You are on page 1of 14

2

3
Locating Centres of Information and Recruitment of volunteers
for the Portuguese Armed Forces:
A decision-analysis case-study

Carlos A. Bana e Costa1,2


João P.C.F.Thomaz2,3

1
Department of Operational Research, London School of Economics
2
Instituto Superior Técnico (CEG-IST), Technical University of Lisbon
3
Portuguese Army & Autonomous Polytechnic Institute (IPA-Lisbon)

The end of conscript military service in Portugal and the option for a volunteer based
military service require a different personnel recruitment and selection system for the
Portuguese Armed Forces. In order to attract as much candidates as possible, it was
decided to implement Centres of Information and Recruitment (CIR) in the Portuguese
territory. This paper describes the decision analysis intervention that helped locate ten
CIR. Developed in a Decision Conferencing framework, the study made use of several
interactive Decision Support Systems, namely Decision Explorer to support the group
cognitive mapping model-structuring process of identifying key location aspects, and
MACBETH (Measuring Attractiveness by a Categorical Based Evaluation Technique) to
develop a simple-additive value model for evaluating CIR location options.

Keywords: Multicriteria Analysis, MACBETH, Decision Conferencing, Location problem

LSE OR Working Paper 01.38 ISBN 0-7530-1486-6

4
1. Problem structuring
CIR mission
The decision problem addressed in this paper is the choice of adequate locations in Portugal
for Centres of Information and Recruitment (CIR) of volunteers for the Portuguese Armed
Forces, in view of achieving the objectives of the CIR strategic mission:
- to support strategic and operational marketing actions to assure a large base of recruitment
in the target-population – 16 to 21 years old (soldiers) and 18 to 26 years old (officers and
sergeants);
- to attract the maximum number of potential candidates to enlist in the Armed Forces;
- to divulge general information about new job offers in the Armed Forces;
- to provide information in a simple, direct and personalised way;
- to process information about candidates and select those whose skills and attributes best fit
the profiles of jobs available in the Armed Forces.

Actors and boundaries of analysis


A work group (WG) of eight experts in Human Resources was created for the specific
purpose of studying the problem and to elaborate recommendations of choice of CIR
locations to the decision-maker – the Portuguese Minister of Defence (MOD). We acted as
decision-analysts and facilitators of the WG, which included a Psychologist and a Sociologist
and a representative of each of the following six organisations: Ministry of Defence, Navy
General Staff, Army General Staff, Air Force General Staff, Army Public Relations Cabinet
and Army Personnel Command. In the first meeting, the following boundaries of analysis
were agreed:
- given the budget previously allocated by the MOD to install CIR, and based on a
preliminary analysis, ten CIR should be located in the Portuguese continental territory,
other two in the Azores Islands and another one in Madeira Islands;
- the scope of the study was limited to the Portuguese continental territory (hereafter simply
referred as “territory”), because the specific characteristics of the islands require a separate
analysis;
- the network of the ten CIR should as much as possible cover the whole territory;
- given that the Portuguese Armed Forces need about 4000 volunteers per year and in
average six candidates are required to enlist one volunteer, 24000 candidates must be
attracted each year; this fact gave rise to the decision of taking medium sized regions as
alternative location options;
- for reasons of information availability, the existing twenty eight territorial units (NUTs III)
in which the territory is divided for statistical purposes (see Fig. 7) were defined as the
“CIR locations options”; ten of them should be selected as a result of the study; in a
subsequent deeper analysis, a site would be chosen within each NUT;
- if chosen, NUT 20 (“Grande Lisboa”) will receive two CIR, because of the dimension of
its target-population when compared with the estimated service capacity of one CIR (about
5000 men/year).

Objectives of the study


The three main objectives of the study were:
- to identify the key location concerns to be taken as evaluation criteria;
- to develop a transparent and consistent model to evaluate the CIR location options (the 28
NUTs) – with regard to each criterion individually and globally, i.e. taking all the criteria
together – from which should result a well-justified selection of ten locations covering as
much as possible the whole territory;
- to contribute to divulge Decision Analysis in the Portuguese Armed Forces.

