MACRAME: A modelling methodology in multiactor contexts

E L S E V I E R Decision Support Systems 17 (1996) 331-343

l)eo on Suppo sgtems

MACRAME: A modelling methodology in multiactor contexts

Fulvio Buffa a Gilberto Marzano b Maria Franca Norese c

a Dip. Sistemi di Produzione ed Economia dell'Azienda. Politecnico di Torino, Torino. Italy b Laboratorio Progetti Speciali, INSIEL, Gorizia, Italy

c Dip. Sistemi di produzione ed economia dell'azienda. Politecnico di Torino Corso Duca degli Abruzzi. 24-10120. Torin, Italy

Abstract

A modelling methodology and a tool that supports the analyst in contexts characterized by a multiplicity of actors and information sources and a limited or absent interaction are presented in this paper. In these situations Problem Formulation becomes the first essential aim of a technical intervention applied to acquire, organize, present and reorganize knowledge of the action system, at both an individual and collective level; Model Structuring and Processing are strictly related in the same global approach. The basic structure of a prototype system, designed to assist the analyst in the structuring and then in the "critical reading process", is also presented.

Keywords: Problem and model formulation; Modelling and validation process

1. Introduct ion

"Defining a problem is a less constrained form of mathematical modelling, but subject to the same concerns: one wants to include all and only the important elements, leaving out unnecessary details without losing significant possibilities" [1]. Elements that must always be included in an explicit way are the analys t ' s perspect ive, that is, the adopted ap- proach, an organizat ional v iew and the process

models o f user involvement [2]. Others have to be identified, selected in relation to their relevance and significance, and structured in a way that allows multiple and flexible conceptualisations and expres- sions, at various levels of analysis [1].

To acquire and identify all the significant ele- ments, formally develop them, individually a n d / o r collectively verify reliability, relevance, consistency and completness, and control the process of problem formulation [3] is a cyclic process: each cycle in- duces a change in the schema that proposes formula-

tion and representation of the problem and a structur- ing of the model.

A number of studies have revealed that an analyti- cal and, to some extent, collective formulation of problem and models significantly improves decision making [4-7]. Knowledge elicitation and processing, conceptual modelling and validation, model sharing, integration and documentation are important func- tions that need support that is pertinent to the context and should be integrable and easily evolvable as the development proceeds. Different approaches and tools have been proposed in recent years (see for instance [8-14]). Some of these have been oriented to specific aims, such as problem identification and representation, problem formulation and structuring, structured modelling and model processing, while others have been more concerned with the global management of problems and models (for a critical review see [15]).

In the presence of various stakeholders, complex problems, incompatible goals and multiple criteria of

0167-9236/96/$15.00 © 1996 Published by Elsevier Science B.V. All rights reserved PII S01 67-9236(96)00008-5

332 F. Buffa et al. / Decision Support Systems 17 (1996) 331-343

success, the analyst has to integrate organizational and political dimensions within a cognitive-oper- ational paradigm (see for instance [3,10,11,16,17]). This implies both an individual level of analysis and reconstruction of the problem and context situation and also individual and /or collective work on the models, which are developed by interacting with different logics of action and the several representa- tions of the involved actors, exploring and redefining until an acceptable level of stability is obtained.

A methodological and technical approach and a tool that supports this approach to non-cooperative multiactor situations are presented in this paper. MACRAME (Multiple ACtor RepresentAtion Mod- Elling), which is analysed in the next section, has been used in different organisational and problematic situations (cf. [18]) as a modelling methodology in multiactor contexts and as a tool to support the process of development in these contexts, and seems to be sufficiently elaborate and free from structural limits to cope with complex situations. It is a multi- dimensional orienting process and schema. It helps the analyst to structure problems and models in multiple actor contexts where differently structured, conflictual and equivocal representations (which have arisen from the process of interaction between the analyst and organization actors or those that the actors wish to involve) become the essential objects of the analyst's concern.

It was initially elaborated after a long and intense intervention in a public sector inquiry [17,19] as an answer to documentation requirements (to help model understanding, use and updating), coordination be- tween analyst teams and support of all their different activities. It was then tested in real situations and made more general and flexible.

It can be used manually when there are few knowledge sources and when data, documents and human knowledge are limited. In data-rich problems this use becomes difficult and time-consuming. Only the latest real-world application has MACRAME been developed using a software tool. This case is still in progress, but it is already clear that this implementation of such a structuring tool can en- hance effectiveness. It enables the obtaining of richer and more detailed results, in less time, and with easier involvement of the actors in the structuring process.

