to be under-represented in the requirements capture
process. To decode; modelers run the risk of ignoring the legitimate valuations of
interested but silent parties.
c) Differing definitions of the system in focus. Conventional requirements that capture
processes and techniques are focused on the establishment of a single simulation
definition, not least because this makes subsequent model-building easier. While
clarity in the simulation definition is much to be valued, work in a variety of
embodiments of the soft system philosophy (Checkland, Jackson, Oliga, Coyle )
shows that the single clarified reality of a model representation is often, in fact, a
misrepresentation, an over-simplification of a multiple reality produced by the varying,
equally valid cognitions and ontologies of stakeholders. Again, to decode: modelers
run the risk of assuming that while various stakeholders may have different valuations
of system outputs, that they nevertheless agree on the definition of the system each is
trying to optimize. This is a risky assumption; the various valuations of stakeholders
often lead them into differing cognitions of the system and therefore of its definition.
At best the single simulation definition is the result of the interplay of power and
communicative capacities of the stakeholders. Now, the nature of the simulation
component of the overall assessment system is such that a single requirement
(definition of the simulation) is needed at any one time, so that the multiple stakeholder
definitions of the system must be dealt with accordingly. There are two options: one is
to reconcile differences in system definition among the stakeholders and the other is to
run variations of the simulations to reflect the varying flavours of system perceived
by the stakeholders. In respect of the reconciliation approach, while this is much to be
favoured in terms of the necessary resources applied to simulation (only one simulation
is needed) we note that the resolution of stakeholders views may be different during
the requirements capture phase from the policy assessment phase. This adds further
31
weight to the need for an overall assessment process which allows for policy
consequences to be included in the early stages of simulation requirements capture.
These perspectives on the plural, even arbitrary nature, of the social and political contexts
surrounding a Human Activity System (HAS) (Checkland 1981), of which health systems are
clear examples, are, of course, widely offered. There is a clear ontological disparity between
the hard, positivist, physical reality of, for example, a supply chain and the socially
constructed, politicized, multi-perspective view of the degree of benefit provided by a health
delivery system. In fact the whole issue of the reconciliation of these ontologies can be argued
to be the concern of the whole corpus of work known as soft OR (Ulrich, 1988; Checkland,
1981; Mingers and Lyons, 2009). The clash of ontologies, then, is well-known and therefore
what is worthy of note here is that there is so little practical mobilization of this corpus of soft
OR methods directly into the world of simulation, particularly at the requirements capture
stage, particularly when the influence of SSM and Strategic Options Development Analysis
(SODA) (Eden and Ackermann 2004) being effective broad cognitive and problem solving
tools, is so great within health system studies.
The explanation, we assert, lies in the level at which systems in focus are treated by the soft
OR methods. The strength of these approaches are in their capacity to treat the broad problem
in a broad manner. It is not their objective to make precise the mechanics of the underlying
system which is the legitimate and complementary focus of the simulation. We hasten to add
that this shortcoming, seen from the point of view of the simulation worker, should not be seen
as a criticism of the soft OR arsenal, any more than a field gunner can be criticized for not
being a sniper, but, from the point of view of achieving helpful, applicable, representative
requirements on which basis simulations can be constructed, an interlocution between the
powerful but broad, multivalent and plural structures of soft OR and the precise, but narrowly
defined structures of simulation would appear highly desirable.
The three issues detailed above provide some basis for declaring the characteristics of a
complementary approach through which simulation requirements analysis, particularly in the
health studies area, can be improved. It needs to be able to
a. Represent and capture expert knowledge of expected system dynamics
b. Represent and capture available expert knowledge of the system structures underlying
those dynamics
c. Allow the co-existence of alternative/conflicting evaluations of system behaviour
d. Allow the co-existence of alternative/conflicting system structures
While a number of the well-known techniques of soft OR, including SSM, SODA (Eden,
1989), drama theory, scenario planning, general causal mapping, theory of constraints and
benefits modeling (Vidal, 2005) provide very powerful insight into the nature and
characteristics of complex problems (often referred to as wicked or messy problems) they
invariably do so not by detailed modeling of the dynamics of a system (as in simulation and
quantitative Systems Dynamics) but by more discursive, verbal, often diagrammatic
32
representations. Two stand out from the field in respect of their ability to support directly the
simulation activities. These are SODAs cognitive mapping and QSD, the qualitative form of
SD, and particularly the variation known as Qualitative Politicised Influence Diagrams (QPID)
(Liddell and Powell 2005; Swart and Powell 2006; Powell and Swart 2010). We do not reject
SODA as a basis for requirement capture here, but offer the observation that because of QSDs
provenance as emerging from the corpus of quantitative SD, it has a natural structural
connection with the vocabulary and representation methods used by SD to scope and specify
its form of simulation, and hence provides a natural interface with the needs of simulation in
the health systems area in particular.
6.
Conclusion and Future Work
Our observations on the literature have led us to consider the relationship between the various
components of a policy assessment regime. We observe a complementarity between broad
policy assessment methods which have to take into account multiple, often ill-defined views
and more focused and precise system representations through use of simulations, a
complementarity which is, perhaps necessarily, an uneasy one. The simulation seeks precision
in requirement, the broad policy method seeks engagement, breadth of representation and
generality, but in order to support the process of sound policy identification and hence
assessment, these two must work together.
We observe a lack of deployment of wider soft OR methods in simulation requirement capture
leading potentially to simulation designs which are sub-optimal with respect to the inclusion
of the widest possible sources of system knowledge and benefit valuations of the system
outputs. While soft system methods are seen as uniquely powerful in eliciting multiple
viewpoints from stakeholders and system owners, we encourage the deployment of those
system representation methods such as QSD which can accommodate multiple viewpoints but
in such a way as to communicate directly with the simulation requirements process. Work is in
hand to establish the practicality of using the toolset of QSD (Powell and Coyle 2005) not only
in the area of health system representation and assessment, but in the associated fields of
(safety and mission) critical human activity systems (CHASs) such as disaster and crisis
management and high risk environments.
References
Barros F J (1995). Dynamic structure discrete event system specification: a new formalism for
dynamic structure modeling and simulation. In: Alexopoulos C, Kang K, Lilegdon W R,
and Goldsman D (eds). Proceedings of the 27th conference on Winter simulation,,. IEEE.
pp781-785
Brard C (2010). Group model building using system dynamics: an analysis of methodological
frameworks. Electronic Journal of Business Research Methods, 8(1):35-46.
Brailsford S C, Harper P R, Patel B, and Pitt M (2009). An Analysis of the Academic
Literature on Simulation and Modelling in Healthcare. Journal of Simulation 3(3):130140.
33
Brailsford SC, Desai S M, and Viana J (2010). Towards the Holy Grail: combining system
dynamics and discrete-event simulation in healthcare. In: Johansson B, Jain S, MontoyaTorres J, Hugan J, and Ycesan E (eds). Proceedings of the 2010 Winter Simulation
Conference.Baltimore, MA, INFORMS.
Chahal K (2010). A Generic Framework for Hybrid Simulation in Healthcare. PhD Thesis.
West London: Brunel University.
Chahal K and Eldabi T (2008). Applicability of Hybrid Simulation to Different Modes of
Governance in UK Healthcare. In: Mason,S, Hill J R R, Monch L, Rose O, Jefferson T,
and Fowler J W (eds). Proceeding of the 2008 Winter Simulation Conference,
Piscataway, New Jersey: IEEE. pp1469 1476.
Chahal, K., and Eldabi T (2010). A generic framework for hybrid simulation in healthcare. In
Proceedings of the 28th international conference of the system dynamics society, 2529
July 2010, Seoul, Korea. pp. 526541.
Checkland P (1981). Systems Thinking Systems Practice, John Wiley & Sons Ltd: Chichester,
UK.
Checkland P and Scholes J. (1990). Soft systems Methodology in Action. John Wiley & Sons
Ltd: Chichester, UK
Demirkan H and Delen D (2013). Leveraging the capabilities of service-oriented decision
support systems: Putting analytics and big data in cloud. Decision Support Systems, 55(1):
412-421.
Djanatliev A and German R (2013). Prospective Healthcare Decision-Making by Combined
System Dynamics, Discrete-Event and Agent-Based Simulation. In: Pasupathy R, Kim SH, Tolk A, Hill R and Kuhl M E (eds). Proceedings of the 2013 Winter Simulation
Conference, Piscataway, New Jersey: IEEE. pp 270-281
Djanatliev A, German R, Kolominsky-Rabas P and Hofmann B M (2012). Hybrid Simulation
with Loosely Coupled System Dynamics and Agent-Based Models for Prospective Health
Technology Assessments. In: Laroque C, Himmelspach J, Pasupathy R, Rose O and
Uhrmacher A M (eds). Proceedings of the 2012 Winter Simulation Conference
Piscataway, New Jersey: IEEE. pp 770-781.
Eden C (1989). Using cognitive mapping for strategic options development and analysis
(SODA). In: Rosenhead J (ed) Rational Analysis for a problematic world, John Wiley &
Sons Ltd: Chichester, UK
Eden C and FAckermann F (2004). Cognitive mapping expert views for policy analysis in the
public sector. European Journal of Operational Research, 152(3):615-630.
Eldabi T, Jun T, Clarkson J, Connell C, and Klein J (2010) Model-driven health care:
Disconnected Practices. In: Johansson B, Jain S, Montoya-Torres J, Hugan J, and Ycesan
34
E (eds) Proceedings of the 2010 Winter Simulation Conference Baltimore, Maryland:

INFORMS
EPSRC/ESRC (2004). "Review of research status of operational research in the UK.."
Available
at
http://www.epsrc.ac.uk/SiteCollectionDocuments/Publications/reports/OpResReview.pdf
(last accessed 27 May 2014).
Jackson M C (2000). System Approaches to Management. New York, NY: Springer.
Katsaliaki K and Mustafee N (2011). Applications of Simulation Research within the
Healthcare Context. Journal of the Operational Research Society, 62(8):1431-1451.
Kotiadis K., Tako A, and Vasilakis C (2014). A participative and facilitative conceptual
modelling framework for discrete event simulation studies in healthcare. Journal of the
Operational Research Society, 65(2):197-213.
Lee T, Cho S-H, Jang H, and Turner J G (2012). A simulation-based iterative method for a
trauma centerAir ambulance location problem. In: Laroque C, Himmelspach J, R
Pasupathy, Rose O and Uhrmacher A M (eds) Proceedings of the 2012 Winter Simulation
Conference, Piscataway, New Jersey: IEEE. pp 955-966
Lehaney B. and Paul R J (1996). The use of soft systems methodology in the development of a
simulation of out-patient services at Watford General Hospital. Journal of the Operational
Research Society, 47(4):864-870.
Lehaney B and Paul R J (1994). Using soft systems methodology to develop a simulation of
outpatient services. Journal of the Royal Society for the Promotion of Health 114(5):248
251.
Lendermann P, Gan B P, and McGinnis L F (2001). Distributed simulation with incorporated
APS procedures for high-fidelity supply chain optimization. In: Peters B A, Smith J S,
Medeiros D J and Rohrer M W (eds) Proceedings of the 33rd Winter Simulation
Conference. IEEE. pp 11381145
Liddell W and Powell J H (2004). Agreeing access policy in a general medical practice: a
case study using QPID. System Dynamics Review 20(1):49-73.
Maria A. (1997). Introduction to modeling and simulation. In: Andradottir S, Healy K J,
Withers D H, and Nelson B L (eds) Proceedings of the 29th conference on Winter
simulation. IEEE. pp 7-13
Mingers J and Lyons L (2009). A review of the recent contribution of system thinking to
operational research and management science. Working Paper series no 197, University of
Kent Business School, Kent, UK. ISSN 1741-7597.
35
Mustafee N, Taylor S J E, Katsaliaki K, and S C Brailsford (2009). Facilitating the Analysis of

a UK National Blood Service Supply Chain Using Distributed Simulation. Simulation:
Transactions of the Society of Modelling and Simulation International, 85(2):113-128.
Mustafee N and Taylor S J E (2009). Speeding Up Simulation Applications Using WinGrid.
Concurrency and Computation: Practice and Experience 21(11):1504-1523.
Mustafee N, Katsaliaki K, Bischoff E E, and Williams M D (2012). Proximity-based
modelling of cross-contamination through agent-based simulation: A feasibility study.
Health Systems 2(1):61-71.
Park H. and Fishwick P A (2010). A gpu-based application framework supporting fast
discrete-event simulation. Simulation, 86(10): 613-628.
Perumalla K S (2006). Discrete-event execution alternatives on general purpose graphical
processing units (GPGPUs). In: Proceedings of the 20th Workshop on Principles of
Advanced and Distributed Simulation pp 74-81. IEEE Computer Society.
Powell J H and Coyle R G (2005). Identifying strategic action in highly politicized contexts
using agent-based qualitative SD. Journal of the Operational Research Society 56(7):787
798.
Powell J H and Swart J (2010). Mapping the values in B2B relationships. Industrial Marketing
Management 39(3):437439.
Swart J and Powell J H (2006). Men and measures: capturing knowledge requirements in
firms through qualitative system modeling. Journal of the Operational Research Society
57(1):10-21.
Robinson S (2011). Choosing the right model: conceptual modeling for simulation. In: Jain S,
Creasey R R, Himmelspach J, White K P, and Fu M (eds) Proceedings of the 2011 Winter
Simulation Conference, Phoenix, Arizona. IEEE pp 1423-1435
Sterman J D (2000). Business Dynamics: Systems thinking and modeling for a complex
world. Boston, MA: McGraw-Hill/Irwin.
Traor M K (2003). Foundations of multi-paradigm modeling and simulation: a meta-theoretic
approach to modeling and simulation. In: Chick S, Snchez P J, Ferrin D, and Morrice D J
(eds) Proceedings of the 2003 Winter Simulation Conference, IEEE.
Taylor S J E, Ghorbani M, Mustafee N, Turner S J, Kiss T, et al. (2011). Distributed
computing and modeling & simulation: Speeding up simulations and creating large
models. In: Jain S, Creasey R R, Himmelspach J, White K P, and Fu M (eds) Proceedings
of the 2011 Winter Simulation Conference IEEE. pp161-175
Ulrich W (1988). Systems thinking systems practice and practical philosophy: a program of
research. Systems Practice 1(2):137-163.
36
Viana J, Brailsford, S C, Harindra V, and Harper P R (2014). Combining discrete-event

simulation and system dynamics in a healthcare setting: a composite model for Chlamydia
infection. European Journal of Operational Research, 237(1):196-206.
Viana, J, Rossiter S, Channon A A, Brailsford S C, and Lotery A (2012). A Multi-Paradigm,
Whole System View of Health and Social Care for Age-Related Macular Degeneration.
In: Laroque C, Himmelspach J, R Pasupathy, Rose O and Uhrmacher A M (eds).
Proceedings of the 2012 Winter Simulation Conference, Piscataway, New Jersey: IEEE.
pp1070-1081
Vidal R V V (2005). Soft OR approaches. Engevista, 7(1):4-20.
Zulkepli J. Eldabi T, and Mustafee N (2012). Hybrid Simulation for Modelling Large
Systems: An Example of Integrated Care Model. In Proceedings of the 2012 Winter
Simulation Conference, edited by Laroque C, Himmelspach J, R Pasupathy, Rose O and
Uhrmacher A M, 758-769.Piscataway, New Jersey: IEEE.
37

Infrastructure
Operational Research Methods for Policy Analysis in the Power

Sector
Afzal S. Siddiqui a,b
a
Department of Statistical Science, University College London, U.K.

Department of Computer and Systems Sciences, Stockholm University, Sweden
afzal.siddiqui@ucl.ac.uk
Abstract
In the past decade, the electric power sector has faced targets for improved sustainability, e.g.,
in the form of more renewable generation. Such requirements are often confounded by the fact
that the sector has been deregulated in most industrialised economies. Thus, power companies
need to comply with increasingly stringent emissions regulations while dealing with aspects of
imperfectly competitive markets. Likewise, policymakers need to take into consideration the
game-theoretic interactions within industry. From this perspective, operational research
methods employing complementarity modelling can provide a framework within which both
power companies and policymakers can assess strategic and regulatory decisions, respectively.
Keywords: Power sector; Deregulation; Sustainability; Game theory; Uncertainty
1.
Introduction
Concerns about climate change have spurred policymakers to implement measures to increase
energy sustainability. For example, the EU has 20-20-20 targets that aim to reduce energy
consumption by 20% relative to 1990 levels, reduce CO2 emissions by 20%, and have 20% of
all energy provided by renewable sources by the year 2020. These ambitious objectives are
often confounded by the deregulation of the power sector in most industrialised countries,
which sought to improve economic efficiency and technological innovation. Prior to
deregulation, vertically integrated utilities provided all functions of the power sector ranging
from generation to retail, thereby serving as monopolies with regulated profits. Meanwhile,
transmission investment decisions were made by a regulated transmission system operator
(TSO). Consequently, there was little incentive to innovate in order to bring new technologies
to the market to meet burgeoning demand (Wilson, 2002). Under this paradigm, capacity
expansion decisions could be made through traditional least-cost optimisation models (Hobbs,
1995).
By contrast, restructuring of the power sector has sought to foster competition by splitting up
the vertically integrated utilities through divestiture of generation assets and separation of
retail services. While the effectiveness of deregulation in terms of delivering benefits for
society is debatable (Hyman, 2010), it has, nevertheless, created an awkward challenge for
policymakers: they have committed to more stringent economic and environmental standards
while having relinquished control over the power sector. Thus, policymakers need to devise
measures carefully because they no longer have the means to intervene directly in the power
sector. Indeed, decisions about generation capacity expansion and transmission investment
that were once made centrally under the auspices of regulated entities are now made by
38
A. Siddiqui / OR56 Keynote Papers, pp.38-43 (2014)
distinct agents with conflicting objectives. As a result, policymakers need to anticipate how
such decisions will be influenced by policy measures. Consequently, models that account for
such game-theoretic interactions are desirable to provide both managerial and policy insights
(Gabriel et al., 2012).
2.
Research Highlight: Renewable Portfolio Standard
A market-based policy instrument to encourage the adoption of renewable energy technologies

is the renewable portfolio standard (RPS). It mandates that a given percentage of power
generated must come from renewable sources, which can be imposed as a constraint on the
power sector. When the market is cleared, the shadow price of this constraint is the renewable
energy certificate (REC) price that is given as a partial subsidy to renewable energy (RE)
plants and serves an effective tax levied on non-renewable energy (NRE) plants. In the U.K.,
this instrument is known as the renewables obligation, while Norway and Sweden run a
common market for so-called green certificates. Typically, the RPS percentage is currently
around 15% in most jurisdictions with increases, e.g., to 33% for the EU as a whole, mandated
by the year 2020.
In assessing the effectiveness of RPS, complementarity modelling is used to capture the
incentives of power companies in a stylised manner and to account for their game-theoretic
interactions. For example, Fischer (2010) assumes that all power companies are price takers in
exploring the impact of RPS on the equilibrium electricity price. She finds that modest RPS
targets may initially lower the equilibrium electricity price because the REC price serves as a
subsidy for RE. However, more stringent RPS targets cause NRE production to reduce
drastically, thereby putting upward pressure on the equilibrium electricity price and providing
consumers with a strong signal to reduce demand. By contrast, Amundsen and Bergman
(2012) allow for power companies to behave la Cournot in both electricity and REC markets.
Taking the Nordic region as an example, they find that the Swedish RE producers incentive to
exercise market power by withholding RECs will increase the retail price of electricity and
reduce electricity consumption. However, this outcome may be mitigated by opening up the
market to include Norwegian producers, which has since been done by the authorities.
Extending this framework, Tanaka and Chen (2013) develop a bi-level model in which a
dominant firm acts as a Stackelberg leader at the upper level constrained by a competitive
fringe at the lower level. The dominant firm is able to influence both electricity and REC
prices by manipulating its production at the upper level. Abstracting from some details of
electricity markets such as intermittency in renewable output and transmission constraints,
they solve a mathematical program with equilibrium constraints (MPEC) analytically. In
particular, they find that a dominant NRE producer can manipulate prices in both markets with
adverse consequences for RE investors. Effectively, it is in the NRE producers interests to
lower the REC price while increasing the electricity price. It is able to do precisely this by
withholding just the right amount of output, which can even lead the REC price to crash to
zero if the RPS requirement is particularly high. Hence, policymakers should be aware of such
incentives when setting RPS targets, especially when there is the risk that a dominant NRE
firm is able to lower the REC price to the extent that it discourages future RE investment.
39
In order to tackle the issue of just how RPS targets should be set, Siddiqui et al. (2014) also
develop a bi-level model but with the policymaker at the upper level and producers at the
lower level. The policymaker is assumed to be a welfare-maximising decision maker who
internalises a convex damage function from carbon emissions due to NRE output. By contrast,
the producers (either RE or NRE) are profit maximisers who do not care about environmental
externalities. Thus, the policymakers decision at the upper level is to set the optimal RPS
target that becomes an equilibrium constraint for the producers at the lower level. Three
settings are considered: a benchmark central planning (CP) framework, in which the entire
power sector is run by the state, perfect competition (PC), in which both types of producers are
price takers, and Cournot oligopoly (CO), in which the NRE producer behaves la Cournot.
Compared to the CP setting, the maximised social welfare is lower in both the PC and CO
settings. However, RE output under CP is surprisingly lower than that for either PC or CO.
This is because a centrally owned industry internalises the cost of emissions, i.e., each plant
produces up to the point where its marginal revenue equals the marginal social cost, and the
lack of REC revenues deters RE production. By contrast, under deregulated settings such as
PC and CO, each producer maximises profit by setting marginal revenue equal to marginal
private cost (plus/minus a REC tax/subsidy). Thus, although the policymaker has no control
over output under deregulation, it is able to set the RPS target in order to subvert the NRE
producer and to encourage RE output. Under PC, the optimal RPS target tends to be around
70% as opposed to around 20% under CP, which is the kind of requirement that is discussed in
policy circles. Consequently, although maximised welfare is higher under CP, greater
reductions in emissions are possible in a deregulated setting. Hence, acknowledging the
incentives of a deregulated industry to maximise profit leads policymakers to drastically
different RPS targets than those mooted in the single-level models.
Moreover, although the optimal RPS target is still higher under CO than under CP, it is lower
than that for PC. This is because the NRE producer exercises market power by withholding
capacity in order to raise the electricity price, thereby lowering energy demand as well as the
REC price. As a result, the policymaker optimally lowers the RPS target in order to reduce the
incentive of the NRE to exercise market power. For this reason, although maximised welfare
decreases with the damage cost in all three market settings, somewhat counterintuitively, it
remains higher under CO relative to PC as the higher profit obtained by the NRE (primarily
from lowering the REC price) more than offsets the lower consumer surplus and the higher
cost of damage from emissions.
While these aforementioned works investigate the strategic interactions among producers or
between producers and a policymaker in the presence of an RPS constraint, they do not take
into account either the topology of the transmission network or the intermittent output of RE
technologies such as wind. Furthermore, they do not address another type of game-theoretic
interaction that involves the entities constructing the transmission line and investing in new
wind farms. Indeed, the former may be done either by a regulated TSO or a private merchant
investor (MI) as proposed by Hogan (1992). The difference is that the TSO invests in new
transmission lines to maximise social welfare, whereas the MI is motivated by maximising its
own profit, which consists of congestion rents collected on transmission flows on the
constructed lines. These congestion rents depend on the differences in the shadow prices for
40
the energy-balance constraints at each pair of nodes in the network. Hence, the TSO and MI
have vastly different incentives for investing in transmission lines, which are needed to
reinforce the adoption of wind farms by power companies.
Taking the perspective of a welfare-maximising TSO, Baringo and Conejo (2012) formulate a
bi-level model in which transmission and wind investment are determined at the upper level
and the market is cleared at the lower level subject to network constraints. Uncertainty about
demand and wind power realisations is captured in the lower level via discrete scenarios. They
convert the resulting MPEC into a single-level problem by first replacing the lower-level
optimisation problems by their Karush-Kuhn-Tucker (KKT) conditions for optimality. Next,
they linearise the complementarity conditions in the resulting problem by using both strong
duality and disjunctive constraints. As a result, they end up with a mixed-integer linear
program (MILP) that may be applied to realistic network sizes and solved using commercially
available solvers. They conclude by showing how intimately transmission reinforcement
drives wind investment.
Using the general framework of Baringo and Conejo (2012), Maurovich-Horvat et al. (2013)
examine how transmission investment and wind adoption are affected by the choice of market
design, i.e., whether there is a TSO or MI at the upper level. They also generalise the
representation of the power sector so that investment in wind farms may be made either by
perfectly competitive or Cournot oligopolistic firms. Thus, the upper-level decision variable is
investment in transmission lines by either the TSO or MI, while lower-level decision variables
are generation investment and operation by the power companies along with resulting power
flows. Uncertainty in wind availability is captured by discrete scenarios, while RPS constraints
are also imposed on the producers. The model is implemented for a stylised three-node
network in which node 1 has the highest potential demand and legacy generation capacity by
an NRE plant, node 2 is a remote location with modest wind potential (but no installed
capacity yet), and node 3 has the highest wind potential (again without installed capacity
initially). Reformulating their MPEC as an MILP for different market designs and conjectural
variations at the lower level as well as a benchmark CP setting, the authors find first of all that
under the CP and TSO (with perfect competition) market designs, the results are identical.
Intuitively, transmission lines linking the most efficient resource with the most valuable
demand nodes are constructed. This results in generation expansion by the NRE producer and
the wind producer at node 2, power flows towards the location of the more expensive wind
producer at node 3, and expected profits are all driven to zero (except for the legacy generation
at node 1 owned by the NRE producer).
The MI market design under perfect competition is qualitatively similar in terms of
transmission investment, generation expansion, and power flows. However, since the MI is
concerned about maximising its own expected profits only, it strategically invests in less
transmission capacity to increase congestion rent revenue, thereby earning positive expected
profits. The expected RE generation is similar to that under the CP and TSO (with perfect
competition) market designs. Allowing for market power at the lower level leads to less
generation investment as producers seek to drive up the market-clearing price. Under the TSO
market design with a Cournot oligopoly, the NRE producers act of withholding investment,
41
indeed, increases its own expected profit and average prices across the network. However, this
withholding could have a deleterious effect on expected social welfare, which the TSO seeks
to remedy by effectively subsidising more transmission investment along lines involving
connections to node 2. Consequently, this countervailing action by the TSO creates an opening
for the wind producer at node 2 to expand generation capacity in order to offset some of the
effects of the market power exercised by the producer at node 1. Somewhat counterintuitively,
expected generation from renewables increases significantly under a Cournot oligopoly as
opposed to under perfect competition, and prevalent power flows are reversed, i.e., the wind
producer at node 2 actually exports power to the location of the NRE producer.
This rather surprising result is also encountered in the MI market design with a Cournot
oligopoly but for different reasons. As with the TSO, the lower level now has producers with
the incentives to withhold generation capacity in order to increase expected profits and prices.
However, such behaviour will conflict with the MIs objective to maximise expected profit,
which consists of the product of congestion rents and power flows. Since the latter are
adversely affected by the producers withholding, the MI takes countermeasures to prevent
power flows from dropping too much. Thus, it also invests in lines connecting the node of the
wind producer at node 2 with the rest of the network, albeit by much less than the TSO. This
enables power generation by not only the wind producer at node 2 but also the wind producer
at node 3. With an RPS constraint, the enhanced role for wind producers supported by
complementary transmission investment and evidenced by the reversal of power flows on
transmission lines is observed even when the lower level does not have producers behaving
la Cournot. Hence, the desirable policy target of significant RE penetration may be attained
(even in the presence of conflicting game-theoretic incentives) without the need to tolerate
market power.
References
Amundsen E and Bergman L (2012). Green Certicates and Market Power on the Nordic
Power Market. The Energy Journal 33: 101117.
Baringo L and Conejo AJ (2012). Transmission and Wind Power Investment. IEEE
Transactions on Power Systems 27(2): 885893.
Gabriel SA, Conejo AJ, Fuller JD, Hobbs BF, and Ruiz C (2012). Complementarity
Modeling in Energy Markets. Springer, Germany.
Fischer C (2010). Renewable Portfolio Standards: When Do They Lower Energy Prices? The
Energy Journal 31: 101-120.
Hobbs BF (1995). Optimization Methods for Electric Utility Resource Planning. European
Journal of Operational Research 83(1): 1-20.
Hogan W (1992). Contract Networks for Electric Power Transmission. Journal of Regulatory
Economics 4(3): 211-242.
42
Hyman LS (2010). Restructuring Electricity Policy and Financial Models. Energy Economics
32(4): 751-757.
Maurovich-Horvat L, Boomsma TK, Fleten SE, and Siddiqui AS (2013). Transmission and
Wind Investment in a Deregulated Electricity Industry. In 10th International Conference on
the European Energy Market (27-31 May 2013), Stockholm, Sweden.
Siddiqui AS, Chen Y, and Tanaka M (2014). Are Targets for Renewable Portfolio Standards
Too Low? A Complementarity-Based Policy Analysis. In 37th International Conference of
the International Association for Energy Economics (15-18 June 2014), New York, NY,
USA.
Tanaka M and Chen Y (2013). Market Power in Renewable Portfolio Standards. Energy
Economics 39: 187196.
Wilson RB (2002). Architecture of Power Markets. Econometrica 70(4): 1299-1340.
43