5
Definition of the problem: discussing the decision-analysis problematic
A fundamental starting point was to achieve a shared understanding of “what is the
problem?” from the decision-maker’s perspective (the “decision-making problematic” – cf.
Bana e Costa, 1996), in order to find a wise answer to the question “how shall the study
approach the problem?” (i.e. which “technical or decision-analysis problematic” should be
adopted?”). This was essential to develop in the WG a sense of common purpose. Let A be the
set of all 28 NUTs. One possible approach would be to compare all the portfolios of ten
NUTs (nine, when Lisbon is included). The technical problem would then be to choose the
best portfolio. However, the search for the “best” would require the (difficult) definition of
criteria to compare portfolios (not individual NUTs) one with the other. More important, this
formulation would not fit the objective of ranking the location options in terms of their
relative attractiveness to receive a CIR. Clearly, the NUTs should be compared one with each
other and therefore the portfolio comparison formulation was abandoned. But, should a
conventional ranking procedure be adopted? In fact, no, because locating a CIR in a NUT
may also benefit (serve) adjacent areas within other NUT(s). How then should this
dependence phenomenon be dealt with? Actually, a procedure of “successive choice” (Bana e
Costa, 1996) was adopted, consisting in ranking the NUTs (indirectly) throughout a sequence
of choices of the “best” location (see Fig. 1): a NUT selected at a choice stage was taken out
of the choice-set for the next stage and, before going ahead, the problem was re-framed to
incorporate in the model the repercussions of the previous choice over the impacts
(performances) of the remaining NUTs.
A A \ A1 A \ A1\ A2

...

A2 A3
A1
st rd
1 choice 2nd choice 3 choice
Figure 1. Ranking by a successive choice procedure

Overview of the methodology adopted


To guide the work process of analysing and evaluating the NUTs, we proposed the adoption
of a Multi-Criteria Decision Aid (MCDA) methodology. It can be described as a package of
activities grouped in three main phases of analysis (cf. Bana e Costa, 1992): Structuring,
Evaluation and Elaboration of Recommendations. In the CIR case-study:
- the Structuring Phase included the activities related with problem-structuring (i.e., the
definition of the problem, as described above), model-structuring (i.e., the definition of the
basic elements of the model) and impact assessment (i.e., the qualitative analysis of the
options);
- the Evaluation Phase included the construction of a quantitative additive value model (cf.
von Winterfeldt and Edwards, 1986, Belton, 1999) with the support of the MACBETH
Approach (cf. Bana e Costa and Vansnick, 1999, 2000), options-evaluation (i.e. to use the
model to compare options or / and analyse their intrinsic attractiveness or
unattractiveness), and sensitivity and robustness analyses (essential to guaranty that the
model built is a “requisite model” – cf. Phillips, 1982 and 1984a);
- Elaboration of Recommendations develops all along the decision-aid process and consists
in the substantive interpretation of the qualitative and / or the quantitative evaluations in
terms of guiding the behaviour of the group in the decision process.

6
All together, these activities constituted a cycling and learning group-process (rather than a
sequence of steps) developed in a Decision Conferencing framework (cf. Phillips, 1984b and
1990, Watson and Buede, 1987). The process evolved during a series of intensive “face-to-
face” facilitated workshops and decision conferences (intermediate with “off-line” data
gathering and processing). The initial meeting with the WG finished with an outline of the
proposed methodology and a discussion on its component activities.

2. Model structuring
Identifying the key concerns to be taken as CIR location criteria
The first workshop devoted to model-structuring comprised several working sessions of a few
hours each. We started by inviting each participant individually to write on “post-its” short
descriptions of aspects he thought to be relevant for CIR location. Then, the “post-its” were
all stamped on a white-board and a discussion was launched to create sub-sets of similar or
related aspects. While the experts spoke, we drew a map in the form of a “means-ends
network-like structure” (Belton et al, 1997), drawing the links that were emerging between
identified aspects. This process was quite helpful; when the group began to understand the
logic of the cognitive mapping technique, it became obvious that the map was helping them to
identify new aspects and relationships. Aspects were re-described, decomposed, linked and /
or grouped, to avoid ambiguity and eliminate redundancies. Essentially, cognitive mapping
(cf. Eden, 1988) was used here as a semi-structured device to generate ideas, facilitated by
questions like “why, or for what, do you think this aspect matters?” The final map (see Fig. 2)
includes forty two concerns (“concerns” in the sense that the WG understood their value
meaning and relevance for comparing location options).
1 Install a CIR in a
2 Install a CIR in a 16 Enlist sufficient
location with low
local with low HR for the Armed
4 Widen the target-population
density of Forces
Recruitment ... adequate
population ... high 19 Decrease the
Structure ... keep location
unemployment rate
the actual structure