The component parts of this orienting schema and some observations arising from the application of the MACRAME software tool are presented in the fol- lowing sections. The concluding remarks are essen- tially a summary of what MACRAME offers.

2. MACRAME methodology

MACRAME is a methodology that guides techni- cal support in relation to complex problems and above all to problematic situations where several actors are involved and perceive the " s a m e " situa- tion differently (cf. [20]). It helps the analyst in the collective phase of problem formulation and structur- ing and proposes multilevel and multiactor represen- tations of the problem at hand. It assists the different activities that characterize the formalized structuring of the problem and the relevant information elements into a consistent model; problem representations and model components are easily modified and reorga- nized in the individual or collective modelling-vali- dation process (cf. [21]).

MACRAME may be defined as a multidimen- sional orienting schema, which is "an active, infor- mation-seeking structure that orients a technical ac- tion" [22]. It helps the analyst to identify, analyse and select information elements (cognitive-construc- tive phase of an individual and/or collective struc- turing process) Then it helps to memorize, in a formalized structure, the elements that have arisen during the investigation and the meaning of each technical choice. It also explores and redefines these abstract representations in order to verify the reliabil- ity (e.g., of the sources and indications), relevance at different generality levels, and consistency and com- pleteness, both in relation to a single information- representation source and to a multiplicity of these. This analysis tries to distinguish between comple- mentary and antithetical representations [23], be- tween collective opinions and subjective points of view and between contextual solutions and more general ideas. Its results orient new investigation cycles, different additional representation develop- ments and the redefinition of conceptual models.

During each cycle the schema may become more precise, with the consideration of more details; more general, when new facts induce an accommodation

F. Buffu et al. / Decision Support Systems 17 (I 996) 331-343 333

of the representation and more complex, because profound modifications are possible or necessary (cf. [24]). Problem formulation, context representation and model structure change in each scheme until a stability level becomes temporarily or definitively acceptable (cf. [8,1]). At this point, the first phase of the modelling process is complete. The final scheme is a synthesis of the different technical results which documents the model (structure, parameters, valida- tion tests, results and information base), the formula- tion and structuring of the problem situation and the relationships between the formal and conceptual model of the problem situation.

MACRAME becomes operative again in a new phase, the collective "critical reading process", when the scheme becomes a documentation and communi- cation tool that is useful for recalling and explaining technical choices in problem structuring and concep- tual and formal modelling-validation. A global and specific use of the schema allows the elaboration of new model classes, by means of the modification of parameters or data, and makes each problem and/or model structure change easier in a negotiation con- text, by means of the analysis of the connections between the actorial representations, problem dimen- sions and model structure.

MACRAME is a schema that operates at different aggregation levels (see Fig. 1); the following are defined for each level, from the most general to the most specific:

a problem formulation, that may be expressed by one or more of the following structures

IO,,IOENERCa. LEVELI

~2, I RRST LEVEL- Mod A I'-~0,,I SECOND LEVEL- Mod. 1 l 1

/ t-[C~SECOND LEVEL- Mod. 2 I~ /

q~,[ FIRST LEVEL- Moa. Z [4o~[SECOND LEVEL- Mod.3 F

~SECONO LEVE,-Mo,., L~O~ISECONO LEVEL- Mod.5 I"

Fig. 1. A mult i level schema to represent MACRAME.

a statement of problem description, declaring the nature of the definite level of structuring and making the adopted technical perspective explicit a network that represents the actorial structure related to the level; each node denotes an actor and his /her role or principal function, the arcs explain the nature of the multiple relationships between the actors (actor structure network) a class of representation network in which nodes represent concepts and their sources and arcs denote relationships between concepts (represen- tation networks); the dimensions of the problem, that is, the disag- gregation of the topic of the level into its terms of reference, so that these can be deduced from the problem formulation (problem dimensions) and the dimensions of model structuring, that is, the transition structure from one part of the schema to another and from one level to another which is activated when a new level of representation structuring becomes possible or necessary, consis- tently with the multi-dimensional exposition of the problem (model structuring dimensions or model dintensions).

