Description
Industrial district was initially introduced as a term to describe an area where workers of a monolithic heavy industry (ship-building, coal mining, steel, ceramics, etc.) live within walking-distance of their places of work.
Research Study on a Model for Simulating Reputation Dynamics in Industrial Districts
Abstract
In this work we try to draw an interdisciplinary framework aimed to integrate a socio-cognitive approach with organizational research about industrial clusters, in order to investigate whether and how social evaluations may affect clusters' dynamics. Industrial districts are sort of "small-worlds" that provide a suitable environment for testing the predictions of our model through Multi-Agent Based Social Simulation (MABSS) experiments. Artificial agents will be allowed to exchange products and to transmit social evaluations, and the main features of a cluster will be implemented. The effects of evaluation transmission on products' quality, partner selection and cheaters' isolation will be discussed.
Keywords: reputation; industrial clusters; social simulation; cognitive modelling.
1. Introduction
The exchange of social evaluations is an extensive activity in human societies. Individuals create, manipulate, modify, and circulate these evaluations, using them to inform, to convince and also to deceive other individuals. This is true also at the supra-individual level, in which institutions, countries, corporations and firms, just to name some, devote a lot of resources to acquire and preserve good reputations. Moving from the individual to the collective level, the complexity of this phenomenon increases, and a simulation model is required to deal with this
complexity. In this work we propose a model for simulating the effects of reputation on profits and quality of production in an industrial cluster. Industrial districts1 are geographically defined production systems [1], common around the world (see the geographical clusters in the US or the milieux innovateur in France). According to Porter, "clusters are geographic concentrations of interconnected companies, specialized suppliers, service providers, firms in related industries, and associated institutions (for example, universities, standard agencies, trade associations) in a particular field that compete but also cooperate" [16]. These networks of firms, suppliers, and institutions have been investigated by several different disciplines, ranging from industrial organization to economic geography and other social sciences. These studies have developed different notions and models, in order to understand the main features of these clusters and to develop effective tools for the amelioration of performance and competitiveness. One key issue in these kinds of studies is innovation, which has been investigated mainly in terms of learning processes [9], allowing firms to improve their knowledge, to develop their know-how and to get good innovation levels. However, better performance could be not recognized as such if other agents do not hold positive evaluations about that cluster, i.e. if it has a bad reputation. Computer-based simulations of industrial clusters are not new: different models have been put forward [10, 19] and several aspects have been addressed, such as their innovation dynamics
1
In this work district and cluster are used as synonyms.
[2], the importance of relationships among firms and institutions [18], and also the evolution of the cluster itself [4]. All these studies share the same interest in economic features of firms and in their spatial agglomeration, which are considered as key factors in determining clusters' competitiveness and innovation strategies. Starting from similar research questions, we want to model the relationships among firms in a quite different way, focusing on social links among agents and on the rich social structure, usually informal, that is a defining feature of industrial clusters. Social evaluations are the building blocks of social and economic relationships inside the cluster; they are used to select trustworthy partners, to create and enlarge the social network and to exert social control on cheaters. In this work we explore the contribution that a cognitive theory of reputation [5, 6] may provide to the studies about industrial clusters and their dynamics. This cognitive approach considers reputation as a highly dynamic process that is deeply rooted in social exchanges and implies several consequences for the subjects involved. Moreover, this model allows us to distinguish between one's own evaluations about other people, namely image, and reputation, which is impersonal. This distinction is not inconsequential and gives rise to different phenomena and dynamic effects that our model is intended to investigate. To test our model in this complex environment we use an agent-based simulation approach, i.e. a computer-based way of formalizing interacting agents in a given environment. Multi-AgentBased Social Simulation (MABSS) is a scientific methodology consisting in the creation of models of artificial societies, i.e. sets of autonomous agents planning, acting and evaluating their actions in an artificial environment with given features. The validity of this methodology has been widely recognized [12, 13, 14]. MABSS is a powerful and flexible tool: it allows the investigation of complex environments where macro-behaviors arise from micro-behaviors, and the resulting properties are not known a priori, as it is the case with reputation. In Section 1 we will put forward the theoretical background and we will discuss the importance of reputation for both natural and artificial societies. The simulation model will be
presented in Section 2 along with some hypotheses regarding the effects of social evaluations in simulated industrial clusters. The scenario and the experiments will be presented in the third section. Finally, a discussion about the model and its results will be provided.
2. Reputation: from natural to artificial societies
If one were to name one of the most influential and recurrent social phenomena, reputation would undoubtedly be one of them. In human societies the exchange of social evaluations is crucial to partner selection, social control, and coalition formation, just to name some of the main functions of reputation. Holding a good reputation is really important and, in some contexts, it is essential to survival, such for instance in competitive markets, in which stakeholders' decisions are often rooted in the firm's reputation [7, 15]. In everyday life, reputation works as a compass to avoid dangerous partnerships and to find reliable collaborators, in both small and large social groups. Reputation is an influential theme also in research, as witnessed by the huge amount of studies carried out on this topic by scholars coming from different disciplines, ranging from psychology [ 3, 20], to management [11, 18], to economics [17] and even to biology [8]. Often, these studies are limited to the specific domain of application they refer to, thus providing neither general explanations nor accepted results. This is partially due to the multiplicity of aspects concerning reputation, and to the difficulty to put all these strands together. An attempt to yield a more comprehensive theory of reputation has been made by Conte and Paolucci [5], who used a socio-cognitive framework to describe how people create, manipulate and transmit social evaluations, and how these evalutions affect individuals' beliefs and behaviours. The social cognitive model is a dynamic approach that considers reputation as the output of a social process that starts in agents' minds. The input to this process is the evaluation that agents (Evaluators) directly form about a given Target during interaction or observation. This evaluation can be transmitted to Beneficiaries that share the goal with regard to which targets are evaluated and thus may use this information as a guide for their behaviour. We distinguish between two kinds