Inventory Research
On the Value of Sharing Demand Information in Supply Chains

Mohammad M. Ali a, John E. Boylan b
a
Coventry University, Coventry, UK

Buckinghamshire New University, High Wycombe, UK
mohammad.ali@coventry.ac.uk, john.boylan@bucks.ac.uk
b
Abstract
Many companies are embarking on strategies to share consumer sales data among supply chain
members. In most cases, this requires huge investments in information systems and training.
Benefits from sharing information have been discussed extensively in the supply chain
literature. However, a steady stream of research papers on Downstream Demand Inference
(DDI) suggest that the upstream member can mathematically infer the demand at the
downstream member. They claim that there is no value in sharing demand information.
Subsequent papers scrutinise the model assumptions in this stream of research and show the
conditions under which the consumer demand can and cannot be inferred mathematically by
the upstream member. Hence, the review of this literature clarifies when information sharing is
and is not valuable. In the DDI literature, the evaluation of the conditions under which
information sharing is valuable can help companies make more informed decisions on such
investments. Under other conditions, where information sharing is not feasible because of such
issues as trust and confidentiality, DDI can prove valuable.
Keywords: Downstream Demand Inference; Supply Chain Management; Demand Information
Sharing.
1.
Introduction
Information sharing has been extensively discussed in the supply chain literature in terms of
its benefits. This is considered by most as a central element of planning and management of
supply chains. Niranjan et al (2011) consider the quality of flow of information as a reflection
of the health of the supply chain. Similarly, Ramanathan (2014) argues that supply chain
designs have moved away from only focussing on the movement of material towards the
coordination of both material and information, which has resulted in reduced costs and
increased service levels.
The literature considers various types of information sharing e.g. demand, stock levels,
shipping quantity, lead time, and capacity. However, the literature on DDI is motivated by the
sharing of demand data or the POS (Point of Sales) data at the consumer level. Hence, in this
paper, information sharing refers to sharing of consumer demand / POS data.
Many companies are embarking on strategies to share consumer demand among supply chain
members. In most cases, this requires huge investments in information systems and training.
Benefits from sharing information have been extensively discussed in the supply chain
literature. However, a steady stream of research papers on Downstream Demand Inference
(DDI), suggest that the upstream member can mathematically infer the demand at the
44
M.M. Ali and J.E. Boylan / OR56 Keynote Papers, pp.44-56 (2014)
downstream member. Hence, they claim that there is no value in sharing demand information.
Subsequent papers scrutinise the model assumptions in this stream of research and show the
conditions where the consumer demand can and cannot be inferred mathematically by the
upstream member. Hence, the review of this literature clarifies when information sharing is
and is not valuable. The evaluation of the conditions, in the DDI literature, where information
sharing is valuable can help companies make more informed decisions on such investments.
On the other hand, DDI would be valuable where information cannot be shared in certain
supply chains due to issues such as trust and confidentiality.
The aim of this paper is to provide insights into when demand information sharing in supply
chains is and is not valuable. This is done by a brief review of the literature on Downstream
Demand Inference (DDI) in supply chains. We start our review, in section 2, by discussing the
Bullwhip Effect, as its existence is one of the motivations for research in DDI. In section 3, we
examine two streams of research in information sharing and specifically discuss their
contradictory conclusions. We then move on, in section 4, to review papers that investigate the
reasons for these contradictory conclusions. Finally, in the last section (section 5), we
summarise the discussions on the value of information sharing and present insights on the
application of this literature in real world supply chains.
2.
The Bullwhip Effect
When demand travels upstream in supply chains, its variation gets amplified and results in the
Bullwhip Effect. Empirical evidence and mathematical models (e.g. Caplin, 1985; Lee et al,
1997; Wong et al, 2007) show that the orders placed by a downstream link to its upstream link
tend to be more variable than the demand on the downstream link. This amplification of
variability propagates up the supply chain. The Bullwhip Effect has been identified as one of
the most important research areas in the field of Operational Research (Fildes et al, 2008).
There is certainly no lack of evidence of the Bullwhip Effect from real world supply chains.
This amplification in demand variability results in high inefficiencies, lack of customer
responsiveness, increase in operating cost and an increase in inventory cost (Ouyang et al,
2006; Lee et al, 1997). In terms of the magnitude of the problem, it has been suggested that
$30 billion per year can be saved by its elimination in the US grocery industry (Lee et al,
1997b; Cooke, 1999). Sterman (2006) remarked that the Bullwhip Effect was the most
significant factor in the inventory write-off of $2.25 billion of obsolete inventory by Cisco
Systems. The following table (Table 1) lists some papers providing empirical evidence of the
Bullwhip Effect.
45
Paper
Industry
Evidence Type
Blackburn (1991)
Various from Automotive to

Cameras
Examples
Kurt Salmon Ass. (1993)
Grocery
Case Study
Hameri (1996)
Paper Making
Case Study
Lee et al (1997a)
Home & Personal Care
Examples
Lee et al (1997b)
Soups, Printers
Case Study
Fine (1998)
Machine Tools
Examples
Fransoo and Wouters (2000)
Perishable Food
Case Study
Greek (2000)
Semi Conductor
Examples
Dejonkheere et al (2003)
FMCG
Examples
Terwiesch et al (2005)
Computer and Semi-conductor
Case Study
Sterman (2006)
Oil Industry
Examples
Disney (2007)
Retail Supermarket
Case Study
Wong et al (2007)
Toys
Case Study
Bray and Mendelson (2012)
4,689 U.S public sector companies
Examples and
analysis
Bottani et al (2010)
Italian FMCG
Case Study
Thompson and Cheng

(2012)
Contract Manufacturing
(Flextronics, Solectron)
Case Study
Akkermans and Voss (2013)
Service Industries (Broadband

services, Glass fibre network)
Case Study
Zotteri (2013)
Personal Care sector (consumer

goods)
Case Study
Shan et al (2014)
1200 Chinese Companies
Examples and
analysis
Lee et al (2014)
Green Bullwhip Effect

Environmental effects
Case Study
Table 1: Empirical Evidence of Bullwhip Effect

A review of the literature shows that the Bullwhip Effect can be reduced if the upstream
member (e.g. a manufacturer) is aware of (and hence utilises) the demand at the downstream
member (e.g. a retailer). This reduction comes into effect due to the use of less variable
46
downstream demand in forecasting and planning by the upstream member. Empirical evidence
in various papers (e.g. Dong et al, 2014; Lu et al, 2013; Bottani et al, 2010; Hosoda et al,
2008; Kulp et al, 2004; Croson and Donohue, 2003; Fransoo and Wouters, 2000) shows the
reduction in the bullwhip effect due to sharing of demand information. In addition, various
papers provide detailed mathematical proof of the reduction in the variability of demand by
shared information. We discuss some of these information sharing models in the next section.
3.
Demand Information Sharing Models
Lee et al (2000) was one of the first papers to quantify the value of information sharing when
the downstream demand is shared using formal information sharing mechanisms in the supply
chain. They show that inventory holdings and cost at the upstream member will be reduced if
the downstream member shares the demand data with the upstream member (via a formal
information sharing mechanism). Raghunathan (2003) and Cheng and Wu (2005) extend the
above supply chain model to a multi retailer-single manufacturer case, while Wu and Cheng
(2003) extend this to a multi-echelon supply chain. Yu et al (2002) further extend the model of
Lee et al (2000) by introducing a unit ordering cost and test this on a Vendor Management
Inventory (VMI) scenario. The above papers show that the main conclusion of Lee et al
(2000), namely lower inventory cost by sharing information, can also be applied to a multi
retailer-single manufacturer supply chain, a three echelon supply chain and also to a supply
chain when taking unit cost and discount factor into account. All the papers show that their
conclusions regarding the value of demand information sharing are also applicable to more
complex supply chain models. The extensions of the Lee et al (2000) model are summarised in
the following figure:
Lee et al (2000)
Multiple
Retailers
Correlated demand
among retailers
Raghunathan
(2003)
Multiple
Echelon
VMI
Independent demand
among retailers
Cheng and
Wu (2005)
Yu et al
(2002)
Ordering Cost
Wu and Cheng
(2003)
Figure 1: Extension of Lee et al (2000)
47
On the other hand, authors in another stream of research (Graves, 1999; Raghunathan, 2001;
Zhang, 2004; Gilbert, 2005) argue that the orders from the downstream member to the
upstream member already contain information about the consumer demand process. By using
their order history, the upstream member can infer the demand at the downstream member.
Thus, these authors claim that there is no value in sharing the demand data and, consequently,
no information sharing is required. We refer to this approach as Downstream Demand
Inference (DDI). According to the DDI approach, the savings in inventory costs from demand
information sharing could be obtained without any formal information sharing with the
downstream member. If this were true, then investments in information systems to allow
demand information sharing would be unnecessary. It is important to note that the supply
chain models in this stream of research papers assume that the process and parameters of the
consumer demand are known across the supply chain.
All the above papers discussed in the previous section assume an invertible demand process.
Gaur et al (2005) and Giloni et al (2014) extend the analysis in this stream of research by
assuming a non invertible demand process. They assume that the demand at the retailer is
modelled as non-invertible ARMA process and show that information sharing will have value
for such demand process. However, it is important to note that for any non-invertible
representation of an ARMA (p, q) process, there exists an invertible representation of the
process and vice versa. Thus, if the order process to a manufacturer is a non-invertible ARMA
(p, q), they may instead use the invertible representation of the demand. Hamilton (1994)
comments that, in order to calculate the noise terms associated with any time series, the
current and past values of demands are required if an invertible representation is used. On the
other hand, in order to calculate the noise terms for a non-invertible representation, the future
values of demands are required. Thus, it is not feasible to use the non-invertible representation
and the manufacturer will instead identify its process as invertible.
We present the following figure (Figure 2) to summarise the links between papers discussing
the DDI approach.
48
Stationary
Non-Stationary
Upstream Translation of
AR(1) demand
Raghunathan (2001)
ARIMA (0, 1, 1) demand
Graves (1999)
AR(1) & ARMA (1, 1)
demand
Alwan et al (2003)
Hosoda & Disney (2006)
ARMA-in-ARMA-out
Property
Zhang (2004)
ARIMA-in-ARIMA-out
Property
Gilbert (2005)
Non-invertibility issues
Gaur et al (2005) Giloni et
al (2014)
QUARMA in
QUARMA-out
Giloni et al (2014)
Figure 2: Downstream Demand Inference in Supply Chains
4.
When is Information Sharing Valuable?
Ali and Boylan (2011) investigate the above papers in the light of the contradictory results
between the two streams of research. They specifically consider the supply chain model
assumptions in view of their relevance to the real world. They argue that the supply chain
models presented in the above papers have very restrictive assumptions. These papers assume
that the manufacturer is aware of the demand process and the demand parameters of the
retailer even when they are unaware of the demand itself. The claim of no value in sharing
demand information in these papers is sensitive to this assumption.
In a real world scenario, supply chain links need to have a formal information sharing
mechanism if they decide to share demand information between downstream and upstream
members. On the other hand, there is no need to invest in a formal information sharing
mechanism if the strategy of the supply chain link is not to share demand information. Thus, it
is highly unlikely that the supply chain links will invest in a formal information sharing
mechanism just to share the information on demand process and parameters and not on the
actual value of demand itself.
In section 2, we mentioned various papers, dating back to 1999, that empirically show how
information sharing can help reduce the bullwhip effect. On the contrary, there is as yet no
empirical evidence to support the successful employment of the DDI strategy. If the
49
mathematical inference of downstream demand were practically implementable, why has it not
yet been adopted? This remains an open question. It may be that the idea is a practical one but
organisations have been very slow to adopt it. On the other hand, it may be that the restrictive
assumptions underpinning inference of downstream demand make it impractical.
In light of the above, Ali and Boylan (2011) consider the possibility of DDI under more
practical assumptions i.e. unknown demand process and parameters. They show that in order
to infer the downstream demand, the upstream member needs to first infer the demand process
and then deduce the required parameters. They present the following two feasibility principles
to show that it is not possible for the upstream member to do both: infer the demand process
and deduce the parameters.
Feasibility Principle I: If the upstream member can infer the demand process at the
downstream link, the demand values at the downstream link cannot be exactly calculated.
Feasibility Principle II: If the upstream member cannot infer the demand process at the
downstream link, then it may be possible for demand values at the downstream link to be
calculated if a certain demand process is assumed from a restricted subset of the possible
processes.
Hence, the feasibility principles show that for an optimal forecasting method and Order-Up-To
(OUT) inventory policy, under more practical settings, information in supply chains has to be
shared via some formal information sharing mechanism rather than being inferred
mathematically.
Ali and Boylan (2012) further considered the application of DDI in real world supply chains
by challenging the use of optimal forecasting method in the literature. They argue that, in the
real world, a forecasting method is not always chosen on the basis of its optimality or accuracy
but rather its simplicity, ease of use and familiarity. Hence, they relax the assumption of
optimal forecasting method in the papers and extend the literature by considering two nonoptimal forecasting methods: Single Exponential Smoothing (SES) and Simple Moving
Average (SMA). In terms of SES, the paper concludes that similar to the optimal method, DDI
is not possible under that method. Hence, a formal information system is required to facilitate
demand information sharing for SES. On the other hand, DDI is found to be possible in the
case of SMA. Thus, if the retailer in a supply chain uses SMA, the next upstream link can
mathematically infer the retailers consumer demand without the need to formally share this
information.
Finally, Cui et al (2013) argue that the assumption of a strict and consistent OUT policy in the
DDI literature is rarely followed in practice. They maintain that the supply chain members will
implement adjustments according to various other factors (e.g. batch production, full truck
load). Hence, they introduce an error term in their empirical model in order to account for the
decision deviations in the order from the downstream member to the upstream members. Their
results show that consumer demand information is required to be shared (rather than inferred)
under these more realistic inventory assumptions.
50
5.
Summary and Insights on Application
Supply chain Management has moved away from the traditional focus on the movement of
material towards planning and coordination of both material and information. Businesses are
continually adopting new approaches to obtain and utilise the end consumer demand
information in their planning. There has been an interesting debate in the Operations Research
literature about the approaches companies can adopt in order to acquire the consumer demand.
One stream of research (Chen et al, 2000; Lee et al, 2000; Yu et al, 2002; Raghunathan, 2003;
Cheng and Wu, 2005) mathematically models the benefits of sharing the consumer demand.
The results of the papers in this stream imply that upstream members will need formal
information sharing mechanisms to share the demand at the downstream link. On the other
hand, another stream of research (Graves, 1999; Raghunathan, 2001; Zhang, 2004; Gaur et al,
2005; Gilbert, 2005) argues that companies at the higher end of the supply chain can
mathematically infer the consumer demand. If this is possible, information sharing
mechanisms will not be required.
Subsequent papers in the literature clarify conditions when formal systems are required to
share consumer demand and when this can be mathematically inferred without the need of
such systems. Firstly, feasibility principles in Ali and Boylan (2011) conclude that information
sharing is always valuable. They show that DDI is not possible when optimal forecasting
methods and Order Up-to-Level (OUT) inventory policy is used. Ali and Boylan (2012)
further show that information sharing is valuable when the supply chain links use the Single
Exponential Smoothing (SES) method. This is because, similar to optimal methods, DDI is not
possible in the case of SES method. However, they also show that DDI is possible in the case
of Simple Moving Averages (SMA) method. Hence, in this specific case, information sharing
will not be valuable. Finally, Cui et al (2013) relaxes the inventory policy assumption by
allowing decision deviations. The empirical results in their paper show that information
sharing is always valuable. We present the summary of the DDI literature with the help of the
following figure.
51
Review of DDI
Literature
There is value in
sharing information
There is NO value in
sharing information
Analysis of noninvertible demand
Analysis of model
assumptions in the
light of application in
the real world
Feasibility Principles
Non-Optimal
Forecasting Methods
Inventory Policy
Application of SMA to
counter lack of
information sharing
Future Directions
Figure 3: Review of DDI Literature in the light of ascertaining value of information sharing
It is important to evaluate the DDI literature with regards to its application to real world
situations. Given the huge investment cost of information systems and relevant training,
companies need to be clearer on expected savings from such initiatives. With regards to the
benefits of information sharing, findings from the literature are clearly aligned with the
evidence from the real world. Various case studies report huge benefits in terms of cost
savings and higher customer level by sharing information. Collaboration programmes between
Wal-Mart and Sara Lee, Schering-Plough Health Care with all their retail partners, and Marks
and Spencer with Gunstones are some examples of these.
As discussed above, one stream of literature in sharp contrast shows no value of sharing
information. However, various successive papers challenge the assumptions in this stream of
research in the light of their application in the real world. Specifically, the assumptions
challenged are known model variables, forecasting methods and inventory policy. The more
practical assumptions in these successive papers have attempted to bring the literature closer
to real life supply chains. Moving forward, a critical need within this area of literature is to
measure the value of information sharing in an empirical setting.
It is also important not to discount the DDI literature in terms of its possible application to
supply chains that do not and cannot share consumer demand information. Evidence from
recent reports shows the magnitude of this problem e.g. Seifert (2010) reports two such
52
surveys on the assessment of the level of data sharing. The first, by CapGemini, of 16 retailers
in Europe and North America found that only 40% of the retailers shared shopper data with all
of their suppliers. The second survey by Forrester Research of 89 retailers showed that 27% of
retailers were sharing such data. In addition, the results from interviews carried out with 15
companies by Allred et al (2011) show that high level collaborations are rare and efforts to
improve information sharing are seldom embraced holistically.
Downstream Demand Inference (DDI) using SMA would be useful for such supply chains
where information sharing does not take place, particularly for products not exhibiting strong
trend or seasonal patterns. This is quite an interesting finding specifically keeping in view that
there are many supply chains where information may not be shared due to constraints such as
compatibility of information systems, information quality, trust and confidentiality. Due to this
unique feature of SMA, as discussed above, Ali et al (2014) propose that it should be
considered for situations where formal information exchange is not possible even though there
may be a willingness to share information from both parties.
In terms of future directions, there are various possible extensions. The first area, as discussed
above, is to validate the applicability of the supply chain models in this area by more empirical
research. Another possible extension is to extend the demand model by assuming seasonality
and / or trend. The issue of non-invertibility may be further explored to ascertain its
practicality. Finally, only two non-optimal forecasting methods (SMA and SES) have yet been
considered and the effect of DDI based on other non-optimal methods may also be explored in
this literature.
References
Akkermans H and Voss C (2013). The service bullwhip effect. International Journal of
Operations & Production Management 33(6):765 788.
Ali M and Boylan J (2011). Feasibility principles for downstream demand inference in supply
chains. Journal of the Operational Research Society 62:474-482.
Ali M and Boylan J (2012). On the effect of non-optimal forecasting methods on supply chain
downstream demand. IMA Journal of Management Mathematics 23:81-98.
Ali M, Babai Z, Boylan J and Syntetos A (2014). A forecasting strategy for supply chains
where information is not shared. Working Paper Series, Coventry University, 2014.
Allred CR, Fawcett SE, Wallin C and Magnan GM (2011). A dynamic collaboration capability
as a source of competitive advantage. Decision Sciences 42(1): 129-161.
Alwan L, Liu J and Yao D-Q (2003). Stochastic characterization of upstream demand
processes in a supply chain. IIE Transactions 35 (3):207-219.
Blackburn J (1991). Time-Based Competition: The Next Battleground in American

Manufacturing, McGraw Hill Professional Publishing, New York.
53
Bottani E, Montanari R and Volpi A. (2010). The impact of RFID and EPC network on the
bullwhip effect in the Italian FMCG supply chain. International Journal of Production
Economics 124(2): 426-432
Bray R and Mendelson H (2012). Information transmission and the bullwhip effect: An
empirical investigation 58(5):765-875.
Caplin A (1985). The Variability of Aggregate Demand with (s, S) Inventory Policies,
Econometrica 53: 1395 1409.
Chen F, Ryan JK and Simchi-Levi D (2000). The Impact of Exponential Smoothing Forecasts
on the Bullwhip Effect. Naval Research Logistics 47: 269286.
Cheng TCE and Wu YN (2005). The Impact of Information Sharing in a TwoLevel Supply
Chain With Multiple Retailers. Journal of the Operational Research Society 56: 1159 1165.
Cooke JA (1999). CPFR: The Countdown Begins, Logistics Management and Distribution
Report 38: 59 62.
Croson R and Donohue K (2003). Impact of POS data sharing on supply chain management:
An experimental study. Production and Operations Management 12(1): 1-11.
Cui R, Allon G, Bassamboo A and Miegham J (2013). Information Sharing in Supply Chains:
An Empirical and Theoretical Valuation. Kellogg School of Management, 2013.
Dejonchkheere J, Disney SM, Lambrecht MR and Towill DR (2003). Measuring and
Avoiding the Bullwhip Effect: A Control Theoretic Approach. European Journal of
Operational Research, 147:567 590.
Disney SM (2007). The Dynamics of Material Flow in Supply Chains. In: OR Forty Nine
Keynote Papers, Shaw, D. [Ed], Operational Research Society.
Dong Y, Dresner M and Yao Y (2014). Beyond Information Sharing: An Empirical Analysis
of Vendor-Managed Inventory. Production and Operations Management 23(5): 817-828.
Fine CH (1998). Clockspeed Winning Industry Control in the Age of Temporary Advantage,
Perseus Books Group, New York.
Fransoo JC and Wouters MJF (2000). Measuring the Bullwhip Effect in the Supply Chain,
Supply Chain Management: An International Journal 5:78 89.
Hosoda T, Naim MM, Disney SM and Potter A (2008). Is There a Benefit to Sharing Market
Sales Information? Computers and Industrial Engineering, 54: 315 326.
Gaur V, Giloni A and Seshadri S (2005). Information Sharing in a Supply Chain under ARMA
Demand, Management Science 51:961 969.
Gilbert K (2005). An ARIMA Supply Chain Model, Management Science 51:305-310.
54
Giloni A, Hurvich C and Seshadri S (2014). Forecasting and information sharing in supply
chains under ARMA demand. IIE Transactions 46(1):35-54.
Graves SC (1999). A Single-Item Inventory Model for a Nonstationary Demand Process,
Manufacturing and Service Operations Management 1:50-61.
Greek D (2000). Whiphand, Professional Engineering 13:43 44.
Hameri AP (1996). A Method to Avoid Demand Amplification in Bulk Paper Distribution,
Paper and Timber Journal 78:102 - 106.
Hamilton JD (1994). Time Series Analysis, Princeton University Press, New Jersey, USA.
Hosoda T and Disney SM (2006). On Variance Amplification in a Three-Echelon Supply
Chain with Minimum Mean Squared Error Forecasting. Omega 34:344 358.
Hosoda T, Naim MM, Disney SM and Potter A (2008). Is There a Benefit to Sharing Market
Sales Information? Computers and Industrial Engineering. 54:315 326.
Kulp SC, Lee HL and Ofek E (2004). Manufacturer Benefits from Information Integration
with Retail Customers. Management Science 50:431 444.
Kurt Salmon Associates (1993) Efficient Consumer Response: Enhancing Consumer Value in
the Grocery Industry, Food Marketing Institute, Wa4shington D.C.
Lee HL, Padmanabhan V and Whang S (1997a). Information Distortion in Supply Chains: The
Bullwhip Effect. Management Science 43:546 559.
Lee HL, Padmanabhan V and Whang S (1997b). The Bullwhip Effect in Supply Chains, Sloan
Management Review 38:93 102.
Lee HL, So KC and Tang CS (2000). The Value Of Information Sharing in a Two-Level
Supply Chain. Management Science 46:626-643.
Lee S-Y, Klassen R, Furlan A and Vinelli A (2014). The green bullwhip effect: Transferring
environmental requirements along a supply chain. International Journal of Production
Economics 156:39-51.
Lu C, Tsai K-M, Chen J-H and Lee W-T (2013). Mitigating the bullwhip effect in the supply
chain of semiconductor assembly and testing through inter-business information platform.
Electronic Business Management 11(3):202.
Niranjan T, Wagner S and Aggarwal V (2011). Measuring information distortion in real-world
supply chains. International Journal of Production Research 49 (11):3343-3362.
Raghunathan S (2001). Information Sharing in a Supply Chain: A Note on its Value When
Demand is Non-Stationary. Management Science 47:605-610.
55
Raghunathan S (2003). Impact Of Demand Correlation on the Value of and Incentives for
Information Sharing in a Supply Chain. European Journal of Operational Research
146:634-649.
Ramanathan U (2014). Performance of supply chain collaboration A simulation study,
Expert Systems with Applications. 41(1):210-220.
Seifert D (2003). Collaborative Planning, Forecasting and Replenishment How to Create a
Supply Chain Advantage. AMACOM, New York.
Shan J, Yang S, Yang S and Zhang J (2014). An Empirical Study of the Bullwhip Effect in
China. Production and Operations Management 23(4):537-551.
Sterman JD (2006). Operational and Behavioural Causes of Supply Chain Instability. In:
Torres OA and Moran FA (2006). The Bullwhip Effects in Supply Chains, Palgrave
Macmillan pp 17 56.
Terwiesch C, Ren JZ, Ho TH and Cohen MA (2005). Forecast Sharing in the Semiconductor
Equipment Supply Chain. Management Science 51:208 220.
Thompson B and Cheng L-C (2012). Bullwhip effect in contract manufacturing industry: case
studies of Solectron and Flextronics. Journal of Academy of Business and Economics
12(3):117-125.
Wong CY, El-Beheiry MM, Johansen J and Hvolby H.-H (2007). The Implications of
Information Sharing on Bullwhip Effects in a Toy Supply Chain, International Journal of
Risk Assessment and Management 7:4 18.
Wu YN and Cheng TCE (2003). The Impact Of Information Sharing in a Multiple-Echelon
Supply Chain, Working Paper, Faculty of Business, Hong Kong Polytechnic University,
Hong Kong.
Yu Z, Yan H and Cheng TCE (2002). Modelling the Benefits of Information Sharing-Based
Partnerships in a Two-Level Supply Chain, Journal of the Operational Research Society
53:436-446.
Zhang X (2004). Evolution of ARMA Demand in Supply Chains, Manufacturing and Service
Operations Management 6:195 198.
Zotteri G (2013). The impact of RFID and EPC network on the bullwhip effect in the Italian
FMCG supply chain. International Journal of Production Economics 143(2):489-498.
56