2.1. Problem formulation

At the general level of MACRAME, a declarative statement introduces the elements that, according to the analyst, globally constitute the problem. The actor-structure network presents the principal actors who are, in some way, involved and the relationship structure that pertains to the specific context (com- munication or information structure, functional, hier- archical, etc.). Representation networks can connect local causes, goals, proposed solutions, definitions and remarks to the proponent sources (actors of the decision process at different levels) and coordinate those concepts which are fragments of information and also problem elements.

Each problem situation requires a specific frame- work of formulation. MACRAME, as a modelling methodology oriented to multiactor contexts, always requires representations of the actorial structure; rep- resentation networks, which are a problem formula- tion structure, are activated when some form of debate is perceived as relevant to the problem or if a multiplicity of representations is solicited for the

334 F. BufJa et al. / Decision Support Systems 17 (1996) 331-343

structuring of the problem (cf. [25]). Problem de- scription may be very simple, but, if necessary, it can include different documents or a longitudinal analysis of the process that caused the problem situation, with specific references.

The representation of the actorial structure may be different and related to the adopted approach. The problem formulation methodology presented in Ref. [8], for instance, is oriented to analyse and represent a structure of interaction between the actors, in terms of general interaction or influence, potential conflict, good, distorted or no communication, good, distorted or no purposeful action, delay in communication or in purposeful action. The "web model" proposed by Kling [10] analyses different levels and modalities of actorial commitment. "Structuration theory" [26,27] specifies that all human interaction is inextricably composed of structures of meaning, power and moral frameworks and that three "modalities" link action and social structure (interpretive schemes, resources and norms). Disposal of data, knowledge and infor- mation structures are often seen as key elements of an actorial structure in a technical/operational per- spective (cf. [19,28]).

Some type of network represents key concepts that the actors considered during previous phases of the decision process or that are proposed during the analyst's intervention. The specific nature of rela- tionships among concepts may generate causally- chained networks (cf. [29,30]), relevance diagrams such as the knowledge maps proposed by Howard [31], or a richer structure that includes consequences and explanations such as the cognitive mapping by Eden [25]. The applications of MACRAME brought to light a multiplicity of different situations which have to be consistently represented, and a need for specification that grows with the problem and model structuring.

2.2. Problem dimensions and model structuring di- mensions

In all situations and at all the levels of the process the disaggregation of the topic of the level into its themes for discussion and terms of reference allows a multidimensional structuring of the perceived prob- lem. The decomposition of a problem situation into sub-problems is common in prescriptive literature [1]

"being needed for techniques to be properly tar- geted" and decompositional strategies and frame- works are proposed for problem solving, decision analysis and problem formulation approaches.

With MACRAME, problem dimensions are mainly related to the adopted approach, a Multiple Criteria approach (see for instance [32,33,38]), char- acterized by a non-problematic disaggregation of the perceived problem, related to the possibility of ag- gregating points of view and problem dimensions at the appropriate moment and by one of the multicri- teria methods (cf. [33]). Problem dimensions and model structuring dimensions are strictly con- nected.The analysis of the problem at the local level, and by the different structures of problem formula- tion, gives rise to consistent problem dimensions. Problem dimensions may induce a new structuring of the perceived problem, or of one upper level of problem formulation, and then a revision of the general schema; or they may stimulate the structur- ing of one or more sub-problems, and then the transition from one level to another. The dimensions of model structuring explain the governing process [34]. This process consists of control and validation activities oriented towards these requirements: verifi- cation, evaluation and selection of different structur- ing opportunities, new exploration processes and ac- tions on the developed representations; formaliza- tions and structuring of the operational model and applications and testing of these formalized models in relation to the conceptual model and then to the reality.

The schema evolves with the sub-processes that it orients. It documents the logic and context of every technical choice and controls the coherence of the sub-processes and the use of models which have arisen from this structured and systematic reading of interactions between the representations. At the end, it provides a document structure in which the Gen- eral Level presents the global approach to the prob- lem as the result of the problem formulation process. The other levels propose local perspectives generally related to subsystems that are logically distinguished in terms of sectoral problems in which only some actors or users are involved. These subsystems be- come more and more specific and are often con- nected to only one information source. The model structure is also broadly defined at the starting levels

F. Buffa et al. / Decision Support Systems 17 (1996) 331-343 335

of the schema and becomes more and more detailed and related to local perspectives and information sources as the process continues.

