ساخت سناریو از طریق دلفی و تجزیه و تحلیل تاثیر متقابل

Since its origins, decision makers have broadly used the Delphi method as a collaborative technique for generating important events and scenarios about what may happen in the future. This is a complex process because of the different interrelations and the potential synergetic effects among the relevant events related to a decision. This fact, along with the uncertainty about the occurrence or non-occurrence of the events, makes the scenario generation task a challenging issue in Delphi processes. In the 1960's, Cross-Impact Analysis (CIA) appeared as a methodological tool for dealing with this complexity. CIA can be used for creating a working model out from a set of significant events. CIA has been combined with other methodological approaches in order to increase its functionality and improve its final outcome. In this paper, the authors propose a new step-by-step model for scenario-analysis based on a merger of Turoff's alternative approach to CIA and the technique called Interpretive Structural Modeling (ISM). The authors' proposal adds tools for detecting critical events and for producing a graphical representation to the previous scenario-generation methods based on CIA. Moreover, it allows working with large sets of events without using large computational infrastructures. The authors present sufficient information and data so that anyone who wishes to may duplicate the implementation of the process. Additionally they make explicit a set of requirements for carrying out a Delphi process for a group to develop a set of significant events, collectively make the estimations of cross impacts, and to support a continuous planning process within an organization. They use two examples to discuss operational issues and practical implications of the model.

The use of scenarios to study the future is well known as an approach to studying situations that can lead to important changes and in which it is difficult to create explicit relationships among the events. Examples are the merger of two companies, extreme disaster or risk situations, major political happenings and/or the long term impacts of new or changing regulations or policies. All the events in the set are of a binary nature: a merger will or will not occur; a new specific policy will be established or not; a company will or will not go bankrupt; a given technological breakthrough will occur or not, etc. By means of scenario generation methods, forecasters make predictions about the occurrence or not of a set of events in time and/or describe a future story, from the present conditions to a set of plausible futures. In both cases, scenarios have been widely used for exploring the detection of future events together, as well as analysis of the path that leads to the desired future or prevents undesirable futures. That is what we call scenario analysis. In this sense, scenario-generation methods have often been used by decision-makers as an instrument to build landscapes of possible futures. Based on these future visions, decision-makers are able to explore different courses of action [1] and [2]. Scenario-generation methods combine a set of behaviors that mix qualitative and quantitative, subjective and objective methodologies in different layers [3]. The number of potential scenario methods is increasing as researchers and consultants from different backgrounds use their particular expertise to create new variations [4] and [5]. The Delphi method is one of the most used techniques for foresight [6], [7] and [8]. By means of Delphi method forecasters, based on the input provided by an expert panel, can make hypotheses about the occurrence or not of singular events. This success is mainly due to two of the main characteristics of the Delphi method: controlled feedback and anonymous interaction among experts. These characteristics help forecasters to avoid several limitations of traditional face-to-face experts' panels, such as unwanted leadership and high time cost [9]. Nevertheless, the inability of the Delphi method to make complex forecasts in which events are not isolated but interrelated is a limitation for scenario analysis. In a basic Delphi process the occurrence or not of an event was usually considered as if it had no effect on the rest of the event set. Cross-Impact Analysis (CIA) [10] was developed to address this limitation. CIA is a powerful tool for taking a set of binary future events and examining the potential causal impacts that the expectation or occurrence of each event may have on the others in the set. CIA was designed to calculate the basic impact of a political, social, or technological event on the occurrence probability of other events in the set. Due to this ability of CIA to analyze complex contexts with various interactions, CIA is one of the most commonly-used techniques for generating and analyzing scenarios. Another success factor of the approach in scenario analysis is that it is a flexible methodology that can be combined with other techniques such as Delphi [11] and [12], Fuzzy [13] or Multi-criteria [14] and [15] methods to allow true collaborative model building and scenario creation by groups. CIA is based on cross-impact questions that allow individuals to easily estimate the relationships among n events taken two at a time (n(n-1)/2 comparisons). It is an approximation to the real world where we do in fact recognize the further possibility of relationships among three, four, etc. events. This same approximation assumption is used in many other modeling areas such as measures of association and payoff matrices. For ten events a complete description among all possible interactions into the future of their occurrence would require gathering approximately ten million estimates [16]. This is calculated by following all possible occurrence sequences in a tree-like expansion for all possible sequences of events. No expert, manager, or team of judges would ever be able to undertake such an estimation process by Delphi or any other collaborative methods to do this. Many different approximation approaches to analyzing the more limited matrix model have been proposed such as an approximated Bayesian model [17] and an approximated systems dynamic representation [18]. There are even some very simplified approaches that directly solicit from experts the degree of impact between each interaction on an arbitrary scale (i.e. ++, +, 0, -, - -) and treat this as the degree of impact. Further discussions about different approximations approaches can be found in [12]. At the moment the method proposed by Turoff [16] is the only one that changes nonlinear probability measurement scales to linear interval measurement scales which makes it much easier for humans to view and understand the degree of influence one event has on other events as a consequence of their estimates. We observe that even for interactions among as few as ten events it is very unlikely that a single human can stay consistent when there is no feedback to show the consequences of the individual's estimates of impacts. To then combine the individual's judgments with the input from other estimators is not an activity that will lead to a trusted model of the situation. The need for improved visualization methods of all forms of complex data intended for supporting human decision processes is currently a major field of Information Systems and Science. In this paper, a new method for building scenarios based on CIA is described. This method is an extension of Turoff's CIA approach [16]. We focus on this CIA approach because of its previously mentioned capability of transforming the nonlinear probability measures to linear interval variables, among other advantages: 1.The estimator supplies a set of probabilities for the n events which are indicative of the non-linear nature of the future occurrence of the events. The estimator is then told to assume the opposite of the initial estimate about each event (i.e. if he or she thinks it is likely, they assume it will not occur) and to indicate how the probabilities of all the other events would change. 