OR Consultancy & Case Studies
The Evolution of OR in the Defence Industry in the UK: A Personal

Perspective
Noel Corrigan
CORDA, BAE Systems, Farnborough, Hants, UK
noel.corrigan@baesystems.com
Abstract
This paper presents a personal perspective on the evolution of Operational Analysis (OA, the
application of OR to military problems) in the UK defence industry over the last thirty years,
and where it may go in the next thirty. This period has seen a shift from product assessments to
service solutions, from conducting independent analysis of how new equipment could improve
military effectiveness in a cold war environment, through to the current drive to deliver cost
efficiencies via availability contracts and capability management. The changes in application
area have been accompanied by changes in the approaches used, embracing developments in
systems thinking and interdisciplinarity. Drawing on experiences and case studies from
CORDAs history as an independent consultancy, and now as part of the UKs largest global
defence company, it reflects on how both the customer and the work have changed
significantly since the 1980s to respond to the changing environment.
Keywords: Retrospective; Defence; Operational analysis; CORDA; History
1.
Introduction
In this paper we will look at how the practice of Operational Analysis (OA, the application of
OR to military problems) has changed (if at all) over the last thirty years. This arbitrary
timeframe has been chosen solely as it happens to be the 30th anniversary of the founding of
CORDA, the company where I have spent much of my OR career. We shall see how the
customers problem types have changed as the world has turned, and how the tools and
techniques have evolved. Finally we will consider how those problems and approaches may
further change in the next thirty years - noting of course that OR is not the business of
predicting the future.
2.
Context
OR as applied in the Defence domain was initially concerned with the analysis of military
operations, with increasing the efficiency and effectiveness of military affairs at the tactical
and strategic levels. After the Second World War, the use of OR expanded into industry,
including the defence industry (Kirby and Capey 1998). With the advent of computers and
their wider uptake in the 1970s, modelling and analysis services began to be provided to the
UK Ministry of Defence (MoD), largely through software houses.
Over the years, the analysis community within the MoD has been reorganised a number of
times, with the current core of analytic capability being found within the Defence Science and
Technology Laboratory (Dstl). The defence industry in the UK has also undergone
57
N. Corrigan / OR56 Keynote Papers, pp.57-67 (2014)
rationalisation. Many of the smaller analysis groups and software houses have been acquired
by larger companies, but there has been a constant regeneration as new companies are
established and grow.
2.1. The world of the 1980s and 1990s

1984, the year of CORDAs founding, did not turn out to be the Orwellian nightmare, though
some involved in the miners strike of that year may beg to differ3. Europe had only just
started the transition from the Industrial Age to the Information Age, a transition from which
defence is not immune, and for a time there was a belief that technology alone could dominate
the battlefield.
The major factor in the world of the eighties, of Reagan and Thatcher, was the strategic
environment - we were still fighting the cold war. The analytic emphasis was on detailed
modelling of the response to the Soviet threat; the setting was well understood, as was how to
get there and assumptions about how long it would last.
The commercial world of Thatchers Britain saw the rise of software houses (EASAMS,
Scicon, Logica, SD, CAP4), that were of a more consultancy bent then the traditional
engineering suppliers; it was within these that many OR services were delivered.
2.2. The world today

In the current environment none of the military or commercial certainties of the 80s and 90s
remain. There is much more emphasis on stabilisation operations and nation building, rather
than high intensity war-fighting. This has an impact on the type of OA that is required, and the
timescales for its delivery.
Information technology and the growth of computing power has made the models faster and
given amateurs the false impression that it is easy to knock together a quick spread sheet
model to conduct a thorough analysis. The speed of communication of ideas and the general
pace of life and business has increased to the extent where it seems no-one has the time to
contemplate a decision.
The industry today is significantly reshaped, and modelling and analysis are embedded
throughout many organisations and business functions, so finding a distinct OR group is
something of a rarity, a trend which started back in the 90s (Fildes et al, 1999).
3.
Changes to the defence and security environment
In the last thirty years, the nature of the military operations has changed significantly. As it
turned out (though few saw it coming) the 1980s proved terminal for the Soviet Union and the
3
The UK miners' strike of 198485 was a major industrial action affecting the British coal industry.
It lasted almost a year and its outcome significantly weakened the British trade union movement.
Arguably its low point was the Battle of Orgreave, widely reported to have involved 5,000 police
officers pitched against 5,000 miners.
4
Note that none of these company names have survived.
58
cold war. The collapse of the Berlin Wall was followed by rapid change, with the events of
11th September 2001 bringing us to the present, far more complex, operational environment.
More recently still, the global economic crisis has left severely constrained defence budgets
throughout the NATO nations (Forder et al 2013).
There will undoubtedly be similar tipping points in the years to come. It is not the role of OR
to predict these tipping points, but rather to ensure that the decision makers of the nation and
its industry have the evidence available to allow them to ensure a robust military and industrial
capability into the future.
Over the last thirty years the size and shape of the British Armed Forces has changed
considerably, most notably in size, both in numbers and budgets. The military have been on
almost constant active service throughout that period, with operations in Iraq, Bosnia, Kosovo,
Sierra Leone, Afghanistan, alongside the standing commitments in Northern Ireland, the
Falkland Islands and Cyprus. Despite the fall of the Berlin wall twenty five years ago the UK
has kept troops stationed in Germany. Only now are they being withdrawn to the mainland
(which, by the way, provides plenty of opportunities for analysis).
The emphasis on operations other than war (a misnomer) has led the Army in particular to
revisit lessons learned in the post-WWII colonial clashes in Kenya, Malaya and Aden (for
example); historical analysis has re-emerged from the shadows. Emphasis on human factors
and influence has come to the fore, as well as the requirement to work alongside not only
allies but also non-governmental organisations, aid agencies, charities and the like. All of
these pose challenges to the analysis community.
Most recently the rise of cyber terrorism as a threat is posing yet another new set of messes,
puzzles and problems for the OA community to address (Pidd 2009).
3.1. Analysis in the MoD

The 1990s saw a transition in the delivery of OA. As the army downsized, there were more
military professionals available to offer their services back to the MoD. This was also coupled
to a reduction in the size of the Civil Service, including the UK MoD, meaning there were
trained analysts available, with work to be done. This lead to the establishment of numerous
independent OA companies, built upon knowledge of both defence operations and the inner
workings of both the UK MoD and the wider Civil Service.
59
By 1994 there were 11 OA groups in the Ministry of Defence research establishments as well
as the Defence Operational Analysis Establishment (DOAE). These groups included:
Royal Armament Research & Development Establishment;

Royal Aircraft Establishment;
Military Vehicles Engineering Establishment;
Admiralty Research Establishment.
There were also OA cells embedded in the commands: in the 1st British Corps in Germany, in
HQ Fleet at Northwood and Strike command in High Wycombe, all of which were supported
by analysts supplied from industry.
All of the analysis functions in these establishments were consolidated into the Defence
Evaluation and Research Agency (1995) and then split into QinetiQ (a privatised company)
and Dstl early in the 21st century.
4.
Changes in OA within the UK defence industry
The development of OR within UK industry has followed 2 paths over the last 30 years.
Firstly, the in-house OR teams working within the major equipment suppliers, to evaluate
technical solutions and inform system investment decisions. Examples are British Aerospace
and Royal Ordnance (now both part of BAE Systems) and Westland (now AgustaWestland,
part of Finmeccanica). Secondly, the specialist consultancies providing niche OR skills to
support decision makers within the MoD, which have developed to complement the skills
available in the research establishments, DOAC, and latterly in Dstl (Moffat et al 2008).
Both these groups saw significant growth at the end of the 20th century, as the need to justify
investment decisions with firm evidence became paramount both within industry and the
MoD. Another factor in their growth was the increasing OR challenge provided by the
expansion in the number and type of operations to which UK forces were committed.
The specialist consultancies involvement in the delivery of OA began in the 1970s with the
use of software houses to develop the computer simulations that were becoming the models of
choice for OA studies. By the start of the 1980s this had become well established. During the
1980s and 1990s the MoD continued to use industry to develop these models, in particular to
help transition to the new Object Oriented programming paradigm. At the same time, in the
major establishments and the MoD Centre, there was growth in support contracts for the
upkeep, update, and upgrade of the model suites, and to assist in the management of the
computer facilities that kept these models up and running.
60
From my perspective, the early to mid-1990s saw a highpoint for Industry OA in the MoD.
Three examples of how embedded this OA support had become:
SD-Scicon led the conduct of the Support Helicopter Combined Operational

Effectiveness and Investment Appraisal (COEIA), a major study supporting the
decision whether or not to proceed with a multi-million-pound equipment programme,
CORDA had a tasking contract with the Assistant Chief Scientific Advisor
(Capability) and did quick support studies at the strategic level, and
A CORDA person filled an Assistant Director post in DSc(Sea) organisation

(Mountain 2008).
Today, industry provides decision support services and OR tools to each of the four
commands (Air, Army, Navy, and Joint), and is helping defence suppliers (of both equipment
and services) to partner more effectively with the MoD. Currently within the UK there is a
trend towards greater risk sharing between the MoD and its prime suppliers and OR is playing
a central role in helping decision-makers across the defence enterprise to understand these
risks and their implications.
5.
Changes to the world of OR practice
5.1. Methods
In 1979 Russell Ackoff published two influential papers (Ackoff 1979a, 1979b) in which he
presented his view that there was an increasing inappropriateness of OR's methodology and a
need for a comprehensive reconsideration of the field, its methodology, the way it is practised,
and the idea that planning is the natural home of OR problem solving. He urged a return to the
original concept of interdisciplinary teams and greater involvement in research (Kirby 2003).
At the same time Checkland (1981) was developing the soft system methodology. These and
others underpinned a move in defence analysis to balance the mathematical-mechanical
treatment of engagements in computer-based simulations and optimisations with a more
holistic problem-centred approach to tackling complex situations, where many different highly
interrelated perspectives, values and beliefs define the situation (Rosenhead 2001).
It is not clear to me whether this was causal or coincidental with the shift in industry towards
the analysis of more strategic issues, and to a more consultant-led approach to working with
the MoD. There was a move away from just doing what the customer requested to helping
them understand what underlying problems needed to be addressed. This helped the customer
structure their problem, not just answer the set question.
61
5.2. Tools
Until the 1980s and the birth of the PC most military OA was focusing on:
Observation and field studies lots of military exercises

Historical analysis
War games (often using probabilities based on historical analysis). 6
weeks to war game 2.5 days of fighting.
Manual analysis analogue / qualitative
Thirty years ago the tools of the trade were a pen, paper and a mainframe. In CORDAs early
days we had one HP9000 series machine, and a VAX mini-computer (so-called because it was
the size of a mini). From 1980s, there was the PC explosion. Mathematical simulations could
take the performance stats from the historical and scientific analysis / research and work out
the outcomes of war games. Visual Interfaces replaced Command Line Coding. By 1999 the
most important tools in the kit bag were spreadsheets (primarily MS Excel), simulation
programmes (which have evolved into programmes like Simul8 and Witness). System
dynamics tools and neural nets were making an impact too.
In OR modelling terms, the transition is from models of the cold war with attrition at their
core, to models of current and future complex operations with information and command
decision-making at their core. The importance of information is also seen as going beyond
providing the basis on which better and faster decisions can be made. The analyst must also be
able to assess the effectiveness of information operations (designed to influence an
adversarys decision-making in a favourable direction), which now play an increasingly
important role in achieving the effect traditionally delivered by force.
5.3. Some things have not changed

Just as there are still Routemaster buses running in London, Boeing 747s carrying us off on
long haul flights, and the Rolling Stones still performing, some things have not changed in the
last 30 years. Despite our pleading of the need for more multi-disciplinarity we still actually
recruit mainly from the fields of Maths, Physics and of course OR. We still largely
communicate the results of our studies in written reports that are designed to be printed and
read. We are still committed to providing the best evidence we can within the constraints of
the study, though these days we call it service excellence rather than total quality or service
with a smile. And yes, I still work at CORDA.
6.
Driving the demands for decision support
The types of decision that OR is required to support in the defence arena have changed over
the last thirty years. Industry now uses OA to understand risks on performance of contracts
and to provide evidence to MoD that their solution is sound. Industry uses OA to support the
sale of equipment to MoD demonstrating performance of a system, including the human
interaction / interface. Now, MoD wants to move the financial risk to industry (part of the
62
Public Finance Initiative5) through the use of mechanisms such as availability contracts.
Therefore industry needs to conduct the OA to help them understand the risks from how the
equipment will be used and demanded. At the same time, the MoD needs OA to help them
assess the bids. But this is a different type of OA to the old days.
In the cold war the MoD challenge was to have the best technology possible to beat a welldefined threat. The Warsaw Pact did not put a price on its military capability they had
strength in numbers and generally in technology too (though this did eventually prove to be
their undoing).
Now thinking has shifted, with the focus on priorities such as:
Spending shrinking budgets wisely;

Security against cyber threats;
Sustainability and the militarys carbon footprint;
Non-equipment aspects such as personnel, training, and infrastructure;
Operational effectiveness and risk; and
Maintaining capability as cost-effectively as possible.
OR now needs to consider a very wide spectrum of possible operational contexts. At one end
of the spectrum are major hostilities, although on a lesser scale than planning the NATOWarsaw Pact cataclysm. Towards the other end of the spectrum we have a range of operations
including stabilisation operations, peace enforcement, protected evacuations and the provision
of humanitarian relief. Increasingly, military means are used only in close conjunction with
other means, such as diplomacy and economic reconstruction. Following the events of 11th
September 2001, counter-terrorism is now also given prominence.
Although the origins of OA lie in direct support to the planning and execution of frontline
military operations, this aspect of the work, from the 1960s onwards, gradually became
overshadowed by the higher-profile activities concerned with future force structure and
procurement planning. However, as well as making the defence planning environment more
complex, the end of the cold war has resulted in a higher level of involvement in actual
operations by the UK. This has put a great deal of emphasis back on OR support to the front
line. The effect has been to raise the profile of frontline OA very considerably.
7.
Case studies through the ages
For me, the major project in the 1980s was Project Foresight. It went on for several years
(through phases 1, 2 and 2A) and delivered literally box-loads of reports. It was a future
battlefield study, looking 10 years out at equipment issues; there was little concern for aspects
such as training, personnel, and logistics in those days. It was a study that attempted to join up
analysis from the low level one-on-one models up to high level campaign games. It involved
nearly all of the MoD analysis groups and a large number of suppliers of OR services. It
A way of creating "publicprivate partnerships" by funding public infrastructure projects with

private capital.
63
involved half of the CORDA staff at the time. Roles included modelling and analysis, data
management (MDAL etc.) and co-ordination across the various parties involved.
It is not clear whether it actually altered any procurement decisions! (Hawkins 2006).
Personally, I worked as the data manager - a huge task. I was based on site at Main Building in
Whitehall for several months. This is something that has not changed over the years; we like
to work closely with our customers if they have the room to accommodate us (and the tyranny
of hot-desking allows).
In the early 1980s, we typically did 'low-level' and 'hard' OR/OA, such as detailed quantitative
studies into individual systems' performance. For example, we did a long running Shallow
Water Anti-Submarine Warfare Study, and long running Pre-Concept and Concept Phase set
of studies into Low Frequency Active Sonar. We traded on analytic modelling rather than
simulation something we still do. Now, the OA we do tends to be much higher-level beyond just system performance, and into woollier things like capability and capability
management.
In the late 80s/early 90s the MoD introduced the Combined Operational Effectiveness and
Investment Appraisal (COEIA) process; this put the cost of a procurement programme front
and centre of the decision making. This has since developed further to take into account the
through life cost of buying and running the equipment (Forder 2008). At the time it was
introduced, the emphasis was on the need to support all investment submissions with OA and
this became a focus for CORDA activities.
Projects seemed to be longer duration then, and less rushed/urgent. Back then, a typical project
might last 9 months to a year or more; today it's likely to be a quarter of that. Turnaround
times for answers were more relaxed too - an answer in 6 months was fine. Today, much faster
response is expected, and there's more variety in the questions being asked. Contracts were
more likely to be time and materials rather than fixed price. Indeed, the pace of life seemed to
offer more time to interact with the customer and mull over problems, I have fond memories
of sessions on the pitch and putt course at the clients site and customer update meetings
followed by a leisurely pub lunch!
Typically we used much more analytical OR/OA then than now. Because we didn't have ready
access to PCs, we used to get our paper and pencils out and write equations, use OR
techniques such as queueing theory and probability theory, and do our calculations (largely)
by hand and calculator. We thought about our problems more back then. In contrast, today we
do (I think) tend to jump straight into a computer model/simulation/numerical analysis
approach to solve a problem, because it's so easy to do. But by doing this I think we
sometimes miss opportunities to produce clearer, simpler, more transparent models/solutions. I
think we must all resist the temptation to rush into churning out impenetrable code that
sometimes obfuscates the problem rather than illuminating it. On the other hand, customers
are more willing to seek a relatively simple logical structuring of a problem, rather than a
proven mathematically-optimised solution.
64
8.
Future Challenges
As we have seen, over the last 30 years OA (at least as practiced by CORDA) has adopted and
adapted new techniques to remain able to deal with the questions its customers need to
address. It is likely that in the coming 30 years there will continue to be increasing
consideration of a much broader spectrum of operations, especially those where the military is
subordinate to other Government levers of power, such as economic or diplomatic
considerations. This will be coupled with more emphasis on the ability to coerce, deter or
deceive an opponent rather than physically defeating him. The modelling of military
operations in the post-cold-war strategic environment presents many intellectual and practical
challenges, but lies at the heart of most applications of OR in the defence domain.
From an industry perspective, the concept of the Total Support Force will gain priority; this
concept aims to achieve an enhanced capability through better exploitation of industrial
expertise and capacity much nearer the front line than has been traditional in the UK military.
All of this will be supported by an increasing understanding of human factors associated with
decision making - what information is used by decision makers, and how they interpret and
exploit it.
The cost of delivering military capability will continue to be a core consideration, but not just
the cost in pure financial terms. The acceptability of casualties will drive strategic decision
making towards greater use of unmanned assets, and the decreasing availability of fossil fuels
will increase the focus on energy efficient systems.
There is an increasing move to considering the peacetime management of defence items
such as personnel planning, training policy, and vehicle fleet management. All are topics that
we have been exploring, but need further development and exploitation with the appropriate
level of decision maker to realise their full benefits.
Big data is very fashionable. The analysis of real operational data could hold the answers to
some puzzles, and provide the key to understanding the drivers of perception and behaviour,
whilst allowing the validation of models of irregular warfare and stabilisation operations
(Appleget and Cameron 2012).
To meet these challenges, we will require method enhancements, increasing the
interdisciplinarity of our study teams to include a greater range of softer skills. This will be
coupled with increasing use of combinations of methods, for example using softer
facilitation and problem structuring methods in concert with hard, numerical methods such
as simulation or mathematical optimisation (Robinson, Pickburn, Forder, 2004).
65
9.
Summary
So what can you take away from this? What does an OR practice require to survive the
rollercoaster of changes in defence policy, the capriciousness of the markets and Events, dear
boy, events (Knowles 2009)? I believe there are a number of strands to be drawn from this
rather rambling personal tour of the last thirty years.
1. You need a sound Weltanschauung an understanding of the context, the world in
which the problem exists and the solution needs to operate;
2. You need to manage the organisational politics;
3. You need good people who can sell themselves, and are able to communicate their
findings at different levels of the customer organisation;
4. Studies dont need to be perfect to be valuable, but they do need to be readable;
5. Best results come from personal relationships and close integration with the customer;
6. Tomorrow is another day.
References
Ackoff RL (1979a). The future of OR is past. Journal of the Operational Research Society 30:
93-103.
Ackoff RL (1979b). Resurrecting the future of OR. Journal of the Operational Research
Society 30: 189-199
Appleget J and Cameron F (2012). Best Practices for Irregular Warfare (IW) Data Quality
Control; proceedings of the 29th International Symposium on Military Operational
Research; London UK
Checkland PB (1981). Systems Thinking, Systems Practice. John Wiley & Sons Ltd.
Fildes R, Ranyard JC and Crymble WR (1999). The management of OR groups: results of a
survey; Journal of the Operational Research Society 50: 563-580
Forder R; (2008) Military OR after the cold war; proceedings of the 25 th International
Symposium on Military Operational Research; Southampton UK
Forder R, Visco G, Military OR and ISMOR (2013). A 30th symposium retrospect;
proceedings of the 30th International Symposium on Military Operational Research;
London UK
Kirby MW and Capey R (1998). The origins and diffusion of operational research in the UK;
Journal of the Operational Research Society 49: 307-326
66
Kirby MW (2003). The intellectual journey of Russell Ackoff: from OR apostle to OR

apostate; Journal of the Operational Research Society 54:11271140
Knowles E (Ed.) (2009) Oxford dictionary of quotations (7 ed.). Oxford University Press;
ISBN-13:9780199237173
Hawkins G (2006). A waste of 37 years; proceedings of the 23rd International Symposium on
Military Operational Research. Southampton, UK
Moffat J, Robinson A, Forder R and Corrigan N (2008). A Review of Defence Operational
Research in the UK, Farnborough, Hampshire, UK. Communications of the OR Society
of Japan, 2008.
Mountain J (2008). From Product Assessment to service solutions; proceedings of the 25 th
International Symposium on Military Operational Research, Southampton UK
Pidd M (2009). Tools for thinking : Modelling in management science 2nd edition. Wiley
2003, ISBN 0-470-84795-6.
Robinson, Pickburn and Forder (2004). Complementarity in Ministry of Defence OR Practice.
In System Modelling: Theory and Practice, M.Pidd (Ed). Wiley, London
Rosenhead J and Mingers J (2001). Rational Analysis for a Problematic World; John Wiley,
England.
67

OR in Sport
Rankings Models in Sport

Ian G. McHale
University of Manchester, School of Mathematics, UK
ian.mchale@manchester.ac.uk
Abstract
The paper presents models used for ranking, and describes them in the setting of sport. Paired
comparisons models and multiple comparisons models are described before a methodology for
estimating time-varying rankings is presented. I use the model to estimate the greatest tennis
player in the Open Era.
Keywords: Rating; Ranking; Bradley-Terry; Plackett-Luce; Dynamic; Sport; Tennis.
1.
Introduction
Perhaps the main objective of professional sport is to rank the competitors. From ranking
football teams over a season to deciding on the gold, silver and bronze medal winners at the
Olympic Games, fans, athletes and stakeholders want to know who is the best?. Of course,
sport is not the only arena in which ranking is of interest. In todays world we rank a lot:
parents are captivated by university and school league tables; Google uses the pagerank
algorithm to order the results of internet searches; different voting mechanisms are proposed
for ranking political parties; marketing analysts study consumer preferences for competing
brands, the list goes on and on.
As a consequence of the abundance of data and general widespread interest in sport, much of
the literature on ranking has been motivated and demonstrated using sport as the setting. In
this paper I will continue that trend. Using various sports as the setting, I will describe some of
the popular models for ranking before presenting some extensions to allow for time-varying
rankings. Lastly, I will give an example of time-varying ranking based on a model for tennis
described in Baker and McHale (2014a).
2.
Rankings Models
Broadly speaking, there are two types of competition for which rankings models exist:
1. paired comparisons: head-to-head matches with lots of players in the population, e.g.
in tennis each match is a head-to-head between two players drawn from a large
population of players;
2. multiple comparisons: many competitors in each match, e.g. golf tournaments have
lots of players all competing simultaneously against one another. Each tournament has
a different set of players drawn from a large population of (professional) golfers.
Other examples are races (motor racing, track athletics and so on).
These two fundamental types of competition require different models for obtaining rankings.
68
I.G. McHale / OR56 Keynote Papers, pp.68-77 (2014)
2.1 Models for Paired Comparisons

Paired comparisons models are formulated so that each competitor being compared is
associated with a strength parameter, a . The probability of one competitor beating another is
then given by a function of the ratio of the strength parameters of the two competitors in
question. More generally, the probability, pij , that competitor i beats competitor j , is given
by pij = F(mi - m j ) , where mi = ln(a i ) , and F is a distribution function. The Bradley-Terry
model (Bradley and Terry, 1952) assumes a logistic distribution for F , and the ThurstoneMosteller model (Thurstone, 1927) uses a normal distribution. In terms of the strength
parameters, the formulae are respectively
pij =
ai
,
(a i + a j )
and
pij = F(ln(
ai
)) ,
aj
where F is the standard normal distribution function.