3. A prototype tool for MACRAME

As was said in the introduction, MACRAME can be used manually, but only in simple situations. A computerized problem and model structuring proce- dure, that is also a model documentation and pro- cessing procedure, is now being implemented. In the first version a Borland Object Vision for Windows software programme was used as a support for both the data base and model base management. Object Vision (OV) proved to be interesting in relation to the authors' needs, but not sufficient for operational needs. Its graphical interface is easy to learn and use and the effort given to its development is rewarded by the quality of results. This is important in the reading process when the problem and model struc- tures are collectively analysed, validated and used as documentation to modify problem and/or model elements. Moreover, in the structuring process, local or global revisions are very frequent and the graphi- cal interface has to be modified or redesigned each time.

This limitation, and other minor difficulties, led us to try a new approach and the first prototype implementation uses the experience acquired through OV and the integration with other tools in a new environment. The prototype has been obtained using QuerySys, a new generation windows-based full text retrieval environment (cf. [35,36]). QuerySys run- time libraries allow the processing of any type of electronic information, such as structured and full text data, images, graphics and voice. This environ- ment integrates three different paradigms of data management: DBMS for processing structured data, information retrieval for processing full text data and hypertext for browsing and linking different chunks of information. QuerySys libraries are appropriate for rapid implementation in document processing, run on network, require a PC Intel 80386 or higher and support the on-line full text database updating. This feature is very important since it allows one to insert, delete or modify records that also contain full text fields, without reindexing the whole database.

The MACRAME prototype is written in C-language; Microsoft SDK libraries for graphic interface imple- mentation and Querysys libraries for data manage- ment have been used.

MACRAME information is organized in a main database and various support databases whose num- ber depends on the problem situation faced. The main database contains complex data objects formed by full text data, structured data and links. Each data object corresponds to the stereotyped representation of a single main problem or a specific subproblem. Single main problems are decomposed into subprob- lems and a subproblem can be decomposed into more specific subproblems. There is no limit to the number of decompositions. The stereotyped descrip- tion is the same whether it deals with a main prob- lem or with a derived subproblem. A mnemonic alphanumerical label can be associated with the vari- ous data objects. The program organizes the subprob- lems into an inclusion hierarchy. The user can navi- gate through this hierarchy by selecting the data object labels, a procedure similar to browsing in a hypertext system.

Each data object contains a full text zone and a formatted zone. The full text zone is formed by three full text fields, Problem Description, Problem Di- mensions and Model Structuring Dimensions. The formatted zone indicates the date of the last data object updating, it specifies whether the specific problem/subproblem has been introduced by the analyst in the "structuring phase" or by an actor in the "critical reading phase", and it contains a syn- thetic judgement, that is selected from a given table, and a reference to local formalized results, locally activated operations and "open problems". Texts included in the full text zone and data in the format- ted zone are stored in the same database. QuerySys run-time functions allow one to browse and view information in a variety of combinations and suitable contexts.

QuerySys manages unidirectional and bi-direc- tional links between data objects belonging to the same or different databases. A label can also be associated to a link and can be used to retrieve and browse over the data objects which share the same label. In the authors' prototype this feature was used in order to build the Actor Structure Network and to link the items of the main database to chunks of

336 F. BuJJit et al. / Decision Support Systems 17 (1996) 331-343

information stored in the support databases. The Actor Structure Network describes the relationships between the actors quoted in the main database in a graphic way. Mnemonic labels have been used to qualify the relationships.

The external links, which are posed between the main database data objects and the items belonging to the support database are also very useful, making it possible to obtain information from other specific databases by means of the reference contained in an item of the main database. An unlimited number of support databases can be used. Since the program adopts a multidocument approach, it is possible to open various search sessions on the screen at the same time and simultaneously visualize information from different databases. A search session is a win- dow containing information retrieved by means of a sequence of queries. The system also offers the opportunity of collecting and storing the outcomes selected from several search sessions in a personal folder. These contents can be saved on disk.

The full text search is an important facility in the prototype. It is easy to obtain access to the data objects by using the keywords contained in the full text fields. Each word of the text is a keyword, with the exception of those that appear in the stopword list. The user can specify topics of interest through a query so that the system can find data objects that deal with the same topics. Keywords are also corn-

bined using Boolean operators to specify topics for a query. Links can then be established between the data objects that satisfy a given query.

Windows SDK is not the simplest GUI develop- ment environment. Moreover it has been chosen to implement the user interface because several data structures and lists of data structures and arrays characterize the MACRAME applications and there- fore pointers to memory objects must be used. The high level languages such as Visual Basic or Power- Builder does not allow this feature. Another reason has been the limited number of windows that can be opened simultaneously in Microsoft Windows. Low level SDK GUI functions permit one to overcome this problem.