of social evaluations: image and reputation. Image is the output of a process of evaluation regarding another agent. In social life, people continuosly assess their colleagues, friends, partners, etc. with regard to their personal features, competences, behaviors and so on. These evaluations are the social images of those agents. In other words, image is an assessment of the positive or negative qualities of a target with regard to a norm, a competence, and so on. The main distinctive feature of image is its being personal, i.e. the identity of the evaluator is always clearly expressed, as for instance in the sentence "I believe John is a good guy". A given image includes three sets of agents: 1. a set of agents who share the evaluation (E) - Evaluators 2. an evaluation target (T) - Target 3. a set of Beneficiaries (B), i.e., the agents sharing the goal with regard to which the targets are evaluated. Often, evaluators and beneficiaries coincide. Social evaluations may concern physical, mental, and social properties of targets; agents may evaluate a target as to both his or her capacity and willingness to achieve a shared goal. After assessing someone, the evaluator may transmit this image to someone else, in response to a specific request (e.g., "Do you know that guy?", "Yes, I met him and he his a very nice guy"), or as proactive information ("Look at that guy, I met him and he is really nice"). The transmission of social evaluations permits beneficiaries to enlarge their social network without bearing the costs of direct interactions, potentially harmful, with unknown agents. People can use others' images regarding an unfamiliar target (provided they trust the evaluators) as guidance for their behavior, their choices, an so on. Reputation is a clearly distinct phenomenon, although strictly interrelated with image. More precisely, image is a set of evaluative beliefs about a given target, while reputation is both the process and the effect of transmission of that image. In fact, reputation is a highly dynamic phenomenon in two distinct senses: it is subject to change, especially as an effect of corruption,
errors, deception, etc.; and it emerges as an effect of a multi-level bidirectional process. In particular, it proceeds from the level of individual cognition to the level of social propagation, and from this level back to that of individual cognition again. Reputation is both what people think about targets and what targets are in the eyes of others. From the very moment an agent is targeted by the community, his or her life will change. Reputation has become the immaterial, more powerful equivalent of a scarlet letter sewed to one's clothes. It is more powerful because it may not even be perceived by the individual to whom it sticks, and consequently it is out of the individual's power to control and manipulate. Reputation is an objective social property that emerges from a propagating cognitive representation. The distinctive feature of reputation is its lack of an identified source, that is, its impersonality, whereas image always requires the specification of who made the evaluation. In addition to the three sets of agents involved by image (Evaluator, Target, Beneficiary), reputation implies other ones: Third Parties or Gossipers. A third party is an agent in the position to transmit reputation without being responsible for that evaluation. Third parties do not evaluate, they just transmit evaluations, thus enlarging the social network. An agent is a (potential) third party if she transmits (is in position to transmit) reputation about a target to another agent or set of agents. A third party may be bluffing: he or she may pretend to be benevolent with regard to beneficiaries, in order to be considered as a part of the in-group by the other evaluators, and therefore gain a good reputation without sustaining the costs of its acquisition (as would be implied by performing the socially desirable behaviour), and avoid the consequences of a bad reputation. Agents may also spread a false reputation, i.e., pretending that a target has a given reputation when this is not the case. Agents do this in order to achieve the aforementioned benefits without taking responsibility for spreading a given social evaluation. The difference between personal and impersonal evaluations is not inconsequential: when transmitting image, the evaluator becomes visible; thus her trustworthiness or competence as a source can make the evaluation more reliable for the beneficiaries. At the same time, this exposes
the evaluator to retaliatory actions when the transmitted image is proven to be inaccurate or even false. On the other hand, reputation is anonymous; it can not be attributed to someone but it is something "people say" about somebody, without making explicit the source of the evaluation. Therefore, reputation is less reliable; it can not be traced back to a specific source, but it is usually spread more easily. In the real world, the differences between image and reputation appear very subtle and they are often very difficult to disentangle. Multi-Agent Based Social Simulation (MABSS) allows us to overcome this problem, to check for specific mechanisms and to test single hypotheses. In what follows we will implement a simulation model of an industrial cluster in which agents are firms that exchange both products and social evaluations and we will test whether the former process is influenced by the latter and in what way. At present we will limit our investigation to the effects of the transmission of personal evaluations, i.e. image, on the cluster's performance.
3. Reputation for industrial clusters: a proposal for modelling
The aim of this work is to verify the socio-cognitive theory of reputation in a novel setting, i.e. an industrial cluster, and to test whether and in what way the exchange of social evaluations can be related to the quality of products delivered by an artificial industrial district. We implemented an artificial environment in which agents can choose among several potential suppliers by relying either on their own evaluations, or on other agents' evaluations. In the latter case the availability of truthful information could help agents to find reliable partners without bearing the costs of potentially negative, i.e. harmful, interactions with bad suppliers. We compared this situation (true information available) with another scenario in which no communication was allowed, in order to test the effects of image transmission on the overall quality of firms' production. At this stage we restricted our investigation to image, without considering reputational dynamics. We tried to answer the following questions: How relevant is image when firms need to
select suppliers, service providers and so on? Does transmission of image promote the improvement of quality in a cluster? How does false information affect the quality of the cluster? Our model is characterized by the existence of two different kinds of interactions among agents: material exchange and evaluation exchange. The former refers to the exchange of products between leader firms and their suppliers, and it leads to the creation of a supply chain network. On the other hand, the flows of social evaluations among the firms create a social network. In this setting, agents can transmit true or false evaluations in order to either help or hamper their fellows searching for a good partner. 3.1 Agents The agents of the model are firms. Firms are organized into different layers, in line with their role in the production cycle. The number of layers can vary according to the characteristics of the cluster, but a minimum of two layers is required. Here, we have three layers, but n possible layers can be added, in order to develop a more complex production process:
• • •
Layer 0 (L0) is represented by leader firms that supply the final product Layer 1 (L1) is represented by suppliers of L0 Layer 2 (L2) are firms providing raw material to firms in L1. When image transmission is allowed, both leader firms and suppliers exchange information
with their fellows regarding their suppliers from the level below, thus creating and taking part in a social network. This process works only horizontally: L0 and L1 are not allowed to talk each other. Agents in both layers can play two possible roles:
• •
The Questioner - asks an Informer, i.e. another firm of the same layer, to suggest a good supplier; The Informer - provides her own image of a good supplier. Honest informers suggest their best rated supplier, whereas cheaters transmit the image of their worse supplier (as if it was a good one).