Stern (1990) gives a formula for the probability of a ranking based on a gamma distribution
for the underlying times to an event. These times can be thought of as the time for each
competitor to run a race in which the quickest wins. One can also consider the random
variable to be a score, where the player with the lowest score wins, as in golf. For Sterns
model,
pij = fi (x)S j (x)dx = Fi (x) f j (x)dx,
(1)
where fi (x) = a i (a i x)b -1 exp(-a i x) / G(b ) (the pdf of the gamma distribution), S j the survival
function and Fi the distribution function. When b = 1 the model reduces to the Bradley-Terry
model. Stern (1990) shows how his gamma comparison model interpolates between the
Bradley-Terry model and the Thurstone-Mosteller model as b and refers to the model as
a continuum of paired comparison models. This is clearly a useful model, but Stern (1992)
concludes that for samples of the size usually encountered, predictions are not sensitive to the
value of b . However, given we are in the era of big data, it is more and more commonplace
to have enough data to determine which value of b fits the data best.
Baker and McHale (2014a) derived a closed form for Sterns continuum of paired
comparisons models in terms of the beta distribution. The probability that player i beats
player j is shown to be equal to
pij =
G(bi + b j ) a i /(a i +a j ) bi -1
b -1
y (1- y) j dy,
G(bi )G(b j ) 0
69
(2)
i.e. the probability that player i beats player j is the distribution function of the beta
distribution. For Sterns model, bi = b j b so that pij is the distribution function of the
symmetric beta distribution. This form of Sterns model is excellent for computation, as the
beta distribution is a special function that is widely available from software platforms such as
fortran, MatLab or R. Analysts fitting paired comparisons models could adopt this generalised
model in (2), rather than choosing between the Bradley-Terry or Thurstone-Mosteller model.
2.1.1 Dealing with ties in paired comparisons models
The paired comparisons models above deal with competitions in which there are two
outcomes (win or lose). This is fine for sports like tennis for which there are just two
outcomes, but there are many situations in which a tie is possible too. For example, the
outcome of a football match can be a win, a loss, or a tie. There are several options available
for those wishing to use a paired comparisons model allowing for ties. Rao and Kupper (1967)
add a parameter to the Bradley-Terry model to allow for ties. Davidson (1970) also adds a
single parameter, n , to the model so that
pi|ij =
ai
a i + a j + n a ia j
p j|ij =
aj
p0|ij =
(3)
n a ia j
where p0|ij is the probability of a tie and pi|ij is the probability that player i wins a match
between players i and j.
2.2 Models for Multiple Comparisons

The Plackett-Luce model (Plackett, 1975) is used to rank competitors when more than two
players compete in a series of contests. It is in fact a generalisation of the Bradley-Terry
model. Under the Plackett-Luce model, when n competitors are to be ranked, the probability
that the i th player wins is
pi =
ai
(4)
j=1
where the j th player has strength a j . After a player has been ranked first, another player
ranks second by winning out of the remaining players, and so on down to the last, who wins
with probability one. For a particular tournament then, the Plackett-Luce model gives the
probability (likelihood) of an ordering of all entrants given their strengths. The probability of
the n competitors finishing in an ordering, w , in a single tournament is given by
70
Pr(w) =
k=1
ak
n
ai
i=k
where a k is the strength of the competitor finishing in k th position.

The essential feature of the Plackett-Luce model is that the probability of a player coming n th
is the same as the probability of that player winning in a contest in which the top n -1 players
did not compete. The simplicity of the Plackett-Luce model is appealing and it has been used
in a variety of studies and settings. Critchlow (1985) for example, uses it to model data where
consumers are asked to rank their three favourite brands in order of preference, from a list of
many brands.
Dealing with ties in the Plackett-Luce model is more complicated than in the Bradley-Terry
model, but it is possible. In the interests of brevity, I do not cover it here.
3.
Time-Varying Rankings Models
The paired comparisons models and the Plackett-Luce multiple comparisons model described
above assume constant competitor strengths over time. Of course, in many applications, time
invariant strengths are not appropriate. For example, when modelling in sport, it is overly
simplistic to assume that team (or player) strengths are constant. Similarly when modelling
consumer preferences for brands, changing fashions dictate brand strengths and they are
unlikely to be constant over time. In response, more recent work on rankings models has
concentrated on extending models to allow for time-varying, or dynamic, competitor strength.
There are two ways in which to incorporate time invariance into these rankings models: one
can assume that competitor strengths evolve either stochastically or deterministically.
Stochastic models assume that the change in competitor strength from time t to time t+1 is a
random variable. For example, Glickman (1999) presents an approximate Bayesian updating
algorithm for estimating dynamic strength parameters of the Bradley-Terry model. The
parameters evolve stochastically and he uses the model to construct a list of the best chess
players of all time, together with a rankings table of the then current tennis players. KnorrHeld (2000) presents a methodology, based on a Bradley-Terry type model, to estimate
stochastically evolving team strengths in football.
An alternative to assuming stochastic evolution of competitor strength is to assume that
strengths evolve deterministically. For individual sports such as tennis, where players compete
as individuals, there is clearly a strong systematic component to the evolution of their strength.
Typically, a players strength increases in the early part of their career, reaches a peak and
then slowly declines until the player withdraws from competing. Undoubtedly a stochastic
element is present, for example when a player is injured. However, injured players normally
stop competing until largely recovered. In either case, the contribution of the deterministic
component to strength evolution outweighs that of the stochastic element. Baker and McHale
(2014a, 2014b and 2014c) present models for estimating rankings models for tennis, football
71
and golf that allow for deterministic evolution of competitor strengths. In the next section we
discuss time-varying rankings in tennis, first presented in Baker and McHale (2014a).
4.
A time-varying paired comparisons model: who is the greatest tennis player?
4.1 Game, set and match: adapting the Bradley-Terry model for tennis
In addition to using paired comparisons models to estimate the strength of players to win
matches, one can also use the models to estimate strength of winning sets, games, or even
points (if you have the data on which player won each point). Of course, one should
acknowledge as much information as possible in modelling. For example, suppose player i
beat player j 7-6, 6-7, 7-6, 6-7, 7-6 and player k beat player j 6-0, 6-1, 6-2, what would
your prediction be for a match between players i and k ? Clearly both players beat player j,
but player k comprehensively beat him. To me, it seems reasonable to believe that player k
would be favourite in a match between i and k.
In terms of the Bradley-Terry model, one way to include this is to treat each game, or set, as a
mini-match within the match. Let us treat each set as a mini-match (it turns out that using a set
as the unit of victory, rather than a game, is better for forecasting and so it is used here). If
player i beats player j 2-0 in sets, then the probability of this would be
pij (2 - 0) =
ai
ai
a i2
=
a i + a j a i + a j (a i + a j )2
This thinking leads us to the general formula for the contribution to the likelihood function for
a match between players i and j in which player i won si sets and player j won s j sets of
L(a i , a j )
a is a j
s +s
(a i + a j )
i
sj
(5)
4.2 A deterministic model for time-varying strengths

To allow for time-varying strengths, we let a i a i (t) in (5), then estimate the strength at each
of several nodes for every player and then interpolate between the nodes to obtain values of a
players strength at any point during his career. We interpolate using barycentric rational
interpolation (BRI) (Floater and Hormann, 2007, and Press et al., 2007).
The time-varying strength is modelled using the barycentric rational interpolant (BRI), so that
the strength of player i at time t is given by
ni
a i (t) =
ik
lik / (t - tik )
k=1
ni
ik
k=1
72
(6)
/ (t - tik )
where lik is the kth fitted strength of player i, i.e. the strength at time t ik . There are ni such
nodes for player i.
There are various methods available for calculating the weights, wik . Using weights wk = (-1)k
when ni 5 , and wk = (-1)k (
1
1
+
), when ni > 5 maximises the likelihood of the
t k - t k-1 t k+1 - t k
model fit to data.

Choosing the node positions and the number of nodes is the same problem faced when fitting
spline curves, where knots are usually equally spaced, or sometimes placed at quantiles of the
distribution of the variable. Here, nodes are spaced equally over a players career, but omitted
a node if it had no support, i.e. no matches played within a year either side of it. The number
of nodes for player i is given by
ni = 1+ Di fi + (Di / M )(1- fi )
where D i +1 is the number of playing years for player i, M = 4 , and

fi =
1- exp(kgi / D i )
,
1+ exp(kgi / D i )
where gi is the total number of games player i played. The choice of a logistic-type function
means that fi goes from 0 for players who never play to 1 for players who play a lot and so
that ni progresses smoothly from 1 every 4 years for players who played few matches, to
annual nodes for those who played frequently.
formula is empirical but seems to work well.
5.
was chosen to minimise the AIC. This
Data
The model described above can be used to obtain rankings in many situations. Here we use the
model to investigate which mens tennis player has been the greatest during the Open Era
(until 1968 players wanting to make a living from tennis had to turn professional and were no
longer allowed to enter the Grand Slam tournaments). To measure greatness we calculate two
metrics here: (i) the maximum one-year strength, and (ii) the maximum lifetime strength of a
player.
We obtained data from the Grand Slam Tennis Archive, http://www.tennis.ukf.net/, on the
results and score lines of all four of mens Grand Slam events since the Open Era began in
1968 and Wimbledon, the US Open, the Australian Open and the French Open all agreed to
allow professional tennis players to compete against amateurs, to the US Open in 2012. Until
1968 the game was somewhat fragmented since top players wanting to earn a living from the
game were not allowed to enter the Grand Slam tournaments. This totalled 20640 matches,
1163 players, and a total of 74722 sets.
73
Table 2: Number of Grand Slam titles for mens tennis since 1968 to June 2014
Ran
k
1
2
2
4
5
5
5
8
8
10
10
10
13
13
Player
Total
Roger Federer
Pete Sampras
Rafael Nadal
Bjrn Borg
Jimmy Connors
Ivan Lendl
Andre Agassi
John McEnroe
Mats Wilander
Stefan Edberg
Boris Becker
Novak Djokovic
Rod Laver
John Newcombe
17
14
14
11
8
8
8
7
7
6
6
6
5
5
Australian
Open
4
2
1
0
1
2
4
0
3
2
2
4
1
2
French
Open
1
0
9
6
0
3
1
0
3
0
0
0
1
0
Wimbledon
US Open
7
7
2
5
2
0
1
3
0
2
3
1
2
2
5
5
2
0
5
3
2
4
1
2
1
1
1
1
Table 1 shows the all-time list of male players with the most Grand Slam titles as of June
2014. Roger Federer is clearly a strong candidate for the title of the best ever tennis player,
with Pete Sampras and Rafael Nadal in joint second place. It is worth noting that as a
consequence of the Grand Slams being held on different surfaces, and of players specialising
in certain surfaces, several of the greats never won a particular tournament. For example, Pete
Sampras never won the French Open (a clay court tournament). Despite these complications
regarding surface, few tennis fans would argue with the rankings in table 1.
However, using the number of Grand Slams as the determining factor in identifying the
greatest player of all time is problematic for several reasons. For example, some players will
have played in an era when competition was extremely tough, preventing that player from
winning more titles. Conversely, competitors may have been of lower quality than in other
eras, so that the strongest player of the era would have won more titles than his ability
suggested he should do. Or perhaps a great player had his career cut short by an unfortunate
injury, yet for a short period, while he was fit, he was playing the best tennis ever. Correcting
for these potential sources of bias is not trivial.
Using the model to obtain rankings has several advantages over table 1. First, the biases
discussed above are automatically corrected for. Second, by using information on margin of
victory, the model-based rankings should be able to produce a more accurate table of tennis
greatest players. Further, the model can be used to identify exactly when, in a players career,
they were at their peak. Lastly, the model can be used to calculate the inferred probability of
one player beating another, even though the two players never played against each other.
74
6.
Results
The first four columns of table 2 gives the top 10 all-time greatest tennis players according to
maximum strength and the final columns show the rankings according to maximum lifetime
strength.
Table 3: All time greatest tennis player: columns 2 to 4 give the ranking according to the
maximum strength achieved by a player during his career with the value of that natural
logarithm of strength and the year it was achieved. The final two columns give rankings
according to total lifetime strength. Standard errors on estimated log-strengths are shown in
parentheses.
Ranking
Player
Year
1
2
3
4
5
6
7
8
9
10
R. Federer
B. Borg
J. Connors
R. Laver
R. Nadal
K. Rosewall
A. Roche
J. Mcenroe
S. Edberg
J. Courier
2006
1977
1976
1968
2007
1970
1968
1983
1990
1992
Maximum
Strength
2.04 (0.17)
1.93 (0.20)
1.87 (0.16)
1.84 (0.21)
1.72 (0.16)
1.69 (0.18)
1.64 (0.21)
1.63 (0.13)
1.62 (0.14)
1.58 (0.13)
Player
R. Federer
B. Borg
J. Connors
R. Nadal
R. Laver
I. Lendl
J. Mcenroe
K. Rosewall
S. Edberg
B. Becker
Lifetime
Strength
1.43 (0.15)
1.35 (0.16)
1.31 (0.13)
1.15 (0.14)
1.12 (0.19)
1.10 (0.92)
1.07 (0.11)
1.06 (0.16)
1.03 (0.12)
1.00 (0.14)
It is pleasing, in terms of the model, that the players listed in table 2 read as a whos who of
tennis since 1968. Further, there is considerable overlap with the players featuring in table 1. It
appears then that Roger Federer is the greatest player of the Open era according to both
measures of greatness. Bjorn Borg and Jimmy Connors are close behind. It is worth noting
that there is a lack of a statistical significant difference in the strengths achieved by the top
players. However, this does not invalidate the rankings, rather we can say that based on the
data observed, the model estimates that Roger Federer attained the highest strength of any
player.
Figure 1 shows the estimated strength trajectories for the top five players in the Open Era.
Players careers tend to follow similar trajectories in the Open Era: a rise to a maximum
strength then a slight decline before retirement. For the entire data set, the mean time from a
player's first Grand Slam appearance to his peak strength is 6.7 years (with a standard
deviation of 3.3 years).
75
100
75
Strength
Name
B. Borg
50
J. Connors
R. Federer
R. Laver
R. Nadal
25
1970
1980
1990
2000
2010
Year
Figure 1: Estimated strength trajectories for five of the greatest tennis players of the Open Era:
Roger Federer, Bjorn Borg, Jimmy Connors, Rod Laver and Rafael Nadal.
7.
Conclusions
Models for ranking have become a key component of modern life. Indeed we are using a
rankings model every time we perform a search on Google. In sport, recent work has
concentrated on time-varying rankings models, which acknowledge the dynamic nature of
competitor strengths. There are two options for allowing strengths to vary through time: (i) use
a stochastic model, or (ii) use a deterministic model. Here I have described a method first
presented in Baker and McHale (2014a) for the latter and use the model to investigate which
tennis player might be regarded as the greatest during the Open Era. Roger Federer looks like
the most likely candidate, with Bjorn Borg and Jimmy Connors close behind. Though we
should note that Rafael Nadal is closing in on Federers record of Grand Slam titles.
Outside the world of tennis, rankings models such as those presented here can be used in a
variety of circumstances. Perhaps the most novel exercise I have heard of is to use the
Bradley-Terry model to assign marks for student projects. Each student is given two
assignments (not their own) and they are asked to choose a winner. The experiment is repeated
and the Bradley-Terry model is used to estimate strengths of the projects.
76
References
Baker RD and McHale IG (2014a). A dynamic paired comparisons model: Who is the greatest
tennis player?. European Journal of Operational Research, 236(2): 677-684.
Baker RD and McHale IG (2014b). Time varying ratings in association football: the all-time
greatest team is. Journal of the Royal Statistical Society: Series A,
DOI: 10.1111/rssa.12060.
Baker RD and McHale IG (2014c). Deterministic Evolution of Strength in Multiple
Comparisons Models: Who is the Greatest Golfer? Scandinavian Journal of Statistics,
DOI: 10.1111/sjos.12101.
Bradley RA and Terry M E (1952). Rank analysis of incomplete block designs I: the method
of paired comparisons. Biometrika 39: 324-345.
Critchlow DE (1985). Metric Methods for Analysing Partially Ranked Data, Springer, New
York.
Davidson RR (1970). On extending the Bradley-Terry model to accommodate ties in paired
comparison experiments. Journal of the American Statistical Association 65:317-328
Floater M S and Hormann K (2007). Barycentric rational interpolation with no poles and high
rates of approximation. Numerische Mathematik 2: 315-331.
Glickman ME (1999). Estimation in large dynamic paired comparison experiments. Journal of
the Royal Statistical Society, Series C (Applied Statistics) 48:377-394.
Knorr-Held, L. (2000). Dynamic rating of sports teams. Journal of the Royal Statistical
Society, Series D. 49: 261-276.
Plackett R L (1975). The analysis of permutations. Applied Statistics 24:193-202.
Press WH, Teukolsky S A, Vetterling W T and Flannery BP (2007). Numerical Recipes, 3rd
ed., Cambridge University Press, Cambridge.
Rao, P. V. and Kupper, L. L. (1967). Ties in paired-comparison experiments: A generalization
of the Bradley-Terry model. Journal of the American Statistical Association, 62, 194204.
Stern, H. (1990). A continuum of paired comparisons models. Biometrika, 77, 265-273.
Stern, H. (1992). Are all linear paired comparison models empirically equivalent?
Mathematical Social Sciences, 23, 103117.
Thurstone, L. L. (1927). The method of paired comparisons for social values. Journal of
Abnormal and Social Psychology, 21, 384400.
77

Problem Structuring
Zooming in on Problem Structuring Interventions: An Affordance

Perspective
L.A. Franco
School of Business and Economics, Loughborough University. Loughborough LE11 3TU, UK
L.A.Franco@lboro.ac.uk
Abstract
Empirical studies examining the complex dynamics between the social and material aspects
of OR interventions are beginning to appear in the literature. Despite these advancements,
we know very little about these dynamics at the micro-level of analysis. This is partly
because of a lack of theories that can inform an empirically grounded understanding of the
interplay between the social and the material as they become interwoven in micro-level
practices. To address this gap, I adopt an affordance perspective to examine how the
materiality of an OR technology can shape, but not fully determine, social actors
behaviours during interaction with that technology. I illustrate the potential usefulness of
the affordance perspective for the case of problem structuring interventions that use Group
Explorer as a group causal mapping technology. I then show, via an empirical case
vignette, how perceptions of affordances called forth by the technology affect social actors
behaviours within a strategy workshop. I conclude with a discussion of the implications of
adopting an affordance perspective for the conduct of OR intervention research and
practice.
Keywords: Affordances; Behavioural OR; Interaction; Materiality; Intervention; Problem
structuring.
1. Introduction
Although it has long been recognised that OR interventions involve complex dynamics
between the social and the technical (e.g. Eden, 1989; Keys, 1989, 1998; Ormerod, 1996),
empirical studies examining these dynamics are just beginning to appear in the literature. For
example, White (2009) draws on actor-network theory (Callon, 1986a, b; Latour, 1987, 2005)
and narrative analysis (Abell, 1993; White & Taket, 2000) to examine how problem
structuring interventions are socially constructed through a bundle of sociotechnical
interactions. More recently, Ormerod (2013) uses the concept of mangle of practice
(Pickering, 1992, 1995) to reassess an intervention undertaken for the UK National Coal
Board during the late 1970s and early 1980s (Ormerod & McLeod, 1984; Plackett, Ormerod,
& Toft, 1982). He shows how the complex intertwining of material and social factors affected
the interventions design, deployment and outcomes. These studies pay close attention to
macro-level interactions between the social agency of intervention actors and the material
agency of the OR technology deployed within an intervention. Despite these advancements,
however, the study of micro-level interactions between the social and the technical elements of
OR interventions remains relatively underexplored in the literature (Franco & Montibeller,
2010; Franco & Rouwette, 2011), with some exceptions (e.g. Ackermann & Eden, 2011b;
Franco, 2013; Horlick-Jones & Rosenhead, 2007).
78
L.A. Franco / OR56 Keynote Papers, pp.78-90 (2014)
The relative neglect of micro-level studies is in part due to a lack of theories that can inform
an empirically grounded understanding of the interplay between the social and the technical as
these aspects become entangled (Orlikowski, 2007) in interaction. In this paper, I address
this gap by drawing upon the theory of affordances (Gibson, 1986; Hutchby, 2001; Norman,
1999) to examine how the materiality of an OR technology can shape, but not fully determine,
social actors behaviours during interactions with that technology. Affordances are aspects of
the materiality of an object that offer action possibilities for those who come into contact with
that object. In the context of OR technology, the principal objects comprise the models,
concepts and physical infrastructures within which these models and concepts are developed
and used; affordances are what these models, concepts and infrastructures enable or constrain
social actors to do within an intervention. I contend that an examination of how these
affordances are realised in practice is likely to produce a more nuanced understanding of why
and how OR technology affects social actors behaviours, and with what effects. This
improved understanding can in turn help design and deploy more effective OR interventions.
The rest of the paper is structured as follows. I begin by introducing the theory of affordances,
highlighting its potential usefulness for the study of OR interventions. I then illustrate this
conceptualisation for the case of problem structuring interventions that use Group Explorer, a
computer-supported group causal mapping technology. Next, I show via an empirical case
vignette how perceptions of affordances called forth by the Group Explorer technology affect
social actors behaviours within a strategy workshop. I conclude with a discussion of the
implications of adopting an affordance perspective for the conduct of OR intervention research
and practice.
2. The Concept of Affordance

The term affordance was first coined by ecological psychologist James Gibson (1986)
following his research on visual perception. His theory of affordances aims to explain how an
individual perceives the behavioural possibilities of an object, noting that the same object can
call for different possibilities for action. Thus, for example, a rock can be used as a weapon or
as a paper weight. Similarly, an apple tree can be used as a shelter or as a source of food.
Gibson claims that what we perceive is not what an object is but rather what kinds of uses it
affords. Obviously we would still be able to discriminate an objects material features (e.g.
colour, shape) if prompted to do so. However, the empirical evidence suggests that what we
normally pay attention to is what the object affords us (Grezes & Decety, 2002; Symes, Ellis,
& Tucker, 2007; Tucker & Ellis, 2004). Beyond the individual, the theory of affordances is
also taken to be relevant to social behaviour (Gaver, 1996; Hutchby, 2001). In this context,
because an object can be perceived to afford multiple uses, it is possible that the same object
can produce multiple behavioural outcomes.
Gibsons formulation has been further elaborated by numerous studies in psychology,
sociology, and information management. Recent developments of the notion of affordance
emphasize its relational character (e.g. Chemero, 2003; Faraj & Azad, 2012; Hutchby, 2001;
Leonardi, 2011). In this view, affordances are constituted in relationships between social
actors and the materiality of the objects with which they interact. The term materiality here
refers to the features of an object, including physical or digital features (Leonardi, 2012;
79
Leonardi, Nardi, & Kallinikos, 2012). In this formulation, materiality exists independent of
social actors, but affordances do not. Because social actors come to the materiality of an object
with diverse goals, they perceive that object as affording distinct possibilities for action.
Furthermore, affordances of an object can change across different contexts of use even though
its materiality does not. Similarly, people may perceive that an object offers no affordances for
action, perceiving instead that it constrains their ability to carry out their goals.
I argue that to gain a deeper understanding of the complex nature of OR interventions at the
micro-level, then it is important to focus research attention on how OR technologys
materiality and its perceived affordances affect social actors behaviour. Take the models
created and used within the so called facilitated modelling interventions (Franco &
Montibeller, 2010; Robinson, Worthington, Burgess, & Radnor, 2013; Rouwette, 2011) as an
example. In this type of intervention a model is developed with a group of participants
working in a workshop environment, and assisted by an OR facilitator. The materiality of the
models created by facilitated modelling interventions enables on-the-spot editability because
participants contributions can be changed almost instantaneously by either the OR facilitator
or the participants themselves. However, the degree of editability enabled by the models
depends on whether they are created with or without computer support (Ackermann & Eden,
1994, 2001). That is, computer-supported models afford higher levels of editability than
models created with manual technology (e.g. pens, flipcharts) because the former allows for
many changes to be made in lesser time than the latter. Perhaps more importantly, participants
will construct individual perceptions of whether editability affords them the possibility to
achieve their goals or not, and take action. Thus materiality and affordance may have direct
consequences for how participants interact in the workshop. To understand what these
consequences are and the conditions under which they are likely, we must first recognize that
a models materiality makes certain actions possible and others impossible, or at least more
difficult to achieve. This insight has been used to show how variations in the level of
affordances called forth by the same model, can lead social actors to engage in similar or
disparate dynamics (e.g Franco, 2013)
3. The Materiality of Group Explorer

In this section I will briefly outline the material features of a particular problem structuring
technology: Group Explorer. Developed by Colin Eden and Fran Ackermann at the University
of Strathclyde, Group Explorer is a computer system that enables the construction and use of
causal maps to support the work of teams engaged in strategy making or problem solving tasks
(Ackermann & Eden, 2010, 2011a; Bryson, Ackermann, Eden, & Finn, 2004; Eden &
Ackermann, 2010). The technology is designed to be used in a workshop room, and in this
sense it can be seen as a particular operationalization of the facilitated modelling intervention
approach discussed earlier. To construct the map, team members typically sit at small tables
arranged in a horseshoe-shaped layout, with a console laptop for each table. The consoles are
connected to a master laptop operated by the OR facilitator, who uses it to control the consoles
and assemble team members contributions, which are then displayed on a large public screen
located at the front of the workshop room. The screen is visible to all team members and
provides the focal point around which discussions take place. Members contributions can be
either entered anonymously through the consoles and instantaneously displayed on the screen,
80
or shouted out loud and entered and displayed on the screen via the facilitator. Contributions
are jointly structured between the team members and the facilitator using causal mapping
coding guidelines (Eden, 1988, 2004), which results in a map that is continuously in transition
as team discussions develop.
Figure 1 shows a typical Group Explorer workshop setting, and Table 1 summarises Group
Explorers key materiality dimensions: tangibility, associability, traceability, and editability.
In the next section, I present an empirical case vignette to help ground and illustrate how
Group Explorers materiality and its perceived affordances interact to shape behaviour.
Figure 1: Typical Group Explorer workshop setting
81
Table 1: Materiality of Group Explorer