4. A session of problem and model structuring

The MACRAME-Problem Structuring and Mod- elling (PSM) system is software designed to assist the analyst in the structuring and then in the commu- nication process. In both these contexts, MACRAME-PSM uses diagrams as user interfaces. Texts, graphs and other stored information elements are accessed via the diagrams through clicking the mouse onto particular elements of the diagrams. The diagrams are data in that they are constructed, as the study proceeds, to describe the general or particular

Fig. 2. Main screen and Multilevel schema.

F. Bul~h et al. / Decision Support Systems 17 (1996) 331-343 337

Fig. 3. Map screen.

problem structuring and modelling; they also provide access to data since these are always attached to some aspects of the diagrams. This section presents the basic structure of a M A C R A M E - P S M session of problem and model structuring and samples of the main screens.

The normal starting point is the " 'Main screen" of Fig. 2. It lists several key topics on the menu bar.

By clicking onto one of the key topics, e.g., " 'appli- ca t ion" , a menu itemlist appears in which the user can choose the commands "~New", for creating a new application, and " O p e n " for opening an already existing application.

For each application of M A C R A M E . a Multilevel schema and a Map are activated in relation to a new " p r o b l e m " object, first in relation to the General

Fig, 4. Module screen.

338 F. Buffa et al. / Decision Support Systems 17 (1996) 331-343

Level, i.e., the global view of the problem situation, and then to the following levels. The "Multilevel Schema" (see Figs. 1 and 2) offers a disaggregation of the problem into levels of growing specification and lesser complexity. One or more modules contain all the basic elements of MACRAME for each level.

The " M a p " is a much more detailed schema. It has been proposed in [37] as a tool of structural modelling which consists of "e lements" and "con- nections"; the "elements" are states of knowledge, either intermediate or final; the "connections" are processes of refinement (called steps) which lead one from one state to the next. In [37] three kinds of activity may occur in each step: 1. an additional assumption is made (A), 2. further information is added ( ~ ) , and /or 3. an algorithm or operation is executed ( * ).

The Map screen (see Fig. 3) is activated by the Multilevel Schema and it organizes states of knowl- edge (the Problem Formulation and the Problem Dimensions) which are essential at the different lev- els of problem structuring and modelling.

All the basic components of the "p rob lem" ob- ject can be called and analysed by clicking corre- sponding elements and steps onto the Map and can also be called by the Multilevel Schema screen, but have to be inserted into the "Module screen" (see Fig. 4). They are then automatically inserted into the

Map and the Multilevel Schema. The Module screen includes, both at the General Level and at the other successive levels, the different structures of Problem Formulation and the Problem Dimensions and Model Structuring Dimensions full text fields.

The first Problem Formulation structure, a Prob- lem Description full text field, allows a synthetic presentation of the problem and context situation. The analysis of the decisional and operational con- text goes through the identification of the main involved actors, their attitudes and roles in relation to the problem, and their potentialities as regards the problem structuring and modelling mainly in terms of specific/general knowledge of the problem, avail- able information elements and levels and modality of commitment.

In the system that is specifically oriented to multi- actor contexts, all the acquired elements are stored in support databases in relation to each individual or collective actor recognized as being an information source and /or a key stakeholder in the past, present or future decision processes. The main elements that characterize the actor's position in relation to the problem situation and the decision context and pro- cess, are also stored by "Actor ' s screens" (see Fig. 5), where a text may include all the relevant ele- ments or may be supported by a connection to more remote files such as data bases or document bases.

Macrame - Problem Structudn Application Actor Mop Stcps

,~ Actor

J E-o Role and main functions

KAG - DANTE

R Actor LIA- DANTE

Past, present and future commitment Interaction structure Information structure l

J°ii N N

g N Actor KA-DANTE

Fig. 5. Ac to r ' s screen.

F. BufJa et al . / Decision Support Systems 17 (1996) 331-343 339

Fig. 6. Actor Structure Network screen.

Each Actor's screen is created starting from the "Main screen" and clicking onto the key topic Actor on the menu bar. All the acquired and stored elements are synthetically presented in the Actor Structure Network (see Fig. 6), that is the second structure of Problem Formulation.

When specific elements of the problem and/or the context are stated or perceived in an equivocal way the third structure of Problem Formulation, the Representation Networks, can be used to structure, analyse and reduce uncertainty (see Fig. 7). In multi- actor contexts, problem structuring and modelling

Fig. 7. Representation Networks screen.