3.2 Actions and Interactions
Agents in L0 have to select suppliers that produce with a quality above the average (Q>0.75) among all L1 agents. Suppliers can be directly tested or they can be chosen thanks to the information received by other L0 firms acting as Informers. Buying products from L1 and asking for information to L0 fellows are competing activities that can not be performed contemporaneously. In turn, once received an order for a product, L1 firms should select a good supplier (above the 0.75 threshold) among those in L2. Also at this level, direct experience or reported evaluations are used to select a partner. L2 agents only supply raw materials when asked for, without either exchanging information or selecting partners. INSERT FIGURE 1 ABOUT HERE Economic transactions2. We model economic variables such as price, cost and profit in a very simple way: the final price (P) is determined by the quality (Q) of that product, that varies in a range between .50 to 1. Leader firms buy from suppliers at at fixed cost (K) that is equal to 0.75, i.e. to the average quality between 0.50 and 1. This average quality is also the threshold between low and high quality products and it works as a reference point for discriminating between good and bad suppliers. Profits (U) are defined as prices minus costs (U= P-K), thus we calculate and record L0 and L1 profits for each transaction. L2 is not taken into consideration since it only sells raw materials without buying anything. Information exchange. After each interaction with a supplier, both L0 and L1 agents create an evaluation, i.e. an image, of it, comparing the quality of the product they bought with the threshold value set at 0.75. If the product's quality exceeds that threshold, the supplier is considered good, otherwise it is labeled as a bad supplier, namely a cheater. Agents are endowed with an "Image Table" in which all the values of the tested partners are recorded and stored for future selections. There are three modalities for partner selection:
• •
2
Experience-based Selection - the best rated supplier among those already tested is chosen; Image-based Selection - the best rated supplier among those suggested by an Informer is
We are indebted to Lucio Biggiero and Enrico Sevi for the modeling of economic transactions.
selected;
•
Random Selection - among the unfamiliar suppliers. At the beginning of the simulation, L0 and L1 firms check whether their "Image Table"
contains a good supplier. The following alternatives are given:
•
The "Image table" contains at least one good rated supplier that is not engaged with another agent (suppliers can deliver one production item per temporal unit). That supplier is chosen. During the simulation, the Image Table is progressively filled with suppliers. Good suppliers are listed according to their quality, therefore the first available good supplier is chosen.
•
The list is empty, or it contains only good rated suppliers that are not available. When the Image Table is useless (because it is empty or the known good suppliers are not available), agents can perform one of three alternative actions: o Asking their fellows. The Questioner asks 10% of firms of the same layer rating the recommended suppliers into the "Candidate Image Table" according to the number of coinciding suggestions. After the suggestion has been tested, the questioner updates both the Image Table (regarding the supplier) and the Informer Table (according to the suggestion). In the latter list, honest informers will be labeled as "+1", while cheaters will receive a "1" and they will not be interrogated anymore. o Looking for a supplier in the "Candidate Image Table", i.e. a record of all those suppliers previously suggested by other firms but not tested yet. The higher the number of suggestions the higher the priority of that supplier. In the case of a tie ranking the firm chooses randomly among all those potential suppliers with the same score. When the Candidate Image Table is exhausted, the agent interrogates her fellows (10% of the firms of the layer), avoiding bad Informers, to know other good suppliers.
o Randomly seeking a supplier. This happens mostly at the beginning of the simulation, when leader firms' tables are still empty and no suggestions are possible. While exploring the social environment, agents are driven by the possibility to make higher profits. At each simulation cycle, firms attempt to interact with the best known suppliers. Every time the best known supplier is unavailable they query their fellows about other high quality suppliers, which will be tested and integrated into the Image-Table. Whenever a higher quality fim is discovered, it becomes the new best known supplier for the following market interactions. 3.3 Scenarios We ran the experiments in two different conditions : a) Scenario A with communication: agents exchanged true or false images. The percentage of cheaters was set by a "cheating rate" parameter: the higher the value of the parameter the higher the number of cheaters in the cluster. This scenario was enriched with the inclusion of retaliation, i.e. the possibility of sending false information in response to false images previously received. When retaliation was on, honest agents responded to cheaters, previously detected, with false information, thus punishing them. On the contrary, when retaliation was not allowed, honest agents always transmitted true information, even to known cheaters. b) Scenario B without communication. In this case, agents were not allowed to communicate and the suppliers' choice was exclusively experience-based. 3.4 Data Structure and Simulation Features Data of companies, products and transactions are loaded into a relational database. At the moment, data for companies and products are randomly generated, but loading of data from real
industrial clusters are expected from the ongoing of the SOCRATE 3 project. In brief, data consist of:
• • • • •
Firms data: identification number, layer, mean quality of production Products data: identification number, layer, market price Economic data: identification number of product and identification number of company that produces it, profits, costs, prices Production orders: identification number of company that receives order, customer (if any), supplier (if any), quality, cost, price etc. Social Information: for each company (same ID) a list of records for others companies with data about image of production quality, credibility as informer etc.
To initialize the simulation we set the following constraints:
• • • • •
300 agents 3 layers with a ratio of 20% L0 agents and 80% suppliers (half L1 agents, half L2 agents) suppliers' production quality has a uniform random distribution ranging between 0.5 and 1.0 each supplier can deliver its product to one customer only once per computation cycle for each computation cycle, each leader firm sends a production order and receives one production item.
Results
We tested the agents' performance in terms of average quality of production and profits, both for single layers and for the cluster as a whole. In the first scenario, in which communication was available (Scenario A), we did not find significant differences in quality of production between
3
SOCRATE is an interdisciplinary research project funded by the Italian Ministry of University and Scientific Research. It involves six partners and is aimed at investigating how reputation affects economic exchanges in industrial districts. A special attention is paid to the italian aerospace cluster.
clusters with different percentages of cheating rate. Average quality was above the 0.75 threshold and around 0.85 in all cases, without appreciable differences. When retaliation, i.e. sending false information to punish dishonest informers, was added, the average quality of the cluster showed a little but not remarkable decrease. On the contrary, it is of some interest the difference between quality levels in Scenario A and Scenario B (Fig. 2 and Fig. 3). In the former Scenario, quality levels reached high values in few ticks, thank to reliable information, whereas without communications both front end (L0 firms) and suppliers (L1firms) needed much more time to find the best partners. INSERT FIGURE 2 ABOUT HERE INSERT FIGURE 3 ABOUT HERE A positive effect of trustful communication was found also on profits. When agents passed on true information (Fig. 4), profits grew in an impressive manner and they were higher, on average, compared to the situation in which all evaluations were false (Fig. 5). INSERT FIGURE 4 ABOUT HERE INSERT FIGURE 5 ABOUT HERE Even more interesting is the comparison between the above scenario and Scenario B, in which agents did not communicate. When only Experience-based selection was possible, gains grew following a different trend, as showed in Figure 6. Here, we can see that maximum profits are lower than those reached in the True information condition and their growth is smoother. True information affected levels of profits for both types of firms in the cluster, leading agents to get better gains. INSERT FIGURE 6 ABOUT HERE
4. Discussion
Although preliminary, this study shows that social evaluations and their reliability matter in a computer-based simulation of an industrial cluster with three layers of firms. Social evaluations, when truthful, help agents to find good suppliers and then to increase their profits. It is especially
interesting to notice the difference in levels of profits but also in their curves. Receiving true evaluations ameliorate performances but it also accelerates the profits growth, compared to the situation without communication. Furthermore, our results show that trusting others' evaluations, provided these are true, leads to better performances in comparison with the situation in which agents rely exclusively on their own images. Honest agents belonging to a social network find good suppliers more easily than lonely agents. Unfortunately, the lack of significance of results regarding levels of quality in the True and False Image conditions does not allow us to make any conclusive remarks on this aspect. A possible explanation can be found in the different size of the layers. Here, for each front-end firm there were two possible suppliers, whereas L1 suppliers were in a 1:1 proportion with L2 firms. In future works we will explore the effects of different populations' distribution on the quality levels of single layers and cluster.