- -
5 Increase Military
visibility ... keep 20 Level of
the actual image development
37 Increase the 17 Increase of work (economics, etc)
number of applicants posts
for the Armed Forces
28 Number of railway
stations in the area
18 Diversity of
Employment

27 Number of bus
stations in the area

25 Facility of
39 Transport flow by transports 3 Locate 10 CIR's in 23 Enlist better 24 Better Quality of
day the territory ... HR in the Armed
quality personnel
install them in any Forces
location
26 Number of roads,
highways etc in the
area 36 Proportion
offer/demand in the
area (historic)

6 To have more
candidates 22 Average base
42 Lower costs for 13 Greater density salary in the area
installation and 45 Range of action 31 Greater density of target-population lower than those of
maintenance of a CIR <> 20 Km of population (16 to 25 years-old) a military under
29 Time to get to a contract
CIR less than 30
minutes
40 Level of
unemployment in the
area
14 Population with 21 Existence of
43 Availability of academic Secondary Schools in
Military facilities 44 Military qualifications of at the area
in the area ... need personnel living in least O or A'levels
10 Existence of a of renting 11 Greater rates of
Military Unit within the area
facilities people enlisted in
a 20Km range the Armed Forces by
7 Greater propensity area
/ motivation for
Military service
8 Family (influence)

41 Positive 32 Greater index


integration of a 33 Awareness 9 Tradition
enlisted / potential
Military Unit within (marketing, etc)
candidates
the local society (historic)
34 Effective 35 "Marketing Plan" 12 Values
Volunteers Reward (resistance,
Policies toughness)
38 "Effect" of the
marketing campaign
in the area

Figure 2. Group map

7
Before the next working session, we performed some background work; namely we “cleaned
up” the group map and introduced it in the Decision Explorer software. The map was
structured in such a way that the relations were oriented towards its centre; “ends” appeared
in top of clusters of closely related “means”. The next working session started with the
analysis and discussion of the map. Again, cognitive mapping helped the participants to
expand their knowledge about the problem and to work out their differences of opinion. It
facilitated the mediation of different perspectives, leading to agreement over a coherent (i.e.,
consensual, operational, exhaustive and non-redundant) family of twelve fundamental
location concerns (FC), grouped in five areas of concern in the value tree of Fig. 3.

Figure 3. Value tree of key concerns

Defining descriptors for impact assessment


A descriptor (i.e., an ordered set of plausible impact levels) was associated with each FC. The
descriptors allowed the WG to appraise, as objectively as possible, the impacts of each
location option with regard to the 12 FC. Nine of the descriptors are “proxy-attributes” (cf.
Keeney, 1992), i.e. indicators that only enable the appraisal of impacts indirectly. These
activities were developed during the second-workshop sessions and a database including the
initial impacts of the 28 options in the 12 criteria was created.

3. Model building
The linear additive value model
The linear additive value model is the simplest and probably the most widely used MCDA
model. In the CIR location case, an additive model was constructed on the basis of experts’
qualitative value judgements. Using the MACBETH approach (Measuring Attractiveness by a
Categorical Based Evaluation Technique – cf. Bana e Costa and Vansnick, 1997 and 1999)
the following model building activities were developed during evaluation workshops with the
experts (cf. Fig. 4):

8
Descriptors

Value Scaling
functions constants

Sensitivity
analyses
Figure 4. Model building activities

- a value function was constructed on each descriptor, based on judgements of the WG. A
value function serves to translate impacts into value scores, indicating the attractiveness
relative to each other of the impacts within a FC;
- scaling constants (“weights”, in common language) were assigned to the plausible impact
ranges of the descriptors, so that the attractiveness values related to each FC could be
harmonised and aggregated;
- several feedback and sensitivity analyses were performed during this interactive process,
so that a requisite model could be built.