Materiality
Tangibility
Description
Ability to make its contents
visible and concrete
Group Explorer illustrative features

Anonymous gathering (i.e.
brainwriting).
Multiple map and text views.
Associability
Ability to relate its contents

based on shared attributes
Linking (via console or public

machine)
Clustering and Hiesets
Domain analysis
Centrality analysis
Loop analysis
Potent analysis
Set analysis
Preferencing and ranking
Traceability
Ability to relate its contents

temporally and structurally
Verbal tag and style search

Consequences analysis
Editability
Ability to modify its contents

instantaneously
Concept editing and deleting (via

console and public machine)
Style editing (via public machine)
4. Empirical Case Vignette

In this section I draw on the use of Group Explorer in a workshop held as part of a strategic
review process at Back2work (a pseudonym), a small company located in England. The
companys mission is to help disadvantaged individuals get back into employment by offering
skills training and placement services. The purpose of the workshop was to help the top
management team at Back2work gain a shared understanding of the key strategic issues they
were facing in 2007. I was invited to act as the facilitator of the workshop by the consultancy
team conducting the strategy review for Back2work, and was able to obtain permission to
video and audio record the workshop. These audiovisual data provided the main source for the
analysis presented in this section.
The interaction segment below shows team members discussion about the cross impacts
between issues of succession planning and company ownership. The segment illustrates how
team members behaviours and Groups Explorers materiality (tangibility, associability,
traceability, editability) become interwoven during interaction. Initially the map is changed
82
because some team members perceive that it constrains their ability to articulate their
individual thinking and knowledge about the strategic issues of concern. Later on, however,
the use of the map produces gradual changes in the way team members interact with one
another, as they perceive that the map affords them the possibility of changing the way they
understand and think about the strategic issues of concern.
1
2
F:
Is that [looks and points at map] related to the previous one somehow? Or is it an
independent?
3
4
5
S:
I think it will be impacted. The thing for me is what we want to do is develop our
people which I think that [points at map] said, but we know that the component is something
that needs to be developed in the next three to five years, it will be magnified with succession.
B:
[looks at S] Yes.
K:
[looks at model, then at S, then at model, then nods].
8
9
S:
[looks at C, then points at map] So, theytheyre interrelated you could take them as
separate components but they are interrelated..
10
11
F:
Sure, can we explore that a little bit I mean can I bring that material into the
previous screen? Aahm just ah[locates and brings contributions into map display].
12
S:
[looks at F, then nods].
In the segment above S begins to articulate his particular understanding and knowledge about how
succession planning and ownership issues may be related (lines 3-6). Team members use Group
Explorers tangibility to highlight certain parts of the map by orienting to and pointing at the map
throughout their discussion. In addition, F uses Group Explorers traceability and tangibility to locate
particular contents of the map and make them visually available to team members, respectively (lines
11-12).
S continues to articulate his perspective until this is challenged by C, who starts surveying the
different (number-tagged) issues displayed on the map (lines 14-18, and evaluates the domain-specific
knowledge collated and displayed in the map at that point (lines 22-25, 27-38).
14
15
16
17
18
C:
[looks at map] 39, 15 and 9 I say are different from that, [points at map] I can see
what 4343 is different from 39, 15 and 9 [looks at S then points at map]. 39, 15 and 9 are
do we sell, do we float, do we stay as we are, and that decision will be made at that time. It is
separate from developing people or whatever else [looks at S and uses hands to indicate
separation].
19
20
21
S: [looks at C, then at map] Do you not think that if 39 occurred that that would affect that?
So you wouldnt change your succession planning in the knowledge that in twelve months
time you sell it.
22
23
C:
[looks at map] But you wouldThe way we are structured at the moment if you were
to put a sell pack together, [looks at S] one of the strengths would be the stability of the
83
24
25
business, the growth of individuals within the business, that would be [extends hands and
claps them once] one of the facets to sell that...
26
S:
27
28
29
30
31
32
33
34
35
36
37
C:
..and that sits there [looks at S while pointing at map]. However, [looks at S, puts fists
together] new ownership of the business would determine its direction. No matter what we as
operating people thought [looks at F], if I just bought your business [looks at S] I am actually
gonna tell you [points at map] and Im gonna give you the point of direction that youre
gonna go in [looks at S]. And thats what the difference we have to get away from here is
[places hands together to his left] what can we impact upon and what is beyond our influence
[places hands together to his right]. And [points at model, then looks at S] when you get
down to ownership and desire of, whether its [looks at S, counting with fingers] a market
thats directing you, whether its a civil owner thats directing you or whatever, a
combination, I think that those three things [points at map] have to be dealt with separately
to
[nods] absolutely agree.
It can be inferred from the above exchange that at this point the map is constraining Cs ability to
express his current understanding of and knowledge about the strategic issues under discussion. A
debate between S and C will ensue, during which F will make changes to the map to capture the
debate while ensuring that S, C and the rest of the team keep their focus on the map. In the segment
below, as C continues to elaborate on his views (lines 40-42, 44-46), S suddenly uses the map to
surface the conditions upon which an accommodation may be possible (lines 39, 47-48), which in turn
causes a shift away from the original contrasting positions (49-50).
39
S:
40
41
42
C:
so [points at map] dependent upon the decision [uses hands] whether its float,
whether its sale, whether it is management buyout, will depend on [points at map] how it can
impact, it can impact in different ways [looks at F].
43
F:
44
45
46
C:
[looks at team, counts using fingers] One could be getting rid of you, one could be
developing you further, and one could be taking over, [looks at S] management buyout, you
become the shareholders [opens hands].
47
48
S:
So those things [points at map, then looks at C] becoming in your way will be
impacted on that.
49
50
C:
[looks at map, then at S] In those ways [raises hands], in those three ways, dependent
on the decision.
[looks at map]
[looks at C] Yes.
84
The preceding segment shows how the map affords S the possibility of reaching an agreement
which signals the emergence of a new collective understanding and knowledge. When this
happens F makes changes to the map to highlight the achievement by using Group Explorers
editability and associability material features (lines 51-54, 59).
F:
[looks at map] Yes, so maybe the issue is the nature between, [looks at S] I mean the
impact between these two reds [locates and points at the two issues on map now coloured
in red] will depend whether it is a buyout or [looks at C] it is a management sell-out decision.
S:
[points and looks at map] It will change the shape of the things that you do within
those comparisons.
F:
[looks at S and nods]
C:
[looks at S, then at map] thats the link.
F:
[draws link].
5. Discussion and conclusion

The argument advanced in this paper is that to gain a more nuanced understanding of the
complex nature of OR interventions at the micro-level, we need to pay attention to how OR
technologys materiality and its perceived affordances affect social actors behaviours. As
illustrated in the case vignette, to the extent that social actors have the ability to change their
goals and the ability to change a model, their perceptions of affordances and constraints may
produce behavioural changes that have important interactional consequences for those
involved. This perspective provides a departure from current theorising about OR intervention
impact in at least two ways, each of which has implications for research and practice.
One contribution is to suggest that the dynamic relationship between the social and the
technical (or material in the terminology used here) within OR interventions may be more
usefully explained as one of affordance than one of causality. The affordance perspective
enables us to build a theory of OR intervention that highlights the relation between its social
and technical aspects without foregrounding one or the other. This move, from causality to
affordance, offers a non-deterministic way to theorize the complex nature of OR intervention
practice (Keys, 1989, 1998; White, 2009). One reason for the relative paucity of micro-level
studies of OR interventions, and for the calls made by scholars such as Ormerod (2013) to
provide more insightful accounts of OR practice in the literature, has been the lack of an
integrative theoretical approach such as the affordance perspective. The case vignette shows
how the concept of affordance can provide a useful lens to help us understand the specific
material features of an OR technology that are relevant to social actors behaviours. More
generally, the affordance perspective can help us understand how the goals of social actors and
the materiality of OR technology interact to shape social actors behaviours. For the OR
practitioner community, adopting the language of affordance pushes practitioners to ask
85
themselves many important questions about the impact of their interventions. For example, is
the intended purpose of a particular intervention to afford behaviours and work practices in
particular ways and, if so, how to embody these intentions in intervention design? Did the
intervention design subsequently have the effects intended? If so, why? If not, why not?
A second contribution is to highlight that the effectiveness of an OR technology is likely to be
dependent on the extent to which that technology is perceived to afford or constrain social
actors ability to realise their goals during interaction. How can OR technologies be perceived
by social actors to afford, rather than constrain, opportunities for goal-oriented action? For the
case of models, one aspect that may significantly affect how social actors perceive a models
affordances is the ability to sustain appropriate levels of interpretive flexibility in the models
created (Franco, 2013). Interpretive flexibility is derived from the level of equivocality
embedded in the model (Belton, Ackermann, & Shepherd, 1997; Eden & Ackermann, 2004;
Eden & Ackermann, 2010) As the case vignette shows, the level of interpretive flexibility
regarding how the issues of concern were related was initially sustained to enable social actors
to reconcile initial understandings and positions and, consequently, change their thinking and
knowledge about these issues. The implication for OR practice is that OR analysts need to
ensure that the models they help create exhibit the required affordances for effective
interaction.
Future work is needed to substantiate the generalizability of the argument I have presented
here. A potentially useful direction for future study would be to implement research designs
that compare the use of identical or similar OR technologies in different contexts, or compare
radically different OR technologies in the same or similar contexts. An example of the former
would be to compare a range of problem structuring methods in small versus largegroup
settings; the latter would involve comparing the use of hard versus soft OR approaches
within the same organisation. This type of research is likely to produce a taxonomy of
affordances (and constraints) that would help to better predict the nature and extent of
behavioural changes enabled (but not necessarily determined) by OR technologies.
Another potentially useful avenue for future research relates to the relational character of the
affordance concept. The notion of affordance always presupposes a perceiving social actor,
and different social actors may be afforded different behaviours by the same OR technology.
Thus future research could consider different types of social actors. For example, familiarity
with a particular OR technology or differences in cultural or professional background is likely
to affect how social actors perceive the affordances (or constraints) called forth by the
technology.
To implement this research agenda would require that empirical accounts of OR interventions
include not just details of the models developed within the intervention but, equally important,
the social interaction context in which the models are produced and used. This means being
able to capture and examine how social actors behaviours and the materiality of OR
technology become entangled during interaction, and with what effects.
86
References
Abell P (1993). Some aspects of narrative method. The Journal of Mathematical Sociology,
18(2-3): 93-134.
Ackermann F and Eden C (1994). Issues in Computer and Non-computer Supported GDSSs.
Decision Support Systems 12(4,5): 381-390.
Ackermann F and Eden C (2001). Contrasting Single User and Networked Group Decision
Support Systems for Strategy Making. Group Decision and Negotiation 10(1): 47-66.
Ackermann F and Eden C (2010). Strategic Options Development and Analysis. In M.
Reynolds, & S. Holwell (Eds.), Systems Approaches to Managing Change: A Practical Guide:
135-190. London: Springer.
Ackermann F and Eden C (2011a). Making Strategy: Mapping out strategic success (2nd ed.).
London: Sage.
Ackermann F and Eden C (2011b). Negotiation in Strategy Making Teams: Group Support
Systems and the Process of Cognitive Change. Group Decision and Negotiation 20(3): 293314.
Belton V, Ackermann F and Shepherd I (1997). Integrated Support from Problem Structuring
through to Alternative Evaluation Using COPE and VISA. Journal of Multi-Criteria Decision
Analysis 6(3): 115-130.
Bryson J M, Ackermann F, Eden C and Finn C (2004). Visible Thinking: unlocking causal
mapping for practical business results. Chichester: Wiley.
Callon M (1986a). The sociology of an actor-network: The case of the electric vehicle In M
Callon, J Law and A Rip (Eds.), Mapping the Dynamics of Science and Technology: 19-34.
London: Macmillan Press.
Callon M (1986b). Some Elements of a Sociology of Translation: Domestication of the
Scallops and the Fishermen of St Brieuc Bay. In PJ Law (ed.), Action and Belief. A New
Sociology of Knowledge. London: Routledge and Kegan Paul. (Ed.), Power, Action and
Belief. A New Sociology of Knowledge. London: Routledge and Kegan Paul.
Chemero A (2003). An outline of a theory of affordances. Ecological Psychology 15(2): 181195.
Eden C (1988). Cognitive Mapping: a review. European Journal of Operational Research
36(1): 1-13.
Eden C (1989). Operational Research as Negotiation. In M C Jackson, P Keys and S A
Cropper (Eds.), Operational Research and the Social Sciences pp 43-50. New York: Plenum
Press.
87
Eden C (2004). Analyzing Cognitive Maps to Help Structure Issues or Problems. European
Eden C and Ackermann F (2004). Use of 'Soft OR' Models by Clients: what do they want
from them? In M. Pidd (Ed.), Systems Modelling: theory and practice pp 146-163. Chichester:
Wiley.
Eden C and Ackermann F (2010). Decision Making in Groups: Theory and practice. In P C
Nutt and D C Wilson (Eds.), Handbook of Decision Making pp 231-272. Chichester: Wiley.
Faraj S and Azad B (2012). The materiality of technology: An affordance perspective. In PM
Leonardi, BA Nardi and J. Kallinikos (Eds.). Materiality and organizing: Social interaction in
a technological world pp 237-258. Oxford: Oxford University Press.
Franco LA (2013). Rethinking Soft OR interventions: Models as boundary objects. European
Franco LA and Montibeller G (2010). Facilitated Modelling in Operational Research (Invited
Review). European Journal of Operational Research 205(3): 489-500.
Franco LA and Rouwette EAJA (2011). Decision Development in Facilitated Modelling
Workshops European Journal of Operational Research 212(1): 164-178.
Gaver WH (1996). Affordances for interaction: The social is material for design. Ecological
Psychology 8(2): 111-129.
Gibson JJ (1986). The Ecological Approach to Visual Perception. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Grezes J and Decety J (2002). Does visual perception of an object afford action? Evidence
from a neuroimaging study. Neuropsychologica 40(2): 212-222.
Horlick-Jones T and Rosenhead J (2007). The Uses of Observation: combining problem
structuring methods and ethnography. Journal of the Operational Research Society 58(5): 588601.
Hutchby I (2001). Technologies, Texts and Affordances. Sociology 35(2): 441-456.
Keys P (1989). OR as technology: some issues and implications. Journal of the Operational
Research Society 40(9): 753-759.
Keys P (1998). OR as technology revisited. Journal of the Operational Research Society 49(2):
99108.
Latour B (1987). Science in Action: How to Follow Engineers and Scientists Through Society.
Milton Keynes: Open University Press.
88
Latour B (2005). Reassembling the Social: An Introduction to Actor-Network-Theory.

Oxford: Oxford University Press.
Leonardi PM (2011). When flexible routines meet flexible technologies: Affordance,
constraint, and the imbrication of human and material agencies. MIS Quarterly 35(1): 147167.
Leonardi PM (2012). Materiality, Sociomateriality, and Socio-Technical Systems: What Do
These Terms Mean? How Are They Related? Do We Need Them? In PM Leonardi, BA Nardi
and J Kallinikos (Eds.). Materiality and Organizing: Social Interaction in a Technological
World: 25-48. Oxford: Oxford University Press.
Leonardi PM, Nardi A and Kallinikos J (Eds.) (2012). Materiality and Organizing: Social
Interaction in a Technological World. Oxford: Oxford University Press.
Norman DA (1999). Affordance, Conventions, and Design. Interactions 6(3): 38-43.
Orlikowski WJ (2007). Sociomaterial practices: Exploring technology at work. Organization
Studies 28(9): 1435-1448.
Ormerod R (1996). On the nature of OR-entering the fray. Journal of the Operational Research
Society 47(1): 1-17.
Ormerod R (2013). The mangle of OR practice: towards more informative case studies of
'technical' projects (in press). Journal of the Operational Research Society.
Ormerod R and McLeod J (1984). The development and use of the NCB strategic model. The
Statistician 33(1): 35-49.
Pickering A (Ed.) (1992). Science as Practice and Culture. Chicago, IL.: University of
Chicago Press.
Pickering A (1995). The Mangle of Practice: Time, Agency, and Science. Chicago, IL.:
University of Chicago Press.
Plackett MW, Ormerod R and Toft F (1982). The National Coal Board strategic model.
European Journal of Operational Research 10(4): 351360.
Robinson S, Worthington C, Burgess N and Radnor ZJ (2013). Facilitated modelling with
discrete-event simulation : Reality or myth? European Journal of Operational Research
234(1): 231-240.
Rouwette EAJA (2011). Facilitated modelling in strategy development: Measuring the impact
on communication, consensus and commitment Journal of the Operational Research Society
62(5): 879-887.
Symes E, Ellis R and Tucker M (2007). Visual object affordances: Object orientation. Acta
Psychologica 124(2): 238-255.
89
Tucker M and Ellis R (2004). Action priming by briefly presented objects. Acta Psychologica
116(2): 185-203.
White L (2009). Understanding Problem Structuring Methods Interventions. European Journal
of Operational Research 99(3): 823-833.
White L and Taket A (2000). Exploring the use of narrative analysis as an operational research
method: A case study in voluntary sector evaluation. Journal of the Operational Research
Society 51(6): 700-711.
90

Project Management
Value Management: From Wishful Thinking to Causal Certainty

Roger H Davies
VGrip, London, UK
Systems Centre, Civil Engineering, University of Bristol
roger.davies@vgrip.co.uk
Abstract
This paper explores how Operational Research (OR) is used in Value Management to find,
quantify, optimise and realise value in the context of strategic change programmes. The
inherent difficulty experienced in quantifying the value of change is introduced and a universal
definition of value, designed to eliminate ambiguity, defined. The scale and financial
implications of the failure of programmes to deliver real value are placed into context. Seven
repeating failure patterns are proposed then key Value Management principles defined in terms
of solutions to these patterns and structured as phases in the IMPACT framework, which is
founded on true causality. The application of OR is introduced in the context of shifts in
mental and physical models demanded for this causal approach and demonstrated for each
phase using a case study. Finally, current research in modelling Learning Journeys with the
purpose of creating sustainable value is outlined.
Keywords: Value Management; Performance Management; System Dynamics; Learning
Journey
1.
Introduction
1.1 The Question from Hell

In the strategy business, it is essential for change programme architects, managers and owners
to be able to deal with tough questions. What is the solution, what will it cost, when will it be
ready and what are the risks? Any professional worth their salt will prepare exhaustively for
these and many other interrogations and show that they are in control using work breakdown
structures, plans, project initiation documents, terms of reference and other trimmings of
todays sound programme management practice. Over the last 20 years there have been some
major advances to this end, particularly the professionalisation of project and programme
management, together with support frameworks, such as PRINCE2, Managing Successful
Programmes (MSP) etc. However, the problem with most programme management support
frameworks is that they focus on the process of doing things right and do not provide a
precise, causal linkage with strategic purpose; doing the rights things.
The implication of this focus is that success of programmes still centres on functional delivery,
rather than value outcome delivered. This is reflected in programme Key Performance
Indicators (KPIs:) cost, time and compliance against specification. There is a particular
difficulty with strategic change where high cost initial components are often enablers of future
deliverables to deliver value to stakeholders. For example, infrastructures cost a great deal of
91
R.Davies/ OR56 Keynote Papers, pp.91-105 (2014)
money but do not generally deliver value in their own right. Value can only be realised
through applications which exploit the infrastructures.
There is one question that exposes this weakness more than any other and guaranteed to inject
fear in even the toughest change warrior. It comprises just two words, So what? This is a
very different kind of question and cuts straight to the core challenge which continues to defy
resolution; the recurring failure of change programmes to deliver value against a clear,
committed purpose. I was faced with this question for my first strategic change programme as
a fresh graduate over 30 years ago and have devoted most of my subsequent career to
researching, developing and field testing the answer.
The so in so what? is stakeholder outcomes and the challenge boils down to three simple
sub-questions in the context of stakeholders:
What benefits will be delivered to whom?
When will the benefits be delivered?
Is it worth it?
The first question relates to the identity and magnitude of benefits for specific stakeholders,
the second concerns timing and the third refers to the overall programme value. For all of
these terms we need clear and precise definitions, the first of which is value.
1.2 Definition of Value

The brutal reality is that delivery of demonstrable stakeholder value is still largely in the realm
of wishful thinking. The first prerequisite in redressing this problem is an unambiguous
universal definition of value. It is revealing that when we ask, what is value? in our Value
Breakthrough workshops, we get almost as many different answers as attendees. Typical
answers include: quality, importance, worth, utility, market price and what the customer is
willing to pay. Whilst all these are undoubtedly relevant, they are inconsistent and vague.
Value also depends upon which stakeholder perspective is being considered; advantageous
price to one is high cost to another. In Value Management, we define value as the difference
between stakeholder outcomes, the benefits realised by a stakeholder, and the cost of
achieving them. This is expressed as the Value Equation:
Value = Benefits Costs

In recognition of the need to redress the failure of programmes to deliver value, there is
increasing focus on benefits and powerful techniques are available to support this shift.
However, the Value Equation demonstrates that we need to be equally concerned with the cost
in relation to the benefits realised. This is called Value Focus and lies at the heart of Value
Management.
Key Shift: Value Management is a shift in focus from accounting for what we spend to
accountability for the value that the spend delivers
92
There is much debate on the viability of measuring items traditionally regarded as intangible,
such as customer behaviour, staff motivation and organisational culture, in monetary terms.
However, these soft factors are increasingly more critical to programme value than hard
factors, such as headcount reduction. Consequently, in Value Management the translation of
all hard and soft factors into financial outcomes is deemed essential so that programme
value can be quantified financially as Net Present Value (NPV) and Internal Rate of Return
(IRR). NPV and IRR are two key Discounted Cash Flow (DCF) Techniques. NPV is the
Present Value (PV) of all net free cash flows over a defined programme life using a discount
rate which reflects the time value of money. IRR is the discount rate at which the Present
Value is zero for the same defined programme life. The result is a dynamic causal linkage
between the programme and DCF, called Programme Value Linkage, as shown in Figure 1.
Figure 1 Programme Value Linkage
2.
Why is Value Management important?
2.1 Scale of the Problem and Opportunity

The scale of the failure of programmes to deliver value is breathtaking. Take Information
Technology (IT). A few years ago Sessions (2009) estimated the annual loss to the US
economy of the direct cost of failed IT projects, together with the loss of benefits, as
$1.2Trillion. For the UK, the annual figure was estimated at 200Bn, enough to wipe out
much of the public debt which is still crippling growth. Despite being approximate, these
figures were derived from first principles using reliable, published data and, if anything, are
conservative. The magnitude of these numbers demonstrates that an inability to quantify then
93
deliver real stakeholder value from change programmes, a crucial vehicle for increasing
wealth and improving the lives of the worlds citizens, represents unsustainable waste.
However, the mirror image of this failure is a monumental opportunity. Reversing just a
fraction of this loss has the potential to transform the economic and social landscape, and a
central tenet of this paper is that much of the waste can be transformed into value, not by
throwing even more money at the problem but through a shift in thinking.
2.2 Elusive Value

The fact that programmes still fail to deliver their intended value to such a large degree is even
more perplexing when we consider the level of advances in technology, processes, IT and
programme management. For example, as a Chief Engineer for a leading defence company
some 25 years ago, I observed how difficult it was to keep major programmes on value, even
when using the best available Computer Aided Design (CAD), Manufacturing Resources
Planning (MRP) and, what was then cutting edge, Earned Value (EV) techniques. Every
month, managers of multi-million pound command and control system programmes faced the
agony of reporting to the Board and many of us left the room in tears after the mauling
received.
The failure to deliver was largely blamed on the limitation of, and lack of integration between,
technologies. However, more recently I undertook an assignment to simulate the entire life
cycle of programmes for a global aerospace company. The technology had changed beyond all
recognition. Complex 3-D designs could be translated into machined components with the
depression of a GO button on the CAD system. Work scheduling software orchestrated
production to a level of precision we had not dared to dream of. Yet, as I researched the
monthly reporting process, founded on the same EV techniques we were pioneering a quarter
of a century earlier, identical failure patterns were present. The technology had made no
fundamental difference. If anything, it merely produced failure more efficiently. On digging
more deeply, it became clear that the problem was not technology. It was certainly not due to
any lack of skills or dedication. Neither was it the programme management techniques, which
were extremely robust. The problem lay in the reporting of performance culture and
measurement itself; the mindset, together with the physical models, policies, processes and
procedures that perpetuated it.
The story went like this. Virtually every programme started late with inadequate resources
because the key people needed were busy fixing problems on previous overdue programmes.
Corners cut early in the programme injected errors which remained dormant like a virus,
whilst creating further errors, often across disciplines due to the high degree of
interdependence between functions, such as Systems, Software, Hardware and Mechanical
Engineering, Manufacturing and Integrated Logistics Support (ILS). Each programme was
tracked using a 24 page monthly report detailing just about every aspect of progress. However,
there was only one page that really mattered from a career viewpoint, the financials,
summarised as Earned Value and projected margin. The pain of reporting any deterioration in
margin was so great that Programme Managers did whatever it took to hide bad news until it
became unavoidable, by which time blame could be shifted or lost in the noise. Consequently,
94
the sophisticated reporting designed to identify and resolve problems early was actually being
used to achieve the opposite, to hide them until it was too late to take effective corrective
action.
System Dynamics was used to demonstrate the extraordinary value potentially released
through a cultural shift where problems were exposed and resolved as early as possible.
Interestingly, all the dynamic patterns modelled by the simulation were recognised by the
Operations Director and by bringing the story to life, OR enabled him to target effective
cultural and procedural interventions.
2.3 Reasons for Failure