5. Concluding remarks
We presented a proposal for modelling the effects of transmission of social evaluations in an industrial cluster. Given the assumption that in this a "small-world", as in the real world, evaluations are crucial to selecting trustworthy partners and to isolating cheaters. We tried to demonstrate how useful this exchange is, especially in terms of global cluster quality and profits. Firms receiving reliable information about potential partners found good suppliers in a faster and more efficient way, compared to firms that were systematically cheated by their fellows. More interesting results are expected after we include an enriched economic structure and the implementation of reputation. We acknowledge that the structure of the cluster we designed is really basic and that further improvements are needed, regarding both economic plausibility and cognitive refinement. In fact, the lack of a true cognitive architecture prevented the possibility of exploring more interesting ways of implementing agents' decision making, since agents' strategies varied only according to the cheating rate.
The work described above is part of a research project ( SOCRATE) on the effects of reputation in industrial districts of small and mid-sized enterprises. At this stage of development of the project, the most interesting part of the story, i.e. the effects of reputation and the passing on of voices and rumours (untested evaluations) on the quality of production, are not yet available. Although preliminary, present findings are of some interest for the governance of social and economic structures and processes. In particular, they show that communication of image adds value to one's own experience of contract partners: it is better to cooperate in spreading the results of direct experience that rely only on experience.
Acknowledgements
This work was partially supported by the European Community under the FP6 programme (eRep Project, contract number CIT5-028575), and by the Italian Ministry of University and Scientific Research under the FIRB programme (Socrate Project, contract number RBNE03Y338).
References
[1] V. Albino, N. Carbonara and I. Giannoccaro, Cooperation and Competition within Industrial Districts: An Agent-Based Computational Approach, Journal of Artificial Societies and Social Simulation 6 (4) (2003) <http://jasss.soc.surrey.ac.uk/6/4/3.html>. [2] V. Albino, N. Carbonara and I. Giannoccaro, Innovation in Industrial Districts: An Agent-Based Simulation Model, International Journal of Production Economics, 104 (2006) 30-45. [3] R. F. Baumeister, L. Zhang and K. D. Vohs, Gossip as Cultural Learning, Review of General Psychology,Vol. 8 (2004) 111-121. [4] T. Brenner, Simulating the evolution of localised industrial clusters—an identification of the basic mechanism, Journal of Artificial Societies and Social Simulation 4 (3) 2001 <http://jasss.soc.surrey.ac.uk/4/3/1.html>. [5] R. Conte, M. Paolucci, Reputation in Artificial Societies. Social Beliefs for Social Order (Kluwer, Boston, 2002). [6] R. Conte, M. Paolucci, Social Cognitive Factors of Unfair Ratings in Reputation Reporting Systems. Proceedings of the IEEE/WIC International Conference on Web Intelligence (WI03 2003). [7] K. Cravens, E. Goad Oliver, and S. Ramamoorti, The reputation Index: Measuring and Managing Corporate Reputation, European Management Journal 21 (2) (2003) 201-212. [8] R. I. M. Dunbar, Gossip in Evolutionary Perspective, Review of General Psychology Vol. 8 (2004) 100-110. [9] M. Enright, Regional Clusters and Firm Strategy, in: A.D. Chandler, P. Hagstrom and O. Solvell, eds., The Dynamic Firm: The Role of Technology, Organization and Regions. (Oxford University Press, Oxford, 1998). [10] G. Fioretti, Agent Based Models of Industrial Clusters and Districts. Contemporary Issues in Urban and Regional Economics (2005). [11] E. Fombrun, Reputation. Realizing Value from The Corporate Image (Harvard Business School Press, Boston, 1996). [12] F. Giardini, R. Conte, M. Paolucci, Reputation for Complex Societies, in: S. Moss and B. Edmonds, eds., Simulating Social Complexity (In press). [13] N. Gilbert, J. Doran (eds.) Simulating Societies. The Computer Simulation of Social Phenomena (UCL Press, London, 1994). [14] N. Gilbert, P. Terna, How to Build and Use Agent-Based Models in Social Science, Mind &Society, I, 1 (2000) 57-72. [15] E.R. Gray, J.M.T. Balmer, Managing Corporate Image and Corporate Reputation. Long Rage Planning, 31(5) (1998) 695-702. [16] M. Porter, Clusters and the New Economics of Competition, Harvard Business Review, 76
(1998) 77-90. [17] P. Resnick, R. Zeckhauser, J. Swanson, K. Lockwood, The Value of Reputation on eBay: A Controlled Experiment, Experimental Economics, 9 (2) (2006) 79-101. [18] C. Rose, S. Thomsen, The Impact of Corporate Reputation on Performance: Some Danish Evidence, European Mangement Journal, 22(2) (2004) 201-210. [19] F. Squazzoni, R. Boero, Economic Performance, Inter-Firm Relations and Local Institutional Engineering in a Computational Prototype of Industrial Districts, Journal of Artificial Societies and Social Simulation, vol. 5, no. 1 (2002) <http://jasss.soc.surrey.ac.uk/5/1/1.html> [20] S.R. Wert, P. Salovey, A Social Comparison Account of Gossip, Review of General Psychology, Vol. 8 (2004) 122-137.
LIST OF FIGURES
FIG. 1. A flow chart explaining the behaviour of agents in the model.
FIG. 2. SCENARIO A: Average quality of production when True Communication was available.
FIG. 3. SCENARIO B: Average quality of production in a cluster without communication.
FIG. 4. SCENARIO A: True Image Situation. Profits of Front End and Supplier firms dramatically increased when all agents exchanged true information.
FIG. 5. SCENARIO A: False Image Situation. Untruthful evaluations negatively affected the
profits. When all agents transmitted false evaluations profits lowered in comparison with a cluster with reliable information. This is true for both layers of the simulation model.