Let (g1(P),…, gj(P),…,gn(P)) be the impact-profile of a NUT “P”. The additive formulation
adopted was:
n
V(P) = V(g1(P),…, gj(P),…,gn(P)) = ∑ k j .v j (g j (P)) (1)
j =1

with
n v j (best j ) = 100
∑k = 1 , kj > 0 (j=1,…,n) and 
 v j ( worst j ) = 0
j
j=1

and where:

V(P) is a measure of the P overall attractiveness for the WG with regard to all FC
simultaneously,
vj(gj(P)) is the value of P measuring its partial attractiveness for the WG with respect to the j-
th FC,
bestj and worstj are the plausible best and worst impact levels of the descriptor of the j-th FC,
respectively, and
kj is the weight of the j-th FC.

In practical terms, V and vj (j=1, …,n) can be viewed as measures or indexes of location
attractiveness. Independence conditions between FC required by the additive model were
verified to be plausible working hypotheses in the framework of a constructive model-
building process (cf. Roy, 1990).

Building value functions with MACBETH


To measure how attractive (or unattractive) each NUT was in terms of relative priority for
installing a CIR, for each FC a 0-100 cardinal value-function was constructed on its
descriptor. A value function serves to translate impacts into value scores, so indicating the
attractiveness relative to each other of the impacts within a FC. The MACBETH approach was
extensively used during a two-day decision conference, in which the value functions were
developed from the answers of the WG to the MACBETH questioning procedure. The full
process is described in detail in (Thomaz, 2000) and Fig. 5 illustrates it for FC2.

9
Figure 5. Developing a value function with MACBETH

The WG was asked to judge qualitatively the difference in attractiveness between each two
impact levels x and y of the descriptor such that x is more attractive than y. This was done by
asking the WG to choose one of the six MACBETH semantic categories “very weak”, “weak”,
“moderate”, “strong”, “very strong”, or “extreme” as a qualitative measure of the difference
of attractiveness between x and y. When a certain judgement is inconsistent with previous
ones, MACBETH detects the problem and gives suggestions to overcome it (for details see Bana
e Costa and Vansnick, 1999). Note also that cases of group judgmental disagreement or
hesitation can also be considered, by choosing more than one category for a specific pair-wise
comparison of options (rather than to force an agreement on the choice of a single category).
For example, in Fig. 5, the agreed WG-judgement between levels “12” and “2” was “very
strong to extreme”. Once all value functions were constructed, they were included in the
database and automatically applied to the impacts of the options to determine their initial
value scores in all criteria.

Weighting with MACBETH


The scaling constants (“weights”) of the additive model (1) correspond to the concept of
“trade-off”: how much would the WG be willing to give up in one FC to achieve more on
another. Weights cannot be assessed by directly comparing FC in terms of their relative
importance, a common mistake to several popular weighting procedures (Keeney, 1992, pp.
147-148 calls it the most common critical mistake). The weights of FC 1 to 11 were assessed
with reference to the impact ranges of the descriptors, based on experts’ MACBETH
judgements. In the one-day decision conference devoted to weighting, the WG was asked first
to consider eleven reference location options (PVF1 to 11 in Fig. 6), each one corresponding
to the swing from the least attractive to the most attractive impact levels in each descriptor,
and rank them in terms of improvement of location attractiveness. Once an agreement on the
ranking of the swings was reached, the WG was then asked to pair-wise compare the
difference of overall attractiveness (“very weak”, “weak”, …, or “extreme”) of each two
swings. Finally, the corresponding MACBETH weights were presented to the WG and adjusted
interactively until an agreement was obtained on final weights.

10
Figure 6. Weighting with MACBETH

4. Elaboration of recommendations
Applying the model
With the value scores and the weights, the additive evaluation model could then be used to
associate an overall score – the priority index – to any NUT. Fig. 7a shows the results of the
model considering only FC1 to 11 (that is, without the territorial dispersion concern). If
territorial dispersion is desired, FC12 must be incorporated into the model and the successive
choice procedure applied. At each choice stage, the impacts (and the corresponding scores) of
the options were updated in the database to take into consideration dependence phenomena.
Of course, two or more NUTs could be selected at the same choice stage of the procedure, if
their overall values were close enough to provoke rank reversals for slight weight variations.
Here, sensitivity analysis played a crucial role.