Many papers and books have been written on why programmes fail to deliver value. This
question is part of the problem. When we ask why? we get reasons, excuses and blame for
the effects. This is human; it makes us feel better by shifting the burden outside ourselves.
The answers also tend to corroborate the question. For example, the most cited reasons for the
failure of programmes to deliver value fall under three broad categories:
Poor specification they did the wrong solution
Lack of buy-in they had the wrong communication and/or culture
Weak programme management they got the implementation wrong
Although these answers are factually correct and make interesting topics for conferences, they
do little to provide pragmatic and effective interventions. We need to dig more deeply by
asking not why? but how?. For example, how specifically are we causing outcomes that
stakeholders do not want? In our commitment to going beyond these surface reasons we
asked a different question, using recent advances in neuroscience, which are behind this shift
from effect to cause:
Key Question: How are we preventing the delivery of value from change programmes?
In addition to researching the many books and papers and attending numerous seminars on the
subject, we approached the question using the same Systems Thinking approach, used for the
aerospace example, on all the programmes that we had been involved with over the previous
30 years. The result of this analysis was the definition of seven repeating failure patterns with
one fundamental underlying common thread. The patterns are listed below and the stories
behind them subsequently explained:
1.
2.
3.
4.
5.
6.
7.
Inadequate specification of stakeholder outcomes

Unrealistic quantification of benefits
Poor causal linkage between programme phases and benefits
Poor value alignment
Imprecise criteria for success and inadequate provision for risk
Inadequate tracking of benefits and overall programme value
Failure to learn and correct
95
The first problem concerns purpose. Of the commonly cited reasons for failure, poor
specification, lack of buy-in and weak programme management, there is one glaring omission,
the wrong problem. It is possible, and quite common, to produce near perfect User and
Functional Specifications for the wrong purpose, defined in terms of intended stakeholder
outcomes (benefits). Secondly, most business cases are still at best fiction, and at worst fraud,
in which benefits are exaggerated and costs understated in order to meet predefined acceptance
criteria. Behind apparently exhaustive detailed data, there is no causal precision underpinning
financial viability. Thirdly, this lack of causal rigour is exacerbated because there is no explicit
traceability between programme deliverables and stakeholder outcomes. For example, whilst
cost escalations, time slips and technical compliance may be fully traceable, the stakeholder
benefit implications of these problems are neither known nor quantified. The fourth problem,
alignment, relates to a mismatch between the programme and real performance drivers in the
business, including soft factors. The fifth and sixth problems concern the inherent
uncertainty and associated risk of strategic change programmes, much of which emanates, not
from the risk itself, but from failure to define precise criteria against which to track and correct
variances from a value perspective. Finally, all these problems are mutually reinforcing and
repeated because there is no process for learning. As we stated previously, most programme
management techniques and frameworks focus on doing things right, rather than doing the
right things. We will discuss later the imperative to shift programme management from a
mechanical process to a Learning Journey from minded purpose to realised performance,
during which value is realised through a combination of right first time and perpetual
corrective feedback.
3.
What are the Key Value Management Principles?
The repeating problems result in a fundamental disconnect between stakeholder benefits and
the programme intended to deliver them. In terms of the seven repeating problems, this means
that stakeholder outcomes are not specified, the performance drivers by which value is created
are not quantified, linked or aligned to the programme, risk is allowed to remain and inject
further problems, value is not tracked and there is no process for learning and breaking these
patterns.
Correcting these problems and delivering value from change programmes demands the total,
non-negotiable commitment to master the quantification of cause and effect. Specifically, this
means the explicit, measurable causal linkage between capabilities delivered by the
programme (deliverables), the business performance drivers and stakeholder benefits.
Key Link: The common thread is causation

3.1 IMPACT Value Management Framework
The flip side of every problem, if reframed appropriately, is an opportunity. In this approach to
Value Management, we reframe the repeating problems into key principles which become
phases in a framework providing the foundation for learning and repeating success. The
96
structure used is an extension of the V Model which is widely used in Systems and Software
Engineering to provide double loop learning to ensure that we are doing the right things
(validation) and doing things right (verification). Variants are also used in performance
coaching and it is one of the most powerful models for change. The IMPACT Value
Management Framework is shown in Figure 2.
Figure 2 Value Management Framework

A major limitation of a typical V Model used for software is the focus on delivering a given
performance against user requirements which, as we have stated previously, is no guarantee of
value delivery. In the IMPACT framework, additional higher level phases are included to
incorporate purpose, defined specifically as stakeholder benefits. In this way, IMPACT
represents a double loop learning structure which provides explicit and perpetual causal
linkage between purpose and performance as outlined below:
Intention: Define the business vision and specific stakeholder benefits, together with high
level programme deliverables, with clear, unambiguous ownership and accountability for both
deliverables and benefits.
Model: Model key cause and effect relationships in the business and focus on the most critical
performance drivers, the leverage points against which to target the interventions, and assign
driver ownership.
Programme: Link performance drivers and consequential benefits dynamically to the
programme deliverables and specific programme phases which output the deliverables.
Alignment: Target, time and align deliverables with the business in order to optimise
stakeholder value most quickly for least cost and risk to produce an Implementation Strategy,
which is a programme plan or portfolio aligned with the business vision.
97
Certainty: Determine the bounds within which we can be as certain as pragmatically possible
that the programme will deliver intended value by subjecting it to extremes using sensitivity,
risk and scenario analyses, along with exhaustive user and other stakeholder reviews. The
resulting output is an approved Baseline Business Case.
Track: In addition to cost, time and quality of deliverables, track changes in performance
drivers, together with consequential benefits, during and post implementation, to ensure that
intended programme value is realised. This is called Value Realisation and is one of the most
neglected, yet critical, aspects of programme management.
Defining purpose is the one place in Value Management where we ask why? Specifically,
why is it important to stakeholders? This question elicits stakeholders values which reflect
purpose. Then in defining the performance criteria for achieving outcomes we ask how?, to
determine the measure of value in the context of purpose?
Key Law: When defining purpose ask why? When performance is your game ask
how?
3.2 Wicked Problems
Todays economic and social landscape is shifting in three interrelated dimensions: scope,
complexity and pace. Scope concerns the breadth and depth of factors which must be
considered, for example the impact of globalisation. Complexity relates to the relational
interdependence between factors and disproportionate influence of apparently small factors on
the entire system, for example behavioural shifts in a small number of customers can create
massive viral effects through social media. Complexity also increases the importance of soft
factors, such as values and beliefs which dictate human behaviour. Pace means the
accelerating rate of change and associated behaviour, for example the explosion in, and
profound effects of, social media.
The result of this increasing scope, complexity and pace is the emergence of wicked problems.
A key distinction of wicked problems is that they are not conducive to complete solutions, a
characteristic that takes us into the domain of Systems Thinking, upon which Value
Management is founded. Some argue that because we cannot solve wicked problems, it is not
worth trying, and is dangerous to do so. In Value Management we take a different view, that
we now have sufficient theoretical and experiential foundation to design effective
interventions, which offer significant improvements provided we gain sufficient understanding
of cause and effect. In order for us to respond effectively within this new reality, we need
mental and physical models that enable us to make sense of it.
3.3 Commitment to Causality

Consequently, to be effective we must know how the universal laws of cause and effect are
manifested under this new reality. For practical purposes this means looking beyond surface
98
events to underlying patterns of cause and effect, then quantifying the relationships to provide
a pragmatic level of certainty upon which to direct, drive and deliver value from change.
Traditionally soft issues, such as values, beliefs and culture, have been designated
intangibles and excluded from financial business cases on the grounds that they cannot be
translated into money. In the world of wicked problems it is no longer valid to separate hard
factors, such as technology and finance. For example, the success or failure of new products is
now dictated less by mass advertising and increasingly by large numbers of interactions
between customers concerning their satisfaction or otherwise using social media. This means
that businesses need to know their customers at an altogether more profound level; what is
important to and motivates them, i.e. values, in order to match their wants and/or needs more
precisely. This takes us into the realms of social sciences and psychology, which in Value
Management are integrated with performance and programme management through a systems
perspective and modelled using OR techniques, most notably simulation.
4.
Application of OR in Value Management Implementation
The application of dynamics modelling in Value Management is illustrated using a case study,
JANET(UK) Business Transformation Programme. JANET(UK) provides one of the world's
leading research and education networks serving 18 million users. There were several reasons
driving the need for strategic change. Most critical was the growing pressure to reduce the cost
of service provision from both the Government, committed to stringent cuts across public
funding, and user institution customers who, in a challenging financial environment, become
increasingly discerning in what they required. In response to these shifting market dynamics,
JANET(UK)'s CEO, Tim Marshall, was driving the transition from a largely centrally funded
model to one where there may be a variety of variable revenue streams. A primary component
in achieving the shift to a commercial model was the development of a Balanced Scorecard
(BSC) to reflect the new value dynamics, with which to track and correct performance across
the company.
4.1 Intention
The BSC is a leading framework for performance management and comprises a number of
stakeholder perspectives, typically four, for which reason they are often referred to as
quadrants. These are: Financial, Customer, Process and Resources, or Staffing as in this case.
(The Process and Resources perspectives are often called Internal and Learning and Growth
respectively). The original thinking behind the BSC was to provide a set of indicators that
reflected all aspects of the business, in addition to traditional financial measures. It was
recognised from the start that to be of real value there needed to be some causal connection
between perspectives and associated measures.
However, defining causality between measures presents a challenge because it is inherently
difficult to capture all the linkages between the many performance measures in a complex
business without getting lost in the complexity. Consequently, instead of linking specific
measures, causal linkages are defined between objectives, which specify what must be
99
achieved in order to realise the vision. As the vision is generally expressed in financial terms,
the causal connections tend to follow paths, called Strategic Themes, routed from Resources
through Process to Customers and finally to Financial. Each objective is assigned a number of
underlying measures of which there are two types, lead and lag indicators. Lead indicators are
generally drivers that must change in order to realise intended outcomes, measured by lag
indicators. The resulting structure is called a Strategy Map, which is shown for JANET(UK) in
Figure 3.
Figure 3 JANET(UK) Strategy Map
100
During this initial phase, simulation was used to develop and test the essential dynamics of the
problem. This is called Causal Hypothesis, a term coined by (Richmond 2004), a pioneer and
world leader in System Dynamics. The Causal Hypothesis is a premise concerning the
dynamic behaviour of one or more key performance parameters. This has two aims, first it
provides precision in defining the problem and secondly, it points to the likely combination of
feedback loops that are causing the behaviour. In the JANET(UK) case, we developed a high
level System Dynamics model framed around the Balanced Scorecard structure as shown in
Figure 4.
Figure 4 JANET(UK) High Level Dynamics Model
101
4.2 Model
For business reporting, the BSC provides a pragmatic tool which can be readily developed into
dashboards with red, amber, green (RAG) notation to reflect the status. This offers great
opportunities for software companies and consultants to build in ever more sophisticated IT
solutions which project the delusion of precision and control. However, from an OR
viewpoint, the absence of explicit causal linkage between specific measures represents a lack
of rigour, which is inadequate for this approach to Value Management. Consequently, we built
a dynamics model which reflects the Strategy Map and not only defines the performance
drivers themselves but also the causal relationships between them. Given that the Strategy
Map is built up from objectives, it makes sense to structure the dynamic Balanced Scorecard
in the same way, which we did for JANET(UK).
4.3 Programme
There were five work streams in the JANET(UK) Business Transformation Programme,
relating to the key outcome deliverables deemed essential to achieve the objectives and overall
vision:
Central Service Delivery
Customer Engagement
Product Portfolio Management
Commercial Culture
Management Culture
A financial business case, using a Discounted Cash Flow (DCF) analysis, was built to quantify
and baseline the transformation programme and the BSC used to track actual changes and
value realised. The causal linkage between the deliverables and financial benefits were derived
with the Operations Director with reference to insights from the simulation and linked
dynamically to the Net Present Value (NPV), using the Value Management Toolset, as
shown in Figure 5.
102
Figure 5 Discounted Cash Flow Analysis
4.4 Alignment
In Value Management, we marry two types of alignment: Business Alignment concerns causal
coherence between value drivers, whilst Programme Value Alignment relates to the structure,
sequence and timing of deliverable phases with the aim of optimising value through changes
in performance drivers. This requires that the performance framework is cascaded throughout
the entire business in order to provide a clear line of sight from any part of the enterprise to
the vision. For JANET(UK) the Strategic level BSC was cascaded to twelve operational level
scorecards.
4.5 Certainty
A critical element in Value Management is defining precise linkages between business value
drivers and programme deliverables. This is achieved by extending the BSC concept of
Strategic Themes from linking objectives to causal relationships between specific underlying
performance measures. This presents two major advantages. First, business managers at the
point of power, where changes actually take pace, have a clear view of expected changes in
those drivers for which they are accountable. Secondly, it provides full causal tracing through
the enterprise for reporting and management purposes.
103
4.6 Track
Driver values for operational BSCs were collected and consolidated for reporting to the
Executive Board Meeting each month. As a further development, an industry strength
Business Intelligence (BI) solution was built with feeds to and from the dynamics model. This
served two purposes, calibration using actual historic data and directing decisions for business
as usual management and strategic investment. This was part of our ongoing research and
development into Dynamic Performance Management.
Tim Kidd, Operations Director, JANET(UK), quoted for the book Value Management (Davies
and Davies 2011) Before applying the Value Management approach, we had tried measuring
performance indicators but we didn't run our business by them because we weren't convinced
we had the key ones. Conversely, the causal approach, in particular dynamics modeling,
enabled us to define the most critical drivers of value against which we are now confident we
are measuring the right things - the things that really influence the outcomes we want and
reflected in our business transformation programme.
5.
The Future: Modelling Learning Journeys
In this paper, we have stressed the need for two shifts in the way we approach strategic change
to deliver real stakeholder value, particularly when related to wicked problems. First, we
discussed the need for a causal approach and proposed a framework, IMPACT, based on the V
Model, used in Systems and Software Engineering, for structuring problems, formulating
interventions and delivering stakeholder value. Secondly, we argued for a shift in how we
perceive and relate to reality through more precise mental and physical models, which then
need to be supported through appropriate policies, processes and procedures.
However, there is a third shift we need in order to facilitate the first two, the capacity to learn
and change in response to that learning, for which there are two key terms, Learning Power
and Learning Journey. Learning Power is the measure of an individual or groups capability
to learn, determined using seven interactive Dimensions, for example, Critical Curiosity, the
desire to learn, and Resilience, the ability to remain committed. Learning Journey refers to the
expedition from purpose to performance. It follows therefore, that Learning Power is a key
measure of the ability to undertake successful Learning Journeys.
The Graduate School of Education at the University of Bristol (Deakin-Crick et al 2014) have
developed the means to measure Learning Power and target interventions based on the results.
The measurement has been validated across large populations of learners, spanning western
school pupils to indigenous Australian Aboriginals. A critical finding is the need for
individuals to take responsibility for their Learning Journey, from selection of purpose to
delivery of performance.
The importance of this work cannot be understated in the context of wicked problems facing
todays world population. All purposeful change requires shifts in human behaviour. As we
have seen, no end of method and tools for the physical delivery of change, doing things right,
104
will succeed in delivering value unless directed to appropriate purpose, doing the right things.
I am currently undertaking doctorial research with the University of Bristol to develop a
generic Learning Journey for the creation of sustainable stakeholder value.
References
Davies R and Davies A (2011). Value Management: Translating Aspirations into
Performance. England: Gower
Deakin-Crick R, Stringher C, and Ren K (2014). Learning to Learn: International Perspectives
from Theory and Practice. England: Routledge.
Richmond B (2004). An Introduction to Systems Thinking with iThink. USA: isee systems,
inc.
Sessions R (2009). Cost of IT Failure: What does IT failure cost us annually? A Lot. Available
at http://simplearchitectures. Blogspot/2009/09/cost-of-it-failure.html, accessed 31 May
2014.
105

Simulation
Cloud Computing for Modelling & Simulation

Simon J. E. Taylora and Anastasia Anagnostoua
a
Modelling & Simulation Group, Department of Computer Science

Brunel University, London, UK
simon.taylor@brunel.ac.uk
Abstract
Cloud computing offers scalable, elastic, on-demand access to a range of computational
resources. Arguably, the time taken to obtain results from a simulation can limit the range and
accuracy of experiments. Low cost cloud computing resources can potentially speedup
simulation experimentation and produce more results more accurately. However, developing
cloud computing solutions for industry is difficult without appropriate expertise. Working
towards a general approach to cloud computing for simulation, this paper discusses how the
CloudSME Simulation Platform is providing a novel approach to cloud computing for
modelling & simulation. An example of the agent-based modelling & simulation system
REPAST is used to show how the platform can make it easy for end users to use the facilities
of multiple clouds.
Keywords: Modelling; Simulation; Cloud Computing; High Performance Computing
1.
Introduction
In modelling & simulation (M&S) we build and then simulate models as we experiment with
different scenarios. Experimentation (and testing) can take a long time. As simulation
experimentation can consist of many independent simulation runs (and replications), one
might save time by running the simulations in parallel across many computers (and multiple
CPUs in these computers). So why not do this? Distributed computing approaches such as
Grid computing and e-Infrastructures have had a major impact on how M&S can be speeded
up in scientific disciplines. Experience has shown that these can be successfully applied in
industry (Kite, et al. 2011). However, despite delivering faster and more accurate results, this
requires a lot of in-house expertise, computers and (potentially) licences that need to be paid
for up front.
Cloud computing attempts to provide convenient access to an on-demand network of shared,
configurable computing resources that can be rapidly provisioned and released with minimal
management effort or service provider interaction (Mell and Grance 2011; NIST 2013). In
terms of the above, a simulation could be implemented as a cloud service using licences
running on cloud computers on a pay-as-you-go basis. However, while this is a potentially
excellent deal for the end user, the simulation company would still need the expensive inhouse expertise to provide this. The Cloud-based Simulation platform for Manufacturing and
Engineering (CloudSME) project (www.cloudsme.eu) is attempting to make cloud computing
more feasible for M&S and has developed the CloudSME Simulation Platform. CloudSME is
initially supporting the development of five different simulation software: discrete-event
simulation (SIMUL8 - www.simul8.com), computational fluid dynamics (ASCOMPs
TransAT - http://www.ascomp.ch/transat/), 3D modelling (INGECONs 3D Scan Insole
106
S.J.E. Taylor and A. Anagnostou/ OR56 Keynote Papers, pp.106-117 (2014)
Designer), aircraft maintenance logistics (2MOROs Bfly software) and agent-based

modelling & simulation (REPAST).
The paper is structured as follows. In section 2 we review aspects of cloud computing.
Section 3 introduces the key features of the CloudSME Simulation Platform that is being used
to implement cloud versions of the above software. Section 4 gives an example of the
platform in action with the agent-based modelling & simulation system REPAST. Section 5
concludes the paper.
2.
Cloud Computing
Cloud computing used as a term to access internet-based computing resources appeared in

the mid-2000s. The National Institute of Standards and Technology (NIST) has made efforts
to standardize the terminology of Cloud computing (NIST 2013) and defines it as a model for
enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable
computing resources (e.g., networks, servers, storage, applications, and services) that can be
rapidly provisioned and released with minimal management effort or service provider
interaction. Elasticity is the main concept of Cloud computing. That is, the use of
computational resources, storage, applications, etc. that can be instantly increased according to
user needs, and ceases when the user does not need these services.
Cloud services can be deployed according to the organizational structure and the security
requirements of the users. Currently, there are four main cloud deployment models (see Figure
1) for cloud resource provision.
Private clouds can be accessed only by a single organization. Services are accessed via
private networks and the cloud consumers are members of the organization.
Community clouds are effectively an extension of a private cloud. A community cloud can
be accessed by more than one organization with common interests. The infrastructure and
datacenters are located on or off site and can belong to one or more of the organizations in
the community or a third party.
Public clouds services can be accessed by the general public. These are available by
public networks, most commonly via the internet. The infrastructure and datacenters are
located on the premises of the cloud provider (e.g. Amazon Elastic Compute Cloud (EC2)
and cloud services provisioned by academic institutions).
Hybrid clouds are a combination of two or all the three of the above. This model allows
service portability by using standardized methods. An issue in hybrid cloud deployment
could be the lack of standards for communicating data and services among different cloud
providers.
107
Figure 1: Cloud deployment models
2.1 The Cloud Computing Stack

The cloud computing stack reflects the cloud computing service models. There are three
defined main service models, as shown in Figure 2. From the bottom-up after the physical
hardware and networking layers, these are Infrastructure as a Service (IaaS), Platform as a
Service (PaaS), and Software as a Service (SaaS). Cloud consumers (users) typically can
access all three services via a web interface or some language supported API.
Figure 2: The Cloud computing stack

IaaS stands at the lowest level of the main provided service models. IaaS does not give access
to the physical layer, however allows access to a virtual infrastructure where the user can
deploy operating systems and application software. Also, an IaaS cloud consumer can deploy
network features, such as security settings and firewalls. PaaS stands on the top of the IaaS
and allows for software deployment and configuration on the existing platform. The PaaS
cloud consumer does not control or manage the operating systems and network features. The
users of this service can deploy existing software or develop software with the provisioned
libraries and compilers by the cloud provider. The most commonly used service model is the
SaaS, which sits atop the PaaS. A SaaS cloud consumer has only application level access.
Users can use the provided applications deployed as online services with no knowledge of the
platform or the infrastructure. The applications can be accessed by web browsers or program
interfaces.
108
We now describe the CloudSME Simulation Platform that has been developed to make Cloudbased M&S accessible to both simulation service providers and consumers.
3.
The CloudSME Simulation Platform
The CloudSME Simulation Platform (CSSP) has been developed by the European Cloudbased Simulation platform for Manufacturing and Engineering (CloudSME) project
(www.cloudsme.eu) that aims to develop cloud-based simulation services and platforms that
enables SME end-users, especially from manufacturing and engineering, to access to
simulation services and speed-up experimentation via Distributed Computing Infrastructures
(DCIs) (e.g., cloud, grid, HPC clusters, etc.). CloudSME is creating a combination of PaaS
and SaaS solutions to support simulation, i.e., a PaaS solution will allow simulation software
providers or consultant companies to build SaaS solutions for end-users and deliver on-line
simulations with access to HPC.
An important feature of this is that given the complexity of cloud-based solutions, the access
mechanism to the simulation software should completely hide the complexity and potential
heterogeneity of the cloud platform and should allow the simulation software to utilize
different types of clouds based on their deployment models and the underlying cloud
middleware. The CloudSME project does this by combining Cloud Brokering services
(provided by CloudBroker, CH (www.cloudbroker.com)) and WS-PGRADE/gUSE workflow
development and deployment services (Kacsuk et al. 2012) into the CSSP. This is the PaaS
that supports the deployment of simulation services and their access to HPC cloud resources in
a user-transparent way.
Figure 3 shows the main parts of the CSSP and the split into three layers: the Simulation
Application Layer, the Cloud Platform Layer and the Cloud Resource Layer. The Cloud
Platform Layer is split into two: CloudBroker and gUSE. To create a simulation service on the
platform the simulation software must be deployed on one or more of the clouds indicated in
the figure. For each of these clouds, CloudBroker must have a cloud adaptor that allows an
instance of the simulation software to be created, run and managed on a virtual instance on
that cloud. The simulation software must be converted for deployment on a cloud. The
approaches to do this vary on operating system. Typically the simulation executable is
uploaded to a repository with runtime scripts that configure the software for execution (when
jobs are submitted for processing). These executables and scripts form M&S-as-a-Service
(MSaaS) Application Patterns and Deployment configurations. The CloudBroker services are
accessed via a web-based workflow management system. Here a workflow is created using
WS-PGRADE that specifies how the software is run and managed. The workflow is stored on
a workflow server (gUSE) and launched either directly via WS-PGRADE, a web-based portal
(using gUSEs ASM Module) or some API (using the REMOTE API). For flexibility the
workflow is agnostic and therefore has the functionality to run on any Distributed
Computing Infrastructure (DCI) as long as the interface is supported by the workflow manager
via the DCI Bridge. To run a simulation service the user runs the workflow and supplies the
appropriate models and data. The workflow manager then manages this across the DCI Bridge
109
on to CloudBroker and across multiple cloud instances. We now describe the current
implementations of CSSP on CloudBroker and gUSE.
3.1 CloudBroker
The CloudBroker (CB) Platform is a web-based application store for the deployment and
execution of compute-intensive scientific and technical software in the cloud. It is provided as
a cloud service that allows for on-demand, scalable and pay-per-use access via the Internet.
CB currently offers the platform as a public version under https://platform.cloudbroker.com,
as hosted or in-house setup, as well as licensed software. This also means that the platform can
be run at different physical places and under different legislation if desired.
CB uses IaaS from resource providers and provides PaaS to software vendors and SaaS to endusers. It offers a marketplace where users can register, deploy, charge for and use their own
cloud resources and application software or use resources and software provided by others
(e.g., CB itself). Surcharges for platform usage are derived as a percentage of the resource and
software prices. From this follows a freemium model, that is, if resource providers and
software vendors set zero prices for their resources and software, the corresponding platform
usage is also free.
Figure 3: The CloudSME Simulation Platform
110
CB incorporates adapters both to public and private cloud infrastructures, and both to compute
and storage resources. Currently supported technologies include:
Amazon EC2 compute and S3 storage resources
IBM SmartCloud Enterprise compute (SoftLayer in preparation) and Nirvanix storage
resources
OpenStack compute resources (storage in preparation)
Eucalyptus compute and storage resources
OpenNebula compute resources (in preparation)
On the application side, CB is cross-domain and supports all kinds of non-interactive, batchoriented applications, both serial and parallel ones, with a focus on software running under
Linux. Windows is also supported. CB has already ported various software in different
scientific and technical fields to the platform, in particular in chemistry, biology, health and
engineering (see www.sci-bus.eu).
CB can be accessed in several ways. Its main two operation modes to manage and use
software in the cloud are either as direct front-end or as back-end middleware service. For the
former, the platform can be accessed with any regular web browser as a service via the
internet. This is fine for first-time and individual usage. However, for frequent, advanced and
automatic usage, API access is provided. These include a REST web service interface, Java
client library and Linux shell command line interface (CLI). Via its different APIs, CB can be
utilized by front-end software as middleware to allow access to applications in the cloud.
CB has been integrated with the WS-PGRADE framework within the FP7 SCI-BUS project
(SCIentific gateway Based User Support www.sci-bus.eu). SCI-BUS aims at making
scientific gateways available that can transparently utilize different distributed computing
infrastructures. The corresponding web portals provide workflows using the WS-PGRADE
framework and can access applications on commercial and private cloud infrastructures using
the CB Platform.
The platform employs industry standard application and server technology and is typically
operated in industry standard secure data centers. The different cloud infrastructures it utilizes
usually also provide authentication mechanisms, isolated virtual machines and securitycertified cloud technologies and data centers.
3.2 gUSE
Overall, to develop an application to run on a Cloud or a Grid a developer will use the
graphical environment of WS-PGRADE to describe the sequence of tasks to run the
application by describing the workflow of the application. A workflow is a directed acyclic
graph of nodes and arcs where a node represents a specific task and an arc (typically) a file
transfer between tasks. Arcs are connected to specific ports on each node to distinguish actions
111
specific to those edges. Nodes can be configured to perform virtually any computing task via a
visual interface. Some specialist nodes have been created to support High Performance
Computing (HPC) tasks. Once an applications developer has created a workflow, it is saved in
the gUSE Services environment. This is a complete environment for workflow support and
execution. It provides secure access and authorization to DCI resources by using appropriate
security certification. In the CSSP architecture this transparently uses the security processes of
the CloudBroker platform. It enables the management, storage and execution of workflows
and has flexible deployment options (single hosting/distributed hosting to optimize resource
usage or increase performance). It also offers data services (user database, proxy credential
database, application code database, workflow database, application history and change log
database, etc.) and control services (e.g., job submission, workflow enactment, etc.). The
gUSE Information System supports workflow discovery and naming services. To execute a
workflow, gUSE uses the DCI-BRIDGE to access the resources of various DCIs (in the CSSP
these are the virtual machines of the Cloud provider selected in the workflow). The DCIBRIDGE is a web service-based application that provides standard access to DCIs via DCI
plug-ins. Workflows are submitted and managed using the Basic Execution Service (BES)
interface. This hides the access protocol and the technical details of the DCIs and uses the
standardised job description language JSDL. BES uses a job registry, an input queue, an
upload manager to manage the execution of jobs generated by a workflow.
WS-PGRADE/gUSE has been developed by MTA SZTAKI as an open source software that is
available in Sourceforge at: http://sourceforge.net/projects/guse. The framework is very
popular among the different user communities. This is proven by the fact that so far there has
been more than 5,000 downloads from more than 40 countries from Sourceforge. It is actively
used by many different user communities and NGIs both in and outside Europe. In the SCIBUS project 13 different user communities and four companies are developing their own
gateway based on WS-PGRADE.
4.
Cloud Computing Example
The implementation of each of the simulation software mentioned in the introduction are close
to being fully ported to the CSSP and will be discussed during the keynote presentation. We
have also ported REPAST SIMPHONY to the CSSP. The Recursive Porous Agent Simulation
Toolkit is a free and open-source ABMS environment developed by The University of
Chicago and the Decision and Information Sciences Division at Argonne National Laboratory
(www.dis.anl.gov). It is widely used for modelling complex and dynamically adaptive
behaviours of a system (North et al. 2013; Taylor 2014). We present a short case study using
the JZombies model, a well-documented ABMS model that is used as a tutorial for the repast
for java edition http://repast.sourceforge.net/docs/RepastJavaGettingStarted.pdf.
The JZombies model is a non-terminating simulation, where zombie agents move towards the
grid point with the most human agents and infect the latter by turning them into zombies. First,
we modified the model to a terminating simulation, where the end point is when all the
humans have turned into zombies. For illustration, each run consists of N replications. We
have multiple runs (experiments) to show how we divide work across a cloud.
112
A model with its required repast plugins are deployed on different cloud resources supported
by the CSSP, i.e., Amazon EC2, BIFI cloud (OpenStack Grizzly provided by the University
of Zaragoza, ES), and UoW cloud (OpenStack Folsom provided by the University of
Westminster, UK). All cloud resources support both Linux and Windows applications apart
from UoW cloud, that currently supports only Linux applications. In order to be able to run the
application in all available cloud resources, we developed both Windows Batch and Shell
Script executables. The only requirement for the cloud resources was to have java runtime
installed. In the following example repast will run in Windows on Amazon EC2 and BIFI and
in Linux on UoW.
The workflow creation is supported by the WS-PGRADE. The first step is to use the graph
editor for creating the workflow. The graph editor provides a user-friendly drag-and-drop
environment, where the structure of the workflow can be created. Figure 4 illustrates the graph
editor, where the bigger squares denote jobs and the attached smaller squares denote input and
output ports (green and grey squares, respectively). Each job can have more than one input
and output. In the example of Figure 4, the initialize node generates two different outputs, that
are then fed to the respective job nodes. Once all the job runs are finished the outputs are
collected to the output collector node. This example shows the jobs being distributed to the
three clouds.
Simulation experiments are usually parametric runs and each set of parameters execute for a
number of replications. CSSP can distribute the parametric runs and the replications to
different nodes on the available cloud resources as jobs. The jobs are distributed and managed
from WS-PGRADE over the DCI Bridge to CloudBroker. CloudBroker then creates and
manages the instances on the different clouds as instructed.
Figure 4: WS-PGRADE graph editor