2. Among other properties of Turoff's model of cross-impact is the result that all the n equations relating the probabilities to one another are solved for a single consistent set of factors that produces an equation for the outcome of each event based upon the values of all the other events. This provides an inferred consistency from the solution method and allows individuals to see the consequences of their model before it is merged with the data of others. 3. The most critical property is that these equations take the non-linear probability factors between 0 and 1 and convert them to linear factors between each event pair that vary from plus infinity to minus infinity for each pair of events, i and j. This conversion of the non-linear probability factors to linear impact factors allows the estimator (or group if a collaborative estimation) to see a consistent set of linear relationships representing the degree to which a given event influences the occurrence of another (positive factors) or the degree to which a given event inhibits the occurrence of another event (negative factors). Specifically we propose combining CIA with the technique called Interpretive Structural Modeling (ISM) [19]. The main goal of this combination is using the previously mentioned linear nature of the cross-impact influence factors in order to integrate the CIA in a smooth incremental iterative manner with the ISM method. This allows the user to determine the stopping point for the condensation of events into scenarios and provides for a linear indication of how much of the impact information has been utilized as well as a graphical representation of the results. There are some antecedents for the application of structural analysis to CIA. Duval et al. [20] suggest that a cross-impact matrix can be structurally analyzed and portrayed by a signed directed graph. Novaky and Llorant [21] propose also an intuitive graphical representation based on CIA concepts. Ishikawa et al. [22] introduce a method for processing CIA outputs and building scenarios based on structural analysis concepts. Godet [23] proposes structural analysis for forecasting the key variables which bear on the future dimension. On the other hand, Martino and Chen [24] combine cluster techniques and CIA analysis in order to create ‘typical’ scenarios. Moreover, there is some previous research focused on using ISM in order to build scenarios and identify key drivers and actors [25]. In this paper we take these ideas and extend them by applying CIA–ISM for building scenarios. This new CIA–ISM approach adds tools for detecting critical events and graphical representation to the previous scenario-generation methods based on CIA. Moreover, it allows working with large sets of events without using great computational infrastructures, being a graphical representation of complex systems following a simplified structured process. This fact makes this methodology highly compatible with other complex systems analysis tools such as System Dynamics, Bayesian Networks, and Fuzzy Cognitive Maps. Traditional CIA approaches need great computational infrastructures (i.e. Monte Carlo simulation process and SMIC). If the problem has more that 10–15 events, traditional simulation processes cannot be easily designed nor run in a personal computer. Our model can be run in seconds by using basic spreadsheet software. The contributions of this paper are: 1. The extension of the original cross impact analysis process by Turoff [16] by merging it with the ISM process developed by Warfield [19]. This allows the users of the cross impact model to decide how to combine individual events into one or more scenarios. This process reduces the complexity of the results and produces a better user understanding of the implications of their estimations in creating the model. 2. The extension of the original model to allow the inclusion of initial condition events or the events which allow a probability of truth to be assigned to initial input events. Further we add output events that measure the degree of accomplishment of certain measures or goals at the end of the basic cross impact time period. 3. We present sufficient information and data so that anyone who wishes to duplicate the implementation of the process as described in this paper and the background theory from [16] will be able to carry out an implementation and have two data sets to be able to check the original model or the extended one. 4. We make explicit a set of requirements for carrying out a Delphi process for a group to develop a set of significant events to apply and point to examples in the literature of how this can be done with either a classical paper and pencil Delphi or a Delphi carried out as a Computer Mediated Communications process on the web. 5. We also make explicit the requirements for a Delphi process to collectively make the estimations in creating the model by the use of experts or knowledgeable people in gathering and verifying with them the resulting estimates. 6. Finally we point out how this new method of cross impact and applications of the Delphi make possible a continuous planning process within an organization. In order to illustrate the CIA–ISM approach step-by-step application, we use Turoff's original example [16]. This will allow others to check the consistency and mathematics of any effort to implement this process and moreover to demonstrate the ability to form meaningful scenarios out of the original event set, using the method of ISM. Based on this example, operational and analytical issues are discussed. Additionally, a new example is also discussed, in which some major extensions to Turoff's CIA approach are provided. This is new 18 event problem of predicting the outcome of software development projects in very different organizational situations. This problem introduces the methodological extensions to CIA of initial condition events and outcome events as two new event types that make CIA much richer in its potential span of application areas. It is thus possible to tailor a single general model to be able to examine very different situations with different initial properties and different outcome goals. These facts have several philosophical and practical implications which are analyzed at the end of the paper. But, firstly, the basics of CIA and ISM will be presented and analyzed in the following section.