FIG.6. SCENARIO B: No communication. Without communication profits were lower and their growth was slower and less steep.
doc_555443587.docx
Industrial district was initially introduced as a term to describe an area where workers of a monolithic heavy industry (ship-building, coal mining, steel, ceramics, etc.) live within walking-distance of their places of work.
Research Study on a Model for Simulating Reputation Dynamics in Industrial Districts
Abstract
In this work we try to draw an interdisciplinary framework aimed to integrate a socio-cognitive approach with organizational research about industrial clusters, in order to investigate whether and how social evaluations may affect clusters' dynamics. Industrial districts are sort of "small-worlds" that provide a suitable environment for testing the predictions of our model through Multi-Agent Based Social Simulation (MABSS) experiments. Artificial agents will be allowed to exchange products and to transmit social evaluations, and the main features of a cluster will be implemented. The effects of evaluation transmission on products' quality, partner selection and cheaters' isolation will be discussed.
Keywords: reputation; industrial clusters; social simulation; cognitive modelling.
1. Introduction
The exchange of social evaluations is an extensive activity in human societies. Individuals create, manipulate, modify, and circulate these evaluations, using them to inform, to convince and also to deceive other individuals. This is true also at the supra-individual level, in which institutions, countries, corporations and firms, just to name some, devote a lot of resources to acquire and preserve good reputations. Moving from the individual to the collective level, the complexity of this phenomenon increases, and a simulation model is required to deal with this
complexity. In this work we propose a model for simulating the effects of reputation on profits and quality of production in an industrial cluster. Industrial districts1 are geographically defined production systems [1], common around the world (see the geographical clusters in the US or the milieux innovateur in France). According to Porter, "clusters are geographic concentrations of interconnected companies, specialized suppliers, service providers, firms in related industries, and associated institutions (for example, universities, standard agencies, trade associations) in a particular field that compete but also cooperate" [16]. These networks of firms, suppliers, and institutions have been investigated by several different disciplines, ranging from industrial organization to economic geography and other social sciences. These studies have developed different notions and models, in order to understand the main features of these clusters and to develop effective tools for the amelioration of performance and competitiveness. One key issue in these kinds of studies is innovation, which has been investigated mainly in terms of learning processes [9], allowing firms to improve their knowledge, to develop their know-how and to get good innovation levels. However, better performance could be not recognized as such if other agents do not hold positive evaluations about that cluster, i.e. if it has a bad reputation. Computer-based simulations of industrial clusters are not new: different models have been put forward [10, 19] and several aspects have been addressed, such as their innovation dynamics
1
In this work district and cluster are used as synonyms.
[2], the importance of relationships among firms and institutions [18], and also the evolution of the cluster itself [4]. All these studies share the same interest in economic features of firms and in their spatial agglomeration, which are considered as key factors in determining clusters' competitiveness and innovation strategies. Starting from similar research questions, we want to model the relationships among firms in a quite different way, focusing on social links among agents and on the rich social structure, usually informal, that is a defining feature of industrial clusters. Social evaluations are the building blocks of social and economic relationships inside the cluster; they are used to select trustworthy partners, to create and enlarge the social network and to exert social control on cheaters. In this work we explore the contribution that a cognitive theory of reputation [5, 6] may provide to the studies about industrial clusters and their dynamics. This cognitive approach considers reputation as a highly dynamic process that is deeply rooted in social exchanges and implies several consequences for the subjects involved. Moreover, this model allows us to distinguish between one's own evaluations about other people, namely image, and reputation, which is impersonal. This distinction is not inconsequential and gives rise to different phenomena and dynamic effects that our model is intended to investigate. To test our model in this complex environment we use an agent-based simulation approach, i.e. a computer-based way of formalizing interacting agents in a given environment. Multi-AgentBased Social Simulation (MABSS) is a scientific methodology consisting in the creation of models of artificial societies, i.e. sets of autonomous agents planning, acting and evaluating their actions in an artificial environment with given features. The validity of this methodology has been widely recognized [12, 13, 14]. MABSS is a powerful and flexible tool: it allows the investigation of complex environments where macro-behaviors arise from micro-behaviors, and the resulting properties are not known a priori, as it is the case with reputation. In Section 1 we will put forward the theoretical background and we will discuss the importance of reputation for both natural and artificial societies. The simulation model will be
presented in Section 2 along with some hypotheses regarding the effects of social evaluations in simulated industrial clusters. The scenario and the experiments will be presented in the third section. Finally, a discussion about the model and its results will be provided.
2. Reputation: from natural to artificial societies
If one were to name one of the most influential and recurrent social phenomena, reputation would undoubtedly be one of them. In human societies the exchange of social evaluations is crucial to partner selection, social control, and coalition formation, just to name some of the main functions of reputation. Holding a good reputation is really important and, in some contexts, it is essential to survival, such for instance in competitive markets, in which stakeholders' decisions are often rooted in the firm's reputation [7, 15]. In everyday life, reputation works as a compass to avoid dangerous partnerships and to find reliable collaborators, in both small and large social groups. Reputation is an influential theme also in research, as witnessed by the huge amount of studies carried out on this topic by scholars coming from different disciplines, ranging from psychology [ 3, 20], to management [11, 18], to economics [17] and even to biology [8]. Often, these studies are limited to the specific domain of application they refer to, thus providing neither general explanations nor accepted results. This is partially due to the multiplicity of aspects concerning reputation, and to the difficulty to put all these strands together. An attempt to yield a more comprehensive theory of reputation has been made by Conte and Paolucci [5], who used a socio-cognitive framework to describe how people create, manipulate and transmit social evaluations, and how these evalutions affect individuals' beliefs and behaviours. The social cognitive model is a dynamic approach that considers reputation as the output of a social process that starts in agents' minds. The input to this process is the evaluation that agents (Evaluators) directly form about a given Target during interaction or observation. This evaluation can be transmitted to Beneficiaries that share the goal with regard to which targets are evaluated and thus may use this information as a guide for their behaviour. We distinguish between two kinds