340 F. Buffet et al. / Decision Support Systems 17 (1996) 331-343

require a long phase of data acquisition and analysis. There are many information elements and these are usually unstructured (documents, interviews, reports and so on). Search sessions in the support databases, where these information elements are stored, single out recurring concepts and sources that constitute the Representation Networks nodes. The relationships between concepts (and sometimes between propo- nent actors or organizations) constitute the arcs, but often their "nature" is not sufficiently clear. In Fig. 7, for instance, Cost reduction and System efficacy may be seen as conflictual goals for two groups of actors or as two contexts of action that may or must generate a shared procedure. A clear definition re- quires individual and collective analysis of the Rep- resentation Networks, and sometimes new investiga- tion cycles or problem disaggregation when different interpretations have to be studied in depth.

Specific Problem Dimensions and Model Struc- ture Dimensions are then induced to disaggregate the topic of the level into its terms of reference and to define the activities that will characterize a new investigation cycle or the structure and elements of a new schema level. The Dimensions are declared in detail in the other two full text fields of the "prob- lem" object (see Fig. 5) where a synthetic judgement specifies the nature of the situation, oriented to a more precise schema (need for further or better information), a more complex schema (one or more new levels; different modules for each level) and a more general schema (sufficient and consistent infor- mation elements induce clear assumptions and model structuring), and a stable schema where the problem and the model are formally represented.

In the Map, the steps that lead from a state (and level) of Problem Formulation (PF) to the element Problem Dimensions (PD) consist of carrying out an operation ( " ) that is, mainly, the analysis of the PF component parts, by search sessions and by individ- ual or collective examination. The steps from a PD element and the next level PF element consist of the activation (") of the transition structure from one level to another (and also from one Multilevel Schema to another), that is the Model Structuring Dimensions (MSD). MSD may be supported by spe- cific operations/activities (" ), mainly the identifica- tion of new knowledge elements, their validity con- trol and a formalized development of modelling ele-

ments, respectively R1 (identification), R3 (control) and R4 (development) activities as described in [3]. PF and PD are MACRAME component parts that are present at each level in each module; MSD is present at each level, but different modules of the same level can propose the same MSD or each module can propose its specific MSD.

Some steps connect elements of PF and PD to local outputs that either represent temporary/defini- tive results or "open" problems which impose de- lays or immediate operations. These steps are charac- terized by different activities, such as additional assumptions (A), new information acquisition ( S ) or operations (*) such as R1 (identification), R3 (control), R4 (development) or R5 (communication), activated alone or as support to MSD ( "3. The output components of each MACRAME application are in- serted by the "application" menu and qualified by roman numbers; a reference to local formalized re- sults and to "open problems" is automatically in- serted into the Module screen formatted zone. The outputs can be called and analysed by clicking into the Map correspondent elements.

Local formalized results can be obtained at almost all the different levels, but only at the last can the global model be formulated in relation to a suffi- ciently structured and therefore reduced complexity. The passages from one level to another have induced a schema that is sufficiently precise, complete, gen- eral and stable. All the elements of the model are shown, in explicit mutual relationships and related to sources and proponent actors. Hypotheses of model variants, in relation to different recognized contexts of action, are generally shown at the last level.

This last Multilevel schema synthesizes the "structuring phase" results. In the second use of MACRAME, i.e., the collective "critical reading phase", the Multilevel Schema offers both a global view of the problem, mainly in terms of disaggrega- tion level, and the possibility of navigation through the scheme modules, in order to analyse, discuss and change some basic elements of the problem and model structure and /or some relationships between these elements and between each element and all the related databases. Modifications of this schema can be frequent in the "structuring phase"; any change to the Multilevel Schema is stored and presented in the "S teps" file attached to any specific application

F. Buffa et al. / Decision Support Systems 17 (1996) 331-343 341

(see Fig. 8). In the "critical reading phase" a de- tailed analysis of a specific problem structure, syn- thesized in the last Multilevel schema or in a previ- ous one, is made possible by clicking onto the first level module (General Level) of the holding atten- tion schema. The General Level module is the start- ing point for a detailed analysis. The passage from one module and level to another module and, if necessary, to another level is activated by the "back- ward" and " forward" buttons (see Fig. 4).

Substantial changes in the scheme require control activities to assure global coherence and can induce a new "structuring phase". These changes are also stored to be used in this modelling phase or in future model-management actions.