The days of discrete, periodic long term plans and short term plans are numbered. Planning today for most organizations has to be made a continuous process. One needs to integrate the type of model described here into a complete planning environment. This would provide the following: 1. A system for professionals in the organization to evolve and update the event set that could be used in various models to evaluate various options for major decisions such as new or improved projects, mergers, investments, etc. 2. Participating groups of experts throughout the organization could enter their estimates about the interactions of the events for which they feel confident in making such judgments. 3. Improving both subjective estimates and the models that events contain as new information provides improved intelligence for the estimators. Given the growing complexity of our world, the problems of dealing with a world economy faced with growing threats to required resources and financial interactions that can lead to unpredictable results, we do need greatly improved methods to build and evolve models that can exhibit the requisite variety to cope with such challenges. We have seen the fallacy of using highly paid executives to make decisions based upon the intuition of one person who cannot, because of his or her status and pay, ask for serious help from other humans. This might well explain the retreat from some of the promises of strategic planning methodologies of the past by most non-technological corporations during the last twenty years. In this sense, planning should not be a discrete event in an organization; it must be a continuous process that is integrated into the fabric of an organization. Planning must also be a highly participative process by all the professionals that represent the professional community of the organization. The current understanding of the strengths and weaknesses of the organization must be understood by all the levels of management and this can only occur if the information of the professionals that carry out the functions of the organization flows freely into the planning process. Given the speed at which things happen in our very highly computerized environment, there is not an option to interrupt and put off work on short or long term plans. The distinction between the two becomes meaningless. The inputs of the events and initial conditions and the outcomes will be continually modified and evolved by an efficient Delphi process to allow hundreds to thousands of professionals to supply and interpret new information for generating new options and considerations to be incorporated into an organizational-wide model forming the basis of the planning process. The ability to test future options by choosing to implement them within the model and see likely forecasts of the outcome consequences becomes the essence of a continuous planning process with organizational wide participation. Collaboration and recognition that one of the consequences of the network integration now possible of all the views of the contributors in the organization is what has brought about a new age in society: The age of participation3[60]. Communications in any type of organization will allow collaborative planning on an organizational wide basis. The communication structures to allow many minds to contribute to the examination of complex problem solving and complex model construction are the major challenge in the design of future organizational systems. To contribute to this aim, a new scenario generation methodology has been proposed. The CIA–ISM approach aims at allowing researchers and practitioners to (1) handle complex systems; (2) obtain a set of plausible snapshots of the future; (3) analyze interaction between events; (4) detect critical events. This scenario-generation method might have several potential applications in scenario-planning, foresight, technology assessment, information systems and management. The main strong points of the authors' proposal are: 1. a strong theoretical background for the techniques on which the proposal is based; 2. the possibility of working with large sets of events; 3. tools for analyzing the key drivers of the scenarios; 4. specific software is not needed for making the calculations (e.g. Excel); 5.graphic output that gives a clear representation about the forecast; 6.compatibility with other techniques such as the Delphi, fuzzy or multi-criteria methods. Additionally, the introduction of the merger of cross-impact with ISM is a major extension in allowing the collaborative development of scenarios out of much larger event sets and this ultimately reduces the complexity of foresight. In the typical large scale event sets approaching a hundred events or more, estimation by groups will usually result in the need to subdivide the estimation process among different areas of expertise. It is generally true in Delphi that one should encourage participants to estimate or judge only those areas they feel confident in judging. We have provided an intelligent approach to the evolution of scenarios from event sets which augments human judgment in an integrated manner for the direct construction of a structural relationship model through the use of a computer in a true direct augmentation process without the need for intermediaries to implement computer programs [19] and [44].