of social evaluations: image and reputation. Image is the output of a process of evaluation regarding another agent. In social life, people continuosly assess their colleagues, friends, partners, etc. with regard to their personal features, competences, behaviors and so on. These evaluations are the social images of those agents. In other words, image is an assessment of the positive or negative qualities of a target with regard to a norm, a competence, and so on. The main distinctive feature of image is its being personal, i.e. the identity of the evaluator is always clearly expressed, as for instance in the sentence "I believe John is a good guy". A given image includes three sets of agents: 1. a set of agents who share the evaluation (E) - Evaluators 2. an evaluation target (T) - Target 3. a set of Beneficiaries (B), i.e., the agents sharing the goal with regard to which the targets are evaluated. Often, evaluators and beneficiaries coincide. Social evaluations may concern physical, mental, and social properties of targets; agents may evaluate a target as to both his or her capacity and willingness to achieve a shared goal. After assessing someone, the evaluator may transmit this image to someone else, in response to a specific request (e.g., "Do you know that guy?", "Yes, I met him and he his a very nice guy"), or as proactive information ("Look at that guy, I met him and he is really nice"). The transmission of social evaluations permits beneficiaries to enlarge their social network without bearing the costs of direct interactions, potentially harmful, with unknown agents. People can use others' images regarding an unfamiliar target (provided they trust the evaluators) as guidance for their behavior, their choices, an so on. Reputation is a clearly distinct phenomenon, although strictly interrelated with image. More precisely, image is a set of evaluative beliefs about a given target, while reputation is both the process and the effect of transmission of that image. In fact, reputation is a highly dynamic phenomenon in two distinct senses: it is subject to change, especially as an effect of corruption,
errors, deception, etc.; and it emerges as an effect of a multi-level bidirectional process. In particular, it proceeds from the level of individual cognition to the level of social propagation, and from this level back to that of individual cognition again. Reputation is both what people think about targets and what targets are in the eyes of others. From the very moment an agent is targeted by the community, his or her life will change. Reputation has become the immaterial, more powerful equivalent of a scarlet letter sewed to one's clothes. It is more powerful because it may not even be perceived by the individual to whom it sticks, and consequently it is out of the individual's power to control and manipulate. Reputation is an objective social property that emerges from a propagating cognitive representation. The distinctive feature of reputation is its lack of an identified source, that is, its impersonality, whereas image always requires the specification of who made the evaluation. In addition to the three sets of agents involved by image (Evaluator, Target, Beneficiary), reputation implies other ones: Third Parties or Gossipers. A third party is an agent in the position to transmit reputation without being responsible for that evaluation. Third parties do not evaluate, they just transmit evaluations, thus enlarging the social network. An agent is a (potential) third party if she transmits (is in position to transmit) reputation about a target to another agent or set of agents. A third party may be bluffing: he or she may pretend to be benevolent with regard to beneficiaries, in order to be considered as a part of the in-group by the other evaluators, and therefore gain a good reputation without sustaining the costs of its acquisition (as would be implied by performing the socially desirable behaviour), and avoid the consequences of a bad reputation. Agents may also spread a false reputation, i.e., pretending that a target has a given reputation when this is not the case. Agents do this in order to achieve the aforementioned benefits without taking responsibility for spreading a given social evaluation. The difference between personal and impersonal evaluations is not inconsequential: when transmitting image, the evaluator becomes visible; thus her trustworthiness or competence as a source can make the evaluation more reliable for the beneficiaries. At the same time, this exposes
the evaluator to retaliatory actions when the transmitted image is proven to be inaccurate or even false. On the other hand, reputation is anonymous; it can not be attributed to someone but it is something "people say" about somebody, without making explicit the source of the evaluation. Therefore, reputation is less reliable; it can not be traced back to a specific source, but it is usually spread more easily. In the real world, the differences between image and reputation appear very subtle and they are often very difficult to disentangle. Multi-Agent Based Social Simulation (MABSS) allows us to overcome this problem, to check for specific mechanisms and to test single hypotheses. In what follows we will implement a simulation model of an industrial cluster in which agents are firms that exchange both products and social evaluations and we will test whether the former process is influenced by the latter and in what way. At present we will limit our investigation to the effects of the transmission of personal evaluations, i.e. image, on the cluster's performance.
3. Reputation for industrial clusters: a proposal for modelling
The aim of this work is to verify the socio-cognitive theory of reputation in a novel setting, i.e. an industrial cluster, and to test whether and in what way the exchange of social evaluations can be related to the quality of products delivered by an artificial industrial district. We implemented an artificial environment in which agents can choose among several potential suppliers by relying either on their own evaluations, or on other agents' evaluations. In the latter case the availability of truthful information could help agents to find reliable partners without bearing the costs of potentially negative, i.e. harmful, interactions with bad suppliers. We compared this situation (true information available) with another scenario in which no communication was allowed, in order to test the effects of image transmission on the overall quality of firms' production. At this stage we restricted our investigation to image, without considering reputational dynamics. We tried to answer the following questions: How relevant is image when firms need to
select suppliers, service providers and so on? Does transmission of image promote the improvement of quality in a cluster? How does false information affect the quality of the cluster? Our model is characterized by the existence of two different kinds of interactions among agents: material exchange and evaluation exchange. The former refers to the exchange of products between leader firms and their suppliers, and it leads to the creation of a supply chain network. On the other hand, the flows of social evaluations among the firms create a social network. In this setting, agents can transmit true or false evaluations in order to either help or hamper their fellows searching for a good partner. 3.1 Agents The agents of the model are firms. Firms are organized into different layers, in line with their role in the production cycle. The number of layers can vary according to the characteristics of the cluster, but a minimum of two layers is required. Here, we have three layers, but n possible layers can be added, in order to develop a more complex production process:
• • •
Layer 0 (L0) is represented by leader firms that supply the final product Layer 1 (L1) is represented by suppliers of L0 Layer 2 (L2) are firms providing raw material to firms in L1. When image transmission is allowed, both leader firms and suppliers exchange information
with their fellows regarding their suppliers from the level below, thus creating and taking part in a social network. This process works only horizontally: L0 and L1 are not allowed to talk each other. Agents in both layers can play two possible roles:
• •
The Questioner - asks an Informer, i.e. another firm of the same layer, to suggest a good supplier; The Informer - provides her own image of a good supplier. Honest informers suggest their best rated supplier, whereas cheaters transmit the image of their worse supplier (as if it was a good one).