The graphical workflow is configured at the WS-PGRADE web portal. The simplest structure
is a one job workflow. An example of a simple one job workflow configuration for a
113
parameter sweep execution of the JZombies repast model is shown in Figure 5. In the job
node, the first step is to select the type of DCI. As mentioned earlier, CSSPs cloud resources
are managed by CB. CB account authentication is needed in order to be able to access the
portal. Then the software that will execute in the workflow (in this case Repast 0.1 is the
repast model) and the executable (Windows batch file) are selected. From the available cloud
resources, once a selection of the instance type is made, an estimation of the job cost is
displayed. The next step is to configure the input and output ports. In repast batch runs, the set
of parameters is indicated in an XML file, therefore this is provided in the input port. The
output port will export the resulting files from the job run. The output files can be downloaded
after the completion of the job.
An example workflow that can utilize all the available cloud resources in CSSP (where repast
is deployed) can be seen in Figure 6. This takes a set of repast simulation runs and distributes
them over three clouds (these could be the same one we do this to show the flexibility of this
workflow approach). The initialize node will output the XML parameter file that is required
as input for every REPAST simulation run. Then, each node can initiate several instances each
in the three CSSP clouds on which repast is deployed. (i.e., Amazon EC2 and BIFI will run
the Windows application and UoW will run the Linux application). Note that for the
commercial cloud there is cost estimation information, while for the academic clouds, the cost
is zero. Also, in the job configuration view, it can be seen that for the BIFI and Amazon EC2
clouds, the Windows batch executables is selected, while for the UoW cloud, it is the shell
script executable.
Furthermore, the initialize node can be configured as a generator node. A generator replicates
the required input file(s) as many times as indicated (or the size of the cloud resources
dictate). In this way, many instances can be instantiated automatically. For example, if the user
wishes to run X number of replications, the generator replicates the input file(s) X times and
the workflow manages the distribution of jobs across the available resources.
The way to create and configure workflows is by no means restricted by the above examples.
These are some indicative examples of the CSSP capabilities. The extended demonstration of
the infrastructure will be the subject of future work.
114
Figure 5: CloudSME workflow configuration
5.
Conclusions
This paper has introduced the CloudSME Simulation Platform and how it is being
used to simplify the development of cloud computing for M&S. It is hoped that
efforts and partnerships such as these will pass on the benefits of high performance
cloud computing to a wide range of end users, even those with models that run quickly
by enabling even more experimentation in the same timescales. Visit the CloudSME
website for more information and updates.
Acknowledgments
This work is funded by the CloudSME Cloud-based Simulation platform for Manufacturing
and Engineering project No. 608886 (FP7-2013-NMP-ICT-FOF).
115
Figure 6: Jobs distribution in different CSSP cloud resources
References
Kite S, Wood C, Mustafee and Taylor SJE (2011) SAKERGRID: Simulation Experimentation
using Grid Enabled Simulation Software. In Proceedings of the 2011 Winter Simulation
Conference. Association for Computing Machinery Press, New York, N.Y. 2283-2293.
Mell P and Grance T (2011). The NIST Definition of Cloud Computing. Accessed May 08,
2014. http://csrc.nist.gov/publications/nistpubs/800-145/SP800-145.pdf.
NIST (2013). NIST Cloud Computing Standards Roadmap. Accessed May 08, 2014.
http://www.nist.gov/itl/cloud/upload/NIST_SP-500-291_Version2_2013_June18_FINAL.pdf.
North MJ, Collier NT, Ozik J, Tatara ER, Macal CM, Bragen M and Sydelko P (2013).
Complex Adaptive Systems Modeling with Repast Simphony. Complex Adaptive
Systems Modeling 1(3). http://www.casmodeling.com/content/1/1/3.
116
Taylor SJE (2014). OR Essentials: Agent-based Modelling & Simulation. Basingstoke,

Hampshire, UK: Palgrave Macmillan Ltd.
Kacsuk P, Farkas Z, Kozlovszky M, Hermann G, Balasko A, Karoczkai K and Marton I
(2012). WSPGRADE/gUSE generic DCI gateway framework for a large variety of user
communities. Journal of Grid Computing 2012 10(4): 601-630.
117

Sustainable Supply Chains
A Methodological Framework for Green Logistics Networks under

Periodic Review Replenishment Policies
Eleftherios Iakovou a, Ioannis Mallidis b, Dimitrios Vlachos c, Rommert Dekker d
a ,b ,c
d
Aristotle University, Department of Mechanical Engineering, Thessaloniki, Greece

Erasmus University, Department of Econometrics, Rotterdam, Netherlands
eiakovou@auth.gr, imallidi@auth.gr, rdekker@few.eur.nl, vlachos1@auth.gr
Abstract
In this manuscript we propose a methodological framework for evaluating the cost and CO 2
emissions trade-offs of implementing periodic review inventory planning policies at each node
of a frequently encountered in practise multi-echelon logistics network design instance. The
proposed methodology determines optimal order delivery frequencies and stock levels at each
node and its application is further illustrated through a real-wold case study of a white goods
retailers logistics network. Various managerial insights are obtained, while it is further
documented that a CO2 optimal solution results in lower transportation frequencies and thus in
lower transportation CO2 emissions and costs, but also, in higher holding and backorder costs.
Keywords: CO2 emissions; Periodic review policy; Sustainable supply chain; Logistics
networks
1.
Introduction
The offshoring of production operations has resulted in the development of extensive and
complex supply chain networks that have to connect distant production locations with various
supply chain stakeholders and demand points around the world. Cost minimization of supply
chain operations has traditionally been of the outmost interest for supply chain executives (at
least for chains dedicated to functional products). Large container vessels, port facilities,
distribution centers, as well as block trains and barges, are employed to minimize
transportation and storage costs per unit exploiting economies of scale. However, supply chain
managers are well aware about the tradeoffs between cost efficiency and sustainability as
increasing transportation volumes lead to the release of large amounts of greenhouse gasses
affecting climate change globally (Mallidis et al. 2012). To this effect, the concept of green
logistics management, defined as the integration of ecological considerations into the design
of logistics networks and operations, has emerged in the C-level executive agenda.
In this work, we propose a green logistics management framework for examining the effect of
both cost and CO2 emissions minimization objectives on tactical inventory planning decisions.
More specifically, a new decision-making model is developed for calculating, in terms of cost
and in terms of CO2 emissions, the optimal order delivery frequencies and base stock
quantities reserved at the nodes of an examined logistics network.
In Section 2 we present a brief literature review, while focusing on green logistics and
inventory management techniques and models. Section 3 describes the problem under study,
118
E. Iakovou et al. / OR56 Keynote Papers, pp.118-133 (2014)
while Section 4 analyses the periodic review decision-making model. Section 5 illustrates the
applicability of the given model through the extensive real-world case study of a white goods
retailer in the market of Greece, using real-world cost and emissions data. Moreover, the
papers results are presented, along with interesting managerial insights. Finally, Section 6
sums-up the findings of the research.
2.
Literature Review
A comprehensive overview of research efforts evaluating the environmental impact of

logistics operations is presented in Dekker et al. (2012). Specifically, with respect to the
inventory planning aspects of logistics, numerous researchers have proposed modelling
methodologies for evaluating their environmental impact. Further, Mallidis et al. (2013)
propose a decision support model for capturing the impact of the evolving pollution regulatory
framework on the design of supply chain networks On this basis, Chen and Monahan (2010),
investigate the effect of a companys production and inventory planning decisions on the
environment. The authors propose a stochastic model in order to determine the optimal
production and inventory planning policies under the constraint of a specific amount of
wastewater generated. This constraint imposes an additional limit on the number of items that
can be produced. Benjaafar et al. (2010) propose a simple inventory planning model that
minimizes fixed and variable ordering, holding and backordering costs per planning horizon
under a constraint on the cap of CO2 emissions produced during the planning horizon. On the
same basis, Hua et al. (2011), propose an inventory planning methodology for managing a
companys CO2 emissions under the carbon emissions trading scheme. According to Hua et al.
(2011), the model of Benjaafar et al. (2010) is too general for any solution algorithm to be
developed. Their proposed model is based on the Economic Order Quantity (EOQ) modelling
methodology taking into account a cap and trade system. The authors determine the optimal
order quantity, while evaluating the impacts of carbon emissions trading, carbon price, and
carbon cap, on order decisions, carbon emissions, and total cost. A multi-echelon extension of
their model was further examined by Bouchery et al. (2012). The authors developed a multiobjective (cost, environmental performance and social impact) formulation of the EOQ model,
thus introducing the sustainable order quantity (SOQ) model. They initially applied the model
in a single-echelon supply chain network and then moved into a multi-echelon extension. For
both models, a set of efficient solutions are analytically identified, through the employment of
a multi-objective optimization methodology that can capture the economic, environmental and
social trade-offs while providing a satisfactory solution.
In this manuscript, a more complex periodic review (R,S) inventory planning model is
proposed, applied in a multi-echelon logistics network, with the transportation frequency to
the retailers and their stock levels being a decision variable.
3.
Problem Under Study
We tackle a logistics network instance frequently encountered by a number of multinationals

that source from Asia to serve the EU market, while employing ports in Northern and S.E.
119
Europe as entry points (EPs). A full discussion and motivation of such logistics networks is
provided by Mallidis et al. (2014).
In the network design instance under consideration, a company considers a single market
served by a number of retail stores. Each retail store receives its orders for a specific product
from one central DC which is in turn served by a manufacturing site (a loading point). These
orders are transported in containers to an entry port, through deep sea shipping, and then by
hinterland transport modes to the central DC. The examined logistics network is exhibited in
Figure 1 for a simplified problem realization with five retail stores.
EP: Entry Port

DC: Distribution Center
RS: Retail Store
RS2
EP
Market
DC
RS1
RS3
RS4
Distant
Loading Point
RS5
Figure 1: Logistics network under study
We assume that the transportation from the DC to each retail store occurs by truck. More
specifically, delivery trucks are employed for short transportation distances within city limits
while heavy duty trucks are utilized for long-haul regional transportation distances.
Additionally, the company is fully accountable for its DC facility and truck emissions, as
trucks and the DC facility are assumed to be leased through long term contracts. Specifically,
with respect to truck transportation this further implies that the company will be also
accountable for the return truck trip CO2 emissions and costs.
Further, the lengthy lead times from the distant manufacturing site to the DC, lead to high
waiting times for an end customer in the case of a stockout at the DC. Thus, the inventory
levels at the DC must be large enough to satisfy practically all orders from its serving retail
stores. This necessitates the operation of a sufficiently large DC that can provide high service
levels (with very low probability of a stockout).
120
The critical decisions for managing the above logistics network involve: (i) the determination
of the stock levels that are to be reserved at the retail stores and, (ii) the determination of the
order delivery frequencies to the retail stores.
The optimization criteria are: (i) the total logistics cost that includes, the holding costs of a
product per time unit at any retail store and the DC, the backorder cost of the product at any
retail store, the transportation costs per truck round trip between the nodes of the network
under study and finally, the DCs operational and leasing costs and, (ii) the total logistics CO2
emissions that include, the transportation emissions per truck roundtrip between the nodes of
the network under study, and the DCs operating CO2 emissions.
Each retail store faces stochastic demand. The analysis focuses on a fast moving stock keeping
unit (SKU), for which the retailer employees a base stock periodic review ordering policy (R,
S) for the replenishment of its stock from the DC.
We argue that the length of the review period has a significant impact on both cost and CO 2
emissions of the retail stores as it may affect the order delivery frequencies, thus the relevant
transportation costs and CO2 emissions, and the inventory-related costs. On this basis, the cost
and CO2 emissions of the retail stores could not be estimated accurately if the review period
was considered fixed. Moreover, as the trucks serving the retail stores are leased, the order
delivery frequencies to each retail store in practise can be quite flexible and depend only on
the stores review period. Based on the above, we treat the review periods of the stores as
independent decision variables.
Regarding the large central DC, its review period is considered fixed and it is commonly
determined by the liner deep sea shipping schedule between the distant manufacturing site and
the entry port.
To this end, since the DC receives orders from its serving retail stores at their review period
epochs, the development of analytical mathematical expressions for modelling these
operations requires that the orders are coordinated. To address this issue we assume that the
allowable values of the review periods of the retail stores are powers of two (Roundy, 1989);
such policies have a well-documented record in improving coordination in multi-echelon
networks.
4.
Model Development
We consider a single market M served by a number of retail stores mM. Each retail store is in
turn served by the central DC, and faces stochastic demand per time unit for a specific product
with a probability density function m () and a cumulative distribution function m (). The
demand of the examined product per time unit at the retail store is denoted by x m , and is
assumed to be an independently normally distributed random variable with a mean m and a
standard deviation m . Each retail store employs a periodic review inventory planning policy
with a review period of Rm time units. This review period further determines the order delivery
frequencies at the retail store (1/ Rm).
121
At the review epochs, the order size is set as order-up-to S m, namely to raise the inventory
position of the product at each retail store to S m. The order arrives after a deterministic lead
time Lm. We further assume that the lead time for the replenishment of the DC is also
deterministic.
Following the standard analysis like the one provided by Silver et al. (1998), the base stock
quantity S m can be expressed as the sum of the mean demand during the protection period,
thus m (Rm+ Lm) plus the safety stock, z m m R m + Lm , where z m represents the safety
stock factor for the examined product at the retail store (further determining its cycle service
level). Thus, the base stock quantity is given by:
S m = m (R m + Lm ) + z m m R m + Lm .
(1)
To determine the expected total cost and CO2 emissions of supplying any retail store, a steady
state analysis has to concentrate on a typical replenishment cycle of length Rm time units,
defined by the times of arrival of two consecutive orders. Specifically, if an order is placed at
time 0 (review epoch), the next regular order will be placed at time epoch R m and it will arrive
at time epoch Rm+Lm. Thus, the replenishment cycle of interest is the time interval starting at
the time epoch Lm and ending at time epoch Rm+Lm.
A typical realization of inventory levels (net stock and inventory position) and orders is
illustrated in Figure 2 for a system with Lm<Rm.
Inventory
Order
Order
Order
Lm
Rm
0
Rm+Lm
t
2Rm
Time
Rm
(Replenishment cycle)
Figure 2: Evolution of the net stock (solid line) and inventory position (dashed line) in a
periodic review system using the base stock (order-up-to S m ) policy
122
The standard expected cost formulae for (R,S) policies found in the literature, charge
backorder quantities once at the end of the review cycle. However, if these formulae are
implemented for the problem under study, where the review period is a decision variable, the
expected backorders of a specific retail store will be charged with different frequencies
depending on the review period. Thus for example, for a planning horizon of one hundred time
units and a review period of ten time units, the backorder cost will be charged ten times during
the planning horizon. If we double the review period, the backorder cost will be charged only
five times during the same planning horizon. Therefore, it would not be possible to compare
the systems performance for different review periods. Thus, in order to compare policies with
different review periods, a model that calculates on-hand and backorder levels continuously
for every time unit is necessary. On this basis, we employ a methodology, initially proposed
by Ray et al. (2010) who compare two periodic review inventory planning models where the
first one charges backorders once at the end of the order cycle, and the second one, every time
unit after the order arrival until the end of the order cycle. We further extend Ray et al.s
model by additionally incorporating transportation and DC parameters.
Consequently, the expected on-hand and backorder quantity of the examined product at any
retail store m at time t is provided by equations (2) and (3) respectively:
Sm
m
m
E(OH m (t)) = 0 (S m xtm ) m
t (xt )dxt
(2)
m
m
E(BOm (t)) = Sm(xtm S m ) m
t (xt )dxt ,
(3)
m
where m
t () and t () are the probability density and the cumulative distribution functions
of the demand of the examined product at retail store m during Lm+ t time units, respectively.
Moreover, S m can now be also expressed as a function of k m (t):
S m = m (Lm + t) + k m (t)m Lm + t
0 t Rm ,
(4)
where k m (t) represents the safety stock factor for the examined product at time t; k m (t) is
then determined given equations (1) and (4) as follows:
k m (t) =
m (R m t)
R m + Lm
+ zm
m Lm + t
Lm + t
0 t Rm .
123
(5)
From equation (5), it is deducted that the safety factor of the examined product at the end of
the order cycle Rm is k m (R m ) = z m .
Following the analysis of Silver et al. (1998, p.721), for systems with normal demand, and by
employing the standard normal random variable um
t =
xm
t m (Lm +t)
m Lm +t
, equations (2) and (3)
can be transformed to equations (6) and (7), respectively:
m
m
m m
E(OH m (t)) = m Lm + t [m
t (k (t)) + k (t) t (k (t))] ,
(6)
m
m
m m
E(BOm (t)) = m Lm + t [m
t (k (t)) k (t) [1 t (k (t))]] .
(7)
The central DC faces the normally distributed collective demand of its serving retail stores.
Thus, the central DCs demand during its replenishment cycle RDC is also normally distributed
as a sum of normal demands and its lead time, LDC is deterministic. Consequently, the
expected time-average on-hand inventory level of the examined product at the DC, and
assuming that the percentage of time that backorders incur is negligible, can be estimated as:
E(OHDC (R DC )) = zDC
2m (R DC+ LDC ) +
R DC m m
,
2
(8)
where zDC represents the safety stock factor for the examined product at the DC.
Finally, in order to determine the DCs operating costs and CO2 emissions, and as these costs
and emissions depend on the DCs size, its capacity needs to be determined. To that effect, the
capacity of the central DC is estimated as: K DC = z DC m 2m (R DC+ LDC ) +
R DC m m consisting of: (i) a quantity reserved for a safety stock that results in a
negligible probability of a stockout occurrence, z DC m 2m (R DC+ LDC ) where zDC = 4,
corresponding to a cycle service level of almost one, and (ii) a capacity large enough to handle
the peak demand of all products from its serving retail stores during a long planning
horizon R DC m m .
Table 1 provides the nonmeclature of the models cost and the CO2 emissions parameters.
124
Table 1: Cost and CO2 emissions parameters

Parameters
Description
Holding cost per time unit.
Backorder cost per time unit.
cDC
Transportation cost per container trip from the distant manufacturing

site (loading point) to the DC. This cost includes the ship transportation
cost to the entry port, the handling cost per container at the entry port
and the hinterland transportation mode cost from the entry port to the
DC.
cm
Truck transportation cost per roundtrip from the DC to the retail store.
C(KDC)
Operational cost of the DC per time unit.
eDC
Truck transportation CO2 emissions per container trip from the loading
point to the DC. These emissions include the ship transportation CO2
emissions to the entry port, and the hinterland transportation mode CO2
emissions from the entry port to the DC.
em
Truck transportation CO2 emissions per roundtrip from the DC to the

retail store.
E(KDC)
DCs CO2 emissions per time unit.
The holding cost per time unit of the examined product is charged on the expected on-hand
inventory level of the product, at any retail store as well as the DC. The backorder cost per
time unit b, is charged on the expected backorder level of the examined product at any retail
store. The transportation cost from the distant loading point to the DC, cDC, is charged per trip,
while the transportation cost from the DC to any retail store, c m is charged per roundtrip since
dedicated leased trucks are employed as discussed earlier. Specifically, for the transportation
segment between the loading point and the DC and since the cargo is transported in containers,
cDC is charged on the total number of container trips required for serving the total shipment to
the DC. The number of container trips needed to transport the total demand of the DC during
its replenishment cycle RDC can be calculated as the ceiling function(R DC m m )FCL,
where FCL represents the full container loading capacity. For the transportation segment
between the DC and the retail stores though, and since containers are deconsolidated at the
DC, the products are transported loose in trucks. Thus, truck roundtrips are considered and
therefore, cm is charged on the total number of truck roundtrips required for serving the total
demand of any retail store during its replenishment period. The number of truck roundtrips is
provided by the ceiling function (R m m )FTL, where FTL denotes the full truck load
125
capacity. Finally, the operational cost of the DC per time unit is assumed as a nonlinear
function of its capacity KDC, denoted by C(KDC).
Additionally, the CO2 emissions parameter, eDC, is charged on the total number of container
trips required for serving the total demand of the DC during its replenishment period, while em,
on the total number of truck roundtrips required for serving the total demand of any store
during its replenishment period from the DC.
Finally, the CO2 emissions related to the operation of the DC per time unit, E(KDC), are further
assumed as a nonlinear function of its capacity.
Consequently, the expected total cost per time unit, E(TC) and expected CO2 emissions E(TE)
per time unit for the examined logistics network are provided by equations (9) and (10),
respectively:

( ()) + ( ())] +

( ) +
( ) +
min () = [
( ( )) + ( ) ,
min () = [
(9)
( ) +
( ) ] .
( ) +

(10)
As our intent is to capture the cost of implementing a CO2 emissions-related optimal solution,
in our mathematical analysis, instead of focusing on obtaining a solution that simultaneously
optimizes the multiple objectives, we examine the two objective functions separately. Through
this approach, a decision-maker would be able to determine accurately the cost of greening
the supply chain before adopting a sustainability-based redesign of his logistics network.
Since the products safety factor at the central DC zDC and its review period R DC are
predefined, the commonly considered decision variables zDC for the examined product and
RDC are now used in the model as input parameters. Further, as the relevant segment of the
objective functions (9) and (10) are constant, and in order to obtain the optimal values of the
decision variables related to the retail store, only the optimization of the first three (out of six)
terms of equation (9) and of the first term of (10) is necessary. This observation leads to
Proposition I.
Proposition I:
For a given Rm, the optimal value of z m = k m (R m ) for the examined product at any retail
store m is obtained from:
126
Rm
tm (k m (t))dt =
bR m
.
h+b
(11)
Proof:
We denote as A(z m ) the expected holding and backorder cost per time unit for the examined
product at a specific retail store,
1
A(z m ) = h R 0 m E(OH m (t))dt + b R 0 m E(BOm (t))dt. By differentiating A(z m ) with

m
respect to z m and using:

k m (t) =
m (R m t)
R m + Lm
+ zm
, we obtain:
m Lm + t
Lm + t
dA(z m )
1 Rm dE(OH m (t))
1 Rm dE(BOm (t))
=
h
dt
+
b
(
(
)
) dt =
dz m
Rm 0
dz m
Rm 0
dz m
h
m
m
1 Rm
dm
dk m (t) m m
t (k (t)) dk (t)
m Lm + t (
+
t (k (t))
Rm 0
dk m (t)
dz m
dz m
dtm (k m (t)) dk m (t)
+ k m (t)
) dt
dk m (t)
dz m
m
m
1 Rm
dm
dk m (t)
t (k (t)) dk (t)
m Lm + t (
Rm 0
dk m (t)
dz m
dz m
dk m (t) m m
m
+
t (k (t)) + k (t)
) dt .
dz m
dk m (t)
dz m
+ b
(12)
Since:
m
m
dm
t (k (t)) dk (t)
+ k m (t)
= 0,
m
m
dk (t)
dz
dk m (t)
dz m
equation (12) can be further simplified as follows:

dA(z m )
1 Rm
R m + Lm m m
=
h
m Lm + t (
t (k (t))) dt
m
dz
Rm 0
Lm + t
+ b
=
1 Rm
R m + Lm R m + Lm m m
m Lm + t (
+
t (k (t))) dt
Rm 0
Lm + t
Lm + t
Rm
Rm
1
m Lm + R m [h tm (k m (t))dt + b [tm (k m (t)) 1]dt].
Rm
0
0
Then, by applying the first order conditions on (9) with respect to z m , one obtains a set of equations that
127
can be solved independently for each retail store m:

Rm
Rm
1
m Lm + R m [h tm (k m (t))dt + b [tm (k m (t)) 1]dt] = 0
Rm
0
0
Rm
Rm
Rm
tm (k m (t))dt + b
tm (k m (t))dt + b
Rm
(h + b)
Rm
[tm (k m (t)) 1]dt = 0

tm (k m (t))dt b
Rm
1dt = 0
tm (k m (t))dt = bR m
Rm
tm (k m (t))dt =
bR m
(A) . m
h+b
Moreover, and since the second derivative of (A) with respect to z m :

Rm
(h + b)
Rm
dk m (t)
m
m (t))
(h
]
dt
=
+
b)
] dt > 0 ,
(k
t
dk m (t)
dz m
dz m
0
an optimal z m value exists that satisfies the first order conditions for a specified value of R m .
By enumerating powers of two values of R m (1,2,4,8.) and for each one of these values, the
optimum value of z m for the examined product can then be determined from equation (6).
The solution methodology proceeds as follows. For each combination of (R m , z m ) decision
variable values, the retail stores total logistics cost is calculated. The optimum (R m , z m ),
combination is then determined as the one that leads to the minimum retail stores costs. For
CO2 emissions minimization E(TE), the safety factor z m does not appear in the total emissions
objective function. This factor is however required in order to be able to determine the on
hand inventory levels and backorders of the product and therefore to further estimate the cost
of implementing a CO2 emissions optimum policy as discussed earlier. Thus, in order to obtain
the optimum z m in terms of CO2 emissions we employ equation (11) and the optimum R m
decision variable value that minimizes the stores transportation CO2 emissions. This R m
value is determined through enumeration, as in the cost optimization process, and corresponds
to the R m value that minimizes transportation CO2 emissions.
5.
A Real-World Illustrative Case Study
In this Section we present the application of the developed methodology on a series of realworld supply chain network realizations of a specific white goods retailer that we have worked
with. Specifically, we consider the three alternative logistics network design options of the
retailer who distributes his/her products throughout Greece.
The first logistics network consists of one entry port (P0) that of Piraeus, one central DC (DC),
located in the wider area of Aspropirgos and served through the Piraeus entry port, and one
large region to serve. The region includes two privately owned retail stores in the wider
metropolitan Athens area (Marousi-RM1, Argiroupoli-RM2), two in the wider Peloponnese
128
area (Tripoli-RM3, Patra-RM4), and one, in each of the wider Metropolitan areas of IoanninaRM5, Larissa-RM6, Thessaloniki-RM7, and Ksanthi-RM8. Transportation from the port of
Piraeus to the central DC occurs by heavy duty trucks with a carrying capacity of one 40ft
container, while from the DC to retail stores (1-2) through delivery trucks with a carrying
capacity of 33 m3, and from the DC to the far away retail stores (3-8) through heavy duty
trucks with a carrying capacity of 66 m3.
The review period of the central DC is dictated by the deep sea shipping schedule from
Shanghai and is equal to 32 days. We examine the supply of three fast moving Type A
products p (refrigerators, washing machines, and ovens) that accounts for approximately 80%
of the companys total sales. Based on historical data for these products we verified that their
daily demand fits the normal distribution with a coefficient of variation (cv) of about 0.3.
Figure 2 illustrates the nodes of the network under study (Mallidis et al. 2014).
RM8
RM7
RM6
RM5
RM1
RM4
RM2
Po
RM3
150 km
Figure 2: Nodes of the network under study
129
5.1 Numerical Results and Managerial Insights

Two instances of the model were solved by combining the examined logistics network design
option with the two optimization criteria, cost and CO2 emissions. Table 2 depicts the
p
optimum order delivery frequencies and the cycle service levels for each product (m , where
p
m =
1 p
m Rm
z ) of
p
m
the retail stores, under the cost and CO2 emissions optimization
criterions respectively, while Tables 3 and 4 the options cost and CO2 emissions break-downs
respectively.
Table 2 Optimal service levels (Type I) and order delivery frequencies (1/Rm) under the CO2
emissions minimization objective for refrigerators
Criterion
Cost
CO2
1m
1/ Rm
1m
1/ Rm
RM1
80%
1/2
61%
1/16
Retail StoresRM2
75%
1/4
61%
1/16
RM3
75%
1/4
51%
1/32
RM4
81%
1/2
51%
1/32
RM5
69%
1/8
51%
1/32
RM6
76%
1/4
51%
1/32
RM7
86%
1/1
52%
1/32
RM8
70%
1/8
52%
1/32
Nodes (m, SDC)
Table 3. Cost breakdown (/day)

Holding Cost
Backorder
Cost
Transportation
Cost
RMs
DC
RMs
DC
RMs
DC
DC
Cost
800
6,546
99
9,247
27,887
1,507
46,086
CO2
4,944
6,546
352
8,579
27,887
1,507
49.815
Opt.
Crit.
130
DC Operational
Cost
Total
Cost
Table 4. CO2 emissions breakdown (tons/day)

Transportation CO2
Opt. Crit.
DC
RMs
DC CO2
Total CO2
Cost
12.8
4.5
0.21
17.51
CO2
12.8
4.2
0.21
17.21
An interesting finding stemming from the numerical analysis is that the CO2 emissions optimal
solutions involve lower order delivery frequencies compared to their respective cost optimal
solutions, thus resulting in lower truck transportation CO2 emissions. With respect to costs, the
lower order delivery frequencies, observed under the CO2 emissions optimum options, reduce
on one hand transportation costs, but on the other hand increase holding and backorder costs.
Thus, order deliveries are more frequent under the cost minimization objective. Moreover, and
since lower order delivery frequencies increase the probability of a stockout occurrence at the
retail stores, the service levels of the products are generally lower under the CO2 emissions
minimization objective. Another interesting outcome of the analysis involves the much higher
cost that the company would pay if it decides to go green, that is by migrating from the cost
optimum solution to the CO2 emissions optimum solution. In this case its emissions would be
reduced from 17.51 tons to 17.21 tons/day, while its cost would increase from 46,086 to
49,815. This in turn, results in a marginal cost per ton of CO2 emissions reduced equal to
(49,815-46,086)/(17.51-17.21)=12,430/ton CO2 (Mallidis et al. 2014).
6.
Summary
This paper proposes a modelling methodology for evaluating the cost and CO2 emissions
trade-offs for implementing a periodic review inventory planning policy in the nodes of a
multi echelon logistics network. The market of Greece was used as a test case for applying the
developed methodology through an extensive real-world case study of a white goods retailer.
The results are intuitively sound and further indicate that the inclusion of CO2 emissions
minimization objectives in periodic review inventory planning decisions results in lower order
delivery frequencies compared to those prescribed by cost minimization, thus leading to
reduced transportation costs and CO2 emissions and increased holding and backorder costs.
Finally, and as lower delivery frequencies increase the probability of stockout occurrences at
the retail stores, the cycle service levels (Type I) of the products stored at the retail stores are
reduced under the CO2 emissions optimization criterion.
131
Acknowledgements
The authors acknowledge the valuable contribution of the companies Orphee Veinoglou, and
Vassilios Kampakis, S.A. that provided detailed and comprehensive data for the case study.
This research paper has been funded by the EU (European Social Fund ESF) and Greek
National Funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework (NSRF) - Research Funding Program: Heracleitus II.
Investing in knowledge society through the European Social Fund. It has also been conducted
in the context of the GREEN-AgriChains project that is funded from the European
Communitys Seventh Framework Programme (FP7-REGPOT-2012-2013-1) under Grant
Agreement No. 316167. All the above reflect only the authors views; The European Union is
not liable for any use that may be made of the information contained herein.
References
Benjaafar S, Li Y and Daskin M (2010). Carbon footprint and the management of supply
chains: Insights from simple models. Automation Science and Engineering, IEEE
Transactions 10(1): 72-90.
Bouchery Y, Ghaffari A, Jemai Z and Dallery Y (2012). Including sustainability criteria into
inventory models. European Journal of Operational Research 222: 229-240.
Chen C and Monahan GE (2010). Environmental safety stock: The impacts of regulatory and
voluntary control policies on production planning, inventory control, and environmental
performance. European Journal of Operational Research 207(3): 1280-1292.
Dekker R, Bloemhof J and Mallidis I (2012). Operations Research for Green Logistics-An
overview of aspects issues and contributions. European Journal of Operational Research
219: 671-679.
Hua TCE, Guowei TCE, Cheng TCE and Wang S (2011). Managing Carbon Footprints in
Inventory Control. International Journal of Production Economics 132 (2): 178-185.
Mallidis I, Dekker R, Vlachos D and Iakovou E (2014). Cost and CO2 emissions Trade-offs in
Continental Supply Chains under Periodic Review Replenishment Policies.
Transportation Research Part E: Under Review.
Mallidis I, Vlachos D and Iakovou E (2013). A Decision Support Model for Capturing the
Impact of Energy Savings and Pollution Legislation on Supply Chain Network Design.
Chemical Engineering Transactions, 35, 325-330, DOI:10.3303/CET1335054.
Mallidis I, Dekker R and Vlachos D (2012). The impact of Greening on Supply chain Network
Design and Cost: A Case for a Developing Region. Journal of Transport Geography 22:
118-128.
132
Ray S, Song Y and Verma M (2010). Comparison of two Periodic Review Models for a
Stochastic and Price-Sensitive Demand Environment. International Journal of
Production Economics 195: 12-28.
Roundy R (1989). Rounding off to powers of two in continuous relaxations of capacitated lot
sizing problems. Management Science 35(12): 1433-1442.
Silver AP, Pyke DF and Peterson R (1998). Inventory Management and Production Planning
and Scheduling (3rd ed.). USA: Willey.
133

Third Sector OR
Whats so Special about the Third Sector?

Ruth Kaufman
Chair of Third Sector Special Interest Group, OR Society, 12 Edward St, Birmingam B1
ruth.kaufman@btinternet.com
Abstract
The OR Societys Third Sector initiative has two main components: a Special Interest Group,
and a Pro Bono scheme, matching O.R. volunteers with third sector organisations needing
their input. But is there really any difference between doing OR in the third sector and doing it
with a government or private sector organisation? Is its for charity really a good enough
reason to work for free? This talk explores these challenges, taking charities as an example of
third sector organisations. It considers three areas of inherent difference between charity,
private and public organisations legal form, governance, and resourcing and other factors
such as organisational size, culture, and business environment. It goes on to consider the
implications for practising OR in three broad areas: strategy, efficiency/effectiveness, and
profitability. Finally, it explores the rationale for volunteering.
Keywords: Third Sector; Practice of O.R; Professional; Strategy
1.
Introduction
The OR Societys Third Sector initiative has two main components: a Special Interest Group;
and a Pro Bono scheme, where OR people offer to do free OR work for a third sector
organisation.
When we established the initiative, the emphasis was on the size of the third sector, the lack of
profile of OR in the sector, and the desire to promote and exploit OR more widely. We did not
explore the rationale for a separate third sector group is there really any difference between
doing OR with a charity and doing it with a government or private sector organisation. Nor did
we explore the rationale for volunteering. This talk begins to address the gap. It examines the
structural differences between sectors, and goes on to look at what this means for O.R. in
practice, and more briefly at the end, what it means for volunteering.
For the most part, this talk will look at charities, and just touch on other third sector forms of
organisation social enterprise, mutual, community amateur sports club, political campaign,
and more at the end. The talk itself will use examples taken from a variety of third sector
practice; the examples are not included in this abstract.
What follows is based largely on direct personal experience, and learning from others who
have been working in the field. It is not intended as a definitive statement of what is, based
on systematic evidence-gathering, but as a trigger for debate and reflection.
134
R.Kaufman / OR56 Keynote Papers, pp.134-139 (2014)
2.
Differences between charity and private or public sector organisations
I will pick out three features that distinguish charities structurally from other organisations:
purpose, governance and resourcing.
2.1 Purpose
Every charity must have charitable purposes for the public benefit, and these must be its only
purposes. This is the single defining characteristic. It is fairly clear how this differentiates
charities from the private sector although many private sector organisations include some
version of public benefit amongst their aims. It is less clear how it differentiates them from
the public sector. Many of the states activities are in fact delivered by charities: around 35%
of charity income comes from statutory sources.
2.2.Governance
In private sector and charities, unlike the public sector, ultimate responsibility lies with the
Board of Trustees. All Boards have to comply with a range of legal constraints, including the
requirement to operate within the rules set by the charitys governing documents. Some may
also be part of a larger organisation that sets constraints (for example, all Womens Institutes
are independent charities and also part of the National Federation of Womens Institutes), or
be accountable to members, or funders where contracts or grants are conditional. But private
sector and charity Boards alike have no democratic mandate, no statutory duty, and no
accountability to citizens at large. Unlike the private sector, though, charity Board members
are (usually) unpaid, and may be elected by members rather than appointed.
2.3 Resourcing
Charity income may come from all the same sort of sources as private and public sector:
revenue for services sold, grants made from taxation, investment interest, loans. But around
one-third of charities cash income is donated from members of the public, corporates,
charitable trusts and foundations or lottery funding. Whats more, charities are massively
dependent on donations in kind, and in particular, volunteers (including Trustees).
Donations and volunteers are not exclusive to the third sector. There are many volunteers
working with the statutory sector, and even the private sector. some in legally designated
volunteer roles, such as magistrate, prison visitor, school governor, others informally offering
their support reading to children, running hospital tea-rooms, or litter-picking. The recent cuts
to the welfare state have blurred the lines further.
So, these characteristics do not rigidly separate charities from other organisations.
Nonetheless, they have significant consequences for how charities behave, and follow-on
consequences for O.R. in charities.
135
As well as these structural factors, its important to be aware of characteristics of the sector
such as organisation size, employee numbers, and activity areas6. But the diversity of the
sector means that each charity has to be considered independently generalisations are
dangerous.
3.
Is OR for charities different from OR for other organisations?
To begin to answer this question, this talk considers three areas of application: strategy,
efficiency/effectiveness, and profitability.
3.1 Strategy
Although the underlying process of strategy development and implementation is no different
for a charity than for any other enterprise, some important issues arise from the nature of the
sector.
Articulating goals can be unexpectedly problematic. Charities are very often set up to
undertake a particular activity which seems to be a Good Thing: for example, providing legal
support to people with disabilities. But one activity can go with many goals, and different
goals may demand different priorities, or delivery models.
Stakeholders may include staff, volunteers, trustees, funders, beneficiaries, members, and
more. Many of these will feel a deep loyalty and commitment to the cause. What differentiates
them from stakeholders in other sectors is that for many of them, that commitment is their only
tie to the organisation. They will walk away if changes violate their view of what the
organisation should be doing, or how it should be doing it. Strategy must take account of these
relationship drivers and risks.
Any charity that is dependent on the UK statutory sector for any of its income or activities, or
that has UK beneficiaries that are in any way disadvantaged, is going through tough times at
the moment. Financially, certainly; also because of greater numbers and needs of
beneficiaries; and because of changes in statutory sector procurement, growth of competition
both between and within sectors, reduction in support from the statutory sector, transfer of risk
from statutory sector to its service deliverers, and pressure towards partnership working with
no recognition of the associated cost.
Charities that do not have an IT-based operating model are similar to most other organisations
that do not have such a model data is valuable, and they may have lots of it, but there are
problems in using it. The only difference with charities is that it is possible for a charity to be
set up and run by an entire team of people with no expertise in business or data management,
and survive for a long time on vision and inspiration alone. Many are highly professionally
Important sources of information are the annually-updated UK Civil Society Almanac (NCVO,
2014) and the Charity Commission http://www.charitycommission.gov.uk/about-charities/sector-factsand-figures/
136
run, of course but there is probably more scope for helping with the basic components of
systematically analysing data, planning, and forecasting.
3.2 Efficiency/effectiveness
Dan Corry, Chief Executive of charity New Philanthropy Capital, has recently identified a
number of barriers that, in his view, militate against charities improving productivity7. Some
of his arguments are contentious, and will be reviewed in the talk. Nonetheless, it is true that
from the O.R. side we have seen relatively little drive for specific efficiency or effectiveness
improvement initiatives. It has been notable within the Pro Bono scheme, for example, that
only a small proportion of project requests have been for improvements in operational
efficiency.
Instead, over the last few years, across the whole sector, there has been growing demand for
evaluation, outcomes measurement, and impact measurement, and this has been reflected in
Pro Bono requests. Indeed Dan Corry prescribes this as a solution to the problems identified.
The assumption is that for the charity, such measurement will help the organisation recognise
what it needs to do to improve; and that for funders, having this information enables them to
channel funds to the more efficient/effective organisations. Significantly, both these
assumptions skip over the step in the process which involves designing the efficiency
improvements. This is the very point where O.R. should have the most distinctive contribution
to make; and it is almost invisible.
Whatever the barriers and the means for their removal, what actually is the potential for
efficiency improvement? Almost all charities are part of the service sector, and have processes
that can be improved, priorities that can be adjusted, resources that can be better matched to
need, just like any other service industry. Scope may be limited: many will have services
which are dependent on one-to-one or one-to-few transactions between staff and beneficiaries,
transactions which cannot be shortened because the nature of the human interaction is key;
many are small organisations operating on a shoe-string already. But that still leaves plenty of
organisations with scope for improving call-centre operations, improving logistics, improving
appointment booking and client follow-up systems, reducing transport costs, and more.
3.3 Profitability
Profitability or, in charitable terms, breaking even or making a surplus, is an area where
charities really are different. There are many possible sources of income which can exist
independently of the service model of the charity. Taking another expression from Dan
Corrys speech, the feedback loops between activity and funding are weak.
One consequence is that fundraising is a significant activity. Fundraisers have become soughtafter professionals. Fundraising is an organisational function, and service process, that doesnt
exist in other sectors, and that can be supported by O.R.
7
Speech to RSA, 8 May 2014: http://www.thersa.org/__data/assets/pdf_file/0018/1543302/Dan-Corry-how-dowe-drive-productivity-and-innovation-in-the-charity-sector-RSA.pdf (accessed 19/6/2014)
137
Another is that financial forecasting can be crucial. Many charities are dependent on a small
number of chunks of funding, much of it one-off, and which may or may not relate to chunks
of cost. This is made more complex by charity accounting rules, which are different from
private and public sector, and do not aid transparency.
A third is that where income is related to activity levels for example where charities are
providing services to local authorities under contract, or have grants to undertake specific
activities, or are running a business costing or pricing may not be straightforward, and may
benefit from modelling support.
Finally, in chasing profitability, charities may follow the money. This is a sensible thing to
do for private sector organisations, but dangerous for charities. There can be a temptation
towards mission drift, away from the objects for which they were founded towards those
which are currently in vogue with funders or donors potentially a strategic disaster.
4.
Where do Values Fit in?
It might be assumed that the key difference when working with a charity is the drive to do
good, and the charitable values, that underpin the organisation and provide meaning to all its
activities.
I have already suggested that public benefit does not reliably distinguish charities from
public or even, in some cases, private sector. People working with any organisation will tend
to seek out meaning in their work. This may be easier to do in a charity, and certainly charities
are more likely to be staffed by people who want meaning, and very unlikely to be staffed by
people looking to make piles of money; but there is a lot of overlap.
Values are even less of a discriminator. Whilst most charities will have strong values relating
to the ends - their direct beneficiaries, or achieving their charitable objects - these values do
not necessarily stretch to the means: the way they treat staff or volunteers, the way they work
with partner agencies, the way they manage their business. In practice, the values guiding
charities day-to-day management are as widely variable as they are in any other sector.
The charitys objects may not make much difference directly to the nature of O.R. work, but
they are absolutely crucial when it comes to the decision to work, and even more to volunteer,
with a charity. If you want to help bring about the same change in the world as the charity
does, then it is exciting to have the opportunity to work with them. Whether you give your
time for free is determined only by whether you can afford to, and how you prioritise this
activity against all the others that you might do.
For Pro Bono O.R. volunteers, there are lots of potential benefits: interest, experience,
contacts, learning, job satisfaction, building self-confidence, helping to market your
profession. For some people, some of the time, these will be enough to justify an unpaid
138
assignment. Nonetheless, it is important to understand the charitys objects8 and how it goes
about achieving them. Charitable purposes cover a wide range. When you volunteer, you are
helping that charity; and you need to assure yourself that what you are helping it to do accords
with your own values and your own vision for a better world. Volunteering with a charity that
does just that can be enormously satisfying.
There is a particular sort of opportunity that charities offer, which is relatively unusual: the
opportunity to become a trustee. The O.R. person as trustee, especially of a small charity, can
become both client and consultant. If well-managed, this resolves the perennial problem of
small organisations in all sectors lack of capacity to manage the consultancy they so badly
need
5.
The Wider Third Sector
The third sector is usually defined by what it is not: neither private sector nor public sector. It
covers a surprising range of organisations, but three distinguishing qualities can be identified:
they are operationally independent of government;
they are value-driven motivated to achieve social aims; and
they reinvest surpluses in pursuit of these social aims.9
What makes such organisations special whether in terms of the nature of the O.R. challenge,
or the reason for wanting to work with them is a mix of the characteristics of charities and
other sectors. This mix will have different components and different weightings, depending
both on the legal form and on the specific organisation itself.
6.
Conclusions
The question was posed at the start: whats so special about the third sector. For the O.R.
analyst as a rational detached professional, the answer lies not in the values, but in the
nature of purpose, governance and resourcing. These are sufficient to generate a range of
characteristics which shape the practice and content of Third Sector O.R. But for the O.R.
analyst as a person with wants, ambitions, and/or a sense of social purpose, what is special is
the richness of the opportunity to satisfy some or all of these.
For those not familiar with charity jargon, the term objects has a similar meaning to objectives,
but carries legal weight. A charitys objects describe what it has been legally set up to achieve; it may
not do anything unrelated to its objects, and can only change them with permission from the regulator
(in the UK, at least).
9
Improvement
Skills
Consulting
for
the
Third
Sector
http://ianjseath.files.wordpress.com/2013/12/isc-third-sector.pdf accessed 19/6/2014.
8
139

Keynote Paper Booklet (2) SH 22072014105003

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Keynote Paper Booklet (2) SH 22072014105003

Uploaded by

Copyright:

Available Formats

OR56 Annual Conference - Keynote Papers

Royal Holloway University of London, 9 11 September 2014

OR56 Annual Conference - Keynote Papers

Data Envelopment Analysis

OR56 Annual Conference - Keynote Papers

OR56 Annual Conference - Keynote Papers

Setting Handicaps to Industrial Sectors in DEA: An Application to

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

xo xok so* , y o y ok so* .

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Figure 1: Sectorial frontiers and projections

2.2.Global evaluation of the projected DMUs

We merge the set X , Y (k 1, , K ) and denote it by X , Y

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

With-handicap group (C)

Figure 2: Meta-frontiers and handicaps

. We judge this sector k has no handicap and classify k to the nok

3.1.Sectorial inputs and outputs

Output from Sector k : y l jk1 y lj (l 1,

Input/output vectors of sector k are defined by

We deal with K DMUs defined by x , y

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Figure 3: S-shaped frontier

3.3.Solving with the non-convex model

We solve the dataset x , y

, K ) with cluster name using the above non-convex

hk SAS score , if the sector belongs to the with-handicap group. (7)

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

4.1.Input (Output) under handicap

Sectorial output lk j 1 yljk

Further we define input/output vectors for each sector as follows:

See Figure 4 for an example.

4.2.Solving the CRS model

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Let an optimal solution to (10) be

, so* . Then we have the projection to the SAS

efficient frontiers as follows:

5. Application to Ethiopian industry

5.1.On Ethiopian industry

5.1.2. The data

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Table 3: Summary statistics of DMUs in their respective industry (non-handicapped data)

5.3.Evaluation of DMU within each sector and projection

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

5.4.Global evaluation of the projected DMUs

Table 5: No-handicap and With-handicap groups before non-convex adjustment

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Table 6: Aggregates of inputs (output) of VRS-projected DMUs in the sector

5.5.2. Solving non-convex model

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

5.7. Global issue

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Table 9: Handicapped Inputs and output

5.7.2. Solving the CRS model

K. Tone and K. Berhe Hailu / OR56 Keynote Papers, pp.1-18 (2014)

Table 10: Efficiency score with handicap model

Tanning, leather and footwear

Paper and printing

Rubber and plastics

Basic iron and steel

Machinery and equipment