The final decision conference


When faced with the result in Fig. 7a, the WG decided to introduce the concern of territorial
covering into the analysis. This was done by analysing the results of applying the model with
a weight of 10% in FC12 (see Fig. 7b – similar results were obtained with 5% and 15%). As
expected, the territorial dispersion of the selected NUTs augmented significantly. At this
stage of the learning process, a participant introduced a very interesting issue. He remarked
that all the analyses so far done were assuming that people always look for the closest CIR.
Indeed, this is not actually true, for people many times prefer to go to a regional
administrative centre even if faraway. The WG decided then to analyse the effects in the
results if this argument was considered. Lower scores in FC12 were given to the NUTs that
are in the same region than NUTs already selected in previous choice phases. However, the
results of the model remained stable. Moreover, extensive sensitivity analyses were
performed on several parameters of the model, to answer other “what-if?” questions. This
stimulate an intensive debate, at the end of which the group finally felt comfortable with a
commitment to action: to recommend to install one CIR in each of the nine NUTs painted
dark-gray in Fig. 7c and one CIR in either NUT 4 or NUT 8 (painted light-gray in Fig. 7c).

11
(a) (b) (c)
Figure 7. Results: a) Without territorial covering (FC12); b) With FC12; c) Final recommendation

5. Final comments
The first important conclusion is that the objectives of the case-study were achieved. Indeed,
key concerns for CIR location were identified and a transparent evaluation model was
developed and extensively used as a support tool, and, finally, the WG was able to
recommend locations for the CIR. Moreover, the socio-technical process enabled the WG to
achieve “a shared understanding of the issues, a sense of common purpose and a mutual
commitment to action” – the three key general aims of a group process according to Phillips
and Phillips (1993). Last but not the least, the case-study and its subsequent report strongly
contributed to increase the interest of applying Decision Analysis concepts and techniques,
and fundamentals of “modelling creativity” (cf. Clemen, 1996, ch. 6), in the Portuguese
Armed Forces. Several other military applications of Decision Conferencing, also using
quantitative models to capture qualitative judgements, can be found in (Watson and Buede,
1987) and (Bresnick et al., 1997). Other recent applications of the MCDA methodology
adopted in the case can be found for instances in (Bana e Costa and Vansnick, 1997), (Bana e
Costa et al., 1998 and 1999), (Bana e Costa, forthcoming) and (Bana e Costa and Oliveira,
forthcoming).

References
Bana e Costa, C.A. (1992), “Structuration, construction et exploitation d’un modèle
multicritère d’aide à la décision”, PhD Thesis, IST – Universidade Técnica de Lisboa,
Lisbon.
Bana e Costa, C.A. (1996), “Les problématiques de l’aide à la décision: vers l’enrichissement
de la trilogie choix-tri-rangement”, Recherche Opérationnelle/Operations Research, 30, 2
(191−216).