3.2 Actions and Interactions
Agents in L0 have to select suppliers that produce with a quality above the average (Q>0.75) among all L1 agents. Suppliers can be directly tested or they can be chosen thanks to the information received by other L0 firms acting as Informers. Buying products from L1 and asking for information to L0 fellows are competing activities that can not be performed contemporaneously. In turn, once received an order for a product, L1 firms should select a good supplier (above the 0.75 threshold) among those in L2. Also at this level, direct experience or reported evaluations are used to select a partner. L2 agents only supply raw materials when asked for, without either exchanging information or selecting partners. INSERT FIGURE 1 ABOUT HERE Economic transactions2. We model economic variables such as price, cost and profit in a very simple way: the final price (P) is determined by the quality (Q) of that product, that varies in a range between .50 to 1. Leader firms buy from suppliers at at fixed cost (K) that is equal to 0.75, i.e. to the average quality between 0.50 and 1. This average quality is also the threshold between low and high quality products and it works as a reference point for discriminating between good and bad suppliers. Profits (U) are defined as prices minus costs (U= P-K), thus we calculate and record L0 and L1 profits for each transaction. L2 is not taken into consideration since it only sells raw materials without buying anything. Information exchange. After each interaction with a supplier, both L0 and L1 agents create an evaluation, i.e. an image, of it, comparing the quality of the product they bought with the threshold value set at 0.75. If the product's quality exceeds that threshold, the supplier is considered good, otherwise it is labeled as a bad supplier, namely a cheater. Agents are endowed with an "Image Table" in which all the values of the tested partners are recorded and stored for future selections. There are three modalities for partner selection:
• •
2
Experience-based Selection - the best rated supplier among those already tested is chosen; Image-based Selection - the best rated supplier among those suggested by an Informer is
We are indebted to Lucio Biggiero and Enrico Sevi for the modeling of economic transactions.
selected;
•
Random Selection - among the unfamiliar suppliers. At the beginning of the simulation, L0 and L1 firms check whether their "Image Table"
contains a good supplier. The following alternatives are given:
•
The "Image table" contains at least one good rated supplier that is not engaged with another agent (suppliers can deliver one production item per temporal unit). That supplier is chosen. During the simulation, the Image Table is progressively filled with suppliers. Good suppliers are listed according to their quality, therefore the first available good supplier is chosen.
•
The list is empty, or it contains only good rated suppliers that are not available. When the Image Table is useless (because it is empty or the known good suppliers are not available), agents can perform one of three alternative actions: o Asking their fellows. The Questioner asks 10% of firms of the same layer rating the recommended suppliers into the "Candidate Image Table" according to the number of coinciding suggestions. After the suggestion has been tested, the questioner updates both the Image Table (regarding the supplier) and the Informer Table (according to the suggestion). In the latter list, honest informers will be labeled as "+1", while cheaters will receive a "1" and they will not be interrogated anymore. o Looking for a supplier in the "Candidate Image Table", i.e. a record of all those suppliers previously suggested by other firms but not tested yet. The higher the number of suggestions the higher the priority of that supplier. In the case of a tie ranking the firm chooses randomly among all those potential suppliers with the same score. When the Candidate Image Table is exhausted, the agent interrogates her fellows (10% of the firms of the layer), avoiding bad Informers, to know other good suppliers.
o Randomly seeking a supplier. This happens mostly at the beginning of the simulation, when leader firms' tables are still empty and no suggestions are possible. While exploring the social environment, agents are driven by the possibility to make higher profits. At each simulation cycle, firms attempt to interact with the best known suppliers. Every time the best known supplier is unavailable they query their fellows about other high quality suppliers, which will be tested and integrated into the Image-Table. Whenever a higher quality fim is discovered, it becomes the new best known supplier for the following market interactions. 3.3 Scenarios We ran the experiments in two different conditions : a) Scenario A with communication: agents exchanged true or false images. The percentage of cheaters was set by a "cheating rate" parameter: the higher the value of the parameter the higher the number of cheaters in the cluster. This scenario was enriched with the inclusion of retaliation, i.e. the possibility of sending false information in response to false images previously received. When retaliation was on, honest agents responded to cheaters, previously detected, with false information, thus punishing them. On the contrary, when retaliation was not allowed, honest agents always transmitted true information, even to known cheaters. b) Scenario B without communication. In this case, agents were not allowed to communicate and the suppliers' choice was exclusively experience-based. 3.4 Data Structure and Simulation Features Data of companies, products and transactions are loaded into a relational database. At the moment, data for companies and products are randomly generated, but loading of data from real
industrial clusters are expected from the ongoing of the SOCRATE 3 project. In brief, data consist of:
• • • • •
Firms data: identification number, layer, mean quality of production Products data: identification number, layer, market price Economic data: identification number of product and identification number of company that produces it, profits, costs, prices Production orders: identification number of company that receives order, customer (if any), supplier (if any), quality, cost, price etc. Social Information: for each company (same ID) a list of records for others companies with data about image of production quality, credibility as informer etc.
To initialize the simulation we set the following constraints:
• • • • •
300 agents 3 layers with a ratio of 20% L0 agents and 80% suppliers (half L1 agents, half L2 agents) suppliers' production quality has a uniform random distribution ranging between 0.5 and 1.0 each supplier can deliver its product to one customer only once per computation cycle for each computation cycle, each leader firm sends a production order and receives one production item.
Results
We tested the agents' performance in terms of average quality of production and profits, both for single layers and for the cluster as a whole. In the first scenario, in which communication was available (Scenario A), we did not find significant differences in quality of production between
3
SOCRATE is an interdisciplinary research project funded by the Italian Ministry of University and Scientific Research. It involves six partners and is aimed at investigating how reputation affects economic exchanges in industrial districts. A special attention is paid to the italian aerospace cluster.
clusters with different percentages of cheating rate. Average quality was above the 0.75 threshold and around 0.85 in all cases, without appreciable differences. When retaliation, i.e. sending false information to punish dishonest informers, was added, the average quality of the cluster showed a little but not remarkable decrease. On the contrary, it is of some interest the difference between quality levels in Scenario A and Scenario B (Fig. 2 and Fig. 3). In the former Scenario, quality levels reached high values in few ticks, thank to reliable information, whereas without communications both front end (L0 firms) and suppliers (L1firms) needed much more time to find the best partners. INSERT FIGURE 2 ABOUT HERE INSERT FIGURE 3 ABOUT HERE A positive effect of trustful communication was found also on profits. When agents passed on true information (Fig. 4), profits grew in an impressive manner and they were higher, on average, compared to the situation in which all evaluations were false (Fig. 5). INSERT FIGURE 4 ABOUT HERE INSERT FIGURE 5 ABOUT HERE Even more interesting is the comparison between the above scenario and Scenario B, in which agents did not communicate. When only Experience-based selection was possible, gains grew following a different trend, as showed in Figure 6. Here, we can see that maximum profits are lower than those reached in the True information condition and their growth is smoother. True information affected levels of profits for both types of firms in the cluster, leading agents to get better gains. INSERT FIGURE 6 ABOUT HERE
4. Discussion
Although preliminary, this study shows that social evaluations and their reliability matter in a computer-based simulation of an industrial cluster with three layers of firms. Social evaluations, when truthful, help agents to find good suppliers and then to increase their profits. It is especially
interesting to notice the difference in levels of profits but also in their curves. Receiving true evaluations ameliorate performances but it also accelerates the profits growth, compared to the situation without communication. Furthermore, our results show that trusting others' evaluations, provided these are true, leads to better performances in comparison with the situation in which agents rely exclusively on their own images. Honest agents belonging to a social network find good suppliers more easily than lonely agents. Unfortunately, the lack of significance of results regarding levels of quality in the True and False Image conditions does not allow us to make any conclusive remarks on this aspect. A possible explanation can be found in the different size of the layers. Here, for each front-end firm there were two possible suppliers, whereas L1 suppliers were in a 1:1 proportion with L2 firms. In future works we will explore the effects of different populations' distribution on the quality levels of single layers and cluster.