5. Concluding remarks

Interaction between the analyst and the problem- context through the knowledge and perceptions of the key actors is essential to elaborate a model for the organization, especially when there is a lack of a formal Information System and a normative Frame of Reference. The elements that characterize the problem situation and the decision context and pro- cess are used in the problem structuring to include

all the different points of view and knowledge ele- ments in an integrated and transparent model formu- lation. Information on the actor's role in the prob- lematic situations is essential to distinguish between the general problem formulation and the different specific sub-problems, in order to correctly model the actor's "posi t ions" that are oriented to offer ideas, solutions or expertise acquired in the field, to reject other actor's points of view, to integrate knowledge or negotiate the terms of the problem and SO o n .

The analyst needs support in all the activities that characterize this interaction. The action that allows the problem to be structured and a consistent model to be formulated and validated must be aided at both an individual and a collective level. MACRAME, as a proposal of methodological and operational support in these contexts, is oriented to multiple functions and two main uses: assisting the analyst in the structuring and then in the documentation and com- munication process. The main functions are:

(a) formulation of the problem at hand, i.e., expla- nation of the client's initial demand and then identi- fication, detailing and structuring of all the relevant elements (such as problematic situation nature and complexity, decisional and organizational context, key stakeholders' roles and commitment, and infor- mation structure),

Fig. 8. Steps screen.

342 F. BufJa et a l . / Decision Support Systems 17 (1996) 331-343

(b) the analysis and selection of information ele- ments oriented to the modelling, i.e., data storage, processing and validation, lack of information recog- nition, etc.,

(c) model structuring, (d) model validation, mainly in relation to all the

relevant formulations and the developed "ad hoc" information base, and model documentation (in terms of structure, parameters, validation tests, results and information base),

(e) model management, to change the structure or modify parameters and data.

These functions are suitable for all the tasks of the global management of problems and models (see [15]). Functions a, b and c, that are oriented to problem and model formulation and representation from the human point of view, are developed mainly in the initial phases of the work when the unstruc- tured problem and the acquired information elements are analysed and structured. Functions d and e, that are oriented to formulation and representation from a system point of view and to model processing mainly from a human point of view, are generally less activated in the first model structuring and become the main functions in the following individual and/or collective model analysis. This methodology aims to be an integrated approach that includes the human view and the systems point of view of problem and model formulation and model representation and processing.

References

[1] G. Smith, Defining Managerial Problems: A Framework for Prescriptive Theorizing, Management Science 35, No. 8 (1989) 963-981.

[2] M. Newman and F. Noble, User Involvment as an Interaction Process: A Case Study, Information Systems Research 1, No. 1 (1990) 89-113.

[3] M.F. Norese and A. Ostanello, A Procedure to Evaluate Domand/Supply of Sport Services and Simulate Intervention Policies, Atti AIRO (Tecnoprint, Bologna, 1985) 467-492.

[4] R. Volkema, Problem Formulation in Planning and Design, Management Science 29, No. 6 (1983) 639-652.

[5] H. Sol, Conflicting Experiences with DSS, Decision Support Systems 3 (1987) 203-211.

[6] A. Balakrishnan and A. Whinston, Information Issues in Model Specification, Information Systems Research 2, No. 4 (1991) 263-286.

[7] I. Vessey and D. Galletta, Cognitive Fit: An Ampirical Study of Information Acquisition, Information Systems Research 2, No. I (1991) 63-84.

[8] K. Bowen, An Experiment in Problem Formulation, Journal Opl. Research Society 34 (1983) 685-694.

[9] A. Geoffrion, An Introduction to Structured Modeling, Man- agement Science 33, No. 5 (1987) 547-588.

[10] R. Kling, Defining the Boundaries of Computing Across Complex Organizations, in: R. Boland and R. Hirschheim, Eds., Critical Issues in Information Systems Research (Wi- ley, Chichester, 1987) 307-362.

[ 11 ] J. Rosenhead, Ed., Rational Analysis for a Problematic World: Problem Structuring Methods for Complexity, Uncertainty and Conflict (Wiley, Chichester, 1989).

[12] W. lsaacs and P. Senge, Overcoming Limits to Learning in Computer-Based Learning Environments, European Joumal Opl. Research 59 (1992) 183-196.

[13] T. Jelassi, M.R. Klein and W.M. Mayon-White, Eds., Deci- sion Support Systems: Experiences and Expectations (North- Holland, Amsterdam, 1992).