12
Bana e Costa, C.A. (forthcoming), “The use of multicriteria decision analysis to support the
search for less conflicting policy options in a multi-actor context: case-study”, Journal of
Multi-Criteria Decision Analysis.
Bana e Costa, C.A., Ensslin, L., Corrêa, E.C., Vansnick, J.C. (1999), “Decision support
systems in action: Integrated application in a multicriteria decision aid process”, European
Journal of Operational Research, 113, 2 (315−335).
Bana e Costa, C.A., Ensslin, L., Corrêa, E.C., Vansnick, J.C. (1999), “Mapping critical
factors for the survival of firms: A case-study in the Brazilian textile industry”, in G.
Kersten, Z. Mikolajuk, A. Yeh (eds.), Decision Support Systems for Sustainable
Development, Kluwer Academic Publishers (197−213).
Bana e Costa, C.A., Ensslin, L., Costa, A.P., (1998), “Structuring the process of choosing rice
varieties at the south of Brazil”, in E. Beinat, P. Nijkamp (eds.), Multi-Criteria Evaluation
in Land-Use Management, Kluwer Academic Publishers (33−45).
Bana e Costa, C.A., Ensslin, L., Zanella, I.J. (1998), “A real-world MCDA application in
cellular telephony systems”, in Th. Stewart, R.C. van der Honert (eds.), Trends in
Multicriteria Decision Making, Springer-Verlag (412−423).
Bana e Costa, C.A., Ferreira, J.A.A., Corrêa, E.C. (forthcoming), “A Multicriteria
methodology supporting bid evaluation in public call for tenders”, OMEGA.
Bana e Costa, C.A., Oliveira, R.C. (forthcoming), “Assigning priorities for maintenance,
repair and refurbishment in managing a municipal housing stock”, European Journal of
Operational Research.
Bana e Costa, C.A., Vansnick, J.C. (1997), “Applications of the MACBETH approach in
the framework of an additive aggregation model”, Journal of Multi-Criteria Decision
Analysis, 6, 2 (107−114).
Bana e Costa, C.A., Vansnick, J.C. (1999), “The MACBETH approach: Basic ideas, software
and an application”, in N. Meskens, M. Roubens (eds.), Advances in Decision Analysis,
Kluwer Academic Publishers (131−157).
Bana e Costa, C.A., Vansnick, J.C. (2000), “Cardinal value measurement with MACBETH”,
in D. Zanakis, G. Doukidis, C. Zapounidis (eds.), Decision Making: Recent Developments
and Worldwide Applications, Kluwer Academic Publishers (317−329).
Belton, V. (1999), “Multi-criteria problem structuring and analysis in a value theory
framework”, in T. Gal, T. Stewart, T. Hanne (eds.), Multicriteria Decision Making,
Advances in MCDM – Models, Algorithms, Theory, and Applications, Kluwer Academic
Publishers (12-1–12-32).
Belton, V., Ackermann, F., Shepherd, I. (1997), “Integrated support from problem structuring
through to alternative evaluation using COPE and V.I.S.A”, Journal of Multiple Criteria
Decision Analysis, 6, 3 (115–130).
Bresnick, T.A., Buede, D.M., Pisani, A.A., Smith, L.L., Wood, B.B. (1997), “Airborne and
space-borne reconnaissance force mixes: a Decision Analysis approach”, Military
Operations Research, 3, 4 (65−78).
Clemen, R. (1996), Making Hard Decisions: An Introduction to Decision Analysis (2nd
edition), Duxbury Press.
Eden, C. (1988),“Cognitive Mapping: a review”, European Journal of Operational Research,
36 (1–13).
Keeney, R.L. (1992), Value-Focused Thinking: A Path to Creative Decisionmaking, Harvard
University Press.
Phillips, L.D. (1982), “Requisite decision modelling: A case study”, Journal of the
Operational Research Society, 33, 4 (303–311).

13
Phillips, L.D. (1984a), “A theory of requisite decision models”, Acta Psychologica, 56 (29−
48).
Phillips, L.D. (1984b), “Decision support for managers”, in H. Otway, M. Peltu (eds.), The
Managerial Challenge of new Office Technology, Butterworths (80−98).
Phillips, L.D. (1990), “Decision analysis for group decision support”, in C. Eden, J. Radford
(eds.), Tracking Strategic Problems, SAGE Publications (142−150).
Phillips, L.D., Phillips, M.C. (1993), “Facilitated work groups: theory and practice”, Journal
of the Operational Research Society, 44, 6 (533−549).
Roy, B. (1990), “Decision-aid and decision-making”, in C.A. Bana e Costa (ed.), Readings in
Multiple Criteria Decision Aid, Springer Verlag (17−35).
Thomaz, J.P.C.F. (2000), “Concepção de um modelo multicritério de apoio à decisão para a
determinação da localização, a nível nacional, de Centros de Informação e Recrutamento
de voluntários para as Forças Armadas”, Ms Thesis, Universidade Lusíada, Lisbon.
Von Winterfeldt, D., Edwards, W. (1986), Decision Analysis and Behavioral Research,
Cambridge University Press.
Watson, S.R., Buede, D.M. (1987), Decision Synthesis: The Principles and Practice of
Decision Analysis, Cambridge University Press.

14

You might also like