5. Concluding remarks
We presented a proposal for modelling the effects of transmission of social evaluations in an industrial cluster. Given the assumption that in this a "small-world", as in the real world, evaluations are crucial to selecting trustworthy partners and to isolating cheaters. We tried to demonstrate how useful this exchange is, especially in terms of global cluster quality and profits. Firms receiving reliable information about potential partners found good suppliers in a faster and more efficient way, compared to firms that were systematically cheated by their fellows. More interesting results are expected after we include an enriched economic structure and the implementation of reputation. We acknowledge that the structure of the cluster we designed is really basic and that further improvements are needed, regarding both economic plausibility and cognitive refinement. In fact, the lack of a true cognitive architecture prevented the possibility of exploring more interesting ways of implementing agents' decision making, since agents' strategies varied only according to the cheating rate.
The work described above is part of a research project ( SOCRATE) on the effects of reputation in industrial districts of small and mid-sized enterprises. At this stage of development of the project, the most interesting part of the story, i.e. the effects of reputation and the passing on of voices and rumours (untested evaluations) on the quality of production, are not yet available. Although preliminary, present findings are of some interest for the governance of social and economic structures and processes. In particular, they show that communication of image adds value to one's own experience of contract partners: it is better to cooperate in spreading the results of direct experience that rely only on experience.
Acknowledgements
This work was partially supported by the European Community under the FP6 programme (eRep Project, contract number CIT5-028575), and by the Italian Ministry of University and Scientific Research under the FIRB programme (Socrate Project, contract number RBNE03Y338).
References
[1] V. Albino, N. Carbonara and I. Giannoccaro, Cooperation and Competition within Industrial Districts: An Agent-Based Computational Approach, Journal of Artificial Societies and Social Simulation 6 (4) (2003) <http://jasss.soc.surrey.ac.uk/6/4/3.html>. [2] V. Albino, N. Carbonara and I. Giannoccaro, Innovation in Industrial Districts: An Agent-Based Simulation Model, International Journal of Production Economics, 104 (2006) 30-45. [3] R. F. Baumeister, L. Zhang and K. D. Vohs, Gossip as Cultural Learning, Review of General Psychology,Vol. 8 (2004) 111-121. [4] T. Brenner, Simulating the evolution of localised industrial clusters—an identification of the basic mechanism, Journal of Artificial Societies and Social Simulation 4 (3) 2001 <http://jasss.soc.surrey.ac.uk/4/3/1.html>. [5] R. Conte, M. Paolucci, Reputation in Artificial Societies. Social Beliefs for Social Order (Kluwer, Boston, 2002). [6] R. Conte, M. Paolucci, Social Cognitive Factors of Unfair Ratings in Reputation Reporting Systems. Proceedings of the IEEE/WIC International Conference on Web Intelligence (WI03 2003). [7] K. Cravens, E. Goad Oliver, and S. Ramamoorti, The reputation Index: Measuring and Managing Corporate Reputation, European Management Journal 21 (2) (2003) 201-212. [8] R. I. M. Dunbar, Gossip in Evolutionary Perspective, Review of General Psychology Vol. 8 (2004) 100-110. [9] M. Enright, Regional Clusters and Firm Strategy, in: A.D. Chandler, P. Hagstrom and O. Solvell, eds., The Dynamic Firm: The Role of Technology, Organization and Regions. (Oxford University Press, Oxford, 1998). [10] G. Fioretti, Agent Based Models of Industrial Clusters and Districts. Contemporary Issues in Urban and Regional Economics (2005). [11] E. Fombrun, Reputation. Realizing Value from The Corporate Image (Harvard Business School Press, Boston, 1996). [12] F. Giardini, R. Conte, M. Paolucci, Reputation for Complex Societies, in: S. Moss and B. Edmonds, eds., Simulating Social Complexity (In press). [13] N. Gilbert, J. Doran (eds.) Simulating Societies. The Computer Simulation of Social Phenomena (UCL Press, London, 1994). [14] N. Gilbert, P. Terna, How to Build and Use Agent-Based Models in Social Science, Mind &Society, I, 1 (2000) 57-72. [15] E.R. Gray, J.M.T. Balmer, Managing Corporate Image and Corporate Reputation. Long Rage Planning, 31(5) (1998) 695-702. [16] M. Porter, Clusters and the New Economics of Competition, Harvard Business Review, 76
(1998) 77-90. [17] P. Resnick, R. Zeckhauser, J. Swanson, K. Lockwood, The Value of Reputation on eBay: A Controlled Experiment, Experimental Economics, 9 (2) (2006) 79-101. [18] C. Rose, S. Thomsen, The Impact of Corporate Reputation on Performance: Some Danish Evidence, European Mangement Journal, 22(2) (2004) 201-210. [19] F. Squazzoni, R. Boero, Economic Performance, Inter-Firm Relations and Local Institutional Engineering in a Computational Prototype of Industrial Districts, Journal of Artificial Societies and Social Simulation, vol. 5, no. 1 (2002) <http://jasss.soc.surrey.ac.uk/5/1/1.html> [20] S.R. Wert, P. Salovey, A Social Comparison Account of Gossip, Review of General Psychology, Vol. 8 (2004) 122-137.
LIST OF FIGURES
FIG. 1. A flow chart explaining the behaviour of agents in the model.
FIG. 2. SCENARIO A: Average quality of production when True Communication was available.
FIG. 3. SCENARIO B: Average quality of production in a cluster without communication.
FIG. 4. SCENARIO A: True Image Situation. Profits of Front End and Supplier firms dramatically increased when all agents exchanged true information.
FIG. 5. SCENARIO A: False Image Situation. Untruthful evaluations negatively affected the
profits. When all agents transmitted false evaluations profits lowered in comparison with a cluster with reliable information. This is true for both layers of the simulation model.
FIG.6. SCENARIO B: No communication. Without communication profits were lower and their growth was slower and less steep.
doc_555443587.docx