[14] M. Basadur, S.J. Ellspermann and G.W. Evans, Eds., A New Methodology for Formulating Ill-Structured Problems, Omega 22, No. 6 (1994) 627-645.

[15] A.A. Baldwin, D. Baldwin and T.K. Sen, The Evolution and Problems of Model Management Research, Omega 19, No. 6 (1991) 511-528.

[16] J. Moscarola, Organizational Decision Process and ORASA Intervention, in: R. Tomlinson and I. Kiss, Eds., Rethinking the Process of Operational Research and Systems Analysis (Pergamon, Oxford, 1984) 169-186.

[17] M.F. Norese and A. Ostanello, Identification and Develop- ment of Altematives: Introduction to the Recognition of Process Typologies, in: A.G. Lockett and G. Islei, Eds., Improving Decision Making in Organisations (Springer- Verlag, Berlin, 1989) 112-123.

[18] M.F. Norese, MACRAME: A Problem Formulation and Model Structuring Assistant in Multiactorial Contexts, Euro- pean Journal Opl. Research 84 (1995) 25-34.

[19] M.F. Norese, A Multidimensional Model by a Multiactor System, in: B.R. Munier and M.F. Shakun, Eds., Compro- mise, Negotiation and Group Decision (Reidel, Dordrecht, 1988) 263-276.

[20] R. Flood and E. Carson, Dealing with Complexity (Plenum, New York, 1988).

[21] M. Landry, J.L. Malouin and M. Oral, Model Validation in Operations Research, European Journal Opl. Research 17 (1983) 207-220.

[22] U. Neisser, Cognition and Reality: Principles and implica- tions of Cognitive Psychology (Freeman, New York, 1976).

[23] R. Mason and I. Mitroff, A Program for Research on Man- agement Information Systems, Management Science 19, No. 5 (1973) 475-487.

[24] B. Espinasse, A Cognitivist Model for Decision Support: COGITA Project, A Problem Formulation Assistant, IFORS- SPC1 March 91, Decision Support Systems 12, Nos. 4 /5 (1994) 277-286.

F. Buffa et al . / Decision Support Systems 17 (1996) 331-343 343

[25] C. Eden, Cognitive Mapping, European Journal Opl. Re- search 36 (1988) 1-13.

[26] A. Giddens, The Constitution of Society: Outline of the Theory of Structure (University of California Press, Berke- ley, 1984).

[27] W.J. Orlikowski and D. Robey, Information Technology and the Structuring of Organizations, Information Systems Re- search 2, No. 2 (1991) 143-169.

[28] J. MEl~se, Approche systemique des organisations (Editions Hommes et Techniques, Paris, 1979).

[29] K. Nakamura, S. lwai and T. Sawaragi, Decision Support using Causation Knowledge Base, IEEE Transactions Syst., Man, Cybern. 12, No. 6 (1982) 765-777.

[30] J. Vennix and J. Gubbels, Knowledge Elecitation in Concep- tual Model Building: A Case Study in Modeling a Regional Dutch Health Care System, European Journal Opl. Research 59 (1992) 85-101.

[31] R. Howard, Knowledge Maps, Management Science 35 (1989) 903-922.

[32] J. De Montgolfier and P. Bertier, Approche multicritere des problemes de decision (Editions Hommes et Techniques, Paris, 1978)

[33] B. Roy, Methodologie Multicritere d'Aide a la Decision (Economica, Paris, 1985).

[34] Ch. Schneeweiss, On a Formalization of the Process of Quantitative Model Building, European Journal Opl. Re- search 29 (1987) 24-41.

[35] G. Marzano, S. Franzin, G. Gregori and E. Silli, Visual Information Retrieval: verso la definizione generale di un approccio operativo, in: Informatica e Diritto (Edizioni Sci- entifiche ltaliane, Firenze, 1993) 83-105.

[36] INSIEL, Querysys Document Retrieval System - - Manuale Operativo (Extralito, Pasian di Prato, 1993).

[37] G. Lendaris, Structural Modeling - - A Tutorial Guide, IEEE Transaction Syst., Man, Cybern. 10, No. 12 (1980) 807-830.

[38] M.F. Norese, A Multiple Criteria Approach to Complex Situations, in: M.C. Jackson et al., Eds., Systems Thinking in Europe (Plenum Press, New York, 1991) 361-369.